Bayesian networks for stock picking

Portfolio managers and investors have to face the perils of the markets and the trade-off between risk and return. A common way to face this environment of uncertainty is to fix some trading rules in order to optimize the returns according to risk profile and investment horizon of each investor. Every market agent would like to predict with accuracy the future price of a stock, unfortunately the constant flow of qualitative information generates noise and distortions and make this unfeasible. In order to mitigate the uncertainty about returns, the investors can use the available historical data referred to a company and integrate efficiently quantitative information with the most recent qualitative data in a systematic way. We propose to combine this information via Bayesian networks. We include in the networks not only the factors indicated as relevant by the fundamental analysis approach, but also market variables. The network is used as a stock picking tool, and to provide a trading recommendation.