{"paper_id":"2510.09785","title":"The Pitfalls of Continuous Heavy-Tailed Distributions in High-Frequency Data Analysis","authors":["Vladimír Holý"],"paper_date":"2025-10-10","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2510.09785v1","page_url":"https://thequant.space/flowcharts/the-pitfalls-of-continuous-heavy-tailed-distributions-in-hig/","scores":{"empirical_rigor":7,"math_complexity":6.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a simple modification to the maximum likelihood estimation procedure that accounts for the discreteness of high‑frequency integer price changes while still using continuous distributions.","quote":"modification to the maximum likelihood estimation procedure that accounts for the discreteness of data while still allowing the use of continuous distributions.","summary":"We address the challenges of modeling high-frequency integer price changes in financial markets using continuous distributions, particularly the Student’s t-distribution. We demonstrate that traditional GARCH models, which rely on continuous distributions, are ill-suited for high-frequency data due…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Price changes are integer-valued and zero is the most likely change.","type":"data","stated":"explicit","quote":"the discreteness of price changes and the fact that zero is the most likely change."},{"assumption":"Observations are rounded to the nearest integer, so the continuous price change falls within an interval corresponding to the observed integer.","type":"model","stated":"explicit","quote":"modeling the log-likelihood in terms of intervals corresponding to the rounding of continuous price changes to the nearest integer"},{"assumption":"The tick‑by‑tick data are cleaned by removing outliers beyond 10 mean absolute deviations within a rolling window of 201 observations.","type":"data","stated":"explicit","quote":"remove outliers with prices exceeding 10 mean absolute deviations within a rolling window of 201 observations."},{"assumption":"Aggregated price changes at fixed frequencies (0.1, 1, 10, 60, 300 seconds) are representative of the underlying high‑frequency process.","type":"data","stated":"implied","quote":"We aggregate observations at 0.1, 1, 10, 60, and 300‑second frequencies using the last tick method."}]},"datasets":{"status":"available","items":[{"name":"IBM tick‑by‑tick transactions","provider":"Refinitiv Eikon","kind":"market-prices","access":"commercial","period":"2024 (252 trading days)","universe":"IBM stock (NYSE)","quote":"Our data sample for the IBM stock consists of tick‑by‑tick transactions from all 252 trading days in 2024. The data source is Refinitiv Eikon.","known_id":"refinitiv"},{"name":"Additional stocks (MCD, CSCO, MSFT)","provider":"Refinitiv Eikon","kind":"market-prices","access":"commercial","period":null,"universe":"MCD (NYSE), CSCO and MSFT (NASDAQ)","quote":"In Appendix A, we also report results for three additional stocks—the MCD stock traded on the NYSE, and the CSCO and MSFT stocks traded on the NASDAQ.","known_id":"refinitiv"}],"known":[{"id":"refinitiv","name":"Refinitiv / LSEG (Datastream, Eikon, Tick History)","provider":"LSEG","kind":"market-prices","access":"commercial","quote":"The data source is Refinitiv Eikon.","mentions":2,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"tick"},"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":"unclear","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"The interval maximum‑likelihood approach yields a fit comparable to the Skellam distribution for modeling high‑frequency integer price changes.","kind":"statistical","quote":"we estimate our GARCH‑like model using the proposed interval maximum likelihood approach and find that it provides a comparable fit to the Skellam distribution."},{"finding":"Out‑of‑sample log‑likelihoods for the Student’s t, Skellam, and zero‑inflated Skellam models are only slightly worse than in‑sample, indicating stable day‑to‑day performance.","kind":"performance","quote":"Table 2 shows that the out‑of‑sample log‑likelihoods are only slightly worse than their in‑sample counterparts, suggesting stable behavior on a day‑to‑day basis."},{"finding":"The normal distribution at 1‑second frequency fails to capture out‑of‑sample outliers on 56 % of days, resulting in numerically zero likelihood.","kind":"negative","quote":"The normal distribution at 1‑second frequency, however, fails to capture out‑of‑sample outliers on 56 percent of days, resulting in numerically zero likelihood, and is therefore unsuitable for forecasting studies."},{"finding":"Zero‑inflated Skellam is most suitable for 1‑second frequency, while the Student’s t‑distribution performs best for 1‑minute frequency.","kind":"performance","quote":"Overall, the out‑of‑sample results confirm the zero‑inflated Skellam model as the most suitable for the 1‑second frequency, and the Student’s t‑distribution for the 1‑minute frequency."},{"finding":"Despite its heavy tails, the Student’s t‑distribution can produce meaningful results when estimated with the interval approach, comparable to dynamic Skellam models.","kind":"theoretical","quote":"The score‑driven model with dynamic volatility based on the Student’s t‑distribution, estimated using this interval approach, produces meaningful results comparable to dynamic models based on the Skellam distribution."}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2404.11722","title":"Beyond the Bid-Ask: Strategic Insights into Spread Prediction and the Global Mid-Price Phenomenon","paper_date":"2024-04-17","authors":["Yifan He","Abootaleb Shirvani","Barret Shao","Svetlozar Rachev","Frank Fabozzi"],"empirical_rigor":7,"page_url":"https://thequant.space/flowcharts/beyond-the-bid-ask-strategic-insights-into-spread-predictio/"},"question":"Do heavy‑tailed log‑return distributions affect spread prediction and global mid‑price modeling?","this_paper_finding":"The normal distribution at 1‑second frequency fails to capture out‑of‑sample outliers on 56 % of days, resulting in numerically zero likelihood.","other_paper_finding":"Log‑return distributions of both TMOBBAS and GMP exhibit heavy tails and significant deviations from normality.","other_paper_quote":"CIs for each depth. This revealed that the log‑return distributions for both TMOBBAS and GMP have heavy tails, indicative of significant deviations from normality.","note":"B shows normal at 1‑s fails on 56 % of days, supporting A’s finding of heavy‑tailed log‑returns for TMOBBAS and GMP.","confidence":0.609}],"contradicts":[],"qualifies":[],"compared_with":34,"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":true,"code_url":"https://github.com/vladimirholy/continuous-high-frequency-garch","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":"2026-11-14T17:24:59Z"},"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/"}