The Profitability of Momentum Trading Strategies in the Irish Equity Market

Transcription

1 The Proftablty of Momentum Tradng Strateges n the Irsh Equty Market Fonnghuala O Sullvan* and Nall O Sullvan** Abstract: We examne the proftablty of momentum based tradng strateges n the Irsh equty market between 1988 and We nvestgate a range of tradng strateges over alternatve backward lookng rankng perods and forward lookng holdng horzons as well as for alternatve sze momentum portfolos. We fnd that returns to momentum based strateges are hghly non-normally dstrbuted gvng rse to concern about the valdty of nferences based on standard statstcal tests of ther abnormal performance. We therefore apply a bootstrap procedure to construct nonparametrc p-values for the portfolo performance measures. Overall, we fnd very lttle evdence that momentum based tradng strateges would have yelded an abnormal rsk adjusted return over the perod. The Irsh equty market appears to be qute effcent n ths respect. Keywords : momentum, tradng strateges, equty returns JEL Classfcaton: G11, G14. * Susquehanna Internatonal Group, Dubln, Ireland. ** Department of Economcs and Centre for Investment Research, Unversty College Cork, Ireland. Correspondng Author : Dr Nall O Sullvan Department of Economcs, Unversty College Cork, Ireland Tel. : +353-(0) E-mal : nall.osullvan@ucc.e We are grateful for fnancal support from the Irsh Research Councl for the Humantes and Socal Scences (IRCHSS).

2 1. Introducton Momentum based nvestment strateges nvolve holdng a portfolo of assets where each perod the portfolo holdngs are decded by a smple rule of buyng past wnner assets and sellng past loser assets (from among the unverse of assets avalable for selecton). The strategy attempts to capture a momentum effect n the prce movements of the underlyng assets over consecutve tme perods. Momentum based tradng strateges are of obvous nterest to nvestors as they may provde an abnormal return at relatvely low cost. Frst, the strategy can be constructed to be low (even zero) cost where short postons fund long postons. Second, t s smple to mplement as t does not requre extensve research nto asset selecton 1. The exstence of proftable momentum strateges among, for example, equty mutual funds s well documented for the US (Jegadeesh and Ttman 2001, 1993) whle Fletcher and Forbes (2002) report evdence that a substantal proporton of UK mutual funds also attempt to capture a momentum effect. In general, however, momentum effects are an under-explored phenomena outsde of the US equty market. In ths study we examne the proftablty of equty based momentum strateges n the Irsh market. The Irsh market s an nterestng case because of ts comparatvely low lqudty and hgh concentraton of stock ownershp whch may permt momentum effects to persst at least n the short term. Furthermore, smaller markets have n the past been found to be less effcent. Studes of momentum nvestment strateges are also of general nterest to researchers because fndngs of abnormal returns would typcally be n breach of the effcent market hypothess. However, t should be noted, that n the fund performance lterature momentum rsk factors are now wdely specfed n regresson models, where the ntercept s a measure of stock selecton skll, n order to control for performance attrbutable to momentum effects whch do not requre skll, per se, on the part of the fund manager to capture. We examne the proftablty of momentum tradng strateges n the Irsh equty market by smulatng and evaluatng several dfferent momentum portfolos based on alternatve sze portfolos, rankng perods (used to select equtes) and holdng perods. We examne the perod February 1988 to December A recent paper by O Donnell and Baur (2009) also examnes momentum tradng strateges n the Irsh case. Over a smlar sample perod the paper fals to fnd evdence of proftable strateges, although some abnormal returns are found durng certan sub-perods. We extend the 1 Of course, transactons costs ncurred wll be related to the degree of portfolo turnover but clearly a portfolo manager has some dscreton here. 1

