The Death of the Overreaction Anomaly? A Multifactor Explanation of Contrarian Returns

Transcription

1 The Death of the Overreacton Anomaly? A Multfactor Explanaton of Contraran Returns Adam Clements *, Mchael E. Drew *,# Evan M. Reedman *, and Madhu Veeraraghavan^ ABSTRACT Are the returns accrung to De Bondt and Thaler s (1985) (DT) much celebrated overreacton anomaly pervasve? Usng the CRSP data set used by for the perod 1926 through 1982, and, for the frst tme, an addtonal two decades of data (1983 through 2003), we provde prelmnary support for the orgnal work of DT, reportng that the overreacton anomaly has not only perssted over the past twenty years but has ncreased, on a rsk-unadjusted bass. However, usng the three factor model of Fama and French (1993) (FF), we fnd no statstcally sgnfcant alpha can be garnered va the overreacton anomaly, wth contraran returns drven by the factors of sze and value, not the behavoral bases of nvestors. It s our conjecture that the anomaly s not robust under the FF framework, wth contraran nvestors followng such a scheme smply compensated for the nherent portfolo rsk held. JEL Classfcatons: Keywords: G11, G12, G14 Overreacton, anomaly, multfactor asset prcng model. # Correspondng author: Correspondng author: Tel: ; Fax ; Emal: m.drew@qut.edu.au; Mal: School of Economcs and Fnance, QUT, GPO Box 2434, Brsbane, Queensland, Australa, *Queensland Unversty of Technology (Clements, Drew & Reedman) and ^Monash Unversty (Veeraraghavan). Reedman acknowledges support provded by the School of Economcs and Fnance, QUT and the Bran Gray Scholarshp (jontly funded by the Australan Prudental Regulaton Authorty and the Reserve Bank of Australa). The authors thank partcpants at the FIRN Doctoral Tutoral 2005, the 18th Australasan Fnance & Bankng Conference, the 5th Global Conference on Busness & Economcs and the 2006 Busness & Economcs Socety Internatonal Conference for helpful comments. All errors reman the sole responsblty of the authors.

2 The Death of the Overreacton Anomaly? A Multfactor Explanaton of Contraran Returns ABSTRACT Are the returns accrung to De Bondt and Thaler s (1985) much celebrated overreacton anomaly pervasve? Usng the CRSP data set used by for the perod 1926 through 1982, and, for the frst tme, an addtonal two decades of data (1983 through 2003), we provde prelmnary support for the orgnal work of De Bondt and Thaler, reportng that the overreacton anomaly has not only perssted over the past twenty years but has ncreased, on a rsk-unadjusted bass. However, usng the three factor model of Fama and French (1993), we fnd no statstcally sgnfcant alpha can be garnered va the overreacton anomaly, wth contraran returns drven by the factors of sze and value, not the behavoral bases of nvestors. It s our conjecture that the anomaly s not robust under the Fama- French framework, wth contraran nvestors followng such a scheme smply compensated for the nherent portfolo rsk held. JEL Classfcatons: Keywords: G11, G12, G14 Overreacton, anomaly, multfactor asset prcng model. 2

