Let the Machine Confirm Theories: A Naive Transfer Learning-Based Retroductive Analysis on the Price Prediction for Crude Oil

Abstract

Due to the time-variance of human behavior, social science theory is constantly testified in various contexts. However, conventional linear methods often neglect the goodness of fit from period to period. In contrast to the conventional one, machine learning techniques are characterized by efficient generalization. Among others, transfer learning integrates structured theoretical knowledge and empirical prediction. Based on the transfer learning method, this paper introduces a novel retroduction framework named machine confirming for testifying theories. Empirically, the Schwartz one-factor model is employed to retroduce the futures price in the context of China’s crude oil market. The results indicate that the theory is validated through our framework and statistically correlated but theoretically irrelevant features are excluded.

Keywords: Retroduction, Machine Confirming, Transfer Learning, Price Prediction