Good News for Value Stoks

7/27/2019 Good News for Value Stoks

1/17

THE JOURNAL OF FINANCE . VOL. LII, NO. 2 . JUNE 1997

Good News for Value Stocks: Further Evidenceon Market EfficiencyRAFAEL LA PORTA, JOSEF LAKONISHOK, ANDREI SHLEIFER, and

ROBERT VISHNY*

ABSTRACTThis article examines the hypothesis that the superior return to so-called valuestocks is the result of expectational errors made by investors. We study stock pricereactions arou nd ea rnin gs annou ncem ents for value and glam our stocks over a 5-yearperiod after portfolio formation. The announc em ent retu rn s suggest tha t a significantportion of the return difference between value and glamour stocks is attributable toearnings surprises that are systematically more positive for value stocks. The evi-dence is inconsistent with a risk-based explanation for the return differential.

MOST FINANCE RESEARCHERS AGREE that simple value strategies based on suchratios as book-to-market, earnings-to-price and cash flow-to-pricehave pro-duced superior re turns over a long period of time.^ Interpreting these superiorreturns, however, has been more controversial. On one side, Fama-French(1992) argue th at these superior returns rep resent compensation for risk alongthe lines of the Merton (1973) intertemporal capital asset pricing model(ICAPM) where portfolios formed on book-to-market ratios are interpreted asmimicking portfolios whose returns are correlated with relevant state vari-ables representing consumption or production opportunities. On the other side,Lakonishok, Shleifer, and Vishny (LSV, 1994) contend that there is littleevidence th at high book-to-market and high cash-flow-to-price stocks are risk-ier based on conventional notions of systematic risk. LSV argue instead thatvalue stocks have been underpriced relative to their risk and return charac-teristics for various behavioral and institutional reasons.A speciflc behavioral explanation pursued in more depth by LSV (1994) isthat the superior retu rn on value stocks is due to expectational errors made byinvestors. In particular, investors tend to extrapolate past growth rate s too farinto the future. Evidence going back to Little (1962) suggests that company

* La Porta is from Ha rvard Un iversity, Lakonishok is from the U niversity of Illinois, Shleifer isfrom Harvard University, and Vishny is from the University of Chicago. We thank Gene Fama,Steve Kaplan, and an ano nym ous referee for helpful co mm ents. Finan cial suppo rt was provided bythe National Science Foundation, the Bradley Foundation, and the National Bureau of EconomicResearch A sset Man agement Research Advisory Group.^ For example, see Basu (1977); Rosenberg, Reid, and Lanstein (1984); De Bondt and Thaler(1985, 1987); Jaffe, Keim, and Westerfield (1989); Chan, Hamao, and Lakonishok (1991); Fama


2/17

860 The Journal of Financeearnings are close to a random walk, with earnings growth rates being pre-dictable only one to two years into the future. Yet the large price-earnings ratiodifferences between value and glamour stocks seem to reflect an expectationthat past growth differences will persist much longer than is reliably predict-able from past data. Value stocks provide superior return s because the marketslowly realizes that earnings growth rates for value stocks are higher than itinitially expected and conversely for glamour stocks. While such extrapolativeexpectations may not be the only source of mispricing, at least they representa testable alternative hypothesis.^In this article, we examine the role of expectational errors in explaining thesuperior return to value stocks. As in Chopra, Lakonishok, and Ritter (1992)and La Porta (1996), we examine the market's reaction to earnings announce-men ts to determine whether investors make systematic errors in pricing. Wetest whether earnings surprises in the 5 years after portfolio formation aresystematically positive for value firm s and negative for glamour firms. This isa direct test of the expectational errors hypothesis. Earnings announcementprice reactions also reveal the time pattern of the resolution of uncertaintyabout the relative prospects of value and glamour firms. Because the superiorreturns to value strategies persist for at least 5 years (perhaps with somepetering out toward years 4 and 5), we would expect a correspondingly longperiod of positive earnings surprises for value stocks.Section I describes our earnings surprise methodology. Section II presentsthe basic results. Section III asks whether the earnings surprise results areconsistent with a risk-based explanation of the return differential betweenvalue and glamour stocks. Section FV concludes.

I. MethodologyData on Wall Street Journal quarterly earnings announcement days (eventdays) become available on COMPUSTAT in 1971. For this reason, our sampleperiod runs from 1971:2 through 1993:1. Our universe offirmsconsists of NewYork Stock Exchange (NYSE), American Stock Exchange (AMEX), and Nasdaqfirms tha t appear on the Center for Research in Securities Prices (CRSP) andCOMPUSTAT tapes with data available for certain income statement andbalance sheet items. We exclude real es tate investm ent tr us ts (REITs), Amer-ican Depository Receipts (ADRs), closed-end mutual funds, foreign stocks, unitinvestment trusts, and American trusts.To be included in our sample, the common stock of a U.S. firm must have aCRSP value of equity in December of year 1 and Jun e of year t. The firmmust also have (DOMPUSTAT data on sales, earnings (before extraordinaryitems), cash flow, and book equity, where cash flow is defined as earnings^ A recent article by Daniel and Titman (1997) casts doubt on the risk factor interpretation ofthe superior returns to high book-to-market stocks. In particular, they find that, while high


