A multivariate model for the prediction of stock returns in an emerging market: A comparison of parametric and non-parametric models

This paper compares the forecasting performance of three structural econometric models, namely the non-parametric, ARIMAX and the Kalman filter models, in predicting stock returns in an emerging market economy using South Africa as case study. The proposed models have different functional forms. Each of the functional forms accounts for specific characteristics and properties of stock returns in general and in a small open economy in particular. The findings of the paper indicate the importance of the US stock returns in predicting stock returns in an emerging market economy. Moreover, the results of the Diebold-Mariano statistics show that the Kalman filter and ARIMAX model both outperform the non-parametric model indicating the dominant characteristics of nonlinearity and Markov properties of stock market returns in South Africa