Annals of the University of North Carolina Wilmington International Masters of Business Administration.

Transcription

1 Annals of the University of North Carolina Wilmington International Masters of Business Administration

2 A COMPARATIVE ANALYSIS OF MARKET EFFICIENCY: THE CASE OF RUSSIA AND THE U.S. Daria Levkina A Thesis Submitted to the University of North Carolina Wilmington in Partial Fulfillment of the Requirements for the Degree of Master of Business Administration Cameron School of Business University of North Carolina Wilmington 2010 Approved by Advisory Committee Edward Graham Peter Schuhmann Nivine Richie Chair Accepted by Dean, Graduate School

3 TABLE OF CONTENTS ABSTRACT... iii LIST OF TABLES...iv LIST OF FIGURES...v INTRODUCTION...1 LITERATURE REVIEW...3 Literature on Market Efficiency... 3 Literature on Event Study... 4 Literature on Post Earnings Announcement Drift... 5 Literature on Market Efficiency in Different Countries... 6 DATA COLLECTION...8 METHODOLOGY...12 EMPIRICAL RESULTS...17 Event Study Market Model Results Mean Adjusted Model Results Event Study Summary Cross Sectional Analysis CONCLUSION...26 FURTHER RECOMMENDATIONS FOR RESEARCH...29 REFERENCES...32 APPENDIX...34 ii

4 ABSTRACT The prime concern of this study is to examine the process by which stock prices adjust to information in two different countries the U.S. and Russia under the assumption that rapid information flows could confuse investors, leaving them little time for analyzing and verifying data. The results suggest U.S. market efficiency and the relative inefficiency of the Russian stock market. The mean adjusted model shows larger cumulative abnormal returns (CARs), except for the event period for U.S. companies, and generally confirms the results from the market model. These results suggest that the increasing speed of information flow improves market efficiency. iii

5 LIST OF TABLES Table Page 1. Descriptive Statistic, Correlation Matrix Market Model CARs for Russian Companies Cumulative Abnormal Return from Market Model for Russian Companies Cumulative Abnormal Return from Market Model for US Companies Mean Adjusted Returns for Russian Companies Market Model for US Companies Cumulative Abnormal Return from Mean Adjusted Model for Russian Companies Mean Adjusted Returns for US Companies Cumulative Abnormal Return from Mean Adjusted Model for US Companies Regression Parameters Estimates for Market and Mean Adjusted Models...50 iv

6 LIST OF FIGURES Figure Page 1. Average Market Model CAR for Russian Companies Average Market Model CAR for US Companies Average Mean Adjusted CAR for Russian Companies Average Mean Adjusted CAR for US Companies...52 v

7 INTRODUCTION In today s technologically advanced marketplace, prices may not adjust to new information at the same speed all over the world. Differences in information technology suggest that the availability of data and the mechanisms by which information is provided to the public are not the same for developed economies relative to emerging economies. The computer era began in developed countries long before it began in developing countries, setting the stage for greater market efficiency in developed markets. On the other hand, the fast pace of information transfer common in most developed countries leaves little time for verifying and analyzing data. Consequently, investors could be misled into making wrong decisions based on false or incomplete data, resulting in decreased market efficiency. If markets are efficient, then stock returns already reflect all available information and thus buying or selling the stock should, on average, return you only a "fair" measure of return for the associated risk (Bodie, Kane, & Marcus, 2008). In this study, I investigate the differences in market efficiency in emerging markets relative to developed markets by comparing Russian market efficiency with U.S. market efficiency. Using event study methodology, I estimate abnormal returns following annual earnings announcements for the 50 largest Russian firms and 50 largest U.S. firms. I find evidence of post earnings announcement drift and pre announcement leakage in the Russian market but no such effects in the U.S. market. The results indicate that market efficiency does indeed differ across countries with different information technology and speed of information transfer. Furthermore, cross sectional analysis reveals that the speed with which stock prices adjust to new information depends on the market capitalization of companies and markets efficiency is

8 more common for the U.S. market rather than for the Russian market. Even after controlling for firm size, Russian markets exhibit market inefficiency that supports the statement that markets are more efficient in developed economies, possibly due to the speed and quality of information transfer in developed economies, and reject the hypothesis that the market efficiency might suffer from modern fast pace information technologies. This study is organized as follows. The next section presents the literature review discussing market efficiency. Data collection and methodology for the event study are then explained, followed by the empirical results. The final section presents the conclusion and recommendations for further research. 2

