Safe Choices
Assessing the efficiency of "safe bets" on prediction markets
Imagine 50 hedge funds, each deploying $100 on a "State Earthquake Trade." Each fund bets on their state NOT being hit by an earthquake (10% annual probability), earning 1.11x if correct.
After 10 years across 100,000 simulations: 17.42 funds survive on average (range: 4-32). These survivors produced 11% annual returns for a decade. In an efficient market, luck becomes the differentiating factor between investment legends and bankruptcy.
Research Questions
We explore the efficiency of "safe choices" (markets trading at 90+ cents before resolution):
- How often will traders making random walks across these markets go bust?
- How often will they beat traditional benchmarks?
- Does stopping at a return threshold (e.g., 12%) improve outcomes?
- Does splitting capital across multiple funds improve survivorship?
Dataset
Data collected via Polymarket's Gamma API and CLOB API, filtering out high-variability categories:
For remaining markets, we log probability snapshots from 7 days to 1 day before resolution. The dataset covers 2024-2025 and is available on Kaggle.
Simulation Model
Strategies
Single Fund
Deploy all capital sequentially across safe markets, reinvesting all winnings. High risk, high reward.
Target Return
Same as Single Fund but stop trading once a target return (e.g., Treasury rate or NASDAQ average) is reached.
Multi Fund
Split capital into N independent funds. Tests whether diversification improves survivorship rates.
Technical Notes
- Uses actual Polymarket data filtered for 2025+ resolution
- Vectorized operations for performance (~1000 sims in 10-20s)
- Binary outcomes: win = 1/probability return, loss = total loss
- Reproducible results via random seeds
Limitations
At $10,000 scale distributed across funds, liquidity isn't a concern. However, scaling this strategy significantly would face real liquidity constraints and potential market impact.