3/17

Good News for Value Stocks: Evidence on M arket Efficiency 861(before extraordinary items) plus depreciation. To minimize the possible im-pact of COM PUSTAT look-ahead bias (see Ko thari, Sh ank en, an d Sloan (1995)and Chan, Jegadeesh, and Lakonishok (1995)), we require the firm to haveCOMPUSTAT data on sales and earnings for fiscal years ending in calendart - 1 through t 5. This ensure s tha t we do not mea sure stock retu rn s for thefirst 5 years that a firm appears in COMPUSTAT, since this data may havebeen back-filled and could not therefore serve as the basis for a measurabletrading strategy available to market participants.To examine earnings announcement return differences between value andglamour stocks, we form portfolios on the basis of two classifications: thebook-to-market ratio favored by Fama-French (1992) and a two-way classifi-cation based on cash-flow-to-price and pa st grow th-in-sales introduce d by LSV(1994). Portfolios are formed in June of each year t using accounting data forfiscal year end in year t - 1 and m ar ke t value of equity from December of yea rt - 1. For the purposes of size classifications, market value of equity ismeasured at tbe end of June of year t.Using tbe ratio of book equity to market value of equity in December of - 1, we sort stocks into deciles using all firms except tbose w itb negative bookvalues of equity. The value portfolio consists of stocks in tbe bigbest decile ofbook-to-market (BMIO) and the glamour portfolio consists of stocks in tbelowest decile of book-to-market (BMl).According to tbe two-way classification of LSV (1994), value stocks aredefined as tbose tbat have shown poor growth in tbe past and are expected bytbe m ark et to continue grow ing slowly. Specifically, valu e stocks have b ad lowsales growth over tbe previous five years and currently trade for low multiplesof current cash flow, presumably because of tbe market's pessimistic expecta-tions for future growtb (LSV, 1994). Each stock is ranked on casb-flow-to-price(CP) and on a weighted average of sales grow tb ra nk s (GS). Tbe weighted salesgrowtb me asure sta rts by ran kin g each firm based on its sales growth in eacbyear t 5 through t 1. Tbe weigbted average sales ran k is tben obtained bygiving tbe weigbt of 5 to its sales growtb ran k in year t - 1, tbe weigbt of 4 toits growtb rate rank in year t - 2, etc. All stocks are divided into 3 groups(bottom 30 percent (1), middle 40 percen t (2) and top 30 percen t (3)) based onCP and, indep ende ntly, 3 groups based on GS. Grou ps formed on CP are b asedonly on firms witb nonnegative casb fiows at tbe time of formation. Tbeglamour portfolio consists of stocks ranked lowest on casb-ow-to-price (CPl)and bigbest on growtb-in-sales (GS3), wbile tbe value portfolio consists ofstocks ranked bigbest on casb-flow-to-price (CP3) and lowest on sales growtb(GSl) .For eacb of our portfolios, we present annual buy-and-bold returns andearnings announcement re turns . Annual buy-and-bold re turns are reportedfor 5 years after formation witb year 1 beginning in July of year t and ending


4/17

862 The Journal of Financemaining stocks in the portfolio. The stock does not appear in the portfoho in thefollowing year. Annual portfolio retu rns are ohtained by equally-weighting th eretu rns on all stocks that belong to the portfolio a t the beginning of July in yeart. Portfolios are rebalanced to equal weights at the end of each year.The focus of this article is on the earnings announcement retu rns. These aremeasured quarterly over a 3-day window (i - 1, + 1) around The Wall StreetJournal publication dates over a period of 5 years after portfolio formation. Foreach quarter, the 3-day, buy-and-hold event returns are equally-weightedacross all stocks in the portfolio to compute a portfolio event return.As a benchmark for the an nua l buy-and-hold retu rns, size-adjusted retu rnsare calculated as follows. For each year, each stock in the sample is sorted intoa size decile where size is measured as market capitalization of equity at theend of June of year t, and decile breakpoints are based on NYSE size decilebreakpoints (excluding REITs, ADRs, etc.). Since a given size decile maycontain a disproportionate number of value or glamour stocks (with the small-est size deciles typically containing a disproportionate num ber of value stocks),we make an attem pt to purge any confounding value effects from our es timatesof size-based re turns (see Chopra, Lakonishok, and Ritter (1992)). This is doneby forming size decile benchmark portfolios using onlyfirms hat are classifiedas neither value nor glamour firms. For the book-to-market analysis, the sizedecile benchmark portfolios are equally-weighted portfolios consisting of allfirm s in tha t size decile tha t are also in deciles 4, 5, 6, and 7 according to BM.For the (CP, GS) analysis, the size decile benchmark portfolios retu rns includeall firms in the size decile except those classified as value (top 30 percentaccording to CP and bottom 30 percent according to GS) or glamour (bottom 30percent according to CP and top 30 percent according to GS). Annual size-adjusted returns are calculated for each stock by subtracting off the return onits corresponding size decile benchmark portfolio for year t. The annual size-adjusted return for a portfolio is then obtained by equally-weighting thesize-adjusted returns for all stocks in that portfolio.

