Forcasting Stock Prices in Nigeria Using Bayesian Vector Autoregression

Abstract

Forecasting the stock market is one of the challenges facing investors and portfolio managers today. These challenges can be overcame by different techniques or analyses in the literature for investment decision-making like, open interest analysis and volatility index on short term basis. However, only a few researchers have employed Bayesian techniques in the forecasts. This study aimed at forecasting the stock prices of five leading banks and the banking sector index of Nigeria using Bayesian Vector Autoregression (BVAR). This research adopted Minnesota priors; Stochastic Search Variable Selection (SSVS) prior; Steady-state with Inverse Wishart prior; steady-state prior with diffuse priors; and Ordinary Least Squares (OLS) procedures. The data were divided into two sets: One set containing 400 datasets for training while the other containing 100 datasets was used for evaluation. Covariance matrices were obtained for these priors as well as the coefficients of the BVAR models. Forecasts for the five priors and their Standard Vector Autoregressions (SVAR) were obtained. The forecast performances for the priors and SVAR were examined using Root Means Square Error (RMSE). The result of RMSE for Minnesota, Minnesota (Normal Inverse Wishart), SSVS, Steady State, Diffuse priors, and OLS obtained were 0.35085, 7.663893, 6.095331, 0.4449004, 11.08892 and 6.174951 respectively. The results showed that the Minnesota prior to BVAR model outperformed the other priors and Ordinary Least Square method of SVAR in predicting stock prices.