# pairs_rotation / 1

Hypothesis. Some pairs of crypto assets share a common stochastic trend (they are cointegrated over windows of a few weeks): the residual of a regression of one log-price on the other is stationary, so when it is far from zero the "cheap" leg should outperform the "expensive" one as the spread closes.

Rule. Pairs: ETH/BTC and SOL/LINK, half of the portfolio each. On the last 720 bars, regress log A on log B (OLS with intercept), take the z-score of the last residual and the Dickey-Fuller t-statistic of the residuals (no lags). If the t-statistic is below -3.34 (Engle-Granger 5 % critical value for two series) and z < -1, the pair's budget goes entirely to A; if z > 1, entirely to B; otherwise it is split equally between the two legs.

This is not a market-neutral pairs trade. A wallet that cannot short cannot isolate the spread: the strategy is always fully invested and carries the whole market risk. All it can do is tilt between two assets it would hold anyway; its fair comparison is the equal-weight benchmark, and the most it can add is the spread convergence.

What would refute it. The cointegration test is repeated every hour on a rolling window (a multiple-testing problem in itself) and rolling-window cointegration between crypto assets is unstable: a "cheap" leg can simply be the one that is losing for a durable reason. Each rotation costs fee + slippage on half the portfolio, twice.