UNIVERSIT DE LA MDITERRANE (AIX-MARSEILLE II)

Facult des Sciences Economiques et de Gestion

Ecole Doctorale de Sciences Economiques et de Gestion dAix-Marseille n372
Anne 2010 Numro attribu par la bibliothque
| | | | | | | | | | | |
Thse pour le Doctorat s Sciences Economiques
Prsente et soutenue publiquement par
Zakaria MOUSSA
le 6 dcembre 2010

Assouplissement quantitatif ; quels enseignements

tirer de lexprience japonaise ?

Directeur de Thse
M. Eric GIRARDIN, Professeur lUniversit de la Mditerrane, GREQAM
Jury
Rapporteurs
M. Patrick FVE Professeur luniversit de Toulouse I, GREMAQ
M. Andrew J. FILARDO Economiste en Chef, Banque des Rglements
Internationaux, zone AsiePacique
Examinateurs
M. Gilles DUFRNOT Professeur lUniversit dAix-Marseille 2, DEFI
M. Michel LUBRANO Directeur de recherche CNRS, GREQAM,
M. Benot MOJON Banque de France, Chef du service de recherche
sur la politique montaire
UNIVERSIT DE LA MDITERRANE (AIX-MARSEILLE II)
Facult des Sciences Economiques et de Gestion
Ecole Doctorale de Sciences Economiques et de Gestion dAix-Marseille n372
Anne 2010 Numro attribu par la bibliothque
| | | | | | | | | | | |
Thse pour le Doctorat s Sciences Economiques
Prsente et soutenue publiquement par
Zakaria MOUSSA
le 6 dcembre 2010

Assouplissement quantitatif ; quels enseignements

tirer de lexprience japonaise ?

Directeur de Thse
M. Eric GIRARDIN, Professeur lUniversit de la Mditerrane, GREQAM
Jury
Rapporteurs
M. Patrick FVE Professeur luniversit de Toulouse I, GREMAQ
M. Andrew J. FILARDO Economiste en Chef, Banque des Rglements
Internationaux, zone AsiePacique
Examinateurs
M. Gilles DUFRNOT Professeur lUniversit dAix-Marseille 2, DEFI
M. Michel LUBRANO Directeur de recherche CNRS, GREQAM,
M. Benot MOJON Banque de France, Chef du service de recherche
sur la politique montaire
LUniversit de la Mditerrane nentend ni approuver, ni dsapprouver les opinions partic-
ulires du candidat: ces opinions doivent tre considres comme propres leur auteur.
En souvenir de mon pre.
A ma famille et Galle.
Rsum
La crise nancire actuelle, en raison de sa similarit avec celle du Japon des annes 1990,
a pouss les autorits montaires des plus grandes banques centrales adopter lassouplis-
sement quantitatif. Seul le Japon, ayant connu une exprience dassouplissement quantitatif
rcente mais depuis susamment dannes pour tre tudie, peut fournir des lments de
solution cette crise.
Cette thse applique les techniques conomtriques les plus appropries et rcentes
lanalyse de lassouplissement quantitatif, appliqu par la Banque du Japon entre 2001 et
2006. En trois chapitres sont traites les questions de savoir sil tait ecace ; sous quelles
conditions ? Par quels canaux ?
Lecacit de cette stratgie de politique montaire stimuler lactivit et stopper
la spirale dationniste a t montre. Cette exprience met en avant le rle important que
la politique montaire peut jouer pour sortir de la crise, mme quand le taux directeur atteint
zro. Le canal des anticipations comme le canal de rquilibrage des portefeuilles ont tous
deux jou un rle important dans la transmission de ces eets. Les principaux enseignements
que lon peut tirer de lexprience japonaise sont, dabord de remdier radicalement et
immdiatement aux fragilits du secteur nancier, deuximement, de mener une politique
montaire particulirement agressive. Enn, dattendre le temps ncessaire pour que les
fruits de cette politique viennent. Lexprience japonaise suggre que la Fed et la banque
dAngleterre doivent reporter leur sortie de cette stratgie, sortie qui doit tre mene dans
le cadre dun programme et selon des objectifs numriques clairs.
Mots cls : Assouplissement quantitatif ; Canaux de transmission ; FAVAR; Markov-
switching ; Time-varying-parameter FAVAR; Modle macro-nance ; Japon.
Abstract
The current nancial crisis has now led most major central banks to rely on quantitative
easing. The unique Japanese experience of quantitative easing is the only experience which
enables us to judge this therapys eectiveness and the timing of the exit strategy. Is quan-
titative easing eective ? Under which conditions ? Through which canal ?
This thesis, consisting of three essays, applies appropriate and recent econometric
techniques to examine the quantitative easing in Japan between 2001 and 2006. We show,
for the rst time, that quantitative easing was able not only to prevent further recession
and deation but also to provide considerable stimulation to both output and prices. Moreo-
ver, both expectation and portfolio-rebalancing channels play a crucial role in transmitting
monetary policy eects. This experience shows that the monetary policy is still potent even
when short-term interest rates reach a zero lower bound.
The Japanese experience suggests that eorts to clean up the banks balance sheets
signicantly improved the eectiveness of quantitative easing. However, this eect, although
considerable, was short-lived ; it became insignicant after one year. The short duration
of this eect conrms the wisdom of the Feds decision to maintain quantitative easing
longer, so that being short-lived, the positive eects could be exploited. In the light of the
Japanese experience, we argue that, in addition to their fast reaction and the huge amount of
CABs employed, which may have helped relieve short-term liquidity pressures in the nancial
system, the Fed was better o postponing its exit from quantitative easing.
Keywords : Quantitative Easing Policy ; Transmission channels ; FAVAR; Markov-
switching ; Time-varying-parameter FAVAR; Macro-nance model ; Japan.
Remerciements
Tant de personnes ont rendu possible laboutissement de ce travail de thse quil mest
aujourdhui dicile de nen oublier aucune. Ce manuscrit conclut quatre ans de travail ; je
tiens en ces trop courtes lignes exprimer ma reconnaissance envers tous ceux qui de prs
ou de loin y ont contribu, et demande par avance excuse ceux que jaurais oublis.
Jexprime en premier lieu ma gratitude Eric Girardin, mon directeur de thse, pour
mavoir propos ce sujet passionnant et mavoir maintenu sa conance tout au long de ces
annes. Je noublie pas son premier message, dcisif, o il me manifestait son intrt et
prsentait sa motivation pour un travail commun sur cette thse. Merci aussi pour les pr-
cieux conseils qui ont suivi, sa constante disponibilit et sa gentillesse. Jai particulirement
apprci les discussions scientiques que nous avons eues et qui mont profondment aid
avancer sur le sujet. Merci galement de mavoir guid, tout en me laissant lautonomie
de choisir mon chemin et mes mthodes.
Pour avoir accept de rapporter ce travail, jassure toute ma reconnaissance Patrick
Fve et Andrew Filardo ; leurs rapports ont grandement contribu amliorer mes travaux,
notamment du point de vue de linterprtation des rsultats des modles exposs. Que soient
remercis galement les autres jurs pour avoir lu mon manuscrit et y avoir port un regard
critique ; messieurs Michel Lubrano, Benot Mojon et plus particulirement Gilles Dufrnot
pour avoir assur le rle de prsident de jury.
Nombreux sont ceux avoir, au l de ma thse, apport leur contribution scientique.
Je tiens ainsi remercier Steve Basen, Anne Pguin, Costin Protopopescu et nouveau
Michel Lubrano, pour leur aide en conomtrie et leurs conseils aviss.
Ce travail a pu voir le jour grce un nancement personnel, puis lobtention dun
demi-poste dATER lUniversit Marseille 2 ; je tiens donc exprimer ma gratitude aux
personnes qui mont aid atteindre ces conditions de travail idales, notamment Domi-
nique Ami et Pierre Granier. Je garde de bons souvenirs de cette exprience durant laquelle
jai collabor principalement avec Dominique, que je remercie normment pour sa bonne
humeur, et le plaisir trouv travailler avec elle. Je tiens remercier galement les membres
de la Facult des Sciences de Luminy pour leur accueil, leur soutien, et surtout pour mavoir
renouvel leur conance une deuxime anne ; cela ma aid nir ma thse dans de bonnes
conditions.
Ce travail de thse a t un long parcours, au sens propre comme au gur ; il ma
mme men lautre bout du monde, au pays du soleil levant. Tout au long de mon sjour au
Japon, jai eu la chance de croiser des personnes de grande qualit, scientique et humaine,
qui mont encourag continuer mon chemin de recherche et mont rendu conance en moi
aprs une premire priode dicile. Jadresse mes vifs remerciements aux professeurs de
luniversit de Musashi Tokyo pour leur accueil, gnrosit et bonne humeur. Dans lordre
alphabtique (franais), je remercie Kimihiro Furuse, Yoshio Kurosaka, Yuko Nikaido, Junko
Nishimura, Sanae Ohno et Eiko Sakai. Je tiens aussi remercier Yuki Teranishi pour lintrt
quil a montr lgard de mon travail et son invitation au sein de la Banque du Japon,
ainsi que le reste de lquipe pour ses remarques et suggestions durant ma prsentation.
Je veux spcialement tmoigner de ma reconnaissance Yusho Kaglaoka qui, au del de
son implication au chapitre 3 ralis conjointement, na cess de montrer sa disponibilit,
sa gentillesse, son souci de mon bien tre et de ma bonne intgration au sein de lquipe ;
que Yusho soit assur de ma reconnaissance pour son indfectible soutien, dans lespoir que
nous retravaillons ensemble bientt.
Bien sr, ce sjour inoubliable au Japon a t facilit par laide nancire procure
par lcole doctorale (n372) et par une bourse accorde dans le cadre du Groupement de
Recherche International en " Connaissance, interactions, dcisions ". Je remercie donc Jean
Benot Zimmerman, directeur du GREQAM et surtout Alain Vendetti pour mavoir procur
les informations utiles en temps et en heure et pour sa bonne humeur sportive et com-
municative. Je remercie galement Nobuyuki Hanaki pour nos changes franco-japonais de
rudiments linguistiques qui mont t dune grande aide quotidienne pour entrer en contact
avec ses concitoyens.
Jadresse galement ma profonde reconnaissance tous les membres de lquipe
administrative et informatique du GREQAM qui mont apport leur indispensable soutien
logistique : Bernadette, Corinne, Grald, Carole, Isabelle, Jean-Paul, Lydie, Pascal. Merci
davoir toujours reu mes demandes avec le sourire et dy avoir rpondu avec autant de-
cacit.
Une mention spciale est donne Marjorie Sweetko pour laide irremplaable quelle
a apporte ma rdaction en anglais ; la tche ntait pas aise et elle la accomplie avec
une comptence et un dvouement remarquables, que je noublierai pas.
La bonne ambiance qui rgne au GREQAM a accompagn la progression de ce tra-
vail ; mes collgues, et leur bonne humeur quotidienne ont grandement contribu faire des
journes au laboratoire un plaisir. Je remercie Benot S. pour son amiti et nos discussions,
Philippe pour sa gentillesse et pour les bons moments passs ensemble lors de notre collabo-
ration dans et en dehors du travail, Renaud pour son sens de lhumour et ses conseils aviss,
Sarra pour nos riches changes et pour sa mticuleuse relecture de lintroduction, Luis pour
11
son sens de lhumour et son aide prcieuse pour la prsentation, Maame et Shamaila pour
leur relecture. Je tiens galement remercier Adreana, Agns, Andreea, Aziz, Benot T.,
Carmela, Chen, Clment, Elvera, Elsa, Gabriele, Gwenola, Jamel, Kalila, Kamila, , Katia,
Leila, Maame, Mandy, Maria, , Mathieu, Maty, Meriem, Morgane Narin, , Ophlie, Paul,
Paul-Antoine, Rabeh, Sonia, Walid et Waqar pour leur soutien.
Mes remerciements vont aussi mes voisins sociologues et anthropologues du centre
Norbert Elias avec qui jai vcu de trs agrables et enrichissants moments pendant les repas
ou en dehors du cadre de travail. Je tiens donc remercier particulirement Jean-Christophe,
Tanguy et Jean-Baptiste pour leur bonne humeur et leurs discussions qui mloignaient
momentanment de lconomie. Je remercie galement Cyril, Karim, Julie et Vincent pour
leurs encouragements en n de parcours.
Jai la chance davoir t solidement accompagn chaque tape de ce priple ; ma
famille, bien quloigne, a toujours t prsente et cest son appui qui ma aid avancer.
Je voudrais remercier spcialement mon pouse, Galle, qui a jou un rle dterminant au
cours de ces annes de thse, et ce depuis le soir o le hasard nous a men dans un restaurant
japonais pour y dcider ensemble de dbuter laventure de cette thse. Elle a accompagn
mes enthousiasmes et mes angoisses, a support mes absences rcurrentes et surtout ma
aid surmonter les moments diciles grce son soutien quotidien indfectible, fourni
avec tout son amour.
Je ddie cette thse ma famille, avec tout mon cur, en souvenir de mon pre qui
a tant souhait voir ses enfants aboutir dans leurs tudes, et qui a t pour moi un modle
de travail, dhonntet et de persvrance. Mes remerciements vont en particulier ma mre
pour son soutien discret et essentiel, mes grandes surs, belles toiles qui veillent sur moi,
ainsi qu tous mes frres pour avoir montr leur optimisme face au partage des dicults.
Jadresse galement ma profonde reconnaissance aux membres de ma belle-famille pour leurs
encouragements constants, leur accueil chaleureux et leur gnrosit.
Table des matires
Table des matires
Table des gures iii
Liste des tableaux v
Nomenclature vi
Introduction gnrale 1
Chapter 1: Quantitative easing works: Lessons from the unique experience in
Japan 23
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.2 Related literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.3 Transmission Channels of QEMP . . . . . . . . . . . . . . . . . . . . . . . 29
1.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.4.1 MS-FAVAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.4.2 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.5 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.5.1 Estimated Structural Factors . . . . . . . . . . . . . . . . . . . . . 42
1.5.2 Traditional MS-VAR . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.5.3 MS-FAVAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.5.4 Is a scal stimulus eective? . . . . . . . . . . . . . . . . . . . . . 52
1.6 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
1.7 Implications and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 56
1.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Chapter 2: Quantitative Easing under Scrutiny: A TVP-FAVAR Model 87
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.2.1 TVP-FAVAR model . . . . . . . . . . . . . . . . . . . . . . . . . . 93
i
Table des matires
2.2.2 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.3 Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
2.3.1 Data and preliminary results . . . . . . . . . . . . . . . . . . . . . 102
2.3.2 Specication tests . . . . . . . . . . . . . . . . . . . . . . . . . . 103
2.3.3 The evolution of the Japanese monetary policy . . . . . . . . . . . 104
2.3.4 Impulse response analysis . . . . . . . . . . . . . . . . . . . . . . . 106
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Chapitre 3 : Quantitative Easing and the Time-Varying Dynamics of the Term
Structure of Interest rate in Japan 123
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.2 Estimating spot rate curves for Japan . . . . . . . . . . . . . . . . . . . . 127
3.2.1 Data construction . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.2.2 Estimation procedure . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.2.3 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.3 Yield-Curve Fitting : The Macro-Finance Model . . . . . . . . . . . . . . . 133
3.3.1 Methodology and Estimation . . . . . . . . . . . . . . . . . . . . 133
3.3.2 Priors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
3.4 Empirical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.4.1 Preliminary Empirical Results . . . . . . . . . . . . . . . . . . . . . 138
3.4.2 Evidence on the expectations hypothesis (EH) . . . . . . . . . . . 140
3.4.3 Time-varying term premium . . . . . . . . . . . . . . . . . . . . . 142
3.4.4 Empirical Results From the Macro-Finance Model . . . . . . . . . 144
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Conclusion gnrale 159
Bibliographie 165
ii
Table des gures
Table des gures
1 Lconomie japonaise avant et aprs le dgonement de la bulle spculative 3
2 Stimulus scal et dette publique au Japon 1990-2008 . . . . . . . . . . . 4
3 Cibles sur le niveau des comptes courants des banques prives . . . . . . . 5
4 Crances douteuses des banques japonaises et pertes dues ces crances . 6
5 Raction de la politique montaire aprs lclatement de la bulle nancire
au Japon et dans le reste des pays du G7 . . . . . . . . . . . . . . . . . . 8
1.1 Regime probabilities for MSIAH-VAR . . . . . . . . . . . . . . . . . . . . 46
1.2 Response to a monetary base shock in MS-VAR regime 1 re-
gime 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.3 Regime probabilities for MS-FAVAR . . . . . . . . . . . . . . . . . . . . . 50
1.4 Response to a monetary base shock in MS-FAVAR . . . . . . . . . . . . . 51
1.5 Estimated factor loadings . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
1.6 The original and corrected M0 . . . . . . . . . . . . . . . . . . . . . . . . 62
1.7 Activity factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
1.8 Price factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
1.9 Interest rate factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
1.10 The JGB issuance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
1.11 Regimes probabilities - MS-FAVAR model . . . . . . . . . . . . . . . . . . 80
1.12 Response to a monetary base shock in MS-FAVAR . . . . . . . . . . . . . 81
1.13 Response to a scal policy shock in MS-FAVAR . . . . . . . . . . . . . . . 82
1.14 Response to a scal policy shock in MS-FAVAR . . . . . . . . . . . . . . . 83
2.1 Posterior mean of the standard deviation of equation residuals . . . . . . . 105
2.2 Impulse response functions . . . . . . . . . . . . . . . . . . . . . . . . . . 107
2.3 Impulse responses - Policy-duration eect . . . . . . . . . . . . . . . . . . 110
2.4 Impulse responses - Portfolio-rebalancing channel . . . . . . . . . . . . . . 112
2.5 Impulse responses - Disaggregated price . . . . . . . . . . . . . . . . . . . 120
2.6 Impulse responses - Disaggregated production . . . . . . . . . . . . . . . . 121
3.1 Japanese Government Bond spot curves 1985-2009 . . . . . . . . . . . . . 131
3.2 Estimated factors and their empirical counterparts . . . . . . . . . . . . . 139
3.3 Estimated Standard deviation of the FAVAR residuals . . . . . . . . . . . . 140
3.4 Extracted expectation component . . . . . . . . . . . . . . . . . . . . . . 141
3.5 Estimated term premium . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
3.6 Unconditional variance - Call rate shock. . . . . . . . . . . . . . . . . . . . 145
3.7 Impulse responses - Call rate shock . . . . . . . . . . . . . . . . . . . . . . 147
iii
Table des gures
3.8 Unconditional variance - level factor shock . . . . . . . . . . . . . . . . . . 149
3.9 Impulse responses - Level shock . . . . . . . . . . . . . . . . . . . . . . . 151
3.10 Variance decomposition due to ination . . . . . . . . . . . . . . . . . . . 154
3.11 Variance decomposition due to the output gap . . . . . . . . . . . . . . . 155
3.12 Variance decomposition due to slope factor . . . . . . . . . . . . . . . . . 156
3.13 Variance decomposition due to curvature . . . . . . . . . . . . . . . . . . 157
3.14 Impulse response functions to slope shock . . . . . . . . . . . . . . . . . . 158
iv
Liste des tableaux
Liste des tableaux
1.1 Eigenvalues and percent of variance of rst four factors . . . . . . . . . . . 44
1.2 Feasible triples for a highly variable Grid . . . . . . . . . . . . . . . . . . . 62
1.3 Unit root tests (Sample period 1985:3 to 2006:3) . . . . . . . . . . . . . . 72
1.4 Unit root tests (Sample period 1985:3 to 2006:3) . . . . . . . . . . . . . . 72
1.5 Linearity test:VAR model . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
1.6 MS specications among various MS-VAR models . . . . . . . . . . . . . . 74
1.7 Lag length test:MSIAH-VAR model . . . . . . . . . . . . . . . . . . . . . 75
1.8 Transition matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
1.9 Linearity test: MS-FAVAR . . . . . . . . . . . . . . . . . . . . . . . . . . 76
1.10 MS specications among various MS-FAVAR model . . . . . . . . . . . . . 77
1.11 Lag length test:MSIAH-FAVAR model . . . . . . . . . . . . . . . . . . . . 78
1.12 Transition matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.1 Model comparison with Deviance Information Criterion (DIC) . . . . . . . 104
2.2 Feasible triples for a highly variable Grid . . . . . . . . . . . . . . . . . . . 116
3.1 Descriptive statistics : Japanese spot rate curves . . . . . . . . . . . . . . 132
v
nomenclature
vi
Nomenclature
BOJ Bank of Japan
CAB Current Account Balances of Financial Institutions held with the Bank of Japan
CPI Consumer Price Index
EH Expectation Hypothesis
EM Expectation-Maximisation
FAVAR Factor-Augmented Vector Autoregression
JGB Japanese Government Bonds
JSDA Japan Securities Dealers Association
M0 Monetary Base
MCMC Markov chain Monte Carlo
MS-VAR Markov-Switching Vector Autoregression
QEMP Quantitative easing Monetary Policy
TSE Tokyo Stock Exchange
TVP-FAVAR Time-Varying Parameter Factor-Augmented Vector Autoregression
VECM Vector Error Correction Model
ZIRP Zero Interest Rate Policy
nomenclature
viii
Introduction gnrale
Dans le contexte actuel de la crise nancire qui se prolonge depuis octobre 2008, les princi-
pales banques centrales ont opt pour la poursuite de politiques montaires expansionnistes
non conventionnelles. La banque du Japon a dcid rcemment de mener une politique mon-
taire dite Comprehensive Monetary Easing qui dire quelque peu par rapport la politique
dassouplissement quantitatif mene entre 2001 et 2006. La Fed, son tour, conrme le
maintien de sa politique dassouplissement des conditions de crdit dbute en 2009. Ces
politiques sont dsormais orientes essentiellement vers la modication de la composition
des actifs des banques centrales par lachat massif de titres long terme dans le but de
baisser leurs rendements. Ceci aurait pour eet de rduire les rendements dautres actifs
nanciers, en apportant davantage de liquidit au systme nancier.
Jusque rcemment la situation particulire de lconomie japonaise daprs 1990 tait
considre comme un cas isol dans lconomie mondiale ; elle sourait dune longue stag-
nation et dune forte tendance dationniste, aggrave par la disparition des instruments de
politique montaire dont dispose traditionnellement la banque centrale. La banque centrale
du Japon (BOJ) a donc d mener une stratgie de politique montaire non conventionnelle,
dite dassouplissement quantitatif. La crise nancire actuelle, en raison de sa similarit avec
celle du Japon des annes 1990, a pouss les autorits montaires des plus grandes banques
centrales adopter ce mme type de stratgie ; celles-ci cherchent donc aujourdhui tirer
partie des leons de lexprience japonaise. Lassouplissement quantitatif tait-il ecace ?
Par quels canaux ? Dans quel dlai ?
En eet, durant la rcession qui suivit le dgonement de la bulle spculative au
1
Introduction gnrale
dbut des annes 1990, des politiques scales et montaires expansionnistes furent menes
dans le but de stimuler lconomie japonaise. Cependant, jusqu dbut 2001, aucun signe
fort de reprise conomique ne se t sentir, du moins au niveau macroconomique. Comme
montr par le graphique 1, lconomie japonaise entra en phase de dpression partir de
1991, entrecoupe de quelques priodes de reprises, puis entra en dation en 1998. Nayant
pas men une politique montaire laxiste immdiatement aprs le dgonement de la bulle
spculative, la BOJ fut critique pour son manque de ractivit. En eet, elle ne rduisit
que progressivement son taux directeur, le rduisant de 6% en 1990 0,5% en 1995, et ne
la amen un niveau proche de zro qu partir de fvrier 1999.
Sans montrer deet satisfaisant, ces politiques ont rduit les marges de manoeuvre
des autorits, qui furent alors contraintes demployer des mesures expansionnistes indites,
comme notamment laccroissement de la dette publique en pourcentage de PIB, passe de
50% en 1991 120% environ en 2001, niveau le plus important parmi les pays industrialiss
(cf. graphique 2), ou encore comme la baisse des taux dintrt nominaux de court terme
jusqu leur niveau plancher zro.
Pour les autorits nippones, la question est de savoir comment faciliter la reprise
conomique, tant donns le poids lev de la dette publique et la contrainte de non-
ngativit des taux nominaux de court terme.
Les politiques budgtaires menes au Japon ont t considres comme inecaces
1
et prsentant le risque daggraver lendettement public, dautant plus quil est dicile dva-
luer le multiplicateur budgtaire pendant les priodes de rcession, comme expliqu par Koo
(2008). Les outils ont alors t cherchs du ct de la politique montaire qui pouvait jouer
un rle crucial pour la reprise. De nombreux conomistes ont donc recommand la BOJ de
renverser durablement les anticipations de dation des agents privs en prenant un engage-
1
Posen (1998) montre que linecacit de la politique scale provient de sa mauvaise application
et de linsusance des montants consacrs par rapport aux objectifs initiaux. Selon lui, lexprience
de 1995 est lexemple dune politique scale expansionniste russie.
2
Introduction gnrale
Figure 1 Lconomie japonaise avant et aprs le dgonement de la bulle spculative
-0.10
-0.05
0.00
0.05
0.10
Eclatement de la bulle financire
QEMP ZIRP
Taux d'inflation (IPC)
Taux de croissance (PI)
Taux directeur
1000
2000
3000
1
9
8
0
1
9
8
3
1
9
8
6
1
9
8
9
1
9
9
2
1
9
9
5
1
9
9
8
2
0
0
1
2
0
0
4
2
0
0
7
TOPIX
QEMP : politique montaire dassouplissement quantitatif ; ZIRP : politique montaire de taux dintrt zro ;
TOPIX 100 : Tokyo Stock Price Index, indice de rfrence sur TSE (Tokyo Stock Exchange), valeurs de n de
mois.
Source : ECOWIN, Banque du Japon.
ment crdible de laxisme, et en accroissant la base montaire courante et future (Krugman
(2000) ; Bernanke et al. (2004) ; McCallum, 2000 ; Svensson, 2000 et 2003). A partir de
mars 2001 la BOJ a ainsi dcid de mener une politique dassouplissement quantitatif qui
consiste en lutilisation simultane de trois stratgies non conventionnelles de politique mo-
ntaire : (i) un accroissement de la base montaire en xant un objectif quantitatif pour
les comptes courants dtenus par les banques auprs de la banque centrale (cf. graphique
3) ; (ii) un engagement public poursuivre une politique montaire laxiste jusqu ce que
lination, mesure par lindice des prix la consommation hors produits alimentaires frais,
ache durablement un taux nul ou positif ; (iii) un soutien de lobjectif quantitatif concer-
nant lencours des comptes courants des banques prives par lachat dobligations dEtat
(Japan Government Bond-JGB).
3
Introduction gnrale
Figure 2 Stimulus scal et dette publique au Japon 1990-2008
-3
-2
-1
0
1
2
3
Consommation Publique
Investissement Public
Total
1
9
9
0
1
9
9
1
1
9
9
2
1
9
9
3
1
9
9
4
1
9
9
5
1
9
9
6
1
9
9
7
1
9
9
8
1
9
9
9
2
0
0
0
2
0
0
1
2
0
0
2
2
0
0
3
2
0
0
4
2
0
0
5
2
0
0
6
2
0
0
7
2
0
0
8
50
100
150
dette publique en % de PIB
Source : Cabinet Oce Japan et OCDE
Il est noter que la politique dassouplissement quantitatif tait prcde dun chan-
gement drastique du systme nancier an de faire face la crise nancire dclenche
suite au dgonement de la bulle nancire. En eet, les systmes nancier et bancaire
ont commenc connatre de srieuses dicults suite laugmentation importante du
ratio de crances douteuses. Le graphique 4 montre que les crances douteuses dtenues
par lensemble des banques ont atteint 5, 5% du PIB en 1996 et que les pertes qui en ont
dcoul reprsentaient plus de 2, 5% du PIB dans la mme anne. Depuis lors, lconomie
Japonaise est entre dans un cercle vicieux de dation, stagnation et augmentation des
prts non performants. De nombreuses banques ont eu des dicults rduire lampleur du
problme et ont fait faillite ; les deux plus importantes taient Hokkaido Takushoku Bank et
Yamaichi Securities Company en 1997. En plus de ces problmes internes, la crise asiatique
de 1997 a provoqu une baisse de lactivit japonaise (cf. graphique 1), exposant ainsi les
4
Introduction gnrale
Figure 3 Cibles sur le niveau des comptes courants des banques prives

Excess reserves
Current Account Balances
Trillon yen

Excess reserves
Current Account Balances
Required reserves
Target range
Target
Trillon yen