3 O Donnell and Baur (2009) analyss by examnng alternatve sze momentum portfolos and also by specfcally nvestgatng the effects of non-normalty n the momentum portfolo returns. The study proceeds as follows: the next secton brefly outlnes some of the key fndngs from prevous studes of momentum strateges, secton 3 descrbes the data and methodology used n ths study, secton 4 descrbes our emprcal fndngs whle secton 5 concludes. 2. Revew of the Lterature Momentum tradng has been wdely examned n the lterature for a number of alternatve markets wth varatons across studes n factors such as, nter ala, the length of hstorcal horzons used to select stocks, holdng perods lengths, sample perods and momentum portfolo szes. For example, Rouwenhourst (1998) found that momentum effects exst n European markets, Moskowtz and Grnblatt (1999) found momentum effects across ndustry-sorted portfolos, and Grundy and Martn (2001) found that momentum strateges have been consstently proftable n the Unted States snce the 1920s. There was some focus on relatve strength strateges (that buy past wnners and sell past losers) n early lterature, most notably Levy (1967). However, as Levy arrved at ths tradng rule after nvestgatng 68 dfferent tradng rules, t was beleved that hs result could be attrbuted to selecton bas (Jegadeesh and Ttman 1993:66). Jegadeesh and Ttman (1993) s a semnal paper n the area of momentum strateges. Usng data from 1965 to 1989 the methodology nvolves selectng stocks based on ther returns over the past 1, 2, 3 or 4 quarters and holdng stocks for perods varyng from 1 to 4 quarters. Specfcally, securtes are ranked n ascendng order at the begnnng of each perod. Based on these rankngs, ten equally weghted decle portfolos are formed. In each perod the strategy buys the top wnner portfolo and sells the bottom loser portfolo, holdng ths poston for h perods. The authors show that stock returns exhbt momentum behavour at ntermedate horzons. They fnd that a strategy that uses a 6 month hstorcal rankng perod can earn profts of about 1% per month for the followng year, after whch the returns begn to dsspate. Ther results ndcate that these profts can be attrbuted to delayed stock prce reactons to frm-specfc nformaton, not common factors. That the strategy s proftable n the medum-term but unproftable n the longer term s seen as evdence that the theores 2

4 of nvestors ether overreactng (n the case of contraran strateges) or underreactng (n the case of relatve strength strateges) to nformaton are too smplstc. Instead, the authors deduce that nvestors who buy past wnners and sell past losers move market prces from ther long-term value temporarly, wth a reversal after about a year. An alternatve deducton s that the market underreacts to nformaton about the shortterm prospects of frms, whch tend to be more ambguous. The results of Jegadeesh and Ttman (1993) prompted a varety of nterpretatons rangng from explanatons of market neffcency, compensaton for rsk and data mnng. Conrad and Kaul (1998) argue that the apparent momentum arses because of cross-sectonal varaton n expected returns n adjacent tme perods and s smply a compensaton for rsk. In drect contrast, others such as Danel, Hrshlefer and Subrahmanyam (1998), Barbers, Shlefer and Vshny (1998) and Hong and Sten (1999) present behavoural models (see Barbers and Thaler 2003 for expanded survey) whch argue that the momentum effect arses because of a delayed overreacton to nformaton that pushes the prces of wnners (losers) above (below) ther long term values and n subsequent perods the returns of losers should exceed that of wnners as prces re-adjust to the over-reacton. Hence such models predct that n the postholdng perod returns to a momentum strategy should be negatve. Jegadeesh and Ttman (2001) evaluate the varous explanatons for the proftablty of momentum tradng strateges dentfed n the lterature followng ther 1993 study. The authors offer evdence to refute the crtcsm that the momentum anomaly s a product of data mnng by demonstratng that proftable momentum strateges perssted n the 1990s after ntally beng dentfed n ther earler study of the 1980s. Jegadeesh and Ttman (2001) do ndeed fnd evdence that the performance of a momentum portfolo n the postholdng perod (13 to 60 months) s negatve as predcted by Danel et al (1998) and others above. Whle the bulk of the extant momentum lterature relates to the US market, Rouwenhorst (1998) s a comprehensve study of the European market, usng data from twelve countres; Austra, Belgum, Denmark, France, Germany, Italy, The Netherlands, Norway, Span, Sweden, Swtzerland and the Unted Kngdom. Employng a smlar procedure to that of Jegadeesh and Ttman (1993), the paper fnds the strategy to be as proftable usng European stock prces as t s usng US stock prces, yeldng approxmately 1.16% per month for the followng year wth reversal thereafter. However, Rouwenhorst (1998) notes that an nternatonal momentum strategy may not 3