9 LITERATURE REVIEW The topic of market efficiency has received a great deal of attention in the academic literature. Four principle areas of research interest are evident: studies about market efficiency in general, literature using event studies to analyze market efficiency, post earnings announcement drift studies, and modern researches related to market efficiency in different countries. I examine research in all four areas. Literature on Market Efficiency The first studies on market efficiency appeared in the 1950s. Kendall (1953) examined market efficiency and showed that there is no predictability in price movements. Subsequent to the publication of this study, Kendall s conclusion was widely investigated and many researchers confirmed the notion of the unpredictable behavior of price movements. Fama (1965) concluded that stock price movements cannot be predicted and follow a random walk. Later on, many studies examined market efficiency from the prospective how different events influence the speed at which stock prices adjust to new information. Beaver (1968) investigates investor reaction to annual earnings announcements and finds that earnings possess information content, since trading volume and stock price volatility increase during the week surrounding earnings announcements. Cornell and Landsman (1989) find evidence that annual earnings announcements have greater stock price impacts with comparison to interim quarters earnings announcements. This position might be supported by the fact that annual earnings are usually audited and are more reliable. They also conclude that the fourth quarter announcements are more informative for the analysts and investors than their interim quarter announcements. 3

10 The results of the research of Kross and Schroeder (1990) suggests that there are differences between the information content of interim quarter and fourth quarter earnings announcements only for small firms which have annual earnings response coefficients less than interim earnings response coefficients. This result was not discovered for large firms. According to their arguments, the accounting information of large firms is subject to continual auditing and monitoring, causing the differences between interim quarters and fourth quarter announcements to be insignificant. Opposite results are found by Salamon and Stober (1994), who suggest some evidence that there is no difference in price response between small and large firms, and that for all firms, fourth quarter earnings response coefficients are smaller than interim quarter response coefficients, meaning that annual earnings announcements bring less information to users than do quarterly announcements. As we can see, there are different and sometimes controversial results on the market efficiency issue, which provides opportunity for further exploration of the topic. Literature on Event Study The event study methodology has been widely used to examine market efficiency. Fama (1969) launched one of the first event studies seeking to analyze how stock prices respond to an event a dividend stock split by using price data from the Center for Research in Security Prices database. The prime focus of his research is to examine the process by which common stock prices adjust to the information of a stock split. Evidence from this study indicates that on average the information of a stock split is fully reflected in prices, at the latest, by the end of month the split takes place. However, in most cases, stock prices adjust to this information almost immediately after the announcement date (Fama, 1969). 4

11 Various methodologies are used to run an event study. Abnormal return is usually calculated by subtracting firm specific average returns from past periods, risk adjusted (based on market model) or market returns from the actual return (Boehmer, Broussard, & Kallunki, 2002). Some studies prove the reliability of the mean adjusted model as well as the risk adjusted model to obtain the results for an events study. Brown and Warner (1985) show that abnormal returns measured with simpler estimation procedures, such as market adjusted and mean adjusted returns, display no significant mean bias. Moreover, Binder (1989) concludes that the market model works well as a measure of the benchmark rate of return. A one factor market model might work at least as well as the alternatives. Literature on Post Earnings Announcement Drift An important field for investigation is the issue of post earnings announcement drift, which represents the anomaly of the sluggish response of stock prices to firms earnings announcements (Bodie, Kane, & Marcus, 2008). Stocks cumulative abnormal returns tend to increase for a certain time after earnings announcements and the direction of the changes depends on the quality of the news. In other words, it continues to be positive for positive earnings surprises and negative for negative news, when actual earnings per share are respectively more or less than was expected. Ball and Brown (1968) are the first to note the anomaly that even after earnings are announced, estimated cumulative abnormal returns continue to drift up for "good news" firms and down for "bad news" firms. Bernard and Thomas (1989) argue that competing explanations for post earningsannouncement drift fall into two categories. The first reason might arise as a result of the delay 5

12 of price response due to the fact that information cannot be disclosed immediately by all market participants. Another reason for post earnings announcement drift is the capital asset pricing model (CAPM), which may not account for all the important factors or may provide wrong estimation of existing factors which, overall, will lead to inaccurate estimation of abnormal returns. As a matter of fact the CAPM model does not fully adjust for the risk. Finally, they conclude that much of the evidence cannot be reconciled with the risk mismeasurement, but it is more consistent with a price response delay. Literature on Market Efficiency in Different Countries Comparisons of market efficiency between countries might give an idea how to diversify the investment portfolio. The modern literature contains a few studies on market efficiency in emerging markets and developed economies. Ojah and Karemera (2005) document evidence that shows that equity prices in major Latin American emerging equity markets, such as Argentina, Brazil, Chile and Mexico, follow a random walk, and generally prove a weak form of efficiency in these markets. Because expected returns cannot be predicted by means of information from the past, as the future returns are not dependent on the past returns, there is little chance for arbitrage opportunities and creating trading schemes for market participants. Another study started by Mecagni and Sourial (1999) examines the behavior of stock returns in the Egyptian stock exchange, the efficiency of the market in pricing securities, and the relationship between returns and volatility. By analyzing the four best known daily indices, the researchers find that there is significant inefficiency in Egyptian stock market. Using data from nine different developing countries (Argentina, Brazil, Chile, Colombia, India, Korea, Mexico, Thailand and Venezuela), Kawakatsu and Morey (1999) indicate that 6

