Emergent and spontaneous computation of factor relationships from a large factor set

We propose a systematic factor analysis approach using the Bayesian Network (BN) framework by taking great advantage of the information conveyed in the large amount of financial data, as well as experts' personal insight and additional evidence. First, we build the BN for a large set of macroeconomic and firm-specific financial factors, describing factor interrelationships in the market. As a subgraph of the learned BN structure, a compact set of influential factors for return are extracted which jointly have high influence on the rate of return and low mutual redundancy among themselves. Then in the individual firm analysis, based on the general model of the market, additional information like human expert judgment can be incorporated in the model to conduct the specific and concrete individual firm study. Empirical results show the efficiency and flexibility of our method.