Pairs trading is one of the most intellectually appealing strategies in quantitative finance. Find two assets that move together over time (cointegrated), and when they diverge, bet on convergence. The spread between the two assets oscillates around a stable mean, and you capture the oscillation. It works beautifully in equities where sector pairs (Coca-Cola and Pepsi, Ford and GM) maintain long-term relationships driven by shared fundamentals.
We built and tested a cointegration pairs strategy for crypto. It failed, and the failure is instructive about the limits of applying statistical arbitrage concepts to crypto markets.
The Engle-Granger Test
Cointegration is a statistical property: two time series are cointegrated if a linear combination of them is stationary (reverts to a mean). The Engle-Granger test checks for this by regressing one series on the other and testing whether the residuals are stationary using the Augmented Dickey-Fuller test.
Our scanner tested all pairwise combinations of our 25-symbol universe, filtering for pairs where the ADF test rejected the null hypothesis of non-stationarity at the 5 percent significance level. We found 9 pairs that passed: various combinations of correlated altcoins like SOL/AVAX, DOGE/SHIB, NEAR/ARB, and similar.
These pairs pass the statistical test because crypto altcoins are highly correlated. When BTC moves, most altcoins move in the same direction with varying magnitude. This correlation creates the appearance of cointegration in the spread, but the underlying mechanism is different from equity pair relationships.
The Half-Life Problem
For a pairs trade to be practical, the spread must revert to its mean within a tradeable timeframe. The half-life measures how many bars it takes for the spread to revert halfway from a deviation to the mean. A half-life of 10 bars means deviations are corrected quickly, generating frequent trading opportunities. A half-life of 500 bars means you wait weeks or months for a single trade to complete.
All 9 of our validated pairs had half-lives exceeding 169 bars on 15-minute candles. At 15 minutes per bar, 169 bars is approximately 42 hours. The shortest half-life was 169 bars. Most were in the 300 to 500 range. This means a spread deviation takes 3 to 7 days to mean-revert, which produces at most a few trades per month.
With so few trades, the strategy cannot generate a statistically meaningful track record within our validation framework. Our five-period validation requires enough trades per period to calculate Sharpe ratios with confidence. At 2 to 4 trades per month, a one-year validation period produces 24 to 48 trades. That is barely enough for a Sharpe calculation and completely insufficient for parameter optimization.
Why Crypto Pairs Revert Slowly
The slow half-life in crypto pairs has a structural explanation. In equities, pair relationships are maintained by fundamental linkages: shared customers, shared supply chains, shared regulatory environments. When Coca-Cola diverges from Pepsi, fundamental investors notice and rebalance, driving convergence within days.
In crypto, pair relationships are maintained primarily by correlation to BTC. SOL and AVAX co-move because they are both high-beta altcoins that respond to the same macro flows. When they diverge, the divergence is typically driven by asset-specific news (a Solana outage, an Avalanche partnership) that affects one but not the other. These asset-specific drivers do not self-correct through fundamental arbitrage. They persist until the next BTC-driven move pulls both assets back into alignment.
The reversion mechanism in crypto pairs is not arbitrage pressure (as in equities) but macro correlation. The spread converges when BTC moves and both altcoins respond, not because anyone is specifically trading the spread. This makes the reversion timing dependent on BTC macro cycles rather than pair-specific dynamics, which is why the half-life is measured in days rather than hours.
What We Tried
We tested several approaches to make pairs trading viable. Shorter timeframes (5-minute, 1-minute) did not improve half-lives because the reversion mechanism operates on macro timescales regardless of the measurement frequency. You can measure the spread every minute, but it still takes days to revert.
Dynamic hedge ratios (recalculating the linear combination periodically) slightly improved out-of-sample results but did not solve the fundamental half-life problem. The ratio changes slowly because the underlying relationship is stable; the reversion is just slow.
Expanding the symbol universe to include smaller tokens (which might have faster reversion due to thinner liquidity) was considered but rejected because smaller tokens have insufficient historical data for our five-period validation framework and their spreads are dominated by exchange-specific pricing differences rather than genuine cointegration.
The Honest Assessment
Cointegration pairs trading in crypto is a dead end for systematic strategies operating on timeframes below daily. The mathematical framework is sound. The statistical tests find genuine cointegration. But the half-lives are too long for the strategy to generate enough trades for validation, optimization, and risk management within practical timeframes.
For traders operating on daily or weekly timeframes with larger capital and longer holding periods, pairs trading might work. The half-lives of 3 to 7 days are tradeable if you are willing to hold positions for weeks and accept the low trade frequency. But this operating profile is fundamentally different from our 15-minute altcoin strategies that generate 50 to 150 trades per year.
We classified cointegration pairs as a dead end in our strategy discovery roadmap and redirected effort toward approaches with higher signal frequency: derivatives strategies using open interest and funding rate data, and macro strategies using cross-asset indicators. Both provide the trade frequency needed for robust validation and risk management.
Lessons for Strategy Development
The cointegration failure reinforced several principles. First, statistical significance in a test does not guarantee practical tradability. All 9 pairs passed the Engle-Granger test at the 5 percent level. None were tradeable in practice.
Second, the reversion mechanism matters as much as the reversion itself. Equities revert through fundamental arbitrage (fast, reliable). Crypto pairs revert through macro correlation (slow, episodic). Understanding why reversion occurs determines whether it will occur fast enough to trade.
Third, trade frequency is a constraint that should be evaluated before building a full strategy. If the expected trade count is below 50 per year, the strategy will be difficult to validate, optimize, and manage. We now check expected signal frequency during initial screening before investing in full parameter sweeps and validation runs.