3 The Death of the Overreacton Anomaly? A Multfactor Explanaton of Contraran Returns The debate surroundng nvestor overreacton and contraran nvestng s one of the most extensve and controversal areas of research n fnance. Despte the fact that there s a general agreement on the evdence of prce reversal, there s no consensus about what s drvng these reversals. From an nvestment management perspectve, the concern regardng contraran strateges relates to ssues of portfolo rsk and the ablty of the anomaly to generate alpha. In the sprt of recent work scrutnzng or dssectng anomales (see Fama and French, 2006), we revst the overreacton anomaly reported by De Bondt and Thaler (1985), updatng the ntal study wth a further two decades of data. Usng a multfactor asset prcng framework, we fnd that contraran returns, partcularly past losers, consstently load on sze and the value factors at economcally meanngful levels (wth past wnners loadng predomnantly on the value factor). It s our conjecture that nvestors followng such a scheme are smply compensated for the nherent portfolo rsk held. THE OVERREACTION CONTROVERSY The overreacton anomaly, evdenced by long-term reversals n stock returns, was frst dentfed by De Bondt and Thaler (1985), who showed that stocks whch perform poorly n the past three to fve years demonstrate superor performance over the next three to fve years compared to stocks that have performed well n the past. The orgnal study performed by De Bondt and Thaler (1985), hereafter DT, enttled Does the stockmarket overreact?, provded evdence that abnormal excess returns could be ganed by employng a strategy of buyng past losers and sellng short past wnners, or the contraran strategy. Usng an array of data for dfferent tme perods and n dfferent markets, support for the fndngs of DT has been provded by, among others, Howe (1986), Fama and French (1988), Poterba and Summers (1988), Chopra, Lakonshok and Rtter (1992) and Campbell and Lmmack (1997). 3

4 Soon after the publcaton of DT, Chan (1988) argued that the work lacked approprate rsk adjustment, and demonstrated that the sngle-factor CAPM had some explanatory power for the returns generated by DT. As asset prcng models developed, Fama and French (1993, 1995, and 1996) showed the relevance of sze and value factors n explanng the crosssecton of stock returns, however, to ths day overreacton studes contnue to gnore ths work n ther methodologcal approach to the anomaly. Ths appears to be a vtal concern, and one whch ths work seeks to rectfy. Further consderaton of the lterature followng DT reveals that overreacton studes are subject to a number of crtcsms. Frst, there s a lack of rsk adjustment n the orgnal study (Chan, 1988; Ball and Kothar, 1989). Second, the mpact of the January effect on returns s not adequately dealt wth (Zarown, 1990). Fnally, there s an ongong dscusson around the role of measurement bases n the sortng and testng perods (Conrad & Kaul, 1993). Our paper drectly consders the mpact of each of these ssues for the U.S. settng from 1926 through 2003 (a further two decades of data followng DT s observaton wndow), fndng that, on rsk-adjusted bass, no statstcally sgnfcant alpha can be garnered through the varous approaches that attempt to explot the overreacton anomaly. To analyse the evdence for long-term reversals, we use the monthly return data from the Centre for Research n Securty Prces (CRSP), the same data set used n the orgnal DT study, for the perod of January 1926 through December 2003 and buld portfolos every perod of the best (wnner) and worst (loser) performng stocks n the prevous n months. The equally-weghted CRSP market ndex s used as our market proxy (a descrpton of the sortng approach s provded n Fgure 1, Appendx A.) We then record the cumulatve average monthly return to these self-fnancng portfolos over our sample perod. 4

5 DECOMPOSING CONTRARIAN RETURNS A. Out-of-sample test of De Bondt and Thaler (1985) The results from the recent sub-perod ( ) provde corroboratng evdence of the overreacton hypothess, and, nterestngly, demonstrate that the magntude of the anomaly, on a rsk-unadjusted bass, has actually ncreased through tme. Durng the perod January 1983 to December 2003, the loser portfolos outperform the market, on average, by 53.7%, 36 months after formaton. The wnner portfolos underperform the market by, on average, 4.03%. These results are dsplayed n Fgure Fgure 2 Average CAR of portfolos formed for three-year sort and test perods for recent sub-perod January 1983 and December C A R Months after portfolo formaton Loser Portfolo Wnner Portfolo 5

