Live

Definition

What is P-value (trading)?

The probability that an observed win rate could occur by chance alone if the trader had no edge — used to separate genuine skill from lucky streaks.

In detail

P-value answers: 'If this wallet had zero edge and was effectively flipping coins, how likely is the win rate I observe?' A p-value of 0.05 means a 5% chance the result is pure luck. Lower p-values mean stronger evidence of genuine predictive skill. In prediction-market analytics, p-value is calculated via the binomial test — given N resolved bets and an expected win rate (usually 50% for binary markets or the category base rate), what's the probability of observing K or more wins?

How CrowdIntel measures it

CrowdIntel calculates p-value for every wallet and investigation using a one-sided binomial test. Wallet profiles display p-value alongside raw and Bayesian win rate. Investigation pages require p-value ≤ 0.1 to be opened; p ≤ 0.01 unlocks higher sitemap priority and 'Cluster Intel' badges on alert cards.

Frequently asked

What p-value threshold is meaningful for trading?

Scientific convention uses p<0.05. For adversarial settings with many wallets tested (multiple-comparison problem), stricter thresholds like p<0.01 or Bonferroni correction are safer.

Can p-value be gamed by a trader who loses early bets?

Yes. A wallet can run losses on a side account and wins on a main account, artificially lowering the main account's apparent p-value. CrowdIntel's cluster detection partly catches this by linking wallets via funding.

Does low p-value prove insider trading?

No. It proves the performance is unlikely under a no-skill null hypothesis. The performance could come from real skill, information asymmetry, or a biased market category. P-value flags candidates for investigation, not verdicts.

Related terms

Last updated 2026-04-25. Sourced from live on-chain Polymarket data via CrowdIntel.
Live
Beta
···v0.1.0