5 be well dversfed. A domnant performance by one country wll subsequently cause the wnner portfolo to be overweght that country. In examnng ths further the paper constructs momentum portfolos that weght by rankng stocks based on past performance relatve only to stocks lsted n the same country. However, momentum portfolo returns usng ths revsed strategy reman hghly proftable at 0.93% per month suggestng that ndvdual country momentum does not explan the success of the European wde strategy. However, the Rouwenhorst (1998) results do show a varaton n excess returns ( wnner portfolo mnus loser portfolo) across European countres. Although momentum effects are present n all countres, the strongest profts were experenced by Span, followed by The Netherlands, Belgum and Denmark. Sweden s the only country that doesn t experence sgnfcant profts n ths perod, wth portfolos earnng 0.16% excess returns per month. Moskowtz and Greenblatt (1999) queston whether the apparent proftablty of momentum strateges arses because of ndustry effects. They formulate a momentum strategy based on returns of dfferent ndustres and test t usng stock prces from 1963 to 1985 on the HYSE, AMEX and NASDAQ ndces. They also test the ndvdual stock prce momentum strategy used by Jegadeesh and Ttman (1993) n order to compare the two strateges. They report that momentum returns exst n ndustry-based portfolos whch are more proftable than ndvdual stock prce momentum strateges clamng that much of the proft derved from the latter s eroded after controllng for ndustry effects. Of course, the further mplcaton here s that momentum portfolos are not well dversfed, as wnners and losers are from the same ndustry, hence momentum returns may be a compensaton for rsk and not a market neffcency. Jegadeesh and Ttman (1993) also nvestgated the hypothess that stock prces overreact to nformaton n an extenson of the contraran strateges developed by De Bondt and Thaler (1985, 1987). Contraran strateges nvolve buyng (sellng) stocks that have been performng poorly (well) n recent months. DeBondt and Thaler (1985) explore the consequences of people s tendences to overreact to nformaton such as unexpected or dramatc news events. They fnd that people tend to emphasse recent nformaton too much and under-weght prevous nformaton. As a result of nvestor overreacton, they beleved t was possble that stock prces mght temporarly depart from ther fundamental values. If ths s the case, buyng past losers would be a more proftable strategy than buyng past wnners. Ther results showed that formng portfolos of past losers reaped exceptonally large January returns as far as fve years on. Ther concluson was that stocks that experenced extreme long term gans or 4

6 losses tended to undergo systematc prce reversals. DeBondt and Thaler (1987) form portfolos of the most extreme losers and wnners as measured by excess cumulatve returns over successve fve year formaton perods. Ther results showed that the loser portfolo outperforms the wnner portfolo by an average of 31.9% over the followng fve year test perod. In order to reconcle the fndngs that both contraran strateges and momentum strateges are proftable, even though they consst of takng opposte actons, Jegadeesh and Ttman (1993) consdered dfferent tme horzons. Contraran strateges are found to be proftable usng returns over the very long term (3 to 5 years) or over the very short term (1 week to 1 month). Relatve strength strateges base ther selecton on prce movements over the medum horzon (3 to 12 months). The Irsh case has been examned recently by O Donnell and Baur (2009). Over the perod 1984 to 2007 the authors examne momentum portfolos (wnner mnus loser) as well as wnner and loser portfolos separately. Over the full sample perod the authors fnd no evdence of proftable momentum strateges although some evdence s found examnng alternatve sub-perods of hgh, low and negatve market growth. Frst, however, O Donnell and Baur (2009) form momentum portfolos comprsng the top and bottom thrd of stocks. In our paper, we look further nto alternatve sze portfolos to dentfy possble proftable momentum strateges among the more extreme wnner and loser stocks. Second, O Donnell and Baur report standard statstcal tests of rsk adjusted return. However, we fnd that portfolos of wnners and loser stocks are both hghly non-normally dstrbuted and serally correlated so much so that questons arse as to the valdty of the standard statstcal tests such as t-tests. To examne the robustness of the O Donnell and Baur conclusons we apply a bootstrap procedure n our paper and derve non-parametrc p-values n our statstcal tests. As past fndngs have been found to be senstve to usng dfferent rankng and holdng perod lengths, the analyss n ths paper s conducted testng alternatve tme horzons n ths regard as well as alternatve sze momentum portfolos n order to capture these dynamcs and examne the robustness of results. In a related area of the lterature, the queston of momentum effects also arses n fund performance evaluaton. As ths s somewhat tangental to the focus of ths study, we do not propose to dscuss t n detal here. Instead, we very brefly refer the nterested reader to some mportant contrbutons to the area. Carhart (1997) demonstrates, nter ala, that momentum effects explan around half of the crosssectonal spread between the top and bottom decle portfolos of mutual funds ranked 5

