Kaustav Sen. Lubin School of Business Pace University, One Pace Plaza New York, NY 10038, USA

Transcription

1 Momentum Strategies and Sophisticated Investor Preferences in India Kaustav Sen Lubin School of Business Pace University, One Pace Plaza New York, NY 10038, USA August 2007 Preliminary: Please do not quote

2 Momentum Strategies and Sophisticated Investor Preferences in India Abstract In this paper, we examine the existence of earnings and price momentum anomalies for a sample of actively traded stocks in the Bombay Stock Exchange during We also examine the relationship between these anomalies and the level of ownership by two distinct categories of sophisticated investors, domestic Mutual Funds (MF) and Foreign Institutional Investors (FII). Our findings indicate that using standard time series models, there is clear evidence of a post earnings announcement drift in the Indian market. In addition, after controlling for value and size, both earnings momentum and a three month price momentum exist. The level of ownership by FIIs mitigates the effect of earnings momentum and accentuates the effect of price momentum on stock returns. In contrast, there is no impact of the level of ownership by MFs. We also find that the changes in holdings by FIIs are driven by the price momentum whereas the changes in holdings by MFs are driven by the earnings momentum. We finally compare the performance of a MF ownership weighted portfolio with a FII ownership weighted portfolio. We find that the FII weighted portfolio has a higher beta and generates a lower beta adjusted return in comparison to the MF weighted portfolio. However, the turnover and holding patterns of the two portfolios are similar.

3 1. Introduction Emerging markets have become increasingly important for the international investor during the last decade, with several countries opening their economies and attempting to integrate their capital markets into the global system. Since 1991, India has started a process of economic liberalization which has put it in the forefront of emerging markets today and has become a favored destination for foreign capital. 1 At the same time, domestic investors in India have increasingly looked towards the equity markets as an attractive investment option. In this paper, we first examine the existence of earnings and price momentum anomalies for a sample actively traded stocks in the Bombay Stock Exchange. We then examine if the level of ownership by two distinct categories of sophisticated investors, domestic Mutual Funds (MF) and Foreign Institutional Investors (FII), mitigate these anomalies. Finally, we examine if there are any differences between these two types of investors in exploiting the alternative momentum based anomalies. We think that the results of our analysis will be useful to investors and regulators alike. It will increase the understanding of stock return determinants as well as help regulators understand the significance of alternative momentum measures in influencing investor behavior. Economic reforms in emerging markets are expected to result in economic growth, better corporate performance and higher stock returns. However, international investors weigh in such expectations against risks at the country level, which can relate to the political environment, economic and currency stability, transparency and accounting standards, legal framework among many others 2. While the considerations in the minds of domestic investors may have common elements with those of the international investors, they may find a different set of indicators to be useful given that they cannot choose the level of country risk. In other words, international 1 Goldman Sachs came up with an influential report that identified four emerging economies that are expected to grow significantly in the future (see Wilson and Purushothaman, 2003). These are Brazil, Russia, India and China (BRIC). Within these four countries, India offers a unique set of characteristics that is attractive to the international investor a democratic political system with a well established legal, banking and business framework inherited from its British legacy; long history of an active stock market; a vibrant domestic economy that has made significant progress into the service sector. Although it lags China in terms of total foreign investments, it complements China in the drivers behind its growth and thus makes an important country to examine. 2 Karolyi and Stilz (2003) survey the literature on international asset pricing. Harvey (1991) and others have concluded the usefulness of extra-market factors. Several papers have examined the role of global macroeconomic factor risks e.g. Chen et al. (1986), Harvey (1995). In addition, currency risk has been examined in various papers e.g. Solnik (1974), Griffin and Stulz (2001); liquidity risk by Bekaret et al (2005); legal protection by LaPorta et al. (1998); transparency by Bhattacharyya et al. (2003). 1

