base-rate-bayes ✓ / poll-lag-fadePublishedBacktested · verifiedv1.4.0Events & politicsIllustrative data
Poll Lag Fade
Fade election markets that haven't caught up to the polling average.
ForkFollowing is saved in this browser during the preview.
Sharpe
1.90
risk-adjusted
Sortino
2.60
downside-adjusted
Max drawdown
11.0%
peak to trough
ROI (net)
+23.8%
after fees & slippage
Win rate
61%
closed trades
Profit factor
1.80
gross win / gross loss
Sample
84 trades
180 days
Out-of-sample
+6.1% · Sharpe 1.5
30% of history · 27 trades
Equity · 180 days · 70/30 in/out-of-sample
Closing-line value +3.4% · verified 2026-10-29$10,207$11,130$12,054$12,977
5 May19 June3 Aug16 Sept31 Oct
Verification
Backtested · verifiedLoreli re-ran the backtest and the out-of-sample window held up. Verification establishes that the results are reproducible and statistically credible. It is not a judgement that the strategy will be profitable.
In-sample Sharpe
1.90
Out-of-sample Sharpe
1.50
Sharpe decay
21%
over 60% fails verification
Reproduced by Loreli on Polymarket + Kalshi history. OOS Sharpe decayed 21% versus in-sample.
About this strategy
Compares Polymarket election prices against a weighted polling average refreshed hourly. Enters when the implied probability lags the model by more than 6 points and the market has at least $50k of liquidity. Exits on convergence or 10 days before resolution.
Rules
- Model P(yes) − market price ≥ 6 pts
- Market liquidity ≥ $50,000
- Not within 10 days of resolution
- Max 8% of equity per market, 40% total exposure
Signals
Probability discrepancy
Your model says one thing, the market another. Trade the gap.
Event-driven
React to scheduled or breaking information before the market fully adjusts.
Changelog
v1.4.02026-10-28Added Kalshi venue; tightened liquidity floor.
v1.3.22026-10-12Fixed look-ahead in poll timestamp alignment.
v1.0.02026-09-01First public release with 2024 cycle backtest.