6 Examnng the full dataset from 1926 to 2003 shows amplfcaton of the anomaly on a rskunadjusted bass, and reveals that f DT were to present the results of ther study today, they would report a dfference n the ACAR s of the wnner and loser portfolos of 42.5%, over 50% larger than that reported n 1985! Ths amplfcaton of the overreacton anomaly suggests that suggests that the overreacton anomaly s, perhaps, alve and well. B. Evdence n Favour of Rsk Adjustment Understandably, DT has been extensvely crtczed for focusng on market-adjusted returns. By any metrc, portfolo managers are constantly focusng on the rsk-adjusted return of ther nvestments. Hence, the core of our study apples varous technques to adjust for rsk usng four technques: frst, by apprasng a sutable asset prcng model; second, through an nvestgaton of the approprateness of beta estmates; thrd; by allowng for the well accepted return premum to small companes; and, fnally, consderng the results n lght of the January effect. 1. In examnng overreacton, Chan (1988) proposes that the rsks of wnner and loser stocks do not reman constant over the combned tme perod of sortng and testng. 2 Ths lne of argument suggests that strkng changes n the rsks of the portfolos, whch are not accounted for n the DT study, assst n explanng the returns from the strategy. Research by such as Chan (1988), Brown, Harlow and Tnc (1988) and Ball and Kothar (1989) show that the when beta s estmated on the approprate test perod, rather than the formaton perod, the strategy earns economcally nsgnfcant abnormal returns. Usng ths method, we model the tme-varyng rsk coeffcents n the data, wth the results presented n Table 1. Our results corroborate the fndngs of earler work, for all tme perods examned, and hence the asset prcng tests for our study are run wth the coeffcents estmated from the test perod. 1 A summary of the sze effect s provded by Schwert (1983), wth further detal n Banz (1981), Regnganum (1981). Kem (1983), Regnganum (1983) and Haug and Hrschey (2006) provde a dscusson of the January effect. 2 The premse of the crtcsms n Chan s (1988) paper are that f beta s estmated n the sort-perod an there s no attempt to model changes n rsk, the estmated beta wll be a based estmate of the beta n the test-perod. Snce the rsk of the loser portfolo ncreases n the sort-perod, the sort-perod beta underestmates the test-perod beta. 6

7 Table I Rsk-Change Test for a 35 stock portfolo aganst an ndex constructed from the CRSP dataset for the perod Tests for abnormal returns under the assumpton that the sort-perod and test-perod betas are not equal Intercept estmates wth t-statstcs from the Chan (1988) model: (R,t - Rf t ) = α 1, (1 D) +α 2, D t + β 1, (Rm t - Rf t )(1-D)+ β 2, ( Rm t - Rf t )D t + ε,t. T-statstcs are n parentheses. Statstcal sgnfcance s denoted at: 1% - **; 5% - *; 10% - #. Losers Sort Perod Test Perod α α β β D Adj R ** ** 0.417* (-5.324) (-0.026) (10.377) (2.505) ** ** 0.307* (-5.209) (0.344) (12.260) (2.060) ** ** (-3.710) (0.409) (10.379) (1.730) Wnners Sort Perod Test Perod α α β β D Adj R ** ** * (6.025) (-0.422) (15.285) (-2.047) ** ** * (5.928) (-0.791) (17.392) (-2.161) ** ** (4.817) (-0.623) (13.044) (-0.875) C. Evdence n Favour of the Three-Factor Model The work of Fama and French (1993, 1996) has demonstrated the relevance of sze and value factors when prcng rsky assets. Investment managers are justly mystfed as to why researchers over the last decade contnue to gnore ths n ther methodologcal approach to the anomaly. 3 Ths study mplements the three-factor model developed by Fama and French (1993) (hereafter FF) on the orgnal dataset used by DT, both n-sample and out-of-sample. We consder performance wth the followng equaton: [ R, t R f, t ] = α + β [ Rm, t R f, t ] + σ [ SMBt ] + η [ HMLt ] + ε, t (1) 3 A vew held by Bowman and Iverson (1998), Bauman et al. (1999), Schereck et al. (1999), Gaunt (2000), Kang et al. (2002), Forner and Marhuenda (2003), Hrschey (2003), La et al. (2003) and Ma et al. (2005). 7