4 investors have a wider menu of investment choices than domestic investors, which may lead to different indicators becoming useful. Many emerging market regulators react to foreign investors by imposing capital controls and worry about hot money impacting the behavior of local markets. While not directly answering such concerns, this paper sheds some light on investor preferences that can appear to be a hot money effect. 3 Emerging economies attract international capital through different routes. Investors can set up businesses, either as wholly owned subsidiaries of a multinational parent company or as a joint venture with a local business partner. This is commonly referred to as a Foreign Direct Investment (FDI). 4 Alternatively, an indirect way of foreign investing includes buying securities of companies listed in the local stock exchanges, either in the primary or secondary markets. In the Indian context, FIIs are allowed to go through this route. The non-institutional foreign investors can invest when locally listed firms access global capital markets by issuing depository receipts overseas. As is the case in several other emerging markets, there is considerable interest in understanding the role of FIIs in affecting return volatility in the Indian market. 5 Condoo and Mukherjee (2004) conclude that the strength and duration of the volatility of day-to-day movements of FII investment and stock returns in India are interrelated; further, the strength of FII volatility is also correlated with the call money rate. However, Bose and Condoo (2004) conclude that economic liberalization that began in 1993 to allow FIIs enter the Indian market has had the desired effect. Starting with the landmark FII legislation in 1995, all subsequent policy revisions, either to promote further liberalization or impose controls to ensure stability of FII 3 Hot money generally refers to rapid movement of capital across international borders to take advantage of attractive short term yields. It is often driven by currency speculation and the fund flows are highly volatile. Chari and Kehoe (2003) find that information frictions together with standard debt default problems cause volatile capital flows, making it appear like a hot money phenomena. There have been several papers that have examined the impact of market liberalization on market returns, return volatility as well as fundamental and market risk factors e.g. Bekaert et al. 2002, Chari and Henry, Choe et al (1999) find that there was herding and positive feedback trading by foreign investors prior to the 1997 economic crisis in Korea, but not during the crisis. They conclude that such behavior did not have a destabilizing effect on the Korean stock market during the crisis. 4 Wikipedia ( has the following insert: Foreign direct investment (FDI) is defined as "investment made to acquire lasting interest in enterprises operating outside of the economy of the investor." [1] The FDI relationship consists of a parent enterprise and a foreign affiliate which together form a transnational corporation (TNC). In order to qualify as FDI the investment must afford the parent enterprise control over its foreign affiliate. The UN defines control in this case as owning 10% or more of the ordinary shares or voting power of an incorporated firm or its equivalent for an unincorporated firm. 5 A speech by the Reserve Bank of India (RBI) governor in 2005 was misinterpreted to imply that FII flows cause stock market volatility. However, volatility and quality of FII flows were the concerns raised in the speech, not its relation to stock return volatility (see 2

5 flows has had positive results or no negative results. Liberalization has increased the level of FII flows and its sensitivity to the market return and FII flow momentum. Market efficiency is one of the major concerns in the mind of any investor. Market regulators in developed and developing countries alike strive to improve it in order to attract more capital. Yet at the same time, sophisticated investors constantly look to exploit any inefficiency in the market, hopefully without taking on any additional risk. This has led to the search and exploitation of price anomalies by such investors. Behavioral finance has explained such anomalies in the context of investor psychology, as either over or under reaction by investors to valuable information. Schwert (2003) reexamined some of the early anomalies discovered in the literature and concluded that most of them had disappeared since the academic research was published. The only one he found to persist was due to market momentum a la Jegadeesh and Titman (1993). Several papers have looked into the role of momentum in the context of international and more specifically, emerging markets. The literature on determinants of stock returns in emerging markets is quite extensive. Rouwenhorst (1999) concludes that the factors that drive cross sectional returns in emerging markets are similar to those in developed markets. Small cap stocks outperform large cap stocks, value outperforms growth and momentum drives returns as well. However, beta does not seem to impact returns. Hou et al. (2006) uses data from 49 countries over to examine the determinants of global stock returns. They find that momentum, cash flow to price and a global market factor explain most of the variation in stock returns at the country, industry and global level. Focusing specifically on emerging economies and Latin American markets in particular, Muga and Santamaria (2007a) find that momentum strategies yield profits. They find that stock type is a more important factor than country in explaining the momentum effect. 6 Ornelas and Fernandes (2005) reexamine the momentum effect during using data from 15 emerging markets. They find that it is reversal, rather than momentum that seems to exist in most of these markets. They argue that this evidence can be explained using either one of the two reasons: (i) the speed of information diffusion has increased over time, thus removing a delayed reaction (ii) increased retail investor participation because of online trading has increased the disposition and contrarian effect on the market. 6 In a separate analysis using Spanish market data, Muga and Santamaria (2007b) find that neither size nor turnover determines momentum effect, although these are associated with momentum based profits. 3

6 Our findings indicate that using standard time series models, there is evidence of a post earnings announcement drift (also referred to as earnings momentum or earnings surprise) in the Indian market. In addition, after controlling for value and size, both price momentum and earnings momentum exists. The level of ownership by FIIs mitigates the earnings momentum effect and accentuates the price momentum effect. In contrast, there is no impact of the level of ownership by MFs on either measure of momentum. We next find that the changes in holdings by FIIs are affected by the price momentum measure whereas the changes in holdings by MFs are affected by the earnings momentum measure. We finally compare the performance of portfolio weighted by the level of ownership by MFs with another portfolio weighted by level of ownership by FIIs. We find that the FII weighted portfolio has a higher beta and generates a lower beta adjusted return. However, the turnover and holding patterns of the two portfolios are similar. The rest of the paper is organized as follows. Section 2 describes the two types of institutional investors. Section 3 outlines the methodology used to measure earnings surprise, which is the basis for understanding the earnings momentum anomaly. It also reviews the literature on the two momentum strategies. Section 4 describes the models we use to test for the existence of momentum and the institutional investor choices. Section 5 describes the sample. Section 6 presents the results. Section 7 concludes the paper. 2. Institutional Investors in India We consider two different categories of institutional investors: Mutual Funds (MF) and Foreign Institutional Investors (FII). 7 The Securities Exchange Board of India (SEBI) has a formal procedure for assessing if an applicant qualifies as a MF or FII. Chapter II of SEBI (Mutual Funds) Regulations 1996 details the eligibility criteria for registering as a MF 8. The sponsor 7 Several distinct categories of owners have been identified by the SEBI. These include Indian Promoters, Foreign Promoters, Private Holdings, Government Holdings (further broken down as Central, State or Other), Institutional Investors (further broken down as Mutual Funds, Banks and Financial Institutions, Foreign Institutional Investors). We exclude Banks and Financial Institutions in our analysis of institutional investors as we are not sure the extent to which their investment strategies are driven by the same considerations as MFs or FIIs. They may have exposure to debt as well as equity securities of the same firm and can engage in commercial banking relationships with these firms. 8 SEBI (Mutual Fund) Regulations 1996 are detailed in THE GAZETTE OF INDIA EXTRAORDINARY PART II - SECTION 3 - SUB-SECTION (ii) PUBLISHED BY AUTHORITY SECURITIES AND EXCHANGE BOARD OF INDIA NOTIFICATION MUMBAI, DECEMBER 9, 1996 (available at 4

