Market Timing on the Johannesburg Stock Exchange
Posted: 12 Mar 2020
Date Written: February 14, 2020
Whether there are significant relationships between independent variables with future returns, of differing horizons, between 1960 and 2010 and finds, using correlation analysis, several significant relationships. These significant relationships are then included in multifactor forecast models, which are estimated using ordinary least squares regression. The findings from these estimations indicate that there is some, albeit small, portion of the market that is predictable by historic variables. Applying these forecasts to three trading strategies, this study finds that returns in excess of 6% above that of the JSE ALSI are possible.
However, there are several look-ahead biases that impact on this initial result. As the beta coefficients and the specification of the model (based on relational strength between variables) are determined based on the full sample of observations, it is possible that limiting the data such that it reflects only the information available to an investor at each point in time could lead to both differing coefficients and different specifications.
However, even when employing a dynamically updating model to eliminate these biases, there is still evidence of market predictability that can be profitably exploited, with an optimal combination of regression type, trading strategy and return horizon generating slightly less than 3% in excess of the JSE ALSI. Even incorporating transaction costs of 150 basis points of transaction value, it is found that it is possible to generate returns of 0.5% in excess of those of the JSE ALSI.
Keywords: Economics, Econometrics, Finance
JEL Classification: G10, G11
Suggested Citation: Suggested Citation