Rezaul Kabir Tilburg University, The Netherlands University of Antwerp, Belgium. and. Uri Ben-Zion Technion, Israel

Transcription

1 THE DYNAMICS OF DAILY STOCK RETURN BEHAVIOUR DURING FINANCIAL CRISIS by Rezaul Kabir Tilburg University, The Netherlands University of Antwerp, Belgium and Uri Ben-Zion Technion, Israel Keywords: Financial crisis, Stock market, Crash, Return behaviour JEL classification: G10, G14, G Address for correspondence: Rezaul Kabir, Department of Finance, Tilburg University, P.O. Box 90153, 5000 LE Tilburg, The Netherlands. Tel: /3043 Fax: Kabir and Ben-Zion

2 THE DYNAMICS OF DAILY STOCK RETURN BEHAVIOUR DURING FINANCIAL CRISIS Abstract Stock market co-movements, specially those of smaller economies, become stronger during financial crises. We demonstrate this by analysing firm-specific data from the Dutch stock market during the period of October 1987 stock market crash. We observe that overnight and intraday stock returns of Dutch stocks respond and interact differently. The results indicate that the daily opening prices of shares listed on the Dutch stock market reflect the influence of information from overnight trading on the U.S. stock market.

3 1 1. INTRODUCTION Financial market crises are continuously receiving attention from academics and regulators. Although these crises seem to disappear after a few days, they reappear again and again. The stock market crash of October 19, 1987 (the Black Monday when the Dow Jones Industrial Average plummeted 22.6% is one major example. Each time stock market reaches a new record level, investors are being cautioned on another plunge. Although a crash of that magnitude did not take place afterwards, there were several mini-crashes. The sharp drop of Dow Jones Index on October 13, 1989, reminded investors of the vulnerability of a panic situation. The Dow Index also dropped by 6.4% on August 29, Recently, on October 15, 1999, the Dow suffered the heaviest weekly percentage drop since All these slides in the U.S. market coincide with fall in European and Asia-Pacific stock markets. With today s integrated world financial markets, a drastic change in financial asset prices in one market is quickly transmitted to other markets. This paper thus aims to extend our understanding of financial market linkages by analyzing the dynamics of stock market return behaviour during financial crises. We specifically investigate the influence of new information arrival on the opening and closing stock price movements. Several studies have earlier analysed the behaviour of stock prices and international stock market linkages using series of daily stock returns. In this paper, we show that splitting the daily stock return (calculated from close-to-close prices in subsequent days into the intraday stock return (using open-to-close on the same day and the overnight stock return (between the closing price of one day to the opening price of the following day can add significantly to

4 2 our understanding of stock return dynamics. We verify the importance of intraday and overnight returns by analyzing a large sample of Dutch stocks listed on the Amsterdam Stock Exchange. Following other works on international linkages (see for example, Roll, 1989; Schwert, 1990; Arshanapalli and Doukas, 1993; Ben-Zion and Kabir, 1993, we use the 1987 October crash period for illustration. A number of papers dealing with the October 1987 crash analyse its causes. Roll (1988 finds that institutional factors like computer directed trading, price limits, margin requirements have no significant relationship with the stock market decline on October 19, Bertero and Mayer (1990, on the other hand, observe that stock markets around the world with circuit breakers in operation experience less decline. In another study, Roll (1989 finds empirical support for a speculative bubble burst in October Amihud, Mendelson and Wood (1990 suggest that investors recognition of a less liquid stock market actually contribute to the decline in stock prices. These and other studies indicate that there is no general agreement as to the causes of this market crash. There is another group of studies which analyse stock return behaviour during the crash period. These studies use data from the United States. Blume, Mackinlay and Terker (1989 examine the intraday returns behaviour of stocks listed on the New York Stock Exchange (NYSE on two days of the crash: October 19 and 20, They observe that there are substantial differences in the returns of stocks that are included in the S&P Composite Index and those that are not. The losses incurred by S&P stocks on October 19 are much greater than the losses on non-s&p stocks. They also find that those stocks with the greatest losses in the afternoon of October 19 experience the greatest gains in the morning of October 20.

