Deep Q-Network Interpretability: Applications to ETF Trading

Abstract

We present an interpretability infrastructure for Reinforcement Learning (RL) based trading
strategies. For all audiences to be able to answer the question of 'how does the algorithm
work?', we provide a visual and user-friendly approach, in contrast to a more quantitative
approach. This allows not only a technical audience to consume insights derived from an
RL-based trading approach. In this application, we introduce a three module approach in
understanding value-based RL, specifically Deep Q-Learning. We demonstrate this
infrastructure and possible derived outcomes of using this infrastructure when applied to
trading a market ETF in a given time interval.