You are on page 1of 39
[NBER WORKING PAPER SERIES GOOD POLICY OR GOOD LUCK? COUNTRY GROWTH PERFORMANCE ‘AND TEMPORARY SHOCKS William Easterly Michael Kremet Lant Pritchett Lawrence H. Summers Working Paper No. 4474 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massschusens Avenue Cambridge, MA 02138, ‘September 1993, We are grateful to an anonymous refer, Robert Barro, Nancy Birdsall, Olivier Blanchard, Michael! Brano, Daniel Cohen, Brad De Long, Staley Fischer, Chad Jones, Robert King, Ross Levine, Johannes Linn, Ramon Marimon, Robert Murphy, Sergio Rebelo, Fabio Schiantareli, Paul Romer, and Nick Stem for comments and useful discussions, to participants in the growth conference in Estoril, Fortgal in January 1993 (especially our diseustant Danny Quah) and in the growth conference in tbe World Bank February 1993 (again especially our discussant Alan Stockman), and to George Clarke and Sheryl Sadberg for research assistance. Any remaining crrors are the responsibilty of the authors. This paper is part of NBER's research programm in Growth. Any opinion expressed are thse of the authors and not those ofthe Department of the Treasury, the World Bank, or the National Bureau of Economic Research. [NBER Working Paper #4474 September 1993 GOOD POLICY OR GOOD LUCK? ‘COUNTRY GROWTH PERFORMANCE ‘AND TEMPORARY SHOCKS ABSTRACT. ‘Much of the new growth literature stresses county characteristics, uch as education levels or political stability, be dominant determinant of grow, However, growth rates are highly unstable over ime, wit a comeaton across decades of to 3, while county characteristic are stable, with cross-decale correlations of 6 10.9. Shocks, especialy those to terms of trade play 4 large tole in explaining variance in growth. These findings suggest iter that shocks are important relative to county characteristics in determining long:run growth, or that worldwide technological change determines long-run growth while county characteristics determine relative income levels William Easterly Michael Kremer ‘The World Bank Department of Economics E52-262B 1818 H Steet, NW Massachusets Institute of ‘Washington, D.C. 20883, ‘Technology ‘Cambridge, MA 02139 and NBER Lant Pritchett Lawrence H, Summers ‘The World Bank ‘Under Secretary for Intemational Affairs 1818 H Steet, NW Department of the Treasury, Room 3432 Washington, D.C. 20433 1300 Pennsylvania Avenue, NW ‘Washington, D.C. 20220 and NBER 2 Torreductin Much ofthe new grow liceraure stresses county characersics as the dominant determinant of grow performance. A vast empirical terre tess the effec of county chartcteriscs on frown This pape preens fact suggesting the emphatis on country characteris i miguidet srowth rates are highly nsble over ime, while county characteris are highly persistent. The correlation acros decade of countries" growth ates of income pe capitis around 119.3, while most county characteris display crost-decade correltions of 610.9. Correlations of growth oss periods a long sto decadet ~ period lengths comparble to those used inthe cross-section pirical trae ~ ae smarty low. With a fw famous exceptions he sme counties donot do el period afer period: unre are “ccs stores" one period and dttpoinents he nex. “The low pessece of growth ate reconciled the enormous variion in grow rate aeons course with the remarabe sabi of elaive incomes arot countries. For each of he astro decades the tandard devation of growth ates been ove 2.5, neal the growth ference ‘ecween Japan andthe US. Yet the coretaton of (Summers and Heston (1991)) GDP per capa in 1960 and 1988 was 92. Even more srking the rank contiaton of GDP per capita forthe 28 courries for which Madson (1985) has daa it 82 over 1670-1988. Malor changes in country ‘income rankings would hve required lrg priest ference ingrowth as; in the even income rankings di not change mich and only a smal raion ofthe grow differences brween counties were persistent. "Anew eur i ri i pin cing pr uy of eget ee ent ‘Toc ei ap tea aN a St Na ena ae ro (icy cma hoy Yi 97, Ey vrei, ly a [Starry trang Seat eet owe Ba oy Ran. mn ‘fermen Aen snd Rot 19, Peon tnd Tibetan (191), maroon py Face (19, vey (Sonia mn Sor Ot wg De Lng. 3 ‘This paper has ree scons and a conason, The frst seton presents the bse fe abot persistence ofeoss county grow ference and of country characteristics. The second section anempts to deny the temporary shocks imperant in explaining low persistence of growth ates across decades, ‘The hin section interpret low persistence under two nes of growth mol models in which