Forecasting the direction of stock return movements using Bayesian networks

This work is aimed at assessing- using Bayesian networks the statistical and economic significance of the predictability of the direction of the stock return movements (sign of return). We applied Bayesian networks and a range of structure training algorithms to daily data series for the Dow Jones and Standard & Poor’s indices for the period January 1992-April 2006. The results were compared as reference to the results for logistic regression and support vector machines for classification. According to our tests, some Bayesian networks had a superior predictive capacity to logistic regression and the support vector machines. Moreover, the Bayesian networks help identify, for the indices analyzed, the circumstances in which a positive movement is likely, and therefore, when an investment is likely to be more profitable.