Fine-Tuning Large Language Models for Stock Return Prediction Using Newsflow

TL;DR

This work investigates fine-tuning large language models to directly map financial newsflow to forward stock returns, bypassing traditional feature extraction. It compares encoder-only and decoder-only LLMs and two token-representation schemes (bottleneck vs aggregated) within a unified framework, using LoRA-enabled fine-tuning and long-horizon backtesting. Across multiple investment universes, aggregated token representations from decoder-only models often yield stronger portfolio performance, with Mistral showing robust results; the LLM-based signals outperform conventional sentiment scores. The approach demonstrates a practical pathway to leverage news-driven text in quantitative investing, with implications for model selection and representation strategy in finance applications.

Abstract

Large language models (LLMs) and their fine-tuning techniques have demonstrated superior performance in various language understanding and generation tasks. This paper explores fine-tuning LLMs for stock return forecasting with financial newsflow. In quantitative investing, return forecasting is fundamental for subsequent tasks like stock picking, portfolio optimization, etc. We formulate the model to include text representation and forecasting modules. We propose to compare the encoder-only and decoder-only LLMs, considering they generate text representations in distinct ways. The impact of these different representations on forecasting performance remains an open question. Meanwhile, we compare two simple methods of integrating LLMs' token-level representations into the forecasting module. The experiments on real news and investment universes reveal that: (1) aggregated representations from LLMs' token-level embeddings generally produce return predictions that enhance the performance of long-only and long-short portfolios; (2) in the relatively large investment universe, the decoder LLMs-based prediction model leads to stronger portfolios, whereas in the small universes, there are no consistent winners. Among the three LLMs studied (DeBERTa, Mistral, Llama), Mistral performs more robustly across different universes; (3) return predictions derived from LLMs' text representations are a strong signal for portfolio construction, outperforming conventional sentiment scores.

Figures (14)

• Figure 1: Comparison of different workflows of utilizing financial news for stock picking. (a) Conventional feature extraction-and-validation process, e.g., financial sentiments. (b) News-to-return forecasting by fine-tuning LLMs.
• Figure 2: Illustration of the LLM-based return forecasting model for the stock-picking process. Assume an investment universe of 3 stocks denoted by $a, b, c$. Each stock has an associated list of news. Then, given the return forecasts and ranks, stocks can be selected into long-only or long-short portfolios.
• Figure 3: Decile Performance of Bottleneck and Aggregated Representations in the North American Universe (best viewed in color). Top Row: Decile RMSE. Middle Row: Decile Precision. Bottom Row: Decile Return. The up (or down) arrow indicates the higher (or lower) values are desirable.
• Figure 4: Cumulative Return Charts of the Portfolios based on Bottleneck and Aggregated Representation Models in the North American Universe (best viewed in color). Top Row: Long-only Portfolios. Bottom Row: Long-short Portfolios.
• Figure 5: Comparison of Encoder-only and Decoder-only LLMs with the Suited Representations in the North American Universe (best viewed in color).