An Empirical Comparison of Asse Pricing Models for he Tokyo Sock Exchange Absrac In his sudy we compare he performance of he hree kinds of asse pricing models proposed by Fama and French (1993), Carhar (1997), and Pasor and Sambaugh (2003), which are ofen used in an empirical research. The primary purpose of his sudy is o deermine he bes asse pricing model for he Tokyo Sock Exchange sample. However, here is no guaranee ha he bes model in a paricular sub-period always ranks he highes in oher sub-periods. In recen years boh legal changes in Japan and a complee sysem replacemen a he Tokyo Sock Exchange migh have made a difference o he asse pricing srucure of Japanese socks. To invesigae hese changes in he long-run, we divide he enire observaion period ino several sub-periods. Then we examine how he saisical significance of hese risk facors and he explanaory power of each model have changed over ime. We find ha all four risk facors, which are relaed o size, book-o-marke, momenum, and liquidiy are associaed wih long-run sock reurns. We also find ha he book-o-marke relaed facor (HML) is no longer significan afer he launch of he arrowhead, while Pasor and Sambaugh s liquidiy innovaion facor remains significan. This laer finding suggess he possibiliy ha he launch of he arrowhead rading sysem drasically changed he asse pricing srucure as well as he price discovery process of Tokyo Sock Exchange socks. Key words: Fama-French hree facor, momenum anomaly, liquidiy innovaion JEL Codes: G11, G12, G14
1. Empirical Evaluaion of Asse Pricing Models The idenificaion of a correc asse pricing model has long been an imporan heme in financial economics lieraure. Such a model no only explains he behavior of sock reurns bu also srenghens he abiliy o forecas abnormal reurns of individual securiies. Previous sudies, such as hose by Fama and French (1993 and 1997), have shown ha hree-facor models based on marke reurns, size, and value can accuraely explain sock reurns. Carhar (1997) has shown ha he fourh facor, oneyear momenum, is significan, while Pasor and Sambaugh (2003) have demonsraed he significance of he liquidiy innovaion facor. Less is known, however, abou Japanese daa samples. Kuboa and Takehara (2003) have shown ha he Fama and French hree-facor model is suiable for Japanese daa, while Chou e al. (2007) have demonsraed he exisence of mid-erm conrarian effecs for sock reurns as well as for abnormal reurns exraced from Fama and French s hree facor model. Kuboa and Takehara (2010) fully invesigaed he condiional five-facor model wih liquidiy and he conrarian facor and found his model performs beer han he Fama-French hree facor model. They confirmed his asserion by he Gibbons, Ross, and Shanken es (1989) and Hansen and Jagannahan s (1997) disance measure es. This sudy invesigaes he exisence of sysemaic risk facor(s) in addiion o he hree idenified by Fama and French (1993), which can beer explain sock reurns of firms lised on he Tokyo Sock Exchange wih more recen daa. We compare hree kinds of asse pricing models proposed by Fama and French (1993), Carhar (1997), and Pasor and Sambaugh (2003). The primary purpose of he research is o deermine he bes asse pricing model for he Tokyo Sock Exchange sample. However, here is no guaranee ha he bes model in a paricular sub-period ranks he highes in oher subperiods. Recenly, muliple legal and insiuional changes affecing rading on he Tokyo Sock Exchange (TSE) as well as a complee sysem replacemen migh have had an impac on he price formaion process of Japanese socks. To carefully invesigae he impac of hese changes over a longer horizon, we divide our observaion period ino several characerisic sub-periods. Then we demonsrae how he significance of risk facors and he explanaory power of each model has changed over ime. The res of he paper proceeds as follows. Secion 2 explains he consrucion mehod of he Fama- French hree facor, conrarian facor, and liquidiy innovaion facors. In Secion 3 we show he descripive saisics of he five facors, and re-examine he adequacy of each as candidaes for priced risk facors. In Secion 4 we compare he performance of hese pricing models based on he resuls of cross-secional regressions and sandard GMM ess. Secion 5 shows he sub-period resuls from asse pricing ess, and we pay paricular aenion o he possible srucural change caused by he launch of he arrowhead. We conclude in Secion 6.