5 3 Mitchell and Netter (1989 also investigate the returns of S&P stocks on the three days preceding October 19 and find that during those days, the U.S. market declined by almost 11% which was far more than the decline by the rest of the world. The rest of the paper is organised as follows. In section 2 the research methodology is outlined. Section 3 reports on the data used in this study. The empirical results are presented in Section 4. Finally, section 5 concludes the paper. 2. METHODOLOGY The separation of intraday and overnight returns is important in the analysis of international stock market linkage, particularly when two stock markets operate in different time zones such as Europe and the North America. For example, the overnight return in Amsterdam could show a response to trading in New York when the Amsterdam market remains closed. The intraday return in Amsterdam, on the other hand, mostly reflects investor behaviour in Amsterdam before the opening of U.S. stock market later in the day. Since the behaviour of overnight and intraday stock returns may be quite different, the factors determining these behaviours may also be different. Combining these two returns into one daily return series basically leads to specification error and estimation bias. The use of overnight and intraday returns also enables us to make a better comparison between the stocks that are traded locally and the cross-listed stocks. Although the overnight return of Dutch stocks may be influenced by the overall results of New York, the cross-listed stocks are expected to be more influenced by actual trading in New York. The effect of an extra session of trading of

6 4 these cross-listed stocks compared to Amsterdam-listed stocks can be analysed from overnight returns. Therefore, in order to properly analyse stock return behaviour empirically, we formulate the following model: Return i, t = f (Returns i, t - j, Risk i, Size i, DumUS where, Return i, t = stock return of firm i in period t, Risk = systematic risk of the firm, Size = size of the firm, and DumUS = dummy variable with value of one if the stock is also traded on the New York Stock Exchange. The model relates each period s stock returns to previous periods returns as well as some firm characteristics. The time period, however, is not a full day (24 hour as was used in previous studies. Rather half-day returns calculated from close-to-open prices and from opento-close prices are now considered. Normally, in an efficient market lagged returns should have no effect on current return. But, in a period of financial crisis, lagged returns can indicate the existence of over- or under-reaction. Therefore, we use lagged returns as explanatory variables in our model. Besides lagged returns, the model also incorporates several firm characteristics: the systematic risk of the firm, the size of the firm and a dummy variable for cross-listed stocks in the US. In the capital asset pricing model framework, individual stock returns are only related to its systematic risk.

7 5 However, stock returns in a cross-section and in times of financial crisis may also be determined by other firm-specific variables. We use firm size as a proxy for omitted firmspecific factors. Overall, the three firm-specific variables included in the model indicate whether riskier firms, larger firms, and firms listed in the United States respond more or less in a period of crash. 3. DATA The empirical analysis used in this paper is based on stock returns during the October 1987 crash period. We analyse a sample of 114 Dutch stocks, which cover about 80% of market capitalisation of firms listed on the Amsterdam Stock Exchange. There are 10 stocks in the sample with cross-listings in the United States. Daily stock returns are calculated after collecting share price data from Datastream. These returns are used to estimate the systematic risk. Each stock's systematic risk (beta is estimated from the standard market model regression. The opening and closing stock prices during the days of the crash period are handcollected from the official newspaper of the Amsterdam Stock Exchange. In order to perform a comparative analysis, we also collect opening and closing price data of the S&P 500 Stock Index. The size of a firm is measured by the natural logarithm of market value of equity. Data on each firm's market value of equity are collected from Datastream. 4. EMPIRICAL ANALYSIS The descriptive statistics of daily stock returns covering pre- and post-crash period are presented in Table 1. We observe that the largest one-day price decline takes place on

