A novel sentiment-enhanced portfolio selection approach based on forecasting for exchange rate market

Purpose This paper aims to investigate the crucial role of the sentiment index constructed based on ChatGPT in exchange rate prediction and exchange rate portfolio investment. This study explores whether the inclusion of the sentiment index constructed by ChatGPT can significantly enhance the prediction accuracy of the forecasting models. Additionally, this paper develops a sentiment-enhanced portfolio selection method to research whether market sentiment is useful in improving portfolio returns and reducing investment risk. Design/methodology/approach This paper proposes a sentiment-enhanced portfolio selection method based on predictions. The asset allocation weights of this method are determined by market sentiment and expected returns. Market sentiment is quantified through a sentiment index constructed by ChatGPT, while expected returns are predicted using the Informer-S model. The forecasting performance of the Informer-S model is compared with machine learning and econometric models such as long short-term memory, backpropagation neural network and autoregressive integrated moving average, and its predictive capability is validated through evaluation metrics including mean absolute error, root mean squared error and mean absolute percentage error. Then, a simulated investment experiment is conducted to compare the proposed portfolio method with the Sharpe ratio-maximizing strategy and equal-weighted portfolio. The superiority of the proposed portfolio method was confirmed through evaluation metrics. Findings Firstly, this study finds that the Informer-S model significantly outperforms the baseline model in exchange rate forecasting, highlighting that the sentiment index derived from ChatGPT effectively reduces prediction errors. Secondly, when compared to alternative portfolio selection methods, the proposed sentiment-enhanced portfolio strategy based on predictions not only achieves superior profitability but also significantly mitigates investment risk. Originality/value This paper contributes to the literature in three aspects. First, we utilize ChatGPT to analyze news text and construct market sentiment indices. Empirical studies have shown that the indices constructed based on ChatGPT significantly enhance the predictive performance of forecasting models. Second, we apply the Informer model to forecast exchange rates. The prediction results for multiple exchange rates verify the superior forecasting ability of the Informer model compared to benchmark models. Third, we propose a sentiment-enhanced investment portfolio approach based on prediction results. The simulation trading results demonstrate the profitability of this portfolio method.