7 should have a sound track record 9 and contribute at least 40% of the net worth of the asset management company. The directors, principal officers, trustees, custodians and the asset management company should be appointed as per the provisions of the regulations. On the other hand, entities such as Pension Funds, Mutual Funds, Insurance Companies, Investment Trusts, Banks, University Funds, Endowments, Foundations, Charitable Trusts/ Charitable Societies that are established or incorporated outside India qualify as FII. In addition, Asset Management Companies, Institutional Portfolio Managers, Trustees and Power of Attorney Holders established or incorporated outside of India who propose to invest on behalf of broad based funds also qualify as FII. 10 A renewable license is given to the FII once it is approved by SEBI. The existence of MFs in India dates back to However, between 1963 and 1987 there was only one MF, a government owned entity called Unit Trust of India. It had about INR 67 billion under management in In the late eighties, other public-sector banks started their own MFs. By the end of 1993, various private groups started offering MFs. At that time, about INR 470 billion was invested in MFs. Regulations were created in that year, with further revisions in As of August 2007, there are thirty-three MF companies in India, with a total of INR 4850 billion under management, of which about INR 1000 billion has been added in the last year alone. The growth in assets under management during the last few years is around 40%, with retail participation growing at 42%. However, most of the growth has come from urban areas and this industry lags behind retail banking in opening of new accounts. 11 Various international asset management companies already operate in India, with several more planning to enter soon. Under the portfolio investment scheme (PIS), FIIs are allowed to invest in both the primary and secondary capital markets. Securities are acquired through the stock exchanges. There is a limit of 24% equity ownership of a company by a FII. The overall limit for foreign investment is at 49%, with the remaining over and above the equity participation route being through FDI. In the case of investments in public sector banks, the limit is 20%. The limit can be raised to a sector or statutory ceiling with the approval of the board and shareholders. Several companies have 9 Sound track record implies (i) at least five years of business experience in financial services, (ii) positive net worth in each of the preceding five years, (iii) net worth in the immediately preceding year greater than the capital contribution by the sponsor into the asset management company, (iv) the sponsor has positive net worth after depreciation, interest and taxes in at least three out the five preceding years, including the fifth year. 10 These details about the eligibility criteria are available at the SEBI website to help FII applicants, 11 Source: 5

8 increased their FII participation using this method. India has increasingly become an attractive destination for FIIs. As of June 2007, there are 1042 registered FIIs as compared to 813 in January, In 2005, 209 new FIIs invested in India and in 2006 there were another 217. As of May 2007, the total amount invested by FIIs in equities is around USD 53 billion, with around USD 6 billion being invested during During Jan 2006 to April 2007, the four most active FIIs generated about USD 4.7 billion of trading in equities Methodology and momentum literature a. Earnings surprise methodology (SUE) The literature on earnings surprises is quite exhaustive. Starting with Ball and Brown (1968), researchers have made significant contributions in estimating earnings surprises. During the initial phase of the work done in this area, researchers had focused on developing time-series models for forecasting earnings (Ball and Brown, 1968; Foster, 1977; Brown and Rozeff, 1978). Considerable effort was spent on improving the forecasting methodology. However, as data on analysts earnings estimates were increasingly available in machine readable form for researchers to use, it was found that they did a better job than the time series models (Brown et al. 1987). In the case of the Indian stock market, unfortunately, the issuance of earnings estimates by analysts is still a nascent industry. Since a small set of firms are followed by analysts, using these estimates would severely impede our sample size. Furthermore, to our knowledge, there is no comprehensive database that captures these estimates. Given these arguments, we have used the time-series approach to assess the extent of the earnings surprise. We feel this is a reasonable approach, in particular because other authors have also used these models to examine postearnings announcement drift in the recent past when analysts estimates were available (e.g. Bernard and Thomas 1990; Bartov et al. 2000). We consider four different models to forecast earnings in order to examine the post-earnings announcement drift. However, in the subsequent analysis, we use one of these models only since 12 Source: The figures presented in this source were in USD, but for comparing it with other figures presented earlier, an approximate conversion rate of USD 1= INR 40 can be used. 6

