Forecasting the time varying-beta of nse-20 share companies: Bi-variate garch (1, 1) model vs kalman filter method

Abstract

This research paper forecasts the time -varying daily beta of ten stocks listed in the Nairobi Securities Exchange 20- Share Index by use of a Bivariate GARCH (1, 1) model and the Kalman filter method. A comparison of the forecasting ability of the GARCH model and the Kalman filter method is made. Forecast errors based on the retUI11 forecasts are used to evaluate the outof- sample forecasting ability of both the GARCH model and the Kalman method. Two measures of error are used: MAE and MSE. The results are inconclusive, based on MSE the Kalman method is superior while based on MAE, the Bivariate GARCH (1, 1) method appears to provide more accurate forecasts of the time-varying beta .