8 Table II Three-Factor Regressons of Performance for a 50 stock portfolo aganst a geometrc average ndex, Fama French 3-factor regressons for monthly excess returns on equal-weghted CRSP portfolos of 50 stocks formed on the bass of past returns: Non-overlappng portfolos for the perod January 1926 to December Intercept estmates wth t statstcs (the regresson coeffcent dvded by ts standard error) from the Fama French 3-Factor model: (R t - Rf t ) = α + β (Rm t - Rf t ) + σsmb t + ηhml t + ε,t. The regresson R 2 s are adjusted for the degrees of freedom. T-statstcs are n parentheses. Statstcal sgnfcance s denoted at: 1% - **; 5% - *; 10% - #. Losers Sort Perod Test Perod α β σ SMB η HML Adj R ** # 0.648* (-0.132) (7.954) (1.677) (2.328) ** 1.546** 0.870** (0.334) (8.386) (6.381) (3.481) ** 1.644** 0.944** (0.026) (8.654) (7.736) (4.752) ** 0.558* 0.643** (0.011) (9.925) (2.333) (2.978) ** 0.504* 0.595** (0.226) (9.392) (1.977) (2.869) Wnners Sort Perod Test Perod α β σ SMB η HML Adj R ** * (-0.474) (12.329) (-0.952) (-1.965) ** 0.809** (-0.835) (14.496) (5.564) (1.110) ** 0.679** (-0.559) (18.338) (6.227) (1.359) ** ** * (-0.677) (18.398) (-2.636) (-2.564) ** * * (-0.761) (-2.270) (-2.350) Table II shows that the three-factor model does an admrable job of explanng the return behavor of the contraran portfolos. For the loser portfolos, we obtaned unformly postve, statstcally sgnfcant loadngs on both the sze and value factors. For the wnner portfolos, the sze and value coeffcents are, on the whole, statstcally sgnfcant and negatve. These fndngs suggest that long-term past losers tend to be small, dstressed stocks and that the wnner portfolos comprse larger, growth stocks and therefore the three-factor model predcts that the long-term past wnners wll necessarly not produce hgher average returns. 4 4 Analyss of the two sub-perods presents smlar results to those detaled for the full study. 8

9 Importantly for our study, the FF model s dong a better job than the CAPM (results not shown) n explanng the future returns generated by a contraran strategy. 5 The average R 2 for the loser portfolos s for the three-factor model, up from n our sngle-factor results. Smlarly, the wnner portfolos the average R 2 ncreases from n the CAPM model to n the three-factor model. D. Evdence of the January Effect The fndngs of the orgnal overreacton study were also challenged on the bass of the wellknown January effect. The crtque by Zarown (1990) ncludes substantal dscusson of seasonalty n the overreacton phenomenon. Ths explanaton s supported by Pettengll and Jordan (1990), who show that almost half of the average cumulatve abnormal return for the year n ther 90-stock loser portfolo s generated n January. 6 Smlarly, Chopra, Lakonshok and Rtter (1992) demonstrate that the overreacton effect was dsproportonately concentrated n January [pp. 249]. In order to study the consequences of the January effect n combnaton wth the three-factor prcng model, and to ensure the robustness of the tests of persstence of the overreacton anomaly, our models were adjusted to allow for a January coeffcent. The three-factor model wth the January coeffcent s specfed: [ R t R f t ] = α + β [ Rm t R f t ] + σ [ SMBt ] + η [ HMLt ] + γ [ θ ] + t (2),,,, ε, Results from these tests are presented n Table III, and show that the loser stocks are stll small, dstressed stocks and the January effect only has a margnal nfluence on some of the portfolos. Interestngly, for the wnner portfolos, the market beta coeffcents appear to be capturng the majorty of the returns from these portfolos, and the January effect s not statstcally sgnfcant n any portfolos. The explanatory power of the models ncreases only margnally wth the addton of January, n the loser portfolos by 3.4% and n the wnner portfolos by only 0.3%. 5 These results are avalable on request. 6 De Bondt and Thaler (1987) concede that they have no satsfactory explanaton for the January effects. 9