Size-adjusted earnings announcement returns are calculated in a similarmanner. For each quar ter in the sample, a size-decile earnings announcementbenchmark portfolio is formed using all stocks in that size decile for whichearnings announcement dates are available and which are neither classified asvalue nor glamour firms. The size-benchmark return is then just an equally-weighted average of these earnings announcement retu rns . In other words, thebenchmark used is not the average 3-day re turn for a firm of comparable size,but rather, the average 3-day return in a (-1, +1) window around thatquarter's earnings announcements for firms in that size decile. Size-adjustedearnings announcement returns for each stock are calculated by subtractingoff the return on its corresponding size decile earnings announcement bench-mark. The size-adjusted earnings announcement return for a portfolio is then


5/17

Good News for Value Stocks: Evidence on M arket Efficiency 863II. Earnings Surprises for Value and Glamour Portfolios

Table I reports results on earnings announcement returns and annualbuy-and-hold returns for value and glamour portfolios using the BM classifi-cation. Panel A contains the key results on earnings announcement returnsover the 5 years after portfolio formation. The 20 quarterly portfolio earningsannouncement returns (Q01-Q20) are equally-weighted 3-day, buy-and-holdretu rns calculated on all stocks for which da ta are available for tha t quarter.These are aggregated into annua l intervals by summing up the four quarterlyearnings announcement returns in each of the five postformation years. Forexample, Q01-Q04 is the sample average over 22 formation periods (June1971-June 1992) of the sum of the 4 quarterly earnings announcement retu rnsoccurring in the first year after portfolio formation, while Q17-Q20 is theanalogous return for year 5 after formation.The results indicate tha t event returns are substantially higher for the valueportfolio than for the glamour portfolio. In year +1, the cumulative eventretu rn is - 0 .5 percent for the glamour portfolio and +3.5 percent for the valueportfolio, indicating a relative disappointment in the earnings performance ofglamour stocks. The difference of +4 percent, realized over only 12 tradingdays, represents one-third of the 12.2 percent total difference in first-yearreturns between the value and glamour portfolios reported in Panel C. Thedifference in year +1 event returns between the value and glamour portfoliosis statistically significant. The i-statistics are calculated using the method ofFama-MacBeth (1973) and assuming year-to-year independence of the earn-ings announcement return differences. For example, we have 22 independentobservations on the difference between Q01-Q04 of the value and glamourportfolios, so we just perform a i-test with standard errors calculated from thistime series.Quantitatively similar results also obtain for year +2 . Substantially higherrelative event retu rns for the value portfolio persist even 5 years after portfolioformation, although the magnitude of the difference in years +4 and +5 isapproximately half that in years +1 and +2. This evidence suggests that thepositive updating on the earnings prospects of value stocks relative to glamourstocks takes place quite slowly. This fits well with the evidence on annual buyand hold returns, since those return differences between value and glamourportfolios also persist for 5 years. Annual buy-and-hold re turn differencesappear to peter out more slowly than the earnings surprises, an issue werevisit shortly.Size-adjusted event returns tell a very similar story, but with somewhatsmaller magnitudes in every case. In year +1, the difference in size-adjustedevent returns between value and glamour portfolios is +3.2 percent, repre-senting approximately 28 percent of the 11 percent total difference in annual


6/17

864 The Journal of FinanceTable IAnnual Cumulative Earnings Announcem ent R eturns and AnnualBuy-and-Hold R eturns on Value and G lamour Portfolios Classifiedby Book-to-Market Ratios, 1971-1992 (Full Sample)At the end of each Jun e be tween 1971 and 1992,10 decile portfolios are formed in asc ending orde rbased on the ra tio of the book value of equity to m ark et v alue of equity (BM). The glamour portfoliorefers to the decile portfolio containing stocks ranking lowest on BM. The value portfolio refers tothe decile portfolio containing stocks ranking highest on BM. The returns presented in the tableare averages over all formation periods. Panel A contains (equally-weighted) earnings announce-ment returns for each portfolio. These are measured quarterly over a 3-day window (i - 1, + 1)around The Wall Street Journal publication date and then summed up over the four quarters ineach of the first five post-formation yea rs (Q01-Q04 Q17-Q20). Pane l B contains the(equally-weighted) size-adjusted earnings announcement returns. For each stock in the portfolio,size-adjusted earnings announcement returns are obtained by subtracting off the earnings an-

nouncement return on a benchmark portfolio consisting of stocks in the same size decile. Panel Ccontains (equally-weighted) annual portfolio returns in year t after formation, = 1, 2, 3, 4, 5.Panel D contains (equally-weighted) size-adjusted annual portfolio returns. For each stock in theportfolio, size-adjusted annual returns are obtained by subtracting off the annual return on abenchmark portfolio consisting of stocks in the same decile.

BM

Q01-Q04Q05-Q08Q09-Q12Q13-Q16Q17-Q20

Q01-Q04Q05-Q08Q09-Q12Q13-Q16Q17-Q20

Y RlYR2YR3Y R 4Y R 5

Y R lYR2Y R 3Y R 4

Glamour1

-0.00472-0.004280.003120.008040.00424

-0.01595-0.01484-0.00822-0.00296-0.00484

0.092540.092840.119790.130630.12274

-0.07810-0 .08011-0.06160-0.06130

2 9Panel A: Event

0.007720.006880.007960.008120.01024Panel B:

-0.00334-0.00419-0 .00411-0.003180.00062

0.032000.028280.024920.021760.01368

Size-Adjusted0.015330.011850.008120.005780.00013

Panel C: Annual0.148110.145900.148350.168360.17032

Panel D:-0.02196-0.02824-0.03947-0.03217

0.225340.200850.241950.231490.22329

Size-Adjusted.0.044120.015690.034020.00610

Value10

Returns0.035320.030120.031360.026440.02432

MeanDifference10-1

0.040040.034400.028240.018400.02008

Event Returns0.016100.012160.013410.009450.00987Returns0.215470.219710.244960.251410.23518

