{"paper_id":"2507.14960","title":"A Comparative Analysis of Statistical and Machine Learning Models for Outlier Detection in Bitcoin Limit Order Books","authors":["Ivan Letteri"],"paper_date":"2025-07-20","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2507.14960v1","page_url":"https://thequant.space/flowcharts/a-comparative-analysis-of-statistical-and-machine-learning-m/","scores":{"empirical_rigor":6,"math_complexity":7,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a unified backtesting environment (AITA-OBS) to evaluate anomaly detection models in Bitcoin limit order books and identifies the Empirical Covariance model as the top-performing approach.","quote":"The primary contributions of this study are threefold: (i) the development of a unified and transparent backtesting environment (AITA-OBS) for evaluating anomaly detection models in LOBs; (ii) a rigorous comparative analysis of thirteen statistical and machine learning models for outlier detection; and (iii) an empiric","summary":"The detection of outliers within cryptocurrency limit order books (LOBs) is of paramount importance for comprehending market dynamics, particularly in highly volatile and nascent regulatory environments. This study conducts a comprehensive comparative analysis of robust statistical methods and…","strategy":"Empirical Covariance (EC)"},"assumptions":{"status":"available","items":[{"assumption":"The data follow a multivariate Gaussian distribution for parametric statistical models.","type":"model","stated":"explicit","quote":"We assume the data follow a multivariate Gaussian distribution or are based on a robust estimate of the covariance."},{"assumption":"Outlier scores can be converted into trading signals using a mean‑reversion strategy.","type":"execution","stated":"explicit","quote":"we provided a transparent and reproducible evaluation of each model’s ability to generate profitable trading signals based on a mean-reversion strategy."},{"assumption":"Feature dimensions are independent for the Histogram‑Based Outlier Score (HBOS) method.","type":"statistical","stated":"explicit","quote":"HBOS score for a data point x is the sum of the inverse logarithms of the estimated densities for each of its feature values, assuming feature independence."},{"assumption":"Transaction costs are approximated by a flat market‑maker fee of 0.08% per trade.","type":"execution","stated":"explicit","quote":"Figure 6 illustrates the estimated total trading fees for each model, assuming a market maker fee of 0.08% per trade."}]},"datasets":{"status":"available","items":[{"name":"Dataset of 26,204 records from a major exchange","provider":null,"kind":"order-book","access":"unknown","period":null,"universe":null,"quote":"An empirical evaluation, conducted via backtesting on a dataset of 26,204 records from a major exchange, demonstrates that the top-performing model, Empirical Covariance (EC), achieves a 6.70% gain"}],"known":[],"sample":{"start":null,"end":null,"market":"Bitcoin limit order books","asset_class":"crypto","frequency":null},"data_resolution":"intraday-bars"},"transaction_costs":{"status":"available","modeled":"partial","headline_net_of_costs":false,"detail":"Costs are estimated using a flat market‑maker fee of 0.08% per trade, shown in a separate cost analysis figure.","quote":"Figure 6 illustrates the estimated total trading fees for each model, assuming a market maker fee of 0.08% per trade.","source":"front_matter","rule":"R3","candidates":{"front_matter":"partial","evidence_card":"partial","evidence_card_quoted":true},"card_detail":"Costs are estimated using a flat market‑maker fee of 0.08% per trade, shown in a separate cost analysis figure.","card_quote":"Figure 6 illustrates the estimated total trading fees for each model, assuming a market maker fee of 0.08% per trade.","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":"trading-pnl","findings":[{"finding":"Empirical Covariance (EC) achieved the highest cumulative profit of $100.47, a 6.70% gain, outperforming the Buy‑and‑Hold benchmark.","kind":"performance","quote":"Table 2 shows that Empirical Covariance (EC) was the most effective statistical model, achieving a cumulative profit of $100.47 (a 6.70% gain)."},{"finding":"CBLOF was the best machine‑learning model with a profit of $75.48 (5.03% gain) but executed nearly double the number of trades of EC.","kind":"performance","quote":"Among the machine learning models (Table 3), CBLOF yielded the highest cumulative profit at $75.48 (a 5.03% gain), though it executed a high number of trades."},{"finding":"OC‑SVM delivered a respectable gain of $43.59 (2.91%) with only 140 trades, indicating higher profit‑to‑trade efficiency.","kind":"performance","quote":"OC‑SVM provided a respectable gain of $43.59 (2.91%) with only 140 trades, suggesting a more efficient strategy when considering transaction costs."},{"finding":"The Buy‑and‑Hold benchmark incurred a loss of $37.06 over the same period.","kind":"performance","quote":"the B&H benchmark, which incurred a loss of -$37.06 over the same period."},{"finding":"High‑frequency models such as K‑Means, CBLOF, and KNN generate substantial trading fees that would erode profits.","kind":"performance","quote":"The high trade frequency of models like K‑Means, CBLOF, and KNN results in substantial costs that would erode profits."}],"reported_performance":{"strategy":"Empirical Covariance (EC)","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":6.7,"max_drawdown_pct":null},"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":40,"citations":{"cited_by_in_corpus":1,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":1},"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":"2026-11-14T17:32:01Z"},"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/"}