9 the drift appears very similar using the different models. The first two models are specified in Brown and Kennelly (1972). The first model assumes that quarterly earnings follow a seasonal random walk model with no drift. This is also referred to as naïve model N1, which essentially assumes that the current quarter s earnings are the same as a year ago. The second model assumes that quarterly earnings follow a seasonal random walk model with a drift term. This is also referred to as naïve model N2, which essentially assumes that the current quarter s earnings are the same as a year ago plus the average change of quarterly earnings over the available data range. N1: E(Q t ) = Q t-4 N2: E(Q t ) = Q t-4 + δ Using data from for US stocks, Beaver (1974) concluded that N2 is mis-specified. He found the forecast errors to have significant first-order positive serial correlation, with adjacent quarterly earnings being not independent of each other. Foster (1977) improved on N2 by assuming that the autocorrelation follows a first-order autoregressive process. Essentially he added an autoregressive term to model N2. Foster: E(Q t ) = Q t-4 + φ 1 (Q t -Q t-5 ) + δ Foster (1977) noted that the main problem to the specification above is the assumption that the first-order autoregressive term describes the quarterly earnings behavior for all firms. Brown and Rozeff (1979) took it one step further, by adding a seasonal moving average term. Brown-Rozeff: E(Q t ) = Q t-4 + φ 1 (Q t -Q t-5 ) + δ + θε t-4 In these models the autocorrelation parameter φ 1 > 0. This captures the positive yet decaying autocorrelations in seasonally differenced earnings at lags 1 through 3. Further ε t-4 is the whitenoise shock for period t and the moving-average parameter θ is sufficiently negative to ensure that the fourth-order autocorrelation in seasonally-differenced earnings is negative. The seasonal moving-average term is included to account for the negative autocorrelation with the fourth lagged term. 7

10 The trend is estimated using data from the most recent 20 quarters, with the estimation period covering quarter t-21 through quarter t-1. Estimates using less than 10 valid observations are discarded. After we estimate expected earnings based on the alternative models specified above, we compute SUE it, the standardized unexpected earnings of firm i in quarter t, by scaling the difference between the reported and expected earnings by the standard deviation of forecast errors over the estimation period. We then rank the SUE it into deciles, based on sample distribution for each quarter and scale them to the range of -0.5 to +0.5 (RSUE it ). Bernard and Thomas (1990) make an important contribution to the understanding of the post earnings announcement drift anomaly. They focus on the autocorrelation structure of forecast errors using alternative time series models used in the prior literature 13. They note that the price drift pattern indicates that given the earnings surprise in one quarter, investors are surprised by the earnings surprise in the subsequent quarter and a significant portion of the drift occurs at the subsequent earnings announcement date. In other words, stock prices fail to reflect the implications of current earnings for future earnings. They attribute it to investors using naïve earnings expectations models when indeed the earnings pattern displays appears to have both auto regressive as well as moving average characteristics. They link the short window price reactions in the subsequent four quarters to the earnings forecast errors using a naïve model and thus conclude that the abnormal returns are predictable. b. Literature on momentum anomalies The behavioral finance literature has identified several anomalies in the stock pricing model. While the initial literature focused on discovery of such anomalies, some of the more recent papers have tried to resolve the common elements among the anomalies. Fama and French (2007) find that accruals, stock buybacks and momentum are the persistent anomalies, after controlling for the size and value (book-to-price) factors. Of these, they conclude that momentum is the premier anomaly. As mentioned earlier, Schwert (2003) also reaches a similar conclusion that momentum still holds in out of sample tests. He finds that the size, value, weekend and turn-ofthe year effect has decreased since the initial evidence was published. While these authors use price momentum to reach their conclusions, others have questioned the overlap of price 13 Watts (1975), Foster (1977), Griffin (1977), Brown and Rozeff (1979), Bathke and Lorek (1984), Brown et al. (1987). 8

