Predicting recessions in South Africa : a comparison of the predictive accuracy of linear and non-linear models

M.Com. (Econometrics)This dissertation investigates the ability of different models to predict a recession in South Africa (SA) by choosing a best performing model based on the smallest prediction errors made by the models. One of the purposes of using econometric models is to predict a recession, with the goal to uncover the probability of a recession or real GDP growth rate as accurately as possible. Although linear and non-linear models prediction strength is frequently compared, none of the studies within SA compare the prediction ability of the four models used in this dissertation. The intent of this research is to ascertain the best prediction model for SA so as to advise policy makers on the soundest model to use if there is suspicion that SA could enter a recession in the future due to global and domestic uncertainty. This is done by comparing the prediction ability of the linear ARIMA, VAR and ARMV models’ and non-linear dynamic probit model; thereby contributing toward the standing literature. It is verified which model outperforms the others in predicting future real GDP growth by comparing the Mean-Square-Error (MSE), Mean-Absolute-Error (MAE) and RMSE percentage. The importance of predicting real GDP growth is accentuated so that policy makers are in the position to develop or apply policies that can stimulate growth in the economy, should a recession occur. By adding dynamics to the system, predictions are improved. The linear VAR model outperforms the other linear and non-linear model based on the RMSE, MAE and RMSE percentage