Polymarket long-shot bets win 52%

Polymarket is supposed to turn public information into prices. The whole sales pitch is that if enough people trade, the market aggregates what the crowd knows. But this new analysis points at a different mechanism in one corner of the platform — some traders may not be forecasting better, they may just know more. The striking number is in military and defense markets. Large “long-shot” wagers on those contracts — bets of at least $2,500 placed when the implied odds were 35% or lower — wound up winning about 51.8% of the time in the Anti-Corruption Data Collective’s review of settled Polymarket markets from January 2021 through mid-March 2026. That is not a small edge. It is a giant gap between what the price said and what actually happened. (acdatacollective.org) Why does that matter so much? Because a 35% contract should not cash more often than a coin flip unless something is badly off. Sometimes markets misprice things. Sometimes a niche trader just has better models. But when the pattern is concentrated in events tied to military action and other insider-heavy political outcomes, the obvious worry is information asymmetry — people with access to sensitive, early, or classified signals trading against everyone else. (acdatacollective.org) The report’s broader baseline makes the contrast sharper. Across Polymarket overall, similar long-shot bets won about 14% of the time. In political markets more broadly, they won about 25% of the time. Military and defense markets were the outlier at 51.8%. So this is not “all prediction markets are noisy.” It is “one category looks weird in a very specific way.” (gizmodo.com) There is also a concentration story here. The researchers say profits on Polymarket are heavily skewed toward a tiny sliver of accounts, with most gains captured by very few wallets. That does not prove insider trading by itself — plenty of markets have whales and power users — but it fits the same picture. If a handful of traders repeatedly win on low-probability, high-impact events, the platform starts to look less like collective intelligence and more like a venue where informed insiders harvest uninformed flow. (acdatacollective.org) The timing matters too. This did not land in a vacuum. Polymarket has already faced scrutiny over war-related contracts and over whether prediction markets can become a monetized leak channel for people inside governments or militaries. After the April 30 report, Polymarket also moved to tighten surveillance, including using Chainalysis to monitor trading activity for signs of market abuse. Separately, Polymarket US is now publishing daily market reports through its regulated U.S. exchange arm. (ground.news) The catch is that statistical evidence is not the same thing as a courtroom case. The report shows a pattern, not a named roster of guilty traders. You would still need wallet tracing, timing analysis, off-chain identity work, and probably subpoena power to prove who knew what and when. But markets do not need proof beyond a reasonable doubt to have a trust problem. If ordinary users start thinking they are trading against people with advance notice of missile strikes or ceasefires, liquidity and credibility both take a hit. (acdatacollective.org) The bottom line is simple. Prediction markets only work as advertised if prices reflect broadly shared information. If the edge comes from access instead of insight, the market stops being a forecasting tool and starts being a transfer machine. That is the real threat hanging over Polymarket now.