واکنش شاخص سهام به تغییرات قیمت بزرگ: شواهدی از شاخص سهام عمده آسیایی

We examine the short-term price behavior of ten Asian stock market indexes following large price changes or “shocks”. Under the standard OLS regression, there is stronger support for return continuations particularly following positive and negative price shocks of less than 10% in absolute size. The results under the GJR-GARCH method provide stronger support for market efficiency, especially for large price shocks. For example, for the Hong Kong stock index, negative shocks of less than − 5% but more than − 10% generate a significant one day cumulative abnormal return (CAR) of − 0.754% under the OLS method, but an insignificant CAR of 0.022% under the GJR-GARCH. We find no support for the uncertainty information hypothesis. Furthermore, the CARs following the period after the Asian financial crisis adjust more quickly to price shocks.

Several empirical studies on short-term price changes or “shocks” show that investors overreact to the arrival of new information about stock prices. Specifically, Bremer and Sweeney (1991) find that large negative price changes give rise to overreaction in U.S. stocks. This price behavior is not consistent with the efficient market hypothesis (EMH). Bowman and Iversion (1998) also find support for the overreaction hypothesis following large weekly price changes in New Zealand stocks, whilst Ferri and Chung-Ki (1996) observe a similar pattern following large one day price changes in the S&P500 index. Cox and Peterson (1994) also find support for the overreaction hypothesis in U.S. stocks. However, most of the price reversals disappear after accounting for the bid-ask bounce. Similarly, Atkins and Dyl (1990) find overreaction in U.S. daily stock prices but they conclude in favor of market efficiency after accounting for the bid-ask spread and market liquidity. In contrast, Park (1995) finds support for overreaction in U.S. stocks and shows that the price reversals are not fully explained by the bid-ask bounce (see also, Bremer et al., 1997). Recently, Lasfer et al. (2003) find support for return continuations in stock price indexes which they attribute to momentum behavior in returns. If short-term return continuations are due to investor biases, these biases should give rise to a price drift as information uncertainty increases (see, Zhang, 2006). Brown et al. (1988) argue that their uncertain information hypothesis (UIH), provides a richer description of short-term price behavior. Under the UIH, the arrival of new market-wide information has two simultaneous effects on a security's price: i) it makes the stock price adjust to the content of impending news; and, ii) it creates a transitory systematic risk component.1 The transitory risk component accounts for the condition that investors often set prices before the full ramification of new and impending information is known. Since risk-averse investors require a higher return for a higher level of risk, the asset's price following the arrival of new information will be lower than its expected value, given the perceived risk associated with the new event. The UIH further predicts that after adjusting for the risk element of the news, the abnormal return (AR) will be zero in line with the EMH. An important advantage of the UIH is that it provides a direct test of the behavior of both risk and expected return around large price changes, unlike other approaches. Brown et al. (1988) find support for the UIH in the U.S. stock market. Specifically, they find that both post-event average returns and volatility increase after the arrival of positive and negative unexpected news. Schnusenberg and Madura (2001) also find support for the UIH, although they are not specifically concerned with large price shocks. Using the symmetric generalized autoregressive conditional heteroscedastic (GARCH) estimation method, Ajayi et al. (2006) also find support for the UIH in U.S. stock indexes.2 Overall, prior empirical studies do not provide conclusive evidence as to whether overreaction, underreaction, market efficiency, or uncertainty information best explains short-term price behavior.3 This weakness of prior studies may be partly due to their research designs. For example, we show later that the magnitude of the price shock is an important factor contributing to differences in the results obtained. We also find that the choice of the estimation method leads to differences in the results.4 This empirical study examines the short-term price behavior of ten Asian stock market indexes following large price shocks.5 We are specifically interested in the unexpected price impacts of positive and negative shocks under uncertainty.6 Most prior studies on short-term price behavior employ the standard OLS method to estimate the ARs. To provide a comparison with prior studies, we estimate the ARs using the standard OLS. However, to avoid the econometric problems associated with the standard OLS estimation method, we also estimate the ARs using the: i) symmetric GARCH, and ii) Glosten et al. (1993) threshold (asymmetric) GARCH, hereafter GJR-GARCH, estimation methods. Both the symmetric GARCH and the GJR-GARCH estimation methods allow us to capture the conditional volatility in returns. There are several important aspects of our research design that deserve consideration. Firstly, Savickas (2003) shows that the use of the symmetric GARCH method to capture heteroscedasticity around event and non-event dates leads to substantially higher rejection rates of the false null hypothesis compared to previous approaches. Whilst the symmetric GARCH seeks to capture the effects of conditional volatility on the returns, the GJR-GARCH method seeks to capture both conditional volatility and asymmetry in returns. That is, in addition to the conditional volatility, the GJR-GARCH also controls for the likelihood that negative shocks have a larger impact on returns compared to positive shocks. The use of GARCH-based estimation methods also leads to greater estimation efficiency compared to the standard OLS method. If the variance of returns increases nearer the event period (see, e.g., Brown and Warner, 1980 and Corrado, 1989) due to uncertainty, failure to capture event-induced variances in the returns can lead to invalid inferences. Secondly, our estimates of ARs and volatility are both obtained using a dummy variable approach similar to that of Karafiath (1988). Karafiath's (1988) approach improves estimation efficiency relative to the two-stage residual method commonly employed in prior studies. Indeed, Karafiath (1988) shows that the dummy variable approach provides correct test statistics in a single step. Finally, unlike previous studies, we show that the magnitude of the price shock has an important effect on the result obtained. To summarize our results, we find that the price reaction to shocks depends on both the magnitude and direction of the price shock.7 Whilst this result holds for all estimation methods, the magnitude of the cumulative ARs (CARs) is substantially different under the OLS and GARCH-based methods. Specifically, the CARs under the OLS method tend to be larger than those of the GARCH-based methods, except for very large negative shocks. Under the OLS method, the CARs following positive and negative shocks give support for return continuations particularly when the shocks are less than 10% in absolute value. In some cases, the CARs are significant for up to ten days. Larger shocks, particularly large negative shocks, are followed by overreaction. The GARCH-based results also provide support for return continuations, but mainly following positive shocks of more than 3% but less than 5%. However, the GARCH-based methods provide much stronger support for market efficiency (compared to the OLS method), particularly when the shocks are large. Our results do not support the UIH. Specifically, the post-event volatility of the ARs does not increase after the price shock. Finally, the CARs appear to adjust more quickly to price shocks after the Asian financial crisis.