0.032050.026990.021620.012400.01471

0.122920.126860.125170.120780.11244

Annual Returns0.032130.032790.034260.02467

0.110230.112890.095850.08597

i-Stat for MeanDifference 10-1

5.657.145.123.674.49

5.035.904.183.053.39

3.843.884.273.823.11

3.503.914.002.78


7/17

Good New s for Value Stocks: Evidence on Market E fficiency 865Table II contains analogous numbers for the (CP, GS) classification. Theresults are similar. In year + 1 , the difference in event returns between valueand glamour portfolios is +3.2 percent, which represents approximately 27

percent of the difference in total year + 1 retu rns. This difference is significantat the 1 percent level. The difference in event retu rns is still 2.0 percent in year+3 and represents approximately 20 percent of the 9.6 percent difference inannual returns between the two portfolios. Interestingly, the differences inboth event returns and an nua l buy-and-hold retu rns die out more rapidly overtime using the (CP, GS) classification. In fact, deterioration of annual returndifferences for the (CP, GS) classification is much more pronounced here thanin the original LSV article. We believe that this is due to the addition ofNasdaq firms as well as to a different sample period. In any case, the peteringout of the earnings surprises is consistent with the petering out of annualre turn differences between th e two portfolios. This is especially evident in thesize-adjusted annual return differences, where year +5 return differences aresignificantly less than half those for year + 1 .

Reconciling the time pa ttern of earnings surprises with the time pattern ofannual buy-and-hold returns is an interesting exercise. The finding that pos-itive relative earnings surprises for value stocks, while relatively long-lived,appear to die out faster than annual buy and hold return differences suggeststhat earnings surprises are not the whole story behind the superior returns tovalue stocks. Other behavioral and institutional factors may play a role in thesuperior retu rns to value strategies (LSV, 1994).So far, we have focused on earnings announcement results for our entireNYSE/AMEX/Nasdaq universe of firms. One interesting question is whetherthese results apply equally well to the larger firms that are more closelyfollowed by market participants. These stocks would presumably be less vul-nerable to the sort of mispricing discussed by LSV (1994). In fact, in a sampleof NYSE and AMEX stocks only, LSV (1994) dofind hat the return differencesbetween value and glamour stocks are approximately 30 percent lower on asubsample of firms with market capitalization above the median for theNYSE/CRSP universe.Tables III and IV present numbers analogous to those of Tables I and II forthe subsample of the largest firms with m arket capitalization above the NYSEmedian in the year of portfolio formation. The difference in earnings announce-ment returns between value and glamour firms is substantially lower than inthe full sample and also accounts for a lower fraction of the annual returndifference between value and glamour portfolios. For example, using the book-to-market classification in Table III, we see that the earnings announcementreturn difference between the value and glamour portfolios is 1.0 percent inyear + 1 , 1.5 percent in year + 2, and 1.2 percent in year + 3. These differencesrepresen t approximately 12 percent, 14 percent, and 11 percent respectively of


8/17

866 The Journal of FinanceTable IIEarning s Announcem ent Re turns and Annual Buy-and-Hold Re turnson Value and Glamour Portfolios Classified by Cash-Flow-to-Priceand Growth-in-Sales, 1971-1992 (Full Sample)At the end of each June between 1971 and 1992, 9 groups of stocks are formed. The stocks areindependently sorted in ascending order into 3 groups ((1) bottom 30 percent, (2) middle 40percent, and (3) top 30 percent) based on each of two variables, the ratio of cash flow to marketvalue of equity (CP) and preformation 5-year average growth rate of sales (GS). The value portfoliocontains stocks ranked in the top group (3) on CP and in t he bottom group (1) on GS. The glamou rportfolio contains stocks ranked in the bottom group (1) on CP and in the top group (3) on GS. Thereturns presented in the table are averages.over all formation periods. Panel A contains (equally-weighted) earnings announcement returns for each portfolio. These are measured quarterly overa 3-day window (i - 1, + 1) arou nd The Wall Street Journal publication date and then summedup over the four q uar ters in each of the first five post-formation yea rs (Q 01-Q04, . . . , Q17-Q20 ).

Panel B contains the (equally-weighted) size-adjusted earnings announcement returns. For eachstock in the portfolio, size-adjusted earnings announcement returns are obtained by subtractingoff the earnings announcement return on a benchmark portfolio consisting of stocks in the samesize decile. Panel C contains (equally-weighted) ann ual portfolio re tur ns in year after formation, = 1, 2, 3, 4, 5. Panel D contains (equally-weighted) size-adjusted annual portfolio returns. Foreach stock in the portfolio, size-adjusted ann ual re tur ns are obtained by subtrac ting off the ann ua lreturn on a benchmark portfolio consisting of stocks in the same decile.