7 by performance. Chen, Jegadeesh and Wermers (2000) examnes the past returns of the current consttuent stock holdngs of wnnng and losng funds and fnds that stocks currently held by wnnng funds have hgher past returns, or momentum, than stocks held by losng funds. The raw returns of the wnnng funds go on to outperform the returns of losng funds for the subsequent two quarters. The rsk adjusted returns of wnnng funds go on to outperform those of losng funds for the subsequent quarter. Grnblat and Ttman (1992) and Hendrcks, Patel and Zeckhauser (1993) report some evdence that the source of the fund performance persstence found n ther studes les n a momentum effect n the fund s stock holdngs rather than n persstent stock pckng ablty on the part of the fund manager. A summary of the man fndngs from the momentum strategy lterature s presented n Table 1. There s some varablty n these fndngs due to varatons n country/ndex, hstorcal horzons, holdng perods and the type of strategy examned as ndcated. All studes fnd momentum strateges to be proftable to some degree. The majorty of the nvestgatons have been carred out on US data, wth very lttle research on European ndces. [ Table 1 Here ] 3. Data and Methodology To construct momentum portfolos, at month t we rank all stocks n ascendng order of raw return based on a past perod of r months. Based on these rankngs two equally weghted portfolos, each of sze s, are formed. The wnner (loser) portfolo comprses the top (bottom) performng stocks. The strategy nvolves buyng (sellng) the wnner (loser) portfolo and holdng for h months. The momentum portfolo return between tme t and t+1 s the return on the wnnng portfolo mnus the return on the losng portfolo over ths holdng perod. Ths s then carred out recursvely monthly to generate a tme seres of returns. In ths study, we examne momentum portfolos for alternatve values of r = 3,6,11, h = 1, 3, 6 and s = top/bottom 30% of stocks, top/bottom 10% of stocks and top/bottom 5 stocks. We then test for abnormal performance n the momentum portfolo by estmatng the rsk adjusted return, α, n the least squares regresson of the CAPM as follows: 6

8 (1) Rt R ft =α +β(rmt R ft ) +ε t R t R mt where s the return on portfolo, s the return on a market factor mmckng portfolo, R ft s a rsk free rate. A statstcally sgnfcant postve value of alpha s taken to ndcate superor performance n the momentum tradng strategy. Here, R mt s the returns on the ISEQ ndex whle R ft s proxed by the one-month nterbank rate. Our entre analyss s conducted on monthly returns. Our data set whch covers the perod February 1988 to December 2006 ncludes all stocks lsted on the ISEQ ndex. Ths also ncludes all de-lsted and dead stocks over the perod. Therefore, portfolos of past wnners and losers are calculated at each tme t based on the full set of stocks that were avalable to fund managers at that tme hstorcally and not just based on the hstorcal tme seres of stocks that exst at the end of the sample perod. Ths avods the possble problem of survvorshp bas. If a stock drops out of the database durng a holdng perod the portfolo s rebalanced to be equally weghted across all the remanng stocks. Our nvestgaton of momentum tradng proftablty extends that of O Donnell and Baur (2009) n two key respects. Frst, we fnd that all the momentum portfolos demonstrate returns whch are hghly non-normally dstrbuted potentally nvaldatng the nferences from standard statstcal tests such as t-tests n (1). Therefore, we apply a bootstrap procedure to generate nonparametrc p-values for the performance estmates of each of the momentum portfolos. To do ths, the performance measurement model s frst estmated by OLS. The estmated coeffcents and OLS resduals, ˆα, ˆβ and ˆε are saved. In the next step a random sample of resduals of t sze T s drawn (wth replacement) from ˆε t. Usng the estmated factor loadngs from step one and the orgnal chronologcal orderng of Rmt and settng ˆ α = 0 under the null hypothess of no abnormal performance, bootstrap smulated returns, R t, are constructed. By constructon, ths bootstrapped or smulated portfolo return has true abnormal performance of zero. Usng these bootstrap ftted returns, the performance measurement model s re-estmated and a bootstrap estmate of abnormal performance under the mposed null hypothess s obtaned, denoted α. Ths represents random samplng varaton around a true value of zero. Ths smulaton α 7