11 momentum with earnings surprise momentum, since they may be capturing the same phenomenon (see for example, Chan et al.,1996). Price momentum, first documented by Jegadeesh and Titman (1993), found that stocks that outperformed in the previous twelve months continue to outperform in the subsequent twelve months as compared to the stocks that underperformed relative to the market. The post earnings announcement drift, first documented by Ball and Brown (1968), found that stocks with positive earnings surprises outperform the market for about nine months, suggesting that investors do not price in all the information contained in the earnings announcement immediately. Chan et al. (1996) investigate if the price momentum is explained by firm-specific elements of earnings surprises. They conclude that each captures returns that are not explained by the other strategy. Based on the evidence, they conclude that the market is slow to react to information in past returns as well as in past earnings. Chordia and Shivakumar (2002) find that the price momentum is related to business cycles. Further, Chordia and Shivakumar (2005) find that the price momentum is due to inflation illusion, where stock investors do not adjust earnings growth due to inflation. As an extension to Chan et al. (1996), Chordia and Shivakumar (2006) explicitly test the relation between these two anomalies and conclude that price momentum is captured by systematic component of the earnings surprise, not the firm-specific component of earnings surprise as used by Chan et al. (1996). They find that the predictive power of past returns goes away after controlling for earnings surprises. They also find that the earnings surprise is related to future macroeconomic information. In an international setting, Hong et al. (2003) find that price momentum is evident only in countries that exhibit earnings surprise. So the evidence suggests that there is a relation between the two momentum anomalies. c. Abnormal returns For firm i, we compute beta using the market model as below. E(R it ) = a i + beta i Mkt ii The Mkt is the return on the Bombay Stock Exchange (BSE) index. Beta i is computed on a weekly basis, using the returns over the previous 103 weeks. Then we compute the weekly betaadjusted return as AR it = R it E(R it ) 9

12 This weekly beta-adjusted return is compounded over various time intervals as specified in the tables to examine the effect of price and earnings surprise as well as the effect of the ownership structure. In addition to examining the returns over several weeks or months, we also examine the short window reaction to earnings surprises. For those measures, we compute Beta i on a daily basis using the returns over the previous 400 calendar days. CAR3 is the cumulative abnormal return (AR it ) over a 3 day interval from -1 to +1 around the event date. AR12 is defined as below. AR12 iq = CAR3 i(q+1) + CAR3 i(q+2) + CAR3 i(q+3) - CAR3 i(q+4) 4. Models After verifying the existence of post-earnings announcement drift in the Indian context, we want to examine what factors are causing such a drift. The initial question we have is whether earnings surprise exists, even in the presence of price momentum. Our first model examines if price momentum is evident for our sample. We specify the following relation to test this issue. AR i = a + b 1 RRet i + b 2 RSize i + b 3 RB_P i (1) where AR i is the beta-adjusted return, RRet i is the rank of the beta-adjusted prior returns, RSize i is the rank of the size and RB_P i is the rank of the book-to-price ratio. For these rank measures, we first compute the deciles and then scale them to the range of -0.5 to We use two different time intervals to determine the price momentum (RRet i ), the prior 3 months and the prior 12 months. In the initial model, we examine the dependent variable (AR i ) over 3, 6, 9 and 12 month intervals. The second model examines if earnings surprise, as captured by the earnings surprise measure, has any contribution to subsequent returns after controlling for price momentum. We include an interaction term between price momentum and earnings surprise in order to understand if there is any common element between the two. Our analysis is slightly different from Chordia and Shivakumar (2006), since we examine the extent to which price momentum at the margin influences the effect of earnings surprise on future returns. If both price and earnings surprise captures the same underlying information, then these are substitutes. On the other hand, if these represent different types of information, then they are complements. A negative sign would indicate that these two measures substitute each other, whereas a positive sign would indicate that 10

13 these are complements. We have no theory to develop expectations regarding the nature of the relation between price and earnings surprise, so the slope b 3 can be either negative or positive. 14 In addition to controlling for size and book-to-price, we also control for transaction costs using the trading volume and closing price similar to Bhusan (1994) and Bartov et al. (2000). AR i = a + b 1 RRet i + b 2 RSUE i + b 3 RRet i *RSUE i + b 4 RSize i + b 5 RB_P i + + b 6 BClosing_Price i + b 7 RAnnVol i (2) In addition to the variables from equation (1), BClosing_Price i is a dummy variable that take a value of 1 if the closing price is greater than INR 5 and takes a value of 0 otherwise. RAnnVol i is the rank of the annual trading volume for firm i. The value of b 2 will demonstrate if the PEAD anomaly indeed exists for this sample, after controlling for price momentum, other risk factors such as size and value, as well as proxies for liquidity. Next, we want to understand if there are sophisticated investors in the Indian market who exploit the PEAD anomaly. In order to understand what types of stocks display earnings surprise, we examine the relation between the PEAD measure and the institutional ownership structure. Bartov et al. (2000) suggest that the level institutional ownership of a stock can be a useful indicator for explaining the extent of the PEAD anomaly. Using US data, they find support of this hypothesis. The additional variable in this model RInstOwn i is the rank of the percentage of equity ownership by the institutions. The value of b 2 will indicate if the level of institutional ownership impacts the anomaly due to earnings surprise. If investigators are sophisticated, we expect b 2 to be negative, indicating that the PEAD effect is priced away for firms with higher levels of institutional owners. We test the result in the Indian context as specified in the model below. AR i = a + b 1 RSUE i + b 2 RSUE i *RInstOwn i + b 3 RSize i *RSUE i + b 4 RB_P i *RSUE i + b 5 BClosing_Price i *RSUE i + b 6 RAnnVol i *RSUE i (3) In our next model, we extend model 3 to further consider the effect of price momentum. In addition to institutional ownership, as in model 2, here we consider the effect of prior stock 14 As we have noted earlier, Chan et al (1996) and Chordia and Shivakumar (2006) have examined the relation between the two types of momentum. The conclusions can be different based on how the momentum measures are defined, systematic or idiosyncratic. 11