Excess reserves
Current Account Balances
Required reserves
Target range
Target
Trillon yen
Les autorits montaires avaient comme cible environ 5 billions de yen la mise en place
de lassouplissement quantitatif en mars 2001, puis 6 billions de yen du mois daout jusqu
dcembre 2001. Elles lont environ double pour tre dans la tranche de 10-15 billions de yen en
dcembre 2001, puis lont leve au niveau de la tranche de 15-20 billions de yen en mars 2003
(+40%) avant datteindre la tranche de 30-35 billions de yen en 2004 (+11%). Les rserves
obligatoires durant cette priode taient de lordre de 5 billions de yen.
Source : Banque du Japon
institutions nancires japonaises une augmentation de leurs crances douteuses et aux
pertes qui en dcoulent (cf. graphique 4). Simultanment, lconomie a connu un phno-
mne dnomm par Koo (2008) "rcession du bilan" par lequel les entreprises, comme les
mnages, ont vu leurs bilans se dgrader suite la chute des cours des actifs nanciers. Les
secours apports par les autorits nippones au secteur nancier, mis en place en plusieurs
tapes, faisaient partie dune politique de drglementation engage en novembre 1996 an
damliorer la transparence du systme nancier et de faciliter sa restructuration (big-bang
nancier). Cargill et al. (2001) montrent que son application tait particulirement ecace
5
Introduction gnrale
Figure 4 Crances douteuses des banques japonaises et pertes dues ces crances
3
4
5
6
7
8
0.5
1.0
1.5
2.0
2.5
1995 1997 1999 2001 2003 2005
QEMP
ZIRP Crise asiatique
En % du PIB
Crances douteuses par l'ensemble des banques de dpts
Pertes bancaires dues aux crances douteuses (chelle de droite)
Source : Financial Services Agency (FSA)
dans la rsolution des dicults des banques en permettant dvacuer de leurs bilans les
prts non performants. Dans le mme temps, la rforme institutionnelle de la BOJ, mise en
application par la nouvelle loi de 1998, a renforc son indpendance par rapport au ministre
des nances ; permettant ainsi dasseoir la crdibilit de la banque centrale et de favoriser
lancrage des anticipations des agents privs.
Un signe avant-coureur de la reprise apparaissait en novembre 2005, alors que le taux
dination commenait tre positif (cf. graphique 1). La BOJ dclara en mars 2006 que
lination demeurerait positive et soutenue et que les eets de la politique dassouplissement
quantitatif commenaient se faire sentir. Dsormais, maintenir trop longtemps cette poli-
tique aurait pu mener une ination leve et soutenue, tant donne la forte expansion de
la base montaire depuis 2001. Par consquent, considrant quil tait temps de mettre n
la stratgie dassouplissement quantitatif, la BOJ a dcid de restaurer le taux dintrt
au jour le jour comme instrument de la politique montaire.
6
Introduction gnrale
La crise nancire actuelle prsente aux moins deux points de forte similitude avec la
crise japonaise. Elles trouvent toutes deux leur origine dans lclatement de bulles spcula-
tives qui ont chacune conduit une baisse des prix, avec spirale dationniste dans le cas du
Japon, spirale qui a t jusque l vite par les autres pays du G7. De plus, dans les deux cas
les taux ont baiss des niveaux proches de zro, suite aux interventions des autorits mo-
ntaires pour grer ces crises. Nanmoins, quelques dirences sont noter au niveau de la
ractivit des banques centrales dans la gestion de la crise et au niveau des politiques mon-
taires non conventionnelles mises en places. Tout dabord, et comme premire leon tire de
lexprience japonaise, les principales banques centrales ont t plus ractives dans la baisse
des taux dintrt et dans la mise en place de politiques montaires non-conventionnelles
ds que les taux dintrt nominaux de court terme atteignirent zro. Cette ractivit a fait
dfaut dans le cas du Japon. Le graphique 5 montre que la BOJ a mis plus de 6 ans pour
baisser les taux dintrt un trs faible niveau et environ 4 ans pour mener des stratgies
de politique montaire alternatives non-conventionnelles quand les taux dintrt nominaux
atteignirent zro. Quant aux autres banques centrales, spcialement la Fed, elles ont r-
agit rapidement, en moins de deux ans elles ont baiss leurs taux directeurs des valeurs
proches de zro et ont appliqu des politiques non conventionnelles. Deuximement, la
dirence de la BOJ, la Banque dAngleterre, la Banque Centrale Europenne et la banque
du Canada ont adopt une politique dassouplissement quantitatif visant atteindre une
cible quantitative de taille du bilan ainsi qu changer la composition du bilan, tout en ne
prenant pas dengagement explicite maintenir les taux directeurs un bas niveau. Quant
la politique de la Fed, qualie dassouplissement de crdit, elle met principalement laccent
sur le changement de la composition du bilan de la banque centrale avec un engagement
explicite maintenir les taux un faible niveau ; la taille du bilan ntant quun objectif
accessoire. La Fed vise alors soutenir dune faon directe les marchs du crdit. Elle a
notamment facilit laccs aux crdits des secteurs choisis, en quantit suprieure ce qui
7
Introduction gnrale
Figure 5 Raction de la politique montaire aprs lclatement de la bulle nancire
au Japon et dans le reste des pays du G7
2
4
6
8
Eclatement
de la bulle
immobilire
- 2 ans 2 ans 4 ans
6 ans 8 ans
Taux directeur
BCE
Fed
Banque de Canada
Banque d'Angleterre
Mise en place de
l'assouplissement
quantitatif par la
banque du Japon
Mise en place de
polititques montaires
non conventionelles
par les plus grandes
banques centrales
10 ans
(%)
BOJ
Source : BOJ, Fed, BCE, Banque du Canada et Banque dAngleterre
serait fourni par des marchs nanciers en dicult. Reste alors savoir si ces direntes
stratgies non conventionnelles de politique montaire sont ecaces, par quels canaux de
transmission et estimer le temps ncessaire. Seul le Japon, qui a connu une exprience
dassouplissement quantitatif rcente, mais depuis susamment dannes pour quelle soit
tudie, peut nous fournir des lments de rponse ces questionnements.
Lassouplissement quantitatif : cadre thorique et preuves empiriques
De nombreux et rcents travaux de recherche, thoriques et empiriques, ont analys lco-
nomie japonaise dans les annes 1990 en essayant dapporter des lments de solution pour
la sortie de la crise.
Lanalyse contemporaine du rle de la politique montaire considre habituellement
que linstrument principal dintervention des autorits montaires est le taux dintrt court
terme. Suite larticle clbre de Taylor (1993), une littrature tendue a cherch identier
8
Introduction gnrale
la politique montaire et ses eets dans le cadre des rgles de taux dintrt, qui peuvent
tre drives explicitement de la fonction objectif retenue par les autorits montaires.
Nanmoins, pour mener des politiques expansionnistes en cas de crise, et dans le cas o
les taux dintrt approchent zro, la rgle de Taylor suggre des taux dintrt ngatifs ; la
politique montaire est donc contrainte. Tenir compte de la contrainte du plancher zro
des taux dintrt reprsente donc un nouveau d pour lapproche de Taylor. La validit
de cette rgle a t ravive dans de nouvelles recherches ; dveloppons lapport des travaux
qui ont mis laccent sur le rle important des anticipations des taux dintrt courts futurs
et de lination pour sortir de la spirale dationniste et stimuler lconomie.
Partant de lide que lconomie japonaise est entre en situation de trappe liquidit,
le paradigme no-Wicksellien, dominant lanalyse de la politique montaire, suggre que la
politique montaire peut toujours inuencer lconomie via lorientation des anticipations
concernant, la fois, la trajectoire des taux courts futurs et lination. Krugman (2000) tait
le premier recommander la banque centrale du Japon dadopter une nouvelle stratgie
de politique montaire visant inuencer les anticipations des agents privs tout en prenant
garde au problme de crdibilit. Une augmentation de lore de monnaie na pas deet
si elle nest pas accompagne dun engagement strict ce que le surplus de liquidit soit
maintenu jusqu ce que les conditions de lengagement soient remplies. Des ranements
et prcisions sont apports par Eggertsson et Woodford (2003) qui arment que le seul
moyen pour sortir de la situation de la trappe liquidit est le contrle des anticipations
des agents privs, en excluant tout eet dune augmentation de la masse montaire ou dun
changement de la composition du bilan de la banque centrale. Ceci est d lhypothse de
parfaite substituabilit entre la monnaie et les actifs non montaires quand les taux dintrt
approchent zro. En eet, quand le taux nominal de court terme devient nul, si les encaisses
relles excdent un certain seuil (ou niveau de satiation), lutilit marginale obtenue des
services de liquidit due des encaisses relles additionnelles devient nulle. Dans ce cas
9
Introduction gnrale
la possibilit dun rquilibrage de portefeuille de lagent priv suite une augmentation
de la base montaire est exclue. Un engagement crdible de la banque centrale pourrait
alors augmenter la demande globale et les prix en stimulant les dpenses courantes via trois
canaux : (i) par le maintien dun taux dintrt un niveau plus bas pour une dure plus
longue que prvue, (ii) en baissant le taux dintrt rel par laugmentation de lination
anticipe, et nalement (iii) par anticipation dune augmentation des revenus futurs.
Svensson (2001) partage le scepticisme neo-Wicksellien lgard de lecacit de
lapproche quantitative et tend ce modle lconomie ouverte. An de sortir de la spirale
dationniste, Svensson (2003) propose, partir de ce quil appelle Foolproof Way, dtablir
pendant un certain temps un sentier cible pour le niveau des prix qui soit arrim un taux
dination positif et de renforcer cette mesure par lannonce dune dvaluation de la monnaie.
Toutefois, Ito et Mishkin (2004) et Ito et Yabu (2007) suggrent que ce type de politique
naura deet que si la BOJ intervient sur le march des changes sans annonce pralable
dune cible de taux de change ; viter la confusion entre les ancres nominales de la politique
montaire renforce la crdibilit de la banque centrale.
A linverse de lapproche neo-Wicksellienne, qui arme que la politique dassouplis-
sement quantitatif ne peut avoir deet que dune manire indirecte au travers des anti-
cipations, lapproche montariste carte la possibilit de trappe liquidit et suggre que
linjection de la liquidit, via laccroissement de la base montaire, peut inuencer lconomie
mme si les taux dintrt approchent zro. Sous cette approche, lination est un phno-
mne essentiellement montaire ; les chocs montaires se transmettent lconomie relle
en provoquant un ajustement du prix relatif des actifs rels et nanciers (de court, moyen
et long terme) et, ainsi, un ajustement des portefeuilles des agents. Malgr la contrainte
due au taux dintrt zro, laccroissement de la base montaire permet donc daugmenter
la consommation via leet de richesse qui incite lagent priv faire des dpenses suppl-
mentaires, stimulant ainsi lactivit (Metzler (1995)). Cela nest bien sr possible que sous
10
Introduction gnrale
condition dimparfaite substituabilit entre les dirents actifs et la monnaie
2
.
Dans la mme ligne, une vue plus rcente de lapproche montariste se focalise
sur la prime de liquidit comme canal de transmission de la base montaire lactivit
(Yates (2004), Goodfriend (2000) et Andrs, Lpez-Salido et Nelson (2004)). Etant donne
limparfaite substituabilit et la dirence qualitative en termes de liquidit entre la monnaie,
les obligations et les actions, une augmentation de la base montaire pousse les agents privs
rduire le niveau exig de la prime de liquidit des actifs non liquides, diminuant ainsi leurs
rendements. Ce mcanisme de transmission a donc la vocation dentraner une relance de
lactivit conomique non pas travers une baisse des anticipations de taux courts futurs,
comme le suggre lapproche neo-Wicksellienne, mais par une baisse des taux dintrt de
long terme.
Koo (2008) dveloppe une analyse dirente, centre sur le secteur priv : pour lui,
la crise japonaise rsultait de ce quil appelle "rcession par le bilan" suite lclatement de
la bulle nancire qui a laiss un grand nombre de socits prives avec un bilan dsquilibr.
Le secteur priv, ayant des dettes dpassant le montant des actifs, est alors un acteur
qui ne cherche plus maximiser son prot, mais minimiser sa dette. Il refuse donc de
soctroyer de nouveaux crdits ou dmettre de nouvelles obligations malgr les faibles taux
dintrt. Selon lauteur, la situation de trappe liquidit qua connue le Japon doit alors tre
vue comme provenant du changement de comportement des emprunteurs et non pas des
prteurs, auquel cas toute politique montaire expansioniste visant augmenter la capacit
des banques octroyer des crdits est voue lechec en raison de labsence demprunteurs.
Nanmoins, il nexclut pas le rle important de la politique dassouplissement quantitatif
faciliter le dsendettement et le fonctionnement des institutions nancires.
Lassouplissement quantitatif, dans son application par la BOJ, nexclut aucun des
2
Metzler (1995) fait lhypothse que, parmi les actifs, seules les obligations sont parfaitement
substituables la monnaie. Les changements des taux dintrt de court terme, tant transitoires,
naectent donc pas les dcisions de consommation.
11
Introduction gnrale
canaux de transmission possibles voqus par les deux approches. Trois types de canaux
de transmission des eets des mesures oprationnelles de la politique dassouplissement
quantitatif ont t mis en exergue :
1. lengagement maintenir des taux dintrt courts futurs un niveau bas peut rduire
les taux dintrt de long terme et les rendements dautres actifs nanciers ;
2. leet de laugmentation de la taille du bilan de la BOJ par la fourniture de rserves
excdentaires aux banques commerciales peut avoir lieu par lintermdiaire de deux
canaux de transmission : (i) le canal de rquilibrage des portefeuilles selon lequel les
agents privs estiment quils disposent dun excdent de liquidits quils transfrent
vers les autres actifs nanciers et rels et vers la consommation ; (ii) le canal deet
du signal qui aecte les anticipations des cours futurs des taux dintrt ;
3. la modication de la composition du bilan de la BOJ par lachat dobligations dEtat
en change de rserves emploie les mmes canaux de transmission que la mesure de
politique montaire prcdente, savoir le rquilibrage des portefeuilles et leet du
signal.
Les travaux empiriques examinant les canaux de transmission ventuels et thoriques men-
tionns prcdemment sont videmment nombreux. Si ces tudes ont mis en vidence la
prsence dun changement de rgime dans les mcanismes de transmission de la politique
montaire japonaise (Fujiwara (2006), Inou et Okimoto (2008), Nakajima et al. (2009a)
et dautres), elles restent partages quant son ecacit (Ugai, 2006 ) ; les rsultats d-
pendent des modles et techniques conomtriques utiliss et galement des canaux de
transmission considrs.
Bernanke et al. (2004) sintressent leet de lassouplissement quantitatif sur les
anticipations des taux dintrt futurs. Les auteurs utilisent un modle macro-nance bas
sur la structure par terme des taux dintrt. Ils montrent que le canal des anticipations,
12
Introduction gnrale
gnr par lengagement de la BOJ, semble bien avoir eu leet escompt sur les taux
dintrt de long terme. Baba et al. (2005) et Oda et Ueda (2007), parviennent spcier
avec prcision le canal des anticipations, et conrment la capacit dun tel canal inuencer
la structure par terme des taux dintrt. Cela dit, lexamen de son eet sur lactivit
et lination est absent de leurs travaux. Oda et Ueda (2007) montrent galement que
leet du canal de rquilibrage de portefeuille, direct par le changement de la composition
du bilan de la BOJ, ou indirect suite laugmentation de la base montaire, na aucun
eet sur la prime de terme. Les travaux empiriques examinant leet de lassouplissement
quantitatif sur les variables macroconomiques utilisent souvent la mthodologie des modles
vectoriels autorgressifs (VAR). Kimura, Kobayashi, Muranaga et Ugai (2003) ont montr
que lecacit des canaux de transmission est fortement incertaine et trs faible. Leur
analyse empirique, se basant sur la mthodologie VAR avec des paramtres volutifs (time-
varying parameters), permet de tenir compte de changements possibles de llasticit de
la demande de monnaie et des mcanismes de transmission quand les taux dintrt se
rapprochent de zro. Aucun eet sur la production ni sur lination na t dtect pendant
la priode de lassouplissement quantitatif.
Fujiwara (2006) estime un modle VAR changements de rgimes markovien (MS-
VAR) sur la priode 1985-2004 en utilisant trois puis quatre variables, savoir lindice de
prix la consommation, la production industrielle, la base montaire et le taux dintrt de
JGB dix ans. Ce modle prsente lavantage de dtecter les ruptures structurelles sans
imposer a priori des contraintes sur les moments auxquels elles se produisent. Il montre que
la politique dassouplissement quantitatif a un eet extrmement faible sur lactivit et sur
les prix en labsence du canal de transmission du taux dintrt. Plus rcemment, Inou et
Okimoto (2008) et Nakajima et al. (2009a) aboutissent dautres conclusions en utilisant
des modles dirents (MS-VAR et TVP-VAR respectivement) ; ils dtectent un eet positif
de lexpansion de la base montaire sur la production pendant la priode de lassouplissement
13
Introduction gnrale
quantitatif. Leet dune telle expansion sur lination reste cependant limit.
Lors des analyses des eets de la politique montaire ralises dans les travaux
empiriques cits prcedemment seul un petit nombre de variables macroconomiques a t
pris en compte, ceci an de maintenir le maximum de degrs de libert possible. Or la
banque centrale, comme les intervenants sur les marchs nanciers, exploitent un ensemble
dinformation contenant un grand nombre de sries de donnes. Le modle VAR traditionnel
montre ses limites parce quil exige une utilisation parcimonieuse du nombre de variables. Une
alternative, dveloppe dans la littrature rcente, a pour objectif dobtenir des exercices
contrefactuels de politique conomique partir de modles fonds sur la thorie conomique,
savoir, les modles dquilibre gnral intertemporels stochastiques (DSGE)
3
. Ces modles,
de plus en plus utiliss par les banques centrales, prsentent plusieurs avantages. Le premier
est li leur fondement microconomique sur lequel est base lanalyse des comportements
de lconomie lchelle macroconomique. Le deuxime avantage est de placer la rationalit
individuelle des agents privs derrire le comportement global, ce qui est utile pour analyser
limpact de la politique montaire sur les anticipations dagents privs ; ceci est en particulier
intressant pour valuer le canal des anticipations de la QEMP. Cette caractristique permet
ces modle dcarter la critique de Lucas (1976). Le troisime avantage rside dans le
caractre parcimonieux de ces modles qui nexigent pas une grande puissance de calcul
et rendent plus facile linterprtation des rsultats. Enn, plusieurs travaux
4
montrent la
supriorit de ce type de modles par rapport au modle VAR structurel en terme de prvision.
Nanmoins, deux problmes surgissent ; premirement, les modles DSGE, tout comme
les modles VAR, emploient un nombre limit de sries macroconomiques qui sont suppo-
ses rsumer toute linformation pertinente pour lestimation. De ce fait, ils sont galement
sujets aux critiques formules prcedemment
5
. Deuximement, et de faon plus importante,
3
Fernndez-Villaverde (2010) fournit une revue de littrature complte et dtaille sur lvolution
du modle DSGE ces dernires annes.
4
Smets et Wouters (2003), Del Negro et al. (2005) et Collard et Fve (2008).
5
Boivin et Giannoni (2007) proposent une mthode destimation des modles DSGE exploitant
14
Introduction gnrale
les travaux empiriques eectus sur le Japon utilisant le modle DSGE la Smets et Wou-
ters (2003) sont bass sur un chantillon de donnes ne dpassant pas 2001, omettant ainsi
la priode de ZIRP et de QEMP (Sugo et Ueda (2008) et Ichiue et al. (2008)). La date
de n dchantillon est choisie prcisemment an dviter la priode durant laquelle les taux
dintrt nominaux ont atteint leur niveau plancher zro, ainsi ne se confrontatant pas
au problme de non-linarit de la rgle de Taylor due la contrainte de non-ngativit des
taux dintrt (Eggertsson et Woodford (2003)). En eet Braun and Shioji (2006) prcisent
que la prsence de la contrainte de non ngativit dans la rgle de politique montaire cre
deux dicults. Dabord, cela complique la rsolution du modle tant donn que la rgle
de Taylor ne peut pas tre approxime par une fonction linaire. La deuxime dicult est
que la contrainte de non ngativit des taux dintrt nominaux change les proprits de
stabilit du modle, comme prcis par Benhabib et al. (2001). Rcemment, Yano (2009)
et Yano et al. (2010) tentent de rsoudre le problme de non-linarit de la rgle de Taylor.
Ils utilisent la mthode de ltrage particulaire, propose par Genshiro (1996), et estiment un
modle DSGE de taille moyenne prsent sous forme dun modle espace-tat non-linaire
et non-gaussien. Lemploi de cette technique dans un modle DSGE la Boivin et Giannoni
(2007) utilisant un chantillon de donnes plus consquent, prsente une piste de recherche
future intressante. Cette dernire raison motive lapproche choisie dans cette thse.
Par rapport aux travaux empiriques existants sur lassouplissement quantitatif utili-
sant essentiellement la mthodologie VAR, nous visons utiliser une structure moins contrai-
gnante et analyser un ensemble de donnes plus riche. A cet eet, nous utilisons des
modles VAR augments des facteurs (FAVAR) introduits dans lanalyse de la politique mo-
ntaire par Bernanke et al. (2005) et Stock et Watson (2005). En eet, Bernanke et al.
(2005) montrent que le manque dinformation dans lanalyse des modles VAR conduit
deux problmes au niveau des rsultats : (i) plus les informations concernant la banque
un grand nombre de variables macroconomiques ; les auteurs montrent ainsi quune combinaison de
lanalyse factorielle et du modle DSGE permet damliorer la performance de ce type de modle.
15
Introduction gnrale
centrale et le secteur priv contenues dans lanalyse sont limites, plus la mesure des chocs
politiques est biaise ; do lapparition dnigmes qui ont caractris jusque l les modles
VAR
6
(Sims (1992)) ; (ii) les fonctions de rponses obtenues pour les variables tudies
ne permettent pas danalyser les eets de la politique montaire sur des concepts cono-
miques gnraux comme lactivit conomique ou linvestissement, qui ne peuvent pas tre
reprsents par une unique variable. An de pallier le problme de limitation du nombre de
variables, les auteurs ont dvelopp un modle VAR augment par des facteurs (FAVAR).
Les facteurs, en nombre restreint, rsument un grand nombre de variables. Les rsultats de
Bernanke et al. (2005) montrent que, mme en utilisant un schma didentication rcursif
la Sims (1992), le problme des nigmes est rsorb, corroborant ainsi la thse que les
enigmes proviennent dinsusance de donnes exploites et non pas du schma didentica-
tion (Carlstrom et al. (2009)). Ces rsultats sont conrms par Forni et Gambetti (2010)
qui montrent que, en utilisant un schma didentication rcursif, lemploi dun chantillon
de donnes plus large produit des rsultats en accord avec la thorie conomique et rsout
donc le problme des nigmes. Les prix baissent immdiatement et de faon continue suite
un choc de politique montaire restrictif, la raction de la production industrielle a la forme
dune courbe en U retant la neutralit de la monnaie long terme. Enn, le choc de
politique montaire impacte dune faon importante les dynamiques des variables relles et
nominales.
Nanmoins, bien que la mthodologie FAVAR permette deectuer une analyse plus
complte des mcanismes de transmission de la politique montaire, elle ignore les change-
ments potentiels des rgimes montaires. En consquence, lutilisation de ce type de modle
linaire aboutit des interprtations errones des eets de la politique montaire, spcia-
6
Dautres tentatives de rconciliation des rsultats empiriques issus des modles VAR avec la
thorie, se focalisent sur la modication du schma rcursif didentication des chocs structurels.
Cela se fait en imposant soit des restrictions de court et long termes en se basant sur la thorie
conomique (Blanchard et Quah (1989) et Kim et Roubini (2000)), soit des restrictions de signe
(Uhlig (2005)).
16
Introduction gnrale
lement au vu des volutions connues par lconomie japonaise durant ces deux dernires
dcennies
7
. Dans la ligne de Sims et Zha (2006), Fujiwara (2006) et Inoue and Okimoto
(2008) utilisent la modlisation MS-VAR dans laquelle le changement de paramtres du
modle VAR dpend des dirents rgimes, qui sont de nature discrte, stochastiques et
inobservables (Hamilton (1994)). Cette mthode permet non seulement de dtecter les
changements de rgime dune faon endogne, mais de le faire uniquement dans le cas o
ils sont statistiquement signicatifs dune faon simultane pour tous les paramtres ; ce qui
permet ainsi de dater les dirents rgimes de politique montaire. Dautre part, les mo-
dles VAR avec paramtres evolutifs (TVP-VAR) prsentent une modlisation alternative du
changement de paramtres et fournissent plus de exibilit, permettant aux dirents para-
mtres, savoir coecients et volatilits, dvoluer sparment chaque date. Malgr leur
dirences, ces methodologies peuvent tre utilises dune faon complmentaire. Une fois
que les rgimes de politique montaire sont dtects moyennant la mthodologie MS-VAR,
il savre intressant de complter lanalyse en utilisant la mthodologie TVP-VAR pour
dtecter les volutions des paramtres, tant permanentes que graduelles. Une extension du
modle FAVAR a t apporte rcemment par Koop et Korobilis (2009). Leur modle non-
linaire (TVP-FAVAR) comporte des paramtres variables dans le temps et permet donc
la fois de tenir compte dun maximum dinformation et de dtecter dventuelles variations
dans le temps de la relation entre les variables macroconomiques. Bianchi et al. (2009)
ont tendu lutilisation de la mthodologie TVP-FAVAR au modle macro-nance appliqu
lanalyse de la structure par terme et de sa relation avec les variables macroconomiques.
A notre connaissance, aucune de ces mthodologies na encore t applique ltude de
lassouplissement quantitatif au Japon.
7
Shibamoto (2007) tait le seul employer un modle FAVAR linaire pour analyser la politique
montaire japonaise. Toutefois, son tude ne couvre pas la priode de lassouplissement quantitatif.
17
Introduction gnrale
Structure de la thse
Dans la continuit des travaux empiriques prcdemment prsents, les principales contri-
butions de la prsente thse seront dappliquer les techniques conomtriques les plus ap-
propries et les plus rcentes au cas bien particulier du Japon. Ce travail gagnera en nesse
danalyse par rapport aux tentatives prcdentes en incorporant le maximum de variables
lies la politique montaire et en dtectant avec prcision les changements de rgime qui
caractrisent cette dernire.
Dans un premier temps, nous analyserons les eets globaux de la stratgie das-
souplissement quantitatif sur lactivit et sur lination. Dans un deuxime temps, nous
chercherons discerner les canaux de transmission suggrs et mesurer leur ampleurs
laide de deux mthodologies distinctes.
Dans le premier chapitre intitul Quantitative easing works : Lesons from the
unique experience in Japan 2001-2006 est explore globalement lecacit de la politique
dassouplissement quantitatif. A-t-elle russi sortir le Japon de la situation de dation
et stimuler son activit ? Toutefois il ne sera pas prcis par quels canaux ces eets ont
t transmis. Nous commenons par proposer un nouveau modle, nomm MS-FAVAR,
qui combine la mthodologie de Markov-Switching et celle de FAVAR an de tenir compte
dventuels changements de rgimes dans la conduite de la politique montaire japonaise. A
la dirence de Bernanke et al. (2005) et suivant Belviso et Milani (2006) nous attribuons
des interprtations prcises aux facteurs utiliss dans le modle, dans la mesure o ils sont
extraits de direntes bases de donnes lies chacune des notions conomiques direntes.
Ces facteurs reprsentent lactivit conomique, les prix et les taux dintrt. Nous montrons
laide des probabilits lisses que le changement de rgime sest produit en deux tapes :
il est apparu lentement partir de la n de lanne 1995 et sest install durablement en
fvrier 1999. Cette priode est considre comme transitoire dans lconomie japonaise
marque par des changements drastiques au niveau du systme nancier (Fujiwara (2006)).
18
Introduction gnrale
Nous montrons galement que laugmentation de la base montaire pendant le deuxime
rgime, qui englobe la priode de politique du taux dintrt zro et celle de lassouplissement
quantitatif, a un eet positif la fois sur la production et sur lination. Bien que cet eet
soit transitoire, il montre quune politique montaire passive aurait nettement aggrav la
rcession : lassouplissement quantitatif a au moins eu le mrite dempcher lactivit de
se dtriorer davantage. Ainsi, quand la BOJ arme que lassouplissement quantitatif na
pas produit les eets dsirs, il est pertinent de se demander si cette politique montaire a
t maintenue assez longtemps. Leet positif transitoire dtect conrme lhypothse que
lassouplissement quantitatif aurait du tre maintenu plus longtemps que le BOJ ne la fait.
An de pouvoir tirer davantage de leons de lexprience japonaise de lassouplisse-
ment quantitatif, une analyse complmentaire savre tre cruciale pour identier les canaux
de transmission et mesurer leur ampleur. Le chapitre 2 , intitul The Japanese Quanti-
tative Easing Policy under Scrutiny : A Time-Varying Parameter Factor-Augmented
VAR Model, a pour vocation de complter le premier chapitre en dtaillant les eets de la
politique dassouplissement quantitatif sur un grand nombre de variables macroconomiques
et nancires. Dans ce chapitre nous utilisons un modle FAVAR avec paramtres variables
dans le temps (TVP-FAVAR) pour analyser des chocs de politique montaire au Japon. Ce
modle prsente deux avantages supplmentaires ceux du modle MS-FAVAR utilis dans
le premier chapitre. Non seulement les ractions de toutes les variables sous-jacentes aux
facteurs peuvent tre explores, mais aussi, grce la variabilit des paramtres chaque
priode de temps, le choix de la priode tudier seectue dune manire ad-hoc. Cela
nous permet donc danalyser la priode dassouplissement quantitatif dune manire prcise.
Quatre rsultats principaux se dgagent. Tout dabord, nous montrons que le modle o
tous les paramtres varient avec le temps est le mieux mme de spcier la politique
montaire japonaise pendant les deux dernires dcennies. En second lieu, leet de lassou-
plissement quantitatif sur lactivit et les prix est plus important que prcdemment trouv ;
19
Introduction gnrale
en particulier, nous dtectons, pour la premiere fois, une raction signicative des prix un
choc sur la base montaire. De plus, contrairement aux travaux prcdents, nous montrons
que le canal de rquilibrage de portefeuille a un rle non ngligeable dans la transmission
des chocs de la politique montaire. Enn, leet positif et signicatif sur les anticipations
des agents privs de lengagement pris par la BOJ en terme de maintien des taux dintrt
des faibles niveaux, bien que transitoire, semble avoir au moins stopp la spirale dation-
niste. Cette dernire observation requiert une analyse supplmentaire laide dun modle
macro-nance de la structure par terme des taux dintrt qui permette dexaminer avec
plus de prcision les eets des anticipations.
Cette analyse fait lobjet du troisime chapitre, intitul Quantitative Easing, Credi-
bility, and the Time-Varying Dynamic of Japans Term Structure, qui se concentre sur
linteraction entre les variables macroconomiques, dont une variable de politique montaire,
et la structure par terme des taux dintrt. Nous rappelons que lobjectif intermdiaire de
la BOJ consiste faire baisser les taux dintrt nominaux de long terme en ancrant, de
manire crdible, les anticipations des taux dintrt futurs un niveau susamment bas,
niveau compatible avec une ination modre et stable dans le futur. Ce canal danticipation,
appel canal de policy duration eect, naura deet que si la BOJ parvient tre crdible
dans son engagement. Cet eet aboutira, dans un deuxime temps, une augmentation
de lination anticipe et une baisse des taux dintrt rels qui son tour stimulera la
demande globale.
Dans ce chapitre nous analysons la capacit de lassouplissement quantitatif at-
teindre lobjectif nal de la BOJ, savoir la sortie de la dation et la reprise de lactivit
relle. Pour ce faire, nous employons un modle de macro-nance la Nelson-Siegel avec
des paramtres variables dans le temps (TVP-VAR), et utilisons lcart de production et
lination comme variables macroconomiques, ainsi que le taux dintrt au jour le jour. La
structure par terme des taux dintrt est ainsi rsume par trois facteurs qui reprsentent
20
Introduction gnrale
le niveau, la pente et la courbure de la courbe des taux. Lavantage de cette approche,
hormis la prise en compte des changements de rgime et lutilisation des nombreux taux
dintrt caractrisant la structure par terme, est quelle permet dexaminer la fois leet
des variables macroconomiques sur la structure par terme et leet de retour.
Ce chapitre dbouche sur trois rsultats principaux. Premirement, nous mettons en
vidence la validit de la thorie danticipations rationnelles, condition ncessaire leca-
cit du canal de policy-duration eect. Linvalidit de cette hypothse, dtecte par les
tudes empiriques prcdentes, est gnralement explique par la variation dans le temps de
la prime de terme qui nest pas prise en compte par ces modles. Deuximement, les rsultats
des estimations de TVP-VAR montrent que les variables macroconomiques ne contribuent
que faiblement la variance de la structure par terme, surtout pendant la priode de las-
souplissement quantitatif. En ce qui concerne leet de retour de la structure par terme sur
les variables macroconomiques, nous dtectons une contribution marginale de la courbe
des taux la variation de lination, indpendamment de la sous-priode considre ; son
eet sur la production savre cependant plus important. Troisimement, en nous focalisant
sur leet de la politique montaire sur la courbe des taux, nous montrons que la baisse du
facteur niveau de la courbe des taux suite un choc positif sur le taux dintrt de court
terme, bien que non signicative, indique que la crdibilit de la BOJ sest renforce pendant
la priode de lassouplissement quantitatif. Ceci est quivalent une hausse du niveau de la
courbe des taux si on considre la politique de maintien de taux dintrt un niveau bas.
Cela implique une augmentation de lination anticipe et donc une ventuelle augmentation
de la demande globale. Dautre part, alors que leet sur la production est signicatif, leet
sur lination reste ambigu, en raison du problme d nigme des prix qui semble tre li
au nombre restreint de variables macroconomiques considres dans lanalyse.
21
Introduction gnrale
22
1
1
Quantitative easing works: Lessons from the
unique experience in Japan 2001-2006
1
1.1 Introduction
The current nancial crisis has now led most major central banks to rely covertly or overtly
on quantitative easing. The unique Japanese experience of quantitative easing is the only
experience which enables us to judge this therapys eectiveness and determine the appropri-
1
This chapter updates work registered as GREQAM working paper n2010-2 submitted and un-
der revision. We thank Stephen Bazen, Martin Ellison, Andrew Filardo, and Michel Lubrano for
their valuable comments and suggestions. We also thank the participants of The European Doc-
toral Group in Economics (EDGE) (Copenhagen, Denmark) conference, the Theory and Method
of Macroeconomics conference (Strasbourg, France) and the Day of Econometrics at University of
Paris X-Nanterre, as well as the seminar participants at GREQAM (Marseille, France). This chapter
also beneted from presentations at Musashi University and Hitotsubashi University, Tokyo, July
2009, and at the Bank of Japan (BOJ) in September 2009. Special thanks go to Professor Yusho
Kagraoka and all the sta of Musashi University for their kind invitation. We are also grateful to all
the seminar participants at the BOJ for their very useful comments and suggestions and express our
special gratitude to Yuki Teranishi for his invitation.
23
24
ate timing of the exit strategy. It is widely believed that during the "lost" decade in Japan,
characterized both by stagnation and by deation, monetary policy was all but impotent.
Available academic work concludes that quantitative easing, based on ooding banks with
base money, did not manage to stimulate activity or revive ination.
The empirical study of output and price eects of monetary policy using the workhorse
in macroeconomic time series analysis, i.e. VARs (vector auto-regressive models), has been
a very intensive area of research over the last decade (Sims et al. (1990a), Sims and Zha
(1998), Bagliano and Favero (1998) and many others). Such works have usually put a lot
of emphasis on the interest rate as the monetary policy transmission channel. However,
in the case of Japan, when the zero lower bound on short-term interest rates is reached,
the room for further stimulus using a short-term interest rate instrument is constrained.
Recent researches, dealing with the issue of the zero-bound for nominal interest rates, ar-
gue that it is still possible to conduct more accommodative monetary policies to aect the
aggregate demand and prices. The neo-Wicksellian approach for monetary policy analysis
mostly focuses on alternative policies to aect expectations of future short-term interest
rates. Krugman (2000) and Eggertsson and Woodford (2003) argue that a zero interest
rate commitment inuences expectations for the future path of the call rate, and then
leads to reduce medium- to long-term interest rates. However, the monetarist approach
suggests that the focus should be on portfolio-rebalancing channel. Metzler (1995) argues
that, given the imperfect substitutability of dierent nancial assets, a massive increase in
the monetary base could lead the private sector to adjust its portfolio lowering yields on
non-monetary assets. By implementing the quantitative easing monetary policy (henceforth
QEMP), by the the Banque of Japan (BOJ) in March 2001, the monetary policy instrument
was changed to current account balances (henceforth CAB) held by commercial banks with
the BOJ. Two transmission channels for the QEMP have been suggested
2
. The rst is the
2
There are several possible ways to classify transmission channels. See also Ugai (2007)
1.1. Introduction 25
expectation channel, consisting of policy-duration (Krugman (2000) and Eggertsson and
Woodford (2003)) and signaling eects, and the second is the portfolio-rebalancing channel
(Metzler (1995)).
On the other hand, the Bank of Japan holds a large fraction of long-term bonds on its
balance sheet. About 60% of Japanese moneatry base is backed by long-term government
bonds. This measure seems to be in line with Bernanke (2003)s recommandation. Bernanke
(2003) suggests that the BOJ dramatically increases its purshases of Japanese government
bonds. This measure would not only lead to a monetary expansion, but would also enable the
government to carry out greater scal stimulus without increasing the private sectors future
tax burden. Moreover, Eggertsson (2003) argues that if government and the central bank
were to cooperate in an attempt to avoid the deationary trap, this would create ination
expectations in the private sector and lead to a rise in output. Therefore, Eggertsson (2003)
interprets the lack of ination despite the large quantity of JGB issuance under zero interest
rates as evidence of lack of cooperation between Treasury ocials and the central bank.
Now the policy question of major importance is to check whether results related to the
monetary policy eectiveness change when the scal policy is simultaneously taken into
account.
In addition, instabilities in the transmission mechanisms of monetary policy are very
likely, particularly in the case of Japan. In a standard stochastic model, Orphanides and
Wieland (2000) show that, when ination is lower than one per cent, non-linearities in the
transmission process of monetary policy arise solely from the presence of the zero bound on
nominal interest rates. Indeed, these eects become increasingly important for determining
the outcome of monetary policy in circumstances with such low ination rates. On an
empirical level, accounting for regime shifts should be a major concern when examining the
transmission mechanisms of monetary policy (Miyao (2000), Fujiwara (2006), Inoue and
Okimoto (2008) and Nakajima et al. (2009a)).
26
The main objective of this chapter is to asses whether the QEMP is eective in
stimulating the economy and to investigate the potential structural changes in transmission
mechanisms of Japanese monetary policy. We will therefore allow for stochastic regime
switching within a vector-autoregressive model.
Moreover, to conserve degrees of freedom, standard VARs rarely employ more than
six to eight variables. This fact is particularly important in the case of a Markov-Switching
(MS) VAR model when the number of estimated parameters rises very quickly if the number
of variables is large or the lag length is long. Moreover, in reality, policymakers work with an
information set which contains many data series. Bernanke et al. (2005) show that lack of
information in the VAR model leads to two related problems : (i) the less the central bank
and private sector related information is reected by the analysis the more the policy shock
measure is biased. This leads to puzzles which characterize the traditional VAR model.
(ii) impulse response functions are not sucient to analyze the eects of monetary policy
on general economic concepts like real economic activity or investment, which cannot be
represented by one variable only. Factor analysis consists in summarizing a large number of
data series to produce a small number of estimated factors. The Factor-Augmented VAR
(FAVAR) model gained in popularity with the work of Bernanke et al. (2005) and Stock
and Watson (2005). This approach attempts to reconcile traditional empirical results with
standard theory by adding further variables to the data set in the VAR system, instead of
questioning the standard recursiveness assumption of the identication scheme. Combining
Factor-Augmented and Markov Switching VAR models would enable us at the same time
to introduce a realistic amount of information, keep the statistical advantages of using a
parsimonious system, and take into account possible structural changes. We suggest this
combination could yield results more consistent with standard theory. Moreover, following
Belviso and Milani (2006), we also attribute a clear economic interpretation to the factors;
each estimated factor will represent one economic concept namely Real activity, Ination
1.1. Introduction 27
and Interest rates.
The combination of these methodologies in a so-called MS-FAVAR model allows us
to establish three major ndings. First, the results obtained with our model are consistent
with the standard theory and contrast sharply with those of the traditional VAR model.
Our results show that the problems of the price puzzle, the non-neutrality of money and
price divergence which characterized the MS-VAR model are solved with the MS-FAVAR.
Second, we propose new empirical evidence supporting that quantitative easing has positive
eect on both output and prices. Given the uncertainties surrounding the measurement of
output and prices during the great stagnation, using factor analysis to characterize these
two macroeconomic concepts by summarizing a large number of variables errs on the side
of caution. Third, proposing the rst Markov-switching analysis of a FAVAR, we are able
to show that the decisive change in regime occurred in two steps: it crept out in late 1995
and established itself durably in 1999 around the time when the BOJ implemented QEMP.
The impulse responses in the second regime should thus describe precisely the eect of
this non-conventional strategy on output and prices. These results remain valid even when
scal policy is simultaneously taken into account in the analysis. However, according to the
Japanese experience, if the quantitative easing can aect the symptoms it cannot aect the
causes of the Japanese disease such as the nancial distress in the banking system and the
excessive indebtness of the corporate sector.
To conduct this analysis we will proceed as follows. Section 2 discusses the related
literature. The MS-FAVAR model is described in section 3. The following two sections
examine data and estimation results and conduct a range of robustness tests. Then Section
6 develops the implications of the chapters main results for management by the Fed of the
global nancial crisis generated by the burst of the United States housing bubble. Finally,
section 7 concludes.
28
1.2 Related literature
Conclusions on the existence and the timing of the structural changes in Japanese
monetary policy appear to be particularly sensitive to : i) the methodology employed, ii) the
variables taken into account, especially the choice of the monetary policy instrument and
iii) the period considered.
Miyao (2000) estimates a four-variable VAR. His monthly data for the call rate,
industrial production, the monetary base and the nominal eective exchange rates span
the period between 1975 and 1998. The structural change point is imposed exogenously
in 1995 by including dummy variables. Such a treatment of structural change is criticized
by Sims and Zha (2006) who argue that structural changes must be treated endogenously
where regimes are considered as stochastic events. Kamada and Sugo (2006) adopt the
VAR methodology to identify monetary policy shocks by imposing sign restrictions on the
impulse response functions. They use ve variables, namely the CPI, industrial production,
the nominal exchange rate, 10-year JGB yields, and a monetary policy proxy. On the other
hand the authors use the Markov Chain Monte Carlo (MCMC) method to detect dates of
possible structural changes between February 1978 and April 2005. The detected structural
change point corresponds to the peak of the asset price bubble in 1990 and results from
a change in VAR parameters. These authors show that during the post-bubble period the
eect of monetary policy on prices and production weakened.
Fujiwara (2006) uses the Markov-switching methodology within a VAR framework
(MS-VAR) with regime-dependent impulse response functions (Ehrmann et al. (2003)). He
examines the period between 1985 and 2003 by including three and then four macroeco-
nomic variables (industrial production, CPI, the monetary base and the 10-year JGB yield).
This model represents the advantage of detecting regime changes without imposing a priori
constraints on the timing of such changes. Smoothed regime probabilities suggest that the
timing of a major regime change is most likely in 1995 and that the period between 1995
1.3. Transmission Channels of QEMP 29
and 1999 is a transition period. However, this study does not uncover any output or price
eect of monetary base shocks during the pre-1999 regime.
In the spirit of Fujiwara (2006), Inoue and Okimoto (2008) employ a MSVAR model
with ve variables, namely industrial production, the consumer price index, the monetary
base, the call rate and the nominal eective exchange rate. The data span the period
between 1975 and 2002. The monetary base and the call rate both account for the monetary
policy instruments. Two regimes are identied. In the rst regime the monetary policy rate
was eective until late 1995. In the second regime which started in 1996, after the interest
rate fell almost to zero, the eectiveness of interest rate shocks collapsed. However, the
monetary base in this regime has a positive and signicant eect on output but a weak
eect on prices. Mehrotra (2009) uses three variables in the estimation of an MSVAR,
specically output, the ination rate and the call rate, as a monetary instrument. Using
data for the period between 1980 and 2003, he detects structural change in 1994. He nds
that monetary policy still has a moderate impact on output in the second regime but the
ination response displays a price puzzle and remains insignicant.
The common point of all these studies is that a limited set of variables is used
in the analysis. In the present chapter, following the spirit of Fujiwara (2006) and Inoue
and Okimoto (2008), we treat the regime change as a stochastic event by using MSVAR
model and we combine this methodology with factor analysis. Our MS-FAVAR represents
an improvement with respect to the standard MS-VAR model since it does not suer from
the omitted-variable bias and allows a parsimonious system.
1.3 Transmission Channels of QEMP
Several factors limited the number of monetary policy transmission channels in Japan.
First, because overnight rates have already hit the zero bound, real interest rates could only
be aected by expected ination. Consequently monetary policy using the traditional channel
30
of the short-term interest rate was inoperative. Second, the collapse of the Japanese banking
system prevented the activation of the credit channel. Indeed, bank lending declined during
the period between 1999 and 2005 in spite of the ample liquidity provided to the banking
system (Ito and Mishkin (2006) and Ito (2006)).
The literature on monetary policy transmission when nominal short-term interest
rates hit the zero bound has focused on two transmission channels through which the QEMP
could be eective. expectation channel, which consists of policy-duration and signaling
eects, and portfolio-rebalancing channel. The expectation channel is strictly connected to
the commitment to maintain a zero interest rate until core CPI ination becomes zero or
positive year-on-year. This channel was suggested by the neo-wicksellian approach (Krugman
(2000) and Eggertsson and Woodford (2003), to cite just a few). This approach suggests
that a credible policy commitment of maintaining nominal short-term interest rate at very
low level, for a longer period than was previously expected, can inuence expectations for
the future path of the nominal rate. This, in turn decreases the long-term interest rates,
simulating aggregate demand and prices. In addition, any monetary expansion or change
in the central bank balance sheet composition is inecient, but a permanent increase in
the monetary base can be a signal strengthening the central bank credibility of maintaining
short-term at a low level. Several empirical studies
3
detect a signicant eect of policy-
duration through a attening of the yield curve. Nonetheless, more recently Nakajima et al.
(2009b) show that there is no evidence that this eect is transmitted to the real economy.
Signaling eects are suggested by all of the three courses of action included in the QEMP.
However, the most pronounced signal sent by the BOJ to the private sector is when it
purchases long-term JGBs. In other words, the BOJ makes the commitment constraining
because it will incur a capital loss when long-term interest rates increase. Surprisingly, Oda
and Ueda (2007) detect a signicant eect of this channel from the increase in CABs but
3
See Oda and Ueda (2007) and Okina and Shiratsuka (2004a)
1.3. Transmission Channels of QEMP 31
no eect from the increase in the long-term JGB purchases.
On the other hand, the monetarist view argue that the potfolio-rebalancing channel
can work directly when the BOJ alters its asset composition or indirectly through the mech-
anism whereby the monetary base excess would lead the private sector to adjust its portfolio
by buying nancial non-monetary assets. This channel can aect the whole spectrum of
relative asset prices and real wealth through share prices (Metzler (1995)). An increase
in money supply thus leads agents to buy equities in order to obtain the cash balances.
The increase in share prices can boost private spending through two channels, involving re-
spectively Tobins q-theory of investment and wealth eects. According to the former, the
increase in stock prices leads to a higher market value of rms relative to the replacement
cost of capital (the q-ratio) generating a rise in investment by rms. The latter channel
implies that the rise in nancial wealth of consumers associated with higher equity prices
leads them to raise their consumption in line with the rise in their lifetime resources. An
alternative monetarist view focusing on the liquidity premium channel. This view argues
that the imperfect substitution between monetary and non-monetary assets comes from
the qualitative dierence between these assets in term of liquidity. Therefore, increasing
money supply should make private sector more willing to hold other illiquid assets on their
balance sheets. The prices of these assets raise and their yields accordingly decrease (Yates
(2004)). In addition, the change in the central bank balance sheet composition by buying
long-term government bonds may reduce premia for illiquidity for these assets (Andrs et al.
(2004) and Goodfriend (2000)). In this case, the decline in long-term interest rates will be
due to the reduction in the term premiums and not to a reduction in the future nominal
short-term interest rate. From the empirical view, Kimura et al. (2003) and Oda and Ueda
(2007) show that the eect of the portfolio-rebalancing channel is insignicant or too small
considering the extensive amount of the CAB expansion and the JGB purchases.
The BOJ committed itself to maintaining this policy until ination (measured by the
32
CPI excluding perishables) is positive and stable. It predicted in March 2006 that ination
would remain positive and judged that the objective was reached and that it was time to exit
the QEMP. Consequently, the BOJ returned to the traditional instrument, the overnight
interest rate, as the operating target. Nevertheless, the ecacy of QEMP has not been
denitively established empirically. We suggest below to evaluate the eects of such a policy
on the real economy through the channels just cited.
1.4 Methodology
Several criticisms addressed to the VAR approach concerning the identication of
the eects of monetary policy focus on the use of a restricted quantity of information. In
order to conserve degrees of freedom, it is rare to use more than eight variables in a classical
VAR model.
Bernanke et al. (2005) show that the lack of information, from which the VAR
approach traditionally suers, leads at least to two problems. First, taking into account only
a small number of variables in the analysis biases the measures of the monetary policy shocks.
The best illustrations of this problem are the price, interest rate, liquidity and exchange rate
puzzles. Second, the impulse response functions are observed only for variables included in
the model. The analysis thus cannot be done on global economic concepts like economic
activity or productivity, which cannot be represented by a single variable. To remedy these
problems, the authors proposed a combination between factor and VAR analysis. This
approach allows us to summarize a large amount of information in a limited number of
factors which will be used in the VAR model. Moreover, it avoids imprecision and possible
biases in the estimates that arise from the fact that any one observable may be a poor
measure of the relevant underlying concept.
However, in Bernanke et al. (2005)s paper the factors do not have an immediate
economic interpretation. Following Belviso and Milani (2006) we provide a clear interpre-
1.4. Methodology 33
tation to these factors. We seek to identify each factor as a basic force that governs the
economy as real activity, price pressure, interest rates and so on. We follow this litera-
ture and attempt to go a step further, seeking to take into account the possible existence of
structural change in the monetary transmission mechanism. We therefore propose a Markov
switching vector autoregression augmented with economically interpretable factors: we label
this novel approach Markov Switching Factor-Augmented VAR (MS-FAVAR).
1.4.1 MS-FAVAR
Let X
t
and Y
t
be two vectors of economic variables, with dimensions (Nx1) and
(Mx1), where t = 1, 2, ...T is a time index. X
t
denotes the large dataset of economic
variables and Y
t
denotes the monetary policy instrument controlled by the central bank. We
assume that variables in X
t
are related to a vector F
t
with (Kx1) unobservable factors, as
follows :
X
t
= F
t
+e
t
(1.1)
where e
t
are errors with mean zero assumed to be either weakly correlated or uncorrelated;
these can be interpreted as the idiosyncratic components. The (NxK) matrix represents
the factor loadings.
We can think of unobservable factors in terms of concepts such as economic activity
or price pressure. But here, following Belviso and Milani (2006) we divide X
t
into various
categories X
1
t
, X
2
t
, ... X
I
t
which represent various economic concepts, where X
i
t
is a (N
i
x1)
vector and
i
N
i
= N. Each category of X
i
t
is thus assumed to be represented by only F
i
t
which is a (K
i
x1) vector (
i
K
i
= K). That means that each variable in the vector X
i
t
is
inuenced by the state of the economy only through the corresponding factors. Hence we
34
obtain :
_

