January 19, 2010 One of the most popular assumptions use in in!estment moelin" is that sto#$ pri#es follo% a ranom %al$. &o%e!er, a "ro%in" 'oy of e!ien#e sho%s that this is a 'a assumption, an that sto#$ pri#es "o throu"h perios of o!er(pri#in" an uner(pri#in" that affe#t future returns. )any of the moels use 'y finan#ial planners are 'ase on the ranom %al$ assumption. This naturally raises the *uestion of %hether %e #an usefully 'uil in!estment moels 'ase on more realisti# assumptions. +n this arti#le, + propose that %e #an 'uil 'etter moels an pro!ie some initial su""estions on ho% to o so. ,ast fall + %rote t%o arti#les for Advisor Perspectives -see here an here. is#ussin" the use of sto#$ mar$et !aluations to prei#t lon"(term sto#$ mar$et performan#e. The parti#ular !aluation measure + use %as a pri#e/earnin"s ratio that smoothes out #y#li#al earnin"s 'y in#luin" ten(year a!era"e earnin"s in the enominator. + referre to this measure as the Shiller P0 'e#ause the main proponent of its use has 'een Professor 1o'ert Shiller of 2ale 3ni!ersity %ho hi"hli"hte it in his 'oo$, 4+rrational 05u'eran#e,4 an in earlier resear#h. This measure is also $no%n as 46ormali7e Pri#e 0arnin"s 1atio,4 4P0 10,4 an 4Cy#li#ally A8uste P0 -or CAP0..4 All referen#es 'elo% to P0 ratios are to Shiller P09s. 1e#ently + e5amine this measure more eeply %ith the "oal of 'uilin" a simple sto#hasti# moel 'ase on the intera#tion of P09s an sto#$ mar$et rates of return. A sto#hasti# moel of this type #oul 'e useful in impro!in" the )onte Carlo pro8e#tions of in!estment returns, an ma$in" impro!ements to one of the prin#ipal tools use 'y finan#ial planners. )y %or$in" hypothesis is that P09s are in!ersely #orrelate %ith future returns, 'ut, on a year(to(year 'asis, any impa#t is mostly ro%ne out 'y ranom influen#es. O!er time these P0 effe#ts a##umulate an si"nifi#antly affe#t performan#e. So +9m assumin" that the sto#$ mar$et almost : 'ut not #ompletely : follo%s a ranom %al$ an that lon"(term in!estors #an 'enefit 'y a8ustin" portfolios 'ase on #urrent le!els of P09s. + 'e"an 'y e5aminin" histori# sto#$ mar$et performan#e ata "oin" 'a#$ to 192; alon"sie Professor Shiller9s ata on histori# P09s. Tests of the relationship 'et%een 'e"innin"(of(year P09s an one(year returns prou#e the e5pe#te < Copyri"ht 2010, A!isor Perspe#ti!es, +n#. All ri"hts reser!e. ne"ati!e #orrelation %ith a #oeffi#ient of (.2= an an 1(s*uare measure of .0=>. A rou"h translation is that the 'e"innin"(of(year P0 e5plains a'out =.>? of that year9s sto#$ mar$et return. The !ariation of the yearly returns aroun the re"ression line %as #onsiera'le : e*ui!alent to a stanar e!iation of 20?, a'out the same as e5pe#te o!erall sto#$ mar$et !olatility. For moelin" one( year sto#$ returns as a fun#tion of the 'e"innin"(of(year P0, + appro5imate the relationship %ith the e*uation@ 1eturn A -.2=.B-.9C.DP0 E 6-0, .20. %here 6-0, .20. is a ranomly "enerate normally istri'ute !aria'le %ith 0 mean an a stanar e!iation of 20?. For e5ample, at a P0 of >Fthe lo%est e5perien#e sin#e 192;Fthe e5pe#te return is 1;? plus or minus the ranom term. At the all(time hi"h P0 of G2, e5perien#e 8ust 'efore the 'urstin" of the ot(#om 'u''le, the e5pe#te return is H? plus the ranom term. Basi#ally, the moel refle#ts the ata, %hi#h sho% that the 'e"innin" P0 e5erts an influen#e on the e5pe#te return for the year, 'ut %ith #onsiera'le ranom noise. Completin" the moel re*uires another e*uation to "enerate the en(of(year P09s. As one mi"ht e5pe#t, ea#h annual #han"e in P0 is #losely relate to that year9s rate of return. Base on this relationship an the histori# ata, + e!elope the follo%in" e*uation to "enerate the enin" P09s@ P0-tE1. A P0-t.B-1E.90B1eturn-t.