← Projects·Quantitative Researcher
Pairs Trading via Cointegration (BTC / ETH)
A cointegration study on BTC / ETH: the most-traded pair in crypto, and the most obvious candidate for mean-reversion. The ADF test is honest about it.
−2.11
ADF test statistic
fail to reject unit root at 5%
208 days
Half-life (estimated)
≈ 0
OOS Sharpe (fade)
as the test predicted
The hypothesis
BTC and ETH move together. If they are cointegrated, a stationary spread should mean-revert — a textbook stat-arb entry.
What the project does
- Pulls BTC/ETH daily prices.
- Estimates a hedge ratio by OLS.
- Runs an Augmented Dickey-Fuller test on the spread (implemented from scratch, no
statsmodels). - Estimates the half-life of mean-reversion from an AR(1) on the spread.
- Fades the spread OOS; applies a 5 bps/side cost.
The result
- ADF test statistic: −2.11 — fails to reject the unit-root hypothesis at the 5% level. The spread is not stationary.
- Half-life: 208 days — too slow to be a trade.
- The OOS fade produces ≈ 0 Sharpe net of cost — and that is the correct answer, given the test.
What’s transferable
Stat-arb begins with the cointegration test, not with the trade. A pre-trade ADF regime filter is the single largest quality lift on a pairs/mean-reversion pipeline.