9 process s repeated B = 1,000 tmes. The 1,000 values of α represent the nonparametrc dstrbuton of the OLS estmate of ˆα nonparametrc p-value for ˆα under the null hypothess. We can then examne where les relatve to the dstrbuton of α to determne a ˆα whch makes no pror assumptons regardng the normalty of returns. So, for example, a p-value = 0.10 ndcates that only 10% of the values of α are greater than observng the estmated value of ˆα suggestng that there s only a 10% chance of ˆα where the true value of α s zero. We can also use the t-statstc of alpha as a measure of abnormal performance. The t-statstc has the advantage that t controls for the standard error and may therefore gve a more relable estmate of abnormal performance relatve to ˆα. The same bootstrap procedure as above can be used to generate t α and hence the nonparametrc dstrbuton of the t-statstc of ˆα, denoted t α ˆ, under the null hypothess. In ths study we report the nonparametrc p-values of t α ˆ. Furthermore, n the calculaton of all t- statstcs n ths study we use New-West seral correlaton and heteroscedastcty adjusted standard errors. Second, the O Donnell and Baur (2009) study nvestgates momentum portfolos comprsng the top and bottom thrd of stocks. One concern s that such large portfolos may dsguse proftable momentum based strateges among more extreme wnner and loser stocks, e.g. top and bottom 10% of stocks or, say, top and bottom 5 stocks. One advantage of these latter cases s that the pursut of the momentum strategy may nvolve lower transactons costs on the part of the fund n rebalancng the fund holdngs each perod. In ths study we also report fndngs for momentum tradng strateges based on the top/bottom 10% and top/bottom 5 stocks. In the next secton we report our fndngs. 4. Emprcal Results Our man fndngs are presented n Table 2 whch shows results for the full sample perod 1988:2 2006:12. Performance estmates are reported for momentum based portfolos for alternatve rankng and holdng perods as ndcated n each column. E.g., the column headed 3-1 refers to a past rankng perod of 3 months and a holdng perod of 1 month etc. Results are also reported for alternatve sze momentum 8