2. Consrucion of Risk Facors 2.1. Candidae Asse Pricing Models We compare he hree kinds of asse pricing models proposed by Fama and French (1993), Carhar (1997), and Pasor and Sambaugh (2003). Firs, we briefly review hese candidae models and he risk facors wihin hem. I is commonly undersood ha he Fama-French hree-facor model is suied o empirical daa, boh for he U.S. (Fama and French, 1993) and for Japan (Jagannahan e al., 1998, and Kuboa and Takehara, 2003, 2010). Fama and French s hree-facor model is composed of: value-weigh excess marke reurns (abbreviaed as EVW), size relaed porfolio reurn spreads (referred o as he SMB facor), and book-o-marke raio-relaed porfolio reurn spreads (also known as he HML facor). The basic Fama and French hree-facor model can be wrien as follows: M SMB HML r rf, i ( rm, rf, ) i SMB i HML (1) In (1) r j, is he reurn of securiy j in monh, r M, is he reurn of he marke porfolio, r f, is he risk-free rae, and SMB and HML are he Fama-French Small-Minus-Big and High-Minus-Low facors, and is error erm, respecively. The second model we examine was developed by Carhar (1997). Iniially, Jagadeesh and Timan (1993) documened an individual sock momenum anomaly in which par of he sock reurns of individual securiies could be prediced by pas one-year realized reurns. To increase he explanaory power of he asse pricing model wih he momenum facor, Carhar added an addiional facor o he Fama-French model and proposed a four facor model, (referred o as he Carhar four facor model). Since a momenum facor named UMD (Upward-Minus-Downward) consruced by Kenneh French was laer used for his purpose, he four facor model in equaion (2) is: r M rf, i rm, rf, ) ( SMB HML UMD (2) SMB i HML i UMD i The final model we es is a pricing model wih a liquidiy innovaion facor proposed by Pasor and Sambaugh (2003), which emphasizes he imporance of marke-wide liquidiy for asse pricing in he U.S. sock marke. We consruc a liquidiy innovaion facor L in he same way as Pasor- Sambaugh and add i o he Carhar four facor model of (2). 1 As a resul, he final model we examine in his sudy is: 1 The consrucion mehod of Pasor and Sambaugh s liquidiy facor for TSE socks is explained in Secion 2.2.
r M i, rf, i ( rm, rf, ) UMD i UMD PSL i SMB i PSL SMB i, HML i HML (3) In equaion (3), PSL denoes Pasor and Sambaugh s liquidiy innovaion facor in monh. 2.2. Consrucion of Risk Facors The following crieria were used o selec he daa. Firs, he sample for sock reurn daa is composed of all socks lised in he firs and second secions of he Tokyo Sock Exchange wih a daily observaion frequency. The primary daa source of daily reurns and rading volume of individual securiies is he NPM Daabase provided by Financial Daa Soluions, Inc. which covers January 1977 o December 2012. Financial saemen daa for consrucing five facors was obained from he Nikkei NEEDS daabase. In order o go back o 1977, daa from each firm s unconsolidaed financial saemens were used for fiscal years 1977-1999 since consolidaed saemens only became a major form of financial saemens in Japan in 1999. For fiscal years afer 2000, we exclusively used consolidaed financial saemen daa. [Fama and French Three Facors, SMB and HML] Fama and French s hree-facor model fis well wih Japanese daa (Kuboa and Takehara, 2003, 2010). As we noed earlier, we used his model as a benchmark and expanded i ino a five-facor model. In order o consruc Fama and French s six benchmark porfolios and wo oher facors for he sample period, we used all firms lised in he firs and second secions of he Tokyo Sock Exchange. The fiscal year-end of more han 90 per cen of he firms lised on he TSE is he end of March. Accordingly, he sample firms lised in he TSE were sored a he end of Augus each year, which is five monhs afer he fiscal year end. This was done o ensure public availabiliy of boh he numbers of shares issued and he book value of equiy daa for invesors. For hose firms ha did no have a March 31 fiscal year-end, earlier daa from heir financial saemens was used. In Augus of each year from 1977 hrough 2012, all firms lised on he TSE were ranked by heir marke value of equiy (MV). Firms were also ranked by heir book-o-marke raios (BM) and he 30h and 70h perceniles of TSE firs secion firms were compued as daa breakpoins. Using he median MV and he 30h and 70h perceniles of BM, he firms were divided ino six MV and BM ranked groups, hus allowing he formaion of six value-weighed porfolios. The Fama- French facors EVW (excess marke reurns), SMB (Small-Minus-Big) and HML (High-Minus-Low) were hen compued by applying a mehod similar o ha of Fama and French (1993). For he risk-free ineres rae, he monhly average of overnigh call-money rae wihou collaeral was used.
