{"paper_id":"2508.19006","title":"Is attention truly all we need? An empirical study of asset pricing in pretrained RNN sparse and global attention models","authors":["Shanyan Lai"],"paper_date":"2025-08-26","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2508.19006v1","page_url":"https://thequant.space/flowcharts/is-attention-truly-all-we-need-an-empirical-study-of-asset/","scores":{"empirical_rigor":9,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes pretrained RNN attention models using additive attention, Luong's three attentions, global self-attention and sliding window sparse attention for asset pricing on 420 large‑cap US stocks.","quote":"This study investigates the pretrained RNN attention models with the mainstream attention mechanisms such as additive attention, Luong's three attentions, global self-attention (Self-att) and sliding window sparse attention (Sparse-att) for the empirical asset pricing research on top 420 large-cap US stocks.","summary":"This study investigates the pretrained RNN attention models with the mainstream attention mechanisms such as additive attention, Luong’s three attentions, global self-attention (Self-att) and sliding window sparse attention (Sparse-att) for the empirical asset pricing research on top 420 large-cap…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Market no‑arbitrage condition holds for evaluating factor models.","type":"market","stated":"explicit","quote":"which satisfies the market no-arbitrage assumption."},{"assumption":"Stocks satisfy a ‘going concern’ condition and are selected from the top 15% by market cap.","type":"data","stated":"explicit","quote":"Thus, this stock selection criterion, ’too-big-to-fail’ and the assumption of ’going concern’, is…"},{"assumption":"Causal masks in attention layers prevent future data leakage.","type":"model","stated":"explicit","quote":"the enforced causal masks in the attention mechanisms address the future data leaking issue ignored by the more advanced attention-based models"},{"assumption":"Trend‑following sign‑signal trading strategy is used for market timing.","type":"execution","stated":"explicit","quote":"This study uses the trend-following sign signal trading strategy as the market timing approach"},{"assumption":"Transaction cost is a flat 50 basis points per trade.","type":"execution","stated":"explicit","quote":"the transaction cost is configured as 50 base points by referring literature [26]"},{"assumption":"Missing values larger than 40% are excluded from the factor set.","type":"data","stated":"explicit","quote":"The 182 factors from Andrew Y.; Zimmermann [5], which excludes the factor with missing values larger than 40% for maintaining a reasonable data quality"}]},"datasets":{"status":"available","items":[{"name":"420 large‑cap US stocks","provider":"WRDS (NASDAQ and NYSE)","kind":"market-prices","access":"commercial","period":null,"universe":"top 15% by market capitalization","quote":"420 large-cap US stocks are selected from the top 15% of stocks ranked by market capitalization (MC), which covers 85% of total market capitalization for these two markets.","known_id":"wrds"},{"name":"182 firm‑characteristic sorted portfolio factors","provider":"Andrew Y.; Zimmermann","kind":"fundamentals","access":"unknown","period":null,"universe":null,"quote":"The 182 factors from Andrew Y.; Zimmermann [5], which excludes the factor with missing values larger than 40% for maintaining a reasonable data quality, are selected as the original input in this study."}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"Flat 50 basis points per trade, applied to the reported portfolio performance.","quote":"the transaction cost is configured as 50 base points by referring literature [26]","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Flat 50 basis points per trade, applied to the reported portfolio performance.","card_quote":"the transaction cost is configured as 50 base points by referring literature [26]","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"In the COVID‑19 period, Model Self‑att and Model Sparse‑att achieve annualized Sortino ratios of 2.0 and 1.80 respectively.","kind":"performance","quote":"they achieve an annualized Sortino ratio of 2.0 and 1.80 respectively in the period with COVID-19."},{"finding":"Model Sparse‑att shows more stable absolute portfolio returns across market‑cap sizes than Model Self‑att.","kind":"performance","quote":"Model Sparse‑att performs more stably than Model Self‑att from the perspective of absolute portfolio returns with respect to the size of stocks' market capitalization."},{"finding":"The vanilla RNN model has the highest out‑of‑sample fitness, followed by Model Self‑att and Model Sparse‑att.","kind":"performance","quote":"The results from model performance evaluation find that the vanilla RNN model has the highest model out-of-sample fitness, followed by Model self att and Model sparse att"},{"finding":"All attention models outperform non‑attention baselines during market shock periods in equal‑weighted portfolios.","kind":"performance","quote":"All attention models show superiority to those models without attention mechanisms during the market shock periods."},{"finding":"In value‑weighted portfolios, annualized returns of all models are significantly higher than in equal‑weighted portfolios, though vanilla RNN returns are lower than its equal‑weighted counterpart.","kind":"performance","quote":"returns of the vanilla RNN model in all periods are considerably lower than those in the equal-weighted portfolio"}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":32,"citations":{"cited_by_in_corpus":1,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":1},"independent_papers":[],"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":"2026-11-14T17:29:07Z"},"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/"}