GPGSQ01-Q04Q05-Q08Q09-Q12Q13-Q16Q17-Q20Q01-Q04Q05-Q08Q09-Q12Q13-Q16Q17-Q20YRlYR2YR3Y R 4Y R 5YRlYR2

Glamour13

-0 .000190.001220.005810.008430.00898Panel

-0.01130-0.00997-0.00526-0.002020.00025

0.117900.123490.139790.157570.15758Panel 1

-0.04562-0.04064

Value31 MeanDifference

Panel A: Event Returns0.032010.029220.025890.020560.01966

B: Size-Adjusted0.012850.011120.008640.004440.00567

Panel C: Annual0.237000.243330.235340.244520.22269

0.032200.028000.020080.012130.01068Event Returns

0.024150.021100.013890.006460.00542

Returns0.119090.119830.095550.086950.06510

y . Size-Adjusted Annual Returns0.051020.05436 0.096630.09499

i-Stat for MeanDifference6.624.144.203.692.895.313.213.082.451.664.254.173.992.832.133.563.62


9/17

Good New s for V alue Stocks: Evidence o n Market Efficiency 867Table IIIAnnual Cumulative Earnings Announcemen t R eturns and Annual

Buy-and-Hold Return s on Value and Glamour Portfolios Classifiedby Book-to-Market Ratios, 1971-1992 (Firms withMarket Cap > ^fYSE Median)At the end of each Ju ne b etween 1971 and 1992 ,10 decile portfolios a re formed in ascending orderbased on the ratio of the book value of equity to m ark et value of equity (BM) from among all stockswith market capitalization greater than the New York Stock Exchange (NYSE) median. Theglamour portfolio refers to the decile portfolio containing stocks ran kin g lowest on BM. The va lueportfolio refers to the decile portfolio containing stocks ranking highest on BM. The returnsprese nted in the table a re averag es over all formation periods. Panel A contains (equally-weighted)earnings announcement returns for each portfolio. These are measured quarterly over a 3-daywindow (i - 1, + 1) arou nd The Wall Street Journal publication date and then summed up overthe four qu art ers in each of the first five post-formation y ears ( Q0 1-Q 04,. . . , Q17-Q20). Panel Bcontains the (equally-weighted) size-adjusted earnings announcement returns. For each stock inthe portfolio, size-adjusted earnings announcement returns are obtained by subtracting off theearnings announcement return on a benchmark portfolio consisting of stocks in the same sizedecile. Panel C contains (equally-weighted) annual portfolio returns in year t after formation, t =1, 2, 3, 4, 5. Pan el D co ntains (equally-weighted) size-adjusted an nua l portfolio re tur ns . For eachstock in the portfolio, size-adjusted annual returns are obtained by subtracting off the annualreturn on a benchmark portfolio consisting of stocks in the same decile.

BMGlamour

1 2 9Panel A:

Value10

Event Returns

MeanDifference1 0 -1-Stat for MeanDifference10-1

Q01-Q04Q05-Q08Q09-Q12Q13-Q16Q17-Q20Q01-Q04Q05-Q08Q09-Q12Q13-Q16Q17-Q20Y R lY R 2YR3Y R 4Y R 5Y R lY R2

0.003150.001890.002650.004740.00230

-0 .00417-0 .00561-0.00566-0.00290-0 .003210.118500.094560.116300.120530.10921

-0 .03286-0 .06261

0.009760.006620.006490.006330.00569Panel B:0.00267-0 .00056

-0.00176-0.001100.00021

0.018400.018190.013410.007570.00498

Size-Adjusted0.011180.010600.005450.00019-0 .00091

Panel C: Annual0.138550.134420.140400.155110.15368Panel D:

-0.01312-0 .02261

0.178100.182200.199850.181500.20022

Size-Adjusted0.023340.02557

0.013480.017170.014680.011720.00182

0.010330.015280.012030.00698-0 .00048

Event Returns0.004760.009460.007410.00470-0 .00346

Returns0.198980.203410.224620.212960.20082

0.008930.015080.012960.00760-0 .000250.080470.108840.108310.092430.09160

Annual Returns0.042110.04220

0.074970.10481

0.802.091.550.93-0 .0 80.692.061.711.03-0 .0 41.772.832.973.322.761.682.84


10/17

868 The Journal of FinanceTable IVEarn ings Announcem ent Retu rns and Annual Buy-and-Hold Re turns

on Value and Glamour Portfolios Classified by Cash-Flow-to-Priceand Growth-in-Sales, 1971-1992 (Firms withMarket Cap > NYSE Median)At the end of each June between 1971 and 1992, 9 groups of stocks are formed from among allstocks with m arke t capitalization gre ater t ha n the New York Stock Exchange (NYSE) me dian. Thestocks are inde penden tly sorted in ascending order into 3 groups ((1) bottom 30 perce nt, (2) middle40 percent, and (3) top 30 percent) based on each of two variables, the ratio of cash flow to m ark etvalue of equity (CP) and preformation 5-year average growth rate of sales (GS). The value portfoliocontains stocks ranked in the top group (3) on CP and in the bottom group (1) on GS. The glamourportfolio contains stocks ranked in the bottom group (1) on CP and in the top group (3) on GS. Thereturns presented in the table are averages over all formation periods. Panel A contains (equally-weighted) earnings announcement returns for each portfolio. These are measured quarterly overa 3-day window (i - 1, + 1) aro und The Wall Street Journal publication date and then summedup over the four qua rte rs in each of the first five post-formation y ears (Q 01- Q0 4,. . ., Q17-Q20).Panel B contains the (equally-weighted) size-adjusted earnings announcement returns. For eachstock in the portfolio, size-adjusted earnings announcement returns are obtained by subtractingoff the earnings announcement return on a benchmark portfolio consisting of stocks in the samesize decile. Pane l C contains (equally-weighted) a nnu al portfolio ret urn s in year t after formation, = 1, 2, 3, 4, 5. Panel D contains (equally-weighted) size-adjusted annual portfolio returns. Foreach stock in the portfolio, size-adjusted an nua l ret urn s are obtained by subtrac ting off the ann ua lreturn on a benchmark portfolio consisting of stocks in the same decile.