10 Table III Three-Factor Regressons of Performance for a 50 stock portfolo aganst a geometrc average ndex, wth January Coeffcent, for the full study perod Fama French 3-factor regressons for monthly excess returns on equal-weghted CRSP portfolos of 50 stocks formed on the bass of past returns: Non-overlappng portfolos for the perod January 1933 to December Intercept estmates wth t statstcs (the regresson coeffcent dvded by ts standard error) from the Fama French 3-Factor model: (R,t - Rf t )= α + β (Rm t - Rf t ) + σsmb t + ηhml t + γ (θ ) + ε,t where the dummy varable θ s set to 1 for January and 0 for all other months. The regresson R 2 s are adjusted for the degrees of freedom. T-statstcs are n parentheses. Statstcal sgnfcance s denoted at: 1% - **; 5% - *; 10% - #. Losers Sort Perod Test Perod α β σ SMB η HML γ JAN Adj R ** # # (-0.586) (7.698) (1.505) (1.732) (1.834) ** 1.507** 0.750** (0.231) (8.397) (6.354) (3.252) (1.556) ** 1.543** 0.806** 0.096* (-0.296) (8.946) (7.483) (4.272) (2.489) ** 0.461* 0.520** 0.094* (-0.102) (10.189) (2.063) (2.590) (2.069) ** 0.485^ 0.536* (-0.086) (9.030) (1.936) (2.488) (1.417) Wnners Sort Perod Test Perod α β σ SMB η HML γ JAN Adj R ** (-0.222) (12.16) (-0.746) (-1.533) (-1.030) ** 0.816** (-0.82) (14.549) (5.570) (1.312) (-0.574) ** 0.701** (-0.418) (18.157) (6.320) (1.474) (-0.958) ** * * (-0.581) (18.433) (-2.407) (-2.345) (-1.283) ** * # (-0.353) (17.743) (-2.066) (-1.869) (-1.625) E. Robustness Tests The focus of our study so far, has been on the DT non-overlappng portfolos. Portfolo managers are able to more effectvely operatonalze the contraran strategy by formng portfolos on the bass of overlappng or rollng wndows. Addtonally, t s recognsed that properly specfed tests of tme seres data can acheve greater effcency by the use of overlappng data. 7 In order to account the problem of autocorrelaton that overlappng 7 In hs work on testng the effcent market hypothess, Glbert (1986) recognzed the mportance of usng a full sample of overlappng data, along wth the nherent problems of heteroskedastcty. 10

11 observatons nduces, all results are approprately modfed va the heteroskedastcty and autocovarance consstent estmator of Newey and West (1987) n order to obtan asymptotcally vald hypothess tests. Table IV reports the average results for the three-factor model rollng wndows tests carred out on all the portfolo combnatons prevously dscussed. 8 TABLE IV Rollng Wndow Tests of Robustness for the Three-factor model Fama French 3-factor regressons for monthly excess returns on equal-weghted CRSP portfolos averaged across 20, 35 and 50 stocks formed on the bass of past returns: Rollng Wndow Portfolos for the perod January 1933 to December Intercept estmates wth t statstcs (the regresson coeffcent dvded by ts standard error) from the Fama French 3- Factor model: (R,t - Rf t )= α + β (Rm t - Rf t ) + σsmb t + ηhml t + ε,t. The standard errors are approprately modfed va the heteroskedastcty and autocovarance consstent estmator of Newey and West (1987) n order to obtan asymptotcally vald hypothess tests on the overlappng data. The regresson R 2 s are adjusted for the degrees of freedom. T-statstcs are n parentheses. Statstcal sgnfcance s denoted at: 1% - **; 5% - *; 10% - #. Losers Sort Perod Test Perod α βι σ SMB η HML Adj R ** * (-0.093) (6.406) (1.547) (2.026) ** 0.870* 1.013** (-0.019) (6.803) (1.929) (3.296) ** 0.787* 0.985** (-0.057) (8.032) (1.662) (3.676) ** 0.662^ 0.693* (-0.027) (8.786) (1.835) (2.478) ** 0.689* 0.735** (0.011) (8.447) (1.983) (2.586) Wnners Sort Perod Test Perod α βι σ SMB η HML Adj R ** ^ (-0.572) (10.729) (-0.630) (-1.760) ** ^ (-0.664) (12.746) (-0.922) (-1.743) ** ^ (-0.750) (14.32) (-0.958) (-1.879) ** * (-0.683) (14.307) (-1.062) (-2.047) ** ^ (-0.603) (14.198) (-1.337) (-1.934) 8 That s, an average of the 20, 35, and 50 stock portfolos. A full presentaton of these tests would be too volumnous for ths paper; however, the results presented n ths secton are representatve of those obtaned from the mplementaton of the ndvdual tests of robustness. 11

