What is statistical arbitrage?
Statistical arbitrage (stat arb) uses quantitative models to identify temporary mispricings between correlated assets, then trades the expected mean reversion. Unlike pure arbitrage (risk-free, locked-in profit), stat arb carries directional and correlation risk — if the spread doesn't mean-revert, the trade loses money. In crypto, common stat arb pairs include ETH vs SOL (both high-beta L1s), BTC vs ETH (different-duration store-of-value bets), narrative baskets vs individual tokens (AI Agents index vs specific AI tokens), and DEX governance tokens vs their parent chain. The core skill is identifying pairs with genuine economic relationship (not just statistical coincidence) and trading the spread when it deviates from its historical mean.
Pairs trading mechanics
A pairs trade longs the relative underperformer and shorts the relative outperformer when their spread widens beyond a threshold — typically 2 standard deviations from the historical mean. Example: historically ETH/SOL has traded in a 0.05-0.08 range (30-day rolling). If ETH/SOL spikes to 0.10 (2 sigma wide), short ETH and long SOL to bet on mean reversion toward 0.065. The trade closes when the ratio returns to its mean (positive PnL) or widens further beyond a stop threshold (negative PnL). Holding periods range from hours to weeks. Position sizing uses dollar-neutral construction — equal notional on long and short legs.
Finding stat arb pairs
Three selection criteria. First, economic relationship — pairs with fundamental reasons for correlation (both are L1 tokens, both are DEX governance tokens, same ecosystem). Second, statistical stability — rolling correlation above 0.7 over multi-quarter windows, stable long-term mean. Third, mean reversion — the spread demonstrably returns to its historical range after dislocations (co-integration tests, half-life estimation). Crypto-specific examples that have worked historically: ETH/SOL, AAVE/COMP, UNI/SUSHI during overlap periods, BTC dominance / ETH dominance ratio, L2 basket vs ARB/OP.
Execution on perpetual futures
Perps are the optimal venue for stat arb — you can short the overperformer without locating borrow, size is unrestricted, and funding rates add a small but meaningful component to returns. Execute both legs simultaneously with dollar-neutral sizing (e.g., $50K long SOL + $50K short ETH). Monitor delta exposure — if ETH and SOL deltas diverge materially, rebalance to stay neutral. Funding rates on both legs can accumulate meaningfully over multi-week holds; compute net funding exposure (long funding − short funding) and include it in the expected return calculation.
Risk management for stat arb
Three risks dominate. First, correlation breakdown — if the historical relationship changes (ETH and SOL stop correlating because of a Solana-specific catalyst), the trade becomes directional rather than spread-based. Second, spread widening instead of narrowing — mean reversion is statistical, not guaranteed. Set a stop at 3-4 sigma from the mean and exit if breached. Third, funding-rate costs — long/short perp positions accumulate funding exposure; during volatile regimes this can erode expected returns. Size positions for the full scenario including funding drag and correlation-breakdown tail risk.
When stat arb outperforms vs underperforms
Stat arb works best during range-bound, low-volatility regimes where spreads oscillate predictably around historical means. It underperforms during trending markets (correlations strengthen in one direction), regime transitions (historical relationships break), and extreme volatility (spreads widen beyond historical bounds). Institutional stat arb funds diversify across 50-200 pairs simultaneously to smooth single-pair variance. Retail stat arb traders running 3-5 pairs have wider return distributions and require longer track records to distinguish skill from variance.