{"paper_id":"2505.01921","title":"Multilayer Perceptron Neural Network Models in Asset Pricing: An Empirical Study on Large-Cap US Stocks","authors":["Shanyan Lai"],"paper_date":"2025-05-03","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2505.01921v2","page_url":"https://thequant.space/flowcharts/multilayer-perceptron-neural-network-models-in-asset-pricing/","scores":{"empirical_rigor":6,"math_complexity":5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes an innovative dynamic MLP structure for asset pricing factor models, including a novel factor investing strategy based on classic MLP models.","quote":"The main contributions of this chapter appear in proposing the innovative MLP structure for asset pricing factor models, moderating the overfitting issue that existed in GKX2020 models via the firm characteristic-sorted factors, introducing a novel factor investing strategy based on classic MLP models...","summary":"In this study, MLP models with dynamic structure are applied to factor models for asset pricing tasks. Concretely, the MLP pyramid model structure was employed on firm-characteristic-sorted portfolio factors for modelling the large-capital US stocks. It was further developed as a practicable factor…","strategy":"MLP with 2 hidden layers"},"assumptions":{"status":"available","items":[{"assumption":"Going‑concern assumption: stocks must have no missing data in the testing period and less than 50% missing during training.","type":"data","stated":"explicit","quote":"They satisfy the conditions of having no missing data in the testing period and less than 50% of values missing during the training period, so it can meet the requirement of the ‘going concern’"},{"assumption":"Too‑big‑to‑fail assumption: only the top 15% largest market‑capitalization stocks are used.","type":"market","stated":"explicit","quote":"These 420 stocks originally are filtered from the top 15% of the largest market capital (MC) which covers 85% of total MC for the two markets."},{"assumption":"No short‑selling is allowed in the back‑test to avoid systematic risk from extreme cases.","type":"execution","stated":"explicit","quote":"the back test of this chapter excludes the short trading option (negative signal for opening then positive for closing the positions)."},{"assumption":"Transaction costs are a flat 50 basis points per trade.","type":"execution","stated":"explicit","quote":"The transaction cost is taken as an average value from the existing literature [55], which is 50 basis points."},{"assumption":"Dynamic MLP structure varies according to its depth as introduced by Coqueret and Guida.","type":"model","stated":"explicit","quote":"employing the dynamic MLP structure introduced by Coqueret and Guida [19], where the MLP structure varies according to its depth."},{"assumption":"Data length limitation prevents deeper MLPs from improving performance.","type":"statistical","stated":"explicit","quote":"with the restrictions of the data size, the MLP models no longer perform \"deeper, better\""}]},"datasets":{"status":"available","items":[{"name":"420 large‑cap US stock excess returns","provider":"CRSP via WRDS","kind":"market-prices","access":"commercial","period":"January 1957 to December 2021","universe":"US NASDAQ and NYSE large‑cap stocks","quote":"The responses, or labels in ML terminology, used in this research contain 420 large‑capital stock excess returns from January 1957 to December 2021. These stocks come from the US NASDAQ and NYSE markets which are derived from CRSP of Wharton Research Data Services (WRDS).","known_id":"crsp"}],"known":[],"sample":{"start":null,"end":null,"market":"large-cap US stocks","asset_class":"equities","frequency":"daily"},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"available","modeled":"unclear","headline_net_of_costs":false,"detail":null,"quote":null,"source":"conflict","rule":"R4b","candidates":{"front_matter":"no","evidence_card":"yes","evidence_card_quoted":true},"conflict_kind":"polarity","card_detail":"Flat 50 basis points per trade, applied proportionally to returns.","card_quote":"The transaction cost is taken as an average value from the existing literature [55], which is 50 basis points. The transaction cost (TC) takes into the proportional wise, can be written as: TC = returns × 50 basis points","meaning":"the platform classifier and the paper-text extraction disagree, so no value is asserted"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"MLP models with two hidden layers achieve higher flexibility and better out‑of‑sample fitness than deeper models under data‑size constraints.","kind":"performance","quote":"the MLP models no longer perform \"deeper, better\", while the proposed MLP models with two and three hidden layers have higher flexibility to model the factors in this case."},{"finding":"MLP models provide superior downside‑risk control compared to pursuing absolute annual returns.","kind":"performance","quote":"This study also verified the idea of previous works that MLP models for factor investing have more meaning in the downside risk control than in pursuing the absolute annual returns."},{"finding":"All examined models generate significant alpha, indicating profitability from factors, with MLP models showing promising performance after transaction costs, especially during COVID‑19.","kind":"performance","quote":"Overall, MLP models bring promising performance on downside risk control and relatively high annual return after transaction cost for factor investing in all scenarios of our case, especially during the COVID-19 period."},{"finding":"The best‑performed MLP model (2 hidden layers) balances investing profitability and market‑fluctuation risk, outperforming GKX2020’s OOS fitness.","kind":"performance","quote":"Our best‑performed models achieve a great balance on the investing profitability and market fluctuation risks whether the COVID-19 period is applied."},{"finding":"Long‑only portfolios have lower Sharpe and Sortino ratios than two‑way portfolios but are more practical for systematic risk management.","kind":"performance","quote":"Albeit, the long‑only setting portfolio has lower Sharpe ratios and Sortino ratios than the two‑way portfolio, it is more practicable for factor investors from the systematic risk management perspective."}],"reported_performance":{"strategy":"MLP with 2 hidden layers","sharpe":0.2651,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":39.03},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":43,"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":"2026-11-14T17:39:21Z"},"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/"}