13 market efficiency of emerging markets is not improved after liberalization of the markets, meaning that increased availability of information for market participants does not affect market efficiency. In fact, they find that the markets were already efficient before liberalization. Looking at Central and Eastern Europe, some researchers conclude that volatility of stock prices does not explain expected returns for countries like Croatia, Czech Republic, Hungary, Poland, Russia and Slovakia and, as a result, future volatility cannot be predicted (Murinde & Poshakwala, 2001). As was indicated by other researchers investigating market efficiency in countries including the most developed nations as Australia, Belgium, Canada, Denmark, Finland, France, Germany, India, Italy, Japan, Netherlands, Norway, Spain, Sweden, Switzerland, the United Kingdom and the United States, stock prices follow a random walk, as well, and all stock markets examined show a weak form of efficiency (Chan, Gup, & Pan, 1997). Overall, there is extensive research on the subject of market efficiency, much of which uses event study as a tool and abnormal return as a measure of market reaction to certain events. However, there is still a niche representing comparison of current market efficiency in various countries. This paper attempts to fill that niche by analyzing price response to annual earnings announcements of large Russian and U.S. companies. 7

14 DATA COLLECTION For the purposes of this study, 50 U.S. and 50 Russian companies are chosen. The sample consists of the large firms included in the RTSI and the S&P500. The criteria for identifying the number of companies is based on the total number of Russian companies included in RTSI. The fifty largest U.S. companies are obtained from S&P500 and represent around 50 percent of its weight (exact number varies on a daily basis). As of April, 22, 2010, the total weight of the 50 largest companies in S&P500 was percent of the index. The most commonly used event for investigation among researchers is the earnings announcement (both quarterly and annual), and less rarely stock splits, dividend payments, mergers and acquisitions and company specific events. Also, when it comes to earnings announcements, it would be more meaningful to choose the period of earnings announcements having the greatest impact on stock prices in order to find noticeable evidence for the hypothetical question. So, in current circumstances, and due to the availability of information for the purpose of present analysis, the event study is based on annual earnings announcements, as the year end earnings announcements have stronger price change responses (Cornell & Landsman, 1989). This position is supported by basic logic, that it is mandatory nowadays for large firms to have their annual data audited, making it more informative for investors. Moreover, annual earnings announcements represent commonplace events which take place in nearly every company, regardless of its relation to any particular country. So the dates of the latest annual earnings announcement were obtained for each company. 8

15 Then the daily prices for the period from 1/1/2008 till 4/21/2010 were collected and used to calculate cumulative abnormal return. In order to estimate parameters for the market model, S&P500 and RTSI prices were collected on a daily basis for the same period. For calculation of the expected return during the event period, as well as before and after, the estimation period of 255 days before the event and ending 46 days prior the event date is used. The estimation period is the period from which the data is taken to estimate the abnormal return for the event period, whereas event period means the period of time, measured in days surrounding the earnings announcement date during which the stock market responds to the earnings announcement (Boehmer, Broussard, & Kallunki, 2002). The longer the estimation period, the more accurate the result of the study. Since there is a probability of information leakage, which might create a situation in which stock prices will have already reflected this information, the models are constructed based on those observations ending 46 days before the event. For the purpose of this study, the event period chosen was 1 day before the event day. It is more common to use a 2 days interval starting with one day prior the event day. If the event window is too long, other information might influence the return thereby making the statistical result less reliable. On the other hand, using too short an event window might create a situation which does not capture the time when investors truly learn about the event (Boehmer, Broussard, & Kallunki, 2002). In order to reveal leakage of information before announcement dates, as well as post earnings announcement drift, a different combination of windows such as ( 30; 21), ( 20; 16), ( 15; 11), ( 10; 6), ( 5; 4), ( 3; 2) before the event period and (1;2), (3;4), (5;10), (11;15), (16;20), (21;30) after the event period are used to capture these trends. 9