_
X
1
t
X
2
t
...
X
I
t
_

_
=
_

f
1
0 ... 0
0
f
2
... 0
... ... ... ...
0 0 0
f
I
_

_
_

_
F
1
t
F
2
t
...
F
I
t
_

_
+
_

_
e
1
t
e
2
t
...
e
I
t
_

_
(1.2)
In this analysis we assume that each segment of X
i
t
can be explained by exactly one factor,
that is K
i
= 1 for all i . Also assume that the dynamics of (Y
t
, F
1
t
, F
2
t
, ..., F
I
t
) is given by a
factor-augmented autoregression (FAVAR):
_

_
F
1
t
F
2
t
...
F
I
t
Y
t
_

_
= (L)
_

_
F
1
t1
F
2
t1
...
F
I
t1
Y
t1
_

_
+
t
(1.3)
A Markov-Switching FAVAR is represented by system (1.4). In its most popular
version (Krolzig (1997)), which we will use here, the regime-switching model is based on
the assumption that the process s
t
is a rst-order Markov process. Hamilton (1989), in
his original specication, assumed that a change in regime corresponds to an immediate
one-time jump in the process mean. We rather consider the possibility that the mean would
smoothly approach a new level after the transition from one regime to another. We do it in
an extension of Hamiltons approach to a regime-switching VAR system (Krolzig (1997)).
Z
t
=
_

1
+B
11
Z
t1
+ ... +B
p1
Z
tp
+A
1
u
t
if s
t
= 1
.
.
.

m
+B
1m
Z
t1
+... +B
pm
Z
tp
+A
m
u
t
if s
t
= m
(1.4)
1.4. Methodology 35
where Z
t
=
_
F
1
t
F
2
t
... F
I
t
Y
t

T
. Each regime is characterized by an intercept
i
,
autoregressive terms B
1i
, ... , B
pi
and a variance-covariance matrix A
i
. We assume that m,
the number of regimes, is equal to two. In this general specication all parameters are
allowed to switch between regimes according to a hidden Markov chain
4
. With Markov-
switching heteroscedasticity, the variance of errors can also dier between the two regimes.
After the change in regime there is thus an immediate one-time jump in the variance of
errors. This model is based on the assumption of varying processes according to the state
of the economy controlled by the unobserved variable s
t
. Here s
t
= {1, 2} is assumed to
follow the discrete time and discrete state stochastic process of a hidden Markov chain and
governed by transition probabilities p
i ,j
= Pr (s
t+1
= j |s
t
= i ), and
2
j =1
p
ij
= 1i , j (1, 2).
The conditional probabilities are collected into a transition matrix P as follows:.
p =
_
_
_
p
11
p
12
p
21
p
22
_
_
_ (1.5)
For a given parametric specication of the model, probabilities are assigned to the unob-
served regimes conditional on the available information set which constitutes an optimal
inference on the latent state of the economy. We thus obtain the probability of staying in
a given regime when starting from that regime, as well as the probability of shifting to an-
other regime. The classication of regimes and the dating algorithm used imply that every
observation in the sample is assigned to one of the two regimes. We assign an observation
to a specic regime when the smoothed probability of being in that regime is higher than
one half. The smoothed probability of being in a given regime is computed by using all the
observations in the sample.
4
In the terminology of Krolzig (1997) this specication is an MSIAH(m)-VAR(p) model.
36
1.4.2 Estimation
Our MS-FAVAR approach retains the advantages of a FAVAR model over a simple
VAR. Moreover, it allows us to take into account the instability of the monetary transmission
mechanism. Factors estimated from the subset databases are the unobserved variables that,
with the policy instrument, enter the MS-VAR (equation 1.4). To estimate the factors, the
variables must be transformed to induce stationarity. By contrast the variables used in a
VAR analysis do not need to be stationary. Consequently, we estimate models using variables
in level and cumulated factors
5
.
In the tradition of Sims et al. (1990a), the specication of a VAR system that we
use considers variables in levels
6
. In the case of such VARs with polynomial functions of
time and one or more unit roots, Sims et al. (1990a) show that, independently of the
order of integration of the variables, one can get a consistent estimation of coecients.
Moreover, as Bernanke and Mihov (1998) argue that a level specication yields consistent
estimates
7
whether or not there is cointegration, but dierence specication is inconsistent
if certain variables are cointegrated. Moreover, focusing on the rate of ination would not
seem adequate when examining a period of overall price stability. Mehrotra (2009) examines
whether price- or ination-targeting would be more adequate in the deationary environment
experienced by the Japanese economy. As Mehrotra (2007) and (2009) argue, movements
in the price level seem to be the relevant variable of interest. In particular, when the BOJ
promised to keep its interest rate at zero until the CPI ination stabilizes at zero percent,
such an ination level, at zero percent, actually corresponds to a price level target. By the
inclusion of the price in level, one could argue that the BOJ has adopted an implicit price
5
Examine the graphs of MS-VAR residuals to nd out whether the residuals are well behaved seems
reasonable. Non-stationarity of variables therefore does not impose problem with the estimation.
6
According to the unit root tests for simple variables and factors shown in tables 1.3 and 1.4 in
Appendix D, all variables and cumulated factors are integrated of order one (I(1))
7
Also see Hamilton (1994) who shows that estimating the VAR in level produces consistent
estimates even in situations where the data are integrated or cointegrated.
1.4. Methodology 37
level target.
Despite the limitations of the quantity theory
8
and the zero lower bound constraint on
nominal short-term interest rate, our model aims to capture portfolio-rebalancing and policy-
duration channel eects on aggregate demand and prices. The former channel suggests
that monetary base expansion, permanent or not, can aect nominal demand and price level
through both wealth and substitution eects on nancial and real assets (Metzler (1995),
Yates (2004) and Andrs et al. (2004)). The latter channel works when the monetary
expansion is understood as permanent, reinforcing the commitment to maintaining zero
interest rates and therefore decreasing long term interest rates, which in turn increases
aggregate demand and price level. Moreover, a permanent increase in monetary policy can
aect price level if the monetary expansion is realized by purchasing government bonds, as
argued in Auerbach and Obstfeld (2005). Altogether, monetary base expansion could in the
long run increase price level and could have a temporary eect on activity level.
In this chapter we consider a two-step approach to estimating 1.2-1.4. The rst
step consists in estimating the factors and factor loadings. The second step is estimating
of the MS-VAR using the factors.
1.4.2.1 Factor estimation
The main approach used for the estimation of factors consists in principal com-
ponent analysis. However, as discussed by Belviso and Milani (2006), the factors thus
estimated have unknown dynamic properties because principal components do not exploit
the dynamics of the factors or the dynamics of the idiosyncratic component. Two standard
principal approaches exploit these features to extract the static factors through dynamic prin-
8
The cointegration relationship between variables was explored here. Results of the VECM model
in Appendix F show that there is no evidence supporting the existence of long-term relationships
between production, price, and monetary base. This can be explained by the fact that the money
multiplier was no longer stable after 1990 and, as mentioned by Fujiwara (2006), there is a lack of
evidence to support the presence of a M2 velocity cointegrating relationship after 1985.
38
cipal components: the static principal components method proposed by Stock and Watson
(1998a) and the Generalized Dynamic Factor Model of Forni et al. (2005) (FHLR) that is a
two-step approach based on dynamic principal components. The rst approach is situated
in the time domain while the second is situated in the frequency domain. Both dier from
static principal component analysis in that they allow for a possibility of autocorrelation
between idiosyncratic components. Nonetheless, there are two main dierences between
Stock and Watsons (2002) method and that of FHLR in the way they estimate the space
spanned by the factors. First, Stock and Watsons (2002) approach estimates the factors
using the standard principal components based on a one-sided lter of the variables. But in
the FHLR approach the common factors are estimated by exploiting information about the
degree of commonality between all variables, obtained from covariance matrices of common
and idiosyncratic components, estimated in a rst step. Indeed, the variables are weighted
according to their common and idiosyncratic variances. The variables having the highest
common/idiosyncratic variance ratio (commonality) are selected. Since the weights are in-
versely proportional to the variance of the idiosyncratic components, this method provides
more ecient estimates of common factors.
For the MS-FAVAR approach employed in this chapter, static factors are estimated
by using the GDFM of Forni et al. (2005). Under the GDFM each variable can be written
as the sum of two unobservable components:
x
it
=
it
+
it
= b
i 1
(L)f
1t
+b
i 2
(L)f
2t
+ +b
iq
(L)f
qt
+
it
(1.6)
where
it
is the common component and
t
it the idiosyncratic component; b
i 1
(L), , b
iq
(L)
(i =0, , s) represent the dynamic loadings of order s; f
1t
, , f
qt
are the q dynamic factors.
Equation 1.6 can be written in vector notation:
x
it
=
it
+
it
= B
i
(L)f
qt
+
it
= B
i
F
t
+
it
(1.7)
1.4. Methodology 39
where F
t
= (f

t
, , f

ts
) and B
i
= B
i
(L). The number of static factors is equal to r =
q(s +1).
As noted above, this approach is a two-step process
9
. First, it uses a frequency repre-
sentation of the time series proposed by Forni et al (2000a) to estimate the spectral density
matrices of the common part (

n
(), < ) and of the idiosyncratic part (

n
()).
Then, the covariance matrices of common and idiosyncratic components (

n0
and

n0
respectively) are obtained by using the inverse Fourier transforms of the respective spec-
tral density matrices. Second, by using estimated covariance matrices, eigenvalues and
eigenvectors are estimated by solving the generalized principal components problem:

n0
V
nj
=

n0
V
nj

nj
s.t.V

nj

n0
V
nj
= I
r
(1.8)
where the columns of the (n x r ) matrix V
nj
correspond to the eigenvectors and
nj
is a
diagonal matrix containing the rst largest eigenvalues of

n0
and

n0
on its diagonal.
The rst generalized principal components are estimated as follows:

F
j
nt
= V

nj
x
nt
(1.9)
The static factor loadings are dened as:
_
(V

nj

T
n0
)
1
_

nj
(

n0
)

(1.10)
where

T
n0
=

n0
+

n0
.