country characterises determine longrun growth res and model in which country caracersies determine relive levels of seady sae income and long-run growth ate re termined by worldwide technological change. A conclusion summarizes the esl 1 Low Persistence of Grom Rate Differences Across Countries () Basie foes The persistence ef growth ae differences arots countries even overlong prods, i low. ‘able | present correlations ofthe lest aquaes growth rte of GDP per worker berwcen 196049, 1970-79 and 1980-84, The Rt cbaned by regressing he cure growth rae on the previous decade's growth was es than 10 percent Lite ofthe variation of growth rs is explained by past, srowih? This low peritene ret i robust over te choice of county sample, time period, and sectoral perfomance menue Fire 1 splay the saserplot ofthe grow rates for 115 couric over to periods, 1960- 73 and 1974-88, The ote lines show the averge in each perio. A larg portion ofthe sample fs contained inthe of iagenal ndrans: above-average in 1960-73 and below average in 197488, or vice vesa. The rank comlaion i 21 in the figure. ‘es tng Won a npr pert ned ot oe ein OO ‘inne OF et seria nu be ance ot pac ane Tee asus for he Se ‘Steam e-ponrecn empna poe is et eae ees pr nd pa ot ere, ‘'Ravuret Sa su, wea eps eyo rcp” geo at mr tse years gy ge a oman sh ve vinta i Lat Sa (81), Levit Ram 9 a Fe ana any pommel ng mney womb fe a tr nn mee 4 ‘The bones inthe comers represent the dies ofthe period growth rates, The notes box ‘represents coureris with growth in he top deciles in both periods. The southwest box shows the counriespessemly inh botem dele. The norest bo (persistent sce) conins Botswana and the famous Aslan Gung of Four (Hong Kong i acu just shor of being in the top decile inthe fee pviag). The Eas Asian sucess sory well now, wile Botwana has benefited from tensive diamond mines and from a democratic goverment that hs avoided some of is neighbors conomie mistakes. The widespread perception of strong country effec in growahis strongly intenced bythe Gang of Four; without them and Boswans, the already low corelaton of growth ‘aes bermeen periods ict in half, In comrast, persistence i not aed much by deleting a small suber of outers Persistence is a low for several subsamples of countries, The second, third and fours rows of Table 1 show the coneation for non-il counries, the OECD countries, andthe noni veloping couurics. The ony exepion thigh comeation between the 60"s and 70s in the small sample of OECD counts, bu this eves o zero beween the 70's and 80's Figure 2 shows tt prsitence sys low a various pera lengths in the postwar data This is confirmed by pari dita on long-run growth res for 30 year periods over 1870-1988 * We have 2 ol of S¢ observations fot 23 OECD and Lain American counties. Figure 3 shows gromt plone agsnst lagged groweh fr these 30 yar periods. Prtgal is ilstraive: decent growth in 187099, negative growh in 190029, average grow in 19909, and one of the highest growth rae in 1960 etry cuad manors rove of cpm co i emda bd (ww msty sason 0 hod dy ras amet bn ae eae, mg ra Sane bee pra 5 88, The corelation of 30year pe capita growth wit pr capita growth in the previous 30 year eid in this dais oly 12 (ne posible explaaton for low persistence in the recent dati instability in agricul due (0 price and weather shocks, Figure 4 shows persistence coefcins fr growth of valve added per ‘worker in agriculture, incustry, and services. The rank persistence of apiculate is zero between the 701 an 808 and is low beeen the 608 ard 70s, However, industry and servis ls have low ersten, “Table 2 shows the lw contemporaneous corelaons of growth rats across sectors (withthe exception of services in he 70, when it fad about x oneal correlation wih both inde and sricuure), The tow coreations could mean that shocks to individual sector, suchas relative price ‘movement that pl factors ike apt an sill entepreneurs ou of one sector ad int thers, are more imgorant than shocks afetng al scor, such a increases ineconomy-wide human capital However, anoter explanation of low cross sector correlations could be that even eonomy-wide shocks ease secon shits because of changing comparative advange Measurement err inthe level of GDP could reste aici low persistence ingrowth ‘nes, by leading to an underestimate of growth in one period and an overestimate inthe next, or vie ‘versa. However, we do st believe tat measurement eror explains low persistence, Fist, growth rates are probably not cosructed by enimaing GDP in subsequent periods - more likely, growth