Latent Segmentation of Stock Trading Strategies Using Multi-Modal Imitation Learning

摘要

While exchanges and regulators are able to observe and analyze the individual behaviorof financial market participants through access to labeled data, this information is not accessible byother market participants nor by the general public. A key question, then, is whether it is possible tomodel individual market participants’ behaviors through observation of publicly available unlabeledmarket data alone. Several methods have been suggested in the literature using classification methodsbased on summary trading statistics, as well as using inverse reinforcement learning methods toinfer the reward function underlying trader behavior. Our primary contribution is to propose analternative neural network based multi-modal imitation learning model which performs latentsegmentation of stock trading strategies. As a result that the segmentation in the latent space isoptimized according to individual reward functions underlying the order submission behaviorsacross each segment, our results provide interpretable classifications and accurate predictions thatoutperform other methods in major classification indicators as verified on historical orderbook datafrom January 2018 to August 2019 obtained from the Tokyo Stock Exchange. By further analyzing thebehavior of various trader segments, we confirmed that our proposed segments behaves in line withreal-market investor sentiments.