{"paper_id":"2410.17212","title":"Neuroevolution Neural Architecture Search for Evolving RNNs in Stock Return Prediction and Portfolio Trading","authors":["Zimeng Lyu","Amulya Saxena","Rohaan Nadeem","Hao Zhang","Travis Desell"],"paper_date":"2024-10-22","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2410.17212v1","page_url":"https://thequant.space/flowcharts/neuroevolution-neural-architecture-search-for-evolving-rnns/","scores":{"empirical_rigor":5.5,"math_complexity":3,"quadrant":"Street Traders"},"proposes":{"status":"available","statement":"The paper proposes using the Evolutionary eXploration of Augmenting Memory Models (EXAMM) algorithm to evolve recurrent neural networks for stock return prediction and to apply them in a simple daily long‑short portfolio trading strategy.","quote":"we propose the use of the Evolutionary eXploration of Augmenting Memory Models (EXAMM) algorithm to progressively evolve recurrent neural networks (RNNs) for stock return predictions.","summary":"Stock return forecasting is a major component of numerous finance applications. Predicted stock returns can be incorporated into portfolio trading algorithms to make informed buy or sell decisions which can optimize returns. In such portfolio trading applications, the predictive performance of a…","strategy":"EXAMM-evolved RNNs with simple long‑short strategy"},"assumptions":{"status":"available","items":[{"assumption":"All 30 Dow‑Jones stocks are given equal weight in the portfolio.","type":"market","stated":"explicit","quote":"The portfolio used for testing consists of the 30 companies in the Dow‑Jones Index (DJI) with each stock have the same weight."},{"assumption":"RNNs are evolved independently for each stock.","type":"model","stated":"explicit","quote":"RNNs are evolved independently for each stocks and portfolio trading decisions are made based on the predicted stock returns."},{"assumption":"Transaction costs have minimal impact and are omitted from the main experimental results.","type":"execution","stated":"explicit","quote":"Due to this, for our results we do not not incorporate transaction costs in the experiments of the following sections."},{"assumption":"Daily long‑short trading decisions are executed at the closing price without slippage.","type":"execution","stated":"implied","quote":"Transactions are only executed at time [t] if all ... stocks demonstrate a positive predicted return, and all ... stocks demonstrate a negative predicted return."},{"assumption":"CRSP data are accurate and suitable for academic research.","type":"data","stated":"explicit","quote":"CRSP data is known for its high level of accuracy and is highly reliable for academic research and financial modeling."},{"assumption":"The two-year test periods (2022 and 2023) are representative of bear and bull market conditions.","type":"market","stated":"implied","quote":"Results show that using these evolved RNNs and a simple daily long‑short strategy can generate higher returns than both the DJI index and the S&P 500 Index for both 2022 (bear market) and 2023 (bull market)."}]},"datasets":{"status":"available","items":[{"name":"CRSP dataset","provider":"Center for Research in Security Prices","kind":"market-prices","access":"commercial","period":"December 1925 to present","universe":"U.S. stock market (NYSE, AMEX, NASDAQ)","quote":"The Center for Research in Security Prices (CRSP) dataset is a comprehensive collection of data on the U.S. stock market. It contains all stocks on the NYSE, AMEX, and NASDAQ, and covers wide range of market indexes","known_id":"crsp"},{"name":"Dow‑Jones 30‑stock portfolio","provider":null,"kind":"market-prices","access":"public","period":null,"universe":"30 companies in the Dow‑Jones Index","quote":"The portfolio used for testing consists of the 30 companies in the Dow‑Jones Index (DJI) with each stock have the same weight."}],"known":[{"id":"crsp","name":"CRSP","provider":"Center for Research in Security Prices","kind":"market-prices","access":"commercial","quote":"Those predictors are directly obtained from the CRSP dataset or calculated by using variables form CRSP dataset.","mentions":8,"used_in":3}],"sample":{"start":null,"end":null,"market":"Dow‑Jones Industrial Index (30 stocks)","asset_class":"equities","frequency":"daily"},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"unclear","headline_net_of_costs":false,"detail":null,"quote":null,"source":"conflict","rule":"R4b","candidates":{"front_matter":"partial","evidence_card":"no","evidence_card_quoted":true},"conflict_kind":"graded","card_detail":null,"card_quote":"Due to this, for our results we do not not incorporate transaction costs in the experiments of the following sections.","meaning":"the platform classifier and the paper-text extraction disagree, so no value is asserted"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Using EXAMM‑evolved RNNs with a simple daily long‑short strategy yields higher returns than the DJI and S&P 500 indices in both 2022 (bear market) and 2023 (bull market).","kind":"performance","quote":"Results show that using these evolved RNNs and a simple daily long‑short strategy can generate higher returns than both the DJI index and the S&P 500 Index for both 2022 (bear market) and 2023 (bull market)."},{"finding":"Portfolio returns without transaction costs are 39.52% in 2022 and –9.44% in 2023; incorporating two different transaction‑cost calculations changes the returns only marginally.","kind":"performance","quote":"Table 2 shows the portfolio trading return using the simple long strategy using 2022 and 2023 DJI company datasets. The first column [RET(%)] shows the portfolio trading return without incorporating any transaction costs. The second column [RET w/ TC1(%)] shows the portfolio trading return using"},{"finding":"EXAMM outperforms other forecasting models (LSTM, MGU, GRU, AR) in the long‑only strategy for both years.","kind":"predictive-accuracy","quote":"Table 3 shows the portfolio returns using the simple long‑only strategy"}],"reported_performance":{"strategy":"EXAMM-evolved RNNs with simple long‑short strategy","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2302.02269","title":"A Modified CTGAN-Plus-Features Based Method for Optimal Asset Allocation","paper_date":"2023-02-05","authors":["José-Manuel Peña","Fernando Suárez","Omar Larré","Domingo Ramírez","Arturo Cifuentes"],"empirical_rigor":7,"page_url":"https://thequant.space/flowcharts/a-modified-ctgan-plus-features-based-method-for-optimal-asset-allocation/"},"question":"Do transaction costs significantly affect portfolio returns?","this_paper_finding":"Portfolio returns without transaction costs are 39.52% in 2022 and –9.44% in 2023; incorporating two different transaction‑cost calculations changes the returns only marginally.","other_paper_finding":"Trading expenses have no significant impact on returns; net returns remain essentially unchanged after accounting for transaction costs.","other_paper_quote":"A comparison between these returns and those shown in Table 9 proves that trading expenses have no significant impact on returns.","note":"Both studies find only marginal impact: A reports unchanged net returns after costs; B shows returns shift minimally when applying two cost calculations (e.g., 39.52% to ~39% in 2022).","confidence":0.531}],"contradicts":[],"qualifies":[],"compared_with":19,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"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":"2026-11-14T17:57:55Z"},"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/"}