cetimates ae prepared fit, and GDP inthe second perio i eximaed from these growth estimates, Second, we calculated pesistence beween periods har id not eomain a common endpoint but nsead "Tine om fe pe in of eer (DEH CIOS C1829 on me wee ONES. CIFDI9 inom cha Guy ts cesar ema pwa te 9 cena ne pe ent ‘cmmmen ens oer ei ne ep ad a tee SET Fe cmp ecm oh acannon oP ‘oot td ato Seema 9 a ie corp my ee es thre 6 ere seprated bya gap of one or sever years. This let persistence unchanged o¢ lowered i raher than raising i as would cur if measurement erot were import. Third 8 Figures 2 ad 3 show. persinence remains low even overlong periods, Finally, while id measurement err in levels would lower persistence, other pes of measurement eror ~ suchas country-specific tendencies to overeport growth rates ~ would raise perience, 1, Are Counery Characins Pertiten? ‘The most raghforward explanation of he fow persistence of growth rates would be hat che county characterises wll thought © determine grow are therelves not persistent. This section shows this explanation tobe untenable: county characteristics are persistent Figure § shows Dersstence of country claractersies beeen the 60's and 70s, and berween the 70's and 0's for sample of 45 countries or which dais available fr all variable and time periods. The variables tosen are those hat appear inthe ease growth regression of Baro (1991), 8 well sever] cers common inthe aur. Alf he country characterise display far higher persistence than frou ates, Many ote country characteris, ike clr and geography, must be even more persistere However, some agrezae index of policy variables could til ave low persistence.” To consrut such an index, we we the variables shown in Figure $ with a pooled kimeseriescrost- section regesion on 10 year averages, Table 3 shows regressions ung te Baro (1991) variables ‘withthe excention of his PPISODEV (evasion ofthe reaive price of investment from te sample ‘mean, which sox avaable in individual decades fra nufcien sample (Our governee consumption variable does not exclude spending on defense and education at Bar's di, duet ack of decade ata on the laee.) We allow the interceps to vary across decades. We also perform 2 “Sm ere cient mo mn evr om pm 7 second renression with abreader set of couse characterises, The fined values from his regression (Genoted Barro Index ané Augmented Barro ides, respectively ar ls fr more perisen han rowth rates, a shown it Figure 5. Races of factor atcumalation are much more perssen than growth ries. To compute an index of factor accumulation, we reresed aggregate growth (oot pr capt) on investment and labor force row, ung «sample of 115 eounies which have daa for all tree decades, Figure 6 shows tat invesonent. labor fore grow, and the fae value of groweh predic by the two ae much smote persistent han growth, The residual rom this regression can be interpreted, under cerain -sssumptons, 38 the deviations of ttl facie productive growth for ech country fom the global ‘pean As shown inthe grips, TFPgrowih rates are even ess persistent han groweh ae 1) Stocks and Poles ‘This section argues dt sbocks, especially sock othe terms of trade, are an import etemninant of variations ingrowth rues over tn Year periods, ad that they can help account for low persistence Below we test how much ofthe variation in growth rates berween countries can be statistically explained in terms of diffrence in policies, and how much is due wo diferenes in shock variable, such asthe terms of trade, external transfer, te change inthe umber of war related canals per capita on rational tertor, andthe presence ofa deb crisis, We show that much ofthe variance in growth aes, even over prio a long a a decade, can be dtedy explained by shocks.” Moreover, “re ingot npn rr nen em me SES cae nh neat cee Gece ow 8 shocks indirectly influence growth by changing policy variables. Thus the low persistence of shocks, aricularty extemal socks helps explain the low persistence of growth rates, ‘Table 4 shows the simple correlations of tres shock variables with growth rtes.