Sources of Industry Momentum Effect – Weekly Data Evidence
Hai-Ching Liu
Lecturer of Department of Finance, Southern Taiwan University of Technology,
17-1, Aly.41,Ln54, Nantai St, Yung-Kang City, Tainan710, Taiwan, R.O.C.
Tel: (06)2542062
E-Mail: hcl @mail.stut.edu.tw
Ying-Fen Fu[*]
Assistant Professor of Department of Finance,TainanUniversity of Technology,
529 Chung-Cheng Rd, , Yung-KangCity,Tainan 710 , Taiwan, R.O.C.
Tel: (06)2548400
E: Mail:
Sources of Industry Momentum Effect – Weekly Data Evidence
ABSTRACT
This study employing the industryweekly return data of Taiwan aims to investigate the industry momentum profit sources of an emerging country.This study, different from the past literature (Pan , Liano and Huang, 2004 and Du, 2008) which adopted weekly data to observe the industry momentum sources of short horizon in that it investigates the industry momentum sources of long horizon as well as short horizon. The result shows there exists significant industry reversal effect in Taiwan, which indicates that investors in Taiwan have the behavior of overreaction but not underreaction. After dividing the industry momentum sources into cross-autocovariances among industries, own-autocovariances of individual industry and cross-sectional variation in mean returns. We find that the sources of short-horizon industry momentum effect in Taiwan mainly come from the negative industry own-autocorrelation returns. The main source of the long-horizon industry momentum effect is the cross-autocovariance before 2002, while the own-autocovariance is the driving force after 2002.
Keywords:industry momentum, autocovariances, momentum profit sources
JEL classification: G10, G11
1 INTRODUCTION
The momentum effect implies that investors can get abnormal return from the stock market.Jegadeesh and Titman(1993)find that the momentum strategy in American stock market can make profit when the formation and holding periods are three to twelve months. Other studies (e.g. Conrad and Kaul,1998; Rouwenhorst,1998 and Schiereck, De Bondt and Weber,1999) also find the significant momentum effect in European and American individual stocks in the mid term. According to the momentum effect of Taiwan’s stock market, Rouwenhorst(1999) finds thatTaiwan exists insignificant mid-term price momentum effect. Chui, Titman and Wei(2000)find that only Hong Kong out of eight Asian stock markets shows the significant momentum effect both before and after the Southeast Asia financial crisis. The momentum effect of Taiwan stock market is however not significant after the Southeast Asia financial crisis. Besides the stock momentum effect on individual stocks, some literature investigates the industry momentum effect. Moskowitz and Grinblatt (1999)investigating the momentum effect of 20 industries in American stock market find that the profit from application of the individual momentum strategy decreases after the industry momentum effect is controlled. That is, the industry momentum can explain the abnormal return of the individual stock momentum. Besides Moskowitz and Grinblatt (1999), Grundy and Martin (2001) also find significant industry momentum effect in America. Swinkels (2002) demonstrates that the industry momentum effect not only exists in America but also in Europe. Moreover, G7 countries except Japan are found to have industrymomentum effect (Bacmann, Dubois, and Isakov, 2001)Taiwan, similar to Japan is also an Asian country and is also found little industry momentum effect (Fu and Kang, 2009). This study further investigates the sources of the industry momentum because as what Du (2008) indicates that industry itself is a kind of portfolio, which reduces the problem of market microstructure and the influence from the firm-specific factors.
Because the success of momentum strategy is based on the fact that when the winner and loser returns persist, investors will make profits by buying winners and selling losers if this momentum effect exists. Thus if the past industry return and future industry return are positively correlated, the significant industry momentum effect will appear.Moskowitz and Grinblatt(1999) decompose industry momentum profits into three sources:(1)positive industry return autocorrelation (2)negative industry return cross-autocorrelation (3) the cross-sectional variance of unconditional average return. They demonstrate that the industry return autocorrelation plays a more important role in momentum strategy profits than cross-autocorrelation. The result of Pan , Liano and Huang (2004) is consistent with what Moskowitz and Grinblatt(1999) propose. Their result shows that the significant industry momentum only exists in the case of significant positive return autocorrelation.
Several studies investigate the industry momentum effect by employing the monthly data of the developed country.Besides Moskowitz and Grinblatt (1999), Du and Watkins (2007) adopt the monthly data of American industries find that the sources of industry momentum vary with time. The serial correlation is the main source of the industry momentum before 1961, while the main source turns into cross-serial correlation after 1961. Lewellen (2002) also adopting monthly data finds that the long-term industry momentum profits mainly come from the cross-serial correlation. However, Pan , Liano and Huang (2004) demonstrate that the weekly data has more samples than the monthly data, which can increase the test power. Moreover, weekly data comes up withless bid-ask problems and less problems of spurious correlation which comes from trading at different periods. Thus, this study investigating the sources of the industry momentum will employs the weekly data.