GPGSQ01-Q04Q05-Q08Q09-Q12Q13-Q16Q17-Q20Q01-Q04Q05-Q08Q09-Q12Q13-^16Q17-Q20YRlYR2Y R 3YR4YR5YRlYR2

Glamour13

0.004560.002450.004450.003740.00388-0.00252-0.00487-0.00278-0.00308-0.00089

0.118400.110890.118570.135510.12846-0.02959-0.03972

Value31

Panel A: Event0.016830.026340.013370.007980.00459

Panel B: Size-Adjusted0.008720.018130.005440.00020-0.00145

Panel C; Annual0.199500.192570.197730.200170.20431

MeanDifferenceReturns

0.012280.023890.008920.004240.00072Event Returns

0.011240.023000.008210.00328-0.00056Returns

0.081090.081680.079160.064650.07584Panel D: Size-Adjusted Annual Returns

0.047220.03650 0.076800.07622

i-Stat for MeanDifference

2.133.801.250.830.16

2.063.681.140.63-0 . 132.582.613.452.112.572.572.55


11/17

Good New s for Value Stocks: Evidence on M arket E fficiency 869represen t approximately 33 percent, 27 percent, and 22 percent of tbe annualreturn differences reported in Panel C of Table L

One interpretation of these results is tha t tbe pricing of the larger firms ismore efficient, leaving less systematic bias in tbe earnings surprises for valueversus glamour firm s. On tbe other hand, since these firms are followed moreextensively by analysts and get much more coverage in the financial press, itmay just be that a greater fraction of fundamental news about tbe larger firmsis impounded into prices outside of quarterly earnings announcements.A related problem is that of late earnings announcements. It is widelybelieved th at when a firm does not announce earn ings when it is expected to,tbe news is more likely to be bad. Hence, for firms announcing late, tbeearnings news may dribble out before tbe actual announcement date as tbemarket realizes that "tbe dog hasn't barked." This could be a source of poten-tial bias in our results if value firms systematically announce bad earnings latemore often than glamour firms, and tbis channel for bad news to be commu-nicated to the mark et is not captured in tbe ( - 1 , +1) window around tbeearnings announcement.To evaluate tbis possibility, we define firms announcing late as tbose an-nouncing more than 2 trading days after the calendar date of tbe announce-ment for the previous year (tbe expected date). For these firm-quarters, in-stead of using tb e ( - 1 , +1) return, we plug in tbe re turn from 4 days before theexpected date to 1 day a fter the actual announcement date to capture the effectof the m arket's learning from tbe announcement delay. While tbis introducesadditional noise, the results give some indication of whether a delayed an-nouncement b ias can account for our resu lts. For tbe BM classification in TableI, we find that 32 percent of value firms and 29 percent of glamour firmsannounce late. When we plug in the return over the extended window for allfirms announcing late, we get an annual earnings announcement return fortbe value firms in year -1-1 of 3.2 percent compared to the 3.5 percent returnreported in Table I, indicating that tbe bias adjustment makes little difference.Extending the event window for the late announcing glamour firms, we get anaverage event return of +0.4 percent compared to the -0.5 percent returnreported in Table I. Extending the event window for late announcing glamourfirms actually increases the event retu rns , which is contrary to the theory th atlate announcement typically m eans bad news. After extending tbe event win-dow for late announcers, tbe event return difference between value and glam-our firms is still +2.8 percent witb a standard error of 1.1 percent.

Wben we perform the same procedure for late announcers on the (CP, GS)portfolios of Table II, we obtain sim ilar results. Extending tbe window for lateannouncers results in an event return of +3.5 percent for tbe value portfolio inyear +1, compared to +3.2 percent reported in Table II, and an event returnof -0.3 percent for tbe glamour portfolio, compared to 0.0 percent in Table II.


12/17

870 The Journal of FinanceAs with most event studies, our measure of earnings surprises focuses onactual announcement dates that are often not availahle to investors in ad-

vance. This means that the announcement returns we measure are not real-izable as part of an implementable trading strategy. For readers interested inan implementable trading strategy, we do have resu lts for a strategy th at doesnot require advance knowledge of the announcement dates. We begin byestimating the expected announcement date simply as the calendar date of theannouncement for the previous year. We then measure event returns begin-ning four days before the expected date, but we only include firms that havenot already announced earlier than day - 4 (since this is observable by inves-tors at the time). Returns are cumulated through one day after the actualannouncement takes place. This event window (expected date - 4 , actual da te+ 1) is of variable length depending on the timing of the actual announcement.The median holding period is five days for all of the BM and (CP, GS)portfolios, w ith a mean of seven days for th e BM portfolios and eight days forthe (CP, GS) portfolios. The results for this trading strategy are consistentwith the results obtained in Tables I and II. For the BM classification, the year+ 1 earnings announcement re turns are 2.2 percent for the value portfolio and0.4 percent for the glamour portfolio, with a difference of 1.8 percent and astandard error of 1.3 percent. For the (CP, GS) classification, the returns are2.7 percent for the value portfolio, -0 .3 percent for the glamour portfolio, witha difference of 3.0 percent and a standard error of 0.7 percent. These resultsare consistent with our earlier findings and suggest th at the earlier results a renot driven by a selection bias from using ex post announcement dates .

