Prediction Market Insider Radar

Polymarket is the world's largest decentralized prediction market, where participants bet real money on real-world outcomes: sports results, geopolitical events, elections, economic data. By design, all trades are public and on-chain. This creates a rare and largely unexploited data source: a continuous, transparent record of every bet placed, including bets made by people who may know something the rest of the market does not.

Traditional financial markets have strict insider trading laws. Prediction markets do not. When a political operative knows a policy announcement is coming, a diplomat has early read on a ceasefire, or a market participant has non-public information about an economic or geopolitical outcome, they can and do bet on Polymarket ahead of the public. This system is designed to detect those bets before the market prices in the new information.

Prediction markets misprice outcomes when informed participants act before public information becomes available. The window between when an insider trades and when the market corrects is measurable from public on-chain data. If we can identify those trades reliably, we can follow them, either as a signal service or by paper trading at similar prices.

The challenge is signal-to-noise. Polymarket sees tens of thousands of trades per day from retail bettors, bots, arbitrageurs, and market makers. Most large bets are noise. The algorithm's job is to score each bet on factors that distinguish informed conviction from routine activity.

Every day, the scanner:

1. Fetches ~5,000 recent public trades from the Polymarket data feed, covering roughly the last 24–48 hours of activity
2. Filters for candidates matching the insider profile: price ≤ 40% (market thinks it's unlikely), minimum $25 notional (meaningful conviction), excluding automated and price-level markets with no plausible information edge
3. Scores each trade using six factors derived from on-chain wallet behavior:
   • Rarity: wallets with few total trades are more suspicious than frequent traders
   • Unlikeliness: bets on lower-probability outcomes carry more information content
   • Size: larger bets signal stronger conviction
   • Size deviation: a $500 bet from a wallet that normally bets $20 is more suspicious than a whale's routine $500
   • Concentration (HHI): wallets focused on one specific market are more likely acting on specific information than broad portfolio traders
   • Cluster: when multiple independent wallets bet the same unusual outcome within hours of each other, the convergence is hard to explain as coincidence
4. Z-normalizes scores against the batch, applies time-decay (fresh signals score higher), resolution-proximity boost (signals near a resolution deadline are more actionable), and a new-wallet multiplier (wallets never seen before in any prior scan)
5. Saves top-scored signals as ALERTs to a database and tracks outcomes as markets resolve

As of May 2026, with 629 resolved ALERT-level signals tracked over approximately 6 weeks:

The results suggest a genuine informational edge, not merely random outperformance. The expected baseline win rate implied by market pricing is 23.8%, because a 23.8¢ bet has a 23.8% chance of paying off in an efficient market. Beating that by 3.9 percentage points means the algorithm is identifying bets that win more often than the market expected.

Statistical status: Under a standard proportion test, the observed edge is approximately 2.3 standard deviations above expectation, modestly exceeding the conventional ~1.96 standard-deviation threshold associated with 95% two-tailed statistical significance. (Calculation: observed edge = 3.9 percentage points; estimated standard error = √(23.8% × 76.2% / 629) ≈ 1.7 percentage points; z-score = 3.9 ÷ 1.7 ≈ 2.3.)

Prediction markets have a natural leverage property: a 10¢ bet that wins pays $9 profit per dollar wagered. A 5¢ bet pays $19. This is not unusual leverage; it's the fair odds on a low-probability outcome. But it means that even modest signal quality (being right slightly more often than the market expects) generates returns that look outsized compared to traditional asset classes.

The current +24.6% ROI comes from a small number of extreme longshot wins:

These are not lucky outliers to be dismissed. They are the structural shape of this strategy. When an informed bettor bets on a 2% outcome and wins, the market dramatically underpriced that outcome, and our algorithm correctly flagged it. The large payoff is the reward for correctly identifying a rare informed bet. A portfolio of these bets, when correctly filtered, generates asymmetric return profiles: most lose small, a few win large.

In rough order of likelihood:

1. Genuine political and geopolitical early information (~40%): the Russia-Ukraine ceasefire trade is the clearest example. Someone bought at 6% before a publicly known outcome. This is the category closest to traditional insider trading.
2. Coordinated information groups (~35%): Discord or Telegram communities that aggregate information and bet together. The Bulgaria Eurovision win at 2% likely reflects this pattern: a well-connected fan community with strong conviction before the public result. The cluster signal (multiple independent wallets, same outcome, same 4-hour window) often reflects this.
3. Informed noise (~25%): large single bets with coincidental signal characteristics. The rarity and new-wallet factors catch some false positives here.

• Statistical uncertainty: Current results are statistically encouraging but the observed edge could narrow or disappear as Polymarket grows and attracts more sophisticated market makers.
• Market structure: Polymarket liquidity is thin in many markets. The paper portfolio's $10 bet has no market impact. A real portfolio at $500–5,000 per trade would move prices in illiquid markets, shrinking the edge.
• Counter-detection: As on-chain signal detection becomes more widely known, informed bettors may adapt by splitting positions across wallets or timing trades to reduce detectability, which would degrade the cluster and rarity signals specifically.
• Legal and platform risk: Polymarket is a decentralized protocol, accessible from most jurisdictions but geo-blocked in the US and a few other markets. Platform risk exists as regulatory environments evolve.

At its current scale (paper $10/trade), the system is a signal tracker and research tool. The paper portfolio is a track record, not a ceiling.

The more interesting near-term opportunity is using the signal data for research: identifying which market categories, wallet profiles, and time patterns produce the strongest and most durable alpha, then concentrating capital there while continuing to build the resolved-outcome database toward statistical certainty.