This empirical study examines the short-term reaction of ten Asian stock indexes following large price shocks. We use both OLS and GARCH-based estimation methods to generate the CARs over a window of ten days after the shocks. Our results based on the symmetric GARCH and GJR-GARCH methods are generally similar; so we focus mainly on the OLS and GJR-GARCH results. There is substantial variation in the effects of shocks across the ten indexes indicating that the price reaction varies by country. This feature can be due to differences in the way investors process unexpected information in those countries, but we leave this aspect for further research. There are also differences in the results due to the estimation method. Specifically, for the full period, the OLS method provides stronger support for return continuations, particularly following positive shocks within + 5% < st+ < + 10% and negative shocks within − 5% ≥ st− > − 10%. There is some support for market efficiency, but this support is not as strong compared with return continuations. There is also mixed support for market efficiency and overreaction following positive shocks of st+ ≥ + 10%. However, overreaction is more pronounced following negative shocks of st− ≤ − 10%. The GJR-GARCH results contrast substantially with those of the OLS for the full period. Return continuations dominate following positive shocks within the range of + 3% < st+ < + 5% whilst market efficiency dominates following negative shocks within − 3% > st− > − 5%. Larger positive and negative trigger values largely provide support market efficiency. This finding is strongest following shocks of st+ ≥ + 10% and st− ≤ − 10% and here, all support for return continuations and the overreaction hypothesis completely disappears. The volatility estimates are typically negative and larger shocks give rise to larger volatility estimates. Our results do not support the UIH. Furthermore, the markets appear more efficiency over the post-Asian financial crisis. We still do not find support for the UIH over the pre- and post-Asian financial crisis. Since we find that return continuations dominate at small trigger values whilst overreaction and/or EMH is more pronounced at larger trigger values, it appears that the results obtained depend on the magnitude of the shock. Specifically, prior studies that support overreaction or market efficiency tend to employ relatively large trigger values (see e.g., Brown and Harlow, 1988).15 In contrast, Lasfer et al. (2003) find support for return continuations using a trigger value within a small range of the standard deviation. If investor over-confidence and self-attribution can explain price overreaction and momentum around public events (see Daniel et al., 1998), our results demonstrate a related reaction in the very short-term for unexpected events. Furthermore, the choice of the estimation method leads to different results about price reaction. The greater support for market efficiency under the GARCH-based methods is in line with Fama's (1998) view that the evidence on short- to long-term market anomalies can be model dependent.16