12 These results show that an nvestor employng a contraran nvestment strategy usng rollng wndows wll only earn returns to compensate for the market rsk of the portfolos, combned wth the rsks of small, value companes, as captured by the SMB and HML coeffcents. 9 CONCLUSION In revstng the overreacton anomaly we have shown that mplementng a contraran strategy for U.S. stocks does not produce alpha. The analyss suggests that the factors of sze and value play a central role n explanng the future returns generated by a strategy of formng portfolos based on past returns. Perhaps the most nterestng fndng s that past losers consstently load on the sze and value factor at statstcally sgnfcant levels, and at levels consstently hgher then ther wnner counterparts. Moreover, for past wnners, ths loadng s prmarly towards the value factor. The long-term past wnners ether load negatvely on the value factor at statstcally sgnfcant levels, or produce no loadngs other than on the market factor, confrmng prevous research that categorzes overreacton as a loser-effect rather than a loser-and-wnner-effect. These conclusons reman robust, even after adjustng for the January effect. Our study shows that portfolo managers could earn returns above the market by constructng portfolos based on the contraran nvestment strategy; however ths addtonal return would come smply at the expense of ncreased rsk a wn for proponents of standard fnance theory. 9 Such fndngs are corroborated by usng the FF model, ncorporatng the mpact of the January effect, whch, agan for reasons of space, are not shown here, but are avalable on request. 12

13 APPENDIX A. CONSTRUCTING CONTRARIAN PORTFOLIOS The frst stage of our study we follow an approach almost dentcal to that of DT, who demonstrate that most reversal evdence s contaned n portfolos constructed for a 3-year tme frame. We use data on stock returns from January 1927 through December 2003 for all stocks lsted on the CRSP tapes. We follow the steps: 1. At every month-end, we rank all stocks accordng to ther return above the market over the prevous m months (perod t-m + 1 to t) where t s on months. 2. Wnner and loser portfolos are formed condtonal upon past excess returns, wth the top 35 stocks (those wth the greatest cumulatve excess returns) formng the wnner portfolo, and the bottom 35 stocks (those wth the smallest cumulatve excess returns) formng the loser portfolo. 3. We then measure the return to each of these portfolos n every month for the next n months (perod t + 1 to n + 1). Over a n-year perod, the cumulatve abnormal (monthly) return (CAR) for each stock s calculated as: 36 CAR = (3) AR, n n= 1 where AR, t s measured by α 1,. 4. Ths step s repeated for all followng non-concdent n-month perods. Varatons to the DT strategy that we use nclude non-equal values for m and n. 5. The cumulatve average resdual returns of all securtes n the portfolos are calculated for the followng n months. Followng ths, the average cumulatve abnormal returns (ACAR) are calculated for months t-m + 1 to t. T-statstcs are then calculated to determne f these ACAR s are statstcally sgnfcant. In summary, Fgure 1 provdes a snapshot of the methodologcal approach central to the study See De Bondt and Thaler (1985) for a more detaled descrpton of the portfolo formaton technque. 13

