{"paper_id":"2407.18103","title":"Fine-Tuning Large Language Models for Stock Return Prediction Using Newsflow","authors":["Tian Guo","Emmanuel Hauptmann"],"paper_date":"2024-07-25","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2407.18103v2","page_url":"https://thequant.space/flowcharts/fine-tuning-large-language-models-for-stock-return-predictio/","scores":{"empirical_rigor":6,"math_complexity":5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes an LLM-based return prediction model that includes text representation and forecasting modules.","quote":"We design an LLM-based return prediction model comprising the text representation and the forecasting modules.","summary":"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…","strategy":"LLM-based return prediction with aggregated/bottleneck representations"},"assumptions":{"status":"available","items":[{"assumption":"An investment universe of stocks is defined and used for stock‑picking.","type":"market","stated":"explicit","quote":"Assume an investment universe consisting of a set of stocks denoted by U = {s}S s=1, where s represents the stock index."},{"assumption":"Return forecasts are n‑step forward returns of each stock.","type":"model","stated":"explicit","quote":"let rs,t+n ∈R be the n-step forward return of stock s w.r.t. timestep t."},{"assumption":"Portfolios are formed by selecting stocks in the top (9th) and bottom (0th) deciles of predicted returns and are equally weighted.","type":"execution","stated":"explicit","quote":"The long-only portfolio is built by taking the stocks with the return predictions falling in the top (9th) decile of prediction rankings. The long-short portfolios take the stocks in the top (9th) and bottom (0th) deciles. The stocks in all portfolios are equally weighted."},{"assumption":"Portfolios are rebalanced monthly.","type":"execution","stated":"explicit","quote":"We perform backtesting to evaluate the portfolios in monthly rebalancing."},{"assumption":"Aggregating token‑level embeddings (e.g., by averaging) provides a valid representation for forecasting.","type":"model","stated":"explicit","quote":"we present a simple alternative to bottleneck representation, i.e., allowing the forecasting module to aggregate the representations of all tokens. This can be done using various methods like averaging..."},{"assumption":"Appending an EOS token to sequences is beneficial for encoder‑only LLMs and is applied to both encoder‑only and decoder‑only models for fair comparison.","type":"model","stated":"explicit","quote":"In experiments, we observed that appending the EOS token is more helpful for encoder-only LLMs. For a comparison on the same ground, we append EOS tokens for both encoder-only and decoder-only LLMs..."}]},"datasets":{"status":"available","items":[{"name":"Company-level financial newsflow data","provider":"financial data vendor","kind":"text-news","access":"commercial","period":"2003-2019","universe":null,"quote":"We use company-level financial newsflow data from 2003 to 2019 provided by a financial data vendor."},{"name":"North American, European, Emerging market investment universes","provider":null,"kind":"market-prices","access":"unknown","period":null,"universe":"NA, EU, EM markets","quote":"we have two investment universe datasets of the North American (NA), European (EU), and Emerging (EM) markets, which consist of dates, stock identifiers, and the true monthly forward returns of corresponding stocks and dates."}],"known":[],"sample":{"start":"2015","end":"2020","market":"North American stocks","asset_class":"equities","frequency":"monthly"},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":false,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"no","evidence_card_quoted":false},"meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Aggregated token‑level representations generally improve the performance of long‑only and long‑short portfolios.","kind":"performance","quote":"aggregated representations from LLMs’ token-level embeddings generally produce the return predictions that enhance the performance of long-only and long-short portfolios"},{"finding":"Decoder‑only LLMs yield stronger portfolios in relatively large investment universes, while no consistent winner appears in small universes.","kind":"performance","quote":"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"},{"finding":"Mistral shows robust performance across different universes compared with DeBERTa and Llama.","kind":"performance","quote":"Among the three LLMs studied (DeBERTa, Mistral, Llama), Mistral performs more robustly across different universes"},{"finding":"Return predictions from LLM text representations outperform conventional sentiment scores for portfolio construction.","kind":"performance","quote":"return predictions derived from LLMs’ text representations are a strong signal for portfolio construction, outperforming conventional sentiment scores"},{"finding":"Aggregated representations produce higher decile returns even when decile RMSE is higher, indicating return predictions can be useful despite lower predictive accuracy.","kind":"statistical","quote":"... the aggregated representation model has a higher decile return, but a higher RMSE, in the 9th decile"}],"reported_performance":{"strategy":"LLM-based return prediction with aggregated/bottleneck representations","sharpe":1.49,"sharpe_benchmark":0.68,"annual_return_pct":27,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"available","supports":[],"contradicts":[],"qualifies":[],"compared_with":38,"citations":{"cited_by_in_corpus":6,"by_independent_authors":5,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":1,"mentions":4,"unclassified":1},"independent_papers":[{"paper":{"paper_id":"2502.17011","title":"Predicting Liquidity-Aware Bond Yields using Causal GANs and Deep Reinforcement Learning with LLM Evaluation","paper_date":"2025-02-24","authors":["Jaskaran Singh Walia","Aarush Sinha","Naman Saraswat","Srinitish Srinivasan","Srihari Unnikrishnan"],"empirical_rigor":8,"page_url":"https://thequant.space/flowcharts/predicting-liquidity-aware-bond-yields-using-causal-gans-and/"},"how":"discusses_findings","quote":"showing that token-level representations extracted from LLMs can serve as strong predictive signals for both long-only and long-short portfolio strategies."}],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T16:36:07Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}