16 For cross sectional analysis, information about market capitalization, debt to equity ratio, operating margin, and price to earnings ratio was obtained on 12/31/2009 in order to utilize recent annual earnings announcement data and to be more specific in the regression analysis. Numbers of institutional holders are represented by current data, due to the lack of availability of historical information. All data was downloaded by means of a Bloomberg terminal. Descriptive statistics are calculated for U.S. and Russian firms to specify what kind of companies are included in both samples. The companies are examined from the perspective of size, leverage, profitability, valuation and number of institutional owners. Table 1 shows that the market capitalization mean in the U.S. sample is $ mln, which ranges from a minimum of $ mln to a maximum of $ mln. The Russian sample of firms has the mean market capitalization of $ mln. This indicator ranges from a minimum of $ mln to a maximum of $ mln. The median for U.S. companies is $ mln, whereas the Russian sample of firms has a median of $ mln. Standard deviations are $ mln and $ mln for U.S. and Russian firms respectively. So we can conclude that in terms of size, U.S. companies are, on average, bigger than Russian companies, but the variety as well as dispersion of companies is more common for the Russian sample, where there are relatively smaller companies, and a few with extremely large market capitalization. U.S. companies are mostly represented by firms with high levels of market capitalization. From the leverage perspective, a higher level of liabilities in capital structure is more common for U.S. companies as the mean of debt to equity ratio for the U.S. sample is reaching a maximum of , while for Russian companies debt to equity ratio is on average with the highest value of Standard deviations of this ratio for the U.S. sample and 10

17 for the Russian sample are and 51.60, so the variability of the leverage level is higher in the U.S. samples. Since the median of for the U.S. companies is smaller than the mean, we can say that a high level of liabilities is common for a small number of U.S. companies. The mean of operating margin for the U.S. sample is ranging from 2.85 to These results are very similar to the Russian sample which has the mean of and ranges from 2.25 to Overall, both samples consist of profitable companies, but profitability of the Russian sample companies is slightly higher than the U.S. sample firms. The mean price to earnings ratio for the U.S. sample is and ranges from 7.37 to with a standard deviation of The Russian sample firms have the mean price toearnings ratio of 57.92, which ranges from 4.71 to with standard deviation of So, Russian companies are generally overpriced, although as the median for this sample is only 15.34, which is smaller than the mean and almost equal to the median for the U.S. companies. We can conclude that only a few companies are overpriced. The average number of institutional owners for U.S. companies is 1,938 ranging from 1,124 to 2,823, whereas for the Russian sample, there are approximately 35 institutional owners on average, and the range extends from a minimum of 3 to a maximum of 76 institutional owners. So, it is obvious that a smaller number of institutions hold Russian stocks relative to U.S. stocks. Overall, the U.S. sample companies are mostly represented by large profitable companies which actively use leverage to finance their assets. On the other hand, the Russian sample companies generally have smaller market capitalization, lower levels of debt, higher profitability and a few of them are overpriced. The U.S. stocks are widely held by institutions, which is not very common for Russian firms. 11

18 METHODOLOGY The main idea of this event study is to estimate which portion of the price change was caused by the annual earnings announcements. Since the key component of this research is to examine price responses, it is necessary to calculate the abnormal returns at the event s period. SAS with Eventus is used to complete the calculations portion of the event study. Abnormal return represents the difference between actual return and expected or normal return in the absence of the event. AR i, t = R i, t E ( R i, t ) (1) There are several models to calculate expected return depending on which benchmark will be used as a measure of expected or normal return. Generally speaking, it could be meanadjusted return, risk adjusted return based on the market model, or and market adjusted return. Mean adjusted returns are calculated by subtracting the average return for stock i during the estimation period from the stock s return during the event period. This method does not explicitly control for the risk of the stock or the return on the market portfolio during the period. Compared to the market model, this approach is at best only slightly simpler, because one rather than two parameters are estimated and no market returns are required. The market adjusted return subtracts market return from actual stock return. This method is also simpler than estimating market model abnormal returns because it is done in one step, rather than two. So when the market model is used, parameters are estimated in the first step and abnormal returns are estimated during the event period in the second step. When the marketadjusted return is used, no statistical parameters are estimated. 12

19 Finally, the market model approach based on risk adjusted returns is straightforward and relatively easy to use. Parameters are estimated using an estimation period sample with ordinary least squares (OLS) regression. The parameter estimates and the event period stock and market index returns are then used to estimate the abnormal returns. This method gives an opportunity to control for the market risk (beta) of the stock and the movement of the market during the event period (Brown & Warner, 1985). It is very important to choose the correct model in order to avoid misspecification, as pointed out in the econometrics literature. Misspecification can occur either because relevant variables have been omitted or irrelevant variables have been included. However, when a large sample of unrelated securities is investigated, the market model can provide unbiased estimations of the average abnormal return (Binder, 1998). Abnormal returns, measured with a marketadjusted model and by a mean adjusted model display no significant mean bias as well (Brown & Warner, 1985). Relying on researchers opinions discussed earlier, a market model and a mean adjusted model are used to calculate abnormal returns. The market model could be expressed as follows: R i, t = α i + β ir m, t + ε i, t (2) where t = 255,, 46, R m,t, is market return, R i,t is actual stock return, α i is constant term for the ith stock and β i is the market beta of the ith stock. Then abnormal return for the event window dates are calculated as: AR i, t = R i, t α i β i R m, t ˆ ˆ (3) 13

