{"paper_id":"2505.08180","title":"Forecasting Intraday Volume in Equity Markets with Machine Learning","authors":["Mihai Cucuringu","Kang Li","Chao Zhang"],"paper_date":"2025-05-13","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2505.08180v1","page_url":"https://thequant.space/flowcharts/forecasting-intraday-volume-in-equity-markets-with-machine-l/","scores":{"empirical_rigor":7.5,"math_complexity":6,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes clustering assets and training specialized models for each cluster, and introduces a DeepLOB‑style deep neural network for intraday volume forecasting.","quote":"To leverage the commonality in intraday trading volumes, we propose a strategy that first clusters assets and then trains a specialized model or “expert” for each cluster.","summary":"This study focuses on forecasting intraday trading volumes, a crucial component for portfolio implementation, especially in high-frequency (HF) trading environments. Given the current scarcity of flexible methods in this area, we employ a suite of machine learning (ML) models enriched with numerous…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Intraday volume follows a U‑shape or J‑shape pattern within each trading day.","type":"market","stated":"explicit","quote":"Previous works from the extant literature have documented the U‑shape or J‑shape pattern in the intraday trading volume."},{"assumption":"The market trades continuously for the selected stocks over the sample period.","type":"data","stated":"explicit","quote":"We restrict our analysis to a universe of 469 stocks, including only those that traded continuously from the start to the end of our sample period."},{"assumption":"Each trading day is divided into 15‑minute intervals (bins) for analysis.","type":"data","stated":"explicit","quote":"we divide each day into 15‑minute intervals, to which we refer as bins."},{"assumption":"Logarithmic transformation is applied to volume before modeling.","type":"model","stated":"explicit","quote":"the logarithmic transformation that we applied in our experiments."},{"assumption":"VWAP tracking error measured in basis points reflects execution performance.","type":"execution","stated":"explicit","quote":"The VWAP tracking error, measured in basis points (bp), is employed as the error metric."}]},"datasets":{"status":"available","items":[{"name":"LOBSTER","provider":"Huang and Polak (2011)","kind":"order-book","access":"commercial","period":"July 2017 to December 2017","universe":"S&P 500 constituents (469 stocks)","quote":"In our analysis, we employ constituents of the S&P 500 index from July 2017 to December 2017. The data calculated from the LOBSTER dataset proposed by Huang and Polak (2011), which contains the best 10 price levels.","known_id":"lobster"}],"known":[{"id":"lobster","name":"LOBSTER","provider":"LOBSTER","kind":"order-book","access":"commercial","quote":"We first enhance the basic CMEM model with a set of auxiliary predictors, such as the number of trades, derived from high-frequency limit order book data from LOBSTER (Huang and Polak, 2011).","mentions":3,"used_in":2}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"l2-book"},"transaction_costs":{"status":"available","modeled":"partial","headline_net_of_costs":null,"detail":"VWAP tracking error in basis points is used to assess execution risk; a matching engine is employed to examine fill ratios of passive orders.","quote":"The VWAP tracking error, measured in basis points (bp), is employed as the error metric. We employ the definition of tracking error stated in Chen et al. (2016) as follows...","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"partial","evidence_card_quoted":true},"card_detail":"VWAP tracking error in basis points is used to assess execution risk; a matching engine is employed to examine fill ratios of passive orders.","card_quote":"The VWAP tracking error, measured in basis points (bp), is employed as the error metric. We employ the definition of tracking error stated in Chen et al. (2016) as follows...","meaning":"costs appear only as a robustness check, a flat fee without spread or impact, or outside the headline numbers"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"Neural networks with auxiliary predictors achieve the highest out‑of‑sample R² of 0.491, outperforming models using only CMEM components.","kind":"predictive-accuracy","quote":"When the predictors used in the prediction are CMEM and the auxiliary predictors, we found that the R2 is improved to 0.491, and the standard deviation is reduced from 0.065 to 0.059."},{"finding":"Incorporating commonality via a universal asset model (UAM) raises XGB R² from 0.532 to 0.622, showing benefits of cross‑stock information.","kind":"predictive-accuracy","quote":"For the XGB model, the transition from a SAM scheme to a UAM scheme results in an improvement in R2 from 0.532 to 0.622."},{"finding":"Auxiliary predictors improve linear model R² from 0.379 (CMEM only) to 0.405, indicating their strong contribution.","kind":"predictive-accuracy","quote":"the prediction with auxiliary predictors achieved an R2 of 0.405, which outperformed the prediction with solely CMEM components, which achieved 0.379."},{"finding":"VWAP replication using the forecasted volume curve reduces tracking error and improves fill rates compared to baseline CMEM forecasts.","kind":"performance","quote":"higher liquidity is associated with improved fill rates. By forecasting the intraday trading volume curve, one can submit more passive orders during periods of high volume..."},{"finding":"Regularization (LASSO, Ridge) yields higher out‑of‑sample R² than OLS for both CMEM‑only and auxiliary‑enhanced models.","kind":"statistical","quote":"we also observe a higher out‑of‑sample R2 in LASSO and Ridge compared to OLS, suggesting that regularization does further aid performance."}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":37,"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":"2026-11-14T16:36:59Z"},"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/"}