8 6 October 19 (R19 = -8.9%. It is followed by another day of stock price decline (R20 = 5.7%. On both days there is one stock in the sample with an extreme decline of almost 30%. The maximum return obtained by a stock on October 19 is below one percent. Of all the stocks listed on the Amsterdam Stock Exchange, 93% experience a decline on October 19 and 79% experience another decline on the following day. These facts indicate the severity of the crash. It is also interesting to observe that no unusual price change occurs on the trading day before the Black Monday (i.e. October 16. The average stock return of the sample of firms is close to zero. The breakdown of daily stock returns to overnight and intraday returns for stocks listed in the Netherlands and in the United States is presented in Table 2. We observe the following major findings. The previously reported largest one-day decline in Dutch stocks (R19 = -8.9% is almost evenly divided between overnight return (Z19 and intraday return (Y19. But, Dutch stocks experience their heaviest decline only after the major decline in the United States on October 19. The S&P 500 Index suffers a decline of 20.4% (Y19 during the intraday trading period. The next day s opening prices of Dutch stocks show a decline of almost 10% (Z20. Since there is no trading on the Amsterdam Exchange during the overnight period, this huge decline undoubtedly indicates a reaction of U.S. stock market s decline. It is interesting to observe that new information released during the trading hours of October 20 on the Amsterdam Stock Exchange leads to an increase of stock returns (Y20 = 4.1%. We also find that while 89% of the stocks decline on the night between October 19 and 20, 82% of the stocks increase during the intraday trading period of October 20.

9 7 We observe that the average overnight return (Z of stocks traded in Amsterdam is larger in absolute value than the average intraday return (Y. This is in sharp contrast to U.S. stock returns where the average absolute overnight return in S&P 500 index is close to zero whereas that of the intraday return is almost 10%. Miller (1989 also reports that the variance of overnight returns for U.S. stocks is far lower than the variance of intraday returns. Another interesting result from Table 2 is that stocks which are traded on the New York Stock Exchange show a stronger overnight response than stocks traded only in Amsterdam (Z20 of -16.7% vs -9.7%. These stocks also undergo a larger correction during the following intraday trading period (Y20 of 10.7% vs 4.1%. This difference can be attributed to the extra session of trading taking place in New York. Had there been no trading in New York, we could have witnessed a lower response in Amsterdam. This implication is relevant for the use of trading suspension and circuit breaker mechanisms during financial crises. In sum, the results presented in Table 2 are consistent with our assertion that international linkage among the United States, the Far-East and Europe is more pronounced in overnight returns. We now proceed to a statistical analysis of stock returns using the model described in Section 2. The regression results of overnight returns (Z and intraday returns (Y for four days during the crash period are presented in Table 3. We observe that the overnight return of October 19 (Z19 is negatively related to the systematic risk of firms, and the relationship is statistically significant. The variable representing stocks traded on the New York Stock Exchange is not significant. But, the intraday stock return of October 19 (Y19 is significantly related to the variable representing U.S. listings. Dutch stocks which are traded on New York exhibit a

10 8 strong negative effect. It implies that the overnight trading in the U.S. has a significant relationship with the following day s trading in Amsterdam. The regression results also indicate that several return coefficients of Y and Z regressions are not equal in magnitude. The very large overnight decline of October 20 (Z20 is a continuation of the overnight decline of October 19 (Z19. The estimated coefficient is positive (0.783 and is statistically very significant (t-value = On the other hand, after the opening of the Dutch market, the intraday return on October 20 (Y20 is mostly a reversal of the overnight decline (Z20 and also of the intraday and overnight returns of the previous day (Y19 and Z19. Stocks that declined more than the average in previous overnight and intraday periods appear to gain relatively on October 20. All these reversal coefficients are also statistically significant. Table 3 also reports the regression results of overnight and intraday returns of October 21 and 22. Once again, we observe negative regression coefficients suggesting a strong overreaction to previous periods' returns. Most of these return coefficients are also statistically significant. These results clearly indicate that intraday and overnight returns of Dutch stocks exhibit different behaviour. Overall, the results demonstrate that the response of two components of daily returns is not the same, and therefore, an empirical analysis combining half-day returns into daily return is subject to specification error. 5. CONCLUSIONS The decomposition of daily stock returns into intraday and overnight returns enables us to

11 9 observe the dynamics of international stock market linkages in a better way. By analysing the period around a stock market crash and using data from a European stock market, we document that intraday and overnight stock returns behave differently. This is due to the fact that overnight returns reflect the information released during stock market trading in the United States. We also find a strong overreaction effect in several periods during the situation like a stock market crash. This may indicate panic behaviour among stock market investors. Although corrections take place in subsequent trading sessions, in a rational market, such corrections are expected to occur immediately. Our empirical analysis also reveals that firm-size has a significant effect on the day of the crash, and like in the United States, larger firms declined more. It is interesting to note that firms systematic risk was significant only during the overnight and the trading periods of the day of the crash (October 19, but not on other days. The paper thus contributes to the literature on the dynamic linkage between international stock markets by showing the importance of analyzing open-to-close and close-to-open returns separately. The decomposition of daily stock returns into its components is especially relevant when one encounters an unusual event like a financial crisis and employs an investment strategy. An interesting extension of this study will be to examine the dynamics where several countries in different time zones are involved.

