Multivariate time-varying parameter modelling for stock markets

Abstract

This paper evaluates the appropriateness of a Linear Market Model (LMM) which allows for systematic covariance (beta) risk. The performance of LMM will be compared against two extensions, a comparison having yet to be undertaken in the literature. The first extension is the Time-varying Linear Market Model (Tv-LMM) which allows for time-varying systematic covariance risk in the form of a mean reverting state space model via the Kalman filter. The second extension is the multivariate Time-varying Linear Market Model (MTv-LMM) which allows for the time-varying systematic covariance risk of country stock market correlation structure via the multivariate KFMR. The comparison between LMM, Tv-LMM and MTv-LMM, is implemented utilising weekly data collected from several developed and emerging markets for the periods; before and after financial crisis in October 2008, and forecasting 2 years forwards. The empirical findings of that process overwhelmingly support the use of the Multivariate Time-varying Linear Market Model (MTv-LMM) when modelling and forecasting stock market returns, especially for developed stock markets.