April 8, 2026
Base Rate Neglect on Prediction Markets
A freshman senator is running against an incumbent. Polls show a tight race. Polymarket has the challenger at 38 cents. The narrative is strong. Traders load up.
Since 1946, U.S. House incumbents seeking re-election have won roughly 93 percent of the time, and Senate incumbents roughly 84 percent (Vital Statistics on Congress, Brookings, 2021 edition). The historical frequency is not the final answer, but it is the starting number. Most traders pricing the challenger never computed it.
That error has a name. Base rate neglectis the documented human tendency to price the specific case while ignoring the category it belongs to. Kahneman and Tversky first described it in 1973 ("On the Psychology of Prediction"). Fifty years later, it is still the single most common calibration error in forecasting research, and the easiest edge to systematize on a prediction market.
Why the market keeps making this mistake
Prediction markets aggregate narrative as readily as they aggregate information. A compelling story, repeated in enough places, produces a flow of small traders willing to pay above the historical frequency. The book moves to the narrative even when the reference class does not support it.
Concrete examples from the reference classes that matter on Polymarket and Kalshi:
- Sports favorites. NBA regular-season favorites by implied probability close to their opening price (covers.com historical closing-line data, 2014-2024). Polymarket often prices them 2-5 points richer than the closing line on Pinnacle because the payoff structure attracts recreational flow.
- Incumbent re-election. House incumbents win ~93% of contested races (Brookings, 2021). When Polymarket prices a challenger above 15 cents on the dollar without a specific scandal or retirement, the position is priced on narrative, not frequency.
- Repeat performers in primaries. A candidate who placed top-three in Iowa and New Hampshire has historically won the nomination ~70% of the time since 1972 (Cohen, Karol, Noel, Zaller, The Party Decides). Contracts pricing the front-runner below that level without a new disqualifying event are typically wrong.
The operator fix: anchor, adjust, size
Philip Tetlock's Good Judgment Project found that the best forecasters start with the base rate and adjust conservatively from there, typically by 5 to 15 percentage points, not 50 (Tetlock and Gardner, Superforecasting, 2015). Prediction-market trading inherits this directly.
1. Anchor. Before pricing any contract, write down the historical frequency for the reference class. Incumbents win contested House seats 93% of the time. NFL home favorites cover ~49%. Bitcoin closes up on the year in 8 of the last 12 calendar years. That number is the anchor.
2. Adjust. Move from the anchor only for specific, named, falsifiable reasons. A sitting incumbent under indictment moves the anchor. The challenger having a compelling stump speech does not.
3. Size to the gap. Your position size should reflect the edge versus market, not the strength of the narrative. A 4-cent gap between the market price and the base-rate-adjusted fair price is a small position. A 15-cent gap in a category where you have real domain knowledge is the kind of concentration the 10,000-wallet cohort shows actually drives profit.
Where this connects to the wallet analyzer
Convexly's 10,000-wallet Polymarket study found that the profit leaderboard is dominated by traders who concentrated on a handful of asymmetric bets, not by the best-calibrated quartile across the full book. Spearman correlation between Brier score and realized PnL across 8,656 ranked wallets was +0.148 (Edge Score Methodology V1). The structural pattern is the same one Kahneman and Tversky predicted: traders pay for narrative, the base-rate-anchored trader wins over time.
Run your own wallet through the analyzer and you will see whether your concentrated bets were base-rate-anchored conviction or narrative trades that happened to work.
See how your wallet scores.
Paste any Polymarket address. Posture, conviction, and discipline percentiles against 8,656 benchmarked wallets. Free, no signup.
Analyze a walletRelated reading
- Stop Optimizing Calibration. Start Optimizing Conviction.
- The Brier Score: Necessary but Not Sufficient on Prediction Markets
- The Pre-Mortem: Running One on a Polymarket Thesis
Sources: Kahneman & Tversky (1973), "On the Psychology of Prediction." Tetlock & Gardner (2015), Superforecasting. Vital Statistics on Congress (Brookings, 2021 edition). Cohen, Karol, Noel, Zaller (2008), The Party Decides. Convexly (2026), 10,000 Polymarket Wallets Scored.