| I. The NYSE extended its trading hours on October 1, 1974, and also on September 30, 1985. These events provide ideal opportunities to examine the sources of volume and return variability, an issue that has become especially relevant given the current move towards continuous trading. We find that the extension of trading time in 1974 did not increase trading activity but trading activity increases after the extension in 1985. These results are consistent with our proposed information cancellation hypothesis. Our finding has useful implication for extending trading periods. It suggests that extending trading time to periods when businesses are closed and information arrival is low may not generate significant increases in trading activity. The study also shows that extending trading hours would not increase return variability. In addition, we find that extending trading time reduces transitory noise in opening prices relative to closing prices, and the extension changes intraday return variances which reflects changes in the arrival of private traders. The latter finding is consistent with the price formation hypothesis and the former supports the private information hypothesis.; II. An accurate estimation of Value at Risk (VaR) requires proper modeling of the unconditional kurtosis of the risk factors as well as appropriate apportioning of the modeled kurtosis between stochastic volatility and the distribution of the risk factors. In GARCH models, the division of the unconditional kurtosis between time varying variances and the distribution is determined by the assumed conditional distribution of the errors. We examine the importance of this by applying normal and Student's t-distributions' filtered historical simulations to five major exchange rates. The study shows that the accuracy of VaR estimates of the British Pound can be improved by using appropriate fat-tailed distributions rather than more general stochastic volatility models. This finding suggests that, for some risk factors, the source of the empirical kurtosis is crucial in appropriately modeling the future distribution of the risk factors. In addition, we find that, for the purposes of forecasting VaRs of direct exposures, a more pertinent measure of kurtosis is the number of standard deviations associated with the particular confidence level. |
