Whale Signals
An event study testing whether large Polymarket bets predict moves in related stocks.
What it does
Prediction-market traders put real money behind their views, so those prices might react to news before stocks do. I measured whether large moves on Polymarket are followed by predictable moves in the stocks exposed to the same events.
Technical details
- Signal
- YES price moving more than 7 points within 60 minutes, on markets with at least $100,000 in volume
- Sample
- 604 resolved markets from 2023 to 2024, giving 2,623 signals after removing repeats and 16,731 observations
- Method
- Market-model event study against SPY, abnormal returns at 1, 3, 7 and 30 days, binomial tests on direction
- Data
- Polymarket APIs and Yahoo Finance. Anyone can reproduce it from public data.
Results
Primary result
Directional accuracy was 46.4% at seven days, below chance. Large bets were followed by small negative returns, not the sector moves I predicted.
Crypto reversal
When crypto-related odds rose, the stocks I expected to benefit (MSTR, COIN, MARA, RIOT) fell, with a -16.46% cumulative abnormal return at 30 days (p < 0.0001, n = 1,028). Taking the opposite side was right 62.8% of the time. This is in-sample and the observations overlap, so it hasn't been tested out of sample.
Geopolitical markets
This was the only category with a positive result: +1.43% at 30 days (p = 0.0037) and 55.5% accuracy at seven days.
Follow-up study
Mapping signals to specific companies instead of broad sectors made accuracy worse in every category, falling from 46.4% to 43.6%.
Getting the data
Polymarket's price-history API returns nothing for resolved markets when you ask for the full range, and that isn't documented. I recovered the full history by requesting each market in 13-day windows, working back from its resolution date.
Data
