Playing with News Context for Algorithmic Trading
Abstract: The application of reinforcement learning for algorithmic trading in the spot market using numerical data is a well-studied problem. However, news data consists of hard-to-quantify information which the investors use to base their trading decisions. Thus factoring in news data for algorithmic trading can improve the trading performance of the RL agent. This paper proposes an RL-based framework that performs algorithmic trading in the futures market by combining news data and price data. We propose two approaches for representing the context of the news data: sentiment-aware approach and context-aware approach. We investigate the effect of these approaches on the trading performance of the RL agent. We further compare the performance of on-policy and off-policy RL algorithms. The models are evaluated by trading in the NIFTY 50 index. The evaluation of the models show that using context-aware approach for representation of news data significantly improves the return (\%) and also reduces the maximum drawdown of the trading model during a trading session.
Paper Type: Long
Research Area: NLP Applications
Research Area Keywords: Language Modeling, NLP Applications
Contribution Types: NLP engineering experiment, Data resources
Languages Studied: English
Submission Number: 5370
Loading