Modeling JSE Stock Returns Dynamics: GARCH Versus Stochastic Volatility

Volatility is an essential parameter in risk management applications, and it can affect economic activity and public confidence. It is also a key parameter in several studies examining the link between the stock market, economic growth, and other financial variables. This paper seeks to broaden our understanding of volatility in a Small Island Developing State capital market by conducting a Bayesian model comparison of several volatility models for modeling stock returns on the Jamaica Stock Exchange (JSE). The paper uses a formal Bayesian model comparison methodology to compare seven generalized autoregressive conditional heteroskedastic (GARCH) type models and seven stochastic volatility (SV) type models using the daily JSE index, All Items, from January 03, 2012, December 31, 2019, from the Jamaica Stock Exchange Database. The models include the standard models of GARCH(1,1) and SV with an AR(1) log-volatility process and models with jumps, volatility in mean, leverage effects, heavy-tailed distribution, and moving average innovations. The results reveal that SV models generally fit the data better than their GARCH counterparts. The heavy-tailed distributed innovations and the jump component substantially improve the performance of the basic GARCH and SV models. It also finds that allowing for the moving average component improves the fit of both GARCH and SV models. The result also indicates that volatility feedback is essential. There is also evidence of a significant inverse leverage effect. In total, the SV model with Student-t innovations is the best. The result presented has a potential value for academics, policymakers, and practitioners. For academics, the evidence of the dominance of the SV models over their GARCH counterparts would suggest that spillover studies involving Jamaica would gain from adopting the SV specification. The result also has a potential value for the Bank of Jamaica (BOJ). The BOJ can adopt the SV-t framework rather than a GARCH framework as a tool for gauging volatility in the JSE. Lastly, using the SV model to compute volatility would provide more accurate pricing and risk management results for practitioners, especially global fund managers who plan to include JSE stocks for asset allocation.