12 10 References Amihud, Y., H. Mendelson and R. Wood (1990, Liquidity and the 1987 stock market crash, Journal of Portfolio Management, 16, Arshanapalli, B. and J. Doukas (1993, International stock market linkages: evidence from the pre- and post-october 1987 period, Journal of Banking and Finance, 17, Ben-Zion, U. and R. Kabir (1993, Trading behavior and firm-specific characteristics during the crash of 1987: evidence from the Netherlands, Journal of Multinational Financial Management, 3, Bertero, E. and C. Mayer (1990, Structure and performance: global interdependence of stock markets around the crash of October 1987, European Economic Review, 34, Blume, M., A. Mackinlay and B. Terker (1989, Order imbalances and stock price movements on October 19 and 20, 1987, Journal of Finance, 44, Mitchell, M. and J. Netter (1989, Triggering the 1987 stock market crash, Journal of Financial Economics, 24, Miller, E. (1989, Explaining intra-day and overnight price behavior, Journal of Portfolio Management, 15, Roll, R. (1988, The international crash of October 1987, Financial Analysts Journal, 44, Roll, R. (1989, Price volatility international market links, and their implications for regulatory policies, Journal of Financial Services Research, 3, Schwert, G. (1990, Stock volatility and the crash of '87, Review of Financial Studies, 3,

13 11 Table 1. Descriptive statistics of Dutch stock returns during the October 1987 crash. Variable Mean St. Dev. Minimum Maximum R R R R R R Notes: The number written next to each variable corresponds to date of October Stock return variables R are calculated based on closing prices of subsequent trading days, are based on all sample firms, and are expressed as percentage returns.

14 12 Table 2. Daily, intraday and overnight stock returns of Dutch and U.S. stocks during the October 1987 crash. Period Firms traded only in the Netherlands Firms with cross-listing in the US Total sample (Amsterdam S&P 500 Index (US R R R R Z Z Z Z Y Y Y Y Note: The number written next to each variable corresponds to specific date of October Variables R, Z and Y are percentage returns calculated from close-to-close prices (daily, close-to-open prices (overnight and open-to-close prices (intraday, respectively. Because of time difference, New York intraday trading is observed after the intraday return in Amsterdam but before the overnight return of the next day. New York overnight return is observed only after the Amsterdam intraday return.

15 13 Table 3. Regression results of overnight stock returns (Z and intraday stock returns (Y. Variable Constant R-1 R-2 Z Z-1 Z-2 Y-1 Y-2 BETA SIZE DUMUS F R 2 Z (0.393 (3.339 (1.845 (3.553 (1.538 (0.600 Y (1.626 (0.502 (1.413 (0.276 (4.034 (2.622 (1.999 Z (0.779 (3.517 (0.491 (1.691 (2.980 (0.311 Y (2.453 (7.997 (2.508 (2.597 (0.644 (0.132 (0.051 Z (1.912 (3.868 (3.409 (4.316 (4.780 (0.771 (1.238 (0.760 Y (0.814 (3.853 (3.358 (1.979 (3.494 (1.995 (1.082 (0.731 (0.331 Z (0.313 (1.111 (1.29 (0.703 (1.794 (2.175 (1.362 (0.281 Y (1.324 (3.126 (4.741 (2.482 (2.561 (3.239 (3.146 (2.421 ( Notes: The table reports the results of ordinary least squares regressions. Variables R, Z and Y are stock returns calculated from close-to-close, closeto-open and open-to-close prices, respectively. Absolute t values are mentioned below each coefficient. R 2 is adjusted for degrees of freedom.