10 portfolos ncludng top 30% mnus bottom 30%, top 10% mnus bottom 10% and top 5 mnus bottom 5 stocks as ndcated. Alpha s the rsk adjusted monthly percentage return from the OLS estmaton of Equaton (1) whle t-alpha s the correspondng t- statstc (Newey-West adjusted for seral correlaton and heteroscedastcty). [ Table 2 Here ] From the t-statstcs t s clear that none of the momentum portfolos yelded statstcally sgnfcant postve returns (at the 5% sgnfcance level) over the full perod and ndeed n several cases returns are negatve. Table 2 also shows results of tests of the normalty of the regresson resduals. Here, we report the Jarque-Bera test statstc, 2 2 JB χ. The χ crtcal value at 5% sgnfcance s It s mmedately evdent df = 2 that n the case of all portfolos the null hypothess of normally dstrbuted resduals s strongly rejected. In turn ths suggests that the alpha estmates are also non-normally dstrbuted thus potentally questonng the relablty of fndngs based on t-tests. Ths motvates our use of the bootstrap procedure to generate non-parametrc p- values n order to nvestgate the robustness of momentum fndngs. In Table 2, these p-values are denoted as Bootstrap p-value. The p-values, all greater than 0.05, ndcate that none of the momentum portfolos yeld postve and statstcally sgnfcant returns at 5% sgnfcance (or even at the 10% sgnfcance level). The full set of results n Table 2 lead us to conclude that, over the full sample perod, momentum tradng strateges dd not yeld a postve rsk adjusted return n the Irsh equty market. Ths fndng s remarkably robust to alternatve rankng wndows and holdng perods as well as to alternatve sze momentum portfolos. It also proves robust to alternatve statstcal testng methodologes whch account for the fndng of non-normally dstrbuted returns data. These overall fndngs are qualtatvely smlar to those of O Donnell and Baur (2009). However, O Donnell and Baur (2009) go on to explore the proftablty of momentum portfolos separately durng perods of low versus hgh growth n the stock market and report evdence of abnormal returns n the latter but not the former. It s n ths analyss that we fnd that results are somewhat senstve to the () non-normalty ssue, () sze of momentum portfolos and () rankng and holdng wndows. Table 3 presents fndngs for the later relatvely hgh stock market growth perod of 1995:9 2006:12 (dates chosen for consstency of comparson wth O Donnell and Baur 9

11 (2009)) 2. The upper panel of Table 3 shows results for the largest momentum portfolos of the top mnus bottom 30% of ranked stocks. Here, accordng to the standard t-statstcs, two of the momentum portfolos yeld a postve and sgnfcant abnormal return at the 10% sgnfcance level,.e. portfolos of 11-3 and 11-6 rankng and holdng perods. However, examnng the non-parametrc p-values of the t-statstcs from the bootstrap procedure we fnd that four of the portfolos are proftable at 10% sgnfcance. Ths s, the parametrc t-tests and the more robustly estmated nonparametrc p-values from the bootstrap procedure gve conflctng nferences regardng the proftablty of some the momentum tradng strateges - hghlghtng the potental for non-normalty n fnancal data to nvaldate the fndngs of many standard statstcal tests. [ Table 3 Here ] From Table 3, we fnd that one of the smaller momentum portfolos (top 10% mnus bottom 10%) wth 3-6 rankng/holdng perod yelds a postve and sgnfcant abnormal return at 10% sgnfcance. However, all other portfolos regardless of sze or holdng and rankng perods yeld nsgnfcant (and sometmes negatve) returns by both the standard t-tests and the non-parametrc p-values. There are some further surprses n the results. Frst, the fndng that proftable momentum strateges are more prevalent among larger rather than smaller portfolos suggests that the momentum effect s not drven by the extreme wnner and loser stocks but nstead s drven by those slghtly further nsde the tal of the cross-secton dstrbuton of stock returns. Alternatvely, there s more nose among the more extreme wnner and loser stock returns whch does not persst, even n the short term. Second, momentum s stronger among portfolos of longer rankng and holdng wndows. That a longer rankng perod provdes a more relable rankng of stocks n the momentum strategy s ntutve but one mght have expected that n an effcent market the momentum effect n stocks would dsspate quckly and hence would be better captured by shorter rather than longer holdng perods. Overall, the fndngs n Table 2 strongly suggest that momentum based tradng strateges n the Irsh equty market faled to yeld abnormal returns over the longer 2 We do not present results for the earler lower growth perod as, smlar to Table 2, they consstently show no sgnfcant return to momentum tradng across all portfolos. 10

