While most traders bet on who wins an election or event, institutional quant desks do something completely different.
They treat prediction markets as fragmented order books and exploit temporary price discrepancies between platforms (mainly Polymarket and Kalshi) using statistical arbitrage and market microstructure.
These strategies are delta-neutral — they don’t care about the final outcome. They only care about the spread closing.
Why Cross-Platform Arbitrage Exists
When two platforms offer markets on the exact same real-world outcome, their prices must eventually converge to the same terminal value ($1 or $0).
However, differences in:
- Liquidity
- User base
- API speed
- Regulatory structure
- On-chain vs centralized settlement
…create temporary divergences that can be systematically exploited.
The Data Advantage: L2 Order Books
Retail traders look at price charts (executions).
Quants look at the full limit order book (L2).
A public 55GB dataset with tick-by-tick L2 snapshots (up to 20 levels deep, sampled at 100ms) going back to late 2024 has been released for this exact purpose. It includes synchronized timestamps across platforms in optimized Parquet format.
This level of data allows proper backtesting of liquidity-aware strategies.
Method 1: Cointegration + Ornstein-Uhlenbeck Mean Reversion
This is the statistical backbone for longer-horizon cross-platform arbitrage.
Step-by-step process:
Test for Cointegration
Let $P_{P,t}$ = mid-price on Polymarket and $P_{K,t}$ = mid-price on Kalshi.
Construct the spread:
$$
S_t = P_{P,t} - \beta \cdot P_{K,t} - \mu
$$
Use the Augmented Dickey-Fuller or Johansen test to confirm the spread is stationary (mean-reverting).-
Model the Spread with Ornstein-Uhlenbeck (OU) Process
$$
dS_t = \theta (\mu - S_t) \, dt + \sigma \, dW_t
$$- $\theta$ = speed of mean reversion
- $\mu$ = long-term mean
- $\sigma$ = volatility of the spread
Calibrate Parameters using Maximum Likelihood Estimation on historical data.
Find Optimal Entry/Exit Thresholds
Solve an optimal stopping problem to maximize expected profit after fees and slippage.
If the mean-reversion speed ($\theta$) is fast enough relative to execution latency, the spread becomes highly tradeable.
Method 2: Order Book Imbalance (OBI) + Micro-Price (High-Frequency)
For shorter timeframes (milliseconds to seconds), statistical models are too slow. Quants go directly to liquidity microstructure.
Key formulas:
Order Book Imbalance (Level 1):
$$
I_t = \frac{V_{bid}(t) - V_{ask}(t)}{V_{bid}(t) + V_{ask}(t)}
$$
Micro-Price (better estimate of “true” value than mid-price):
$$
P_{micro}(t) = \frac{V_{bid}(t) \cdot P_{ask}(t) + V_{ask}(t) \cdot P_{bid}(t)}{V_{bid}(t) + V_{ask}(t)}
$$
This can be rewritten as:
$$
P_{micro}(t) = P_{mid}(t) + I_t \cdot \frac{\Delta spread}{2}
$$
Cross-Venue Alpha
Because Polymarket (hybrid/on-chain) and Kalshi (centralized) have different processing speeds, a strong imbalance on one platform often predicts a move on the other within a short window (e.g., 200ms).
Quants build predictive models (sometimes using Markov chains) on historical L2 data to front-run these micro-moves.
Core Philosophy: Process Over Prediction
The biggest edge isn’t better forecasting of events.
It’s:
- Continuous risk harmonization via calibrated statistical models (OU parameters)
- Microstructure dominance — understanding the mathematical relationship between order book imbalance and price movement
- Latency capitalization — acting on cross-platform signals before retail systems react
These desks are completely agnostic to politics, news, or fundamentals. They only trade the structural relationship between two order books.
Practical Takeaways
| Approach | What They Focus On | Time Horizon | Edge Source |
|---|---|---|---|
| Retail | Event outcome / news | Hours–Days | Prediction |
| Quant (Cross-Platform) | Price spread between venues | Milliseconds–Hours | Statistical + Microstructure |
| High-Frequency | Order book imbalance & micro-price | < 1 second | Latency + Liquidity |
The release of large L2 datasets and open repositories is making these previously institutional-only strategies more accessible for serious builders.
Would you rather focus on event-driven trading or structural/statistical arbitrage between platforms?
If you have more questions, please feel free to contact me at any time: https://t.me/FatherSon97
Top comments (0)