[Momenum Facor, UMD] An addiional facor in (2) which is inended o be added o Fama and French s hree facors is he Upward-Minus-Downward (UMD) facor. Jagadeesh and Timan (1993) found he predicabiliy of momenum relaed variables for U.S. daa. For Japanese daa, however, a conrarian sraegy, raher han a momenum sraegy, generaes abnormal reurns. Ilhara e al. (2004) found a one-monh shorerm reversal for raw reurns. Chou e al. (2007) found a reversal of abnormal reurns from one o hree monhs and for longer horizons of more han wo years using he Fama-French hree-facor model. In view of hese findings, we consruced a momenum facor, referred o here as he UMD (Upward- Minus-Downward), by using he realized reurns in he pas one o five years. The consrucion of he UMD facor closely follows ha of he HML facor in Fama and French, excep ha he porfolio formaion frequencies are he same as French s UMD facor. The difference in he mehod is ha i is consruced once a year in he HML facor, as is he case wih he original Fama and French HML facor, while he UMD facor is re-consruced a he beginning of every monh. The oal sample is divided ino hree sub-groups by he pas (-2, -k) cumulaive reurns for k = 12, 24, 36, 48, and 60. A he same ime, i is divided ino wo by size, and he upper 30 per cen of he sample is classified as he winner porfolio and he lower 30 per cen as he loser porfolio. The firs horizon is 11 monhs (and he las one 59 monhs) because daa from he previous monh is deleed, as is sandard procedure. Because porfolios are simulaneously divided by size and pas reurns, he UMD facor porfolio reurns are consruced from he differences beween equal weighed reurns of wo value-weighed winner porfolios of differen sizes and he equal weighed reurns of wo valueweighed loser porfolios of differen sizes. [Liquidiy Innovaion Facor, PSL] Nex, he liquidiy facor was based on Pasor-Sambaugh, among many differen definiions of liquidiy measures (Consaninides, 1986). Le r e d, be he excess reurn of securiy j over he reurns of value-weighed marke reurns consruced from our sample universe for day d during monh of securiy j; le v d, be he rading volume of his sock for day d measured in million yen; and le sign ( ) be he signum funcion, in his case, he sign of he realized reurn from he previous day. The following regression equaion is hen wrien for monh, where D days of ransacion are observed during monh, and he final denoes he residuals of he fied values. The esimaed slope j, d 1, coefficiens for he sign variable γ hen becomes he liquidiy measure for sock j as in Pasor- Sambaugh. r e r d, sign ( r d, ) v d, d 1,, d 1,,. (4) e d 1, D The sign of his slope coefficien γ above is expeced o be minus wih over-reacion over he day
as a resul of illiquidiy reasons or bid-ask rading coss facors. I shows degrees of he daily reversion of a paricular sock. I is expeced ha he higher he absolue value of γ, he lower he liquidiy of he sock. Furhermore, Pasor and Sambaugh consruced he following aggregae marke liquidiy measures based on he equal weighed average of an individual sock s liquidiy measure. In equaion (5), N is he number of socks during monh. N 1. (5) N j 1 Using his aggregae measure as a saring poin, Pasor and Sambaugh consruced he innovaion of his liquidiy measure as follows. Le mv 1 be he oal marke value of firms a he beginning monh of he daa saring poin, January 1977, and le mv be he oal marke value of he monh. The monhly changes in he consruced marke liquidiy measure, afer being adjused for he overall growh of he sock marke size, is N mv ( 1 ). (6) mv 1 j 1 Given he changes in his measure, he following regression equaions are compued where u represens he residuals. mv 1 a b 1 c 1 u. (7) mv Afer exracing he ime-series of he residual erm u in (7), we define he Pasor and Sambaugh liquidiy innovaion facor PSL as: 1 PSL u. (8) 100 1 3. Behavior of he Five Facors Before conducing he asse pricing es, we check he behavior of five risk facors: EVW, SMB, HML, UMD and PSL. We also divide he enire daa period from Sepember 1977 hrough December 2012 ino hree characerisic sub-periods; Sepember. 1977 o December 1989; January 1990 o December 2009; and January 2010 o December 2012. The firs sub-period includes he Japanese sock-price bubble, and he second sub-period is referred o as he Los Two Decades in Japan. In he firs monh of he hird sub-period, he nex generaion rading sysem, he arrowhead, was launched and high frequency rading (HFT) became mainsream a he TSE.