® The ‘arabes are (1) the grewch in dollar export pices times the iia share of expors in GDP minis the fromth in import prices times the initial sare of imports in GDP (tems of trade change); (2) the change in war casualties per capita on national eritory; and (3) a dummy measuring counties likely to have a debt crisis in he 19805." Growth is strongly corelated with terms of trade improvements and high extemal deb nthe 8°, ae with wat in the 70° (and weakly wih war inthe 807). ‘When shock variables are added toa regression witha small set of sgifcan country chanctristies from secon I, thy have substantial explanatory power compared to policy variables (Table 5), We add te hee variables from te previous paragraph and, for completeness, the per num inerease in offal transfers, The parial Rt ofthe policy variables (enrollments, black market premium, M2/GDP) inthe 1970s was 26 and ofthe shocks 14, while inthe 19805 the paral of the policy variables was 10 verss 15 for shock variables. ‘The terms of trae effec is large and strongly significant in both periods. Inthe 1980s a favorable terms of trade shock of 1 percentage point of GDP per anmum raises the growth rate by 85, percentage point per anmum. Recall that GDP is measured in constant price, s0 there is no direct, effect of a terms of trade shock on growth. ‘This increase in growth i far larger than would be creued simply through the effect ofthe lnreaed income on savings, Even if al the shock pase ou tenn se pope ct tind ei otf Mary al Dare) a ny ele amg ee tn CDP wae 0 pre 19 nd mie tne es ws mere ch nec pee ac a © Tp ay er ering of pn cmt ax arog Ta cme coor dost ol Bop Beeb ml ef CO ° Into saving. and the rate 9 rerum capital were (optimistically) 20 percent, growth would only inerease by 2 percentage pois Factor movemene ae one poten explanation of large growth ees from terms of trade shocks. For example, bbor or capital might flow within the counry to the sector receiving a favorable shock, capital nih low in trom abroad co the expor set, of domes savings might respond o improved exper opportunites. tn order to generate large growth elec through factor ‘movements, however, fadors and export demand must be elastic, and tems of trade shocks mast be 2 least somewhat persistent. Excel shock vribles eter than he terms of wade have smaller effects on growth, panty refecing substantial mlicliesriy mong the shocks an beeween shock and policy variables. The ‘aribl forthe increase in war asus i marily significant in the 70's but ot nthe 80's: we ‘it detect significant spare effec of tates and debt crtea. The maginede ofthe coetfcen, om the war casual variable implies relatively modest eects of war in most cases, Violence in Chile associated with the ovenbrow of Allende an its afermath are extinted to have cost.3 percentage oins of gromh per annum inthe 70s. Isr’ wars during he 70's are estimated 0 have lowered gromh dung the decade by 2 percenage pins per aanum. Highest asus per ‘oii in the sample were fom the civil war in Uganda, which was estimate fo have reduced growth in he 70' by 3 percentage points per annum. Civen the distribaon of various shock variables (with 2 few large values for canbe, transfer, and rms of trade movements) the results for dividual ‘arabes ar sestive fo oie of sample "amar gins et pn mt de neve vn rh 9 of op (i Chen neers deg emo sy A ams pa pb ae Shot cnuages peta pres ar ena ber yond nyo ae oye De {ej ind Semen 9, Fey nip como wcs Keean eps acest ne ‘pogo ny pred 10 ‘The shocks help explain the low persistence of the observed growah rates. The corelaion of the growth rates between the 1970s and 19808 is 0S in this sample of countries, while he persistence of the component of predicted gromh dependent on the non-shock variables was 63, The Correlation berween decides of the fied growth component du 1 shocks was -08 andthe pessisenc ofthe Sted growth rates including all variables was 37% ‘The shock variables influenced grow not only directly, but alto indie, through policy ‘arabes. Table 6 repr regression of the black market premium on shock variables. War is scsociated wih a hgh tack market premium and favorable terms of trade changet witha lower premium ‘This cass doubt on the widespread imerpreatin of the black market premium aan indicator of bad policies, If shock variable ae omined,eximates of the effet ofthe back market prem on growh wll faltely aribue externally induced adversity to policy. Table $ demonstrates thatthe inclusion of shock variables in the regression rtuces the coefficient and significance levels onthe lack market prema, especially inthe 1980s, when it cus the coefficient in half ‘To summarize, shocks are imporas over decade-long periods, they help explain ihe Aiference beween the peristence of acmal and predicted growh, ant they influence “policy” ‘variables, and thus etnates ofthe impact of police. IM) Persistence and Growth Theory ‘This section examine the ierpreation of low persistence under two types of grow modes In the frst eype of mod, long-run growrh