{"paper_id":"2405.18938","title":"HLOB -- Information Persistence and Structure in Limit Order Books","authors":["Antonio Briola","Silvia Bartolucci","Tomaso Aste"],"paper_date":"2024-05-29","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2405.18938v3","page_url":"https://thequant.space/flowcharts/hlob----information-persistence-and-structure-in-limit-order/","scores":{"empirical_rigor":7.5,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a novel large‑scale deep learning model named HLOB for forecasting Limit Order Book mid‑price changes.","quote":"We introduce a novel large-scale deep learning model for Limit Order Book mid-price changes forecasting, and we name it ‘HLOB’.","summary":"We introduce a novel large-scale deep learning model for Limit Order Book mid-price changes forecasting, and we name it `HLOB’. This architecture (i) exploits the information encoded by an Information Filtering Network, namely the Triangulated Maximally Filtered Graph, to unveil deeper and…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The tick size (θ) is $0.01 and the lot size (ψ) is 1 on the NASDAQ exchange.","type":"market","stated":"explicit","quote":"In the NASDAQ exchange, which is the source of the data used in the current research work (see Section 3), θ = $0.01 and ψ = 1."},{"assumption":"Mid‑price changes are considered only when the movement is larger than or equal to the tick size θ.","type":"data","stated":"explicit","quote":"at 3 different horizons (i.e., H∆τ ∈{10, 50, 100}) when such a movement is larger than or equal to θ."},{"assumption":"Order submissions follow the standard LOB mechanics: market orders have higher transaction fees, limit orders have lower fees and no execution guarantee.","type":"execution","stated":"explicit","quote":"A market order expresses the necessity to buy or sell a certain amount of a given asset at the current best available price on the opposite side of the LOB; it is typically subject to higher transaction fees. A limit order"}]},"datasets":{"status":"available","items":[{"name":"LOBSTER order‑book data","provider":"LOBSTER","kind":"order-book","access":"commercial","period":"2017‑2019","universe":"15 NASDAQ stocks (small‑, medium‑, and large‑tick classes)","quote":"For each stock, high-resolution, tick-by-tick LOB data obtained from the LOBSTER provider [31] are employed.","known_id":"lobster"}],"known":[{"id":"lobster","name":"LOBSTER","provider":"LOBSTER","kind":"order-book","access":"commercial","quote":"tick-by-tick LOB data obtained from the LOBSTER provider [31] are employed.","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"l2-book"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R4a","candidates":{"front_matter":"n/a","evidence_card":"partial","evidence_card_quoted":true},"card_detail":"The paper describes that market orders incur higher transaction fees while limit orders incur lower fees, but does not incorporate these costs into the reported performance metrics.","card_quote":"... is typically subject to higher transaction fees. A limit order","meaning":"the paper reports no trading or execution result; the cost the extractor found is a model parameter"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"At horizon H∆τ = 10, HLOB outperforms state‑of‑the‑art alternatives in 73.3% of cases, achieving average F1 scores of 0.42 (small‑tick), 0.41 (medium‑tick) and 0.48 (large‑tick).","kind":"performance","quote":"... The HLOB average F1 score is equal to 0.42 for small‑tick stocks, 0.41 for medium‑tick stocks, and 0.48 for large‑tick stocks."},{"finding":"At horizon H∆τ = 50, HLOB outperforms SOTA alternatives in 60% of cases, a 10% drop compared to H∆τ = 10.","kind":"performance","quote":"Looking at Table 5, we notice that, at H∆τ = 50, HLOB model outperforms SOTA alternatives in the 60% of cases (10% less than what happens at H∆τ = 10)."},{"finding":"The average gain of HLOB over DeepLOB is modest: +0.03 F1 for small‑tick, +0.02 for medium‑tick, and +0.003 for large‑tick stocks.","kind":"predictive-accuracy","quote":"Comparing HLOB performances with DeepLOB ones, we observe that (i) the average gain in F1 score is equal to 0.03 for small‑tick stocks, 0.02 for medium‑tick stocks, and 0.003 for large‑tick stocks."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":15,"citations":{"cited_by_in_corpus":7,"by_independent_authors":6,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":5,"unclassified":2},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":true,"code_url":"https://github.com/FinancialComputingUCL/LOBFrame","note":null},"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":"2026-11-14T18:08:38Z"},"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/"}