F
j
nt
are consistent estimates of the unknown common factors in equation 2.2.
9
The representation theory of the dynamic factor model can be found in Forni et al. (2005)
40
1.4.2.2 MS-FAVAR estimation and identication
In the second step the model is estimated with the EM
10
(ExpectationMaximization)
algorithm. Estimated factors are introduced in 1.4 instead of simple variables in a classical
MS-VAR model. Finally, the condence intervals around the impulse responses are computed
by bootstrapping techniques.
In a Markov-switching VAR, with regime-dependence in the mean, variance and
autoregressive parameters, a large number of parameters can potentially switch between
regimes. Ehrmann et al. (2003) propose using regime-dependent impulse response func-
tions in order to trace out how fundamental disturbances aect the variables in the model,
dependent on the regime. As a result, there is a set of impulse response functions for each
regime. Such response functions are conditional on a given regime prevailing at the time
of the shock and throughout the duration of the response
11
. They facilitate the interpreta-
tion of switching parameters by providing a convenient way to summarize the information
contained in the autoregressive parameters, variances and covariances of each regime. This
approach combines Markov-switching and identication in a two-stage procedure of esti-
mation and identication. First, a Markov-switching unrestricted VAR model is estimated,
allowing means, intercepts, autoregressive parameters, variances and covariances to switch.
Second, in order to identify the system, restrictions can be imposed on the parameter es-
timates to derive a separate structural form for each regime, from which it is possible to
compute the regime-dependent impulse response functions.
The choice of identication assumptions is controversial and has been the subject
of numerous debates in the literature. Dierent sets of identication assumptions can
lead to very dierent conclusions in the policy debate. The dierences between theoretical
prediction and empirical results are known as puzzles. A classical example is the recursive
10
The estimation method, identication and impulse response are detailed in Ehrmann et al. (2003)
11
As shown by Ehrmann et al. (2003) regimes predicted by the transition matrix must be highly
persistent in order to have useful regime dependent impulse functions.
1.4. Methodology 41
structure VAR, initiated by Sims (1980), which fails to nd evidence supporting economic
theory
12
.
The issue of reconciling empirical results and standard theoretical model predictions
is receiving greater attention. There are two approaches to the question Could puzzles
be due to an identication failure or to a decient information set?. The rst approach
focuses on the identication scheme itself and proposes alternative identication schemes,
either using VAR models with non-recursive short- or/and long-run zero restrictions based
on theory (Kim and Roubini (2000), Blanchard and Quah (1989)
13
and Clarida and Gali
(1994) among others), or using sign restriction identication methodology (Uhlig (2005)).
On the other hand, the second approach, rather than questioning the validity of the standard
recursive scheme, explains the presence of puzzles by the deciency of information considered
in VAR models. Sims (1992) shows that adding commodity price as an additional variable
in the VAR system solves the price puzzle. The FAVAR model proposed by Bernanke et al.
(2005) and used in our work generalizes this approach by adding further variables related
to activity and nancial market. This could lead all economic variables, not only prices, to
respond in accordance with theory. We therefore use a recursive structural VAR la Sims
(1980) in order both to check the expected advantages of the fact that our model reects
the economic theory even when a standard Cholesky scheme is adopted, and to facilitate
comparison with VAR results, particularly with Fujiwara (2006)s MS-VAR results.
12
While economic theory predicts that monetary policy has a sizeable eect on prices and pro-
duction and that following a restrictive monetary policy prices decrease immediately at all horizons
and production decreases, assuming an inverted hump shape, recursive VARs generally lead to price
increase (price puzzle) and limited and permanent decrease in production.
13
Note that imposing zero long-run eect has been questioned by Faust and Leeper (1994) who
argue that in nite samples the long-run eect of shocks is imprecisely estimated and the inferences
regarding impulse responses are biased. Moreover, this methodology requires variables to enter into
the model in rst dierences, which can be problematic for the reasons explained in the beginning of
the section (1.3.2).
42
1.5 Empirical Analysis
In the following, we report the results from the estimation of a MS-FAVAR model
on a data set including 3 sub-groups of factors, representing 3 economic concepts, and
a monetary policy instrument. Our vector X
t
contains 135 variables. Since we focus our
empirical analysis on the quantitative easing period our sample spans the period between
1985 : 3 and 2006 : 03 at a monthly frequency. A full description of the database is provided
in appendix B. The standard method to evaluate monetary policy through a VAR model is to
consider the uncollateralized overnight call interest rate as the monetary policy instrument.
In the special case of Japan, where interest rates were almost zero, this method cannot be
applied, because interest rates contained no more information concerning monetary policy.
Theoretical work investigated alternative variables, so-called intermediate variables, which
are not directly controlled by the central bank. These variables can be the long-term interest
rate, the exchange rate, the interest rate spread and a monetary policy proxy (Kamada and
Sugo (2006)). Nevertheless, intermediate variables can be inconvenient as far as they can
react to their own shocks, thereby complicating the identication of monetary policy shocks.
In this chapter, we use the monetary base
14
as the monetary policy instrument to measure
the eects of the quantitative easing policy in Japan. The monetary base thus represents
the only observed factor included in Y
t
.
1.5.1 Estimated Structural Factors
Since subsets of similar variables are considered to extract factors, the comovement
observed in these macroeconomic time series should be strong. A small number of factors
14
The seasonally adjusted M0 was corrected for the Y2K eect related to the temporary surge
in liquidity demand in December 1999 and January 2000. As argued in Juselius (2006) transitory
shocks in the model generate residual autocorrelations and violate the independence assumption of
the VAR model. As the Y2K eect appears as an additive outlier we removed it by estimating an
ARMA model with transitory intervention dummies (see gure 1.6 in appendix A.)
1.5. Empirical Analysis 43
therefore account for a relevant percentage of the overall panel variance. The rst obvious
check of the t of our factor model is to see how well each factor represents each sub-group
of data series. In particular we examine the assumption according to which every sub-group
is represented by only one factor. Following Bernanke et al. (2005)and Belviso and Milani
(2006), this chapter determines the number of static and dynamic factors in an ad hoc way.
For the purpose of statistical identication, Stock and Watson (2005) estimate the number
of static and dynamic factors included in the VAR using Bai and Ng (2002)s criterion which
determines the number of factors present in the data set. However, as Bernanke et al.
(2005) point out, Bai and Ng (2002)s criterion, using the percentage of the variance of
the panel accounted for by common factors, describes comovements among series but does
not determine the number of factors to include in the MS-FAVAR model. In addition, the
number of parameters to estimate in the models depends on the number of variables, lags
and states and can quickly be explosive. We then extract one factor from each sub-group in
order to employ the more parsimonious system. Table 1.1 gives the results on the relative
importance of the rst four factors in explaining the variance of all variables. The rst factor
explains about 34, 59 and 97 percent of the data variability respectively for activity, prices,
and interest rates. Even when an additional factor is added, there is relatively little gain in
the share of variance explained. This conrms the robustness of our assumption considering
only one factor for each sub-group. Figure 1.5 in Appendix A illustrates the estimated
loadings plotted as bar charts for each factor. The numbers on the horizontal axis refer to
the ordering of the series of each subgroup and the factor loadings are on the vertical axis.
The interest rate factor loadings are high (0.6 or higher), while price and activity factor
loadings have a lower level for some variables. This is due to the fact that activity and
price variables are more heterogeneous than interest rates. Nonetheless, it appears that all
variables are involved in constructing the factors since loadings are spread across all series.
Furthermore, Figures 1.7, 1.8 and 1.9 in Appendix C show that cumulative factors clearly
44
Table 1.1: Eigenvalues and percent of variance of rst four factors
Activity factors
F1
a
F2 F3 F4
Eigenvalue 1.73 1.12 0.23 0.14
Percent variance 34.04 8.44 4.87 4.02
Price factors
Eigenvalue 2.04 0.87 0.57 .34
Percent variance 59.39 21.52 9.63 5.85
Interest rate factors
Eigenvalue 2.37 0.76 0.43 0.32
Percent variance 97.18 1.36 0.62 0.27
a
F
i
, (i = 1...4) denotes i th factor.
represent the corresponding variables in level.
1.5.2 Traditional MS-VAR
We rst evaluate Japanese monetary policy using the MS-VAR model following Fu-
jiwara (2006) with four observed variables namely output Y
t
, the price level P
t
, the money
base M0
t
and the 10-years JGB yields BY
t
, but using a longer sample. Identication
achieved through a Cholesky (lower triangular) factorisation of the variance-covariance ma-
trix. The ordering Z = [Y
t
, P
t
, M0
t
, BY
t
] implies that the measure of output, is the most
exogenous variable, the measure of price level can respond contemporaneously to output
only, whereas the instrument of monetary policy, can respond contemporaneously to both
ination and real activity but not to the long term interest rate. The third equation in the
structural VAR is interpreted as a contemporaneous policy rule.
First and foremost, we need to determine the optimal number of regimes to charac-
terize the behavior of the time series studied. Second, the best specication among various
MS-VAR models has to be determined. We tested for linearity by taking the linear model as
the null hypothesis (there is a single regime) and the two-regime model as the alternative.
1.5. Empirical Analysis 45
In this case the usual tests, namely LR, LM and Wald tests, cannot be conducted since the
nuisance parameter is identied only under the alternative. The problem of statistical infer-
ence when the nuisance parameters are unidentied under the null hypothesis has frequently
been addressed. Hansen (1992) and Garcia (1998) propose a non-standard likelihood ratio
test (NSLR) which is calculated as a correction on the p-value of a standard likelihood ratio
test. However, this method does not give exact critical values but only a lower bound for
the limiting distribution of a standard LR statistic and is not developed for VAR models but
for a univariate process. Since the null parameter space contains only two subsets, Cho
and White (2007) show that the NSLR test is not valid if boundary conditions are ignored.
Moreover, Cho and White (2007)s test (QLR) is only applicable on specic models which do
not include the MSVAR. In this chapter we therefore use other tests like the Log-likelihood
or information criteria. The null hypothesis can easily be rejected as shown in Table 1.5 in
Appendix D. Moreover, the plots
15
of the nonlinear model estimation residuals indicate the
absence of residual autoregression and almost all of the standardized residuals fall within
two standard deviations of a zero mean. The two-regime model is therefore supported.
Next, the best specication among various MS-VAR models has to be identied.
In this case the LR test suggested by Krolzig (1997) can be performed without causing
problems. The alternative hypothesis MSIAH-VAR specication
16
, where all parameters
switch between regimes, is tested against the other possible specications. We then test
the hypothesis of no regime dependence in the variancecovariance matrix (MSIA-VAR), in
autoregressive terms (MSIH-VAR) and in both the variancecovariance matrix and autore-
gressive terms (MSI-VAR) for dierent lags.
The likelihood ratio test (Appendix D, Table2.1) suggests that an MSIAH-VAR model
15
Plots are not reported here in order to conserve space and are available upon request from
authors.
16
According to Krolzig (1997)s notation, MSI means that only intercepts are assumed to switch
between regimes, MSIA means that intercepts and coecients are assumed to switch, MSIH means
that intercepts and variance covariance matrices are assumed to switch and MSIAH means that all
parameters are assumed to switch.
46
better ts the data than other MSI-VAR specications for two and three lags. Consequently,
this study applies the Markov switching MSIAH-VAR model in which all parameters, namely,
intercepts, autoregressive terms and variance-covariance matrices are allowed to switch
between regimes. The lag length of three is chosen in order to have serially-uncorrelated
residuals. This lag length is supported by AIC and HQ criteria (Appendix D, Table1.7).
Moreover, according to Table 1.8 in Appendix D, showing the transition matrix, the two-
regimes are highly persistent. Regime dependent impulse responses are therefore an useful
tool to analyze monetary policy of Japan.
Figure 1.1: Regime probabilities for MSIAH-VAR
1990 1995 2000 2005
0.25
0.50
0.75
1.00
Smoothed prob., Regime 1
1990 1995 2000 2005
0.25
0.50
0.75
1.00
Smoothed prob., Regime 2
Figure 1.1 plots smoothed regime probabilities. The Japanese economy was in regime
one up to 1997 and has been in regime two since then, with an advanced warning in 1997 and
early 1998. This result is similar to that of Fujiwara (2006); the 2000 break date coincides
neither with the beginning of ZIRP or QEMP, but lies in-between. The period between
1.5. Empirical Analysis 47
1997 and 2000 can be interpreted as a transition period. This result conrms the choice of
non-absorbing two-state using MS-VAR model. In other words, once a state moves from
State 1 to State 2, it can return to State 1. Smoothed probabilities show that the state
evolution between these two regimes should be modeled as a transitory change. Assuming a
permanent structural change, using dummy variables or subsample analysis (Miyao (2000)),
cannot take into account reversible changes between regimes as it is the case during the
transition period.
The stylized facts on the eects of an expansionary monetary base shock were es-
tablished by Christiano et al. (1999), using impulse response functions. They conclude that
plausible models of the transmission mechanism of a monetary expansion should be consis-
tent at least with the following evidence on price, output and interest rate : i ) the aggregate
price level initially responds very little, ii ) output initially rises, with an inverted j-shaped re-
sponse, with a zero long-run eect of the monetary impulse, and iii ) interest rates initially
fall. Figure 1.2 presents the impulse response functions to a positive shock on the monetary
base. The condence intervals are generated using the 10th and 90th percentile values cal-
culated on the basis of 999 bootstrap replications
17
. Over the 1985-2000 period points i )
and iii ) are almost matched, while understandably, ii ) does not hold. The non-neutrality of
money and the divergence of prices after a shock on the monetary base are striking. Indeed,
output responds immediately in a persistent way, while adjustment in prices takes more than
twice as long. The 2000-2006 regime is characterized by insignicant eects of monetary
base shocks on output. The price level initially decreases insignicantly, a result known as
price puzzle
18
. Evaluating the reaction of long-term interest rates reveals important results.
17
We refer to Davidson and MacKinnon (2000) who considered the problem of choosing the
number of bootstrap replications.
18
Carlstrom et al. (2009) argue that the price puzzle is due to the choice of the standard recursive
specication which is a wrong assumption. However, as shown in Bernanke et al. (2005) and more
recently in Forni and Gambetti (2010), price puzzle can be solved within the FAVAR approach even
when Cholesky identication is employed. Price puzzle therefore is due to a decient information,
including small number of variables in the VAR system, rather than to a wrong identication scheme.
48
Figure 1.2: Response to a monetary base shock in MS-VAR
regime 1 regime 2
0 1 2 3 4 5
0.5
0.0
0.5
Y
0 1 2 3 4 5
0.00
0.25
Y
0 1 2 3 4 5
0.0
0.1
0.2
0.3
P
0 1 2 3 4 5
0.0
0.1
P
0 1 2 3 4 5
0.5
1.0
1.5
MB
0 1 2 3 4 5
0
1
MB
0 1 2 3 4 5
0.05
0.00
0.05
Bond yields
0 1 2 3 4 5
0.05
0.00
0.05
Bond yields
Note: Responses of industrial production (Y), CPI (P) and 10-year JGB yields (Bond yields) to
expansionary monetary policy shock increasing the monetary base (MB) by one standard deviation.
The impulse reaction period is chosen to be 5 years. Solid lines show impulse responses, while
dotted lines represent condence intervals using the 10th and 90th percentile values calculated on
the basis of 999 bootstrap replications.
In regime one the response of the interest rate is negative but insignicant. In regime two
the reaction of bond yields is more substantial but remains insignicant. A look at the
interest-rates reaction reveals that policy-duration and signaling eects could aect prices
in the expected way, even though they remain weak. There is thus little evidence that the
transmission mechanism of Japanese monetary policy at a time of near-zero interest rates
would work essentially through the eects on the term-structure of interest rates.
The MS-FAVAR estimate results can shed some light on this question.
1.5. Empirical Analysis 49
1.5.3 MS-FAVAR
In the following, we present the estimated eects of the QEMP within the aforemen-
tioned specications of model 1.4. Since we identify monetary shocks by using the Cholesky
decomposition, the factor ordering must be determined carefully. The interest rate factor
includes several long-term rates that contain expectations on the economy. Because the
monetary authorities can react only to the current state of the economy, the interest rate
factor is ordered after the monetary base. We consider therefore the following ordering: real
activity factor, price factor, monetary base and the interest rate factor. Information criteria
(Appendix E, Table1.9) suggest that the model is non-linear.
From table 1.10 and table 1.11 in Appendix E, an MSIAH-FAVAR specication, in
which all parameters switch between regimes, is suggested by the LR test and the lag length
supported by two dierent information criteria is two. The transition matrix (Appendix E,
Table 1.12) implies that the regimes are highly persistent. As shown in Figure 1.3, the
change in regime occurred in two steps: it rst appeared in May 1996 and established itself
durably in February 1999. Regime two thus corresponds precisely with the beginning of the
non-conventional monetary policy strategy namely the ZIRP consolidated by the QEMP.
A comparison of Figure 1.4 to Figure 1.2 indicates that broad patterns are roughly
similar, but there are some important dierences between the two gures.
Figure 1.4 shows that, unlike in a classical MS-VAR, the stylized facts aforementioned
are veried in all points in both regimes. By contrast with Fujiwara (2006), Kamada and
Sugo (2006) and Kimura, Kobayashi, Muranaga and Ugai (2003) we detect a positive and
signicant eect on real activity even in the second regime under QEMP. In the pre-1996
regime (regime 1), the response of the output factor is moderate and short lived, while
the response of the price factor is half as large, as quick, but hardly signicant. Under the
second regime, after its initial rise the monetary base subsequently falls smoothly towards its
initial level; within eight months approximately half of the initial innovation has disappeared.
50
Figure 1.3: Regime probabilities for MS-FAVAR
1985 1990 1995 2000 2005
0.25
0.50
0.75
1.00
Smoothed prob., regime 1
1985 1990 1995 2000 2005
0.25
0.50
0.75
1.00
Smoothed prob., regime 2
The response of the output factor is three times as large as under the rst regime, and fty
percent longer-lived. The 90% condence interval indicates that the eect lasts signicantly
for thirteen months. The peak increase is found within 6 months. The monetary shock is
equivalent to a 1% increase in the monetary base. For reference, the total stock of CABs
was about 4.6 trillion yen at the beginning of regime two, so an increase of 1% represents
46 billion yen, leading to an increase in real activity of about 0.15% and 0.1% after six
months and one year, respectively. Moreover, the considerable successive increases in CABs
by 25%, 20%, 100% and so on, will have caused a sizable rise in Japans activity, respectively
of about 3.75%, 3% and 15% after six months. Granted, the magnitude and duration of
this estimated eect seem small in absolute terms and the response of output remains short-
lived. However, even though this eect becomes insignicant at the end of the rst year,
it shows that a passive monetary policy would have made the recession even more severe:
1.5. Empirical Analysis 51
Figure 1.4: Response to a monetary base shock in MS-FAVAR
0 1 2 3 4 5
0.0
0.1
0.2
Regime 2
Activity
0 1 2 3 4 5
0.00
0.05
0.10
Regime 1
Activity
0 1 2 3 4 5
0.025
0.000
0.025
Price
0 1 2 3 4 5
0.025
0.000
0.025
0.050 Price
0 1 2 3 4 5
0
1
2
3 MB
0 1 2 3 4 5
0
1
2
3 MB
0 1 2 3 4 5
0.01
0.00
Interest
0 1 2 3 4 5
0.005
0.000
0.005
Interest
Responses of activity factor (Activity), price factor (Price) and interest rate factor (Interest) to
expansionary monetary policy shock increasing the monetary base (MB) by one standard deviation.
The impulse reaction period is chosen to be 5 years. Solid lines show impulse responses, while
dotted lines represent condence intervals using the 10th and 90th percentile values calculated on
the basis of 999 bootstrap replications.
quantitative easing must have at least prevented a further fall in output. The response of
the price factor, while slightly smaller, is much longer-lived (up to nine months) than under
the pre-1996 regime. The impulse responses indicate that a 1% increase in the monetary
base results in a cumulative 0.05% rise in prices over 5 years. Thus, when the BOJ contends
that quantitative easing did not produce the desired eects, it is at least worth considering
whether the policy was taken far enough. For example, an interesting question is whether
more positive results could have been obtained by extending the duration of quantitative
easing beyond the rst signs of economic recovery in 2005, say until 2008. As argued in
52
Koo (2008), the corporate sector had just nished repaying its debts at the end of 2005,
and this would have given them time to reap the benets.
As compared to the standard MS-VAR, it is possible to see the contribution of
the information contained in the factors and it is then noteworthy that the non-neutrality
of money, the price divergence and the price puzzle, which characterized the MS-VAR
model, disappear with the MS-FAVAR
19
. From the viewpoint of the liquidity premium (Yates
(2004),Andrs et al. (2004) and Goodfriend (2000)), the signicance of the output eect
tends to imply that, at near-zero interest rates, base money and nancial assets are not
perfect substitutes. Portfolio-rebalancing channel could therefore stimulate the economy.
In other words, an increase in the monetary base reduces the liquidity premium and leads
economic agents to adjust their portfolios away from the monetary base to nancial assets,
stimulating investment. On the other hand, policy-duration and signaling eects seem to
be stronger on long-term interest rates in regime two than under regime one; the decline in
the interest rate factor becomes signicant with a delay of one year. However, the positive
eect of this expectation channel remains small since the response of the interest rate factor
veers to be insignicant from the beginning of the second year.
1.5.4 Is a scal stimulus eective?
Most studies on the Japanese scal policy eectiveness during the last two decades argue
that scal policy was impotent or at the best would have prevented deeper depression.
Kuttner and Posen (2001) examine the hypothesis that scal policy was ineective using
a VAR model. They conclude that scal policy was actually eective but when it is tried.
19
This result conrm the view that puzzles can be solved by introducing further information in the
VAR system (Bernanke et al. (2005) and Forni and Gambetti (2010), to cite aonly a few.). VAR
model with standard recursive indentication gives results consistent with standard theory when a
maximum of information related to central bank and private sector is taken into account.
1.5. Empirical Analysis 53
According to Posen (1998) the Japanese scal stimulus was not enough
20
. Guerrero and
Parker (2010), using VAR and VECM models for the period between 1955 and 2009, show
that the scal stimulus may have helped to prevent a more severe balance-sheet recession.
Ihori et al. (2003) argue that the Keynesian scal policy was not eective and thus the
eect of scal policies was too marginal to help macroeconomic activity recover.
However, Bernanke (2003) points out that a scal stimulus could be important if
the BOJ increased dramatically its purchases of Japanese government bonds. He asserts
that this measure would not only lead to an monetary expansion, but would also enable
the government to carry out greater scal stimulus without increasing the private sectors
future tax burden. During the QEMP the BOJ increased its purchases of long-term bonds.
About 60% of Japanese monetary base is backed by long-term government bonds. Morever,
Eggertsson (2003) argues that if government and the central bank were to cooperate in
an attempt to avoid the deationary trap, this would create ination expectations in the
private sector and lead to a rise in output. But if the government and the central bank do
not cooperate and the central bank maximizes an independent objective function, ination
expectations would not form. Therefore, Eggertsson (2003) interprets the lack of ination
despite the large quantity of JGB issuance under zero interest rates as evidence of lack of
cooperation between Treasury ocials and the central bank.
In this section we examine the eectiveness of the scal stimulus and we check
whether results related to the monetary policy eectiveness change when scal policy is
simultaneously taken into account. In order to take into account scal policy we introduce
JGB issues variable
21
ordered rst
22
along with the extracted three factors representing
20
The highest annual structural decit in 1990s was 3,8 percent of GDP in 1996. But this was
comparable to or less than the highs of the united states (3,4 percent), Germany (4 percent), France
(3,6 percent) when none of these countries suered a great a recession.
21
For the reasons explained above we use monthly data as for the scal policy proxy. As most
of data on scal variables are available only at yearly and quarterly frequency we use Japanese
government bond (JGB) issues, which is available at monthly data, as proxy for the budget decit.
22
This ordering of variables means that shock in monetary policy, activity and price have no
54
activity, prices and interest rates. Ideally, we would like to estimate a ve-variable model
including the activity factor, the price factor, the monetary base, the interest rate factor and
JGB issues. However, as explained above, in the Markov-Switching VAR model the number
of parameters to estimate can quickly be explosive when we add variables, lags or states.
For this reason instead of estimating a ve-variable model we estimate two four-variable
models; a model without interest rate factor (JGBissues-activity-price-M0) to evaluate the
eect of monetary policy and scal policy shocks on output and prices and a model with
the interest rate factor but without monetary base (JGBissues-activity-price-interestfactor)
to asses the eect of scal policy shocks on the interest rate factor.
Figure 1.10 (in Appendix F) presents the JGB issues in level and in variation. It is
noteworthy that after a jump following the stimulus package of 24 April 1998 the government
bond issues tended to decrease after 2001. This date coincides with the implementation of
the quantitative easing strategy. This leads us to think that scal and monetary policy went
in opposite direction.
Figure 1.11 (in Appendix F) plots smoothed probabilities. The timing of regimes
does not change and still have a clear interpretation as before. Clearly from the plot, the
regime change coincides with the implementation of the ZIRP and the quantitative easing
strategy in 1999. The results of the JGBissues-activity-price-M0 model are shown in gure
1.12 and 1.13. Figure 1.12 presents the impulse response functions to a positive shock to the
monetary base (M0). It is noteworthy that the results are consistent with the main results
obtained from the model without scal policy variable. The results for the period 1999-2006
imply that quantitative easing is eective in helping activity recover and stimulating prices,
while the response of JGB issues is insignicant for that period. The impulse response
functions to a similar shock to JGB issues are displayed in gure 1.13. The results imply
contemporaneous eect on JGB issues. As argued in Blanchard and Perotti (2002), this delay
assumption reects the fact that in short term government may be unable to adjust its bond issue
in response to changes in monetary and macroeconomic conditions.
1.5. Empirical Analysis 55
that the eect of scal policy is too marginal to revive output and prices. The responses
of macroeconomic variables are insignicant in the two regimes. The results of the model
with JGBissues-activity-price-interestfactor are displayed in gure 1.14. The reactions of
activity, prices and interest rate factors are insignicant in the two regimes except a very
short-lived increase in the interest rate factor in the second regime.
These results are consistent with the main results obtained by most empirical studies
dealing with the eectiveness of Japanese scal policy. The failure of scal policy can be
due to many factors. First, as argued in Koo (2008), there is a problem with calculating the
scal multiplier during a recession. In other words, following a scal policy expansion and
starting from a situation of recession, GDP could remain steady. This can happen thanks to
the scal stimulus. But in this case the econometric models suggest that the scal multiplier
is very low or null when the scal stimulus could have prevented the economy from further
depression and the resultant multiplier would be huge. Second, if the scal stimulus is lower
than the deationary gap the remaining headwind will tend to push the economy into a
deationary spiral and thus will lower the measured multiplier eect of the stimulus. An
alternative interpretation, which is in connection with the last explanation, is that there
is a recurring tendency for overstatement of Japanese government scal packages. Posen
(1998) argues that during the period between 1990 and 1998 only the stimulus package
implemented in the second half of 1995 and the early part of 1996 was large and thus
eective. According to Posen (1998), a scal policy stimulus can work when it is tried.
Moreover, the combined contractionary policies of 1996 and 1997
23
completely oset the
positive eects of the 1995 packages.
23
The 1996 budget was contractionary. Cutting government spending was one of the main targets
of scal reconstruction movements started by Fiscal Restructuring Target in 1996.
56
1.6 Robustness
Our results are based on the four variables which are arranged in order of output,
price, monetary base and long-term interest rate. To check the robustness of the reported
results, we estimated two additional types of models, price-outputmonetary basebond
yields and outputpricemonetary base. The three variables model was previously estimated
by Fujiwara (2006). Neither the change in the ordering of variables (and factors) nor the
exclusion of bond yields change the dominance of the MSIAH-VAR specication. The timing
of regime change in the model price-outputmonetary basebond yields is similar to that
found in the model reported here. However, the exclusion of bond yields (and interest
rate factor) did not change the timing of regimes for both the MS-VAR and MS-FAVAR
in a signicant way. The results obtained from all models are qualitatively similar to the
results presented in the previous section. Under the traditional MS-VAR all models indicate
that the output and price reactions to a positive shock on the monetary base are positive
and signicant during regime one. In the MS-FAVAR the additional models conrm our
basic nding that monetary base shocks still have a positive eect on output and price even
during the second regime. Moreover, we applied the method proposed by Stock and Watson
(1998a) to estimate static factors. The results obtained from the MS-FAVAR using these
factors are very similar to those of using Forni et al. (2005)s methodology. Thus, our basic
ndings remain unaltered even if we include static factors in the estimation.
1.7 Implications and Discussion
The attempt to ght the eects of the global crisis generated by a credit boom built
around the subprime-bubble has led most major central banks to rely on quantitative easing.
The Fed, ECB, and Bank of England announced the adoption of quantitative easing in,
1.7. Implications and Discussion 57
respectively January, March and 2009
24
,
In order to draw lessons from the unique experience of quantitative easing in Japan,
a comparison between the quantitative easing programs implemented by the Fed and the
BOJ is useful. Dierences between the two experiences can be classied into two categories:
those that are related to the preconditions for implementation and those that are related to
the implementation of the quantitative easing itself.
With respect to preconditions, the Japanese experience demonstrates that quantita-
tive easing should be seen as a symptomatic treatment which stimulates activity and prices.
It was preceded by a treatment that addressed the cause of the problems of the Japanese
economy and it is note worthy that the quantitative easing policy was adopted in Japan after
a dramatic change in the nancial framework dealing with nancial distress. Cargill et al.
(2001) investigate changes in the Japanese nancial system and the BOJs evolution since
the early nineties. They argue that the smooth implementation of the big-bang announce-
ments succeeded in establishing an infrastructure
25
for the resolution of bank failures. These
changes in the regulatory environment were combined with a commitment of 60 trillion yen
(roughly 460 billion US$) to clean up the banks balance sheets, a process which had not
been completed yet for the American
26
and European nancial systems. In addition, the
US Treasury Department announced only 30 billion US$ to remove toxic assets from the
24
The Fed has boosted its balance sheet to US$ 2.04 trillion from US$ 946 billion in September
2008. For more details see Credit and Liquidity Programs and the Balance Sheet on the boards
public website at www.federalreserve.gov/monetarypolicy/bst_reportsresources.htm.
25
The establishment of the Financial Supervisory Agency and the Financial Reconstruction com-
mittee in June and October 1998 respectively should provide more transparent reporting of nonper-
forming loans and more direct control over managing the nancial crisis.
26
In the case of the US, the nancial crisis induced the collapse of the nancial markets and
particularly the securities market, which in turn caused a decline in the capacity and willingness of
the nancial system to support lending, thus tightening credit. In this context, the nancial rescue
needed to be oriented mainly towards reating the securities market and particularly the Mortgage-
Backed Securities market (MBS). As explained by Ben Bernanke, the Fed chairman, at the Kansas
City Federal Reserve Symposium in Jackson Hole, the Fed increased its portfolio of mortgage-backed
securities (MBS) in order to reduce their yields and indirectly, to reduce the yields of other assets
(through the portfolio-rebalancing channel).
58
banks balance sheets, an intervention which now seems insucient in view of the size of
the current crisis and the Japanese experience.
As regards the implementation itself, the BOJ and the Fed used dierent approaches.
One principal dierence is related to the timing; it took 10 years after the bubble burst in
Japan for the authorities to take on quantitative easing, while the Fed rapidly adopted this
strategy, just one year after the USA entered into nancial crisis in 2007. The Fed, therefore,
was more reactive. The second dierence concerns the total amount of CABs devoted to
this strategy. After only one year the increase in the reserves held by the banks with the
Fed, roughly 8% of GDP, already exceeded the level reached by those with the BOJ during
the ve years of quantitative easing, between 2001 and 2006 (6% of GDP). The BOJ also
had a commitment to a clear numerical target for ination and a xed 5-year timetable. In
contrast, the Fed prefered exibility. This ruled out a clear commitment, probably reduced
uncertainty and allowed for better control of ination expectations.
This comparison is interesting against the background of the Feds discussions about
exit strategies at the time. One issue involved choosing between increasing either short-
term or long-term interest rates. Assuming the Fed raised short-term rates, it would face
the decision of whether or not to reduce the excess reserves in the banking system. If the
decision were taken to reduce excess reserves, the magnitude and timing of such a reduction
would need to be considered. A similar debate occurred within the BOJ in late 2004, at the
end of the series of increases in the CABs. The BOJ chose to raise short-term rates at the
end of the quantitative program while CABs were sharply reducing prior to this.
The Japanese experience suggests that eorts to clean up the banks balance sheets
signicantly improved the eectiveness of quantitative easing. However, this eect, although
considerable, was short-lived; it became insignicant after one year. The short duration
of this eect conrms the wisdom of the Feds decision to maintain quantitative easing
longer, so that being short-lived, the positive eects could be exploited. In the light of
1.8. Conclusion 59
the Japanese experience, we argue that, in addition to their fast reaction and the huge
amount of CABs employed, which may have helped relieve short-term liquidity pressures in
the nancial system, the Fed was better o postponing its exit from quantitative easing.
1.8 Conclusion
Facing zero lower bound interest rates, the BOJ was an early convert to QEMP. In
this chapter we propose an FAVAR approach combined with a Markov-Switching method in
order to analyze the eectiveness of the Japanese monetary policy. We implement a two-
step approach. First, structural factors are estimated from subset databases representing
dierent economic concepts. Second, a Markov-switching model is estimated.
Three main conclusions can be drawn from this work. First, we show for the rst time
that when the Bank of Japan began QEMP, this strategy had a positive eect on activity
and prices. However, this eect was short-lived: it lasted only one year. Our results contrast
with almost all available empirical evidence on the eects of this policy. The contrast does
not stem from our use of regime-switching analysis, but rather from our use of factor analysis
in order to account for the myriad of variables which may have been interacting under this
new monetary policy of the BOJ. The transient positive eect found, even when sizable,
bears out the hypothesis that quantitative easing needs to be maintained longer than the
BOJ did, and should be seen as a symptomatic treatment. Recession and deation were
the symptoms and not the sources of the disease of the Japanese economy, suggesting that
QEMP needed to be coupled with the necessary restructuring of the nancial system.
Second, in contrast to the MS-VAR approach, our MS-FAVAR allowed us to detect
changes in monetary policy mechanisms in a reliable way; structural change occurred in
February 1999 after a period of transition starting in May 1995. Third, we show that the
MS-FAVAR model yields results consistent with standard theory. Thus, the price puzzle, the
non-neutrality of money and the price divergence in the pre-1995 regime, which characterized
60
the MS-VAR model, disappear with the MS-FAVAR. Our ndings thus conrm the idea
that exploiting a larger and more realistic information set proves a more reliable way to
model monetary policy behavior. Our conclusion is that quantitative easing, when coupled
with nancial reforms, can have positive eects on the economy. However, the Japanese
experience suggests that we should not expect quantitative easing to deal with such a serious
crisis in the short term, since it needs to be applied long enough for the benets to work
thier way through to activity and prices.
In the subsequent chapter we will investigate in detail the transmission mechanisms
of Japanese monetary policy. The Interest rate factor seems to be operative and responsible
for monetary policy inuence. However, this factor can be aected both by the expectation
and the portfolio-rebalancing channels. It will therefore be interesting to determine to what
degree each factor aects every transmission channel.
1.8. Conclusion 61
Appendices
A- Factor loadings
Figure 1.5: Estimated factor loadings
Note: The loadings are spread across many series. The numbers on the horizontal
axis refer to the ordering of the series of each subgroup and correlations between the
variables and the rst factors (factor loadings) are on the vertical axis.
62
Figure 1.6: The original and corrected M0
Source: Bank of Japan
Note: The monetary base (M0) is corrected for the Y2K eect when the BOJ had
provided an exceptionally large amount of funds in the market.
B- Data description
Table 1.2: Variable list
Data are extracted from Reuters EcoWin database. The transformation codes (T) are:
1 no transformation; 2 rst dierence; 4 logarithm; 5 rst dierence of
logarithm.
N Description T
Real activity factor
1 Industrial Production Total Index 5
2 Production, Capital goods, SA, Index 5
3 Production, Ceramics, stone and clay products, SA, Index 5
4 Production, Chemicals, SA, Index 5
5 Production, Construction goods, SA, Index 5
6 Production, Consumer goods, SA, Index 5
1.8. Conclusion 63
7 Production, Domestic vehicle, total 5
8 Production, Durable consumer goods, SA, Index 5
9 Production, Fabricated metals, SA, Index 5
10 Production, Food and tobacco, SA, Index 5
11 Production, General machinery, SA, Index 5
12 Production, Iron and steel, SA, Index 5
13 Production, Manufacturing, SA, Index 5
14 Production, Mining and manufacturing, SA, Index 5
15 Production, Non-durable consumer goods, SA, Index 5
16 Production, Non-ferrous metals, SA, Index 5
17 Production, Other manufacturing, SA, Index 5
18 Production, Petroleum and coal products, SA, Index 5
19 Production, Plastic products, SA, Index 5
20 Production, Precision instruments, SA, Index 5
21 Production, Producer goods, SA, Index 5
22 Production, Pulp, paper and paper products, SA, Index 5
23 Production, Semiconductor devices, SA, Index 5
24 Production, Textiles, SA, Index 5
25 Production, Transport equipment, SA, Index 5
26 Shipments, Capital goods excl transport equipment, SA, Index 5
27 Shipments, Capital goods, SA, Index 5
28 Shipments, Construction goods, SA, Index 5
29 Shipments, Consumer goods, SA, Index 5
30 Shipments, Durable consumer goods, SA, Index 5
31 Shipments, Mining and manufacturing, Index 5
32 Shipments, Mining and manufacturing, Index 5
64
33 Shipments, Non-durable consumer goods, Index 5
34 Shipments, Producer goods total, Index 5
35 Shipments, Producer goods, for mining and manufacturing, Index 5
36 Shipments, Producer goods, for others, Index 5
37 Capacity Utilization, Operation Ratio, Fabricated metals, Index 5
38 Capacity Utilization, Operation Ratio, General machinery, Index 5
39 Capacity Utilization, Operation Ratio, Iron and steel, Index 5
40 Capacity Utilization, Operation Ratio, Machinery industry, Index 5
41 Capacity Utilization, Operation Ratio, Manufacturing excluding machinery industry, Index 5
42 Capacity Utilization, Operation Ratio, Manufacturing, Index 5
43 Capacity Utilization, Operation Ratio, Petroleum and coal products, Index 5
44 Capacity Utilization, Operation Ratio, Pulp, paper and paper products, Index 5
45 Capacity Utilization, Operation Ratio, Textiles, Index 5
46 Capacity Utilization, Operation Ratio, Petroleum chemicals products, Index 5
47 Capacity Utilization, Operation Ratio, Rubber products, Index 5
48 Capacity Utilization, Operation Ratio, Transport equipment, Index 5
49 Hours Worked, Average Per Month, Electricity, gas, heat and water 1
50 Hours Worked, Average Per Month, Manufacturing 1
51 Hours Worked, Average Per Month, Mining 1
52 Unemployment, Rate 1
53 Labour Productivity, Foodstu and tobacco (30 employees or more), Index 5
54 Labour Productivity, Furniture (30 employees or more), Index 5
55 Labour Productivity, Manufacturing (30 employees or more), Index 5
56 Labour Productivity, Textiles (30 employees or more), Index 5
57 Employment, Overall, Total 5
58 Sales at Deapartement Stores (Total) 5
1.8. Conclusion 65
59 Wholesale Trade, Food and beverages, JPY 5
60 Wholesale Trade, Furniture and house furnishing, JPY 5
61 Wholesale Trade, General merchandise, JPY 5
62 Wholesale Trade, Machinery and equipment, JPY 5
63 Wholesale Trade, Minerals and metals, JPY 5
64 Wholesale Trade, Others, JPY 5
65 Wholesale Trade, Textiles, JPY 5
66
Wholesale Trade, Total, JPY 5
67 Housing Starts, Housing built for sale 4
68 Housing Starts, Private homes 4
69 Housing Starts, Rental homes 4
70 Housing Starts, Total 4
71 Inventory Mining and manufacturing, Index, JPY, 2000=100 5
72 Inventory Construction goods, Index, JPY, 2000=100 5
73 Inventory Capital goods, Index, JPY, 2000=100 5
74 Inventory Durable consumer goods, Index, JPY, 2000=100 5
75 Inventory Non-durable consumer goods, Index, JPY, 2000=100 5
76 Inventory Consumer goods, Index, JPY, 2000=100 5
77 Inventory Producer goods, Index 5
78 New Orders, Construction, State organizations 5
79 New Orders, Construction, Total, big 50 constructors 5
80 New Orders, Construction, Works abroad 5
81 New Orders, Construction, Works executed 5
82 New Orders, Construction, Works yet to be executed 5
83 New Orders, Machine Tools, Total demand 5
66
Price factor
84 Japan, Consumer Prices, Nationwide, All Items, General, Index, JPY, 2000=100 5
85 Japan, Consumer Prices, Industrial products,All, Index, JPY, 2000=100 5
86 Japan, Consumer Prices, Industrial products,Textile, Index, JPY, 2000=100 5
87 Japan, Consumer Prices, Electricity, gas & water charges , Index, JPY, 2000=100 5
88 Japan, Consumer Prices, Services , Index, JPY, 2000=100 5
89 Japan, Consumer Prices, Durable goods , Index, JPY, 2000=100 5
90 Japan, Consumer Prices, Non Durable goods , Index, JPY, 2000=100 5
91 Japan, Consumer Prices, Food , Index, JPY, 2000=100 5
92 Japan, Consumer Prices, Reading and Recreation , Index, JPY, 2000=100 5
93 Japan, Consumer Prices, Reading and Recreation, Recreational durables , Index, JPY,
2000=100
5
94 Japan, Consumer Prices, Reading and Recreation, Recreational goods , Index, JPY, 2000=100 5
95 Japan, Consumer Prices, Reading and Recreation, Recreational Services , Index, JPY,
2000=100
5
96 Japan, Consumer Prices, Nationwide, Clothing and Footwear, Hats and caps, Index, JPY,
2000=100
5
97 Japan, Consumer Prices, Nationwide, All Items, General excluding imputed rent, Index, JPY,
2000=100
5
98 Japan, Consumer Prices, Nationwide, Miscellaneous Goods and Services, Durable goods, Index,
JPY, 2000=100
5
99 Japan, Consumer Prices, Nationwide, Transport, Private transportation, Index, JPY, 2000=100 5
100 Japan, Consumer Prices, Nationwide, Transport, Public transportation, Index, JPY, 2000=100 5
101 Japan, Consumer Prices, Nationwide, Communication, Communication, Index, JPY, 2000=100 5
102 Japan, Corporate Goods Prices, Domestic demand products, nondurable consumer goods,
Index, JPY, 2000=100
5
103 Japan, Corporate Goods Prices, Domestic demand products, total, Index, JPY, 2000=100 5
1.8. Conclusion 67
104 Japan, Corporate Goods Prices, Domestic, capital goods, Index, JPY, 2000=100 5
105 Japan, Corporate Goods Prices, Domestic, chemicals, Index, JPY, 2000=100 5
106 Japan, Corporate Goods Prices, Domestic, consumer goods, Index, JPY, 2000=100 5
107 Japan, Corporate Goods Prices, Domestic, total, Index, JPY, 2000=100 5
108 Japan, Corporate Service Prices, All items, Index, JPY, 2000=100 5
109 Japan, Corporate Service Prices, Transportation, Index, JPY, 2000=100 5
110 Japan, Corporate Service Prices, Finance and insurance, Index, JPY, 2000=100 5
Interest rate factor
111 Call Rates, Collateralized Overnight (a)/Average(b) 1
112 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed
Banks,Stock/Short-term Loans/City Banks
1
113 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed Banks,
Stock/Short-term Loans/Regional Banks
1
114 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed Banks,
Stock/Short-term Loans/Regional Banks II
1
115 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed Banks,
Stock/Long-term Loans/City Banks
1
116 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed Banks,
Stock/Long-term Loans/Regional Banks
1
117 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed Banks,
Stock/Long-term Loans/Regional Banks II
1
118 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed Banks,
Loans/Regional Banks II
1
119 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed Banks,
Discounts/Shinkin Banks
1
120 Average Contracted Interest Rates on Loans and Discounts of Domestically Licensed Banks,
Stock/Total/Shinkin Banks
1
121 (Discontinued)Average Interest Rates on Certicates of Deposit (New Issues)/Total (through
February 2000)
1
68
122 (Discontinued)Average Interest Rates on Certicates of Deposit (New Issues)/60 days - 89
days (through February 2000)
1
123 Japan, Interbank Rates, BBA LIBOR, 3 Month, End of Period, JPY 1
124 Japan, Interbank Rates, Collateralized Overnight, Average, JPY 1
125 Japan, Treasury Bills, Bid, 3 Month, Yield, End of Period, JPY 1
126 Japan, Prime Rates, Discounts, Regional Banks II, End of Period, JPY 1
127 Japan, Prime Rates, Discounts, Regional Banks, End of Period, JPY 1
128 Japan, Prime Rates, Discounts, Shinkin Banks, End of Period, JPY 1
129 Japan, Prime Rates, Finance Corporations, Key Lending Rates, - 5 Year, End of Period, JPY 1
130 Japan, Prime Rates, Loans, City Banks, End of Period, JPY 1
131 Japan, Prime Rates, Prime Lending Rate, Long Term, End of Period, JPY 1
132 Japan, Prime Rates, Prime Lending Rate, Short Term, End of Period, JPY 1
133 Japan - Benchmark bond - Japan 10-year Government Benchmark bond yield - Yield, average
of observations through period - Japanese yen
1
134 Government Bond Yield, 10 Year, Average 1
135 10-year interest-bearing Government Bonds 1
136 10-year Local Government Bonds 1
137 10-year Government Guaranteed Bonds 1
138 5-year interest-bearing Bank debentures 1
1.8. Conclusion 69
C
-
E
s
t
i
m
a
t
e
d
f
a
c
t
o
r
s
F
i
g
u
r
e
1
.
7
:
A
c
t
i
v
i
t
y
f
a
c
t
o
r
1
9
8
5
1
9
9
0
1
9
9
5
2
0
0
0
2
0
0
5
4
.
3
4
.
4
4
.
5
4
.
6
4
.
7
4
.
8
4
.
9
L
r
g
d
p
L
c
o
n
s
t
g
o
o
d
s
L
d
u
r
c
o
n
g
d
L
M
i
n
i
n
g
m
a
n
u
f
L
c
a
p
u
t
o
r
c
h
L
c
a
p
u
t
o
r
m
i
n
d
L
c
a
p
u
t
o
r
f
a
b
m
L
c
a
p
u
t
o
r
m
a
n
g
L
c
a
p
u
t
o
r
p
p
p
p
L
c
a
p
u
t
o
r
t
e
x
t
L
c
a
p
i
t
a
l
_
g
o
o
d
s
L
c
o
n
s
g
o
o
d
s
L
m
a
n
u
f
L
N
o
n
d
u
r
a
c
o
n
s
g
d
L
c
a
p
u
t
o
r
c
c
s
p
L
c
a
p
u
t
o
r
e
l
e
c
m
a
L
c
a
p
u
t
o
r
g
e
n
m
L
c
a
p
u
t
o
r
n
f
m
e
t
L
c
a
p
u
t
o
r
t
r
a
n
s
c
u
m
f
a
c
t
i
v
i
t
y
N
o
t
e
:
T
h
e
r
e
a
l
a
c
t
i
v
i
t
y
f
a
c
t
o
r
i
s
e
s
t
i
m
a
t
e
d
f
r
o
m
t
h
e
v
e
c
t
o
r
o
f
a
l
l
t
h
e
8
3
t
h
e
r
e
a
l
a
c
t
i
v
i
t
y
r
e
l
a
t
e
d
v
a
r
i
a
b
l
e
s
.
T
h
e
b
o
l
d
l
i
n
e
s
h
o
w
s
t
h
e
c
u
m
u
l
a
t
i
v
e
a
c
t
i
v
i
t
y
f
a
c
t
o
r
,
w
h
i
l
e
t
h
e
d
o
t
e
d
l
i
n
e
s
r
e
p
r
e
s
e
n
t
s
i
m
p
l
e
v
a
r
i
a
b
l
e
s
i
n
l
o
g
l
e
v
e
l
r
e
l
a
t
e
d
t
o
t
h
e
r
e
a
l
a
c
t
i
v
i
t
y
o
u
t
l
i
n
e
d
i
n
t
a
b
l
e
2
.
70
F
i
g
u
r
e
1
.
8
:
P
r
i
c
e
f
a
c
t
o
r
1
9
8
5
1
9
9
0
1
9
9
5
2
0
0
0
2
0
0
5
4
.
2
4
.
4
4
.
6
4
.
8
5
.
0
L
c
p
i
d
u
r
L
c
p
i
s
e
r
v
i
c
e
L
c
p
i
c
o
m
m
u
n
i
c
a
t
i
o
n
L
c
p
i
t
r
a
n
s
p
c
o
m
L
c
p
i
s
u
b
g
r
f
L
c
o
r
p
o
r
a
t
e
2
c
u
m
f
p
r
i
c
e
L
c
p
i
p
r
o
d
a
f
L
c
p
i
m
e
d
i
c
i
n
e
s
L
c
p
i
l
o
g
e
L
c
p
i
s
u
b
g
r
m
i
s
c
L
c
o
r
p
o
r
a
t
e
1
L
C
o
r
p
o
r
a
t
e
3
N
o
t
e
:
T
h
e
p
r
i
c
e
f
a
c
t
o
r
i
s
e
s
t
i
m
a
t
e
d
f
r
o
m
t
h
e
v
e
c
t
o
r
o
f
a
l
l
t
h
e
2
7
p
r
i
c
e
r
e
l
a
t
e
d
v
a
r
i
a
b
l
e
s
.
T
h
e
b
o
l
d
l
i
n
e
s
h
o
w
s
t
h
e
c
u
m
u
l
a
t
i
v
e
p
r
i
c
e
f
a
c
t
o
r
,
w
h
i
l
e
t
h
e
d
o
t
e
d
l
i
n
e
s
r
e
p
r
e
s
e
n
t
p
r
i
c
e
r
e
l
a
t
e
d
v
a
r
i
a
b
l
e
s
,
e
x
p
r
e
s
s
e
d
i
n
l
o
g
l
e
v
e
l
,
d
e
s
c
r
i
b
e
d
i
n
t
a
b
l
e
2
.
1.8. Conclusion 71
F
i
g
u
r
e
1
.
9
:
I
n
t
e
r
e
s
t
r
a
t
e
f
a
c
t
o
r
N
o
t
e
:
T
h
e
i
n
t
e
r
e
s
t
r
a
t
e
f
a
c
t
o
r
i
s
e
s
t
i
m
a
t
e
d
f
r
o
m
t
h
e
v
e
c
t
o
r
o
f
a
l
l
t
h
e
2
8
i
n
t
e
r
e
s
t
r
a
t
e
r
e
l
a
t
e
d
v
a
r
i
a
b
l
e
s
.
T
h
e
b
o
l
d
l
i
n
e
s
h
o
w
s
t
h
e
c
u
m
u
l
a
t
i
v
e
i
n
t
e
r
e
s
t
r
a
t
e
f
a
c
t
o
r
,
w
h
i
l
e
t
h
e
d
o
t
e
d
l
i
n
e
s
r
e
p
r
e
s
e
n
t
s
o
m
e
i
n
t
e
r
e
s
t
r
a
t
e
r
e
l
a
t
e
d
v
a
r
i
a
b
l
e
s
o
u
t
l
i
n
e
d
i
n
t
a
b
l
e
2
.
72
D- MS-VAR estimation results
Table 1.3: Unit root tests (Sample period 1985:3 to 2006:3)
ADF
a
PP ERS KPSS SIC lag
b
DET
c
LS Break dates
CPI -1.326
d
-1.541 -0.503 0.017* 1 C
IP -0.361 -1.451 -0.716 4.987* 5 C
M0 -0.419 -1.109 -0.298 2.089* 3 C
JGB Y -1.045 -0.349 -1.013 1.398* 1 C
CPI -5.016* -15.013* -2.907* 1.513* 0 C -12.58* 88:05-91:10
IP -6.245* -21.459* -2.716* 0.134 4 C
M0 -5.815* -14.491* -5.261* 0.656* 3 C -6.3280* 01:08-04:02
JGB Y -9.180* -12.841* -5.053* 0.145 1 C
a
The 5% critical values for the tests including a constant are -2.89 for the Augmented
DickeyFuller (ADF) and the PhillipsPerron (PP) test, -1.95 for the Elliot, Stock and Roten-
berg (ERS) test and 0.46 for the Kwiatkowski, Phillips, Schmidt and Shin (KPSS) test.
Whereas the ADF test, the PP test and ERS test have the null hypothesis that the vari-
able tested is nonstationary, the null hypothesis for the KPSS test is stationarity. Lee and
Strazicich (2003)s model (LS) allows for two endogenous breaks both under the null hypoth-
esis of a unit root and the alternative one. The critical values of LS depends the location
of breaks. While rst dierences in industrial production (IP) and 10-year JBG yield appear
to be stationary, those in consumer price index (CPI) and monetary base (M0) are stationary
according to ADF, PP and ERS tests but non-stationary according to the KPSS test. The LS
test indicates that rst dierences in CPI and M0 are stationary with break in intercept.
b
Number of lags included in the test was chosen by the Schwarz information criterion (SIC).
c
This column indicates whether a constant (C) or a trend and a constant (T) are included
in the test regression.
d
* The rejection of the null hypothesis at the 5% level.
Table 1.4: Unit root tests (Sample period 1985:3 to 2006:3)
ADF
a
PP ERS KPSS SIC lag
b
DET
c
Price -4.576*
d
-13.302* -0.907* 0.34 0 C
Activity -3.78* -8.783* -3.28* 0.08 4 C
Int rate -9.731* -9.264* -8.396* 0.05 1 C
a
Price (Price), activity (Activity) and interest rate (Int rate)
factors appear to be stationary
b
Number of lags included in the test was chosen by the Schwarz
information criterion (SIC).
c
This column indicates whether a constant (C) or a trend and a
constant (T) are included in the test regression.
d
* The rejection of the null hypothesis at the 5% level.
1.8. Conclusion 73
Table 1.5: Linearity test:VAR model
Lags IC Two regimes
a
single regime
Lag1
AIC -20.4985 -20.2435
HQ -20.3584 -20.0434
SC -19.7938 -19.6435
Lag2
AIC -20.4987 -20.2834
HQ -20.3402 -20.0345
SC -19.7058 -19.7003
Lag3
AIC -20.2358 -20.1348
HQ -19.8345 -19.7905
SC -19.3345 -19.3104
a
Four variable MSVAR with output, price level,
monetary base and bond yield. All information crite-
rion (values in bold font) for all number of lags support
the presence of regime shifts.
74
Table 1.6: MS specications among various MS-VAR models
IC MSI
a
(2) MSIA(2) MSIH(2) MSIAH(2)
Lag1
Log-L 2675.3495 2702.3459 2833.4954 2843.8345
Parameters 36 52 46 62
LR test
b
336.97 282.9772 20.6782 -

2
(R) 38.885 18.307 26.296 -
Lag2
Log-L 2683.8454 2738.3245 2826,3455 2860.3432
Parameters 52 84 62 94
LR test 353.9894 244.0374 67.9954 -

2
(R) 58.124 18.307 46.194 -
Lag3
Log-L 2686.3485 2740.3428 2846.4328 2888.2394
Parameters 68 116 78 126
LR test 403.7818 295.7932 83.6132 -

2
(R) 76.778 18.307 65.171 -
a
According to Krolzig (1997)s notation, MSI means that only inter-
cepts are assumed to switch between regimes, MSIA means that intercepts
and coecients are assumed to switch, MSIH means that intercepts and
variance covariance matrices are assumed to switch and MSIAH means
that all the parameters are assumed to switch.
b
All the calculated values of Likelihood Ratio test, except for lag = 1,
are greater than Chi2 tabulated values. All the specications are thus
outperformed by the MSIAH.
1.8. Conclusion 75
Table 1.7: Lag length test:MSIAH-VAR model
AIC
a
HQ SC
Lag = 1 -20.4985 -20.3584 -19.7938
Lag = 2 -20.4987 -20.3402 -19.7058
Lag = 3 -20.2358 -19.8345 -19.3345
a
The lag length supported by the IC (values
in bold font) is three.
Table 1.8: Transition matrix
Regime 1
a
Regime 2
Regime 1 0.9227 0.0773
Regime 2 0.0563 0.9437
a
Note that p
i ,j
= Pr (s
t+1
= j |s
t
= i )
76
E- MS-FAVAR estimation results
Table 1.9: Linearity test: MS-FAVAR
Lags IC Two regimes
a
Linear FAVAR
b
Lag1
AIC -24.2349 -23.5437
HQ -23.8645 -23.5889
SC -23.5984 -23.4787
Lag2
AIC -24.4375 -24.4048
HQ -24.1653 -24.1648
SC -23.7375 -23.7861
Lag3
AIC -24.4348 -24.3904
HQ -24.0849 -24.0394
SC -23.5103 -23.4938
a
The presence of two regimes is supported
by all the information criterion for all number of
lags except SC criteria for two lags.
b
The four variables MS-FAVAR consist of
real activity, price and interest rate factors and
monetary base.
1.8. Conclusion 77
Table 1.10: MS specications among various MS-FAVAR model
IC MSI(2) MSIA MSIH MSIAH
Lag1
Log-L 3123.5672 3145.2763 3239.2340 3254.1346
Parameters 36 52 46 62
LR test
a
261.1348 217.7166 29,8012 -

2
(R) 38.885 18.307 26.296 -
Lag2
Log-L 3167.3458 3243.5745 3345.9074 3354.3409
Parameters 52 84 62 94
LR test 304.2359 256.4347 57.4375 -

2
(R) 58.124 18.307 46.194 -
Lag3
Log-L 3164.3341 3222.7817 3328.0644 3372.5083
Parameters 68 116 78 126
LR test 416.3484 299.4532 88.8878 -

