Empirical Evaluation of Deadline-Resolved Information Leakage on Documented Polymarket Insider Cases
Maksym Nechepurenko · 2026
Read Empirical Evaluation of Deadline-Resolved Information Leakage on Documented Polymarket Insider Cases by Maksym Nechepurenko (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-02286.txt.
Introduction
Maksym Nechepurenko evaluates a deadline-aware version of the Information Leakage Score on documented insider cases from Polymarket, a prediction market. The score aims to detect trading that front-loads information before a public event. Deadline-resolved contracts, which settle by a set date, are the dominant form among publicly documented insider cases on the platform.
The study anchors on a 2026 United States-Iran conflict cluster, the largest documented deadline cluster in the inventory used. Parts cover hazard-rate fitting by category, a single-case score computation, cross-market wallet analysis, and method refinements. Military-geopolitics markets fit an exponential hazard well, with a half-life near three days. A corporate-disclosure fit is preliminary, and the regulatory-decision category is rejected as bimodal.
Catalog record
| CREATOR | Maksym Nechepurenko |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.02286
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Empirical Evaluation of Deadline-Resolved Information Leakage on Documented Polymarket Insider Cases Maksym Nechepurenko Abstract This paper reports an end-to-end empirical evaluation of the deadline-Information Leakage Score (ILS dl {}^{\text{dl}} ) extension introduced in the companion methodology paper ( 2 ) . The deadline-ILS extends the original ILS to deadline-resolved prediction-market contracts—the dominant structural form of publicly documented insider trading on Polymarket. We anchor the evaluation in the 2026 U.S.–Iran conflict cluster of the ForesightFlow Insider Cases (FFIC) inventory, the largest documented deadline cluster in the public-reporting record. The evaluation has four parts: per-category exponential-hazard rate estimation for the time-to-event distribution; a single-case ILS dl {}^{\text{dl}} computation on the cleanest applicable FFIC market; cross-market wallet analysis across two cluster contracts; and the methodological refinements that the evaluation surfaces. The hazard-rate estimation produces an adequate exponential fit for military-geopolitics markets (KS p = 0.426 p=0.426 , half-life 2.9 2.9 days, n = 18 n=18 ) and a preliminary fit for corporate-disclosure markets ( n = 5 n=5 ). The regulatory-decision category is rejected as bimodal ( p = 0.023 p=0.023 ) and requires sub-categorization. On the largest applicable FFIC contract (“US forces enter Iran by April 30,” $269M cumulative volume), the article-derived T event T_{\text{event}} anchor yields ILS dl = + 0.113 {}^{\text{dl}}=+0.113 versus a resolution-anchored proxy value of − 0.331 -0.331 : a 0.444 0.444 shift in magnitude on opposite sides of zero, demonstrating that the extension distinguishes signal from proxy artefact. The pre-event drift is mild rather than concentrated, and short-window variants (30-min, 2-h) are exactly zero, ruling out a last-minute informed spike. Cross-market wallet analysis identifies 332 332 wallets active in both major Iran-cluster markets, but the available trade history covers only the resolution-settlement window; the cross-market signal is therefore settlement arbitrage rather than pre-event coordination, and converting this diagnostic into a coordination-signal diagnostic requires continuous per-trade collection from T open T_{\text{open}} . We treat the present evaluation as a methodological proof of concept supported by a single illustrative case, not a population-level evaluation. Five concrete refinements bound any future scaling: a positive- τ \tau requirement, regulatory sub-categorization, corporate-disclosure sample expansion, continuous CLOB price collection, and continuous per-trade collection from T open T_{\text{open}} . Both data resources used here—the FFIC inventory and the resolution-typology classification of the 911,237-market Polymarket corpus—are released as separate datasets at https://github.com/ForesightFlow/datasets . Keywords: prediction markets, informed trading, deadline contracts, hazard-rate estimation, Polymarket, blockchain forensics, market microstructure, empirical evaluation. JEL Classification: D82 (Asymmetric and Private Information), G14 (Information and Market Efficiency), G18 (Government Policy and Regulation), C58 (Financial Econometrics). 1 Introduction Decentralized prediction markets such as Polymarket have, since 2024, accumulated a substantial public-reporting record of suspected informed trading. 1 and 3 jointly document hundreds of millions of dollars in anomalous profits across this period, with case studies spanning military operations, corporate proprietary disclosures, and regulatory decisions. The companion methodology paper ( 2 ) develops an information-theoretic framework—the Information Leakage Score (ILS) and its deadline-resolved extension ILS {}^{\text{dl}} —for quantifying informed flow on these markets in a way that is interpretable, replicable, and connected to the proper-scoring-rule literature via the Murphy decomposition of the Brier score. The methodology paper specifies the framework end-to-end: definition, scope conditions, resolution-typology classification, the FFIC validation inventory, and the deadline-ILS extension that addresses the deadline-contract structure of the documented cases. It does not, however, carry the framework through to a quantitative empirical claim on a documented case. The present paper does so. We implement the extension end-to-end, fit per-category hazard parameters on the time-to-event distribution, and apply the pipeline to the eighteen substantive markets in the 2026 U.S.