Impact of day of the week effect on All Share Price Index (ASPI) and a comparison of forecastability of GARCH and NARX models

Abstract

Forecasting volatility in the stock market is an important research topic that has been immensely reviewed over the years. However, in Sri Lankan context this research topic is studied by only few researchers and they did not incorporate the day of the week effect in their studies. Therefore, this study examines the impact of the day of the week effect on All Share Price Index (ASPI) of Colombo Stock Exchange and compares forecastability of Generalized Autoregressive Conditional Heteroskedastic (GARCH) model and ‘Nonlinear Autoregressive model with exogenous inputs’ (NARX), for the period of 1st of January 2007 to 28th of February 2015. This study uses four other exogenous factors namely gold price, world crude oil price, Sri Lanka Inter Bank Offer Rate (SLIBOR) and the exchange rate in modelling the conditional return and conditional volatility along with the day of the week effect. The results indicate that there exists a day of the week effect in Colombo Stock Exchange. It suggests that first two days of the week has a thin trading behaviour while the last three days of the week shows an active trading behaviour. Therefore, it is more recommended to buy shares on Tuesday and sell shares on Thursday and Friday. Moreover, this study finds that, the difference between the gold price at time t-1 and t-2 shows a negative impact on conditional stock returns at time t. The results also indicate that gold, Sri Lanka Inter Bank Offer Rate (SLIBOR) and exchange rate returns tend to have a positive effect on conditional stock market variance. However a better forecast can be obtained through the GARCH(1,1) model with day of the week effect, three past return terms of ASPI and one lag return of gold price in the mean equation while current gold return in conditional variance equation. Finally, a comparison between the built GARCH and NARX model has been carried out on the basis of the forecasted results. It has found GARCH model is comparatively powerful in predicting the direction whereas both NARX and GARCH models have their own pros and cons.