In sum, the evidence indicates that a significant portion of the returndifference between value and glamour stocks is attributable to earnings sur-prises that are systematically more positive for value stocks.III. Do Differences in Event Returns Represent Differences in RiskPremia?

An arden t defender of the risk premium story might argue a t this point thatthe foregoing event returns evidence is inconclusive. The sizeable differencesin event returns between value and glamour portfolios may just representdifferences in ex ante risk premia realized around a small number of importan tinformation events. If a disproportionately high fraction of the annual uncer-tainty about a stock is realized around quarterly earnings announcements,then perhaps a disproportionate share of the risk premium is as well. Since therisk premium for value stocks is higher, these stocks should exbibit higherevent returns than glamour stocks.Do the data support such an explanation? It does not appear so. Recall that

the return on glamour stocks around earnings announcements in both years+ 1 and +2 after portfolio formation is actually negative. This is clearly not


13/17

Good News for Value Stocks: Evidence on Ma rket E fficiency 871Table VCross-Section R egression Tests of Difference b etw een Event andNonevent Returns for Value and Glamour Portfolios, 1971:2-1993:1(Full Sample)

For each quarter in the sample (1971:2-1993:1), we run cross-sectional regressions of the dailyreturn for each portfolio on the Center for Research in Securities Prices (CRSP) value-weightedma rket re turn and a dummy variable for whe ther the day belongs to the ( - 1 , +1) window aroundthat quarter's earnings announcement. Regressions are run separately for value and glamourportfolios, with new value and glamour portfolios formed at the end of each June. As in Fama-MacBeth (1973), the coefficients reported are the averages of the coefficients from the 88 quarterlycross-sectional regressions with standard errors computed according to the time series of thosecoefficients. In Panel A, portfolio formation is based on the book-to-market ratio (BM). Theglamour portfolio consists of firms in the bottom decile based on BM and the value portfolioconsists of firms in the top decile according to BM. In Panel B, portfolios are formed on the ratioof cash flow to market value (CP) and on the preformation 5 year growth rate in sales (GS). Theglamour portfolio consists of firms ran ked in the bottom 30 percent on CP an d in the top 30 p ercenton GS. The value portfolio consists of those stocks ranked in the top 30 percent on CP and in thebottom 30 percent on GS.

Intercept Event Day DummyPanel A: Regressions for Portfolios Formed on BM

Low BM Portfolio Re tur n 0.000128(Glamour) (2.00)High BM Portfolio Re tur n 0.001104(Value) (6.77)

Low CP, High(Glamour)High CP, Low(Value)

-0.000661(-3.44)0.001945(5.45)

Pan el B: Regressions for Portfolios Formed on (CP, GS)GS Portfolio Re turn 0.000161(2.40)GS Portfoho Re turn 0.000764(7.35)

-0.000399(-2.56)0.001769(7.05)

Market Return

1.0670(73.08)0.6502(30.67)1.0276(76.12)0.6751(32.30)

stocks is equal to the T-bill return. There is a more powerful test of the riskpremium hypothesis, however. The risk premium view, that maintains thatearnings announcement days contain a disproportionately large fraction of astock's risk premium, implies that, for botii glamour and value stocks, eventreturns should be higher than nonevent returns. In contrast, if the behavioralview is correct, and the information revealed about glamour stocks on eventdays is sufficiently negative, event returns could be significantly lower thannonevent returns, despite a higher ex ante risk premium. A comparison ofevent returns and nonevent returns for glamour stocks can potentially dis-criminate between these two views.Table V presents the numbers for this test using both BM and (CP, GS)classifications. For each quarter in the sample, we run cross-sectional regres-sions of the daily return for each stock on the value-weighted market return


14/17

872 The Journal of Financethe end of each Jun e. As in Fama-MacBeth (1973), the coefficients reportedare the averages of the coefficients from the 88 quarterly cross-sectionalregressions (1971:2-1993:1) with standard errors computed according to thetime series of those coefficients.We begin with the results for the BM classification. Regressions for low BM(glamour firm s) show an intercept of 1.3 basis points per day and a coefficientof 1.07 on the market return (beta). More importantly, the coefficient on theevent dummy is -6.6 basis points per day with a standard error of 1.9 basispoints. On an annualized basis, this difference in return is approximately 16percent per year. Event day returns are significantly below nonevent dayreturns despite the higher ex ante risk premium that should be required tohold stocks over event days. This result can only be explained by the hypoth-esis that, on average, the market receives negative surprises for glamourstocks on earnings announcement days. Results for value stocks are also quiteinteresting. The mean event retu rn for value stocks on event days is 19.4 basispoints higher than the nonevent return with a standard error of 3.6 basispoints. On an annualized basis this difference is approximately 50 percent peryear. This result is consistent with either a very high risk premium realized onevent days or positive earn ings surp rises for value stocks or some combinationof the two.

