Towards Dynamic Structure Changes Detection in Financial Series via Causal Analysis

This is a preliminary paper describing the concepts and principles for a sequential approach towards causal detection in a financial system presented by large-scale data. In particular, we focus on both the regime-switching and causal discovery detection models. This is to address the problem of heterogeneous conditions when analysing nonlinear characteristics from the financial markets. Thus handling the dynamics of multiple regimes in a series and new data to obtain valid answers to causal queries of interest. The availability of large-scale time series data presents new opportunities in knowledge discovery because the insight that can be gained from a causal perspective in a nonlinear system would be tremendous for asset allocation. However, large-scale series are prone to biases, including sampling selection. For decades, the main ways to study nonlinear time series analysis has been isolated to statistical analysis, largely restricted to parametric models. We here present an approach for handling a nonlinear system, infused with a causal solution in a temporal mining task.