14 returns on the impact of earnings surprise in influencing subsequent returns. Unlike the mitigating effect of institutional ownership on the PEAD anomaly, we do not have an expectation of the effect of institutional ownership on price momentum. One can argue that both these are momentum anomalies and the institutional investor exploits them, so the effect should be similar. However, given the literature on institutional investors and positive feedback trading 15, we do not have any expectations on sign of b 5. As earlier, we expect the slope b 4 to be negative if institutional investors are sophisticated. AR i = a + b 1 RSUE i + b 2 RRET_B3 i + b 3 RSUE i * RRET_B3 i + b 4 RSUE i *RInstOwn i + b 5 RRET_B3 i *RInstOwn i + b 6 RSize i *RSUE i + b 7 RSize i *RRET_B3 i + b 8 BClosing_Price i *RSUE i + b 9 BClosing_Price i * RRET_B3 i + b 10 RB_P i *RSUE i + b 11 RB_P i *RRET_B3 i + b 12 RAnnVol i * RSUE i + b 13 RAnnVol i *RRET_B3 i (4) The models specified in equations 3 and 4 above examines whether a PEAD phenomenon exists even in the presence of price momentum and holdings by sophisticated investors. The models examine what explains subsequent returns after an earnings announcement. If the value of b 4 is negative, we can infer that for stocks with higher institutional ownership, the PEAD anomaly is muted probably because these investors exploit it. However, the last conclusion is by inference only. Whether institutional investors actually exploit this opportunity cannot be concluded from the models above. In addition, the value of b 5 will indicate how institutional owners exploit the price momentum measure. Ke and Ramelingegowda (2005) use a model to directly examine whether transient institutional owners exploit the earnings surprise measure to generate profits. InstOwn it = a i + 3 q= 0 b q RSUE it-q + b 4 Size it + b 5 B_P it + b 6 InstOwn it-1 + b 7 RETQ24 it + b 8 RETQ1 it + b 9 RETQ0 it + b 10 PWInstOwn it-1 (5) This model examines the change in the level of institutional ownership due to the earnings surprise from the last four periods. In addition to size and book-to-market, the level of institutional ownership and the weight of the stock in the institutional investor s portfolio are 15 For example, see Nofinger and Sias (1999), Lee et al. (1999) and Lakonishok et al. (1992). 12

15 controlled for in explaining the change in ownership. Ke and Ramelingegowda (2005) use this model to understand how transient institutional investors (i.e. those actively trading to maximize short term profits) exploit the PEAD strategy. We use this model to understand which type of investor actually invests using the earnings surprise strategy. The list of variables and their definitions are summarized in Table Insert Table Sample The Bombay Stock Exchange (BSE) is the oldest stock exchange in Asia, established in Of the twenty three stock exchanges in India, the BSE is the largest. Around six thousand Indian firms are listed with the stock exchange and accounts for two thirds of the trading volume in India. The BSE groups listings into ten different groups. 16 The A group consists of the most liquid stocks, followed by B1 and B2. The Z group consists of those stocks that do not comply with BSE listing requirements. Stocks with unusual price movements are put in the T group. --- Insert Table 2a --- Our data source is Prowess, which is a database compiled by the Centre for Monitoring the Indian Economy. It contains financial statement as well as market price information for all publicly 16 The groups have been developed based on (i) compliance with SEBI norms and (ii) trading and settlement cycles. Groups A, B1, B2, C, S, TS, T and Z refer to equities and F, G refers to debt and government securities. With effect from April 1, 2003 all trades are settled on a T+2 basis. The A group includes securities that can be carried forward to the next settlement cycle. The securities in this group are selected on the basis of equity capital, market capitalization, number of years of listing on the exchange, public share holding, floating stock, trading volume etc. B1 and B2 are a subset of the other listed shares that have higher market capitalization and liquidity. C group covers the odd lot securities in A, B1 and B2 groups. Odd lot refers to those securities where the number of transactions is less than the market lot, a minimum number of transactions set by the exchange. T represents securities that are settled on a trade to trade basis as surveillance measure, where no netting will be allowed and trades will be settled in the T+2 cycle. Essentially, this category allows BSE to flag securities to counter any unwarranted price movements. Z group consists of the companies which does not comply with the rules and regulations of BSE and are at times suspended from trading. The S and TS category has been created effective Jan 1, All B1 and B2 securities which are exclusively listed in the regional stock exchanges with market capitalization between INR 30 million and 300 million qualify as the S group. TS are those within the S group which are under surveillance. 13