3.1. Fama-French Three Facors Table 1 shows he descripive saisics of Fama-French hree facors. Firs of all, he mos remarkable feaure for Japanese daa is ha he average of he HML facor is much larger han ha in oher counries. In he enire observaion period, he arihmeic average of HML is 0.624 percen per monh (which is equivalen o 7.488 percen per annum.) [TABLE 1 ABOUT HERE] On he oher hand, he average of SMB is very low a 0.071 percen per monh and is no saisically significan (p-value=0.668). Alhough he small-cap effec is one of he well known anomalies explained in finance exbooks (Berk and DeMarzo, 2012, for example), i is diminished in he Japanese sock marke. The average of he excess value-weighed marke reurn (EVW) is 2.396 percen per annum (0.200 percen per monh). The low equiy risk premium a his level seems o be quie reasonable because our sample period includes a long economic recession from he 1990s. In he firs sub-period (Sepember 1977 o December 1989), he average of he HML facor is 0.685 percen per monh which is significan a a 1% level. EVW is very high a 1.036 % per monh (12.432% per annum). The average of he SMB facor in he firs sub-period is high a 0.217% per annum alhough i is no significan. In he second sub-period, he HML facor sill remains srongly significan. However, signs of EVW and SMB facors become negaive. I raises he possibiliy ha EVW and SMB were no priced facors even hough hey are associaed wih sock reurns. In he las sub-period, afer he launch of he arrowhead, he average of he HML facor grows smaller o 0.529% per monh and i is no longer significan. Many porfolio managers of Japanese equiy funds argue ha value sraegies did no work well afer he launch of he arrowhead, and our finding in he hird subperiod seems o confirm his. 3.2. Momenum Facors Table 2 shows he descripive saisics of UMD facors wih 5 differen daa horizons. The basic saisics of spread reurns are repored for monhs of differen lenghs. The case for 60-monh reurns is demonsraed, so he resuls sar from January 1982. The able repors he mean, p-values for he null hypohesis of zero means, and sandard deviaion, as well as percenile numbers. Similar o Table 1, we divide he sample period ino hree characerisic sub-periods. [TABLE 2 ABOUT HERE] The null of he zero mean is rejeced for 24 monhs and for longer horizons of 36, 48 and 60 monhs a he 5% significance level. However, i should be noed ha he sign of he average of he UMD
facor is negaive over any horizons wihou excepion. This implies ha he conrarian sraegy from one o five years works well in Japan, bu he momenum sraegy does no, despie he abundan evidence of one year momenum profis around he world. So, i seems ha he Tokyo Sock Exchange is an excepion among developed counries sock markes. In he firs sub-period (Sepember 1977 o December 1989), he average of he UMD facor wih he las 12 monhs of daa (UMD12) becomes posiive a 0.269% per annum and he average of he UMD facor wih more han 24 monhs of daa remains negaive. Alhough he average of UMD12 was no significanly differen from zero, i implies ha fund managers were able o exploi abnormal profis from a one year momenum sraegy, a leas in he 1980s. In he second and hird sub-periods, he average reurns from five UMD facors wih a differen daa horizon are all negaive. Furhermore, average values of UMD24 hrough UMD60 are all significanly negaive in he second sub-period. In he hird sub-period, which is afer he launch of he arrowhead, all he UMDs remained negaive bu were no saisically significan. Figure 1 shows cumulaive reurns from UMDs over he enire observaion period. As is shown here, all five UMDs behave quie similarly over ime. [FIGURE 1 ABOUT HERE] Panel B of Table 2 shows he correlaion marix among he UMDs, and we found ha he correlaions are very high. Based on hese findings, we decided o use he one year momenum facor (UMD12) in he following asse pricing es. The negaive and significan average reurn from he UMD12 implies ha he one year momenum facor canno be a priced facor for TSE socks. As we explained, however, he primary purpose of his sudy is o invesigae he effeciveness of asse pricing models. By aaching significance o he comparabiliy of asse pricing es resuls beween he U.S. and Japan, we employ he one year momenum facor. 3.3. Liquidiy Innovaion Facor Our porfolio universe is all socks lised on he TSE firs and second secions wih a sampling period of January 1978 hrough December 2012. Using daily volume and reurn daa for his period in he same way as Pasor and Sambaugh (2003) we compued he liquidiy innovaion facor PSL. Recall ha Pasor and Sambaugh s liquidiy innovaion facor is an unanicipaed change in and is defined as a residual erm u in he regression equaion (7). As is obvious from he definiions of marke-wide illiquidiy measure γ in (5) and PSL in (7), resuls from he esimaion heavily depend on he universe and he sampling period. 2 We infer ha he small caps lised on he TSE second secion 2 On he conrary, in he case of compuaion of he Fama-French hree facors and he UMD facor, hey are less subjec o he difference of he universe because he inermediae six size and book-o-
and he sample during he world financial crisis show differen behavior wih he esimaed liquidiy innovaion facor. The descripive saisics of he esimaed PSL over he enire period and for he hree sub-periods are shown in Table 3. [TABLE 3 ABOUT HERE] The mean of PSL for he enire observaion period becomes zero by is definiion. The average PSL facor is posiive and significan a he 1% level and he sandard deviaion is small a 2.09 in he firs period. However, he average of he PSL facor urns negaive in he second and hird sub-periods. Also in he hird sub-period, he average of he PSL is very low a -1.046 alhough i is no significan due o a high sandard deviaion a 5.627. 3.4. Correlaion among Five Facors Table 4 shows a correlaion marix among five facors for he enire observaion period and he hird sub-period afer he launch of he arrowhead. For he enire period, correlaions among hese facors are very low. [TABLE 4 ABOUT HERE] This implies ha he five disinc risk facors behave differenly. Afer he launch of he arrowhead, however, we find he correlaions among he facors drasically change. The Pearson correlaion beween EVW and HML became higher a 0.581 and he one beween HML and UMD became -0.518. Noe we have confirmed in Table 1 ha he mean of HML was no saisically differen from zero only in he hird sub-period, and, in paricular, we find he correlaions among five facors have changed significanly around and afer he launch of he arrowhead. 4. Resuls of Asse Pricing Tess In his secion, we make comparisons among asse pricing models described in equaions (1), (2) and (3) by using wo represenaive esing mehodologies; Fama and MacBeh s (1973) wo-sage regression and he GMM es using he Hansen and Jagannahan (1997) disance measure. 4.1. Resuls of Fama-MacBeh Regressions Since Fama and MacBeh s wo-sage regression analysis is a well-known mehod in empirical marke benchmark porfolios are value weighed. Also, hey are no subjec o he difference in he sampling period as benchmark porfolios are consruced in each year or each monh.