20 where t= 1,0, ά i and β i are estimated parameters. After calculating the abnormal return for each stock in its event period, it is necessary to estimate average abnormal return for N=50 events for each day in the event window: AR t = 1 N N i = 1 AR i, t (4) Finally, as it is suggested by Fama (1998), post event drift of abnormal returns is a common finding. Most important is that this anomaly can be due to methodology, and it tends to disappear with reasonable changes in technique. He argues that theoretical and statistical considerations suggest that cumulative abnormal return should be used, rather than buy and hold abnormal returns. The reason is that the mean of the CARs increases with N, the number of months summed, but the standard error of the CAR increase more slowly, like the root square of N (Fama, 1998). So CAR is calculated over the event windows instead of using only abnormal return in order to achieve proper estimates. CAR = e t = b AR t (5) The same procedure is repeated for mean adjusted return, with exception that abnormal return is calculated as follows: AR i, t = R i, t R i (6) In order to estimate the level of significance of the event study results, two additional tests are implemented. First is a standardized cross sectional test (STDCSECT). This option 14

21 selects the standardized cross sectional test, which compensates for a possible variance increase on an event date by incorporating a cross sectional variance adjustment (Cowan, 2007). It represents an extension of the Patell test and was introduced by Boehmer, Musumeci and Poulsen (1991). Second is a cross sectional test (CSECTERR). The standard error for this test for each date (or window) in event time is computed across securities, not across time (Cowan, 2007). In order to understand the reasons for differences in the level of market efficiency between two countries, cross sectional analysis is launched and the regression models on CAR are estimated. The impact of different companies characteristics such as size, capital structure, profitability, stock valuation and ownership structure are analyzed. More specifically, both riskadjusted CARs and mean adjusted CARs for the period (1,30) obtained from each firm are regressed on market capitalization, debt to equity ratio, operating margin, price to earnings ratio, number of institutional owners and a dummy variable for countries. The CAR for this period reaches the highest amount after the event date. Increasing this window would not make sense as it is too far from the event. Correlation coefficients are calculated between independent variables to reveal the potential for multicollinearity. This issue would not decrease the predictive reliability of the model as a whole, but it would confound the individual parameter results, making it unclear which independent variables are statistically significant in explaining the variations in dependent variable in the multiple regression model. Table 2 shows that within the chosen variables, two pairs which are noticeably correlated are Market capitalization and Price to earnings ratio, Number of institutional owners and a dummy variable for countries with correlation coefficients of 31 and 94, respectively. 15

22 The results, based on comparison of R Square, test statistics and p value, are obtained from regressions of risk adjusted CARs and mean adjusted CARs for the period (1,30) from each firm on market capitalization, debt to equity ratio, operating margin, price to earnings ratio, number of institutional owners and dummy variable for countries. After running a battery of regressions, the most statistically significant variables are market capitalization and the dummy variable for Russian companies, although some models show evidence of multicollinearity. The multiple regression model, which most accurately expresses the findings, could be specified as follows: CAR i =µ 0 + µ 1 MC i + µ 2 D i +ε i,, where CAR i cumulative abnormal return for window (1,30) for each company, MC i market capitalization for each company at 12/31/2009 1, D i a dummy variable equal to 1 for Russian companies, 0 otherwise, µ s the ordinary least square (OLS) regression coefficients, ε i error term. 1 When data were not available for this date currently available data were used. 16

23 EMPIRICAL RESULTS Event Study In order to analyze market efficiency in Russia and in the U.S., the event study is launched and the price response to the latest annual earnings announcements of 50 companies from each country is examined. Market Model Results Table 3 shows the cumulative average abnormal returns from the market model for Russian firms for the event window as well as for different pre, post event periods. It shows that there are subsequent CARs exceeding 1.5 percent per 2 days starting 5 days prior the event and exceeding 2 percent per 2 days for the next 4 days after the event. Parameters for market models are estimated by using the stocks and index prices from the estimation period ( 255; 46) prior to the earnings announcements. Table 4 shows day to day abnormal returns, as well as CARs from the market model for the sample of Russian companies, and indicates that for the whole analyzed period ( 30; +30), CAR reaches 28 percent. Daily abnormal returns are especially noticeable on the day of the earnings announcement, 9 and 4 days before the event with the 1 percent level of significance, and 2 3 days after, it with the level of significance of 10 percent. Moderate CARs could be seen from 8 to 11 days after the announcement with significance level between 5 and 10%. Figure 1 provides graphical representation of CAR from the market model for Russian companies from Table 3 and 4. It shows that the average market model CAR for Russian companies rises steadily starting from around 20 days prior to the earnings announcement, and 17