Du (2008) indicates that the profit sources of the industry momentum strategy are different between the short-horizon momentum and long-horizon momentum effects[1]. Thus this study employing the industry return data of Taiwan aims to investigate the industry momentum profit sources of an emerging country. Moreover, we adopt the weekly data to analyze the industry momentum sources.This study, different from the past literature which adopted weekly data to observe the industry momentum sources of short horizon (Pan , Liano and Huang, 2004 and Du, 2008) in that it investigates the industry momentum sources of long horizon as well as short horizon[2]. Furthermore, both of the above studies employ the data of America. This study employs the industry index data of Taiwan which is a developing country not a developed county. A lot of past literature has shown that the stock and industry momentum effect in developed countries is different from that in developing countries. Therefore, whether the industry momentum sources of the developing countries are different from those of the developed countries is in need of further investigation. This study complements this gap which is not analyzed in the literature.
Consistent with Fu and Kang (2009)[3], the result of this study shows that there exists littleindustry momentum effect in Taiwan. We find the significant industryreversal effect in some cases. Some cases show the short-term significant industry reversal effect, while both of the long-term industry momentum and reversal effects are not remarkable. This result is not totallyconsistent with the studies of Barberis, Shleifer and Vishny(1998)(BSV thereafter),Daniel, Hirshleifer and Subrahmanyam (1998) (DHS thereafter), and Hong and Stein(1999)(HS thereafter). The three behavioral models of BSV, DHS and HS demonstrates that the underreaction of investors in the short run and the overreaction in the long run result in the momentum effect in the short run and the reversal effect in the long run. Although the lack of long-term reversal of Du (2008) is a challenge to these three behavioral models, the result of this study is not totallyconsistent with Du (2008). This study adopting the weekly data of Taiwanese industry index returnsfinds that there exists significant reversal effect when the formation and holding periods are under 52 weeks. The reversal effect is not significant when the formation and holding periods are longer than 52 weeks. This result indicates that investors in Taiwan have the behavior of overreaction in the short run butnot in the long run.
When we divide the sources of the momentum profits into three components, we find that the momentum sources mainly come from the own-autocovariance when the formation period is very short (1 and 4 weeks). However, the main source of momentumeffect changes with time when the formation periods are long (104 and 156 weeks). In the long-horizon cases (104 and 156 weeks), the main source of the momentum effect is the cross-autocovariance before 2002, while the own-autocovariance is the driving force after 2002. The cross-sectional variation in mean returns has little influence on the industry momentum or reversal profits.
The rest of this paper is organized as follows. Section 2 is the data description and analysis of the sources of the industry momentum profits. Section 3 presents empirical results. Finally, section 4 presents conclusions.
2 METHODOLOGY
2.1 The Data
Weekly data of nineteen industry indexes in Taiwan was obtained from Taiwan Economic Journal (TEJ).The sample period ran from January, 1995 through February, 2009. Table 1 is the summary statistics of the data. The top three industries with the highest average weekly returns are electronics, rubber and plastics industries and with the highest standard deviations are construction, cement and electrical & cable industries.Because the industry momentum effect sources are related to the return autocorrelation of the industries and the formation and holding periods are set to be 1, 4, 13, 26, 52, 104 and 156 weeks in the empirical analysis, Table 1 also lists the coefficients of autocorrelations for these lags (1, 4, 13, 26, 52, 104 and 156 weeks). Table 1 shows that much more negative return autocorrelations appear in the shorter-lag groups (1-,4- and 13- week lags), while much more positive return autocorrelations appear in the longer-lag groups (26-, 52-,104 and 156- week lags). If the sources of the industry momentum profits mainly come from the own-autocovariance, we can infer that the short-term momentum profit will be negative and the long-term momentum profit will be positive.
Table 1 Summary statistics
Notes: This table is the summary statistics of 19 industries in Taiwan. The study period is from January, 1995 to February 2009. Weekly data of nineteen industry indexes in Taiwan was obtained from the Taiwan Economic Journal (TEJ).
2.2 The Industry Momentum Effect
We first construct a momentum strategy to test the industry momentum effect in Taiwan. The momentum strategy is to construct a zero-cost portfolio by buying winners and selling losers of the formation period. The momentum effect is investigated by testing the holding period return of the zero-cost portfolio. The momentum (reversal) effect exists when the return of the zero-cost portfolio is significantly positive (negative). Most of the studies adopt the momentum method proposed by Lo and Mackinlay (1990) (thereafter LM) or Jegadeesh and Titman(1993) (thereafter JT). These two methods differ in the classification method and the weight of the investment portfolio. According to the LM method, all the industries are grouped into either winner or loser groups. As equation (2) demonstrates, the weight of every industry is different. However, Jegadeesh and Titman(1993) divide the stocks into deciles. According to the JT method, the momentum strategy buys the winner decile (the best performance decile) and sells the loser decile (the worst performance decile). Based on the JT method, the industry numbers in the winner and loser decile are the same. And the investment weight of every industry in the winner (loser) decile is the same. Lewellen (2002) indicates that the LM method is better than the JT method because the LM method invests in all the industries rather thanonly investing in the winner and loser deciles. Moreover, it is easy to investigate whether the test result is consistent with the three behavioral models (BSV, DHS and HS) because the LM method decomposes the momentum sources into serial correlation, cross-serial correlation and cross-sectional variation in mean. This study adopts the method of LM (1990) because the purpose of this study is to investigate the sources of the industry momentum effect in Taiwan. There are only 19 industries in Taiwan, making the industry numbers too small in every decile when JT (1993) method is adopted. Moreover, it is convenient to analyze the profit sources of the momentum effect by adopting the LM method.
