Continuous variable based Bayesian network structure learning from financial factors

In this paper, for the discovery the interrelationship of financial factors, we present a two-step accelerated method in learning the structure of Bayesian networks without making parametric assumptions for continuous domains. Our approach divides the high dimensional space into an uniform grid, over which the density can be estimated in an efficient way by using compact support kernels. Local scores are then estimated by the iterative Monte Carlo approximation method with rigorous relative error control. Empirical studies on 15 US financial factors show the efficiency and effectiveness of our method.