Why mean-reversion strategies are sleeping in 2026
Pair trading, cointegration baskets, and HMM regime models have all underperformed over the past 12 months. Here's why — and why that's exactly what you'd expect.
By Yuko Xiu, CFA
If you've been watching OpenAlpha's Performance page lately, you've probably noticed something that looks bad: 4 out of our 6 strategies are flat or slightly negative over the past 3 years. Pair Trading: -1.0%. Cointegration Basket: -0.9%. HMM Regime: -1.5%. These aren't disasters, but they're not exactly setting the world on fire either.
Meanwhile, our two trend-following strategies — TSMOM and XGBoost — are crushing it: +12.4% and +13.3% respectively, with Sharpe ratios above 0.5. So what's going on? Are some of our strategies broken?
Short answer: no. They're regime-dependent.
Mean-reverting strategies — pair trading, cointegration baskets, statistical arbitrage in general — make money when prices move around but eventually return to some equilibrium. They thrive in range-bound, choppy markets where the same level keeps getting hit and bounced off.
Trend-following strategies — TSMOM, XGBoost momentum prediction — make money when prices move in one direction and keep moving. They thrive in directional, sustained-trend markets.
These two regimes are mutually exclusive. When markets trend hard, mean reversion gets killed (the spread keeps widening, never reverts). When markets chop sideways, momentum gets killed (every breakout fails, every position whipsaws). 2023-2026 has been overwhelmingly a trend regime — strong dollar, sustained gold rally, persistent equity drift higher with low realized volatility.
How the original papers performed across regimes
Gatev, Goetzmann & Rouwenhorst's 1999 paper on pair trading reported a Sharpe of 1.4 over 1962-1997. That was a market with much more sector rotation, more banking crises, more volatility regime shifts. The same strategy tested over 2010-2020 (a period of relentless central bank liquidity and trend) returned closer to a Sharpe of 0.3.
Moskowitz, Ooi & Pedersen's TSMOM paper (2012) showed annualized Sharpe of 1.0+ across 58 instruments from 1985-2009. But within that 24-year sample, there were multi-year stretches where TSMOM was flat or negative — most notoriously the 1990 oil price spike and the 2009 'risk-on' rebound after the GFC. Same model, different regimes.
Key insight: 'Strategy works' and 'strategy works right now' are two different claims. The first is about long-run statistical edge across many regimes. The second is about which regime the market is in this quarter.
What this means for OpenAlpha
We deliberately built OpenAlpha around 6 strategies covering both trend and mean-reversion exactly because no single approach works in all markets. Looking at the Performance page today and concluding 'pair trading is broken' would be like looking at a weather forecast in July and concluding 'snow shovels are useless equipment.' They're useless right now. They will not be useless in February.
If you're using OpenAlpha to follow signals, the rational allocation right now is to overweight TSMOM and XGBoost. If you're using it to learn quant trading, the rational use right now is to study why mean reversion is failing — that understanding becomes valuable the moment volatility returns.
When will mean reversion come back?
We don't know — and anyone who tells you they do know is selling something. But here's what we'd watch for: rising realized volatility (currently sub-10% on EUR/USD), widening cross-asset correlations breaking down, and central banks losing the ability to anchor expectations. When those conditions reverse, the strategies that look 'broken' today will start firing again. We'll be ready.
Like this kind of analysis? Upgrade to Learner for weekly articles, real-time signals, and full educational content.
See Learner plan