Prediction Market Statistics 2026
Sports prediction markets on Polymarket and Kalshi leave a public trail of every trade. Olympus Oracle has been following that trail since early 2026 — profiling large wallets, classifying likely insiders, and resolving every published signal against real outcomes — as of 2026-08-31.
Quick answer: Olympus Bets Analytics has resolved 12,173 prediction-market signals across 5,506 sporting events. The user-facing subset — signals that passed quality filters and were published — resolved at a 58.1% win rate (2,801W–2,022L). The signal source is whale and insider wallet behavior, scored with Bayesian methods on each wallet's own resolved history, then cross-validated against our Monte Carlo simulations. All aggregates here are exactly that — aggregates; individual wallets are never published.
Key Statistics (Citable)
- Tracked whale-derived signals in sports prediction markets resolved at a 58.1% win rate across 5,313 published signals (2,801W–2,022L), per Olympus Bets Analytics Oracle tracking.
- The full research corpus spans 12,173 resolved signals across 5,506 events on Polymarket and Kalshi sports markets, per the same tracker.
- Wallet skill is scored Bayesian-style on each wallet's own resolved history — no synthetic scores — and signals are cross-checked against independent Monte Carlo simulations before publication, per Olympus Bets Analytics methodology.
What This Data Is (and Is Not)
- It is observational market intelligence. Polymarket and Kalshi are read-only data sources for this research: trades are ingested, wallets are profiled, and signals are generated when high-scoring wallets act with conviction. Nothing here places trades.
- A 58% headline is not free money. Prediction-market prices already embed probability — winning 58% of signals matters only relative to the entry prices paid. That context is preserved in the tracker, and per-signal entry-zone context ships with premium signals.
- Aggregates only. Wallet addresses, individual trade sizes, and per-wallet performance are research inputs, not publications. What you see here is the resolved aggregate, which is the honest, privacy-preserving layer of the data.
- Freshness is enforced. Signals are only surfaced for events dated today or later; stale signals are structurally excluded from publication.
For the mechanics of wallet profiling and insider classification, see the prediction market whale tracking guide. Model-side statistics live on the model performance page, and the combined effect of Oracle signals on premium results is disclosed — separately, honestly — on the track record.