2
(R) 76.778 18.307 65.171 -
a
Since Likelihood Ratio statistic values are greater than Chi2 tabulated
values, the null hypothesis of linearity is rejected. MSIAH FAVAR speci-
cation is thus supported to perform better the data.
78
Table 1.11: Lag length test:MSIAH-FAVAR model
AIC HQ SC
Lag = 1 -25.2042 -24.3240 -24.2305
Lag = 2 -25.4534 -25.3941 -24.4375
Lag = 3
a
-25.4649 -24.3485 -23.3458
a
This lag length is supported by only AIC.
Table 1.12: Transition matrix
Regime 1
a
Regime 2
Regime 1 0.9517 0.0483
Regime 2 0.0671 0.9329
a
Note that p
i ,j
= Pr (s
t+1
= j |s
t
= i )
1.8. Conclusion 79
F- Fiscal policy
Figure 1.10: The JGB issuance
1985 1990 1995 2000 2005 2010
14.5
15.0
15.5
JGB issuance in level
1985 1990 1995 2000 2005 2010
-0.01
0.00
0.01
0.02
JGB issuance in variation
Source: Bank of Japan
Note: The monthly data of JGB issuance are seasonally adjusted and
transformed in log.
80
Figure 1.11: Regimes probabilities - MS-FAVAR model
1990 1995 2000 2005
0.25
0.50
0.75
1.00
0.25
0.50
0.75
1.00
Smoothed prob., regime 1
1990 1995 2000 2005
0.25
0.50
0.75
1.00
Smoothed prob,. regime 2
Results from MS-FAVAR model including activity and price factors, the JGB issuance
variable and the monetary base.
1.8. Conclusion 81
Figure 1.12: Response to a monetary base shock in MS-FAVAR
0 1 2 3 4 5
-0.1
0.0
0.1
Regime 1
Fiscal policy
0 1 2 3 4 5
0.0
0.2
0.4
Regime 2
Fiscal policy
0 1 2 3 4 5
0.0
0.5
Activity
0 1 2 3 4 5
-0.2
0.0
0.2 Activity
0 1 2 3 4 5
-0.005
0.000
0.005
0.010 Price
0 1 2 3 4 5
0.00
0.01
0.02
Price
0 1 2 3 4 5
0
1
M0
0 1 2 3 4 5
0
1
2
M0
Responses of the JGB issuance variable (Fiscal policy), the activity factor (Activity) and
the price factor (Price) to an expansionary monetary policy shock increasing the mon-
etary base (MB) by one standard deviation. The impulse reaction period is chosen to
be 5 years. Solid lines show impulse responses, while dotted lines represent condence
intervals using the 10th and 90th percentile values calculated on the basis of 999 boot-
strap replications. The impulse reaction period is chosen to be 5 years. Solid lines show
impulse responses, while dotted lines represent condence intervals using the 10th and
90th percentile values calculated on the basis of 999 bootstrap replications.
82
Figure 1.13: Response to a scal policy shock in MS-FAVAR
0 1 2 3 4 5
0.5
1.0
1.5
Regime 1
Fiscal policy
0 1 2 3 4 5
0.5
1.0
1.5
Regime 2
Fiscal policy
0 1 2 3 4 5
-0.025
0.000
0.025
Activity
0 1 2 3 4 5
-0.025
0.000
0.025
Activity
0 1 2 3 4 5
-0.001
0.000
0.001
Price
0 1 2 3 4 5
-0.0010
-0.0005
0.0000
0.0005
Price
0 1 2 3 4 5
-0.0025
0.0000
0.0025
0.0050
M0
0 1 2 3 4 5
0.000
0.005
M0
Responses of the activity factor (Activity), the price factor (Price), and the monetary base
(M0) to a positive shock to the JGB issuance variable (Fiscal policy) (one standard de-
viation). The impulse reaction period is chosen to be 5 years. Solid lines show impulse
responses, while dotted lines represent condence intervals using the 10th and 90th per-
centile values calculated on the basis of 999 bootstrap replications.
1.8. Conclusion 83
Figure 1.14: Response to a scal policy shock in MS-FAVAR
0 1 2 3 4 5
0
2
4
Regime 1
Fiscal policy
0 1 2 3 4
0.0
0.5
1.0
Regime 2
Fiscal policy
0 1 2 3 4 5
-0.05
0.00
Activity
0 1 2 3 4
-0.025
0.000
0.025
Activity
0 1 2 3 4 5
-0.0025
0.0000
0.0025 Prices
0 1 2 3 4
-0.002
-0.001
0.000
0.001
Prices
0 1 2 3 4 5
-0.05
0.00
Interest rate factor
0 1 2 3 4
-0.025
0.000
0.025
Interest rate factor
Responses of the activity factor (Activity), the price factor (Price), and the interest rate
factor (interest) to a positive shock to the JGB issuance variable (Fiscal policy) (one
standard deviation). The impulse reaction period is chosen to be 5 years. Solid lines
show impulse responses, while dotted lines represent condence intervals using the 10th
and 90th percentile values calculated on the basis of 999 bootstrap replications.
84
G- VECM estimation results
Before estimating a VAR in level, we explored the possibility of using a VECM
27
. We started
by testing for the number of cointegrating relationships in the system and estimating the
long run relations. Juselius (2006) recommends that the stationarity tests on any single time
series should be conducted with a chi-square-distributed likelihood ratio statistic. This should
be done within a modeled system that is restricted for rank. Juselius cautions against using
univariate tests such as the Dickey-Fuller tests, because she argues that the (non)stationarity
of a series is not independent of the rank of the error-correction terms.
Following Nielsen (2004) and Juselius (2006) we rst examined the plotted
28
logged
levels, except for the interest rate, and rst dierence of the data. There is no mean
reversion and the examination of rst-dierenced data suggests that there are a number
of observation-specic non-normal outliers eects. Therefore the specication includes
a linear trend, and a number of various appropriately specied observation-specic dummy
variables to account for outliers. Then the unrestricted VAR in levels, denoting a VAR
model in logged levels, was estimated with a restricted trend and three lags. The standard
misspecication tests showed that the residuals were not well behaved. The multivariate
normality test strongly rejected the null hypothesis of normality.
We followed a procedure for the examination and analysis of potential outliers rec-
ommended by Juselius (2006) and Nielsen (2004). An observation is considered an outlier
if it generates a standardized residual with an absolute value which should be larger than
3.6 given our sample size. Looking at the standardized residuals there are no outliers in the
industrial production variable. Outliers in the consumer price index are present on: 1989:04
(+3.7), 1989:11 (-3,2) and 1997:04 (+6.8). In the monetary base variable the outliers are
on: 1999:12, 2000:01, 2000:02 and 2006:04 . The government bond yields show outliers
27
OxMetrics and the econometrics package CATS in Rats are used the VECM analysis.
28
Plots are not reported here in order to conserve space
1.8. Conclusion 85
on 1998:10 (-3.7), 1998:12 (+4.5), 2003:06 and 2003:08. Outliers which can be explained
by economic events are:
1989:04 : the consumption tax and the consumption Tax Law took eect from 1
April 1989. There is a dramatic change in slope of the price variable starting on this
date. We can consider this outlier as permanent. Specication considerations include
a permanent shift variable for the post 1998:04 part of the sample.
1997:04 : Prime Minister Hashimoto decided to increase the consumption tax from
3 to 5 percent and to put an end to temporary income tax cuts. Specication con-
siderations include an impulse dummy variable to allow for the shock caused by this
intervention.
1998:10 : corresponds to a sharp decline in 10-year yields generated by the Russian
crisis which led to ight to quality and pulled down the term premium. This event
seems to have been temporary and a blip dummy variable included in the short run
deterministic component is specied for this event.
1998:12 : the sharp increase in 10-year yields reects an increase in the public debt;
Moodys reduced Japans debt rating from its highest Aaa to Aa1 on November 17,
1998. This increase can also be explained by the announcement by the Trust Fund
Bureau that it would stop outright purchases of government bonds in December 1998.
In 2001:9 there is a permanent shift asociated with an important decision taken by
the BOJ: a change in the guideline for monetary market operations; CAB rose from
5 to 6 trillion yen. At the same date there was an increase in outright purchases of
long-term government bonds, from 400 billion to 600 billion yen per month.
Outliers which are present on 1999:12, 2001:01 and 2000:2 correspond to the pro-
vision of extra liquidity by the BOJ to deal with the potential Year 2000-related
86
problems. As such outliers are of opposite signs: they are considered as the transitory
eects of a shock (+4, +3 and 7).
2006:04 corresponds to the end of QEMP (the eect seems to start here rather than
in 2006:03).
Permanent shift dummies were restricted to the cointegrating space to allow for the
possibility that the events may have had a permanent eect. However, outliers which do
not correspond to an economic event and which seem to be due to transitory eects of
shocks or simple mistakes, are likely to be additive and should therefore not be modeled.
Following Nielsen (2004) we chose to leave the additive outliers in the data set. The
inclusion of the shift, blip and transitory dummies in an unrestricted VECM does improve
the misspecication tests signicantly but the multivariate normality test still rejects the
null hypothesis of normality. This is caused by the non-normality of the residuals of the
monetary-base equation in spite of major attempts to improve the specication. In the next
step, to calculate the rank test statistics we used a simulated Bartlett test. This is because
the 95 percent fractile values are adjusted for the restriction of permanent shift dummies
included in the cointegration space (Juselius (2006)). The null hypothesis corrected for the
shift dummies suggest that the null hypothesis of at least r =0 is accepted. Hence the rank
test statistics suggest that the system has no cointegrating relation and the four variables
do not share common trends. The same analysis was conducted for cumulated factors.
Since the factors are estimated from dierent subsets of variables they are not orthogonal
to each other and can be cointegrated. In spite of the ability of factor analysis to eliminate
idiosyncratic shocks and therefore outliers from simple variables we cannot detect any long
term relationship between the factors and the monetary base.
2
2
Quantitative Easing under Scrutiny: A
Time-Varying Parameter Factor-Augmented VAR
Model
2.1 Introduction
The eectiveness of the quantitative easing monetary policy (QEMP) remains a much de-
bated issue. Since this strategy is adopted by most major central banks, namely the Fed, the
Bank of England and the European Central Bank, it is crucial to know whether this strategy
can be used as an active tool to stimulate prices and foster growth, and, if so, through
which transmission channel it works. The problem of quantifying the empirical relevance of
the dierent channels of transmissions through which QEMP exerts its inuence on output
and prices has received wide and increasing attention in recent years. A growing body of
empirical macroeconomic literature using VAR methodology has tried to gauge the eects
87
88
of the Japanese monetary policy either in the very low interest period from 1995 or more
specically for the QEMP period. This Japanese use of QEMP, the only experience we can
learn from, still requires exploration.
Earlier VAR studies have often been concerned with measuring monetary policy and
its macroeconomic eects. See e.g.Christiano et al. (1999), Leeper et al. (1996), and
Bernanke and Mihov (1998) for studies of the U.S., and Teruyama (2001) for the research on
the Japanese monetary policy transmission mechanisms. Moreover, many researchers have
investigated possible structural breaks which can characterize the monetary transmission
mechanisms. More particularly, in the study of Japanese monetary policy all empirical studies
are fairly consensual on the fact that examining the impact of such a policy should take into
account the instability in the transmission mechanism. Structural breaks have been treated
either exogenously, by including dummy variables or by using subsample analysis (e.g Miyao
(2000)), or endogenously, by using Markov Switching VAR (MS-VAR) (e.g. Fujiwara (2006),
Inoue and Okimoto (2008) and chapter 1 above) or Time-Varying-Parameters VAR (TVP-
VAR) model (e. g. Kimura et al. (2003), Nakajima et al. (2009a)).
Miyao (2000) estimates a recursive VAR model and concludes, by using
2
testing
procedure, that the eect of the monetary policy weakens from 1990 onwards. On the
other hand, Kimura et al. (2003) employ a time-varying VAR model for the period between
1971-2002 and detect a structural change point in 1985 after which the ination rate is less
responsive to an expansion in the monetary base. More recently, Fujiwara (2006), Inoue and
Okimoto (2008) and Mehrotra (2009) estimate an MS-VAR model where the regime states
are considered as stochastic events. All the parameters of the models are stochastic and
switch according to a hidden Markov chain. Both Fujiwara (2006) and Inoue and Okimoto
(2008) conclude that the monetary policy is eective until around 1995-1996, when the
call rate approaches the zero boundary and subsequently weakens. In addition, the period
between 1995 and 1996 is considered as a transition period. The only work that covers
2.1. Introduction 89
the total period of QEMP is that of Nakajima et al. (2009a). To estimate the TVP-VAR
they use quarterly data, namely the call rate, industrial production, the consumer price index
and the monetary base, for the period between 1981 and 2008. Despite the existence of
puzzles, their ndings conrm to a certain extent those of Fujiwara (2006) and Inoue and
Okimoto (2008) and show a change in the eect of monetary policy on activity and prices
when interest rates become very low.
Usually, the overall eects of QEMP are examined for a single channel or a subset
of channels
1
; typically, one or a subset of the following channels are considered: portfolio-
rebalancing channel; signaling eect; policy-duration eect and also exchange rate channel.
All empirical studies are relatively consensual on the fact that the portfolio-rebalancing chan-
nel does not work. Empirical studies dealing with the eectiveness of such a transmission
channel, for instance Oda and Ueda (2007) and Kimura et al. (2003), show that the eect
of a portfolio-rebalancing channel is insignicant or too small considering the huge amount
of current account balances (CABs) expansion and the Japanese Government Bond (JGB)
purchased by the Bank of Japan (BOJ). Referring to the signaling eect, Oda and Ueda
(2007) detect a signicant eect of this channel from the increase in CABs but no ef-
fect from the increase in long-term JGB purchases. The empirical studies dealing with the
policy-duration eect nd that it signicantly lowers long-term interest rates. Among these
studies we can quote Baba et al. (2005), Oda and Ueda (2007), Okina and Shiratsuka
(2004a) and more recently Nakajima et al. (2010). The later work uses a TVP-VAR model
and shows that the signicant eect of the policy-duration on the yield curve and market
expectations is not transmitted to the real economy. On the other hand, Svensson (2003)
oers what he calls a foolproof way of escaping from a liquidity trap. The author mostly
focuses on alternative policies in a liquidity trap to aect private-sector expectations of the
future price level via the exchange rate channel. However, Ito and Mishkin (2006) and Ito
1
For more detail about the transmission channels suggested by the QEMP the reader is referred
to the paper of Ugai (2007).
90
and Yabu (2007) argue that this channel can work if the BOJ neither sterilizes the interven-
tion in the foreign exchange market ordained by the Ministry of Finance, nor announces an
exchange rate target, sending a signal that the main objective remains the price level. On
the other hand, Girardin and Lyons (2008) show some eects of this channel even though
the BOJ/MOF intervention is technically fully sterilized.
All these empirical works use models with a small number of variables either to
examine the existence of structural change or to quantify the possible transmission channels
of the QEMP. However, for the reasons explained in Bernanke et al. (2005) and Stock and
Watson (2005), using limited information can lead to a biased policy shock measurement.
In other words, when information related to the central bank and the private sector is
omitted, the measurement of the unsystematic part of monetary policy may be incorrect.
This problem can be illustrated by the puzzles that characterize VAR results as obtained
in most of the papers cited above. Moreover, the limited information means that transition
channels are examined separately, and hence the possible interaction between channels is
not considered. Of course, the challenge in assessing the strength of any particular channel
of monetary transmission comes from the concurrent operation of multiple channels. For
example, it is hard to tell how much of the long-term interest rate decline to attribute to
a decline in stock prices (portfolio-rebalancing channel) and how much to the reduction
in private sector expectations about the path of future short-term interest rates (policy-
duration eect). However, a complete model in which a maximum of information will be
taken into account will allow us to capture most of the structure underlying the economy
and will reliably reveal what are the mechanisms through which the QEMP could aect the
economy.
In this chapter, following Bernanke et al. (2005) and Stock and Watson (2005) we
use the factor augmented VAR (FAVAR) model in order to complement the empirical works
on Japanese monetary policy cited above, specically with introducing further variables to
2.1. Introduction 91
the VAR data set. To our Knowledge, only one study so far has been conducted on the
Japanese economy using the FAVAR model. Shibamoto (2007) was the rst to employ
a FAVAR model on Japanese data. However, since he uses data from January 1985 to
March 2001, he does not examine the QEMP period. In addition, his results should be
interpreted with great care since, as mentioned above, examining Japanese monetary policy
without taking into account structural breaks could be misleading. In the previous chapter we
combine MS-VAR methodology and factor analysis in what we call MS-FAVAR to examine
Japanese monetary policy. The MS methodology allows us to detect discrete jumps for all
parameters simultaneously; it permits us to date breaks and assess whether a new regime
appears. Our ndings on regime change timing are similar to those of Fujiwara (2006) and
of Inoue and Okimoto (2008) ; the second regime corresponds to the adoption of the Zero
Interest Rate Policy (ZIRP) and QEMP. In this chapter our objective is twofold. First, we
use TVP-VAR methodology to allow for more exible and independent variation in FAVAR
parameters and to detect permanent and even gradual variations. Given the conrmation of
regime changes in chapter 1, to go one step further, TVP-VAR methodology allows us to
examine the evolution of Japanese monetary policy at each point in time, more particularly
inside the second regime detected in chapter 1. Therefore, we will be able to focus precisely
on the QEMP period and more reliably examine the eectiveness of this strategy. Second,
it is true that the MS-FAVAR allows us to derive impulse responses for structural factors,
since they are identied, representing clear economic concepts namely, activity , prices and
interest rates. However, we cannot examine the dynamics of all the variables explained by
the factors. Therefore, we employ here the Bayesian Markov chain Monte Carlo approach
(MCMC) to the estimation of time-varying parameters in the FAVAR model (TVP-FAVAR),
developed by Koop and Korobilis (2009). With these motivations and considerations in mind,
we aim to use this complete model in order to endogenously treat the possible structural
changes in the Japanese economy and provide a more complete and detailed analysis on
92
how monetary policy shocks in Japan aect a large range of macroeconomic time series.
After analyzing a period ranging between 1978:1 and 2008:4 we obtain four main
results. First, the best model to specify the monetary policy during the last two decades
is a model where all of parameters vary over time. This corroborates our choice of a time
varying parameters model. Second, the eect of QEMP on activity and prices is stronger
than previously found. In particular, we nd a signicant price reaction to a monetary
policy shock. Third, in contrast with previous work, there is a detectable eectiveness of
the portfolio-rebalancing channel, which could have a role in transmitting the monetary
policy shocks. Finally, even though the eect on expectation channel is short-lived, the
policy commitment might prevent a downward spiral of expectations but were not able to
generate an inationary pressure to escape from the deationary spiral and to revive the
economie.
The remainder of this chapter proceeds as follows. In section two the TVP-FAVAR
model is described. Section 3 contains the data description, specication tests and results.
Section 4 concludes.
2.2 Methodology
In the previous chapter we combined MS-VAR methodology and factor analysis in MS-FAVAR
to examine the Japanese monetary policy. MS model allows for state shifts in the FAVAR
parameters only when they are signicant and permits detecting simultaneous discrete jumps
for all parameters. This model not only enabled us to know whether a signicant new
monetary policy regime appeared, but also permitted to date regime changes. A second
regime appeared in February 1999, covering both ZIRP and QEMP periods. The objective
of this chapter is to complement the analysis in chapter 1 by using TVP-FAVAR model,
allowing state shifts in the FAVAR parameters at the dierent point of the sample and not
2.2. Methodology 93
for subsamples. By doing this, we will be able to analyse the Japanese monetary policy at
each time in the sample and especially QEMP period.
2.2.1 TVP-FAVAR model
Following Koop and Korobilis (2009), this subsection shows the econometric framework of
the TVP-FAVAR. This model is a generalization of the FAVAR model developed by Bernanke
et al. (2005) and Stock and Watson (2005). Factor dynamics are given by the following
time varying parameters FAVAR:
Y
t
=
t
+
P

p=1

t,p
Y
tp
+
t
(2.1)
where Y
t
= [F
t
R
t
]

. This means that along with the unobserved factors, Y

t
contains an
observable factor R
t
of dimension (x1), which represents the monetary policy instrument.
The ((K +)x1) vector of error terms
t
is mean 0 with covariance matrix
t
of dimension
((K +)x(K +)). However, Equation 2.1 cannot be estimated directly because the factors
are unobserved. We need, therefore, as a rst step, to estimate factors using a singular
value decomposition of data. Factors, becoming observable, are included in a second step in
the equation. We assume that the X
t
is (Nx1) economic variable vector can be decomposed
into a (Kx1) unobservable factor vector F
t
. The unobservable factors are reected in a wide
range of economic variables. We can think of unobservable factors in terms of concepts such
as economic activity or price pressures. Assume that X
t
are related to the unobservable
factors F
t
and the observable factors R
t
with drifting parameters, as follows :
X
t
=
f
t
F
t
+
R
t
R
t
+e
t
(2.2)
where e
t
are errors with mean zero and variance-covariance matrix =diag(exp(
1,t
), , exp(
n,t
)).
The term error e
t
are assumed to be either weakly correlated or uncorrelated; these can be
94
interpreted as the idiosyncratic components.
f
and
R
are the (NxK), (Nx) matrices of
factor loadings. The implication of the diagonality of the covariance matrix is that the pa-
rameters in equation (2.2) can be estimated equation-by-equation. This approach is needed
for reasons that will be explained below.
A Choleski decomposition of the reduced form covariance matrix
t
can be used to
orthogonalize the reduced form innovations and to identify the structural model:

t
= A
1
t
H
t
_
A
1
t
_

(2.3)
The time-varying matrices H
t
and A
t
are dened as follows:
H
t

_

_
h
1,t
0 0
0 h
2,t
0
.
.
.
.
.
.
.
.
.
0 0 h
(K+),t
_

_
(2.4)
A
t

_

_
1 0 0
a
21,t
1
.
.
.
.
.
.
.
.
.
.
.
.
0
a
(K+)1,t
.
.
.
a
(K+)k,t
1
_

_
(2.5)
As suggested by Primiceri (2005) and Koop and Korobilis (2009) we assume that all
the parameters evolve as random walks
2
augmented with the mixture innovation specication
of Giordani and Kohn (2008). Therefore, the innovations of the random walk evolution of
2
As explained in Primiceri (2005) the random walk assumption has the advantages of focusing
on permanent shifts and reducing the number of parameters in the estimation procedure. However,
a random walk model is non-stationary and it is obviously "more explosive" than the number of
observation increases. By choosing quarterly data for the period between 1978 Q1 and 2008 Q4 our
sample contains no more than 120 time series observations. Using such a short period alleviates this
problem.
2.2. Methodology 95
the parameters is dened as a mixture of two normal components (see koop et al 2009 and
Koop and Korobilis (2009)):
_

t
=
t1
+J

i ,t

i ,t
=
i ,t1
+J

i ,t

t
=
t1
+J

i ,t

t
a
t
= a
t1
+J
a
i ,t

a
t
lnh
i ,t
= lnh
i ,t1
+J
h
i ,t

h
t
(2.6)
where = [
t

t,p
] and h
i ,t
evolve as geometric random walks and we assume that the
innovation vectors are independent from each other and are distributed as
_

a
t

h
t
_

_
N(0, Q), where Q =
_

_
Q

t
0 0
0 Q

t
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Q

t
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Q

a
t
0
0 0 Q

h
t
_

_
(2.7)
The error terms in equation (3.12) are allowed, to some extent, to be mutually correlated.
However, we assume for parsimony that all error components in equations (1.1)-(1.8) are
uncorrelated with each other.
Note that the monetary policy variables are ordered last in the FAVAR (equation
(2.1)). Then by imposing some normalization as in (3.13) the unobservable factors do
not respond to the monetary policy shocks contemporaneously, and the innovations in the
equations of R
t
are treated as the monetary policy shocks.
Suppose that J
t
are binary random variables that control structural breaks in the
respective error term of the time varying parameters. As in Koop and Korobilis (2009) we
96
assume that J
t
Bernoulli (), where is the probability
3
corresponding to each of the
parameter vectors , , , a and lnh. Therefore, if J
t
= 0 or J
t
= 1 that means that
the data indicated constant and time varying parameters specications, respectively, for all
(t = 1, ..., T). Otherwise, data can also determine a time varying parameters specication
for some subsamples only; J
t
=1 for some t. The choice of either specication is motivated
by the Bayesian procedure selection model based on marginal likelihoods. Following Koop
and Korobilis (2009), we choose the more exible model allowing J

t
to be dierent for
each row of in equation (2.2) such that J

it
= J

jt
. This is the reason why equation (2.2)
is estimated equation-by-equation. We assume also that hyperparameters Q

a
t
are block
diagonal in which each block corresponds to parameters belonging to separate equations
4
.
A particular advantage of the factor-augmented framework is that we can derive
impulse responses not only for the fundamental factors, but also for all the variables included
in the factors. We provide impulse responses to a monetary policy shock for some of the
most interesting variables. Equation (2.1) can be written as

(L)

Y
t
=
t
(2.8)
where L is a lag operator of order p,

the coecient matrix including ,

Y
t
=
_

F
t
R
t

and
t
is a ((K +)x1) vector of structural innovations. As the estimator of X
t
using (2.2)
is

X
t
=
f
t

F
t
+
R
t

R
t
, impulse response functions of

X
t
are obtained as follows:

X
t
=
_

f
t

R
t

F
t
R
t
_

_ =
_

f
t

R
t

(L)
t
(2.9)
where

(L) =
_

(L)
_
1
.
3
Also we assume that () is distributed as a Beta(
0
,
1
) and all probabilities have the same prior
values (
0
=
1
) and they are common for all parameters.
4
We have then (K + ) 1 blocks, namely a
block1
= {a
21,t
}, a
block1
= {a
31,t
, a
32,t
}, ...,
a
block((K+)1)
=
_
a
(K+)1,t
, ..., a
(K+)k,t
_
2.2. Methodology 97
2.2.2 Estimation
This section gives an overview of the estimation strategy and the algorithm used
in estimation. The Bayesian methods described by Kim and Roubini (2000) is used to
estimate the model in equations (2.1)-(3.12) for two reasons. First, if the variance of the
time varying coecients is small, then the maximum likelihood estimator (MLE) is biased
towards a constant coecients FAVAR. As a consequence, numerical optimization methods
are very likely to get stuck in uninteresting regions of the likelihood (Stock and Watson
(1996)). Second, multiple peaks are highly probable in a non-linear FAVAR model with
highly dimensional parameters. This makes maximum likelihood estimation quite unreliable
if in fact a peak is reached at all. Therefore, the Gibbs sampler is appropriate to deal with the
problem of estimating a highly dimensional parameter model, by allowing to divide the task
in smaller and simpler ones. In addition, given that Gibbs sampler is a stochastic algorithm,
it is more likely to escape local maxima.
Before summarizing the basic algorithm we need to clarify the choice of the factor
estimation method. If factors form a part of the unknown parameters of the TVP-FAVAR
model we need additional restrictions to identify it. Nonetheless, factors cannot be directly
identied since we cannot attribute a clear economic interpretation to them. On the other
hand, the main advantage of the static representation of the dynamic factor model, described
by equation 2.2, is that the factors can be estimated by the principal component method.
However, as discussed by Belviso and Milani (2006), the factors estimated by principal
component have unknown dynamic properties because principal components neither exploit
the factor nor the idiosyncratic component dynamics. There are two principal approaches
that exploit these features to extract the static factors through principal components. The
rst is the tow-step approach situated in the frequency domain proposed by Forni et al.
(2005) and employed in the chapter 1. The second approach is a two-step strategy in the
parametric time domain introduced by Stock and Watson (2005). Therefore, we use Forni
98
et al. (2005)s
5
method to estimate the space spanned by the factors
6
. In order to choose
the appropriate number of estimated factors, we consider the sensitivity of the results to
the inclusion of a dierent number of factors. As explained in Bernanke et al. (2005), this
ad hoc way is justied by the fact that the statistical identication determines the number
of factors present in the data set but it does not determine the number of factors to use in
the model.
2.2.2.1 Prior distribution and starting values
In the choice of prior distribution of unknown parameters, we follow the specications of
Primiceri (2005) and Koop and Korobilis (2009). Following the Bayesian literature, , H
t
and A
t
will be called parameters and the covariance matrices of the innovations, i.e. the
elements of Q, and the break probabilities hyperparameters.
All the hyperparameters Q

except Q

t
are assumed to be distributed as independent
inverse-Wishart random matrices. The Wishart distribution can be thought of as the multi-
variate analog of -square, and used to impose positive deniteness of the blocks of Q
/
.
Finally, the diagonal elements
i
of Q
0

have univariate inverse Gamma distributions as

each
i
is a scalar.
Q
0

IW(l

.(1+m

).V
OLS

, 1+m

).
Q
0

IG(l

.(1+m

).V
OLS

, 1+m

)
where V
OLS

denotes the variance of the OLS estimate of and l

are tuning constants. In

our case we do not use a training sample
7
to estimate V
OLS
h
as in Primiceri (2005), hence
V
OLS
h
and V
OLS

are assumed to be null matrices of dimension (m

) and (m

),
5
For details of the dynamic factor model the reader is referred to Forni et al. (2005).
6
This method is, in addition, appropriate for samples with relatively small numbers of time ob-
servations. The choice of this method is therefore particularly appropriate since we use a quarterly
data sample with no more than 150 observations.
7
In this paper we do not use informative priors from training sample because our sample is already
relatively short and we are not prepared to sacrice observations.
2.2. Methodology 99
respectively; m is the number of elements in the state vectors. IW(Sc, df ) and IG(Sc, df )
represent respectively the inverse-Wishart and the inverse-Gamma with scale matrix Sc and
degrees of freedom df . As in Primiceri (2005), l

and l

are assumed to be equal to 0.07.

For all the parameters governing the structural break probabilities we assume that (
0
)
Beta(0.5, 0.5), which indicates that there is a 50%
8
chance of a break occurring in any time
period. Using uninformative priors we do not impose any constraint on the number of breaks
and we let the data speak for themselves.
The priors for the initial states of the regression coecients, the covariances and
volatilities are assumed to be normally distributed, independent of each other and of the
hyperparameters. Let
0
= [
0

i ,0

0
a
0
lnh
i ,0
]

N(0, 4I ), where I is the identity

matrix with dimensions of each respective parameter and 0 is a vector of 0s. The choice
of zero mean reects a prior belief that our variables will show little persistence since they
are used in rst dierence and are stationary. The variance scaling factor 4 is arbitrary but
large relative to the mean 0.
2.2.2.2 Simulation method
Conditional on using the conjugate priors and a Kalman lter, the Gibbs sampler is repeated
until convergence to the true posterior densities of the parameters. Note that at time t =1
we do not need to choose an initial value of J

1
since whether we assume all parameters
are constant (J

1
= 0) or all are varying (J

1
= 1) does not aect the posterior results.
The states in J

t
are updated in the subsequent periods. Let a superscript T denote the
complete history of the data (e.g.
T
=

1
, ... ,

T
). We summarize the applied Gibbs
sampler involving the following steps:
1. Initialize the parameters (
0
) and the estimated factors.
8
E() =

0

1
+
1
.
100
2. Draw
T
from p(
T
|Y
T
,
0
) using Carter and Kohn (1994)s algorithm, except for
h and which are simulated using Kim et al. (1998)(1998)s algorithm.
3. Draw hyperparameters Q
T

using the inverse gamma distribution and the remaining

Q
T

hyperparameters are drawn from an inverse Wishart distribution.

4. Simulated the binary random variables J

using the Gerlach et al. (2000) algorithm.

5. Simulate

(
0
,
1
), where
0
=
0
+
T
t=1
J

t
and
1
=
1
+T
T
t=1
J

t
.
6. Go to step 2
9
Conditional on initial values for the parameters (
0
), except for
i ,0
and lnh
i ,0
,
the estimated factors and the data Y
T
, the state-space form given by (2.1) and (2.2) is
linear and Gaussian. Therefore, the conditional posterior of
T
is a product of Gaussian
densities and
T
can be drawn using a forward-backward sampling algorithm from Carter
and Kohn (1994). Our objective is to characterize the marginal posterior densities of
T
.
To obtain an empirical approximation to this density, the Gibbs sampler simulates
T
from
the conditional density p(
T
|Y
T
,
0
, F
T
). This consists rst, in updating the parameters
at time t conditional on data at time t (from t = 1 to T, each
t
is consecutively updated
conditional on data at time t). Then, the Kalman lter produces a trajectory of parameters
by again updating the estimated
t
using information in the subsequent periods (t +1).
Finally, from the terminal state
T
, a backward recursion produces the required smoothed
draws by updating
t
conditional on information in previous periods from t = T 1 up to
t = 1, using the information from the whole sample.
However, drawing from the conditional posterior of
i ,0
and lnh
i ,0
is dierent because
the conditional state-space presentation for
i ,0
and lnh
i ,0
is non-normal. A Gibbs sampling
technique that extends the usual Gaussian Kalman lter, developed by Kim et al. (1998),
9
Note that only factor loadings are considered as time-varying parameters. For this reason we
do not need to go back to step 1 in the algorithm. As explained above, factors are considered as
known parameters in the absence of theoretical justication of additional identication.
2.2. Methodology 101
consists of transforming the non-Gaussian state-space form into an approximately Gaussian
one, so that the Carter-Kohn standard simulation smoother can be employed.
In this second step, drawing parameters proceeds as follows. First, factor loadings
(
T
) are simulated conditional on prior distributions of estimated factors and data X
T
(p(
T
|X
T
, F
T
)). Second, conditional on the sampled values of
T
, a set of values of
T
are drawn from the conditional distribution p(
T
|X
T
, F
T
,
T
). Third, coecients (
T
)
are simulated from the conditional density p(
T
|Y
T
,
0
, a
0
, lnh
0
). Fourth, the elements of
A
t
are drawn from p(A
t
|Y
T
,
T
, a
0
, lnh
0
). Finally, the diagonal elements of H
t
are drawn
from p(A
t
|Y
T
,
T
, a
T
, lnh
0
).
In step 3, conditional on Y
T
, estimated factor and
T
, drawing from the conditional
posterior of the hyperparameters Q
T
/
is standard, since it is a product of independent
inverse-Wishart distributions. However, since we have constrained the hyperparameter ma-
trix Q
T

to be diagonal, its diagonal elements Q

T

i
have univariate inverse-Gamma distribu-
tions. For the structural break probability parameters, the independent sequence of Bernoulli
variable J

is simulated non-conditional on data using Gerlach et al. (2000) algorithm

10
.
Finally, in step 5 the conditional posterior for the break probabilities is sampled from Beta
distributions.
Given these marginal posterior densities, estimates of parameters and hyperparam-
eters can be obtained as the medians or means of these densities. The algorithm uses 60
000 sampling replications and discards the initial 40 000 as burn-in. When the posterior
moments vary little over retained draws, this means that the Gibbs sampler does converge
to the true posterior densities of the parameters.
10
The algorithm proposed by Carter and Kohn (1994) draws J conditional on states Y
T
, but in
the presence of structural breaks or additive outliers J and Y
T
become highly correlated, making
this sampler very inecient. The Gerlach et al. (2000) algorithm retains a high degree of eciency
regardless of correlation between J and Y (Giordani and Kohn (2008)).
102
2.3 Empirical results
2.3.1 Data and preliminary results
In our application of the TVP-FAVAR methodology, the set of information variables is of
a balanced panel of 139 macroeconomic time series for Japan. The data are at quarterly
frequency and span the period from 1983:Q2 through 2008:Q4. The data set consists of
variables related to the real activity, consumer and producer price indexes, nancial markets,
private and business anticipations and interest rates. As in Bernanke et al. (2005) our
data are classied into two categories of variables: we distinguish between slow-moving
variables which are predetermined in the current period and fast moving variables which
react contemporaneously to the economic news or shocks. The series have been demeaned
and standardized and seasonally adjusted when it is necessary and, as usual, the series are
initially transformed to induce stationarity. Our data set with the complete list of variables,
its sources and the relevant transformations applied, is presented in Table 1 in Appendix A.
As for the choice of monetary policy instrument for Japan, indicators vary from
study to study. As discussed in Inoue and Okimoto (2008), this choice is between the call
rate (Miyao (2000) and Nakajima et al. (2009a))
11
and the monetary base (Shioji (2000)).
Inoue and Okimoto (2008) argue that the best choice is jointly considering the call rate and
the monetary base as policy indicators. This is because from 1995 onwards and particularly
from the introduction of QEMP in March 2001 to March 2006, interest rates were almost
zero and the monetary policy target was explicitly the monetary base. However, Inoue and
Okimoto (2008) nally consider only data spanning the period between January 1975 and
December 2002. This is because from October 2002 onwards the call rate was zero, in
which case the normality assumption is invalidated. Here, since our objective is to focus
on the QEMP period and for the reasons given in Inoue and Okimoto (2008) we assume
11
Note that all of these studies use data from 1975 and 1977 to 1995 and 1998 and hence the
period of zero interest rate policy and QEMP are excluded.
2.3. Empirical results 103
that the monetary base is the only observable factor and then the only monetary policy
instrument.
In the rst step, we need to determine the number of factors that characterize our
data set. Our results are not materially aected whether we choose three or four factors.
Bernanke et al. (2005) and Stock and Watson (2005) argue that three factors perform well
and since parsimonious modeling is always preferred, in our case we will also assume that
the data set can be described by three factors.
2.3.2 Specication tests
To carry out subsequent model selection, we opted for the Deviance Information Criterion
(DIC) statistic (Spiegelhalter et al. (2002)). The problem with the TVP-VARs is that it is
not easy to use the marginal likelihood, which is a typical measure for the Bayesian model, as
we have stochastic volatility which makes likelihood evaluations dicult and cumbersome.
The problem becomes more severe for the TVP-FAVAR model which has an additional
equation. The DIC takes into account two important features of the model: the complexity
(based on the number of the parameters) and the t (typically measured by a deviance
statistic). DIC examines the two features together and gives a measure which balances
between the two. Table 2.1 shows the values of DIC estimated on 20,000 posterior means
draws for 5 dierent models with 3 factors and 2 lags: (i) a model with constant param-
eter (FAVAR), (ii) a model with only varying factor loadings (TVPL), (iii) a model with
varying factor loadings and auto-regressive terms (TVPLB), (iv) a model in which factor
loadings, auto-regression terms and covariance elements are assumed to vary (TVPLBA),
(v) a model where factor loadings, auto-regression terms and Log volatilities are assumed to
vary (TVPLBS) and (vi) a model in which all of the parameters are assumed to vary (TV-
PLBAS). Except FAVAR model all the other models are estimated for two kinds of priors:
uninformative priors (Beta(0.5, 0.5)) and tightened priors (Beta(0.01, 10)) for the transition
104
probabilities. With the latter priors we constrain the model to have few breaks (one or two
breaks) while with the uninformative priors the number of breakpoints is determined by the
data. Not surprisingly, the FAVAR model shows the highest DIC value, indicating that we
Table 2.1: Model comparison with Deviance Information Criterion (DIC)
FAVAR TVPL(2) TVPLB(2) TVPLBA(2) TVPLBS(2) TVPLBAS(2)
- 10421.3
Few breaks - 10528.3 - 10530.0 -10531.7 -10610.9 -10651.3
a
uninformative -10529.1 -10530.4 10543.0 -10607.2 -10654.1
a
Results are based on 60,000 iterations after a burn-in period of 40,000. The model
with smaller DIC would better predict a replicate datasets of the same structure.
need to take into account breaks in the model. All the other models perform clearly better,
corroborating the validity of a TVP approach. Then we test whether all parameters or few
of them vary over time. The resulting DIC of the unconstrained model (TVPLBAS-FAVAR)
is the lowest, hence all parameters do change over time. Next, we test whether the Japanese
economy is characterized by only a small number of breaks (e.g., among others, Fujiwara
(2006), Inoue and Okimoto (2008) and chapter 1 above). The comparison between models
with uninformative and informative priors tend to conrm the existence of more than two
breaks in the data (Nakajima et al. (2009a)). Even with informative priors results still indi-
cate a gradual evolution of the parameters. These outcomes tend to conrm our choice of
uninformative priors where the number of breakpoints is determined in a data based fashion.
2.3.3 The evolution of the Japanese monetary policy
Before examining the eectiveness of QEMP and its transmission channels, we need to
analyse the evolution of the Japanese monetary policy during the last three decades. In
Figure 2.1 we present the time-varying standard deviations of the errors in the equations
for the three factors, ination, activity and the monetary base (i.e. the posterior means of
the square roots of the diagonal element of
t
). Figure 2.1 shows that there is evidence of
2.3. Empirical results 105
time variation in error variances in all equations.
Figure 2.1: Posterior mean of the standard deviation of equation residuals