finance used o esimae parameers for asse pricing models, we skip he explanaion. As in Fama and French (1993), we analyze reurns of he size and book-o-marke ranked 25(=5 5) porfolios. The sampling period is again January 1978 o December 2012 (420 monhs) which is an inersecion of he daa consrucion period of Fama-French hree facors, he UMD facor, and he PSL facor. In he firs sage we compue he beas of size and he book-o-marke ranked 25 porfolios by running univariae regressions. In his case, reurns from each size and book-o-marke ranked porfolio are regressed ono one of he five risk facors. In he second sage, reurns from he porfolios are regressed ono beas esimaed in sage 1 o measure he risk premium for each risk facor. Table 5 presens resuls for candidae facor models: ha is, 1, 3, 4 and 5 facor models, respecively. I is worh menioning ha esimaed premiums for he HML are all posiive and srongly significan in he 3, 4 and 5 facor models. [TABLE 5 ABOUT HERE] In he cases of he four facor model by Carhar (1997) and he five facor model, he slope coefficiens for he UMD bea are posiive and significan a he 1% level. In hese four and five facor models he slope coefficiens for he SMB bea are posiive and significan a he 5% level alhough i is negaive in he case of he hree facor model. PSL is also significan in he five facor model. Based on hese findings, we infer ha all five risk facors are meaningfully associaed wih reurns of he Japanese sock marke. The adjused R-squared figure for he Fama and French (1993) hree facor model is sufficienly high a 0.724, while ha of he sandard CAPM is low a 0.222. When we add he UMD and PSL o he hree facor model, he adjused R-squared slighly increases o 0.744 and 0.760. Alhough he slopes of he UMD and PSL beas are significan, we find he addiional explanaory power of hese wo facors is marginal. A large fracion of he explanaory power of he five facor model is due o he conribuion of HML and SMB facors. 4.2. Resuls of he GMM es When Euler condiions are esed and he bes model is o be chosen, he use of an opimal weighing marix is no appropriae because he sochasic discoun facor wih he larges errors may show he smalles value in he quadraic form of he pricing error. Accordingly, Hansen and Jagannahan (1997) sugges using he alernaive covariance marix, and when doing so, he es saisics become he sum of N-K χ 2 (1) variables under suiable regulariy condiions. The esing form of he Euler condiions for he five-facor case is: E r r ) 1 ( r r ) SMB HML UMD PSL 0 (9) ( p, f, 1 M, f, 2 3 4 5
The null hypohesis is ha he mispricing of he given asse pricing model is zero, and he rejecion indicaes a significan mispricing error in he model. Hansen and Jagannahan s disance measure is more appropriae for comparing model performance among nesed models, and he comparison of disance measures can provide a unique and unambiguous mehod for choosing he bes asse pricing model. By using reurns from he size and book-o-marke ranked 25 porfolios, we compare he hree pricing models. Table 6 shows he resuls of he GMM es using he Hansen and Jagannahan (1997) disance measures. [TABLE 6 ABOUT HERE] For he Fama and French hree facor model, he disance measure is found o be 0.343 and for he four and five-facor models hey are 0.316 and 313, respecively. The five-facor model achieves he minimum disance measure of all. However, based on he p-values (as shown in he second row in each cell of he righ-mos column), all models are rejeced wih a 5% significan level. Wih regard o he significance of each facor variable, he p-values are shown in he second row of each cell. In he case of he five facor model, HML and UMD are significan a he 1% level while oher facors, EVW, SMB, and PSL, are no significan. In he GMM es, he signs of delas in (9) should be negaive if each facor in he pricing model has a posiive risk premium. The resuls in Table 6 show ha his is indeed he case wih all facor variables excep he PSL and i suggess ha he overall resuls are quie inuiive. 5. Sub-Period Resuls of Fama-MacBeh Regressions In he previous secion we horoughly invesigaed he behavior of he five risk facors as well as he explanaory power of hree kinds of pricing models. We will now invesigae how averages and sandard deviaions of risk facors are differen beween he sub-periods. If i is he case, i suggess he possibiliy of a srucural change in asse pricing over ime for Japanese socks. We show he subperiod resuls of he Fama-MacBeh ess a he end of his sudy. 3 In Table 7 we divide he observaion period ino hree sub-periods which are he same as in previous secions. [TABLE 7 ABOUT HERE] 3 Since he number of monhly observaions in he hird sub-period is 36 which is no sufficienly large compare o he number of asses (25), i is hard o conduc he GMM es o check srucural change in he hird period.