–Iran conflict cluster of the FFIC inventory. We report a single-case ILS {}^{\text{dl}} computation in full, a cross-market wallet analysis on the two cluster markets with available price coverage, and the methodological refinements that the evaluation surfaces. The empirical sample is small by design. Deadline-ILS is a methodologically demanding score: it requires recovered article-derived event timestamps, full CLOB price coverage from market opening, scope-condition compliance, and positive lead time between market opening and the underlying event. Of the eighteen Iran-cluster markets at which we attempt T_{\text{event}} recovery, only one satisfies all four requirements end-to-end. We treat this case in full as a methodological demonstration, do not generalize from it to a population-level claim, and identify the infrastructure constraints that gate any future scaling. The single most important finding for the methodology developed in the companion paper is that the article-derived anchor materially changes the substantive reading of the score: ILS {}^{\text{dl}}=+0.113 at the article-derived T_{\text{event}} versus -0.331 at the resolution-anchored proxy ( T_{\text{resolve}}-1\,\text{h} ), with the two values on opposite sides of zero and differing by 0.444 in magnitude. The proxy fails because by the proxy lookup time the market price has already collapsed on participants’ belief that the event will not occur; the lookup window captures the collapse rather than the pre-event period. This is the empirical content of the proxy-quality problem identified abstractly in 2 . 1.1 Outline Section 2 provides a brief notational and methodological recap; readers familiar with the companion paper may skip directly to Section 3.1 . Section 3.1 reports the per-category hazard-rate estimation. Section 3.2 reports the single-case ILS {}^{\text{dl}} computation on the largest applicable Iran-cluster market. Section 3.3 reports the cross-market wallet analysis. Section 3.4 states the five methodological refinements that the evaluation surfaces and that bound any future application of the extension at scale. Section 3.5 concludes. 2 Methodological Recap We refer the reader to 2 for the full development of the framework. Three definitions are needed for the empirical evaluation that follows. Information Leakage Score (ILS). For a resolved binary market M with first trade at T_{\mathrm{open}} , public news event at T_{\mathrm{news}} , formal resolution at T_{\mathrm{resolve}} , and binary resolution outcome p_{T_{\mathrm{resolve}}}\in\{0,1\} , \Delta_{\text{pre}}\;=\;p(T_{\mathrm{news}})-p(T_{\mathrm{open}}),\qquad\Delta_{\text{total}}\;=\;p_{T_{\mathrm{resolve}}}-p(T_{\mathrm{open}}),\qquad\mathrm{ILS}(M)\;=\;\frac{\Delta_{\text{pre}}}{\Delta_{\text{total}}}. The score is interpretable only when |\Delta_{\text{total}}|\geq\varepsilon for a threshold \varepsilon>0 (we use \varepsilon=0.05 ); when the market opens with |p(T_{\mathrm{open}})-0.5|\leq 0.4 (the edge-effect scope condition); and when ILS is robust to the choice of T_{\mathrm{news}} anchor offset (the anchor-sensitivity scope condition). The companion paper develops a Murphy-decomposition reading of ILS as the share of the Brier-score resolution component accumulated before T_{\mathrm{news}} . Resolution typology. Polymarket markets are classified into three resolution types via question-text and description analysis: event-resolved (resolution triggered by a publicly observable event), deadline-resolved (resolution triggered by the elapsing of a contractual deadline), and unclassifiable . Documented insider-trading cases are systematically deadline-resolved, of the form “Will event X occur by date Y ?”. Deadline-ILS (ILS {}^{\text{dl}} ). For a deadline-resolved YES market with recoverable event timestamp T_{\mathrm{event}} falling within [T_{\mathrm{open}},D] where D is the contractual deadline, \Delta^{\text{dl}}_{\text{pre}}\;=\;p(T_{\mathrm{event}}^{-})-p(T_{\mathrm{open}}),\qquad\Delta^{\text{dl}}_{\text{total}}\;=\;p_{T_{\mathrm{resolve}}}-p(T_{\mathrm{open}}),\qquad\mathrm{ILS}^{\text{dl}}(M)\;=\;\frac{\Delta^{\text{dl}}_{\text{pre}}}{\Delta^{\text{dl}}_{\text{total}}}. Here p(T_{\mathrm{event}}^{-}) is the market price one minute before public observation of the event. The companion paper adopts \theta_{T_{\mathrm{open}}}\equiv p(T_{\mathrm{open}}) as a conservative baseline and a per-category exponential survival function S(\tau)=\exp(-\lambda\tau) for the time-to-event distribution, with hazard rate \lambda fitted by maximum likelihood on resolved deadline markets in the same target category. The scope conditions of the original ILS apply unchanged. T_{\mathrm{event}} recovery. For YES-resolved deadline markets the empirical T_{\mathrm{event}} is recovered through an LLM-assisted multi-source verification pipeline (Claude Haiku 4.5 with web-search tool access). The retrieval target is the timestamp at which the underlying event first publicly