Screening Polymarket M&A Markets for Insider-Shaped Trades

The tell, borrowed from the options desk

Insider trading around mergers has a well-documented signature in equity options: an account that has never traded options buys a block of cheap, short-dated, out-of-the-money calls, and days later the target announces a deal. Every element of that signature has a prediction-market analogue. The account is a wallet. The out-of-the-money call is a YES contract at long odds. The expiry is the market's resolution date. And on Polymarket's global exchange, unlike at an options clearinghouse, every fill is public on a blockchain.

Bloomberg's analysis (with analytics firm Polysights) scored trades across eight metrics — stake size, account recency, entry odds, volume concentration, and four others they did not fully enumerate. We defined our own eight in the same spirit, published the exact anchors below, and ran them over every merger market we could find that is tied to a publicly traded company on either side of the deal.

The universe

Discovery was a keyword sweep of Polymarket's search API (acquire, merger, buyout, takeover, and variants), followed by an editorial pass: does the market concern a corporate acquisition, and is at least one side a listed equity? That produced 32 events. Markets from before 2024 (Twitter×Musk, SVB, FTX×Robinhood) had to be excluded — the public trade API serves no fills from that era. Active markets (GameStop×eBay, Stripe×PayPal, the Warner Bros. close race) are pulled but not scored until they resolve.

Each scored market's full lifetime trade history was pulled from the public data API — both maker and taker legs, because Polymarket's exchange can match a YES limit buy directly against NO buyers, so a patient limit-order buyer may never appear on the taker side at all. That is 45,938 fills across 28 resolved markets, about $2.9M in both-sides notional. Announcement "jumps" are detected from the trades themselves: the first sustained move from 40¢ or below to 90¢ or above. The detector is deliberately conservative — only three markets show a clean jump; deals that repriced gradually contribute to every metric except timing.

Eight metrics, fixed anchors

Each (wallet, market) position on the deal side is scored 0–1 on eight metrics. Anchors are fixed, not fit to the data, so a score means the same thing in every market. The composite is a weighted mean (identity and bet-shape tells weighted up), and a position is flagged only if the composite reaches 70, at least five metrics exceed 0.6, and the stake exceeds $200 — the stake floor exists because airdrop-farming dust wallets otherwise dominate the top of the list.

One correction the data forced on us: wallets whose history exceeds the 50,000-trade pull cap initially looked brand new — the capped window's start masqueraded as their first-ever trade, and a $3.7M-volume whale briefly topped the suspicion rankings. Established whales now have their newness and breadth scores pinned to zero by construction.

What the screen found

Across 28 resolved markets and 4,096 scored positions, the screen flags exactly one.

It is a textbook match for the insider signature: a wallet whose first-ever Polymarket action was a single $731 buy at roughly 10¢, 100% of its lifetime volume in one market, no prior history, pure accumulation, never hedged. The market was "Tesla and xAI merger officially announced by June 30?" — and the merger was never announced. The position expired worthless. The screen's single flag is a perfect insider shape attached to a bet that lost everything.

The near-misses tell the same story from different angles. The highest-scoring position in a market where a deal was announced: a day-old wallet that bought the Netflix side of the Warner Bros. Discovery race 19 hours before the price jumped — for $7. The largest high-suspicion stake anywhere in the data, $66,000 concentrated in the Pizza Hut market, bought at 33¢ and exited completely flat before resolution — the shape of a market-maker scratching a position, not of informed money riding a deal to payout.

The control test

Here is the part a headline screen cannot do without resolved losers. Twenty-two of our markets are deals that were never announced — Qualcomm×Intel, Honda×Nissan, Chubb×AIG, United×American, Apple×Mistral. If insider-shaped trading reflects actual information, high-suspicion profiles should concentrate in the markets where deals really happened. So we computed an outcome-blind profile score (stake, odds, account age, concentration, breadth, conviction — no timing, no payoff, nothing a winner gets for free) for every profiled deal-side buyer in both cohorts, selected by an identical rule.

The prevalence is not higher where deals happened. It is lower — about 3% of positions in announced-deal markets fit the high-suspicion profile versus about 5% in the no-deal controls, a ratio of roughly 0.56, with overlapping confidence intervals. And the aggregate economics are decisive: the 42 high-suspicion positions across both cohorts staked about $67,200 and collectively lost money (−2% return). The 36 in no-deal markets lost 100% of everything they staked. Sixty-seven positions came from wallets less than a day old; in aggregate they lost roughly half their stake.

In this universe, the insider shape is overwhelmingly the gambler shape: fresh wallets making cheap, concentrated, conviction bets on longshot mergers that mostly never happen. The tell that works on an options desk does not, by itself, separate information from hope in Polymarket's M&A markets.

Why this differs from the Bloomberg finding

This is a domain finding, not a refutation of the Bloomberg analysis. Their flagged volume concentrated in geopolitical and military markets (roughly $45M around Iran alone), where events are decided by small groups of people on known timelines, and where Polymarket's liquidity is deep. Corporate M&A on Polymarket is a different habitat: markets are thin (median both-sides notional in our universe is well under $100k), listed deals skew toward speculative rumor pairings, and anyone with genuine deal information has far deeper and more familiar venues — options, CDS, the stock itself — in which to monetize it. A rational insider may simply never show up here. The one prosecuted Polymarket insider case to date involved a soldier trading on military operations, not a banker trading on a merger.

Limitations

• Twenty-five of 28 markets never produced a detectable announcement jump, so the timing metric is silent for most positions; composites renormalize over available metrics.
• Wallet profiling is capped (top-25 per market by prescore, plus every position of $500+ at 30¢ or less). The control comparison applies the identical rule to both cohorts, but absolute prevalences are conditional on that selection.
• On negRisk events, an equivalent position can be built by buying NO on every sibling; we do not score synthetic baskets.
• Wallets are not people. One person can run many wallets, and copy-traders can make one informed wallet look like a cluster.
• The public trade API serves nothing before roughly 2024, so era-defining cases (Twitter×Musk) are unreachable.
• With six announced deals, every inference here is small-N and suggestive, not statistical proof of anything.

Reproducibility

The pipeline is five commands over two public, unauthenticated APIs: discover (keyword sweep), pull (full trade history per market, time-window cursor pagination), profile (wallet history aggregates), score (the eight metrics), report (the aggregates rendered on this page). Every number above regenerates from the public record. The one human step is the roster: deciding which markets are genuinely M&A tied to a listed equity. That judgment is recorded per event, with reasons, in the roster file.