{"paper_id":"2312.08927","title":"Limit Order Book Dynamics and Order Size Modelling Using Compound Hawkes Process","authors":["Konark Jain","Nick Firoozye","Jonathan Kochems","Philip Treleaven"],"paper_date":"2023-12-14","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2312.08927v5","page_url":"https://thequant.space/flowcharts/limit-order-book-dynamics-and-order-size-modelling-using-com/","scores":{"empirical_rigor":8,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a novel Compound Hawkes Process model for limit order books that samples order sizes from calibrated distributions and ensures a positive spread.","quote":"We propose a novel methodology of using Compound Hawkes Process for the LOB where each event has an order size sampled from a calibrated distribution. The process is formulated in a novel way such that the spread of the process always remains positive.","summary":"Hawkes Process has been used to model Limit Order Book (LOB) dynamics in several ways in the literature however the focus has been limited to capturing the inter-event times while the order size is usually assumed to be constant. We propose a novel methodology of using Compound Hawkes Process for…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The model assumes nothing about the price process of the equity, letting price emerge as a jump process from order flow.","type":"model","stated":"explicit","quote":"In this formulation of the Hawkes Process, we assume nothing about the Price Process of the equity and rather let it rather emerge as a jump process by modest assumptions about the order flow itself."},{"assumption":"Spreads are mean‑reverting, so in‑spread order intensity depends on the current spread.","type":"market","stated":"explicit","quote":"We formulate the in‑spread order arrivals as a function of the current spread - utilizing the well known fact that spreads are mean‑reverting."},{"assumption":"Order sizes follow calibrated distributions rather than being unit size.","type":"model","stated":"explicit","quote":"We make use of calibrated distributions for sampling the order sizes of the events rather than assuming unit order‑size."},{"assumption":"Intensities are conditioned on time‑of‑day to capture higher trading activity around open and close.","type":"model","stated":"explicit","quote":"We further condition the intensities to be dependent on time‑of‑day in order for the simulator to not be ignorant of the increase in trading intensities around open"},{"assumption":"The top six price levels (best bid/ask and one level each side) sufficiently capture LOB dynamics.","type":"model","stated":"implied","quote":"denotes the queue size, is what we define as the state of the order book."},{"assumption":"In‑spread order intensity can be approximated as piecewise constant over small time bins.","type":"statistical","stated":"explicit","quote":"Now we make another approximation, as in Kirchner 2017, that the grid Δ is small enough so that the intensity is constant over"}]},"datasets":{"status":"available","items":[{"name":"Equity stock LOB on NASDAQ","provider":"NASDAQ exchange","kind":"order-book","access":"unknown","period":null,"universe":"single equity stock","quote":"We showcase the results and quality of fits for an equity stock’s LOB in the NASDAQ exchange and compare them against several baselines."},{"name":"Apple Inc. (Medium‑Tick) LOB data","provider":null,"kind":"order-book","access":"unknown","period":null,"universe":"Apple Inc.","quote":"In addition to Apple Inc. (Medium‑Tick), we also calibrate and simulate the 12D Hawkes Process for Intel (Large‑Tick), Tesla (Small‑Tick) and Amazon (Very Small‑Tick) stocks."},{"name":"Intel (Large‑Tick) LOB data","provider":null,"kind":"order-book","access":"unknown","period":null,"universe":"Intel","quote":"In addition to Apple Inc. (Medium‑Tick), we also calibrate and simulate the 12D Hawkes Process for Intel (Large‑Tick), Tesla (Small‑Tick) and Amazon (Very Small‑Tick) stocks."},{"name":"Tesla (Small‑Tick) LOB data","provider":null,"kind":"order-book","access":"unknown","period":null,"universe":"Tesla","quote":"In addition to Apple Inc. (Medium‑Tick), we also calibrate and simulate the 12D Hawkes Process for Intel (Large‑Tick), Tesla (Small‑Tick) and Amazon (Very Small‑Tick) stocks."},{"name":"Amazon (Very Small‑Tick) LOB data","provider":null,"kind":"order-book","access":"unknown","period":null,"universe":"Amazon","quote":"In addition to Apple Inc. (Medium‑Tick), we also calibrate and simulate the 12D Hawkes Process for Intel (Large‑Tick), Tesla (Small‑Tick) and Amazon (Very Small‑Tick) stocks."