24 continues to rise for the 20 days following the announcement. The observed information leakage, as well as post earnings announcement drift, prove the presence of market inefficiency in Russian market which is inconsistent with findings of some researchers on market efficiency of another emerging market such as Latin American equity markets who confirm the presence of market efficiency in those countries (Ojah & Karemera, 2005). On the other hand, Table 5 points out that with 5 percent level of significance, CAR for U.S. companies from market model reaches a maximum of 0.70 percent at the most at the event window, while Table 6 gives more specific results of 0.50 percent for the event day, highlighting that for the 30 days before and after the annual earnings announcement, CAR on average decreases only by 1.53 percent, reaching 1.62 percent. Figure 2 provides graphical support and shows that the average market model CAR for U.S. companies remains near zero before as well as after the earnings announcement, indicating that the market is more efficient in the U.S. than in Russia. There is no noticeable informational leakage before the event. It confirms stock market efficiency in the U.S. which is consistent with the opinion of Chan, Gup and Pan (1997). To summarize, according to the results from the market model, the Russian market responds to the annual earnings announcement with a delay, whereas U.S. stock prices adjust to this information on the same day, giving us powerful evidence to reject the hypothesis that markets become inefficient due to the fast pace of information and lack of time for analyzing and verifying the data. Mean Adjusted Model Results Table 7 represents CARs from the mean adjusted model for Russian companies and points out that the highest CAR of more than 3 percent occurs on the 2 days following the annual 18

25 earnings announcement, and for the period of 5 days prior to 4 days after the event, CARs exceed 1.5 percent per 2 days. In Table 8 we can find average daily returns for Russian companies from the mean adjusted model and see that the highest abnormal returns occur 9 and 4 days before the event, with 99 percent level of confidence and 2, 8 through 11 days after the event with the level of confidence as low as 90 percent. The results from the mean adjusted model are also illustrated in Figure 3, and generally confirm the findings from the market model. The only difference with the market model is the magnitude of CARs. Mean adjusted CAR reaches 36 percent by the end of post event period, whereas the market model CAR ends up at 28 percent. Existing information leakage and the post earnings announcement drift suggest Russian market inefficiency, and, as a result, the higher likelihood that price movements can be predicted. The findings are contrary to the results of Murinde and Poshakwala (2001), who do not reveal any predictable pattern in price movement in Central and Eastern European countries. By contrast, the results from the mean adjusted model for the sample of U.S. companies shown in Table 9 indicate the biggest CAR of 1.59 percent only during the 2 days after the earnings announcement. It appears to be slightly different from the market model findings in Table 5, where the highest CAR is during the event window ( 1;0). So, the market model confirms information leakage, whereas the mean adjusted model finds slight post earnings announcement drift. Table 10 shows the dynamic of CARs from the mean adjusted model for U.S. companies and points out the CAR of 2.27 percent for the whole investigated period. Figure 4 captures the mean adjusted model CAR and confirms these findings, but demonstrates greater CAR volatility from +1 to almost 4 percent with comparison to the market model where CARs ranges from to 1.30 percent. 19

26 Generally speaking, the results illustrate market efficiency in the U.S. There are variations of CARs near zero with the highest abnormal return of 1.06 percent on the next day after the announcement, which is significant at the 5 percent level. Some leakage of information might be seen from 6 to 3 day before the event, as CARs are slightly higher during this time with the level of significance 5 10 percent, but the magnitude is very low. These results are consistent with the findings of Chan, Gup and Pan (1997), who found evidence of the random walk of stock price and market efficiency in the U.S. Overall, the results allow us to reject the hypothesis that increasing speed of information negatively affects market efficiency. U.S. companies CAR has insignificant volatility around zero with the tendency to decline during all the periods, which suggests efficiency, whereas the CAR for the Russian market increases constantly and more sharply, giving more opportunities for arbitragers to take advantage of market inefficiency. According to the market model, by the day of the annual earnings announcement, on average, CAR reach and 0.30 percent for Russian and U.S. markets, whereas for the meanadjusted model it was and 0.99 percent, respectively. It means that the Russian market overreacted on annual earnings announcements significantly more than the U.S. market, which allows the hypothesis of decreasing level of market efficiency with rising speed of information flow to be rejected. This provides further evidence for the efficiency of markets in developed countries suggested by Chan, Gup and Pan (1997), and contradicts the findings of Murinde & Poshakwala s (2001) who noted unpredictable volatility of prices in Central and Eastern Europe. 20

