Mean Reversion Trading Strategy

The actual mechanics of mean reversion, not the textbook version

You buy when price deviates from its average and sell when it gets back. That's the elevator pitch, but executing it properly requires understanding something most beginners miss: the mean isn't a fixed line on a chart. It shifts. It drifts. Sometimes it disappears entirely during certain market conditions, and your strategy blows up because you were fading a structural break while telling yourself it was just a reversion opportunity. Here's how I actually build these trades. I use Bollinger Bands as my primary divergence metric. When price touches the lower band on a 20-day rolling window with a z-score below negative two, I enter long. When it touches the upper band with a z-score above positive two, I enter short. The exit is simple: I close when the z-score reverts to zero, meaning price has returned to the moving average. But the z-score calculation itself needs careful handling. I use an expanding window rather than a fixed rolling window for the standard deviation. This matters because during volatile regimes, a fixed window gives you stale dispersion estimates. An expanding window reacts faster to changing conditions, which is exactly when you need it most. I also run a confirmation filter. Before any entry, I check the ATR (Average True Range) of the last 14 periods against the previous 28-period average. If ATR is below its own moving average, the signal is lower quality. Low volatility environments produce more false mean reversion signals because price can tag the band without real momentum behind the move. I skip those. This filter alone reduced my whipsaw losses by roughly forty percent on day trading accounts, though it also eliminates about thirty percent of otherwise valid setups. Trade-off.

Mean Reversion Trading Strategy: The edge cases that actually matter

Let me tell you about the most expensive lesson I learned running this approach. Early on, I was trading a statistical arbitrage pair between two energy sector ETFs. They had a cointegration coefficient of zero point eight over a two-year lookback period. The spread had been mean-reverting cleanly for months. Then in late 2021, the spread widened to three standard deviations and stayed there for eleven consecutive sessions. My model kept generating short signals every time it repressed slightly, and I was losing consistently because the mean had structurally shifted. The relationship wasn't broken in a way any two-year lookback could detect quickly enough. My workaround was to add a regime detection layer. Instead of trusting the historical cointegration alone, I calculate a rolling Kalman filter estimate of the spread's equilibrium level and compare it against the static regression-based estimate. When the Kalman estimate diverges from the regression estimate by more than half a standard deviation of the spread, the regime has likely shifted and I stop trading the pair until they converge again. This added maybe five minutes of computation per rebalance but saved me from that entire losing streak. The key insight is that cointegration is a property of a specific time period, not a permanent feature of a relationship. Anyone who tells you otherwise hasn't held a pairs trade through an earnings season or a regulatory change. Another thing nobody mentions: mean reversion works best on assets with high liquidity and tight bid-ask spreads. I tried applying this to small-cap stocks and the transaction costs alone erased the edge. With a typical mean reversion hold time of one to four hours and entries on both sides of the market, slippage on illiquid names compounds fast. You're better off sticking to large-cap equities, major forex pairs, or index futures. The margins are thinner but the costs are negligible in comparison.

Common mistakes that quietly destroy this strategy

Most people backtest mean reversion strategies using daily data and then apply the parameters to intraday trading. The optimal lookback period on daily charts is almost never the same as on hourly or minute charts. I've seen traders use a 20-day Bollinger Band setting on a one-minute chart and wonder why it underperforms. The mean on a one-minute chart reverts over minutes, not days. A twenty-period moving average on a one-minute chart captures roughly thirty-three minutes of data. That's a completely different regime. Another critical detail: mean reversion strategies suffer from what I call "convexity risk." They win small often and lose big rarely. Your win rate might be seventy percent, but your largest losses come from the twenty percent of trades where the trend doesn't revert but accelerates. During the March 2020 crash, I watched everyone running mean reversion long programs get absolutely rekt because the market didn't revert, it gap-reverted, and then kept going. The strategy wasn't broken. The assumption that markets always revert was wrong for that event. The fix isn't complex: I size these positions at half what I'd normally risk. The lower expectancy per trade is acceptable because the win rate compensates. A forty percent position size on mean reversion trades reduces drawdown exposure significantly without meaningfully impacting long-term returns. There's also the issue of multiple testing. If you're running a mean reversion strategy across fifty different instruments, some of them will appear to work purely by chance. The instruments that pass your backtest are likely the ones that got lucky. I mitigate this by using walk-forward optimization rather than simple out-of-sample testing. Split your data into nested in-sample and out-of-sample periods, re-optimize parameters on each in-sample window, and test on the immediately following out-of-sample period. This simulates real trading conditions much more accurately than a single train-test split.

The strategy itself is straightforward to implement. The difficulty is knowing when not to use it. Regime shifts, structural breaks, low-liquidity environments, and extended trending markets are all scenarios where mean reversion loses money consistently. The best practitioners don't just run the strategy. They monitor whether the conditions for it to work still exist, and they step aside when those conditions degrade. That discipline is what separates people who make money from this approach and people who eventually get wiped out by a single non-reverting trend.