The firs sub-period included he so-called Japanese sock bubble and he second sub-period is labeled Los Two Decades in Japan. The hird sub-period is afer he launch of he arrowhead, in which high frequency rading emerged on he cener sage of sock rading. In he firs sub-period from January 1978 hrough December 1989, he Fama and French hree facor model works very well. The esimaed risk premiums for EVW, SMB and HML beas are all posiive and significan. The famous Jensen s alpha defined as an inercep erm in regression is no saisically significan. The adjused R-squared figure is sufficienly high a 0.654. Carhar s four facor model also works well in he firs sub-period. The risk premium for he UMD bea is posiive and significan a he 1% level, alhough he marke facor EVW becomes insignifican. In he case of he five facor model, he risk premium for he PSL bea is posiive bu no saisically significan. The explanaory power of pricing models measured by he adjused R-squared figure says sufficienly high, bu i is hard o explain he behavior of risk facors in he second sub-period. The premium for he HML bea remains posiive and significan in he Fama-French model, however, i becomes insignifican when he addiional wo facors are added. The premium for he SMB bea is no significan in he second sub-period. The PSL bea is now disincively negaively associaed wih sock reurns and is significan a he 1% level. Finally, he significance of Jensen s alpha in he es of all he pricing models suggess ha he cross-secional variaion of sock reurns in he second subperiod canno be explained very well by hese pricing models. The observaions change remarkably in he hird sub-period and he HML facor is no longer significan. We confirmed in Table 1 ha he average of he HML facor was no saisically differen from zero only in he hird sub-period. As a resul, here is a possibiliy ha he HML facor does no work as i has been used in he Japanese marke. In conras, he risk premium for he SMB bea becomes significanly posiive in he hird sub-period. As for he five facor model, he risk premium for he PSL bea is also significanly posiive bu Jensen s alpha is no significan (p-value=0.846). I suggess ha he five facor model for his sub-period can well explain he cross-secional variaion of sock reurns excep for he HML and UMD facors. 6. Conclusion In his sudy, we compared hree kinds of asse pricing models proposed by Fama and French (1993), Carhar (1997), and Pasor and Sambaugh (2003). The resuls of Fama-MacBeh regressions and he GMM es suggess ha all candidae risk facors: SMB, HML, UMD and PSL are associaed wih Tokyo Sock Exchange long-run reurns. We furher spli he observaion period ino hree characerisic sub-periods and invesigaed he srucural change in asse pricing over ime. The Fama-French model can well explain he crosssecional variaions of Japanese socks especially in he 1980s. In conras, he HML facor was no longer significan afer he launch of he arrowhead, while Pasor and Sambaugh s liquidiy innovaion
facor became significan. This finding suggess he possibiliy ha he launch of he arrowhead rading sysem drasically changed he asse pricing srucure, liquidiy, and informaion asymmery of socks lised on he Tokyo Sock Exchange. References: Amihud, Y. (2002), Illiquidiy and sock reurns: Cross-secion and ime-series effecs, Journal of Financial Markes, Vol. 5 No. 1, pp.31 56. Berk, J and DeMarzo, P. (2011) Corporae Finance. (Second ediion) New York: Pearson Educaion, Inc. Carhar, M. (1997), On persisence in muual fund performance, Journal of Finance, Vol. 52 No. 1, pp. 57 82. Chou, P-U., Wei, K. C. J and Chung, H. (2007), Sources of conrarian profis in he Japanese sock marke, Journal of Empirical Finance, Vol. 14 No. 3, pp. 261 286. Cochrane, J. (2001), Asse Pricing, Princeon Universiy Press, N.J. Consaninides, G. (1986), Capial marke equilibrium wih ransacion coss, Journal of Poliical Economy, Vol. 94 No. 4, pp. 842 862. Fama, E. F. and French, K. R. (1993), Common risk facors in he reurns on sock and bonds, Journal of Financial Economics, Vol. 33 No. 1, pp. 3 56. Hansen, L. P. and Jagannahan, R. (1997), Assessing specificaion errors in sochasic discoun facor models, Journal of Finance, Vol. 52 No. 1, pp. 557 590. IIhara, Y., Kao H. and Tokunaga, T. (2004), The winner loser effec in Japanese sock reurns, Japan and he World Economy, Vol.16 No. 4, pp. 471 485. Jagannahan, R., Kuboa K. and Takehara, H. (1998), Relaionship beween labor-income risk and average reurn: Empirical evidence from he Japanese sock marke, Journal of Business, Vol.71 No. 1, pp. 319 347. Jagadeesh, N., and Timan, S. (1993), Reurns o buying winners and selling losers: implicaions for sock marke efficiency, Journal of Finance, Vol. 48 No.1, pp. 65 91. Kuboa, K. and Takehara, H. (2003), Financial secor risk and he sock reurns: Evidence from Tokyo Sock Exchange Firms, Asia-Pacific Financial Markes, Vol. 10 No.1, pp. 1 28. Kuboa, K. and Takehara, H. (2010), Expeced reurn, liquidiy risk, and conrarian sraegy: Evidence from he Tokyo Sock Exchange, Managerial Finance, Vol. 36. No.8, pp. 655-679. Pasor L. and Sambaugh, R. (2003), Liquidiy risk and expeced sock reurns, Journal of Poliical Economy, Vol. 111 No. 3, pp. 642 685.
