Date: Approved:

In this paper, we propose a new learning algorithm for non-stationary dynamic Bayesian networks. Although a number of effective learning algorithms for non-stationary DBNs have previously been proposed and applied in signal processing and computational biology, those algorithms are based on batch learning and thus cannot be applied to the online time-series data that often arise in financial applica-tions. Therefore, we propose a novel learning algorithm based on a particle filtering approach so that we can apply our algorithm to online time-series data. To evaluate our algorithm, we apply it to a simulated data set and a real-world financial data set. The result on the simulated data set shows that our algorithm estimates the structures of networks accurately and detects structural changes reasonably. Apply-ing our algorithm to the real-world financial data set, we observe several features consistent with what has been reported in previous research, which also suggests that our algorithm is effective. i