depends on country characteristics. For example, inthe AK model of Rebelo (191), growth depends on tax rts. In closed economy versions of Romer (1990) or Aghion ad Howie (1992), technological change, and therefore economic grown, depend ctu nas poe na eas aioe mega on mane Por Oh ley 3 sh nt u on a counsy's patent system and market size, In simple versions ofthese models, low persistence of srowth rates implies tat andom shock are imporant in determining the long-run pth of ouput. In the second type of mos, which includes both the neoclassical model with exogenous technological change and some models of technological dfn. growth isa world-wide process and country chuaceriticsdtemine the reative level of income, Ia these els low persistence is consistent ‘vith shocks of any size and shocks may play only a minor role in determining the long-nin path of ‘ouput, despite being ar imponantdeteminan of variance in decade tong growth res, 4. Models in Which Cowsry Characteristics Determine Long-Run Growth Ina simple mol in which country characeristics determine growth, the prsience coefficient canbe inerrted a eflecing the magitue of variance inundesying growth rates Across counties relative tthe variance of random shocks. To see his, dot he long-run growth rat associated with the policies of country iasg, This can be represented asthe word average sowie, g, ps a unary specific component ¢, determined by coun charters, Grow for country | in period equals its underiying growth rte, pls a cour specifi, period specie shock, (A period specie aggregate shock could also be added, Dut would not afet heres.) Thus, Be BE TE, va) = of varie.) = oy a ‘Th simplest assumption one can make it that ¢ and. 4 are independent normal variables, and is serially uncorrelated, Under tis asumption, the persistence coefficient, denoted», is 26,-B VE, 8 VE, 1-8 ® ‘This simple model ofecuty fined effecss does no allow for changes in policy over vine. However since policies change ony slowly, ic may be a reasonable approximation over periods that ae not 0 long. Unie this model, te best freast ofa country’s growth rte will ea weighed combination of its ow past growth mie and the average growth ate ofall other counties" Under this mode of fixed county effect, low persistence bounds he penal R tha an be hieved in grow regeatons. Even if polices were perfectly meaated, and all polices and ther facor affecting growth were taken ino account, the expect R? ina thiny-year growth regression ‘would be only about 0.6. To see this, aoe that she expected R? fom regressing growth over pevods oa a perfect measure of policies that determine the country’s long-run grow rate willbe ERX] = it ol a, o m0, Ady ‘This simplifies to o aoe From the definition ofthe persistence coefficient, a: = (I)af. Hence the expected R from ‘egrestng growth over n periods ona perfer measure of the polices that determine the country specific underying growh rate will be iRim) = ot _ ae 2 parte 6 on Ione defines a perio a ten year, then ois approximazely 1/3, 50 the expected R? over a thiy-year period, given perfect measure of policies sony 0.6, Thus if this model of fixed country eects esribes te du, sulky that we wil ee much increase over the current R? si the Hera whlen are alresdy inthis range. For example, Levine and Rent (1992, p. 947) report Rs fom 46 10 62 in thei bate regresons for groweh over 1960-69. OF course Risa random variable 9 in the process of many autor running regressions, it would not be surprising if some obtained higher Rs. On the other hand, the expected R given exiting imperfect measutes of policy would be les tan 0.6. ‘The fading that tis model even perfect measure of policies would explain nly 60% of variance in growth rates ove a thiny-year period has economic as well as econometric iplisions {nhs simple mode, low persistence implies that ek i imporane eave to police in determining the longnin path of out, This model in which country characteris determin long-run growih ‘hus leaves mich of growth unexplained. “Tis simple model asus socks are verily uncontested, I there were negative serial correltion inthe shocks. o if groweh came in spurs for deterministic reasons, persistence would be “4 lower forgiven variance in underlying growth rates across countries, Thus, policies would pay a ‘mote imporar ole in determining the long-un pth of ouput. snot clear why one should cxpetsubuanal egaive serial corelation over successive ten-year periods, For the spurs hypothesis i itereting to ate tat forthe courries hat have four decades of dain the Suramer nd Heston (1991) sample, on average around 6O% of thee growth rom 19501 1988 is Achieved in he fastest growing decade within that period. However, iis not lear wheter this is ve to determine spre of rows of high random variation. Models in which Worldwide Techrological Progress Determines Long-Run Growh Under a difersne typeof model, worldwide technological progres determines lone-ren row and country characteristics determine steady sate relative level of income, This esteory includes not ony the ecelariel model (Solow (1956), but ao some models of ehnologial p).