12 sample perod under nvestgaton. Results presented n Table 3 do show some evdence of momentum tradng proftablty but hghlght the senstvty of results to nonnormalty, momentum portfolo szes and rankng and holdng perod lengths. In any case, these abnormal returns are found only n condtons of relatvely hgh market growth. As these condtons are comparatvely rare and ther persstence unrelable, our overall analyss fnds aganst the exstence of abnormal returns from momentum based equty tradng n the Irsh market. 5. Concluson Ths study examnes the proftablty of momentum based tradng strateges n the Irsh equty market between 1988 and Investgatng a range of tradng strateges over alternatve backward lookng rankng perods and forward lookng holdng horzons as well as for alternatve sze momentum portfolos, we fnd that returns to momentum based strateges are hghly non-normally dstrbuted gvng rse to concern about the valdty of nferences based on standard statstcal tests. We apply a bootstrap procedure to construct nonparametrc p-values for the momentum portfolo performance measures. Overall, we fnd very lttle evdence that momentum based tradng strateges would have yelded an abnormal rsk adjusted return over the perod. Our overall results are qualtatvely smlar to those of O Donnell and Baur (2009) but contrbute to ths lterature by hghlghtng that () the non-normalty of stock returns, partcularly n the tals of the cross-sectonal dstrbuton, must be consdered f robust nferences are to be drawn from ths type of study and () the most extreme wnner and loser stock returns appear to be nosy and detract from rather than drve proftable momentum portfolos where these exst at all. 11

13 Table 1: Summary of Internatonal Fndngs from Momentum Studes Author and Year Jegadeesh and Ttman (1993) Momentum Strategy Indvdual Stock Prce Country/Index USA: NYSE and AMEX Sample Perod Hstorcal Horzon (Months) Holdng Perod (Months) Profts , 9, 12 3,6,9,12 Around 1% per month for followng year. Unproftable under 1 month and over 1 year. Rouwenhorst (1998) Indvdual Stock Prce 12 European countres , 6, 9, 12 Around 1% per month for followng year wth reversal thereafter. Jegadeesh and Ttman (2001) Indvdual Stock Prce USA: NYSE, AMEX, NASDAQ , 6, 9, 12 Around 1% per month for followng year. Unproftable under 1 month and over 1 year. Grundy and Martn (2001) Indvdual Stock Prce USA: NYSE and AMEX % per month George and Hwang (2004) Indvdual Stock Prce USA: CRSP % per month Marshall and Cahane (2005) Indvdual Stock Prce Australan Stock Exchange (ASX) % per month 12

14 Moskowtz and Grnblatt (1999) Industry USA: NYSE, AMEX and NASDAQ % per month for followng year. Strongest at 1 month horzon. Unproftable after a year. George and Hwang (2004) Industry USA: CRSP % per month Marshall and Cahane (2005) Industry Australan Stock Exchange (ASX) % per month George and Hwang (2004) 52-week hgh USA: CRSP % per month wthout a reversal after a year. Marshall and Cahane (2005) 52-week hgh Australan Stock Exchange (ASX) % per month 13

15 Table 2: The Proftablty of Momentum Tradng Strateges Full Sample Perod Table 2 presents performance estmates for momentum based portfolos for alternatve rankng and holdng perods as ndcated n each column over the full sample perod 1988:2 2006:12. E.g., 3-1 refers to a past rankng perod of 3 months and a holdng perod of 1 month. Alpha s the portfolo performance 2 measure from the OLS estmaton of Equaton (1). In addton, we report the Jarque-Bera normalty test statstc, JB χ. The table also reports the bootstrapped p-value of the t-statstc of alpha as descrbed n Secton 3. All t-statstcs are based on New-West seral correlaton and heteroscedastcty adjusted standard errors. Performance estmates and assocated test statstcs are shown for momentum portfolos comprsng the top/bottom 30% of stocks, top/bottom 10% of stocks and top/bottom 5 stocks as ndcated Top 30% Bottom 30% Alpha t-alpha JB Normalty Statstc Bootstrap p-value df = Top 10% Bottom 10% Alpha t-alpha JB Normalty Statstc Bootstrap p-value Top 5 Bottom 5 Alpha t-alpha JB Normalty Statstc Bootstrap p-value