(.0;GE6-0,.0>.. These e*uations #an 'e #om'ine in an 05#el moel, an repeate simulations #an 'e run usin" 05#elIs ranom num'er "enerator. Ta'le 1 pro!ies a #omparison of a#tual !ersus test results. 0a#h of the tests in!ol!e 10 runs of ;1(year -192; : 200;. !ersions of the moel, so the a!era"e of all tests is 'ase on C0 simulations. O!erall, the moel prou#e a "oo fit %ith the a#tual ata. The a!era"e P0 from the moel #ame in sli"htly hi"her than the a#tual a!era"e P0, 'ut it9s un#lear %hether %e nee to fine(tune the moel or a##ept that the a#tual num'ers %ill #ontain some statisti#al !ariation. -The a#tual histori# num'ers are really 8ust a sample, an may or may not represent a 'est estimate of num'ers "oin" for%ar.. Jhat is perhaps most interestin" in the #omparison is that the #orrelations an 1(s*uares are hi"her for the a#tual than for the a!era"e of the tests. )y "uess is that there are li$ely #loser ties 'et%een P09s an su'se*uent performan#e than are #apture 'y my simple moel, 'ut + ha!e no %ay of pro!in" this. +n "eneral, it loo$s li$e this moel pro!ies a "oo start to%ar 'uilin" a %or$a'le simulation moel to sho% the effe#t of P09s on performan#e. < Copyri"ht 2010, A!isor Perspe#ti!es, +n#. All ri"hts reser!e. Ta'le 2 pro!ies a #omparison of test results %ith the asset allo#ation results + e!elope in the first arti#le referen#e a'o!e. Base on a small sample of 8ust 10 #ases, the results for the test #ases are similar to the a#tuals, althou"h the a#tuals sho% a 'i""er impa#t from in#orporatin" P09s in the asset allo#ation. This result is #onsistent %ith the lo%er #orrelations for the test results sho%n in Ta'le 1. The 'eha!ioral influen#e on sto#$ pri#es has 'een es#ri'e in the popular press as an e5#ess of "ree %hen sto#$ pri#es are hi"h, an an e5#ess of fear %hen pri#es are lo%. The result is sto#$ pri#e mo!ements that o not follo% a ranom %al$. This moel also re8e#ts the ranom %al$ assumption, 'ut it is 'ase on a ifferent #hara#teri7ation of in!estor 'eha!ior : some fear -'ut not enou"h. %hen pri#es are hi"h, an some "ree -'ut not enou"h. %hen pri#es are lo%. To the e5tent %e #an impro!e our unerstanin" of in!estor 'eha!ior an 'uil simulation moels that refle#t its effe#t on in!estment returns, %e #an pro!ie more useful tools for finan#ial planners an their #lients. < Copyri"ht 2010, A!isor Perspe#ti!es, +n#. All ri"hts reser!e. Table 1 - Shiller PE's and Returns--Actual versus Test Results
Averages PE Return Std Dev of Return Loest PE !ighest PE "orrelation PE and 1#-$r Returns R- S%uare Test &1 19.G= 10.>G? 20.G>? >.9 GH.2 (0.C> 0.HG Test &' 1=.GG 11.21? 21.09? =.1 HH.; (0.CH 0.H2 Test &( 1;.1C 11.0>? 20.;G? >.1 G0.2 (0.=G 0.G2 Test &) 1=.H2 11.G1? 19.;>? =.; HH.; (0.=1 0.H9 Test &* 21.0; 11.G2? 20.G1? >.2 GG.9 (0.=1 0.H; Average All Tests 1;.29 11.1>? 20.CG? >.0 H9.2 (0.C9 0.H> Actual 1+',-'##, 1>.H> 11.09? 20.H9? >.G G2.C (0.=9 0.G> Table ' - Asset Allocation Results--Test versus Actual
*#-*# .ond-Stock .u/ and !old *#-*# Annual Rebalance PE-.ased Allocation .u/ and !old PE-.ased Allocation Annual Rebalance Average of 1# test "ases >.;G? >.99? ;.G0? ;.>1? Difference vs0 *#-*# .u/ and !old 0.00? 0.1C? 0.C=? 0.;>? .ased on Actual 1+',-'##, ;.H0? ;.G1? 9.GC? 9.=C? Difference vs0 *#-*# .u/ and !old 0.00? 0.11? 1.1C? 1.HC? %%%.a!isorperspe#ti!es.#om
For a free su's#ription to the A!isor Perspe#ti!es ne%sletter, !isit@ http@//%%%.a!isorperspe#ti!es.#om/su's#ri'ers/su's#ri'e.php < Copyri"ht 2010, A!isor Perspe#ti!es, +n#. All ri"hts reser!e.