{"paper_id":"2403.09267","title":"Deep Limit Order Book Forecasting","authors":["Antonio Briola","Silvia Bartolucci","Tomaso Aste"],"paper_date":"2024-03-14","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2403.09267v4","page_url":"https://thequant.space/flowcharts/deep-limit-order-book-forecasting/","scores":{"empirical_rigor":7,"math_complexity":6,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper introduces LOBFrame, an open‑source framework for processing large‑scale limit order book data and proposes an operational framework that evaluates the practicality of forecasts by focusing on the probability of correctly executing transactions.","quote":"we introduce ‘LOBFrame’, a novel, open-source software framework designed to offer a standardized protocol for processing large-scale LOB data.","summary":"We exploit cutting-edge deep learning methodologies to explore the predictability of high-frequency Limit Order Book mid-price changes for a heterogeneous set of stocks traded on the NASDAQ exchange. In so doing, we release `LOBFrame’, an open-source code base to efficiently process large-scale…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Backtesting a trading strategy based on model outputs requires live data processing, always‑executed orders, zero market impact, and zero transaction costs.","type":"execution","stated":"explicit","quote":"several assumptions are needed, including but not limited to (i) having the technical and infrastructural potential to record and process live data, produce forecasts and execute them in due time; (ii) being always executed; (iii) having a zero market-impact; (iv) having zero transaction costs."}]},"datasets":{"status":"available","items":[{"name":"High‑quality LOB data for 15 NASDAQ stocks","provider":null,"kind":"order-book","access":"unknown","period":"2017‑2019","universe":"15 liquid NASDAQ stocks","quote":"we first collect high-quality LOB data for a heterogeneous set of 15 liquid stocks traded on the NASDAQ exchange over a period of 3 years (2017 −2019)"}],"known":[{"id":"fi-2010","name":"FI-2010","provider":"Ntakaris et al.","kind":"order-book","access":"public","quote":"FI-2010 consists of 10 trading days LOB data from 5 Finnish companies traded on the NASDAQ Nordic stock market.","mentions":2,"used_in":1},{"id":"lobster","name":"LOBSTER","provider":"LOBSTER","kind":"order-book","access":"commercial","quote":"For each of them, we use high quality, tick-by-tick, LOB data from the LOBSTER provider [77].","mentions":1,"used_in":1}],"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":"Limit orders have lower transaction costs, market orders have higher transaction costs; the paper discusses these costs but does not incorporate them into headline trading results.","quote":"... A market order is typically subject to higher transaction costs since it reduces the liquidity available in the LOB.","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"partial","evidence_card_quoted":true},"card_detail":"Limit orders have lower transaction costs, market orders have higher transaction costs; the paper discusses these costs but does not incorporate them into headline trading results.","card_quote":"... A market order is typically subject to higher transaction costs since it reduces the liquidity available in the LOB.","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":"forecast-accuracy","findings":[{"finding":"Large‑tick stocks are easier to predict and exhibit a stronger prediction signal than small‑ or medium‑tick stocks.","kind":"performance","quote":"large-tick stocks are easier to predict, they are overall characterized by a stronger"},{"finding":"The Matthews Correlation Coefficient (MCC) performance of DeepLOB varies significantly across tick‑size groups and prediction horizons.","kind":"predictive-accuracy","quote":"Our results demonstrate that the chosen deep learning model performs significantly differently depending on the stocks’ tick size."},{"finding":"High traditional ML scores (F1 > 0.7, accuracy > 0.9) do not necessarily translate into actionable trading signals.","kind":"negative","quote":"achieving high scores on these traditional machine learning metrics does not necessarily translate into the generation of actionable trading signals."},{"finding":"Assessing the probability of executing a correct transaction is a more robust evaluation than PnL‑based backtests.","kind":"theoretical","quote":"We argue that this approach is more general than the one based on estimating the PnL of a single strategy on historical data"},{"finding":"Deep learning models achieve high forecasting power but their practical utility depends on low‑latency infrastructure.","kind":"performance","quote":"they are overall characterized by a stronger prediction signal, and their forecast performance is robust across prediction horizons. However,"}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"available","supports":[],"contradicts":[],"qualifies":[],"compared_with":33,"citations":{"cited_by_in_corpus":6,"by_independent_authors":5,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":1,"mentions":3,"unclassified":2},"independent_papers":[{"paper":{"paper_id":"2410.08744","title":"No Tick-Size Too Small: A General Method for Modelling Small Tick Limit Order Books","paper_date":"2024-10-11","authors":["Konark Jain","Jean-François Muzy","Jonathan Kochems","Emmanuel Bacry"],"empirical_rigor":6,"page_url":"https://thequant.space/flowcharts/no-tick-size-too-small-a-general-method-for-modelling-small/"},"how":"discusses_findings","quote":"Similarly, [CITED] , also test the distribution of bid-ask spread, volumes at the top of the LOB and the distance between the top of the LOB and the 10th best top of the LOB against the three categories of relative tick-size. They find that each of these three stylized facts differ remarkably across these three categories."}],"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":"2026-11-14T18:13:12Z"},"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/"}