Table 1. Descripive Saisics of Fama-French Three Facors EVW: Excess reurns from he value-weighed marke index, SMB: Small-Minus-Big facor, HML: High-Minus-Low facor. Mean is an arihmeic average of monhly reurn from facors (in %), and p-value is a probabiliy value from Suden s -es in which he null hypohesis is he arihmeic average of risk facor equal o zero. S.D. is a sandard deviaion of risk facors. 25%ile, Median and 75%ile denoe 25 percenile, median and 75 percenile of risk facors, respecively. Enire Period: 09/1977-12/2012 Mean p -value S.D. 25%ile Median 75%ile EVW 0.200 0.427 5.173-2.634 0.374 3.443 SMB 0.071 0.668 3.427-1.805 0.150 2.216 HML 0.624 0.000 3.051-0.953 0.513 2.195 Sub-period A. 09/1977-12/1989 EVW 1.036 0.001 3.861-1.040 0.766 2.846 SMB 0.217 0.499 3.893-1.502 0.674 2.732 HML 0.685 0.009 3.147-1.294 0.370 2.513 Sub-period B. 01/1990-12/2009 EVW -0.320 0.393 5.793-4.007-0.464 3.625 SMB -0.076 0.722 3.280-2.144-0.096 2.040 HML 0.600 0.003 3.066-0.835 0.599 2.063 Sub-period C. 01/2010-12/2012 EVW 0.225 0.796 5.181-2.780 0.324 3.596 SMB 0.452 0.206 2.104-0.861 0.454 1.773 HML 0.529 0.229 2.591-1.757 0.459 2.017
Table 2. Summary of UMD Facors UMD12, UMD24, UMD36, UMD48, and UMD60 denoe Upward-Minus-Downward facors consruced by using he pas 12, 24, 36, 48 and 60 monhs cumulaive reurns. (Uni: in %.) In Panel B, Pearson correlaions among UMDs are shown in he lower-lef riangular par of he marix and Spearman rank correlaions are shown in he upper-righ riangular par. Panel A. Descripive Saisics Mean (p -value) S.D. 25%ile Median 75%ile Enire Period: 01/1982-12/2012 UMD12-0.103 0.674 4.712-2.157 0.352 2.591 UMD24-0.528 0.031 4.694-2.552-0.205 2.065 UMD36-0.761 0.001 4.529-2.748-0.379 1.773 UMD48-0.842 0.000 4.315-2.939-0.409 1.651 UMD60-0.997 0.000 4.402-3.240-0.483 1.486 Sub-period A. 01/1982-12/1989 UMD12 0.269 0.547 4.359-2.121 0.677 3.107 UMD24-0.182 0.699 4.597-2.567-0.042 2.411 UMD36-0.530 0.252 4.499-2.795-0.560 2.016 UMD48-0.510 0.218 4.032-2.944-0.620 1.953 UMD60-0.676 0.105 4.040-3.667-0.670 1.979 Sub-period B. 01/1990-12/2009 UMD12-0.261 0.426 5.072-2.291 0.174 2.484 UMD24-0.702 0.028 4.924-2.626-0.276 2.054 UMD36-0.853 0.006 4.740-2.716-0.309 1.752 UMD48-1.029 0.001 4.534-2.939-0.344 1.360 UMD60-1.120 0.000 4.656-2.949-0.476 1.203 Sub-period C. 01/2010-12/2012 UMD12-0.038 0.935 2.760-1.152 0.550 1.441 UMD24-0.289 0.588 3.172-1.513-0.146 1.896 UMD36-0.763 0.134 2.982-2.501-0.477 1.533 UMD48-0.484 0.411 3.491-2.481 0.038 2.050 UMD60-1.030 0.091 3.562-3.861 0.033 1.024 Panel B. Correlaion Marix UMD12 UMD24 UMD36 UMD48 UMD60 UMD12 1.000 0.824 0.712 0.664 0.617 UMD24 0.723 1.000 0.906 0.862 0.807 UMD36 0.593 0.863 1.000 0.940 0.894 UMD48 0.531 0.785 0.895 1.000 0.942 UMD60 0.490 0.725 0.841 0.922 1.000