We divide nineteen industries into winner and loser groups based on their past returns during the formation period. The winner (loser) group includes industries whose returns are greater (less) than the average during the formation period.The investment weight of the industry is greater when the return difference between the industry and the average of all the industriesisgreater.The zero-cost portfolio is constructed by selling loser and buying winner groups.We then duplicate the processin the next calendar week. The profit of the momentum strategy is the return of winners minus that of losers during the holding period for each calendar week.At last we average the momentum profit of each calendar week and test whether the average momentum profit is significantly different from zero. The t value is as follows:
(1)
where denotes the average momentum profit of each calendar week, is the variance of the momentum profit of each calendar week and N is the duplication number of the momentum strategy during the sample period.
2.3 Sources of the Industry Momentum Profit
This paper adopts the method presented by Lo and Mackinlay(1990) to decompose the profit of momentum strategy into (1) cross-autocorrelation among industry returns, (2) autocorrelation in industry returns, and (3)cross-sectional variation in mean returns of the individual industry. Following Lo and Mackinlay (1990), the weight of everyindustryin the zero-cost portfolio at time t is as follows:
where Ri,t-kdenotes the return of industry i at time (t-k) during the formation period, Rm,t-k is the return of equal-weighted market portfolio at time t-k, k is the lags, n is the number of industries and is the weight of industry i at time t. The momentum strategy is to construct a zero-cost portfolio by buying (selling) winner (loser) industries at time t.The industry which outperforms (underperforms) the market portfolio at time (t-k) is grouped in the winner (loser) group. The profit of the momentum strategy is as follows:
where denotes the momentum profits at time t during the holding period and Ri,t is the return of industry i at time t during the holding period. We multiply industry i’s return during of the holding period by the investment weight of industry i. The return of the momentum strategy is the sum of the above product of each industry. We can get the decompositions of the expected momentum strategy profit by taking expectations on both side of equation (3):
wheredenotes the expected momentum profit at time t. Equation (4) indicates that the expected momentum profits come from three sources: (1) the cross-autocovariances among industries (Ck) (2) the own-autocovariances of individual industry (Ok), and (3) the cross-sectional variation in mean returns ()[4].
3.EMPIRICAL RESULTS
3.1 The Industry Momentum Effect in Taiwan
Table 2 lists the industry momentum test result by employing weekly returns of industries’indexes. The holding periods and formation periodsare set to be 1, 4, 13, 26, 52, 104 and 156 weeks[5]. The statistics in Table 2are the average weekly returns of the zero-cost portfolio. The resultin panel A of Table 2 shows no significant industry momentum effect in Taiwan, while the significant reversal effect appears in eight cells. This result indicates that the industry momentum effect in Taiwan is not remarkable, but the industry reversal effect is significant[6].To observe the industry momentum effect during the different periods, we divide our sample period into two sub-periods. The first period (second) is from 1995/1/1-2001/12/31(2002/1/1-2009/2/28) and the industry momentum effect test result is presented in Panel B (C) of Table 2. The result in Panel B shows that 4 cells show the significant reversal effect and 4 cells show the significant momentum effect. In Panel C, 5 cells show the significant reversal effect and only 1 cell shows the significant momentum effect. In general, Table 2 shows that the industry momentum effect varies with time. Among the three panels, more significant negative momentum effect is shown when the holding period is 1 week. Thus, we further analyze the sources of the industry momentum profits when the holding period is 1 week.
Table 2 The industry momentum effect in Taiwan (weekly data)
Table 2 The industry momentum effect in Taiwan (weekly data) (continued)
Note: 1.This table employs the momentum strategy (Lo and Mackinlay, 1990) to investigate the existence of industry momentum effect in Taiwan with weekly returns for the time period from January 1995 to February 2009. 2. We divide 19 industries into winner and loser group based on their lagged returns during formation period. The winner (loser) group includes industries that outperform (underperform) the market portfolio. The zero-cost portfolios are constructed by buying winner and selling loser and held in holding period. 3.The average weekly returns of zero-cost portfolio are presented in this table. The numbers in parentheses are tstatistics. 4. **significantly at 0.05 level, *significantly at 0.1 level