}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"l2-book"},"transaction_costs":{"status":"not_extracted","modeled":"unclear","headline_net_of_costs":null,"detail":null,"quote":null,"source":"none","rule":"R0","candidates":{"front_matter":"unclear","evidence_card":"unclear","evidence_card_quoted":false},"meaning":"the classifier was not confident"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"The simulator replicates the empirically observed concave market impact function.","kind":"performance","quote":"Finally, we conduct a market impact study of the simulator and show the empirical observation of a concave market impact function is indeed replicated."},{"finding":"Power‑law kernels fit the LOB data better than exponential kernels.","kind":"statistical","quote":"We find that the power law kernels again fit the data better than exponential kernels."},{"finding":"The model captures right‑tail of inter‑event time distributions but is less accurate in the high‑frequency (left‑tail) domain.","kind":"statistical","quote":"their model is able to capture the right tail of the distribution quite well in all the events we simulate however the model is insufficiently accurate in the high frequency domain (i.e. the left tail)."},{"finding":"For Intel the Hawkes model struggles to capture high‑frequency order events and overestimates spread; for Tesla and Amazon it fits inter‑arrival times better but underestimates spread.","kind":"statistical","quote":"For Intel, we see that the Hawkes Process model is not able to capture the high frequency domain of the order book events"},{"finding":"The proposed formulation maintains a positive spread throughout simulation.","kind":"theoretical","quote":"The process is formulated in a novel way such that the spread of the process always remains positive."}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2503.14814","title":"Modelling High-Frequency Data with Bivariate Hawkes Processes: Power-Law vs. Exponential Kernels","paper_date":"2025-03-19","authors":["Neal Batra"],"empirical_rigor":3,"page_url":"https://thequant.space/flowcharts/modelling-high-frequency-data-with-bivariate-hawkes-processes-power-law-vs/"},"question":"Which kernel, power‑law or exponential, provides a better fit for high‑frequency order‑book data?","this_paper_finding":"Power‑law kernels fit the LOB data better than exponential kernels.","other_paper_finding":"The power‑law kernel provides a significantly better fit to the HFT data than the exponential kernel, capturing sharp peaks in buy intensity and smoother sell trends.","other_paper_quote":"The power-law kernel demonstrates a significantly better fit, as evidenced by the sharp peaks in the buy intensity (blue) and the smoother yet responsive trends in the sell intensity (orange).","note":"Both studies find the power‑law kernel superior: A on LOB data and B on HFT data, showing better fit and capturing intensity patterns.","confidence":0.893},{"paper":{"paper_id":"2503.14814","title":"Modelling High-Frequency Data with Bivariate Hawkes Processes: Power-Law vs. Exponential Kernels","paper_date":"2025-03-19","authors":["Neal Batra"],"empirical_rigor":3,"page_url":"https://thequant.space/flowcharts/modelling-high-frequency-data-with-bivariate-hawkes-processes-power-law-vs/"},"question":"Which kernel (power‑law vs exponential) better models limit order book intensity dynamics?","this_paper_finding":"Power‑law kernels fit the LOB data better than exponential kernels.","other_paper_finding":"The exponential kernel shows limitations, producing fewer and less pronounced peaks in buy intensity and more uniform sell intensity.","other_paper_quote":"The exponential kernel, while providing a simpler and more tractable framework, shows limitations in fully representing the observed data. The buy intensity (blue) under the exponential kernel exhibits fewer and less pronounced peaks, while the sell intensity (yellow) appears more uniform with occasional spikes.","note":"Power‑law kernels fit LOB data better (A); exponential kernels show limitations, producing fewer and less pronounced peaks (B).","confidence":0.74}],"contradicts":[],"qualifies":[],"compared_with":43,"citations":{"cited_by_in_corpus":7,"by_independent_authors":4,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":3,"unclassified":4},"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":"2026-11-14T18:18:02Z"},"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/"}