14 Fgure 1. Tradng Method for Contraran Investment Strategy Calculate the return of each stock for each month of the sort perod Subtract market return from stock return to provde excess return Cumulate the abnormal returns over the sort perod Rank the stocks by cumulatve abnormal returns of the sort perod Long the stocks wth the lowest cumulatve abnormal returns Short the stocks wth the hghest cumulatve abnormal returns Contraran Sort Perod n Months Portfolo Test Perod n Months t-n (Month t-n to Month t-1) t (Month t to Month t+n-1) n = 36, 48, 60 tme In ths study, we also examned the contraran nvestment strategy wth the followng varatons to Step 2: 2a Portfolos were formed contanng 20 stocks and 50 stocks. Addtonally, our methodology acknowledges the numerous papers that have replcated, examned, extended and crtqued the orgnal overreacton study by also conductng tests on many alternate portfolo compostons. 11 These alternatves are not so much areas of crtcsm, rather a sensble procedure for provdng more robust results. 12 To overcome the perceved measurement shortcomngs n the earler work our methodology ncludes: portfolos that were examned on the bass of 20, 35 and 50 stocks; portfolos that were formed for both symmetrcal wndows (e.g. 3 year sort; 3 year test), and non-symmetrcal wndows, (e.g. 3 year sort; 4, and 5 year test); and, testng undertaken for the DT tme perod ( ), the recent perod ( ) and the full sample ( ). 11 For example, Pettngll and Jordan (1990) examne 90 stock portfolos, Chopra, Lakonshok and Rtter (1992) use 20 stocks, many studes use decle portfolos, Levs and Lodaks (2001) examne top and bottom one-thrd, De Bondt and Thaler (1987) use 50 stocks and Schereck, De Bondt and Weber (1999) use portfolos rangng from 10 to 40 stocks. For sort and test perods, Kryzanowsk and Zhang (1992) who use perods rangng from 1 to fve years, Campbell and Lmmack (1997) who mantan a three year sort perod but test over 1 to 5 years, and Schereck, De Bondt and Weber (1999) who use much smaller sort perods of 1, 3, 6 and 12 months. 12 In fact Schereck, De Bondt and Weber (1999) concede the ponts made n Ball, Kothar & Shaken (1995) and, when referrng to the orgnal DT study, state that profts may be llusory, a product of methodologcal and measurement problems [p104]. 14

15 REFERENCES Ball, R., and S. P. Kothar, 1989, Nonstatonary expected returns: Implcatons for tests of market effcency and seral correlaton n returns, Journal of Fnancal Economcs 25, Ball, R., and S. P. Kothar, and J. Shaken, 1995, Problems n measurng portfolo performance: An applcaton to contraran nvestment strateges, Journal of Fnancal Economcs 38, Banz, R., 1981, The relatonshp between return and market value of common stocks, Journal of Fnancal Economcs 9, Bauman; W. S., C. M. Conover, and R. E. Mller, 1999, Investor overreacton n nternatonal stock markets, Journal of Portfolo Management 25, Bowman, R., and D. Iverson, 1998, Short-run overreacton n the New Zealand stock market Pacfc-Basn Fnance Journal 6, Brown, K. C., W. V. Harlow and S. M. Tnnc, 1988, Rsk averson, uncertan nformaton and market effcency, Journal of Fnancal Economcs 22, Campbell, K., and R. Lmmack, 1997, Long-term over-reacton n the UK stock market and sze adjustments, Appled Fnancal Economcs 7, Centre for Research n Securty Prces, 2003, CRSP Data descrpton Gude, Unversty of Chcago, Chcago. Chan, K. C., 1988, On the contraran nvestment strategy, Journal of Busness 61, Chopra, N., J. Lakonshok, and J. R. Rtter, 1992, Measurng abnormal performance: Do stocks overreact?, Journal of Fnancal Economcs 31, Conrad, J., and G. Kaul, 1993, Long-term market overreacton or bases n computed returns?, Journal of Fnance 48, De Bondt, W. F. M., and R. Thaler, 1985, Does the stock market overreact?, Journal of Fnance 40, De Bondt, W. F. M. and R. Thaler, 1987, Further evdence on nvestor overreacton and stock market seasonalty, Journal of Fnance 42, Fama, E. F., and K. French, 1988, Permanent and temporary components of stock prces, Journal of Poltcal Economy 96, Fama, E. F., and K. French, 1993, Common rsk factors n the returns on stocks and bonds, Journal of Fnancal Economcs 33, Fama, E. F., and K. French, 1996, Multfactor explanatons of asset prcng anomales, Journal of Fnance 51, Fama, E. F., and K. French, 2006, Dssectng Anomales, SSRN Dscusson Paper d < Forner, C., and J. Marhuenda, 2003, Contraran and Momentum Strateges n the Spansh Stock Market, European Fnancal Management 9, Gaunt, C., 2000, Overreacton n the Australan equty market: , Pacfc-Basn Fnance Journal 8, Glbert, C. L., 1986, Testng the effcent market hypothess on averaged data, Appled Economcs 18, Haug, M., and M. Hrschey, 2006, The January Effect, Fnancal Analysts Journal 62, Hrschey, M., 2003, Extreme return reversal n the stock market, Journal of Portfolo Management 29,