The gures show the time-varying standard deviations of the errors in the equations for the
three factors, ination, activity and the monetary base.
The sharp increase in 1989 and in 1997 can be explained by the introduction of the
consumption tax (the consumption Tax Law took eect from 1 April 1989) and its increase
from 3 to 5 percent in April 1997. However, after early 1998 and until 2005 the volatility
is greatly reduced, reecting the deationary period experienced by the Japanese economy.
The volatility of GDP keeps increasing from the mid-1990s until 2001. This conrms the
106
ndings of Nakajima et al. (2009a) that the variance of real GDP becomes higher in the
1990s than it was in the 1980s. One possible explanation is the increased uncertainty
that characterized the period after the burst of the asset price bubble and inuenced the
investment. We particularly note the sharp decline in GDP volatility after the implementation
of the QEMP. In a similar way, we can think that during the QEMP period monetary
policy was more widely understood, and reducing the volatility of investment, reinforced the
perception that the business cycle had become less severe. Finally, the increase in monetary
base volatility from the end of 1995 corresponds to the decrease in the call rate to a lower
level in 1995 (0.5 %) and to nearly zero under the zero interest rate policy and QEMP.
2.3.4 Impulse response analysis
This section examines the dynamic relationships between variables through impulse response
functions which can be implemented for all series included in our database. We conduct
our analysis for three periods and dates are chosen in ad-hoc way: 1989 Q4, 1995 Q1 and
2002 Q1. The rst date corresponds to the burst of the asset price bubble, the second
date represents the end of the use of the call rate as a monetary policy instrument and the
last date represents the beginning of the QEMP and the period when short-term interest
rate reached zero. The shock is normalized so that it increases the monetary base by its
standard deviation at all dates.
2.3.4.1 Was the QEMP eective?
Figure 2.2 displays impulse responses of key variables in the model to a monetary policy
shock over dierent dates chosen arbitrary: (i) 1989 Q1, before the burst of the asset
bubble and when interest rates were high, (ii) 1996 Q1, after the decline in the short term
interest rates to 0.5% and (iii) 2002 Q1, over the QEMP period. The posterior median is
the solid line and the broken lines are the 10th and 90th percentiles.
2.3. Empirical results 107
Figure 2.2: Impulse response functions
3 6 9 12 15 18 21
-0.5
0
0.5
1
1.5
M0, 89-Q1
3 6 9 12 15 18 21
-0.1
0
0.1
0.2
CPI Inflation, 2002-Q1
3 6 9 12 15 18 21
-0.1
0
0.1
0.2
CPI Inflation, 95-Q1
3 6 9 12 15 18 21
-0.1
0
0.1
0.2
Output, 89-Q1
3 6 9 12 15 18 21
-0.5
0
0.5
1
1.5
M0, 95-Q1
3 6 9 12 15 18 21
-0.1
0
0.1
0.2
CPI Inflation, 89-Q1
3 6 9 12 15 18 21
-0.1
0
0.1
0.2
Output, 2002-Q1
3 6 9 12 15 18 21
-0.1
0
0.1
0.2
Output, 95-Q1
3 6 9 12 15 18 21
-0.5
0
0.5
1
1.5
M0, 2002-Q1
The gures show the reactions of ination and GDP to a shock to M0 over 21 quarters for three
dierent dates . The solid lines show the impulse responses implied by the time-varying FAVAR
(posterior median) and dashed lines represent the 10
th
and 90
th
percentiles.
It is not surprising that the eect of the monetary base shock on ination and
GDP until 1995 is very weak and insignicant, indicating that monetary policy has been
considered as interest rate policy. However, from the second half of 1995 (second row) the
eect of the monetary base shock becomes positive but hardly signicant. These results
are consistent with the evolution of the monetary base stochastic volatility from the end of
1995. During this period the interest rates fell to 0.5 percent and then declined further to
almost zero percent during the ZIRP period. It is then plausible to think that interest rates
108
being extremely low, the monetary base began to be used as an alternative policy instrument.
Interestingly, and in contrast with Fujiwara (2006) and Inoue and Okimoto (2008), during
the QEMP period (third row) ination displays a positive and signicant response, which
becomes statistically insignicant only after 3 quarters. This eect, though it is short-
lived, shows that the QEMP has an inationary eect. The eect of the monetary base
shock on GDP is more pronounced. Production displays a temporary and not persistent
positive response, which veers to be insignicant after one year. This positive eect on
activity is unanimously detected in empirical studies. This temporary impact put together
with the decline in the output volatility leads us to think that monetary policy might be
the source of output uctuations during the QEMP period. Note that the disconnection
between traditional VAR results and the standard theory predictions, that is revealed by
puzzles, price divergence and non-neutrality of money arising in Fujiwara (2006), Inoue and
Okimoto (2008) and Nakajima et al. (2009a), disappears under our rich-data model. As
shown in Bernanke et al. (2005) and Forni and Gambetti (2010), our results corroborate
the idea that a FAVAR methodology, which exploits a large set of information, improves the
accuracy of econometric models in predicting the eects of monetary policy, and, therefore,
could address puzzling eects observed otherwise.
In order to go further in our analysis we exploit the advantage of using TVP-FAVAR,
allowing us to observe the impulse responses to shocks for all the economic series included
in the construction of the factors. In doing so, we are able to detect the origin of the
QEMP eect. Figure 2.5 (in Appendix B) displays the reaction of disaggregated prices.
Except for two producer price indexes the reaction of the remaining prices is signicantly
positive
12
. These results are in line with theory and are opposite to the so-called price
puzzle observed by Nakajima et al. (2009a) and Inoue and Okimoto (2008). An interesting
result that emerges from Figure 2.5 is that the monetary base shock has a positive eect
12
We do not report the condence intervals for lack of space.
2.3. Empirical results 109
on house prices (CPIHWEGFH), which are strongly correlated to the land price. According
to Kwon (1998), a large fraction
13
of business investment nanced by bank loans is secured
by land. It is therefore plausible to think that movements in land prices, whose values may
serve as collateral, can improve nancing conditions and may play a signicant propagating
role in the monetary transmission mechanism. As for disaggregated production, as shown in
Figure 2.6 (in Appendix B), except mining, a positive shock to the monetary base increases
all industrial production components, capacity utilization rates, shipments and to a lesser
extent earnings and employment. The employment rate remains fairly unaected.
This result raises the question of the transmission mechanism through which the
QEMP aected the output and ination. The QEMP can work through either policy-
duration channel or the portfolio-rebalancing channel, or both of them.
2.3.4.2 Policy-duration eect
The empirical validity of the policy-duration eect implied by theoretical studies is still an
open question. As shown by Eggertsson and Woodford (2003) and Jung et al. (2005), a
central bank can lower long-term interest rates by committing to the future zero interest
rates in advance, and so lower the real interest rates thanks to the ination expectation.
Eggertsson and Woodford (2003) argue that this expectation channel is the only way to
escape deation and stimulate an economy under a liquidity trap situation. Note that
lowering long-term interest rates is an intermediate objective and the ultimate objective
of monetary policy is price stabilization, which will hopefully facilitate economic growth.
Therefore, if this expectation channel is eective the economic recovery should increase
expected ination and thus future short-term interest rates, which, in turn, will raise long-
term interest rates. From Figure 2.3, we see that during the period of QEMP the reaction
13
Of total secured bank loans, about 45% have been collateralized by land, while only about 3%
have been backed up by stocks and bonds. Thus, land prices might be closely related to real activities
in Japan.
110
of private-sector (HHE) and business-sector (DIBSE) expectations is signicant but short-
lived.
Figure 2.3: Impulse responses - Policy-duration eect
3 6 9 12 15 18 21
-0.2
-0.1
0
0.1
0.2
JGB 10Y, 89-Q1
3 6 9 12 15 18 21
-0.4
-0.2
0
0.2
JGB 10Y, 95-Q1
3 6 9 12 15 18 21
-0.2
-0.1
0
0.1
0.2
JGB 10Y, 2002-Q1
3 6 9 12 15 18 21
-0.5
0
0.5
LT 5Y, 89-Q1
3 6 9 12 15 18 21
-0.4
-0.2
0
0.2
0.4
LT 5Y, 95-Q1
3 6 9 12 15 18 21
-0.2
0
0.2
0.4
LT 5Y, 2002-Q1
3 6 9 12 15 18 21
-0.4
-0.2
0
0.2
DIBSE, 89-Q1
3 6 9 12 15 18 21
-0.4
-0.2
0
0.2
DIBSE, 95-Q1
3 6 9 12 15 18 21
-0.4
-0.2
0
0.2
DIBSE, 2002-Q1
3 6 9 12 15 18 21
-0.4
-0.2
0
0.2
0.4
HHE, 89-Q1
3 6 9 12 15 18 21
-0.4
-0.2
0
0.2
0.4
HHE, 95-Q1
3 6 9 12 15 18 21
-0.4
-0.2
0
0.2
0.4
HHE, 2002Q1
The gures show the reactions of ve-year JGBss yields (LT 5Y), long-term JGBss yields (JGB
10 Y), private sector (HHE) and business-sector (DIBSE) expectations to a shock to M0 over
21 quarters for three dierent dates. The solid lines show the impulse responses implied by the
time-varying FAVAR (posterior median) and dashed lines represent the 10
th
and 90
th
percentiles.
In contrast with Nakajima et al. (2009a) and Kimura et al. (2003), the impulse
responses of medium- (LT 5Y) and long-term (JGB 10 Y) interest rates are insignicant.
However, these results need to be interpreted carefully and should not be taken as evidence
against the expectation channel (neo-Wicksellian view). The positive eect on private and
business sector expectations, even short-lived, can also be interpreted as a successful BOJ
policy commitment in preventing a downward spiral of expectations. However, as argued
in Nakajima et al. (2009a), the policy commitment, alone, is not sucient to generate
signicant inationary pressure to escape from the trap of deationary phase and to lead to
2.3. Empirical results 111
upward shifts in the trend growth path. In order to better analyze the policy-duration eect,
a more appropriate model examining the interactions between the macroeconomic variables
and the yield curve is needed. This will be the subject of the next chapter.
2.3.4.3 Portfolio-rebalancing channel
The portfolio-rebalancing channel is supposed to be induced indirectly by the increase in the
CAB or directly by the increase in BOJs JGB purchases. As prices rise for JGBs their yields
will fall relative to those of other assets. Households and companies may be encouraged
to switch into other type of assets in search of higher returns. That would push up other
asset prices as well. Similarly, households and companies use the additional money injected
by the central bank to buy alternative non-monetary assets, increasing their prices. The
stock price (TOPIX), which serves as a proxy for nancial asset prices, increases in reaction
to monetary base expansion but becomes insignicant only after around 6 quarters (Figure
2.4). As investors demand for alternative assets such as equities increases, the ability
of businesses to raise nance in capital markets improves and the cost falls. By contrast
with Oda and Ueda (2007) and Kimura et al. (2003), these results show that the portfolio-
rebalancing channel could have a role in transmitting monetary policy shocks. It is likely that
the QMEP was eective through the stock price channel. As explained in chapter 1, there
are four possible channels through which higher stock prices boost output: an increase
in consumption through a rise in households wealth (the wealth eect); an increase in
investment through higher Tobins q; an increase in bank lending through a decline in the
external nance premium of borrowers (the balance sheet eect); and an increase in bank
lending through an improvement in the banks capital-to-asset ratios.
112
Figure 2.4: Impulse responses - Portfolio-rebalancing channel
3 6 9 12 15 18 21
-0.5
0
0.5
Consumption, 89-Q1
3 6 9 12 15 18 21
-0.5
0
0.5
Consumption, 95-Q1
3 6 9 12 15 18 21
-0.5
0
0.5
Consumption, 2002-Q1
3 6 9 12 15 18 21
-0.05
0
0.05
Bank lending, 89-Q1
3 6 9 12 15 18 21
-0.05
0
0.05
Bank lending, 95-Q1
3 6 9 12 15 18 21
-0.05
0
0.05
Bank lending, 2002-Q1
3 6 9 12 15 18 21
-0.2
0
0.2
0.4
TOPIX, 89-Q1
3 6 9 12 15 18 21
-0.2
0
0.2
0.4
TOPIX, 95-Q1
3 6 9 12 15 18 21
-0.2
0
0.2
0.4
TOPIX, 2002-Q1
The gures show the reactions of the consumption, the bank lending and asset prices (TOPIX) to
a shock to M0 over 21 quarters for three dierent dates. Solid lines show the impulse responses
implied by the time-varying FAVAR (posterior median) and dashed lines represent the 10
th
and
90
th
percentiles.
While bank lending does not react signicantly to the monetary base shock, con-
sumption
14
increases signicantly during the QEMP period but this reaction is short-lived.
Therefore, we suppose that the stock price channel is driven mainly by the wealth eect
and investment
15
. The increase in the stock price may have helped Japanese rms restore
their balance sheets, which were destroyed after the asset price bubble burst and land prices
collapsed in the early 1990s
16
. Companies therefore started investing their prots instead
of using them to repay debts.
Our ndings suggest that QEMP is eective and works through both monetary policy
14
This correponds to the total consumption for 2 or more persons (variable number 49 in the list
of variables in Appendice A.)
15
The data for private investments are available only from 1994
16
As argued in Koo (2008), the corporate sector was busy repaying debt until 2004; net debt
repayments fell to zero by the end of 2005.
2.3. Empirical results 113
commitment and portfolio-rebalancing channel. This is in line with Bernanke and Reinhart
(2004)s suggestions that the neo-Wicksellian policy commitment needs to be complemented
with more aggressive use of monetarist approaches to monetary policy. The authors also
argue that the BOJ should not have to limit changes to the composition of its balance
sheet to only focus mainly on purchases of government securities but that it should extend
its open market purshases to a wide range of securities. The recommendations addressed
by Bernanke and Reinhart (2004) to the BOJ were put into practice by Ben Bernanke, as
chairman of the Federal Reserve System, in order to combat the current nancial crisis.
The non-conventional monetary policy strategy adopted by the Fed called credit easing, is
similar to QEMP in its explicit commitment to maintaining the nominal short-term interest
rate at low levels. However, the main dierence between the two strategies is that the Fed,
through its Credit Easing, focuses on the change in the composition of its balance sheet by
purchasing a wide range of securities
17
, yet the size of the balance sheet remains a secondary
objective. Moreover, Gagnon et al. (2010) show that credit easing mainly worked through
the portfolio-rebalancing channel, the decline in long-term interest rates being attributed
to the decline in term premia and not to the expectation of low future short-term interest
rates. The authors argue that the large-scale asset purshases (LSAPs) implemented by the
Fed not only reduced longer-term yields on the assets being purshased (agency MBS and
Treasury securities), but also reduced yields on other assets (corporate bonds and equities).
This complementarity between the portfolio-rebalacing channel and the expectation
channel is, moreover, corroborated by the fact that the BOJ, building on its past experience
with QEMP, recently implemented Comprehensive Monetary Easing (CME). This strategy
17
The Feds experience of credit easing comprises two courses of action. First, there is an ex-
plicit commitment to maintaining the nominal short-term interest rate at low levels. Second, the
Fed implements large-scale asset purchases (LSAPs), which range from housing agency debt and
mortgage-backed securities (MBS) to long-term Treasury securities. However, the Bank of England
and the ECB associated their operating procedure on a monetarist view of the transmission process.
They began a programme of large-scale asset purchases in 2009 without any explicit commitment
to maintaining their policy rates at low levels.
114
focuses more on changes in balance sheet composition and on the extension of open market
purchases to a wide range of securities
18
.
2.4 Conclusion
Recent research has employed VAR models, accounting for regime changes, leading
to advances in the measurement of the eect of Japanese quantitative easing. These models
permit researchers to verify whether or not the Japanese monetary policy has undergone
structural changes. This issue is particularly important for the Japanese economy in the last
two decades. The main shortcoming of this literature has been the inability to incorporate
larger and more realistic information sets related to central banks and the private sector.
This chapter employed a time-varying parameters FAVAR (TVP-FAVAR) model to overcome
these limitations. This model allowed us both to take into account regime changes and to
measure the eects of monetary policy shocks on numerous variables.
Our analysis delivers four main results. First, unsurprisingly, our results suggest
that the best model to specify Japanese monetary policy during the two last decades is a
model where all parameters vary over time. This corroborates our choice of a time varying
parameters model. Second, the eect of QEMP on activity and prices is stronger than
previously found. In particular, we nd a signicant price reaction to a monetary policy
shock. Moreover, the problem related to the price puzzle, the price divergence and the
non-neutrality of money that arises in previous works disappears under our data-rich model.
18
In October 2010, the BOJ announced the adoption of the new monetary strategy called Com-
prehensive Monetary Easing in reference to its past experience of QEMP. This strategy approaches
credit easing as implemented by the Fed, consisting of the following two principal courses of action.
First, as in QEMP, the BOJ commits to maintaining short-term interest rates at around 0 to 0.1
percent. Second, the BOJ increases the amount of outright purchases not only of government securi-
ties, but also of commercial paper, corporate bonds, exchange-traded funds and Japanese real estate
investment trusts. Note that in contrast to QEMP, CME puts the emphasis on the composition of
the BOJs balance sheet without any explicit reserve level target.
2.4. Conclusion 115
Third, by contrast with previous work, there is a detectable eectiveness of the portfolio-
rebalancing channel, which could have a role in transmitting monetary policy shocks. The
weak reaction of bank lending and the signicant increase in consumption, even short-lived,
lead to think that the positive and signicant asset price reaction generates two main eects:
it means lower yields, reducing the cost of borrowing for households and companies, leading
to higher consumption and investment spending. It also means that the wealth of the asset
holders increases, which should boost their spending. Fourth, while the policy commitment
succeeds in controlling private and business expectations, the reaction of medium to long-
end of the yield curve remains insignicant.
Moreover, one interesting result that emerges from the price reaction is that the
monetary base shock has a positive eect on house prices, which are strongly correlated to
the land price. A large fraction of business investment nanced by bank loans is secured
by land. It is therefore plausible to think that movements in land prices, whose values may
serve as collateral, can improve nancing conditions and may play a signicant propagating
role in the monetary transmission mechanism.
These results shoud not be taken as evidence in favor of portfolio-rebalancing channel
against the expectation channel. The positive but short-lived eect on private and business
sector expectations may not be sucient to restore the previous trends in prices and out-
put, but might prevent downward spiral of expectations. Therefore, the two channels are
complementary rather than exclusive. On the other hand, since the expectations hypothesis
of the term structure of interest rates is a necessary condition for the eectiveness of the
expectation channel, we think that a macro-nance model is more appropriate to better
analyse the eectiveness of the policy-duration eect. This will be the issue of the next
chapter.
116
Appendix A : Data and transformations
Table 2.2 Variable list
Data are extracted from Reuters EcoWin database. The transformation codes (T)
are : 1 no transformation ; 2 rst dierence ; 4 logarithm; 5 rst dierence of
logarithm. In this database VRAI means seasonally adjusted.
# Mnemonic T Description
Slow moving
1 IPT 5 Industrial Production Total Index
2 IPSCP 5 Production, Ceramics, stone and clay products, Index
3 IPCH 5 Production, Chemicals, Index
4 IPVEH 5 Production, Industrial vehicle, Index
5 IPDVEH 5 Production, Domestic vehicle, total
6 IPFM 5 Production, Fabricated metals, Index
7 IPFT 5 Production, Food and tobacco, Index
8 IPGM 5 Production, General machinery, SA, Index
9 IPIS 5 Production, Iron and steel, Index
10 IPMANUF 5 Production, Manufacturing, Index
11 IPMMANUF 5 Production, Mining and manufacturing, Index
12 IPNFM 5 Production, Non-ferrous metals, Index
13 IPOMUNUF 5 Production, Other manufacturing, Index
14 IPPCP 5 Production, Petroleum and coal products, Index
15 IPPP 5 Production, Plastic products, Index
16 IPPI 5 Production, Precision instruments, Index
17 IPIP 5 Production, By industry, paper, Index
18 IPCE 5 Production, Communication Equipment, Index
19 IPSD 5 Production, Semiconductor devices, Index
20 IPTEXT 5 Production, Textiles, Index
21 IPTRANSPE 5 Production, Transport equipment, Index
22 SHIPMCGEXTE 5 Shipments, Capital goods excl transport equipment Index
23 SHIPMAG 5 Shipments, Capital goods, SA, Index
24 SHIPMCE 5 Shipments, Communication Equipment , Index
25 SHIPMCONSTG 5 Shipments, Construction goods,Index
26 SHIPMCONSUMG 5 Shipments, Consumer goods, Index
27 SHIPMDCG 5 Shipments, Durable consumer goods, Index
28 SHIPMING 5 Shipments, Investment goods , Index
29 SHIPMMANUF 5 Shipments, manufacturing, Index
30 SHIPMMMANUF 5 Shipments, Mining and manufacturing, Index
31 SHIPMNDCG 5 Shipments, Non-durable consumer goods, Index
32 SHIPMPG 5 Shipments, Producer goods total, Index
33 SHIPMPGMMANUF 5 Shipments, Producer goods, for mining and manufacturing, Ind
34 SHIPMPGOTHERS 5 Shipments, Producer goods, for others Index
35 CAPUORCH 5 Capacity Utilization, Operation Ratio,Chimicals
36 CAPUORFM 5 Capacity Utilization, Operation Ratio, Fabricated metals
37 CAPUORGM 5 Capacity Utilization, Operation Ratio, General machinery
38 CAPUORIS 5 Capacity Utilization, Operation Ratio, Iron and steel
39 CAPUORMINDUS 5 Capacity Utilization, Operation Ratio, Machinery industry
40 CAPUORMNUF 5 Capacity Utilization, Operation Ratio, Manufacturing
41 CAPUORPC 5 Capacity Utilization, Operation Ratio, Petroleum and coal
42 CAPUORPPP 5 Capacity Utilization, Operation Ratio, Pulp, paper and pap
43 CAPUORTEXT 5 Capacity Utilization, Operation Ratio, Textiles
44 CAPUORTE 5 Capacity Utilization, Operation Ratio, Transport equipment
45 HWAVGC 5 Hours Worked, Average Per Month, Construction
46 HWAVGMANUF 5 Hours Worked, Average Per Month, Manufacturing
47 HWAVGMIN 5 Hours Worked, Average Per Month, Mining
2.4. Conclusion 117
48 CONSGENEXCLHA 5 Japan, Index of Consumption Expenditure Level, 2 or more persons, ge-
neral excl housing, automobiles, money gifts & remittance, Vrai, Index,
JPY, 2000=100
49 CONSGENERAL 5 Japan, Consumer Surveys, Index of Consumption Expenditure Level, 2
or more persons, general, Vrai, Index, JPY, 2000=100
50 CONSHOUSING 5 Japan, Consumer Surveys, Index of Consumption Expenditure Level, 2
or more persons, housing, Vrai, Index, JPY, 2000=100
51 CONSTRANSCOM 5 Japan, Consumer Surveys, Index of Consumption Expenditure Level, 2
or more persons, transportation & communication, Vrai, Index, JPY,
2000=100
52 UNEMP 5 Unemployment, Rate, SA
53 EMPTRATE 5 Employment, Overall, Total
54 EMPCONST 5 Employment, By Industry, Construction, Index
55 EMPGOV 5 Employment, By Industry, Government
56 EMPMANUF 5 Employment, By Industry, Manufacturing
57 EMPALLINDUST 5 Employment, By Status, Regular employees, all industries
58 JALFT 5 Japan, Activity, Labour Force, Total
59 SDST 5 Sales at Deapartement Stores, Total, Index
60 CPIALL 5 Japan, Consumer Prices, Industrial products,All, Index, JPY, 2000=100
61 CPIINDP 5 Japan, Consumer Prices, Industrial products,Textile, Index, JPY,
2000=100
62 CPIINDT 5 Japan, Consumer Prices, Electricity, gas & water charges , Index, JPY,
2000=100
63 CPIEGW 5 Japan, Consumer Prices, Services , Index, JPY, 2000=100
64 CPISERV 5 Japan, Consumer Prices, Durable goods , Index, JPY, 2000=100
65 CPIDG 5 Japan, Consumer Prices, Non Durable goods , Index, JPY, 2000=100
66 CPINDG 5 Japan, Consumer Prices, Food , Index, JPY, 2000=100
67 CPIFOOD 5 Japan, Consumer Prices, Reading and Recreation , Index, JPY,
2000=100
68 CPIRR 5 Japan, Consumer Prices, Nationwide, Miscellaneous Goods and Ser-
vices, Durable goods, Index, JPY, 2000=100
69 CPIGSDG 5 Japan, Consumer Prices, Nationwide, Transport, Private transporta-
tion, Index, JPY, 2000=100
70 CPITPT 5 Japan, Consumer Prices, Nationwide, Transport, Public transportation,
Index, JPY, 2000=100
71 CPITPUBT 5 Japan, Consumer Prices, Nationwide, Communication, Communica-
tion, Index, JPY, 2000=100
72 CPICC 5 Japan, Consumer Prices, Nationwide, Housing, Water, Electricity, Gas
and Other Fuels, Electricity, Index, JPY, 2000=100
73 CPIWEG 5 Japan, Consumer Prices, Nationwide, Health, Medical treatment, In-
dex, JPY, 2000=100
74 CPIHMT 5 Japan, Consumer Prices, Nationwide, Health, Medical care, Index, JPY,
2000=100
75 CPIHMC 5 Japan, Corporate Goods Prices, Domestic demand products, consumer
goods, Index, JPY, 2000=100
76 PPIDDPCG 5 Japan, Corporate Goods Prices, Domestic demand products, nal
goods, Index, JPY, 2000=100
77 PPIDDPFG 5 Japan, Corporate Goods Prices, Domestic demand products, nondu-
rable consumer goods, Index, JPY, 2000=100
78 PPIDDPNCG 5 Japan, Corporate Goods Prices, Domestic demand products, total, In-
dex, JPY, 2000=100
79 PPIDDPT 5 Japan, Corporate Goods Prices, Domestic, capital goods, Index, JPY,
2000=100
80 PPIDCG 5 Japan, Corporate Goods Prices, Domestic, chemicals, Index, JPY,
2000=100
81 PPIDCH 5 Japan, Corporate Goods Prices, Domestic, consumer goods, Index,
JPY, 2000=100
82 PPIDT 5 Japan, Corporate Goods Prices, Domestic, total, Index, JPY, 2000=100
83 PPISERVALL 5 Japan, Corporate Service Prices, All items, Index, JPY, 2000=100
118
84 PPISERVT 5 Japan, Corporate Service Prices, Transportation, Index, JPY,
2000=100
85 PPIFINS 5 Japan, Corporate Service Prices, Finance and insurance, Index, JPY,
2000=100
86 EXPORT 5 Japan, Exports, Volume, Total, Index, JPY, 2000=100
87 IMPORT 5 Japan, Imports, Volume, Total, Index, JPY, 2000=100
Fast moving
88 CONSTSTARTEDP 4 Japan, construction started, Private
89 CONSTSTARTEDPUB4 Japan, construction started, Public
90 CONSTSTARTEDT 4 Japan, construction started, Total
91 HSBS 4 Housing Starts, Housing built for sale
92 HSRH 4 Housing Starts, Rental homes
93 HST 4 Housing Starts, Total
94 NEWORDCONSP 5 Japan, New Orders, Construction, Private sector, JPY
95 NEWORDCONST 5 New Orders, Construction, Total, Big 50 constructors, JPY
96 NEWORDIM 5 Japan, New Orders, Machine Tools, By industry, machine and equip-
ment industries, industrial machinery, JPY
97 NEWORDMTT 5 Japan, New Orders, Machine Tools, By industry, machine and equip-
ment industries, total, JPY
98 NEWORDCMANUF 5 Japan, New Orders, Construction, Manufacturing, JPY
99 JDFFTSET 5 Japan, Daiwa, Free oat, TSE, Total Index, JPY
100 JDFFTSETU 5 Japan, Daiwa, Free oat, TSE, Transportation & Utilities Index, JPY
101 TOPIX 5 Japan, Tokyo SE, Topix Index, Price Return, End of Period, JPY
102 DOLLARYEN 5 US.Dollar/Yen Spot Rate, Average in the Month, Tokyo Market
103 EFFEXCHANGE 5 Japan, BIS, Nominal Narrow Eective Exchange Rate Index, Average,
JPY
104 M1 5 Japan, M1, outstanding at end of period, Vrai, JPY
105 M2CDs 5 M2+CDs/Average Amounts Outstanding/(Reference) Money Stock
106 M3 5 Japan, M3, outstanding at end of period, JPY
107 BOJAAL 5 Japan, BOJ accounts, assets, loans, JPY
108 BOJAAT 5 Japan, BOJ accounts, assets, total, JPY
109 DLBABD 5 Japan, Domestically Licensed Banks, Assets, bills discounted, JPY
110 DLBACL 5 Japan, Domestically Licensed Banks, Assets, call loans, JPY
111 DLBACLBD 5 Japan, Domestically Licensed Banks, Assets, loans and bills discounted,
JPY
112 DLBCBALBD 5 Japan, Domestically Licensed Banks, City banks, assets, loans and bills
discounted, JPY
113 DLBRBALBD 5 Japan, Domestically Licensed Banks, Regional banks, assets, loans and
bills discounted, JPY
114 DLBAL 5 Japan, Domestically Licensed Banks, Assets, loans, JPY
115 DLBCBAL 5 Japan, Domestically Licensed Banks, City banks, assets, loans, JPY
116 DLBRBAL 5 Japan, Domestically Licensed Banks, Regional banks, assets, loans,
JPY
117 INVINVG 5 Inventory Investment goods, Index
118 INVMMANUF 5 Inventory Mining and manufacturing, Index
119 INVFM 5 Inventory Fabricated metals, Index
120 INVCG 5 Inventory Construction goods, Index
121 INVCAPG 5 Inventory Capital goods, Index
122 INVNDCG 5 Inventory Non-durable consumer goods, Index
123 INVCONSUMG 5 Inventory Consumer goods, SA, Index
124 INVPG 5 Inventory Producer goods, Index
125 PLRLT 1 Japan, Prime Rates, Prime Lending Rate, Long Term, End of Period,
JPY
126 PLRST 1 Japan, Prime Rates, Prime Lending Rate, Short Term, End of Period,
JPY
127 TB3M 1 Japan, Treasury Bills, Bid, 3 Month, Yield, End of Period, JPY
128 TIOR3M 1 Tokyo interbank oered rates (3 months)
129 JGB10 1 Yield of Government Bonds (10 Y)
2.4. Conclusion 119
130 SP10TIOR3M 1 Spread rate : Yield of Government Bonds (10 Y) - Tokyo Interbank
Oered Rate (3 M)
131 IBGB10 1 10-year interest-bearing Government Bonds
132 LGB10 1 10-year Local Government Bonds
133 GGB10 1 10-year Government Guaranteed Bonds
134 IBBD5 1 5-year interest-bearing Bank debentures
135 CALLRATE 1 Japan, Interbank Rates, Uncollateralized, O/N, Average, JPY
136 SPIBBD5TIOR3M 1 Spread between the Yield on long-term and short-term : Yield of Go-
vernment Bonds (5 Years) - Tokyo Interbank Ored Rate (3 months)
137 SPGGB10TIOR3M 1 Spread between the Yield on long-term and short-term : Yield of Go-
vernment Guaranteed Bonds (10 Years) - Tokyo Interbank Ored Rate
(3 months)
138 DIBSE 1 DI/Business Conditions/All industries/Forecast
139 HHE 1 Consumer Surveys, Consumer Condence, Including one-person house-
holds, total
120
Appendix B : Impulse response functions for price and activity
variables
Figure 2.5 Impulse responses - Disaggregated price
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
0.75
1
M0
89-Q1
95-Q2
01-Q1
3 6 9 12 15 18 21
0
0.25
IRF of
CPITTC
3 6 9 12 15 18 21
0
0.25
IRF of
CPICC
3 6 9 12 15 18 21
0
0.25
IRF of
CPIHWEGFH
3 6 9 12 15 18 21
0
0.25
IRF of
CPIMGSM
3 6 9 12 15 18 21
0
0.25
IRF of
PPIDDPCG
3 6 9 12 15 18 21
0
IRF of
PPPDDPDCG
3 6 9 12 15 18 21
0
0.25
IRF of
PPIDDPNDCG
3 6 9 12 15 18 21
0
0.25
IRF of
PPIDDPPGT
3 6 9 12 15 18 21
0
0.25
0.5
IRF of
PPITEXT
3 6 9 12 15 18 21
0
0.25
IRF of
PPIDT
3 6 9 12 15 18 21
-0.25
0
IRF of
PPIDTE
The gures show the reactions of some selected prices to a shock to M0 over 21 quarters for
three dierent dates . The solid lines show the impulse responses implied by the time-varying
FAVAR (posterior median) and dashed lines represent the 10
th
and 90
th
percentiles. Details on
nomenclatures are given in Appendice A.
2.4. Conclusion 121
Figure 2.6 Impulse responses - Disaggregated production
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
0.75
1
M0
89-Q1
95-Q2
01-Q1
3 6 9 12 15 18 21
0
0.25
IRF of
IPCE
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
0.75
1
IRF of
IPGM
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
0.75
1
IRF of
IPV
3 6 9 12 15 18 21
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
IRF of
IPMANUF
3 6 9 12 15 18 21
0
IRF of
IPMIN
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
IRF of
IPTRANSE
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
0.75
1
IRF of
SHCG
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
0.75
1
IRF of
SHCONSUMG
3 6 9 12 15 18 21
0
0.25
0.5
0.75
1
IRF of
SHDCG
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
0.75
1
IRF of
SHINVG
3 6 9 12 15 18 21
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
IRF of
SHMMANUF
3 6 9 12 15 18 21
0
0.25
0.5
0.75
1
IRF of
CAPMI
3 6 9 12 15 18 21
0
0.25
0.5
0.75
1
IRF of
CAPMANUF
3 6 9 12 15 18 21
-0.5
-0.25
0
0.25
0.5
IRF of
CAPTRANSE
3 6 9 12 15 18 21
0
IRF of
EAVGMALL
3 6 9 12 15 18 21
0
IRF of
EMPT
3 6 9 12 15 18 21
0
IRF of
EMPGVT
3 6 9 12 15 18 21
0
0.25
0.5
0.75
1
IRF of
NEWJOB
3 6 9 12 15 18 21
0
IRF of
UNEMPR
The gures show the reactions of some selected variables related to activity to a shock to M0
over 21 quarters for three dierent dates . The solid lines show the impulse responses implied by
the time-varying FAVAR (posterior median) and dashed lines represent the 10
th
and 90
th
percen-
tiles.Details on nomenclatures are given in Appendice A.
122
3
3
Quantitative Easing and the Time-Varying
Dynamics of the Term Structure of Interest rate
in Japan
3.1 Introduction
In a zero interest rate environment the short-term interest rate is no longer a policy instru-
ment under the direct control of the central bank. The alternative monetary policy used by
most major central banks is monetary easing. The goal of the central bank is therefore to
impact the economy across the yield curve, bringing down long-term interest rates, thereby
boosting the economy. The aim of this chapter is to examine the eectiveness of such a
policy in aecting the yield curve using the Japanese experience of QEMP. We are particu-
larly interested in analyzing the possible bi-directional feedback from the yield curve to the
economy.
123
124
The QEMP as implemented by the BOJ comprised three courses of action namely,
(i) injecting ample liquidity into the market using the current account balances (CAB) as the
main monetary policy instrument, (ii) making a commitment to maintain short-term rates at
around zero until the CPI ination stabilized at zero percent or increased year after year and
(iii) purchasing more of long-term Japanese government bonds (JGBs). The transmission
mechanisms suggested by this policy are the portfolio-rebalancing channel (Metzler (1995))
and the expectation channel which consists of the policy-duration eect (Krugman (2000)
and Eggertsson and Woodford (2003)) and the signaling eect.
This chapter evaluates the eectiveness of the QEMP, focusing on the expectation
channel. The eectiveness of such a channel depends totally on the credibility of the central
banks policy of maintaining the future short-term interest rate at a near zero level. The
desired intermediate eect of the monetary policy is that the reduction of expected future
short-term rates will be transmitted to the long end of the yield curve. The decline in the
long-term interest rates will, in turn, lead to increased expectations of ination and stimulate
activity.
Several papers have examined the eectiveness of this transmission channel by fo-
cusing on the term structure of interest rates. Oda and Ueda (2007), Okina and Shiratsuka
(2004a) and Baba et al. (2005) show that policy duration has a clear and signicant eect,
lowering the yield curve. The common point of these papers is that they do not examine
the transmission of this positive eect to the real economy. On the other hand, Evans and
Marshall (2007) and Ang and Piazzesi (2003) examine the joint dynamics of bond yields and
macroeconomic variables in a vector autoregression. Ang and Piazzesi (2003) show that a
substantial portion of short- and medium-term bond yields is explained by macroeconomic
variables. In contrast, Evans and Marshall (2007) nd that macroeconomic variables do also
explain much of the long-term bond yield dynamics. For the Japanese case, Nakajima et al.
(2010), using a time-varying-parameter VAR model (TVP-VAR), show some evidence of the
3.1. Introduction 125
eectiveness of policy duration in lowering ve-year JGB yields, although this eect is not
transmitted to the real economy. One potential drawback of such models is that few specic
yields are taken into consideration. The results may therefore not reect to full range of term
structure. Diebold et al. (2006) examine the interactions between the macroeconomy and
the yield curve by means of the one-step Kalman lter approach. They aggregate information
from a large set of yields using latent factors, which represent the level, the slope and the
curvature, based on Nelson and Siegel (1987)s model. Using this model, which allows for
correlated latent yield factors and observed macroeconomic variables, Diebold et al. (2006)
show that macroeconomic variables have strong eects on future movements of the yield
curve, while latent interest rate factors have a relatively small impact on macroeconomic
variables.
Needless to say, it could be unrealistic to assume time-invariance either in monetary
policy transmission mechanisms or in the structure of the Japanese economy. Indeed, as
already documented in several papers for the Japanese economy (Miyao (2000), Fujiwara
(2006), Inoue and Okimoto (2008), Nakajima et al. (2009a) and chapter 1), there is clear
evidence of signicant structural changes in the last two decades. As shown in Cargill et al.
(2001), the Japanese economy has become very unstable having experienced signicant
institutional and monetary strategy changes during the lost decade. For these reasons and
for the rst time to the best of our knowledge, we employ a generalized Nelson-Siegel model
with time-varying coecient and stochastic volatilities, as described in Bianchi et al. (2009),
for the Japanese case. This study diers from previous studies on the JGB market on three
points. Firstly, it focuses on a more complete set of possible macroeconomic variables and
monetary policy instruments and their possible eects on the term structure of interest
rates. Secondly, it allows us to take into account potential instabilities both in monetary
policy transmission mechanisms and in the relationship between yield curve and the structure
of the economy. Finally, the data sample in this study is signicantly larger ; data span the
126
period between February 1985 and October 2009. This allows us to analyze the relationship
between yield curve and macroeconomic and monetary variables under dierent economic
conditions, i.e. periods with dierent monetary policy strategies.
The objective of this work is twofold. First, we propose a constrained smoothing
B-splines method to estimate the term structure of spot rates using JGB prices. Given
that short term interest rates were even lower than usual during the QEMP period, using
traditional yield curve models could result in negative values for yields with short maturity
during this period. In order to overcome this problem we incorporate non-negative restrictions
in the smoothing B-splines method. Second, we apply a time-varying parameter macro-
nance model to data on JGBs with 17 dierent maturities as well as three macroeconomic
variables, namely, output gap, ination and monetary policy instrument.
Contrary to the results of standard term-structure models with time-invariant term
premia, our results show that the expectations hypothesis of the term structure of interest
rates is generally supported even during the QEMP period. This is a necessary condition for
the eectiveness of the expectation channel. Moreover, the estimation results reveal that
the contribution of macroeconomic variables to the yield curve variation is relatively small,
especially during the QEMP period. As for the feed-back eect, the yield curve factors
contribute only marginally to ination variation. They account for a greater part of output
gap dynamics during the period of high interest rates, but these eects revert to a much
lower level during the QEMP period. These results indicate that during the deationary
period and under the zero lower bound of interest rates the relationship between nancial
markets and macroeconomic variables becomes very weak. These ndings are corroborated
by the impulse response functions. The monetary policy shock has a signicant eect on the
yield curve level factor only during the high interest rate periods. However, during the QEMP
period the decline in the level factor, while insignicant, indicates strengthening credibility
of the BOJ and thus the eectiveness of its policy.
3.2. Estimating spot rate curves for Japan 127
The rest of the chapter is structured as follows. The next section presents the
term structure estimation and the data construction. Section 3 describes the time-varying
parameter macro-yield model. The results are presented in section 4 and nally section 5
concludes.
3.2 Estimating spot rate curves for Japan
3.2.1 Data construction
We base our analysis on data originating from two dierent sources and covering the
period from 1985 :02 to 2009 :10. Data were provided by Tokyo Stock Exchange (TSE)
and Japan Securities Dealers Association (JSDA)
1
. The data set comprises beginning-of-
month observations of the ocially quoted prices, remaining maturities and coupons of a
total of 374 listed public debt securities. The data used for estimating zero coupon yield
curves cover three categories of government issues : 10-year JGB
2
, 3-month treasury bills
(3m TB) and 3-month nancing bills
3
(3m FB). In order to obtain a more homogeneous
set of data, TB and FB prices are adjusted for withholding tax which is levied at issuance
and repaid at redemption
4
. The number of debt securities available for each month varies
considerably until the end of 2000, between 48 and 89 and it grows sharply from 2001 to
vary only between 86 and 95. In order to select the most accurately priced bonds, we apply
several data lters. We eliminate the data for 10-year JGBs with remaining maturity of less
1
From 1985 :02 to 2001 :12 data are provided by TSE while from 2002 :01 to 2009 :10 data are
provided by JSDA.
2
The 10-year JGBs are most liquid bonds in the Japanese bond market, and they work as bench-
marks for bond investors.
3
The issue of 3-month TB was terminated in 2000 :03 while maturity of nancing bills (FB) is
extended to three months in 1999 :04. As for three-month yield, we treat therefore 3-month TB and
3-month FB interchangeably.
4
Prices are adjusted according to this formula : P
adj
=
PF
F+(FP
0
)t
, where P represents the market
price, P
adj
the tax-adjusted market price, F the face value, P
0
issue price and t the tax rate (0.18).
128
than half a year not only because of the existence of a redemption fee
5
but also because
these JGBs are not actively traded and appear to be signicantly inuenced by their low
liquidity and therefore they are not accurately priced. Moreover, in order not to distort real
yields we eliminate outliers from data set bonds. We exclude then bonds whose yields dier
greatly from these at nearby (similar) maturities.
3.2.2 Estimation procedure
Our objective is to estimate macro-dynamic yield model from the spot rates. In order to
estimate the spot rate curve, given a set of bond prices, we apply the B-spline method and
parametrize the spot curve in terms of a cubic B-spline. Moreover, we apply smoothing
splines that incorporate a roughness penalty parameter, as the Bank of Japan approach and
use Fisher et al. (1995) method to estimate spot rate curve. The advantage of using this
method is that the cubic B-spline has base functions which have compact support (non-zero
function value) in a knot interval, and we can impose non-negative spot rate restriction in
the estimation procedure. This is even more important when using Japanese data since, in
the special case of the zero-interest-rate environment characterizing the Japanese economy
since 1999, short maturity bond yields can be negative
6
.
Given the wealth of literature detailing the use of smoothing cubic spline to extract
spot rates, we provide only the essential elements of the method
7
. Consider a set of N bonds
traded on one date. Let P
i
be the market price of bond i ,
_
C
i ,j
_
1j k
i
be its principal and
interest payment, which is paid at a set of cash ow dates {t
i ,j
}
1j k
i
. Under the classical
5
Bonds with a remaining time to maturity below half a year are usually excluded to estimate yield
curves using Japanese government securities. This because the redemption fee could lead to negative
yields. This issue doe not impose problem in our work since we impose positive spot rate constraint
in estimation procedure, which is explained below.
6
An other solution consists of replacing negative values of yields with zeros as in Ueno et al.
(2006) that combine Black model with Gorovoi and Linetsky (2004) model to estimate yield curve
where interest rates are considered as options and negative values are replaced by zeros.
7
See Fisher et al. (1995) for more complete description of this model.
3.2. Estimating spot rate curves for Japan 129
assumptions that there is no taxes or transaction costs, absence of arbitrage implies that
the observed bond price (market price plus accrued interest a
i
) is equal to the present value
of its future cashows :
P
obs
i
= P
i
+a
i
=

P
i
+
i
=
k
i

j =1
C
i ,j
(t
i ,j
) +
i
(3.1)
=
k
i

j =1
C
i ,j
exp(t
i ,j
(
i ,j
)) +
i
(3.2)
=
k
i

j =1
C
i ,j
exp(

t
i ,j
0
f (s)ds) +
i
(3.3)
where (.) is the discount function,
i
are independent and normally distributed with mean
of zero and variance
2
. It is widely known that the discount function (t) is related both
to the spot rate ((t)) and to the instantaneous forward rate (f (t)) respectively by (t) =
ln((t))/t and f (t) =(t)

/(t).
When using the smoothed cubic B-spline we place the B-spline bases on the spot
rate curve :
(t) =

=1

(t) (3.4)
where (t) = (
1
(t),
2
(t)...