16 traded as well as a large number of private firms in India. 17 The steps for selecting our sample is presented in Table 2a. We collected earnings announcement dates for 201 group A firms and 595 group B1 firms for the period between March 31, 2001 and June 30, 2006, giving us an initial sample size of 796 firms. However, we only used 582 firms in the final sample as the remaining 214 firms have more than 2 quarters with missing earnings announcement dates. Out of the final sample of 582 firms, 202 have earnings announcement dates for all twenty two quarters included in our sample period; 286 have one and 94 have two missing values for earnings announcement dates. We verified the data provided by Prowess with publications in the financial press following an earnings announcement. After a board meeting where quarterly earnings are announced, the firms inform the stock exchange of these results. In the case of group B1 firms, which are comparatively smaller than the group A firms, this information is captured by the news media in the following day and gets published the day after that. For group A firms, the news media capture the information on the same day as the board meeting is held and publishes it the following day. --- Insert Table 2b --- The characteristics of the firms in our sample are presented in Table 2b. The mean (median) annual trading volume is INR (0.08) trillion. They have a mean (median) book to price ratio of 2.6 (0.84) and a mean (median) market capitalization of INR 13.2 (1.89) billion. We notice that our proxy for firm size, log of market capitalization has a mean value of 21.5, but a low standard deviation of 1.9. So while there is variation in growth versus value characteristics, our sample represents primarily large cap stocks, without much variation in size. The median book value is INR and the mean price is INR The mean (median) of earnings forecast error using the Brown and Rozeff model is INR 1.27 (0.20), with standardized unexpected earnings of 0.13 (0.08). The mean ownership by MFs are 3.75% and by FIIs are 4.52%, although the median values are 2.26% and 0.56% respectively. The maximum values of ownership by MF and FIIs are 25% and 53%. We thus notice that there is sufficient variation in earnings surprise as well as institutional ownership for these firms. While these ownership levels hold for this sample, it should be kept in mind that as reported in an earlier section, the total 17 This database has been used for several empirical studies on India e.g. Khanna and Palepu (2000), Bertrand et al. (2002) and Gopalan et al. (2007) 14

17 amount invested by MFs in the Indian equity market is significantly more than that invested by FIIs Results In order to understand the time series behavior of the quarterly earnings figures, we first examine the auto correlation structure of alternative earnings based measures. These results are presented in Table Insert Table The seasonally differenced earnings process indicates that there is a positive correlation for three consecutive quarters, with decreasing magnitude. Starting from the fourth quarter, there is a negative correlation. These results are consistent with the evidence from the US markets (see table 1 of Bernard and Thomas, 1990). We then examine the autocorrelation of the earnings surprise and the ranks of the earnings surprise measures. The results are slightly different, with the third quarter showing a negative and the fourth quarter showing a positive autocorrelation. Interestingly, this phenomenon is again evident in the eight quarter. While this particular result does not affect the results presented later in the paper, it does suggest that earnings surprise based on these time-series models shows some positive serial correlation on an annual basis unlike the reversal observed in the US markets. --- Insert Table 4 and Figure We next examine the existence of the post earnings announcement drift (PEAD) effect. Using the alternative models to forecast earnings, we form a long-short portfolio based on the standardized earnings surprise (SUE) measure. This portfolio captures the return when the investor buys stocks with the largest SUE decile and shorts stocks with the lowest SUE decile. The PEAD effect has been examined extensively in the US context, but to our knowledge, this is the first time it has 18 The following figures (not reported in the Table 2b) will give a better idea about distribution of some of the variables. The minimum trading volume is INR 7 million, INR142 million and INR 582 million at the 1 st percentile, 5 th percentile and 10 th percentiles respectively. Similarly, the per share book value is INR , and 16.7, the book-to-price value is INR -0.37, 0.12, 0.18, the closing price is INR 2.06, and at the 1 st, 5 th and 10 th percentiles respectively. In addition, ownership percentages at the 1 st quartile for MFs and FIIs are 0.14% and 0 respectively. 15

18 been examined for the Indian market. As reported in Table 4 and evident from Figure 1, we conclude that for this sample, there is strong evidence of an earnings surprise anomaly. The PEAD effect continues to pay off for about fourteen to sixteen weeks, depending on which forecast model is used. The mean (median) interval between earnings announcements for this sample is 93 (91) days. The cumulative return over twelve weeks is about 10 percent due to this PEAD effect, depending on which model is used. The cumulative return levels of around 12 percent over a slightly longer interval of up to twenty weeks. Next we want to understand what factors are causing the PEAD effect. Various explanations have been provided in the literature. The most common explanation is that of investor under reaction to new information and the inability of investors to understand the implications of current earnings for future earnings (Bernard and Thomas, 1990). The first model to understand the determinants of returns includes a rank measure of the prior returns in addition to size and book to price ratio, the two factors that have been found by Fama and French (1996) to be relevant in addition to the market return. In essence, this rank measure captures the price momentum. --- Insert Table Using the model specified in equation 1, we examine the returns subsequent to earnings announcement over three, six, nine and twelve months (see Table 5). There are two flavors of the price momentum measure we consider, one using a prior three month interval and the other using the prior year s return. The dependent variable is the beta-adjusted return, where the beta is computed using weekly BSE 500 returns over 104 weeks. From table 5, we notice that the three month price momentum is significant in explaining the subsequent three to nine month returns. Using a one-year price momentum measure, we find it to be significantly negative in explaining the six to twelve month returns. In other words, there is a strong reversal effect over the next six to twelve months if we consider the previous twelve month returns. We also notice that the value factor, book to price ratio, is significant for the six to twelve month window. However, the coefficient for size is not significant. Given that our sample primarily consists of large cap stocks, this result is not unexpected. These results indicate that for our sample of firms, depending on how price momentum is defined, there is either a positive or negative effect in determining future returns. A short window (three month) price momentum continues to pay off in for a short to medium term into the future, whereas a longer window (twelve month) price momentum reverses itself over the medium to long term into the future. 16