27 Event Study Summary Analysis of different pre event windows reveals that some leakage of information occurs in the Russian stock market on average 9 and 4 days before the announcements, which is confirmed by both the market model and the mean adjusted model with at least a 99 percent level of confidence. For U.S. companies the most noticeable leakage occurs later approximately 1 day before, according to the market model, and in the range of 3 6 days before the announcement date, according to the mean adjusted model, with a level of confidence of percent. So, as we can see, information leakage is common for both samples of companies, with the tendency to be higher and earlier for Russian firms. Important results are derived from analyzing post event windows. Post earningsannouncement price drift is an anomaly first discovered by Ball and Brown (1968), which shows that even after the announcement date, stock prices continue to adjust to the information of earnings announcements slowly, and noticeable CARs take place. For the Russian companies CAR rises by approximately 17 percent for the last 30 days after the announcement date to percent, according to the market model, whereas the U.S. market CAR decline by only 1.92 percent for the same period. The direction of changes in CARs depends on the quality of the news brought by the annual earnings announcements. Estimated cumulative abnormal returns continue to rise for firms with good news and fall down for firms with bad news (Ball & Brown, 1968). For Russian companies, the most statistically significant drift occurs on average 2 3 and 8 11 days after the announcement, confirmed by both the market model and the risk adjusted model. The U.S. market does not show post earnings announcement drift, according to the market model. As for the mean adjusted model, the drift exists for the first 2 days after the event. 21

28 Generally speaking, stock markets in both countries confirm the hypothesis of postearnings announcement drift, but this drift is significantly more noticeable in magnitude and duration in the Russian market. In the U.S. market, abnormal returns are too small to cause market inefficiency. This is quite inconsistent with the findings of Ojah and Karemera (2005), who suggest the efficiency of some emerging markets. Overall, the U.S. market is more price efficient in terms of the magnitude of CARs and the speed with which it adjusts to new information. The Russian market responds to the annual earnings announcement with delay and higher resonance. It allows us to reject the hypothesis of the negative influence of high speed information flow on the ability of prices to adjust it. Cross Sectional Analysis In order to reveal what causes the differences in the levels of market efficiency between two countries, cross sectional analysis is launched and the regression models of CARs are estimated with the following independent variables: market capitalization, debt to equity ratio, operating margin, price to earnings ratio, number of institutional owners and dummy variable for countries. The number of observations for both models is less than the number of companies in two samples due to the availability of information and the fact that CARs for some companies were not calculated by SAS as long as there were some non trading days among the event days or due to the shortage of information available for estimating the parameters. Table 11 shows the results of 5 multiple regressions of CAR from the market model and from the mean adjusted model using different combinations of independent variables which produce the most reasonable results. It should be mentioned that although some multiple regression models illustrate the potential for multicollinearity, there is a very noticeable tendency 22

29 that market capitalization, as well as the dummy variable, are statistically significant in all models. The results from the first cross sectional regression of CAR from the market model by using all 6 variables mentioned above shows that the coefficient on price to earnings ratio is , and statistically significant at the 1 percent level. The inverse relationship between CAR and the price to earnings ratio indicates that market efficiency increases when companies are overpriced. The results from the regression of mean adjusted CARs on all variables on are more significant. The coefficient for price to earnings ratio is with a 1 percent level of significance, and the coefficients for market capitalization and debt to equity ratio are and , respectively, with a 10 percent level of significance. Inverse relationships with a dependent variable mean that the market is more efficient when firms are overpriced, and have a high level of debt and market capitalization. R Squares for the market model and for the mean adjusted model are and percent, respectively, which suggests that the regression of mean adjusted CARs has slightly better fit. The second regression of the market model CAR on the same independent variables, except price to earnings ratio, reveals that the coefficient for market capitalization is and is significant at the 10 percent level. The inverse relationship means that the bigger the companies, the more efficient the market. The same model for mean adjusted CAR shows that the coefficients for market capitalization and debt to equity ratio are and with a 5 and 10 percent level of significance, respectively. A negative sign means that the market is more efficient in the presence of large companies with a high level of leverage. R Square values for the market model CAR and mean adjusted CAR are and 12.23, which means that the regression of mean adjusted CAR produces better fit. 23

30 The third regression of CAR from the market model using operating margin, number of institutional owners, market capitalization and dummy variable for companies shows that the coefficient for market capitalization is again significant at the 10 percent level and reaches Other coefficients are not statistically significant. The same regression model of mean adjusted CAR reveals that the coefficients for market capitalization and the dummy variable for companies are and with 5 and 10 percent levels of significance. Since the coefficient for the dummy variable is positive, we can conclude that market efficiency depends on the country and that there is a tendency for the Russian market to be inefficient. R Square values for this market model CAR and mean adjusted CAR are and percent, so the regression with mean adjusted CAR is marginally better. The next regression model excludes operating margin from the previous, and indicates significance of the coefficients for market capitalization and the dummy variable for companies, which is and , respectively at the 10 percent level. Direct relations between former variable and CAR reveal that market inefficiency with a higher level of CAR is common for the Russian market. The regression of mean adjusted CAR shows that with a 95 percent level of confidence, the coefficients for market capitalization and dummy variable for companies are and , which confirms the result from the regression on the market model CAR. R Square is again higher for mean adjusted regression and reaches percent with only 9.55 percent for the market model regression. For the last cross sectional model, in order to eliminate the effects of multicollinearity, only market capitalization and the dummy variable for companies are regressed on CARs, obtained from both models. The results of the regression on the market model CAR show that with a 90 and 95 percent level of confidence, all coefficients are significant and reach 24

