A Comparison Of Linear And Nonlinear Models In Predicting Stock Returns At The Nairobi Stock Exchange

Abstract

Empirical literature shows that stock returns could be nonlinear. However, studies on the nonlinear behavior of stock returns in emerging markets are limited. This study aims at filling this knowledge gap by comparing linear and nonlinear models in predicting stock returns at the NSE. The study compared the Random Walk Model, Moving Average Models, Autoregressive model ARMA models, Autoregressive conditional Heteroskedasticity (ARCH) models. The Nairobi stock Exchange index was used as a proxy for stock prices and hence changes in the NSE index represented stock returns. The sample period consisted of daily observations of the NSE index. The Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) were used to select the best fitting model from each type of models. Then the best fitting model from each type of models were used to predict returns over the sample period of three months. The mean absolute error (MAE) and the Root mean square error (RMSE) were used to select the best model. The results indicate that (ARCH (1) performs better than the other models. Therefore this study concluded that nonlinear models are better than linear models in predicting stock returns on the NSE. Thus stock returns are nonlinear.