Table 3. Descripive Saisics of Pasor-Sambaugh Liquidiy Innovaions Enire Period 01/1978-12/2012 Sub-period A 01/1979-12/1989 Sub-Period B 01/1990-12/2009 Sub-period C 01/2010-12/2012 Mean (p -value) S.D. 25%ile Median 0.000 1.000 5.539-2.305 0.310 0.578 0.001 2.091-0.181 0.608-0.136 0.757 6.783-3.335-0.682-1.406 0.143 5.627-4.547-1.016 Table 4. Correlaions among Five Facors Pearson correlaions among five facors are shown in he lower-lef riangular par of he marix and Spearman rank correlaions are shown in he upper-righ riangular par. Enire Period: 01/1978-12/2012 EVW SMB HML UMD PSL EVW 1.000-0.098-0.170-0.146-0.082 SMB -0.052 1.000 0.144-0.205 0.115 HML -0.177 0.108 1.000-0.324 0.012 UMD -0.085-0.141-0.235 1.000-0.030 PSL -0.086 0.084-0.007-0.039 1.000 Afer he Launch of arrowhead: 01/2010-12/2012 EVW SMB HML UMD PSL EVW 1.000-0.424 0.589-0.336-0.158 SMB -0.326 1.000-0.111 0.246-0.017 HML 0.581-0.029 1.000-0.543-0.128 UMD -0.178 0.116-0.518 1.000-0.012 PSL -0.181-0.030-0.161 0.022 1.000
Table 5. Resuls of Fama-MacBeh Regressions EVW, SMB, HML, UMD and PSL denoe esimaed risk premium for beas. Jensen s alpha is defined as an inercep erm. Inercep EVW SMB HML UMD PSL Adjused R 2 Coef. 3.824-3.466 0.222 (p -value) 0.000 0.000 Coef. 0.449-0.136 0.077 0.916 0.724 (p -value) 0.480 0.846 0.688 0.000 Coef. -0.374 1.694 0.473 1.707 2.991 0.744 (p -value) 0.559 0.039 0.030 0.000 0.002 Coef. -0.269 1.462 0.867 1.716 3.534-5.032 0.760 (p -value) 0.668 0.070 0.001 0.000 0.000 0.007 Table 6. GMM Tes wih he Hansen and Jagannahan Disance Measure In he able, δ 1, δ 2, δ 3, δ 4, δ 5 are he parameer values in equaion (20) of he main ex. HJ-Dis denoes he Hansen-Jagannahan disance measure. The rows denoed as p-value are he significance of he coefficiens, excep in he column HJ-Dis, in which he significance of he Hansen and Jagannahan disance measure for he GMM es is shown. In he es T = 333, N = 25, and K = 1, 2,.., 5 and he p-values are compued by numerically generaing χ 2 (1) values for 10,000 imes. δ 1 (EVW) δ 2 (SMB) δ 3 (HML) δ 4 (UMD) δ 5 (PSL) HJ-Dis Coef. -1.155 0.409 (p -value) 0.232 0.000 Coef. -1.778 0.571-7.951 0.343 (p -value) 0.060 0.693 0.000 0.000 Coef. -3.929-1.247-13.135-8.785 0.316 (p -value) 0.003 0.452 0.000 0.009 0.017 Coef. -3.730-2.050-13.035-9.120 4.169 0.313 (p -value) 0.007 0.277 0.000 0.009 0.397 0.030
Table 7. Sub-Period Resuls of Fama-MacBeh Regressions Inercep EVW SMB HML UMD PSL Adjused R 2 Sub-period A. 01/1979-12/1989 Coef. -0.209 2.450 1.709 1.205 0.654 (p -value) 0.727 0.012 0.004 0.001 Coef. 0.427 1.244 1.924 1.417 3.151 0.733 (p -value) 0.463 0.203 0.002 0.000 0.004 Coef. 0.649 1.436 1.827 1.221 2.322 1.004 0.746 (p -value) 0.296 0.134 0.003 0.003 0.024 0.107 Sub-period B. 01/1990-12/2009 Coef. 1.806-1.993-0.061 0.793 0.760 (p -value) 0.010 0.014 0.804 0.001 Coef. 2.201-2.998-0.059 0.398-1.190 0.766 (p -value) 0.002 0.029 0.810 0.464 0.444 Coef. 1.879-1.975 0.443 0.822 0.888-6.393 0.820 (p -value) 0.007 0.089 0.192 0.072 0.418 0.007 Sub-period C. 01/2010-12/2012 Coef. -0.822 1.648 1.130 0.389 0.706 (p -value) 0.271 0.447 0.040 0.634 Coef. -0.823 1.709 1.180 0.244-0.257 0.698 (p -value) 0.270 0.407 0.022 0.767 0.825 Coef. -0.129 1.352 0.942 0.314-0.435 3.788 0.872 (p -value) 0.846 0.502 0.049 0.703 0.705 0.042
Figure 1. Cumulaive Reurns of UMD Facors