16 Hong, H., and J. C. Sten, 1999, A unfed theory of underreacton, momentum tradng and overreacton n asset markets, Journal of Fnance 54, Howe, J. S., 1986, Evdence on stock market overreacton, Fnancal Analysts Journal 42, Jacobs, B. I., K. N. Levy, and D. Starer, 1998, On the optmalty of long-short strateges, Fnancal Analysts Journal 54, Kang, J., M. Lu, and S. X. N, 2002, Contraran and momentum strateges n the Chna stock market: , Pacfc-Basn Fnance Journal 10, Kem, D. B., 1983, Sze-related anomales and stock return seasonalty: further emprcal evdence, Journal of Fnancal Economcs 12, Kryzanowsk, L., and H. Zhang, 1992, The contraran nvestment strategy does not work n Canadan markets, Journal of Fnancal and Quanttatve Analyss 27, La, M., B. K. Guru, and F. M. Nor, 2003, Do Malaysan nvestors overreact?, Journal of Amercan Academy of Busness 2, Lakonshok, J., A. Shlefer and R. Vshny, 1994, Contraran nvestment, extrapolaton and rsk, Journal of Fnance 49, Levs, M., and M. Lodaks, 2001, Contraran strateges and nvestor expectaton: The UK experence, Fnancal Analysts Journal 57, Ma, Y., A. P. Tang, and T. Hasan, 2005, The Stock Prce Overreacton Effect: Evdence on Nasdaq Stocks, Quarterly Journal of Busness and Economcs 44, Merton, R. C., 1987, On the current state of the stock market ratonalty hypothess, n Macroeconomcs and Fnance: Essays n Honor of Franco Modglan, M.I.T. Press: Massachusetts. Newey, W. K. and K. D. West, 1987, A smple, postve sem-defnte, heteroskedastcty and autocorrelaton consstent covarance matrx, Econometrca 55, Pettengll, G.N., and B. D. Jordan, 1990, The overreacton hypothess, frm sze, and stock market seasonalty, Journal of Portfolo Management 16, Poterba, J. M., and L. H. Summers, 1988, Mean reverson n stock returns: Evdence and mplcatons, Journal of Fnancal Economcs 22, Renganum, M. R., 1981, Msspecfcaton of captal asset prcng: emprcal anomales based on earnngs yelds and market values, Journal of Fnancal Economcs 9, Renganum, M. R., 1983, The anomalous stock market behavour of small frms n January: emprcal tests for tax-loss sellng effects, Journal of Fnancal Economcs 12, Schereck, D., W. De Bondt, and M. Weber, 1999, Contraran and momentum strateges n Germany, Fnancal Analysts Journal 55, Schwert, G. W., 1983, Sze and stock returns, and other emprcal regulartes, Journal of Fnancal Economcs 12, Zarown, P., 1990, Sze, seasonalty, and stock market overreacton, Journal of Fnancal and Quantatve Analyss 25,