31 and for market capitalization and dummy variable for companies, respectively. The output of the regression on mean adjusted CAR confirm the findings from the previous model and indicates that with a 99 and 95 percent level of confidence, the coefficients for market capitalization and dummy variable for companies are and , respectively. The R Square for the multiple regression on the market model CAR and meanadjusted model CAR are 9.54 and percent. The results suggest that there is insignificant impact on the companies characteristics such as profitability and leverage. Overall, although cross sectional models do not explain a significant part of the variation in CAR, it is still possible to conclude that market efficiency is country related and depends on the size of the market participants. Since the U.S. companies are mostly represented by larger companies and market capitalization is inversely related with CAR, it could partly explain the efficiency of the U.S. market. Significance of the dummy variable for countries and its direct relations with CAR, confirm general inefficiency of Russian stock market. It is important to point out that the data for the regression is mostly from the recession period, which might partly influence the results of the regression, suggesting that this type of analysis could be repeated after the recession. 25

32 CONCLUSION The results derived here provide researchers and investors with valuable information concerning market efficiency and arbitrage opportunities in stock markets, as well as with information for investors to formulate their international diversification strategy. The prime concern of this study is to examine the process by which the stock prices adjust to the information in two different countries the U.S. and Russia under the assumption that rapid information flows could confuse investors, leaving them only a little time for analyzing and verifying the data, which might negatively affects their perception and decision making process. The research indicates U.S. market efficiency and the relative inefficiency of the Russian stock market with statistically significant results. The mean adjusted model shows larger CARs, except for the event period for U.S. companies, and generally confirms the results from the market model. So, at the current time, an increasing speed of information flow is a good sign for market efficiency. These relations could be subject to change during the time, giving wide area for further research. The result is consistent with the previous researches illustrating market efficiency in developed countries (Chan, Gup, & Pan, 1997) and contrary to the findings of some other researchers who confirm a random walk of stock prices and the lack of a predictable pattern in emerging markets (Ojah & Karemera, 2005). Both markets confirm the presence of information leakage starting from 1 and 9 days for the U.S. and Russia, respectively, as well as post earning announcement drift, which lasts up to 11 days for Russian companies and a couple of days for U.S. firms. Generally, the drift appears to be more common for the Russian market where the CAR from the market model reaches 17 26

33 percent 30 days after the announcements and only 1.92 percent for the same period for the sample of the U.S. companies. The main reasons for differences in market efficiency between countries, revealed by the regression analysis based on cross sectional data, are the size of the company and presence of inefficiency in the Russian market. The coefficients from the regression on CARs from the market model and the mean adjusted model show the significance at least at the 10 percent level. Market capitalization is inversely related with CAR, which means that it directly relates to market efficiency. This finding is quite intuitive, as the firms with a higher level of capital, due to their reliability, are considered to be more desirable for investors who are interested in reacting more quickly to any information concerning the companies in order to pursue profit. Also, the average company s size is higher for the U.S. sample compared with the Russian companies and, as a result the U.S. market appears to be more efficient. The significance of the dummy variable for Russian companies suggests general inefficiency of the Russian market, which represents an argument for rejecting the hypothesis that increasing speed of information negatively affects market efficiency in developed countries relative to developing countries. Surprisingly, the profitability of the companies and their leverage do not appear to play very important roles for market efficiency. This might be caused by the fact that the data analyzed here was taken from the recession period after the world economic crisis, when other market anomalies might take place. To conclude, the research results provide grounds with which to reject the hypothesis that high speed of information flow negatively influences the ability of price to adjust all data. There is market efficiency in the U.S. and the presence of inefficiency in the Russian stock market, which could provide its participants with arbitrage opportunities as stock prices do not 27

34 immediately reflect information brought by annual earnings announcements and rather absorb it moderately during the time period. Results derived here confirm the result of some previous research suggesting market efficiency in developed countries (Chan, Gup, & Pan, 1997), but contrary to those that discover market efficiency in some emerging markets (Murinde & Poshakwala, 2001). Modern information technology, as well as the increasing speed of information transfer, provide reliable information for market participants and maintain market efficiency. The level of information technology has great influence on the speed of information transfer. Different measures such as the number of internet users, number of subscriptions for business newspapers or more advanced ways of finding information like Bloomberg terminals can be implemented to compare the level of information industry development. All these can help to continue the investigation of the differences in the level of market efficiency in international scope. 28