19 --- Insert Table We next examine the effect of earnings surprise on future stock returns after controlling for the Fama French factors, size and book-to-price as well as price momentum. These results are presented in Table 6. We also control for the closing price and the trading volume as in Bartov et al. (2000). Similar to the analysis in Table 5, we consider two alternative measures of price momentum i.e. return over the previous three months and that over the previous one year. The results presented in Table 6 are based on the model specified in equation 2. It should be noted that Bartov et al. (2000) do not present any results for a model similar to equation 2, i.e. without the effect of institutional ownership. For the three month measure of momentum, we find that the earnings surprise element is significant at less than 1 percent for the subsequent three through twelve month returns. Price momentum is only significant for the next six month window at the 5 % level. Interestingly, we find that over the next 3 month period, the interaction term is significantly negative, indicating that these two measures are substitutes. We find the value measure (book-to-price) continues to be significant for the 6 to 12 month return window. In addition, the closing price of the stock shows a significant negative slope, indicating that low price stocks (at below INR. 5) require a liquidity premium. Furthermore, the annual trading volume is also found to be significantly negative, again indicating that lower volume stocks require a higher return. However, the annual trading volume can represent liquidity or efficiency of price formation. A surprising finding is the strongly positive slope for size, which basically suggests that in this model, larger size generates higher returns. At first it appears counterintuitive and is the opposite of the results from Table 5. However, in the absence of closing price and annual trading volume, if we believe that size captures liquidity and pricing inefficiency, then the positive sign for size in Table 6 is not unexpected. 19 In the lower panel of Table 6, we report the results when price momentum is measured over a prior twelve month interval. We find that the signs and significance levels continue to be the preserved for earnings surprise, size, book-to-price, closing price and annual volume. The negative sign for the interaction term between price momentum and earnings surprise is 19 Size as a risk factor has been explained to capture primarily distress (Reignanum, 1983; Kross, 1985) and information risk (Roll, 1981; Bhardwaj and Brooks, 1992). Book-to-price has also been attributed to capture distress risk. If closing price and trading volume capture information and liquidity risk, then risk elements that size normally represents are being controlled by these variables. It is our belief that in this model, size is capturing something else, perhaps linked to visibility or market presence. 17

20 significant for both three and six months. As is the case in Table 5, we find that the slope for the twelve month price momentum is significantly negative. This confirms that reversal effect continues to hold even after controlling for closing price and annual trading volume when price momentum is measured over prior twelve months. --- Insert Table In the next analysis, we replicate the basic test conducted by Bartov et al. (2000). The results are presented in Table 7. The model specified in equation 3 is used to examine the incremental effect of institutional ownership, size, book-to-price ratio, closing price and volume on earnings surprise. The interaction terms indicate whether these factors reduce the effect of earnings surprise on subsequent stock returns. There are four alternative measures of subsequent returns that are examined. The drift in the returns over twelve and sixteen weeks, the three day returns around the earnings announcement and the net effect of the three day reactions over four quarters. As mentioned earlier, we consider two separate types of institutional investors, MFs and FIIs. We find that the three day returns as well as AR12 are significantly lower for larger firms. In addition, high book-to-price also leads to lower AR12. However, the interaction term with institutional ownership offers some interesting results. The analysis using MF ownership indicates that the earnings surprise increases for AR12. In all other measures of subsequent returns, the ownership variable does not make a difference to the earnings-surprise effect. The results using FIIs are quite different. Similar to the findings of Bartov et al. (2000), we find that PEAD measured over the next twelve and sixteen weeks is significantly lower in the presence of FIIs. However, for the three day and AR12 windows, there is no effect of FII ownership on PEAD. So these results indicate the stocks which have higher FIIs exhibit a lower PEAD. However, the short window return around earnings announcements is not affected by it. --- Insert Tables 8a and 8b --- Given the recent literature (Chan et al.,1996 and Chordia and Shivakumar, 2005b) on the relation between price momentum and earnings surprise, we extend the model used in Table 7 to examine the role of price momentum in the same setting. These results are presented in Tables 8a and 8b. Essentially, we are interested in finding out if price momentum explains away the relation between the PEAD and institutional ownership. The model in equation 4 is used to obtain the results in Table 8a and 8b. In addition to the terms considered in equation 3, in this model we 18