(t))

is a cubic B-spline basis, it is an -dimensional vector

constructed from a set of basis functions (
j
(t); j = 1, ..., ), and = (
1
, ...,

) is an
unknown parameter vector to be estimated and is the number of knot points plus 2.
We impose positivity constraints in the estimation of the spot rate curve. Since the cubic
B-spline basis functions take their maximum at a center of adjacent knot points, it suces
to verify positiveness of the sport rates at middle points of adjacent knot points so as to
assure positive spot rates. Therefore, the spot rate curve is chosen to be the cubic B-spline
130
which minimizes the objectif function for a given , with respect to as follows :
Min

i =1
(P
i

P
i
())
2
+

T
0

(s)ds
SC (t) 0(t [0, T])
(3.5)
where T is the maximum maturity. The rst term measures the error in the pricing of input
bonds. The second term represents the roughness penalty parameter that sets the level of
smoothing in term structure. It is the size of the roughness penalty parameter () that
controls the tradeo between the smoothness in the curve and the goodness of t. The
value of is determined by Generalized Cross-Validation (GCV), it is chosen to minimize
the expression
() =

N
i =1
(P
i

P
i
(

()))
2
(N enp())
2
(3.6)
where enp() is the eective number of parameters, the cost or tuning parameter. For
each we solve

() and then calculate (). In this paper we impose non-negativity

constraint when estimating to guarantee the positiveness of the spot rates
8
, otherwise
the estimated spot rates may be negative at shorter maturities. In the Japanese case this
problem could arise as short term interest rates are almost zero starting from 1999.
As for the number of knot points selected here, Fisher et al. (1995) suggest choosing
the number of knot points to be roughly one third of the sample size. Applied to Japanese
bond market this gives us 15 knot points. It is also necessary to decide on their location ;
although the knots could be distributed evenly over time to maturity it is common to concen-
trate them towards the short end to capture the greater complexity of the curve at shorter
maturities. The maturities at which our knots are located are 0, 0.25, 0.5, 0,75, 1, 2, 3, 4,
5, 6, 7, 8, 9 and 10 years. The size of the roughness penalty parameter () depends on that
of the tuning parameter , which is xed by discretion. If this parameter was almost equal
8
The estimates are made using Matlab software.
3.2. Estimating spot rate curves for Japan 131
to zero there would be no smoothing of the curve and the resulting forward curve could
oscillate wildly. Alternatively, if it was large, the estimated forward curve would be smoother
at the expense of goodness of t. We follow BIS (2005) and use a tuning parameter of 3.
3.2.3 Summary statistics
Figure 3.1 provides a composite picture of Japanese Government Zero-coupon bond yield
curves over the sample period between 1985 :02 and 2009 :10. It shows that during the
period of quantitative easing, from March 2001 to March 2006, yields for very short-term are
zero. The non-negativity constraint prevents from possible negative yields for these years.
Figure 3.1 Japanese Government Bond spot curves 1985-2009
Note : Spot rates are estimated using Fisher et al. (1995) model where we impose
positive constraint. Sources : data on prices and bond specications are provided by
both TSE and JSDA.
In oder to summarize the statistical properties of the estimated zero-coupon yields
over the sample period, we focus on the 1-, 5- and 10-year spot rates as representative short-
132
, medium- and long-term interest rates. We dene the level, the slope and the curvature as
1
3
_
y
(12)
t
+y
(60)
t
+y
(120)
t
_
, y
(12)
t
y
(120)
t
and 2y
(60)
t
y
(120)
t
y
(12)
t
, respectively. Table 3.1
provides a summary of the descriptive statistics on the four measures of the spot rate curve.
The downward level shift and the reduction in volatility of spot rates are apparent, especially
at the long end of the curve. The average spot rate curve is upward sloping, meaning
that spot rates rise as the maturity of bonds lengthens ; standard deviations of spot rates
generally decrease with maturity ; and spot rates are highly autocorrelated, with decreasing
autocorrelation at longer maturities.The spot rate levels show mild excess kurtosis at short
maturities which decreases with maturity, and positive skewness at all maturities. Excess
kurtosis is, however, more pronounced for rst-dierenced spot rates (for example, 4.145
for the 1-year spot rate). We reject the normality hypothesis at the 5% level for all measures
of the spot rate curve. These descriptive statistics are consistent with some stylized facts
in bond pricing.
Table 3.1 Descriptive statistics : Japanese spot rate curves
Central moments Autocorrelations
Mean Std.Devn Skewness Kurtosis J-B test Lag 1 Lag 2 Lag 3
L
e
v
e
l
s
1-year 0.0192 0.0193 1.5282 1.3529 0.000 0.9602 0.921 0.873
5-year 0.039 0.024 0.964 0.333 0.000 0.968 0.94 0.907
10-year 0.0507 0.0165 0.8649 0.2194 0.000 0.930 0.881 0.83
Slope -0.0315 0.0102 0.6199 0.4957 0.000 0.831 0.70 0.613
Curvature 0.0052 0.0098 0.9708 0.8497 0.000 0.853 0.778 0.71
1
s
t
-
d
i

e
r
e
n
c
e
s
1-year -0.000339 0.003503 -0.396 4.145 0.000 -0.0161 0.228 -0.171
5-year -0.0004 0.0036 0.663 1.99 0.000 -0.016 -0.011 -0.123
10-year -0.000177 0.0059 0.333 2.861 0.000 -0.1647 -0.295 -0.286
Slope -0.00016 0.0054 0.322 1.704 0.000 -0.137 -0.105 0.099
Curvature -0.00015 0.0051 -0.927 5.407 0.000 -0.26 -0.126 -0.034
3.3. Yield-Curve Fitting : The Macro-Finance Model 133
3.3 Yield-Curve Fitting : The Macro-Finance Model
3.3.1 Methodology and Estimation
The recent macro-nance literature has convincingly advocated the case for the
existence of a bi-directional link between the term structure and the rest of the economy
(Ang and Piazzesi (2003), Evans and Marshall (2007), Diebold et al. (2006) and Rudebusch
and Wu (2008)). Moreover, as documented in several papers for the US and UK
9
, there is
strong evidence of instability in the dynamics of the yield curve. What makes the Japanese
case particularly interesting is that its economy has become very unstable, having experienced
signicant institutional and monetary strategy changes during the lost decade. However,
even though earlier empirical contributions based on Japanese data (Oda and Ueda (2007),
Okina and Shiratsuka (2004a), Baba et al. (2005) and Ugai (2007)) show that macro
variables have a clear and signicant lowering eect on yield curve, they do not examine the
potential reverse inuence from the yield curve to the real economy. In addition, to the best
of our knowledge no study has yet tried to model time variations in both the yield curve and
the economy, simultaneously for the Japanese case.
The model presented here is proposed by Bianchi et al. (2009) ; it is set in state-
space form and can be seen as a time-varying extension of the approach used by Diebold
et al. (2006) and developed by Nelson and Siegel (1987). In particular, it allows for time
variation in the state equation, thus revealing possible recurring structural breaks in the time
series dimension of the underlying yield curve factors.
The observation equation is a description of the phenomenon that is being investi-
gated. In the case of the NelsonSiegel model, the observation equation would be the yield
curve equation. The state equation is a description of how the underlying factors evolve over
9
Prominent examples include Diebold et al. (2006), Bianchi et al. (2009), Rudebusch and Wu
(2008).
134
time ; these factors are taken to be unobserved and will be estimated using the Kalman l-
ter. The following equations represent the NelsonSiegel model written in state-space form.
First, the observation equation :
y
t
() = L
t
+
_
1e

_
S
t
+
_
1e

_
C
t
+e
t
() (3.7)
where y
t
() is the bond yield to maturity at time t ; L
t
, S
t
and C
t
are the unobservable
level, slope and curvature factors of the yield curve, the factor loadings 1,
_
1e

_
and
_
1e

_
are for level, slope and curvature factors, respectively. The unity coecient
is a constant, so that it does not decay to zero in the limit. The loading of S
t
is an exponential
function that starts at one and decays monotonically towards zero. The loading of C
t
starts
at zero, increases with the maturity and then declines approaching zero. is a parameter
that governs the exponential decay and determines for which maturity the function assumes
its maximum. We calculate
10
using Japanese data and we set it equal to 0.04215. This
implies that the curvature factor loading reaches its maximum at a maturity of 30 months.
Following Bianchi et al. (2009) we assume that the idiosyncratic component e
t
() is serially
correlated and heteroskedastic but uncorrelated across maturities E
_
e
t
(i )

e
t
(j )
_
= 0 for
i = j . In particular : e
t
() =()e
t1
() +
1/2
t
()
t
where the volatility
t
() follows a
geometric random walk log(
t
) = ln(
t1
()) +
t
. Second, the state equation describes
the dynamics of these factors as a time-varying VAR :
Z
t
=
t
+
P

p=1

t,p
Z
tp
+v
t
(3.8)
where the n1 vector Z
t
=[L
t
, S
t
, C
t
,
t
, Y
t
, R
t
]

denotes macro and yields data matrix. The

errors v
t
are assumed to be normally distributed with 0 mean and time-varying covariance
matrix
t
. Following Primiceri (2005) and Cogley et al. (2005), we use a triangular reduction
10
Diebold et al. (2006) choose =30 months as a reference maturity for the medium term and
set equal to 0.077 for US data.
3.3. Yield-Curve Fitting : The Macro-Finance Model 135
of the state error covariance as follows :

t
= A
1
t

t

t
A

1
t
(3.9)
where A
t
is a lower triangular matrix with ones on the main diagonal and
t
is a diagonal
matrix. the time-varying matrices
t
and A
t
are dened as follows :
A
t
=
_

_
1 0 0
a
21,t
1
.
.
.
.
.
.
.
.
.
.
.
.
0
a
n1,t
.
.
.
a
n(n1),t
1
_

_
and
t
=
_

1,t
0 0
0
2,t
0
.
.
.
.
.
.
.
.
.
0
n,t
_

_
(3.10)
The vectors a
t
=
_
a
21,t
, (a
31,t
a
32,t
), , (a
n1,t
a
n(n1),t
)

are the equation-wise stacked

free parameters of A
t
and h
t
=log(diag(
t
)). As suggested by Primiceri (2005) and Bianchi
et al. (2009) among others, all parameters are assumed to be independent random walks
11
:
_

t
=
t1
+

t
a
t
= a
t1
+
t
h
i ,t
= h
i ,t1
+
h
t
(3.11)
where = [
t

t,p
].
11
As explained in Primiceri (2005) the random walk assumption has the advantages of focusing
on permanent shifts and reducing the number of parameters in the estimation procedure. However,
a random walk model is non-stationary and it is obviously "more explosive" than the number of
observation increases. Our sample contains no more than 200 time series observations. Using such
a short period alleviates this problem.
136
The variance-covariance matrix of innovations is block-diagonal :
_

t
v
t

h
t

t
_

_
N(0, V), where V =
_

_
R 0
.
.
.
0
0
.
.
.
.
.
.
.
.
. Q
.
.
.
.
.
.
.
.
. G 0
0 0 S
_

_
(3.12)
where G = diag(
2
1
, ... ,
2
n
). For simplicity, it is assumed that the matrix S is also block-
diagonal with respect to the parameter blocks belonging to each equation.
On the other hand, Z
t
contains a set of unobservable factors of the yield curve next
to observable macroeconomic factors, namely, the output gap (Y
t
), ination (
t
) and the
monetary policy instrument (R
t
). As noted by Diebold et al. (2006), the intuition behind
this ordering is the fact that the yield curve observations are dated at the beginning of the
month. Under this identication scheme, yield factors are assumed to be contemporaneously
unaected by the macro factors.
To estimate the model we use the procedure originally suggested in Kim and Roubini
(2000) and used by Bianchi et al. (2009), whereby we employ the Gibbs sampling algorithm
that exploits the fact that given observations on Z
t
, the model is a time-varying parameter
model. In this paper preference is given to the Bayesian one-step method rather than the
two-step Diebold-Li approach because in the two-step approach the uncertainty of parameter
estimation and signal extraction in the rst step may aect the second step computations.
However, simultaneous estimation of all parameters results in correct inferences. We follow
the same basic steps of the algorithm as in Bianchi et al. (2009) whereby, rst, given initial
values for the factors, the VAR parameters and hyperparameters are simulated. Second, given
data on Z
t
and y() and a value for , the variance and covariance matrices are simulated.
Third, conditional on all simulated parameters, factors are simulated again. Finally, the
3.3. Yield-Curve Fitting : The Macro-Finance Model 137
chain is started again, going back to the rst step. This iterative procedure converges to an
invariant density that equals the desired posterior density. Gibbs-sampling is implemented in
such a way that the rst 45,000 draws in the Gibbs simulation process are discarded, then
the next 5,000 draws are saved and used to calculate moments of the posterior distribution.
3.3.2 Priors
We follow Primiceri (2005) and calibrate some of the prior distributions using a training
sample and estimating a time-invariant VAR model by OLS. To initialize the factors and
calibrate priors for the VAR, a pre-sample of three years starting in February 1985 is used.
Therefore, the results presented in the following section refer to the period February 1988-
October 2009. The remaining prior distributions are also chosen in a manner similar to
Primiceri (2005) and Bianchi et al. (2009). The prior choices can be summarized as :

0
N(

OLS
, Var(

OLS
))

A
0
N(a
OLS
, Var(a
OLS
)
h
i ,0
N(Log
0
, I
n
10)
log
0
() N(log
0
(), I
n
10)
Q
0
IW
_
Var(

OLS
) 10
5
, T
0
_
S
i ,0
IW(

S
i
, K
i
)

2
i
IG
_
10
4
2
,
1
2
_
R
i ,0
IG
_
10
4
2
,
1
2
_
(3.13)
where T
0
is a training sample.
0
are the diagonal elements of v
OLS
, which is the OLS
estimate of the VAR covariance matrix estimated on the training sample data. a
OLS
denotes
the o diagonal elements of v
OLS
. i =1, ... , n indexes the blocks of S.

S
i
is calibrated using
138
a
ols
. Specically,

S
i
is a diagonal matrix with the relevant elements of a
ols
multiplied by
10
3
. Note that factor priors are obtained using the least squares estimator employed by
Diebold et al. (2006) and thus e
t
() is obtained using the initial least squares estimates of
the factors.
3.4 Empirical results
3.4.1 Preliminary Empirical Results
For the empirical analysis, we use data on the following three variables : the output gap
12
, the
collateralized overnight call rate as an indicator of monetary policy ; and ination, measured
as annualized monthly changes in the consumer price index excluding fresh food. To obtain a
parsimonious specication, we choose a lag order of one, as the computations are otherwise
very burdensome.
We start the analysis by examining the estimated factors which are displayed in Figure
3.2 together with their empirical proxies dened above. In order to evaluate the uncertainty
characterizing the factors Figure 3.2 also plots the 16-th and 84-th quantile intervals of
the standard deviation. In our case, the error bands are almost indistinguishable from the
estimated series, indicating that factors are precisely estimated. However, when it comes
to their empirical counterparts, the t is at best satisfactory. Correlations between the level
factor and the slope factor and their empirical counterparts are 0.76 and 0.82, respectively.
Even worse, the estimated curvature does not t its empirical proxy well. This indicates that
level and slope factors can be approximated by their corresponding empirical counterparts
whereas approximations of the curvature factor are rather dicult. It is interesting to note
that estimated factors start to wander from their empirical counterparts from about 1992
12
The output-gap data, prepared and provided by the BOJ sta, are related to paper of Hara et al.
(2006)
3.4. Empirical results 139
and to catch up with them again from 1999. We point out that this nding is not dierent
from that of previous work as in Tam and Yu (2008).
Figure 3.2 Estimated factors and their empirical counterparts
Note :In the top panel, together with factors (blue lines), the graph shows the 68% probability
bands (red band) and empirical counterpart level, slope and curvature (dotted lines).
We recall that the macroeconomic interpretation to factors are as follows : the level
factor is typically associated with some measure of long-run expectations of ination
13
. The
slope factor contains information about the expected stance of monetary policy and thus is
a predictor of future economic activity and the curvature factor could also be informative
about the evolution of the economy.
The error volatility has gained increasing prominence in macro-nance models. Our
results support our choice of the heteroskedasticity assumption. Figure 3.3 plots the esti-
mated diagonal elements of the time-varying covariance matrix. The estimated stochastic
volatility of the structural shock both to factors and to the macroeconomics variables shows
that the time variation of volatilities has been signicant. Therefore, using the homoskedas-
ticity assumption would result in biases in the covariance matrix for the disturbances. The
time-varying volatilities of the level-and-curvature-factor shocks display a stable declining
path. The volatilities of the shocks to the slope and curvature factors drop to close to zero
during the ZIRP and QEMP periods. The stochastic volatility of the shocks to ination
13
It is particularly dicult to compare this factor with actual data, as information on long-run
expectations is not available for Japan. Ination-index 10-year bonds were only introduced in March
2004
140
declines during the same period. We observe some short-lived increases in shocks to the
output gap. The largest increase occurred around the global nancial crises of 2008. The
volatility of the shock to the call rate is typically of the evolution of Japanese monetary
policy. Call rate volatility declines up to 1995 to near zero and disappears afterwards. This
reects the decline of the call rate to 0.25% in 1995 and then within 0.25% in 1998 be-
fore the implementation of the ZIRP and the QEMP when the monetary policy instrument
shifted from the call rate to current account balances.
Figure 3.3 Estimated Standard deviation of the FAVAR residuals

The gures show posterior means of the estimated standard deviation of the structure shocks.
The solid lines show the median and the band areas represent the 68% bands.
3.4.2 Evidence on the expectations hypothesis (EH)
Empirical evidence has recently challenged the validation of the EH using Japanese data.
This issue is especially important for the Japanese economy because the principal channel
suggested by either ZIRP and QEMP is the expectation channel. One important channel
through which monetary policy works is long-term interest rates, shaping them so that in turn
3.4. Empirical results 141
Figure 3.4 Extracted expectation component
EH consistent yields (solid lines) together with their actual counterparts (dashed lines). Solid
lines show the median and the band areas represent the 68% bands.
they aect the level of economic activity. The expectation that a policy of low short-term
interest rates may be maintained for a substantial period of time will likely lower medium- to
long-term interest rates. The crucial link between a central banks instrument and long-term
interest rates is the EH of the yield curve theory. However, the empirical support for the EH
and the eectiveness of the policy commitment is rather mixed. Thornton (2004) applies a
bivariate VAR for long-term and short-term interest rates for the period from March 1981
to January 2003. He shows that the EH does not hold for the Japanese case. The author
also argues that one reason why EH fails is because the term premium varies over time.
In this chapter we review the validity of EH and the eects of the BOJs expectations
management on the JGB yield curve using a more exible approach. Using a time-varying
coecient model we can check whether EH consistent yields track actual yields well or not.
142
However, examining the evidence on EH requires separating expectations of future interest
rates from the term premium in the term structure. According to the EH, a long bond yield
is the average of the expected short-term rates :
y
t
()
EH
=
1

i =0
E
t
y
t+i
(1) (3.14)
On the other hand, the interest rate risk means that investors could require additional
compensation, and EH ignores this risk. The term premium, therefore, refers to this com-
pensation and any other deviation from the EH :
TP() = y
t
() y
t
()
EH
(3.15)
The implicit assumption behind our time-varying coecient VAR model is that agents review
their beliefs about uncertainty regarding ination, real activity and monetary policy at each
period. This assumption allows us to perform accurate predictions since it makes the model
more exible and more realistic. In addition, the term premium could vary with the business
cycle, as investors might be more risk-averse in recessions than in booms. Figure 3.4 provides
some selected maturity EH consistent yields together with their actual counterpart. The
theoretical yields tracking actual yields well, despite limited deviation which occurred between
1992 and 1998, indicate that the expectations hypothesis of the term structure of interest
rates is generally supported even during the QEMP period.
3.4.3 Time-varying term premium
According to the expectations hypothesis the term premium should simply be a function of
maturities, but not a function of time. However, the empirical investigation of the expecta-
tions theory has been unsuccessful, and the hypothesis has almost always been rejected. The
particular case of Japan conrms this result (Thornton (2004)). One possible explanation for
3.4. Empirical results 143
the empirical failure of the EH is the presence a time-varying term premium. Time-variation
in term premia might arise because of changes in market participants preferred risk aversion.
Moreover, a standard nding in the literature is that term premia are countercyclical ; they
seem to be highest during and immediately after recessions and lowest in booms (see, for
example, Cochrane and Piazzesi (2005)).
Figure 3.5 Estimated term premium

The solid lines show the median and the blue areas indicate the 16th and 84th percentiles of
the term premium.
Estimated term premia of some selected maturities are shown in Figure 3.5. There
appears to be a structural break in the behavior of the term premium for all maturities in
particular, volatility and magnitude have subsided since 1998. The particular episode of large
decrease in term premia between 1989 and 1992 may be explained by the heavy demand for
JGBs during the burst of the asset price bubble. A second episode of positive term premia
coincides with Japans economy stagnation from 1992 to 1999. When real GDP showed a
144
slight recovery phase in 1999, term premia declined and even started to uctuate around
zero over the rest of the sample for longer maturities. It is interesting to note that during
the quantitative easing period between 2001 and 2006 term premia decline to a lower level.
This could be due to the heavy demand for JGBs from the BOJ during that period. Another
explanation of the decline of term premia is that it could reect a global decline in risk
and term premia in developed and emerging market economies at that time. Shin and Shin
(2010) argue that global banking sector liabilities due to foreign creditors, called non-core
liabilities, increased rapidly from 2003 up to the nancial crisis in 2008. The authors show
that the banking sector relies more on funding from foreign creditors when global economic
conditions are favorable since private sectors deposits are not enough to sustain banks
desired balance sheets expansions. These additional funds are invested in a variety of assets
such as corporate and government securities, lowering yields and risk premia in these assets.
3.4.4 Empirical Results From the Macro-Finance Model
3.4.4.1 Are monetary policy shocks an important source of variation in the
yield curve ?
Given the focus of this chapter on the eects of monetary policy on the yield curve, the
discussion is focused on the decomposition of the unconditional variance of selected endo-
genous variables into contributions from the monetary policy shocks
14
.
The variance decomposition of the call rate
15
, shown in Figures 3.6, identies the
contribution of the monetary policy shock to variations in the yield curve and macroeconomic
14
The proportion of the unconditional variance accounted for by monetary policy shock is calculated
as the ratio of the unconditional variance due to the shock of interest rate and the total unconditional
variance. For more details see Bianchi et al. (2009).
15
Using call rate as monetary policy instrument during the QEMP may be subject to criticism
since the BOJ is targeting the monetary base. But, as demonstrated in Nagayasu (2004), the BOJ
had to practice a strong smoothing of interest rates in order to keep the short-term rate near to
zero during this period.
3.4. Empirical results 145
variables. The variance decomposition implies that a variation in the call rate is largely due
to independent monetary policy shocks. However, during the QEMP period (between 2001
and 2006) the contribution of the policy shock is negligible. This is consistent with the
change in monetary policy instrument from call rate to current account balances by the
adoption of the quantitative easing strategy. After 2006 the contribution of monetary policy
shocks to the policy interest rate becomes again important. The monetary policy rate shock
accounts at most for 20% of the uctuations in output until 1995, around the time when
the BOJ dropped its interest rate to 0.5%. The contribution has been negligible afterwards.
Innovations to call rate makes the largest contribution to the variance of ination during
the period between 1988 and the end of 1994, after which it veers to almost zero. A similar
patern emerges in its contribution to the uctuations in level, slope and curvature factors.
Figure 3.6 Unconditional variance - Call rate shock.

The gures show the variable contributions to the monetary policy rate variation. The solid line
denotes the median estimate, while the band indicates the 16% and the 84% quantiles of the
posterior distribution of variance decompositions.
146
Looking at the variance decomposition of ination shocks, displayed in Figure 3.10,
it is apparent that ination does not explain a large part of yield curve factors. The largest
contribution was in the early 1990s after which it became negligible. However, Figure 3.11
shows that variances in level and slope factors are signicantly explained by the variance in
the output gap until 1994, which accounts for approximately 30% and 40%, respectively.
This indicates that news about the future evolution of output might be more important for
the dynamics of the yield curve than inationary concerns for that period. This contribution
quickly shifted to very low levels over the period between 1995 and 2006 after which it rose to
similar levels as before. Altogether, these results suggest a negligible role of macroeconomic
variables in inuencing the yield curve during the long-lasting economic stagnation between
1995 and 2006.
Figure 3.7 complements the variance decomposition by displaying the impulse res-
ponse functions of the yield-curve factors and the macroeconomic variables to a monetary
policy shock
16
. We recall that the ultimate objective of the Japanese monetary policy is to
aect the yield curve level in order to stimulate the economy and to achieve low and stable
ination. More precisely, we focus on the eectiveness of the QEMP in aecting the long-
end of the yield curve. Before turning to the impulse responses following a surprise change
of the monetary policy rate, it is worth calling attention to the dicult interpretation of the
level factor reaction. Indeed, the success of monetary policy could be dened as a decrease in
the long-end of the yield curve via either expected short-term rates (policy-duration eect),
term premium (portfolio-rebalancing channel) or both of them. However, this represents only
an intermediate target in an attempt to generate economic recovery and to stop deation.
Therefore, the denition of a successful policy may be subject to criticism since economic
recovery, the nal goal of the BOJ, is expected to increase ination expectations and thus
16
For the sake of brevity, we only report the impulse response functions of the key variables to a
chock to monetary policy. Results of impulse response functions to shock to the slope, curvature,
output and ination are available upon request from the authors.
3.4. Empirical results 147
future short-term interest rates, which in turn will raise long-term interest rates. As argued
by Nagayasu (2004) monetary policy mechanisms take one to two years to achieve their
full eects. It seems appropriate, at the time of writing, to expect that the eectiveness of
QEMP, if any, would result in an increase of the level factor.
Figure 3.7 Impulse responses - Call rate shock
The gure shows the reactions of ination, output and level factor to a shock to the call rate over
25 months for three sample periods. The solid lines show the impulse responses implied by the
time-varying VAR following a rise by 100 basis point in call rate. The impulse responses in each
sub-sample are average of the impulse response in each month in that sample. The band areas
represent 68% error bands.
We report the responses for four sub-periods covering dierent main measures imple-
mented by the BOJ to stimulate the economy. The period between 1988 and 1994 represents
the period of high short term interest rate. The 1995-2001 period covers the so called tran-
sition period and the ZIRP period. The third and fourth sub-periods represent respectively
the QEMP period and after the exit from this strategy.
Consider the reaction of key variables to a monetary policy shock. Impulse response
148
functions of the level factor indicate a signicant and persistent increase of the level factor
during the 1988-1995 period. During that period short-term interest rates were still high
and had a potent dynamic eect on the level of the yield curve. However, during the period
of low interest rates, specially the QEMP period, the eect of the call rate on the level
factor becomes insignicant. This nding is consistent with previous research (Okina and
Shiratsuka (2004a), Nagayasu (2004) and others) and corroborates the idea that under
the zero lower bound, and according to the expectation hypothesis, the expected future
short-rate becomes equal to zero and then the long-term interest rate becomes equal to
the expected future term premium. In this case it becomes more dicult for the central
bank to inuence long-term interest rates. However, while statistically insignicant, the
decline in the level factor, which is more pronounced during the QEMP period, may reect
the strengthening credibility of the BOJ and thus the eectiveness of its policy. Indeed, as
argued in Diebold et al. (2006) and Bianchi et al. (2009), if monetary policy is credible the
level factor, everything being equal, should fall after a positive shock to call rate, because
the expectation of future ination declines. Since the BOJ commits itself during the QEMP
period to maintaining the short-term rates to a zero level, the decline in the level factor after
an increase in the call rate is by analogy equivalent to a rise in this factor to a monetary
policy expansion. This can be due to an expectation of an economic recovery and an ination
rise, indicating a monetary policy success.
The reaction of ination suggests a strong evidence of a price puzzle regardless of the
sample period
17
. The magnitude of the positive response is smaller over the high-interest-
rates period (1988-1995) ; the ination response veers to zero and becomes insignicant
more rapidly. The call rate shock still has a negative eect on the output gap with a more
persistent response over the periods before 1995 and after 2006.
17
The price puzzle problem can be due to the lack of information included in the VAR system as
explained in Bernanke et al. (2005), or to the small number of lags chosen given the model complexity
3.4. Empirical results 149
3.4.4.2 Is there a feedback eect from the yield curve to macroeconomic
variables ?
The variance decomposition of the level factor, shown in Figure 3.8, implies that the shock to
the level factor has the strongest impact on the long end of the yield curve. During the period
between 1995 and 2006 level shocks account for more than 90% of all level-factor variation.
This suggests a large amount of idiosyncratic variation in the long end of the yield curve
that is unrelated to macroeconomic fundamentals. This shock explains at most 5% of the
variance in ination and 10% of the variance in the output gap. The increasing contribution
Figure 3.8 Unconditional variance - level factor shock

