Abstract
This study examines the predictability of stock returns on the Johannesburg Stock Exchange (JSE) by comparing a linear parametric model to the non-parametric and Bayesian models. These two models are compared to the linear model, as they are believed to capture its shortcomings (normality assumption, endogeneity and persistence), particularly in predicting stock returns. The out-of-sample performance of these models is compared using the Predicted Mean Square Error (PMSE), Mean Absolute Error (MAE) and the Diebold-Mariana (DM) test. Predictability using the DM test is examined over a range of forecast horizons. In all three models, the JSE stock return data is regressed against the dividend yield, JIBAR, consumer price inflation, S&P 500 returns and FTSE returns. The variables were chosen based on model selection criteria and empirical evidence of their usefulness in other studies, as there is little theoretical guidance on appropriate variables for forecasting stock returns. Examining stock return predictability in South Africa is important as the majority of the literature on this topic focuses on developed markets (Kadilli 2014, Apall and Gaarde 2011, Masih et al. 2010, Campbell and Thompson 2007, amongst others). Results indicate that model performance in forecasting the JSE returns depends on the performance criteria used. Using the PMSE and MAE, the linear model is found to have better forecasting ability than the nonparametric model. This is because PMSE and MAE are more appropriate to use when model errors follow a normal distribution (Chai and Draxler 2014). However, using the DM test, the nonparametric model shows better forecasting ability, as this test is applicable to non-quadratic loss functions, multi-period forecasts, and forecast errors that are not normally distributed, non-zero mean and contemporaneously correlated. In comparing the linear model to the Bayesian model, the latter out-performs the former, when the PMSE and MAE are considered. However, the DM test shows that both these models have the same forecasting ability.
M.Com. (Financial Economics)