Sammendrag
This research studies the effects from merger and acquisition events on the volatility and distribution of stock returns. The GARCH family of stochastic processes has shown to represent well the stochastic volatility of stock returns. Moreover, using the GARCH family models are standardised residuals can be tested for appropriate model assumptions. Using the Box-Ljung Q-test for autocorrelation in residuals and squared residuals, the Kolmogorov-Smirnov Z-test statistic for the equality of 2 independent standardised residuals, and the Brock, Dechert, and Scheinkman (BSD) test statistic for i.i.d. standardised residuals, a student t-density ARMA(0,1)-GARCH(1,2)-in-Mean model show support for the null hypothesis, that is non-significant test statistic. Therefore, using GARCH models for the variance of event, non-event, trading volume firm portfolios and a market index returns, conditioned on the non-synchronous trading structure in the conditional mean, distribution differences can be studied. The positive curtosis and skewness to the right for the two shortest event periods for the selling firm and the shortest event period for the acquiring firm portfolios, suggest a positive abnormal return in these periods compared to non-event and trading volume portfolios. Moreover, the conditional variance is considerably higher during event periods than during normal periods. The increase can partly be explained by increased trading, especially for the acquire portfolio and confirm my earlier results of trading induced volatility (Solibakke, 1996)in the Norwegian equity market. The constant term in the ARCH/GARCH conditional volatility process specification show a strong increase, indicating non-explained conditional volatility. That is, neither past squared errors nor past conditional variance can explain the high present conditional variance. This result confirms the event induced variance hypothesis. Moreover, the past squared error of the selling firm event firm portfolios form s higher influence on the conditional variance than highly traded and the overall market portfolios. This influence show increasing effects the shorter the event period. The past conditional variance exerts a greater influence over the current conditional variance in the case of lowly traded firm portfolios. For the selling firm event portfolios I find a decreasing GARCH-effects the shorter the event period. That is less emphasis on past conditional variance and more emphasis on past squared errors. The shortest event period for selling firms show a result far below the highly traded firm portfolio and the market index. Moreover, compared to non-event portfolios for the same selling firms, the decrease in GARCH-effects is very strong. The combination of these ARCH- and GARCH-features of the conditional variance suggest that although shocks to the volatility of lowly traded and non-event firm portfolios have less impact than shocks to the volatility of highly traded and event firm portfolios, they are (much) more persistent. Finally, by using a multivariate GARCH approach assuming a multinormal log-likelyhood function, more market dynamics can be modelled. I find the same high kurtosis and the positive skewness for the shortest event periods as for the univariate study. However, the multinormal GARCH model does not report standardised residuals that show non-significant Kolmogorov-Smirnov tests and report more significant BDS test statistic for data dependence. The validity of model is therefore disputed and the results should be interpreted with caution. Owing to these results I suggest two new approaches for event studies in the financial markets in the future.
Vis fullstendig beskrivelse
