Causal Inference in Machine Learning for Accurate Prediction in Dynamic Pricing Models

Abstract: In recent years, there has been a noticeable increase in the use of Machine Learning (ML) methods in a wide range of domains, including recommendation systems, text categorisation, picture analysis, and predictive credit scoring. It begins by providing an overview of the conventional asset pricing models and analysing how well they represent the intricacies of financial markets. Furthermore, the Granger causality test—which is inappropriate for non-stationary and non-linear stock factors—is a major determinant of causal link inference. Furthermore, the majority of current research does not take into account the influence of confounding factors or further confirmation of causal linkages. Unfortunately, correlation does not imply causation, because causality—rather than merely correlation—drives the actual world. For example, recommender systems may suggest a battery charger to a user after they purchase a phone, even if the latter may be the source of the former; this kind of causal relationship is irreversible. This was accomplished by incorporating causal diagrams from the structural causal model (SCM) into the stock data analysis. The possible values of closing prices were then predicted using a sliding window method in conjunction with Gated Recurrent Units (GRUs), and confounding factors were controlled using a grouped architecture modelled after the Potential Outcomes Framework (POF). Using the non-linear Granger causality test, the architecture was used to identify causal links between closing price and pertinent parameters. Lastly, comparative experimental findings show that adding causative elements to the prediction model significantly improved the performance and accuracy of closing price forecasts.