News sentiment and stock price ratio dynamics: implications for pairs trading strategies

Abstract This study examines the predictive relationship between sentiment extracted from Google News headlines and the log spreads of cointegrated or highly correlated stock pairs, aiming to enhance pairs trading profitability. Pairs trading is an important strategy in quantitative finance due to its market neutral nature, offering opportunities for profit in both bullish and bearish conditions. Despite its popularity, incorporating sentiment analysis into pairs trading strategies remains an underexplored area. Sentiment analysis captures investor emotions and market psychology, which can drive price deviations from fundamental values. Understanding how sentiment influences stock pair log spreads in pairs trading can improve predictive models and enhance trading strategies. Research on sentiment analysis in pairs trading remains limited, and there is little consensus on the temporal effects of sentiment on stock pair log spread movements. Key discussion points include optimal sentiment sampling periods, the lag between sentiment shifts and price responses, and the duration of temporal effects. While prior studies demonstrate the predictive power of LSTM models augmented with sentiment for stock prices, their application to stock pair log spread prediction and pairs trading is largely unexplored. By decoding the emotional drivers behind market movements, sentiment analysis could reveal both the strength and fleeting nature of stock pair correlations, offering a decisive edge for optimizing pairs trading strategies. In this study, six stock pairs from different industries within the S&P500 were selected, and price data and news sentiment were collected via web scraping. Linear regression models assessed sentiment’s explanatory power, followed by the development and testing of novel pairs trading strategies incorporating sentiment-based predictions. An LSTM model was finetuned and trained to predict future stock pair log spread bounds using sentiment and price data. Findings indicate that sentiment’s predictive influence varies by time horizon, peaking within short-term periods (2-4 days). None of the proposed sentiment aware pairs trading strategies consistently outperformed the standard Bollinger bands based approach with respect to average returns, or Sharpe ratio. However, the sentiment-aware “fully-predicted" strategy generated consistently higher Sortino ratios, and lower maximum draw-downs, implying that the inclusion of averaged news sentiment data can decrease risk in pairs trading strategies. Despite these improvements, the inclusion of sentiment data did not consistently improve the predictive accuracy of the LSTM models. These results suggest that news sentiment contributes to price movements, and its integration into pairs trading deserves further optimization.