Trading the Twitter Sentiment with Reinforcement Learning

TL;DR

This paper investigates using Twitter sentiment and reinforcement learning to develop stock trading strategies, showing that sentiment signals can improve prediction and that RL-based strategies outperform traditional machine learning approaches.

Contribution

It introduces a novel approach combining Twitter sentiment analysis with reinforcement learning to optimize stock trading decisions.

Findings

Twitter sentiment significantly predicts stock returns during major events.

Reinforcement learning-based trading strategies outperform machine learning prediction-based strategies.

Sentiment signals are more predictive when driven by company growth expectations.

Abstract

This paper is to explore the possibility to use alternative data and artificial intelligence techniques to trade stocks. The efficacy of the daily Twitter sentiment on predicting the stock return is examined using machine learning methods. Reinforcement learning(Q-learning) is applied to generate the optimal trading policy based on the sentiment signal. The predicting power of the sentiment signal is more significant if the stock price is driven by the expectation of the company growth and when the company has a major event that draws the public attention. The optimal trading strategy based on reinforcement learning outperforms the trading strategy based on the machine learning prediction.