16 Table 3: The Proftablty of Momentum Tradng Strateges Hgh Stock Market Growth Table 3 presents performance estmates for momentum based portfolos for alternatve rankng and holdng perods as ndcated n each column. Results are reported for the later sample perod of relatvely hgh stock growth from 1995:2. E.g., 3-1 refers to a past rankng perod of 3 months and a holdng perod of 1 month. Alpha s the portfolo performance measure from the OLS estmaton of Equaton (1). In addton, we report the Jarque-Bera normalty test statstc, JB 2 χ df = 2. The table also reports the bootstrapped p-value of the t-statstc of alpha as descrbed n Secton 3. All t-statstcs are based on New-West seral correlaton and heteroscedastcty adjusted standard errors. Performance estmates and assocated test statstcs are shown for momentum portfolos comprsng the top/bottom 30% of stocks, top/bottom 10% of stocks and top/bottom 5 stocks as ndcated. Top 30% Bottom 30% Alpha t-alpha JB Normalty Statstc Bootstrap p-value Top 10% Bottom 10% Alpha t-alpha JB Normalty Statstc Bootstrap p-value Top 5 Bottom 5 Alpha t-alpha JB Normalty Statstc Bootstrap p-value

17 References Barbers, N. and Thaler, R. (2003). A survey of behavoural fnance, n Handbooks of the economcs and fnance, edted by G.M. Constantnds, M. Harrs and R. Stulz, Elsever Scence B.V. Barbers, N., Shlefer, A. and Vshny, R. (1998). A model of nvestor sentment, Journal of Fnancal Economcs, 49, pp Carhart, M. (1997). On persstence n mutual fund performance, Journal of Fnance, 52, pp Chen, L., Jegadeesh, N. and Wermers, R. (2000). The value of actve mutual fund management: an examnaton of the stock holdngs and trades of fund managers, Journal of Fnancal and Quanttatve Analyss, 35, pp Conrad, J. and Kaul, G. (1998). An anatomy of tradng strateges, Revew of Fnancal Studes, 11, pp Danel, K., Hrshlefer, D. and Subrahmanyam, A. (1998). Investor psychology and securty market under- and over-reactons, Journal of Fnance, 53, pp De Bondt, W.F.M. and Thaler, R. (1985). Does the stock market overreact? Journal of Fnace, 40, pp De Bondt, W.F.M. and Thaler, R. (1987). Further evdence of nvestor overreacton and stock market seasonalty, Journal of Fnance, 42, pp Fletcher, J. and Forbes, D. (2002). An exploraton of the persstence of UK unt trust performance, Journal of Emprcal Fnance, 9, pp George, T.J. and Hwang, C.Y. (2004). The 52-week hgh and momentum nvestng, Journal of Fnance, 59, pp Grnblatt, M. and Ttman, S. (1992). The persstence of mutual fund performance, Journal of Fnance, 42, pp Grundy, B.D. and Martn, J.S. (2001). Understandng the nature of the rsks and the source of the rewards to momentum nvestng, The Revew of Fnancal Studes, 14, pp Hendrcks, D., Patel, J. and Zeckhauser, R. (1993). Hot hands n mutual funds: short run persstence of performance, , Journal of Fnance, 48, pp Hong, H. and Sten, J. (1999). A unfed theory of under-reacton, momentum tradng and over-reacton n asset markets, Journal of Fnance, 54, pp Jegadeesh, N. and Ttman, S. (1993). Returns to Buyng Wnners and Sellng Losers: Implcatons for Stock Market Effcency, Journal of Fnance, 48, pp Jegadeesh, N. and Ttman, S. (2001). Proftablty of Momentum Strateges: An Evaluaton of Alternatve Explanatons, Journal of Fnance, 56, pp Marshall, B.R, and Cahane, R.M. (2005). Is the 52-week hgh momentum strategy proftable outsde the US? Appled Fnancal Economcs, 15, pp

18 Moskowtz, T.J. and Grnblatt, M. (1999). Do Industres Explan Momentum? Journal of Fnance, 54, pp O Donnell, D. and Baur, D. (2009). Momentum n the Irsh Stock Market, Appled Fnancal Economcs Letters, Vol. 16, Issue 11, pp Rouwenhorst, K.G. (1998). Internatonal Momentum Strateges, Journal of Fnance, 53, pp