Results for the (CP, GS) classification are similar, but less dramatic. Re-gressions for glamour stocks (low CP, high GS) show an intercept of 1.6 basispoints per day, and a coefficient on the market return of 1.03 (beta). Theestimated return on event days is 4.0 basis points below the return on non-event days, with a standard error of 1.6 basis points. Regressions for valuestocks show an estimated event day return tha t is 17.7 basis points above thenonevent day return with a standard error of 2.5 basis points.In sum, comparisons of event and nonevent day returns do not support therisk premium explanation of the superior return on value stocks. The riskpremium hypothesis implies that event returns should be higher than non-event returns for both glamour and value stocks. The data show that eventretu rns are lower tha n nonevent re turns for glamour stocks despite the higherex ante risk premium posited by the theory. This can only be explained bynegative earnings surprises for glamour stocks.

rV. ConclusionThe evidence in this article suggests that expectational errors about futureearnings prospects play an important role in the superior return to valuestocks. Postformation earnings announcement returns are substantiallyhigher for value stocks than for glamour stocks. Event returns for glamour

stocks are significantly lower than glamour returns on an average day, whichis inconsistent with the risk premium explanation for the re turn differences


15/17

Good News for Value Stocks: Evidence on M arket Efficiency 873glamour stocks in the first two to three years after portfolio formation andapproximately 15-20 percent of return differences over years four and fiveafter formation. Results for firms larger than the NYSE median are weaker,which may be due to a tendency of widely-followed stocks to adjust to newsmore continuously rather than have information events heavily concentratedon quarterly earnings announcement days.The persistence of positive relative earnings surprises for value stocks longafter portfolio formation is consistent with the finding of various researchersthat the superior returns to value stocks persist long after portfolio formation.However, the magnitude of earnings surprises does diminish more rapidlythan the annual return differences, indicating that learning about futureearnings prospects may not explain all of the difference in returns betweenvalue and glamour stocks.What does explain the long-lived component of these differences in averagereturn? LSV (1994) suggest various possibilities. First, investors may simplyhave a preference for investing in "good" companies with high levels of profit-ability and superior management. Unsophisticated investors may equate agood company with a good investment irrespective of price. They may evenperceive such stocks to be less risky, as allegedly w as the case with IBM beforeinvestors were exposed to its vulnerability. Finally, sophisticated institutionalinvestors may gravitate toward well-known, glamour stocks because thesestocks are easier to justify to clients and superiors as prudent investments.Although a complete and satisfying explanation for the superior return tovalue stocks is beyond the scope of the present article, our evidence suggeststhat behavioral factors (and expectational errors in particular) play an impor-tant role.

REFERENCESBanz, Rolf W., and William J. Breen, 1986, Sample dependent results using accounting andmarket data: Some evidence. Journal of Finance 41, 779-793.Basu , S., 1977, Inve stme nt performance of common stocks in relation to thei r price earn ings r atios:A test of the efficient markets hypothesis. Journal of Finance 32, 663-682.Chan, Louis K. C , Yasushi H amao, and Josef Lakonishok, 1991, Fund am entals in stock re turn sin Japan, Journal of Finance 46, 1739-1764.Chan, Louis K. C , N arasim han Jegadeesh , and Josef Lakonishok, 1995, Evaluating the perfor-mance of value versus glamour stocks: The impact of selection hias. Journal of FinancialEconomics 38, 269-296.Chopra, Navin, Josef Lakonishok, and Jay R. Ritter, 1992, Measuring abnormal performance: Dostocks overreact?. Journal of Financial Econom ics 31, 235-268.Daniel, Kent, and Sheridan Titman, 1997, Evidence on the characteristics of cross-sectionalvariation in stock returns, Journal of Finance 52, 1-34.Davis, James L., 1994, The cross-section of realized stock returns: The pre-COMPUSTAT evi-dence. Journal of Finance 49, 1579-1594.De Bondt, Werner F. M., and Richard H. Thaler, 1985, Does the stock market overreact?. Journalof Finance 40, 793-805.


16/17

874 The Journal of Finance

Fama, Eugene, and James D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests.Journal of Political Economy 81, 607-636.Jaffe, Jeffrey, Donald B. Keim, and Ran dolph Westerfield, 1989, Ea rnin gs yields, mar ket values,and stocks returns. Journal of Finance 44, 135-148.Kotha ri, S. P., Jay Sha nken , and Richard G. Sloan, 1995, Ano ther look at the cross-section ofexpected stock returns. Journal of Finance 50, 185-224.La Po rta, Rafael, 1996, Expec tations and the cross-section of stock retu rns . Journal of Finance 511715-1742.Lakonishok, Josef, Andrei Shleifer, and Robert Vishny, 1994, Contrarian investment, extrapola-tion, and risk. Journal of Finance 49, 1541-1578.Little, L M. D., 1962, Higgledy piggledy growth. Bulletin of the Oxford University Institute ofEconomics and Statistics 24, 387-412 .Me rton, Robert, 1973, An intertem pora l a sset pricing model, Econometrica 41, 867-887.Rosenberg, Barr, Kenneth Reid, and Ronald Lanstein, 1984, Persuasive evidence of marketinefficiency. Journal of Portfolio Management 11, 9-17.


17/17

Good News for Value Stoks

Documents

Transcript of Good News for Value Stoks