The gures show the variable contributions to the level factor variation. The solid line denotes
the median estimate, while the band indicates the 16% and the 84% quantiles of the posterior
distribution of variance decompositions.
of this shock to the call rate since 1995 is matched by the contribution of the slope factor.
These results reect the increasing emphasis on long-term interest rates by the monetary
150
policy stance during the quantitative easing period. Taken together, nancial shocks do not
explain much of the variance in macroeconomic variables. These results are not surprising
in the case of Japan as the relation between the yield curve and macroeconomic variables
largely depends on nancial system conditions. As argued in Koo (2008) corporate sector
was busy paying down debt to improve its balance sheets, which were destroyed following
the asset price collapse. Then, corporate sector was reluctant to borrow new loans or to
issue new bonds for new investments. In this case real activity reacts less to the nancial
markets shocks, in particular to the yield curve shocks.
Figure 3.9 summarizes impulse response functions to an unexpected increase of the
level factor. A positive surprise change of the level factor indicates an increase of ination
during the period before 1995 and after 2006, but this eect remains insignicant. However,
by contrast with the conventional wisdom, in both periods of low interest rates, including
the QEMP period, ination decreases immediately in reaction to the level factor shock and
reverts towards zero. The level shock has a positive eect on output gap, although its impact
seems to exhibit time variation. During the period between 1995 and 2006 output gap reacts
signicantly reinforcing our nding that the contribution of macroeconomic variables to level
factor variation, if any, comes from output gap.
Figure 3.14 plots responses to an unexpected positive change of the slope factor. An
increase in the slope factor means a reduced spread between long-term and short-term bonds,
which indicates a monetary policy tightening and thus a decline of economic activity
18
. The
direction of the reaction of the output gap corroborates this view, while its responses are
short-lived and hardly signicant. The reaction of ination looks qualitatively similar to the
response to a level shock. An unexpected increase of the slope factor is followed by an initial
decrease of ination for all sub-samples. The call rate rises after the slope shock but its
18
Normally a decreasing yield curve slope announces an economic slowdown. But since the loading
of the slope factor in our model decreases with maturity and corresponds to the dierence between
short- and long-term yields, an increase in this factor corresponding therefore to a decrease in the
term spread.
3.5. Conclusion 151
reaction is not signicant except during the rst period of high interest rates. Altogether,
the impulse response functions reinforce results from variance decompositions that the yield
curve is not informative about macroeconomic variables when interest rates decline to a
very low level and specially during the quantitative easing period.
Figure 3.9 Impulse responses - Level shock
The gures show the reactions of ination, output gap and call rate to a shock to the level factor
over 25 months for three sample periods. The solid lines show the impulse responses implied by
the time-varying VAR following a rise by 100 basis point in call rate. The impulse responses in
each sub-sample are average of the impulse response in each month in that sample. The band
areas represent 68% error bands.
3.5 Conclusion
This chapter has examined the eects of the quantitative easing strategy in Japan on
the yield curve and the possible feed-back eect by applying a macro-nance model that
allows for time-varying parameters. This model provides maximum exibility in measuring
the eect of macroeconomic variables on the yield curve and vice-versa. It incorporates three
152
macroeconomic variables, namely the output gap, ination and the call rate, and three yield
curve factors which represent level, slope and curvature, summarizing the term structure
of interest rates. Before estimating this model, we constructed a database of zero-coupon
yields using Japanese government bond price data for the period between 1985 and 2009.
In order to estimate the spot rate curve we applied the smoothing B-splines method and
imposed non-negative spot restrictions in the estimation procedure. We thus avoid having
negative spot rates during the lower short-term interest rate period.
Contrary to the results of standard term-structure models with a time-invariant term
premium, our results show that the expectations hypothesis of the term structure of inter-
est rates is generally supported, even during the QEMP period. Empirical results from a
macro-nance model show that the relationship between the macroeconomic and nancial
variables has changed signicantly over time. There is hardly any relationship at the zero
lower bound on interest rates and deation, and especially during the quantitative easing
period. The structural decomposition of the yield curve into its macroeconomic components
shows that, by contrast with conventional wisdom, ination, activity and monetary policy
play a less prominent role in explaining the yield curve. They play no role at all particularly
during quantitative easing. The variance decomposition of the level factor indicates that the
main part of the variation in this factor comes from the yield curve factors, limiting the
contribution of macroeconomic variables. Conversely, the relative importance of yield curve
factors in the variation of ination is relatively small and even inexistent during quantitative
easing. A more pronounced eect of yield curve factors on the output gap is detected du-
ring the period of high interest rates, and this eect disappears during quantitative easing.
Moreover, the increasing contribution of the level factor in the variation of the call rate
during quantitative easing reects the increasing importance attributed by monetary policy
to long-term interest rates. These nding are corroborated by impulse response results.
As the objective of the BOJ during the quantitative-easing regime is to aect long-
3.5. Conclusion 153
term interest rates in order to stimulate the economy, a credible and successful expansionary
policy is a policy resulting in a rise in the level factor due to a recovery and ination expecta-
tions. This is equivalent to a decline in the level factor following a positive shock to the call
rate. Impulse response analysis shows that during this period, while statistically insignicant,
the level factor declines in response to a positive shock to the call rate. This may be due to
an expectation of an ination rise and an economic recovery, indicating increasing credibility
leading to monetary policy success.
On the other hand, the insignicant eect of the call rate on the level factor can
be explained by the fact that, at the zero lower bound of interest rates, expected future
short-term rates are almost zero and long-term interest rates become largely determined
by the forward term premium, which the BOJ nds dicult to inuence. A focus on term
premium and expectation components would better capture the eect of monetary policy on
yield curve. In addition, according to Bernanke et al. (2005), simple variables alone cannot
represent economic concepts like activity or price. The alternative is then to use a model
with macroeconomic factors summarizing a larger set of macroeconomic variables. These
issues could well benet from research attention in the future.
154
Figures
Figure 3.10 Variance decomposition due to ination
Unconditional variance due to the ination shock. The graph shows the percentage of the variance
of each of the variables that explained by the ination shock.
3.5. Conclusion 155
Figure 3.11 Variance decomposition due to the output gap
Unconditional variance due to the output gap shock. The graph shows the percentage of the
variance of each of the variables that explained by the output gap shock.
156
Figure 3.12 Variance decomposition due to slope factor
Unconditional variance due to the slope factor. The graph shows the percentage of the variance
of each of the variables that explained by the slope factor shock.
3.5. Conclusion 157
Figure 3.13 Variance decomposition due to curvature
Unconditional variance due to the curvature factor. The graph shows the percentage of the variance
of each of the variables that explained by the curvature shock.
158
Figure 3.14 Impulse response functions to slope shock
The gures show the reactions of level factor, ination, output and call rate to a shock to the
slope factor over 25 months for three sample periods. The solid lines show the impulse responses
implied by the time-varying VAR following a rise by 100 basis point in call rate. The impulse
responses in each sub-sample are average of the impulse response in each month in that sample.
The band areas represent 68% error bands.
Conclusion gnrale
La dcision de recourir des politiques montaires non conventionnelles an de faire face
la crise nancire de 2008 fait lunanimit au sein des principales banques centrales ; ces
politiques visent stimuler les conomies dans lesquelles les taux dintrt atteignent leur
niveau plancher zro. La Fed, en dcembre 2008, a men une politique dite dassouplisse-
ment des conditions de crdits en se substituant aux banques et au march nancier pour
nancer directement lconomie. Dans le mme temps, la banque dAngleterre, ainsi que la
BCE, ont adopt des politiques semblables consistant acheter massivement des actifs
long terme, jusqu atteindre une cible de taille du passif de leurs bilans. Plus rcemment,
en octobre 2010, la banque du Japon a dcid de reprendre sa politique dassouplissement
quantitatif mene entre 2001 et 2006 sous une autre forme (Comprehensive Monetary Ea-
sing), en mettant davantage laccent sur lachat des actifs long terme. Face la lenteur
de la reprise, la Fed dcide en novembre 2010 denclencher une deuxime tape dans lap-
plication de sa politique non conventionnelle en annonant lachat des bons long terme du
Trsor amricain, dans le but de baisser les taux dintrt de long terme. Ces interventions
des banques centrales nous ont incit questionner dans cette thse lecacit de ce type
de politique montaire au regard de lexprience japonaise dassouplissement quantitatif de
2001-2006. Les trois chapitres du prsent travail apportent des lments de rponse aux
interrogations suivantes :
La politique dassouplissement quantitatif, telle quapplique par la BOJ, tait-elle
ecace pour stimuler lconomie japonaise et lextraire dune situation de dpression
et de dation ? Daprs la BOJ, cette stratgie naurait fait que stabiliser les marchs
159
160
nanciers et empcher le prolongement de la spirale dationniste, sans pour autant
relancer lconomie. Ce constat signie-t-il que la BOJ aurait d injecter davantage
de rserves aux banques ? Ou bien aurait-elle d maintenir la mme politique plus
longtemps ?
Par quels canaux les eets de cette politique peuvent-ils tre transmis lconomie
relle ? Lapproche neo-wickselienne met laccent sur le canal des anticipations. Elle
considre que lconomie japonaise est tombe dans une situation de trappe liquidit.
Elle suggre donc que la banque centrale sengage explicitement maintenir le taux
dintrt nominal de court terme un niveau bas pendant une priode signicative, an
de pouvoir agir sur la courbe des taux. Dautre part, lapproche montariste se focalise
sur le canal du rquilibrage de portefeuille. Laugmentation de la base montaire par
lachat des titres long terme a pour eet direct de baisser les rendements de ces
actifs. De plus, tant donne limparfaite substituabilit entre la monnaie et les actifs
non montaires, la liquidit additionnelle fournie par la banque centrale pousse les
agents privs rquilibrer leurs portefeuilles par lachat dactifs non montaires, et
baisse ainsi leurs rendements.
Pour tre en mesure de rpondre ces questionnements, nous nous sommes eorcs dans
ce travail de thse dutiliser les avances les plus rcentes de lconomtrie de la politique
montaire pour les appliquer lexprience japonaise dassouplissement quantitatif. Les co-
nomtres ont pris conscience du fait que les banques centrales fondent leurs dcisions sur une
multitude dindicateurs conomiques alors que les estimations conomtriques des dtermi-
nants et des consquences de ces dcisions utilisent des modles vectoriels autorgressifs
(VAR), o seul un trs petit nombre de variables agrges est pris en compte, et quelles
sourent ds lors de nombreux biais (nigme de prix, non-neutralit long terme). Dautre
part, il parat vident que faire lexamen de la politique montaire japonaise sans tenir compte
des changements de rgimes aboutit des rsultats biaiss, et mne des recommandations
Conclusion gnrale 161
errones.
Le premier chapitre de cette thse a t consacr lexamen de la capacit de las-
souplissement quantitatif stimuler lactivit et sortir le Japon de la spirale dationniste
qui sest installe suite au dgonement de la bulle spculative. Ce travail complte la lit-
trature existante en combinant la mthodologie de Markov-Switching VAR avec celle de
lanalyse factorielle en un modle appel MS-FAVAR. Ce modle a non seulement rendu
possible lintroduction dun grand nombre de variables dans lanalyse mais a aussi permis de
prendre en compte les changements de rgimes qui caractrisent la structure de lconomie
japonaise de ces deux dernires dcennies. Des facteurs communs ont t extraits des va-
riables lies lactivit, aux prix et aux taux dintrt et ont t introduits dans le modle
MS-VAR; ils reprsentent des mesures de concepts conomiques gnraux comme lactivit
relle et les prix.
Le modle prsent dans le chapitre 1 nous a permis didentier et de dater le change-
ment de rgime de la politique montaire. Le deuxime rgime, dtect en 1999, correspond
la priode de politique du taux dintrt zro et celle de lassouplissement quantitatif. De
plus, labsence du problme dnigme de prix et de la non-neutralit de la monnaie dans nos
rsultats montre que la prise en compte du maximum dinformation dans lanalyse donne
des rsultats en accord avec les prdictions thoriques. Ce chapitre a montr en particulier
que leet de lassouplissement quantitatif sur lactivit et sur les prix, bien que transitoire,
est signicatif. Ce caractre transitoire peut expliquer pourquoi la BOJ juge que limpact de
cette politique sur lactivit relle et sur les prix reste modeste. Durant la majeure partie de
la priode dassouplissement quantitatif, jusquen 2005, les rmes japonaises taient rti-
centes avoir recours des crdits ou mettre des obligations pour nancer de nouveaux
investissements, car elles taient proccupes par le paiement des dettes accumules suite
au dgonement de la bulle nancire. La politique dassouplissement quantitatif ayant t
stoppe juste aprs la n du paiement de ces dettes par les rmes, nous nous autorisons
162
penser que le prolongement de cette politique dassouplissement quantitatif aurait t
protable, et nous en dduisons donc que la BOJ en aurait rcolt les fruits si elle lavait
maintenu plus longtemps.
Lobjectif du deuxime chapitre tait lidentication des canaux de transmission et
la mesure de leur ampleur. A laide dun modle FAVAR avec paramtres volutifs dans le
temps (TVP-FAVAR) nous avons pu analyser la priode dassouplissement quantitatif dune
manire prcise. Les rsultats obtenus confortent ceux du premier chapitre en dmontrant
lecacit de lassouplissement quantitatif sur lactivit et sur les prix. Le canal de rqui-
librage de portefeuille savre jouer un rle important dans la transmission des eets de
lassouplissement quantitatif. En particulier, la raction positive, la fois de la consomma-
tion et de lindice boursier, montre que limpact de laugmentation de la base montaire a
t transmis par lintermdiaire de leet de richesse. Dautre part cet eet de laugmen-
tation de la base montaire sur les anticipations des agents privs, bien que transitoire,
conforte la dclaration de la BOJ que lengagement maintenir les taux de faibles niveaux
a pu stopper la spirale dationniste, sans pour autant gnrer de pression inationniste
consquente.
Lapport du chapitre 3 est le prolongement de lanalyse du canal des anticipations
dans le cadre dun modle macro-nance reliant les variables macroconomiques la courbe
des taux. Nous sommes dans un premier temps parvenus estimer la courbe des taux zro-
coupon en utilisant des prix dobligations dtat japonais. Puis, un modle macro-nance
avec paramtres volutifs dans le temps a t employ an danalyser les interactions entre
la structure par terme des taux dintrt et les variables macroconomiques. La prise en
compte de lvolution des anticipations des agents privs a permis de dmontrer la validit
de lhypothse danticipations rationnelles, sans laquelle le canal des anticipations ne peut
gure fonctionner. Ltude de la dcomposition de la variance et des fonctions dimpulsion
a montr la faible interaction entre les variables macroconomiques et la courbe des taux
Conclusion gnrale 163
durant la priode dassouplissement quantitatif. Quant lanalyse de leet de la politique
montaire, la raction du niveau de la structure par terme suite un choc sur le taux
dintrt de court terme a mis en vidence le renforcement de la crdibilit de la BOJ dans
son engagement maintenir des taux dintrt de faibles niveaux. Ce rsultat conforte
ainsi les conclusions concernant le canal des anticipations trouves dans le chapitre 2.
Lapport de cette prsente thse centre autour du cas du Japon est ainsi de mettre
en lumire des enseignements utiles la comprhension des politiques montaires non
conventionnelles rcemment appliques par les principales banques centrales. Le travail ra-
lis montre la capacit de la politique dassouplissement quantitatif inuencer lactivit
et les prix, ecacit conditionne par une dure susante dapplication, et met aussi en
vidence la complmentarit entre le canal de rquilibrage de portefeuille et celui des anti-
cipations.
Le choix de la variable de politique montaire demeure problmatique pour le cas
du Japon. Partant de lide que la banque centrale peut agir sur le taux dintrt ainsi
que sur le march des rserves, Bernanke and Mihov (1998) dveloppent un modle VAR
semi-structurel qui extrait les informations relatives la politique montaire des donnes
concernant les rserves bancaires et les taux dintrt. Ce modle autorise les relations entre
les variables macroconomiques dans le systme non contraint mais il impose des restrictions
didentication contemporaines sur lensemble des variables relatives au march de rserves
des banques commerciales. Pour identier ce modle les auteurs sappuient sur les modalits
dintervention de la banque centrale et valuent les dirents indicateurs de lorientation de la
politique montaire qui dcoulent des modalits dintervention en testant les restrictions de
suridentication imposes. Dans le cas du Japon Il apparaiterait ainsi intressant demployer
le modle TVP-FAVAR en prenant en compte direntes variables de politique montaire,
et en vitant ainsi de faire lhypothse quune variable unique constitue le meilleur indicateur
de la politique montaire.
164
De plus, ltude des eets de la politique montaire nationale dans les direntes
rgions a montr que, dans des conomies de grande taille, les eets agrgs peuvent cacher
une grande diversit rgionale dans la rpartition activit/ination (sur les Etats amricains :
Carlino et DeFina, 1998 ; Owyang et Wall, 2003). Il pourrait tre fructueux destimer des
eets rgionaux de la politique montaire japonaise an de dterminer dans quelle mesure la
politique dassouplissement quantitatif sest transmise de manire dirente entre les rgions
japonaises, aussi bien pour la production que pour les prix.
Bibliographie
Andrs, Javier, Lpez-Salido, J. David, and Nelson, Edward. 2004. Tobins Imperfect Asset
Substitution in Optimizing General Equilibrium. Journal of Money, Credit and Banking,
36(4).
Ang, Andrew, and Piazzesi, Monika. 2003. A no-arbitrage vector autoregression of term
structure dynamics with macroeconomic and latent variables. Journal of Monetary Eco-
nomics, 50(4), 745 787.
Auerbach, Alan J., and Obstfeld, Maurice. 2004. Monetary and Fiscal Remedies for Dea-
tion. American Economic Review, 94(2), 71 75.
Auerbach, Alan J., and Obstfeld, Maurice. 2005. The Case for Open-Market Purchases in
a Liquidity Trap. American Economic Review, 95(1), 110137.
Baba, Naohiko, Nishioka, Shinichi, Oda, Nobuyuki, Shirakawa, Masaaki, Ueda, Kazuo, and
Ugai, Hiroshi. 2005. Japans Deation, Problems in the Financial System, and Monetary
Policy. Monetary and Economic Studies, 23(1), 47111.
Bagliano, Fabio C., and Favero, Carlo A. 1998. Measuring monetary policy with VAR
models : An evaluation. European Economic Review, 42(6), 10691112.
165
166
Bai, Jushan. 2004. Estimating cross-section common stochastic trends in nonstationary
panel data. Journal of Econometrics, 122(1), 137183.
Bai, Jushan, and Ng, Serena. 2002. Determining the Number of Factors in Approximate
Factor Models. Econometrica, 70(1), 191221.
Bai, Jushan, and Ng, Serena. 2004. A PANIC Attack on Unit Roots and Cointegration.
Econometrica, 72(4), 11271177.
Bayoumi, Tamim. 2001. The morning after : explaining the slowdown in Japanese growth
in the 1990s. Journal of International Economics, 53(2), 241259.
Belviso, Francesco, and Milani, Fabio. 2006. Structural Factor-Augmented Vars (SFAVARs)
and the Eects of Monetary Policy. Topics in Macroeconomics, 6(3), 14431443.
Benhabib, Jess, Schmitt-Grohe, Stephanie, and Uribe, Martin. 2001. The Perils of Taylor
Rules. Journal of Economic Theory, 96(1-2), 4069.
Bernanke, Ben, Boivin, Jean, and Eliasz, Piotr. 2005. "Measuring the Eects of Monetary
Policy : A Factor-augmented Vector Autoregressive (FAVAR) Approach". The Quarterly
Journal of Economics, 120(1), 387422.
Bernanke, Ben S. 2000. Comment on Americas Historical Experience with Low Ination.
Journal of Money, Credit and Banking, 32(4), 99497.
Bernanke, Ben S. 2003. Some Thoughts on Monetary Policy in Japan. Remarks before the
Japan Society of Monetary Economics, Tokyo, Japan. May 31.
Bernanke, Ben S, and Blinder, Alan S. 1992. The Federal Funds Rate and the Channels of
Monetary Transmission. American Economic Review, 82(4), 90121.
Bernanke, Ben S., and Mihov, Ilian. 1998. The liquidity eect and long-run neutrality.
Carnegie-Rochester Conference Series on Public Policy, 49(1), 149194.
Bibliographie 167
Bernanke, Ben S., and Reinhart, Vincent R. 2004. Conducting Monetary Policy at Very
Low Short-Term Interest Rates. American Economic Review, 94(2), 85 90.
Bernanke, Ben S., Reinhart, Vincent R., and Sack, Brian P. 2004. Monetary policy alter-
natives at the zero bound : an empirical assessment. Tech. rept. 2004-48. Board of
Governors of the Federal Reserve System (U.S.). Finance and Economics Discussion
Series.
Bianchi, Francesco, Mumtaz, Haroon, and Surico, Paolo. 2009. The great moderation of the
term structure of UK interest rates. Journal of Monetary Economics, 56(6), 856871.
BIS. 2005. Zero-coupon yield curves : technical documentation. Bank for International
Settlements, technical documentation, No 25, October.
Blanchard, Olivier, and Perotti, Roberto. 2002. An Empirical Characterization Of The
Dynamic Eects Of Changes In Government Spending And Taxes On Output. The
Quarterly Journal of Economics, 117(4), 13291368.
Blanchard, Olivier Jean, and Quah, Danny. 1989. The Dynamic Eects of Aggregate De-
mand and Supply Disturbances. American Economic Review, 79(4), 655.
Boivin, J., and Giannoni, M. 2007. DSGE Models in a Data-Rich Environment. Tech. rept.
Braun, R. Anton, and Shioji, Etsuro. 2006. Monetary Policy and the Term Structure of
Interest Rates in Japan. Journal of Money, Credit and Banking, 38(1), 141162.
Braun, R. Anton, and Waki, Yuichiro. 2006. Monetary Policy During JapanS Lost Decade.
The Japanese Economic Review, 57(2), 324344.
Cargill, Thomas F. 2002. Japan passes again on fundamental nancial reform. FRBSF
Economic Letter, Sep.
168
Cargill, Thomas F., Hutchison, Michael M., and Ito, Takatoshi. 1995. Lessons from nancial
crisis : the Japanese case. Proceedings, 101109.
Cargill, Thomas F., Hutchison, Michael M., and Ito, Takatoshi. 1997. The Political Economy
of Japanese Monetary Policy. MIT Press Books, vol. 1, no. 0262032473. The MIT Press.
Cargill, Thomas F., Hutchison, Michael M., and Ito, Takatoshi. 2001. Financial Policy and
Central Banking in Japan. MIT Press Books, vol. 1, no. 0262032856. The MIT Press.
Carlstrom, Charles T., Fuerst, Timothy S., and Paustian, Matthias. 2009. Monetary policy
shocks, Choleski identication, and DNK models. Journal of Monetary Economics, 56(7),
10141021.
Carter, C. K., and Kohn, R. 1994. On Gibbs sampling for state space models. Biometrika,
81(3), 541553.
Chinn, Menzie D. 2006. The (partial) rehabilitation of interest rate parity in the oating
rate era : Longer horizons, alternative expectations, and emerging markets. Journal of
International Money and Finance, 25(1), 721.
Cho, Jin Seo, and White, Halbert. 2007. Testing for Regime Switching. Econometrica,
75(6), 16711720.
Christensen, Jens H. E., Diebold, Francis X., and Rudebusch, Glenn D. 2009. An arbitrage-
free generalized NelsonSiegel term structure model. Econometrics Journal, 12(3), C33
C64.
Christiano, Lawrence J., Eichenbaum, Martin, and Evans, Charles L. 1999. Monetary policy
shocks : What have we learned and to what end ? Chap. 2, pages 65148 of : Taylor,
J. B., and Woodford, M. (eds), Handbook of Macroeconomics, vol. 1. Elsevier.
Bibliographie 169
Clarida, Richard, and Gali, Jordi. 1994 (Feb.). Sources of Real Exchange Rate Fluctuations :
How Important are Nominal Shocks ? NBER Working Papers 4658. National Bureau of
Economic Research, Inc.
Cochrane, John H., and Piazzesi, Monika. 2005. Bond Risk Premia. American Economic
Review, 95(1), 138160.
Cogley, Timothy, Morozov, Sergei, and Sargent, Thomas J. 2005. Bayesian fan charts for
U.K. ination : Forecasting and sources of uncertainty in an evolving monetary system.
Journal of Economic Dynamics and Control, 29(11), 1893 1925.
Collard, Fabrice, and Fve, Patrick. 2008. Modles VAR ou DSGE : que choisir ? Economie
& prvision, 2-3(183-184), 153174.
Collin-Dufresne, Pierre. 2001. The Determinants of Credit Spread Changes. Journal of
Finance, 56(6), 21772207.
Davidson, Russell, and MacKinnon, James G. 2000. Bootstrap Tests : How Many Boots-
traps ? Econometric Reviews, 19(1), 5568.
Del Negro, Marco, Schorfheide, Frank, Smets, Frank, and Wouters, Rafael. 2005 (Jan.).
On the Fit and Forecasting Performance of New Keynesian Models. CEPR Discussion
Papers 4848. C.E.P.R. Discussion Papers. http://ideas.repec.org/p/cpr/ceprdp/
4848.html.
Diebold, Francis X., and Li, Canlin. 2006. Forecasting the term structure of government
bond yields. Journal of Econometrics, 130(2), 337364.
Diebold, Francis X., Piazzesi, Monika, and Rudebusch, Glenn D. 2005. Modeling Bond
Yields in Finance and Macroeconomics. American Economic Review, 95(2), 415420.
170
Diebold, Francis X., Rudebusch, Glenn D., and Borag[caron]an Aruoba, S. 2006. The
macroeconomy and the yield curve : a dynamic latent factor approach. Journal of Eco-
nometrics, 131(1-2), 309338.
Diebold, Francis X., Li, Canlin, and Yue, Vivian Z. 2008. Global yield curve dynamics and
interactions : A dynamic Nelson-Siegel approach. Journal of Econometrics, 146(2), 351
363.
Doz, Catherine, Giannone, Domenico, and Reichlin, Lucrezia. 2006. A Two-step estimator
for large approximate dynamic factor models based on Kalman ltering. THEMA Working
Papers. http://ideas.repec.org/p/ema/worpap/2006-23.html.
Eggertsson, Gauti B. 2003 (Apr.). How to Fight Deation in a Liquidity Trap : Committing
to Being Irresponsible. IMF Working Papers 03/64. International Monetary Fund.
Eggertsson, Gauti B., and Woodford, Michael. 2003. The Zero Bound on Interest Rates and
Optimal Monetary Policy. Brookings Papers on Economic Activity, 34(2003-1), 139235.
Eggertsson, Gauti B., and Woodford, Michael. 2004. Policy Options in a Liquidity Trap.
American Economic Review, 94(2), 76 79.
Eggertsson, Gauti B., and Woodford, Michael. 2006. Optimal Monetary and Fiscal Policy
in a Liquidity Trap. Pages 75144 of : NBER International Seminar on Macroeconomics
2004. NBER Chapters. National Bureau of Economic Research, Inc.
Ehrmann, Michael, Ellison, Martin, and Valla, Natacha. 2003. Regime-dependent impulse
response functions in a Markov-switching vector autoregression model. Economics Let-
ters, 78(3), 295299.
Evans, Charles L., and Marshall, David A. 2007. Economic determinants of the nominal
treasury yield curve. Journal of Monetary Economics, 54(7), 19862003.
Bibliographie 171
Faust, Jon, and Leeper, Eric M. 1994. When do long-run identifying restrictions give reliable
results ? Tech. rept.
Favero, Carlo A, Flabbi, Luca, and Giavazzi, Francesco. 1999. The Transmission Mechanism
of Monetary Policy in Europe : Evidence from Banks Balance Sheets. CEPR Discussion
Papers, Nov.
Fernndez-Villaverde, Jess. 2010. The econometrics of DSGE models. International Review
of Economics, 1(1), 349.
Fisher, Mark, Nychka, Douglas, and Zervos, David. 1995. Fitting the term structure of
interest rates with smoothing splines. Finance and Economics Discussion Series 95-1.
Board of Governors of the Federal Reserve System (U.S.).
Forni, M., Hallin, M., Lippi, M., and Reichlin, L. 2005. The generalized dynamic factor
model : one-sided estimation and forecasting. Journal of the American Statistical Asso-
ciation, 100, 830840.
Forni, Mario, and Gambetti, Luca. 2010. The dynamic eects of monetary policy : A
structural factor model approach. Journal of Monetary Economics, 57(2), 203 216.
Fujiwara, Ippei. 2006. Evaluating monetary policy when nominal interest rates are almost
zero. Journal of the Japanese and International Economies, 20(3), 434453.
Gagnon, Joseph E., Raskin, Matthew, Remache, Julie, and Sack, Brian P. 2010 (March).
Large-Scale Asset Purchases by the Federal Reserve : Did They Work ? Sta Report No
441. New York : Federal Reserve Bank of New York.
Ganduuml ;rkaynak, Refet S., Sack, Brian, and Swanson, Eric. 2005. The Sensitivity of Long-
Term Interest Rates to Economic News : Evidence and Implications for Macroeconomic
Models. American Economic Review, 95(1), 425436.
172
Garcia, Rene. 1998. Asymptotic Null Distribution of the Likelihood Ratio Test in Markov
Switching Models. International Economic Review, 39(3), 76388.
Genshiro, Kitagawa. 1996. Monte Carlo lter and smoother for non-Gaussian nonlinear
state space models. Journal of Computational and Graphical Statistics, 5(1), 125.
Gerlach, R., Carter, C., and Kohn, R. 2000. Ecient Bayesian inference for dynamic mixture
models. Journal of the American Statistical Association, 75, 819828.
Giordani, Paolo, and Kohn, Robert. 2008. Ecient Bayesian Inference for Multiple Change-
Point and Mixture Innovation Models. Journal of Business & Economic Statistics, 26(Ja-
nuary), 6677.
Girardin, Eric, and Lyons, Richard K. 2008. Does Intervention Alter Private Behaviour ?
Working Papers. http://faculty.haas.berkeley.edu/LYONS/wp.html.
Girardin, Eric, and Moussa, Zakaria. 2010. Quantitative easing works : Lessons from the
unique experience in Japan 2001-2006. GREQAM working paper, 2010-02. http://
www.greqam.fr/spip.php?article7099&lang=fr.
Goodfriend, Marvin. 2000. Overcoming the Zero Bound on Interest Rate Policy. Journal of
Money, Credit and Banking, 32(4), 10071035.
Gorovoi, Viatcheslav, and Linetsky, Vadim. 2004. Blacks Model of Interest Rates as Op-
tions, Eigenfunction Expansions and Japanese Interest Rates. Mathematical Finance,
14(1), 4978.
Gosselin, Marc-Andr, and Tkacz, Greg. 2001. Evaluating Factor Models : An Application
to Forecasting Ination in Canada. Bank of Canada. Working Papers. http://ideas.
repec.org/p/bca/bocawp/01-18.html.
Bibliographie 173
Guerrero, Federico, and Parker, Elliott. 2010. The Eect of Government Purchases on
Economic Growth in Japan. University of Nevada, Reno, Department of Economics,
Working Papers. 10-004.
Hamilton, James D. 1989. A New Approach to the Economic Analysis of Nonstationary
Time Series and the Business Cycle. Econometrica, 57(2), 35784.
Hamilton, James Douglas. 1994. Time Series Analysis. 1st edn. Princeton University Press.
ISBN 0691042896.
Hansen, Bruce E. 1992. The Likelihood Ratio Test under Nonstandard Conditions : Testing
the Markov Switching Model of GNP. Journal of Applied Econometrics, 7(S), S6182.
Hara, N., Hirakata, N., Inomata, Y., S., Ito, Kawamoto, T., Kurozumi, T., Minegishi, M.,
and Takagawa, I. 2006. The New Estimates of Output Gap and Potential Growth Rate.
Bank of Japan, Research and Statistics Department. Working Papers.
Hattori, Masazumi, and Shin, Hyun Song. 2007. The Broad Yen Carry Trade. IMES
Discussion Paper Series, Oct. no 07-E-19.
Hrdahl, Peter, Tristani, Oreste, and Vestin, David. 2006. A joint econometric model of
macroeconomic and term-structure dynamics. Journal of Econometrics, 131(1-2), 405
444.
Ichiue, H., Kurozumi, T., and Sunakawa, T. 2008. Ination Dynamics and Labor Adjust-
ments in Japan : A Bayesian DSGE Approach. Working Paper Series 08-E-9. Bank of
Japan.
Ihori, Toshihiro, Nakazato, Toru, and Kawade, Masumi. 2003. Japans Fiscal Policies in the
1990s. The World Economy, 26(3), 325338.
174
Inoue, Tomoo, and Okimoto, Tatsuyoshi. 2008. Were there structural breaks in the eects
of Japanese monetary policy ? Re-evaluating policy eects of the lost decade. Journal of
the Japanese and International Economies, 22(3), 320342.
Ito, Takatoshi. 2006. Japanese monetary policy : 1998-2005 and beyond. Pages 105132
of : for International Settlements, Bank (ed), Monetary policy in Asia : approaches and
implementation. BIS Papers chapters, vol. 31. Bank for International Settlements.
Ito, Takatoshi, and Mishkin, Frederic S. 2006. Two Decades of Japanese Monetary Policy
and the Deation Problem. Pages 131202 of : Monetary Policy under Very Low Ination
in the Pacic Rim, NBER-EASE, Volume 15. NBER Chapters. National Bureau of
Economic Research, Inc.
Ito, Takatoshi, and Yabu, Tomoyoshi. 2007. What prompts Japan to intervene in the Forex
market ? A new approach to a reaction function. Journal of International Money and
Finance, 26(2), 193212.
Jung, Taehun, Teranishi, Yuki, and Watanabe, Tsutomu. 2005. Optimal Monetary Policy
at the Zero-Interest-Rate Bound. Journal of Money, Credit and Banking, 37(5), 81335.
Juselius, Katrina. 2006. The Cointegrated VAR Approach : Methodology and Applications.
Draft, forthcoming textbook on the economometrics of the cointegrated vector autore-
gression model. Economics Institute, University of Copenhagen.
Kamada, K., and Sugo, T. 2006. Evaluating the Japanese Monetary Policy under the Non-
Negativity Constraint on Nominal Short-Term Interests Rates. Bank of Japan Working
Paper Series. forthcoming, Sep.
Kamada, Koichiro, and Masuda, Kazuto. 2001. Eects of Measurement Error on the Output
Gap in Japan. Monetary and Economic Studies, 19(2), 10954.
Bibliographie 175
Kim, Sangjoon, Shephard, Neil, and Chib, Siddhartha. 1998. Stochastic Volatility : Likelihood
Inference and Comparison with ARCH Models. Review of Economic Studies, 65(3), 361
93.
Kim, Soyoung, and Roubini, Nouriel. 2000. Exchange rate anomalies in the industrial coun-
tries : A solution with a structural VAR approach. Journal of Monetary Economics, 45(3),
561 586.
Kimura, Takeshi, Kobayashi, Hiroshi, Muranaga, Jun, and Ugai, Hiroshi. 2003. The eect of
the increase in the monetary base of Japans economy at zero interest rates : an empirical
analysis. Tech. rept. in Monetary Policy in a Changing Environment.
Koo, Richard C. 2008. The Holy Grail of Macroeconomics : Lessons from Japans Great
Recession. John Wiley and Sons (Asia) Pte. Ltd. ISBN 974-0470-82387-3.
Koop, Gary, and Korobilis, Dimitris. 2009 (Sept.). Bayesian Multivariate Time Series Me-
thods for Empirical Macroeconomics. MPRA Paper 20125. University Library of Munich,
Germany. http://ideas.repec.org/p/pra/mprapa/20125.html.
Krolzig, Hans-Martin. 1997. Markov-Switching Vector Autoregressions : Modelling, Statis-
tical Inference, and Application to Business Cycle Analysis (Lecture Notes in Economics
and Mathematical Systems). Springer. ISBN 3540630732.
Krugman, Paul. 2000. Thinking About the Liquidity Trap. Journal of the Japanese and
International Economies, 14(4), 221237.
Kuttner, Kenneth N., and Posen, Adam S. 2001. The Great Recession : Lessons for Ma-
croeconomic Policy from Japan. Brookings Papers on Economic Activity, 32(2001-2),
93186.
176
Kwon, Eunkyung. 1998. Monetary Policy, Land Prices, and Collateral Eects on Economic
Fluctuations : Evidence from Japan. Journal of the Japanese and International Economies,
12(3), 175 203.
Lagana, Gianluca. 2004. Measuring monetary policy in the UK : a factor augmented vector
autoregressive approach. Money Macro and Finance (MMF) Research Group Conference
2004, Sept. http://ideas.repec.org/p/mmf/mmfc04/64.html.
Lee, Junsoo, and Strazicich, Mark C. 2003. Minimum Lagrange Multiplier Unit Root Test
with Two Structural Breaks. The Review of Economics and Statistics, 85(4), 10821089.
Leeper, Eric M., Sims, Christopher A., and Zha, Tao. 1996. What Does Monetary Policy
Do ? Brookings Papers on Economic Activity, 27(1996-2), 178.
Litterman, JC., and Scheinkman, J.A. 1991. Common Factors Aecting Bond Returns.
Journal of Fixed Income, 1(1), 5461.
Lucas, Robert Jr. 1976. Econometric policy evaluation : A critique. Carnegie-Rochester
Conference Series on Public Policy, 1(1), 1946.
Maeda, Eiji, Fujiwara, Bunya, Mineshima, Aiko, and Taniguchi, Ken. 2005. Japans Open
Operations under the Quantitative Easing Policy. Bank of Japan Working paper Series.
No.05-E-3.
Marcellino, Massimiliano, Favero, Carlo A., and Neglia, Francesca. 2005. Principal compo-
nents at work : the empirical analysis of monetary policy with large data sets. Journal of
Applied Econometrics, 20(5), 603620.
McCallum, Bennett T. 2000. Theoretical Analysis Regarding a Zero Lower Bound on No-
minal Interest Rates. Journal of Money, Credit and Banking, 32(4), 870904.
Bibliographie 177
Mehrotra, Aaron. 2009. The case for price level or ination targetingWhat happened to
monetary policy eectiveness during the Japanese disination ? Japan and the World
Economy, 21(3), 280291.
Mehrotra, Aaron N. 2007. Exchange and interest rate channels during a deationary era
Evidence from Japan, Hong Kong and China. Journal of Comparative Economics, 35(1),
188 210.
Metzler, Allan. 1995. Monetary, Credit and (Other) Transmission Processes : A Monetarist
Perspective. Journal of Economic Perspectives, 9(4), 4972.
MIHIRA, T., YAMASAWA, N., SEITANI, H., and SAITO, J. 2006. Was Quantitative Easing
Policy Eective ? An Empirical Analysis of the Monetary Policy in Japan during 2001-
2006. ESRI International Conference 2006, Sept.
Miyao, Ryuzo. 2000. The Role of Monetary Policy in Japan : A Break in the 1990s ? Journal
of the Japanese and International Economies, 14(4), 366384.
Miyao, Ryuzo. 2005. Use Of The Money Supply In The Conduct Of JapanS Monetary Po-
licy : Re-Examining The Time-Series Evidence. The Japanese Economic Review, 56(2),
165187.
Moench, Emanuel. 2008. Forecasting the yield curve in a data-rich environment : A no-
arbitrage factor-augmented VAR approach. Journal of Econometrics, 146(1), 2643.
Nagayasu, Jun. 2004. The Term Structure of Interest Rates and Monetary Policy during a
Zero Interest Rate Period. Monetary and Economic Studies, 22(2), 1943.
Nakajima, Jouchi, Kasuya, Munehisa, and Watanabe, Toshiaki. 2009a. Bayesian Analysis
of Time-Varying Parameter Vector Autoregressive Model for the Japanese Economy and
Monetary Policy. IMES Discussion Paper Series, May.
178
Nakajima, Jouchi., Shiratsuka, Shigenori., and Teranishi, Yuki. 2009b. Eect of commitment
Policy : Based on Japans Experience. Institute for Monetary and Economic Studies, Bank
of Japan, August.
Nakajima, Jouchi, Shiratsuka, Shigenori, and Teranishi, Yuki. 2010. The Eects of Monetary
Policy Commitment : Evidence from Time- varying Parameter VAR Analysis. IMES
Discussion Paper Series. Institute for Monetary and Economic Studies, Bank of Japan.
Nelson, Charles R, and Siegel, Andrew F. 1987. Parsimonious Modeling of Yield Curves.
Journal of Business, 60(4), 47389.
Nielsen, Heino Bohn. 2004. Cointegration analysis in the presence of outliers. Econometrics
Journal, 7(1), 249271.
Oda, Nobuyuki, and Ueda, Kazuo. 2007. The Eects Of The Bank Of JapanS Zero
Interest Rate Commitment And Quantitative Monetary Easing On The Yield Curve : A
Macro-Finance Approach. The Japanese Economic Review, 58(3), 303328.
Okina, Kunio, and Shiratsuka, Shigenori. 2004a. Policy commitment and expectation for-
mation : Japans experience under zero interest rates. The North American Journal of
Economics and Finance, 15(1), 75100.
Okina, Kunio, and Shiratsuka, Shigenori. 2004b (Oct.). Policy Duration Eect under Zero
Interest Rates : An Application of Wavelet Analysis. CESifo Working Paper Series 1138.
CESifo Group Munich. http://ideas.repec.org/p/ces/ceswps/_1138.html.
Orphanides, Athanasios. 2004. Monetary policy in deation : the liquidity trap in history
and practice. The North American Journal of Economics and Finance, 15(1), 101124.
Orphanides, Athanasios, and Wieland, Volker. 2000. Ecient Monetary Policy Design near
Price Stability. Journal of the Japanese and International Economies, 14(4), 327365.
Bibliographie 179
Posen, Adam S. 1998. Restoring Japans Economic Growth. Washington, DC 20036-1207 :
Institutes for International Economics. ISBN 0-88132-262-8.
Primiceri, Giorgio E. 2005. Time Varying Structural Vector Autoregressions and Monetary
Policy. Review of Economic Studies, 72(3).
Rudebusch, GlennD., and Wu, Tao. 2008. A Macro-Finance Model of the Term Structure,
Monetary Policy and the Economy. Economic Journal, 118(530), 906926.
RUGE-MURCIA, Francisco J. 2002. Some Implications of the Zero Lower Bound on Interest
Rates for the Term Structure and Monetary Policy. Cahiers de recherche 2002-06.
Universite de Montreal, Departement de sciences economiques. http://ideas.repec.
org/p/mtl/montde/2002-06.html.
S., Bernanke Ben. 2010. The Economic Outlook and Monetary Policy. At the Federal
Reserve Bank of Kansas City Economic Symposium, Jackson Hole, Wyoming, August
27.
Shibamoto, Masahiko. 2007. An Analysis of Monetary Policy Shocks in Japan : A Fac-
tor Augmented Vector Autoregressive Approach. Japanese Economic Review December
2007, 58(4), 484503.
Shin, Hyun Song, and Shin, Kwanho. 2010. Procyclicality and Monetary Aggregates. Wor-
king paper. http://www.princeton.edu/~hsshin/working.htm.
Shioji, Etsuro. 2000. Identifying Monetary Policy Shocks in Japan. Journal of the Japanese
and International Economies, 14(1), 22 42.
Sims, Christopher A. 1980. Macroeconomics and Reality. Econometrica, 48(1), 148.
Sims, Christopher A. 1992. Interpreting the macroeconomic time series facts : The eects
of monetary policy. European Economic Review, 36(5), 9751000.
180
Sims, Christopher A., and Zha, Tao. 2006. Were There Regime Switches in U.S. Monetary
Policy ? American Economic Review, 96(1), 5481.
Sims, Christopher A., and Zha, Tao A. 1998. Does monetary policy generate recessions ?
Tech. rept. 98-12. Federal Reserve Bank of Atlanta. Working Paper. http://ideas.
repec.org/p/fip/fedawp/98-12.html.
Sims, Christopher A, Stock, James H, and Watson, Mark W. 1990a. Inference in Linear
Time Series Models with Some Unit Roots. Econometrica, 58(1), 11344.
Sims, Christopher A, Stock, James H, and Watson, Mark W. 1990b. Inference in Linear
Time Series Models with Some Unit Roots. Econometrica, 58(1), 11344.
Smets, Frank, and Wouters, Raf. 2003. An Estimated Dynamic Stochastic General Equi-
librium Model of the Euro Area. Journal of the European Economic Association, 1(5),
11231175.
Spiegelhalter, D, Best, N, Carlin, B, and van der Linde, A. 2002. Bayesian measures of
model complexity and t. Journal of the Royal Statistical Society, 64.
Stock, James H., and Watson, Mark W. 1996 (Aug.). Asymptotically Median Unbiased
Estimation of Coecient Variance in a Time Varying Parameter Model. NBER Technical
Working Papers 0201. National Bureau of Economic Research, Inc. http://ideas.
repec.org/p/nbr/nberte/0201.html.
Stock, James H., and Watson, Mark W. 1998a (Aug.). Diusion Indexes.
NBER Working Papers 6702. National Bureau of Economic Research, Inc.
urlhttp ://ideas.repec.org/p/nbr/nberwo/6702.html.
Stock, James H., and Watson, Mark W. 1998b. Diusion Indexes. NBER Working Papers.
Bibliographie 181
National Bureau of Economic Research, Inc, Aug. http://ideas.repec.org/p/nbr/
nberwo/6702.html.
Stock, James H., and Watson, Mark W. 2005 (July). Implications of Dynamic Factor Models
for VAR Analysis. NBER Working Papers 11467. National Bureau of Economic Research,
Inc. http://ideas.repec.org/p/nbr/nberwo/11467.html.
Sugo, Tomohiro, and Ueda, Kozo. 2008. Estimating a dynamic stochastic general equilibrium
model for Japan. Journal of the Japanese and International Economies, 22(4), 476502.
Svensson, Lars-E-O. 2001. The Zero Bound in an Open Economy : A Foolproof Way of
Escaping from a Liquidity Trap. Monetary and Economic Studies, 19(S1), 277312.
Svensson, Lars E. O. 2003. Escaping from a Liquidity Trap and Deation : The Foolproof
Way and Others. Journal of Economic Perspectives, 17(4), 145166.
Svensson, Lars E.O. 1994 (Sept.). Estimating and Interpreting Forward Interest Rates :
Sweden 1992 - 1994. NBER Working Papers 4871. National Bureau of Economic Re-
search, Inc. urlhttp ://ideas.repec.org/p/nbr/nberwo/4871.html.
Tam, Chi-Sang, and Yu, Ip-Wing. 2008. Modelling sovereign bond yield curves of the US,
Japan and Germany. International Journal of Finance and Economics, 13(1), 8291.
Teruyama, Hiroshi. 2001. Analysis of monetary policy using VAR : A perspective. Financial
Review, 59(1), 74140.
Thornton, Daniel-L. 2004. Testing the Expectations Hypothesis : Some New Evidence for
Japan. Monetary and Economic Studies, 22(2), 4569.
Ueno, Y., Baba, N., and Sakurai, Y. 2006. The Use of the Black Model of Interest Rates
as Options for Monitoring the JGB Market Expectations. Working paper. Institute for
Monetary and Economic Studies, Bank of Japan.
182
Ugai, Hiroshi. 2007. Eects of the Quantitative Easing Policy : A Survey of Empirical
Analyses. Monetary and Economic Studies, 25(1), 148.
Uhlig, Harald. 2005. What are the eects of monetary policy on output ? Results from an
agnostic identication procedure. Journal of Monetary Economics, 52(2), 381419.
Yano, Koiti. 2009. Dynamic Stochastic General Equilibrium Models in a Liquidity Trap and
Self-Organizing State Space Modeling. Tech. rept.
Yano, Koiti, Iidaz, Yasuyuki, and Wagox, Hajime. 2010 (Aug.). Estimating New Keynesian
DSGE Models in A Liquidity Trap Using the Monte Carlo Particle Filter : An Application
to the Japanese Economy. ESRI Discussion Paper Series. Economic and Social Research
Institute, Cabinet Oce.
Yates, Tony. 2004. Monetary Policy and the Zero Bound to Interest Rates : A Review.
Journal of Economic Surveys, 18(07), 427481.