* In this model, the relative steady state level of income is determined by policy, bu except for those countries Iarge and advanced enough to generate a significant share of word techaology, long-run growth i exogenously determined. Under ca difsion which cither a neoclassical mode! of capil accumulation of models of tet “Fo smn apc oat Pip 40) ror ah 9) Bet an es 1s incorporate advantages of bckwardnss, persistence depends onthe distribution of countries! incomes relative to their steady ste income Adding an nderendent norma error tem linearize version ofthese models allows persistence tbe characrtasd, Ifthe isa wide daperson of distances berween cours nti Incomes and their steady sates, then transitional namic wil dominate the effect ofthe random crvor tem, ‘The counties fahest below thei steady sate wil grow the fastest. Relative growch ‘es wil inialy be Righy persist. However, all counties approach hit steady sate evel of income, persistence wil il because trasitional dynamics will Become les important relive othe random err erm. Asynprotiallycoueis will onverge to an ergo distbtion around the steady state, in which pensence wil be negative since counties which receive a positive random ‘shock one pesod will tend to fallback towards the steady state tbe net period “his can be easly een in Baro and Sulas+Matn’s linearized version of the neoclassical ‘model with an aed random shock, but similar res hold Inthe dffsion model, fa the rneclassical model, y,,, = yyewiy*-y,)ty WHEE Yu dences 1g income of country i atime &, 1 denotes steady stat income, gis a random shock, and v & (0,1) measures the speed of adjustment to he steady sate, which depends on a hos of parameter, including the capital share." Thus groweh berween «and +1, denoted g. equals v4y"-y,)+44, erating, a = Y=) a) Even hs, is straipeorward to wie perience as a we cou pf tn en py nse ei ty ett on oj eal ts be i ay ae 6 fercion ofthe erossection variance of income, Since gf, ~ W-y,)"W, 2a os Bo © WGEI)YG I) oreo, a8 Fle DE] = Weak o where 9 denotes the crss-setonal variance of log income at time t and , denotes he variance of the shock, As Barro and Sala-i:Mantn show, ay = -yfel +08 ® Sie yyy = yO “HDehy- iene defen of perience 8 cme Bowl (ovat = voe o Since the numerator increases more than proportionally ing? and the denominator increases lst than proportionally ino? persistence increstes with the cross-section variance of income Baro and Sulni-Manin show tht a goes to infinity, the cross-section variance monotonically” approaches the steady state value, o «of (2v-v2). Thus if he inkl crossetionvarace is gener than is sendy state vale, peristence will decline overtime. Persistence it asymprotically negative, since the limi of the covariance berween a country’s growth at and 2 +1 i co) which must be negative, Nove that even ifthe random shock are arbitrarily small, thee models predict that persistence wil asymptotically become negative. Under this model, 2 county's ime path of income could be stermined almost complaely by worldwide technological change and is police, but if it were close to ts steady sate income a lrg percentage of te time series variance in is grow rate would be explained by random shocks, In this ete the grown rate would Just represen Mucratons around 3 steady state income “This mode couldbe generalized by allowing ech county to have its own steady state level of income depending on plies, and by allowing for exogenous technological change. In this cate, persistence depends not ox varlance of income, Sut on varlance in the gap berween actual Income and steady state income relative othe level of technology. If counties vary greatly in their dstanee fom their eave steady stats, persistence willbe high. Te counties fr below tei eave steady ome wl inallybaveperistenl high growh rates. As they approach the steady sate, ‘hee grow ae wil all. Asymptotically, ere is a sharp distinction berween models in which county harass

You might also like