{"paper_id":"2504.19623","title":"Multi-Horizon Echo State Network Prediction of Intraday Stock Returns","authors":["Giovanni Ballarin","Jacopo Capra","Petros Dellaportas"],"paper_date":"2025-04-28","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2504.19623v1","page_url":"https://thequant.space/flowcharts/multi-horizon-echo-state-network-prediction-of-intraday-stoc/","scores":{"empirical_rigor":8,"math_complexity":4.5,"quadrant":"Street Traders"},"proposes":{"status":"available","statement":"The paper proposes a return prediction framework for intraday returns at multiple horizons using Echo State Network models with most parameters randomly drawn and never trained.","quote":"We propose a return prediction framework for intraday returns at multiple horizons based on Echo State Network (ESN) models, wherein a large portion of parameters are drawn at random and never trained.","summary":"Stock return prediction is a problem that has received much attention in the finance literature. In recent years, sophisticated machine learning methods have been shown to perform significantly better than ‘‘classical’’ prediction techniques. One downside of these approaches is that they are often…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"A portfolio rebalance takes place every 10 minutes from 9:30 AM to 3:50 PM and positions are not kept overnight.","type":"execution","stated":"explicit","quote":"Accordingly, we assume that a portfolio rebalance takes place every 10 minutes, from 9:30 AM to 3:50 PM."},{"assumption":"Intraday data are downsampled from 1‑minute to 10‑minute resolution for empirical exercises.","type":"data","stated":"explicit","quote":"Due to the large volume of data associated with 1‑minute frequency, we downsample observations to 10‑minute resolution in our empirical exercises."},{"assumption":"If a stock return or any of its signals are not observed at time t, that observation is discarded.","type":"data","stated":"explicit","quote":"If the i‑th stock return or any of its signals are not observed at time t, we discard it for period t in our empirical exercise."},{"assumption":"The conditional expected return is measurable with respect to the sigma‑algebra generated by past signal realizations.","type":"statistical","stated":"explicit","quote":"We are going to assume throughout that rpiq t,h is measurable with respect to this filtration, and that any observable predictive information associated to past returns is already fully embedded in the signals."},{"assumption":"Effective ESN models require echo state and fading memory properties, enforced via hyperparameter normalization.","type":"model","stated":"explicit","quote":"We underline here that simply drawing A, C, and b from some prescribed distribution will generally not yield an effective ESN model. This is because specific properties, such as the so‑called echo state and fading memory properties"},{"assumption":"Regression coefficients linking signals to returns are common across the stock cross‑section.","type":"model","stated":"explicit","quote":"As with the linear model, here too we assume that returns can be predicted from signals with regression coefficients that are common across the stock cross‑section."}]},"datasets":{"status":"available","items":[{"name":"AlgoSeek high‑frequency intraday stock data","provider":"AlgoSeek","kind":"market-prices","access":"commercial","period":"January 2007 to October 2020","universe":"500 US stocks","quote":"Intraday stock data is sourced from AlgoSeek, a market‑leading platform providing high‑quality high‑frequency trading data"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"intraday-bars"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"n/a","evidence_card":"no","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"The ESN‑based forecasting strategy dominates benchmark models at all intraday horizons, achieving up to a 0.87% reduction in mean‑squared forecasting error (MSFE).","kind":"performance","quote":"Our results show that our ESN‑based forecasting strategy for the cross‑section of returns dominates the benchmarks at all intraday horizons"},{"finding":"Per‑horizon Diebold‑Mariano and model confidence set tests confirm that the ESN gains are statistically significant.","kind":"statistical","quote":"Per‑horizon Diebold‑Mariano and model confidence set (MCS) tests confirm that these gains are all statistically significant."},{"finding":"At the 10‑minute horizon the ESN reduces MSFE by more than 0.87%, with the improvement diminishing for longer horizons and vanishing for end‑of‑day returns.","kind":"predictive-accuracy","quote":"our ESN model is able to achieve a consistent reduction in MSFE of more than 0.87% at the 10‑minute horizon"},{"finding":"Although R² values are negative, ESN forecasts improve relative to linear benchmarks, especially at 10, 30, and 60‑minute horizons.","kind":"predictive-accuracy","quote":"while all values are negative, ESN predictions show improvement, especially at 10, 30, and 60 minute horizons."},{"finding":"The full forecasting pipeline can process one calendar year of data in a handful of minutes, demonstrating numerical efficiency.","kind":"performance","quote":"Finally, our full forecasting pipeline can be run over one calendar year of data in a handful of minutes, highlighting the numerical efficiency of our ESN proposal."}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":45,"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":"2026-11-14T17:40:00Z"},"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/"}