An Integrated Approach Incorporating Nonlinear Dynamics and Machine Learning for Predictive Analytics and Delving Causal Interaction

Abstract

Development of predictive modeling framework for observational data, exhibiting nonlinear and random characteristics, is a challenging task. In this study, a neoteric framework comprising tools of nonlinear dynamics and machine learning has been presented to carry out predictive modeling and assessing causal interrelationships of financial markets. Fractal analysis and recurrent quantification analysis are two components of nonlinear dynamics that have been applied to comprehend the evolutional dynamics of the markets in order to distinguish between a perfect random series and a biased one. Subsequently, three machine learning algorithms, namely random forest, boosting and group method of data handling, have been adopted for forecasting the future figures. Apart from proper identification of nature of the pattern and performing predictive modeling, effort has been made to discover long-rung interactions or co-movements among the said markets through Bayesian belief network as well. We have considered daily data of price of crude oil and natural gas, NIFTY energy index, and US dollar-Rupee rate for empirical analyses. Results justify the usage of presented research framework in effective forecasting and modeling causal influence.