Select pairs with an economic link
A shared ecosystem, collateral type, protocol function, liquidity source, or demand driver can motivate a pair. Quantitative screens can then test liquidity, listing overlap, return co-movement, and residual behavior. The economic link does not prove convergence, but it makes the hypothesis interpretable and the failure modes easier to name.
Use point-in-time universes so failed or delisted assets are not erased from history. Avoid pairs whose apparent relationship is created by stale prices or different trading hours. Token migrations, redenominations, and quote-currency differences need explicit adjustment.
Construct a reproducible spread
Define the price source, timestamp alignment, transformation, hedge-ratio method, estimation window, and rebalance schedule. A common structure uses log prices and an estimated beta, but rolling regression, state-space models, and factor residuals produce different spreads. Faster adaptation can follow change or simply chase noise.
Correlation can help screen pairs, yet mean reversion usually requires stronger evidence. Cointegration tests can evaluate whether a weighted price combination appears stationary in the sample. Treat the result as evidence, not permanence. The relationship can break after a token-specific event or market-structure change.
Convert deviations into stateful signals
A z-score standardizes the spread relative to a rolling mean and volatility. Entry, exit, and stop thresholds should be chosen with costs and holding time in mind. Hysteresis can reduce rapid flipping. Time stops, model stops, and loss stops address different risks and should be recorded separately.
The strategy should know whether it is flat, entering, holding, reducing, or exiting. Repeated signals in one direction or lengthening holding periods can indicate relationship decay. A model should enter quarantine when diagnostics challenge the original hypothesis.
Size and execute both legs
Equal dollars are not necessarily hedged. Use the modeled ratio, target spread risk, and executable depth to size the legs, then cap gross exposure. Include broad crypto beta, ecosystem factors, funding, borrow, and instrument basis. A spot-perpetual pair is also a derivative trade.
Calculate entries at the ask for the long and bid for the short, then reverse sides for the exit. Model partial fills and the thinner leg. The order policy needs timeouts, cancellation, residual hedging, and a loss budget if simultaneous execution fails.
Monitor the relationship after launch
Track residual volatility, rolling hedge ratio, correlation, test statistics, realized slippage, funding, holding time, and stop frequency. Structural events such as unlocks, exploits, governance changes, or delistings can invalidate one leg immediately.
A live pair is not entitled to remain active because its original backtest looked strong. Define retirement and reapproval criteria. The cleanest result of monitoring may be a smaller universe of relationships that remain explainable and executable.