Mean reversion trading strategy: how it works and where it fails

What a mean reversion strategy is, the four parts every one needs, where it breaks (trends, averaging, no stop) and how to backtest it without fooling yourself.

On this page

A mean reversion strategy bets that price, after stretching far from its average, tends to come back. You sell the stretch up, buy the stretch down, and take profit on the way back. It works until a stretch turns into a trend, and that's the whole risk.

I trade this way myself, so below is the version I'd have wanted before I started. No returns are quoted here. Past results, mine included, don't predict anything.

The four parts

Every mean reversion system, from a ten-line script to a full strategy, answers the same four questions:

  1. What's the mean? A moving average, a smoothed price, VWAP, a regression line. The choice matters less than keeping it fixed, so you can say afterwards what "far" meant.
  2. How far is far? A distance from the mean, measured in standard deviations, in average true range, or with an oscillator like RSI or Bollinger bands. This is the entry threshold. Too close and you trade noise. Too far and you trade twice a month.
  3. When is the trend not your enemy? A filter. Mean reversion loses money in strong trends, so a decent system checks the direction on a higher timeframe and skips stretches that go with it.
  4. How do you exit? At the mean, at a fixed target, on a time limit or on a stop. This is the part people skip, and it decides how the drawdown looks.

In Meanward Revert, these are settings you can read: the mean line and two bands in the channel (period 200, band multipliers 1.0 and 2.415 by default), a higher-timeframe Supertrend as the direction filter, and exits by target, channel line or opposite signal.

Where it breaks

Mean reversion has a particular shape: many small wins and a few deep drawdowns. Three things produce it.

  • Trends. Price that leaves the average and doesn't come back looks, for a long time, like a very good entry. The filter in part 3 reduces this and doesn't remove it.
  • Averaging down. Adding to a losing position raises the win rate and the exposure together. In the Revert Strategy the defaults take a 500 USDT first order to 8,500 USDT of position after two adds (source). If you run a cascade, size the account for the whole cascade on every symbol, not for the first order.
  • No stop. Without a stop, a trade that doesn't recover stays open, so the closed-trade win rate looks great while the open loss grows. The honest number is net result including open positions.

If a description of a mean reversion strategy says "90% win rate" and doesn't mention open positions or drawdown, close the tab.

How to backtest it without fooling yourself

These are the checks I use, and the ones I'd ask anyone to show before I believed their numbers:

  • Costs in. Commission on every order, including every add, plus slippage. My runs use 0.04% per order and 1 tick of slippage in the TradingView strategy tester.
  • Fill rule stated. Signal on the bar's close, order at the next bar's open. If a backtest fills at the close it signalled on, it's looking at the future.
  • Drawdown, not just net. Report the maximum drawdown with the number. For a strategy with open positions, the tester's closed-trade list isn't enough: mark open positions to market at every minute. How I do it is on Backtest methodology.
  • One account, not many. Running 78 symbols in the tester gives each its own 10,000 USDT. Add them up and you've invented 780,000 USDT. Put everything on one timeline and one capital.
  • Know what's missing. Funding on perpetuals, real fills, delisted symbols, and the fact that you kept the variant that looked best on this data. All of that flatters the result.

Then rerun it on your own pairs.

Which indicators to use

People ask for "the best indicator for mean reversion". There isn't one; there are tools for parts 1 and 2. Bollinger bands and RSI are the common ones, a channel built from a smoothed mean and smoothed range is another. Pick by what you can explain: if you can't say why an entry fired, you can't tell a bad market from a bad setting.

A simple way to start

  1. Pick one instrument and one timeframe. Not twenty.
  2. Write down the four answers before you open the tester.
  3. Run it with realistic costs and a fixed start date.
  4. Read the drawdown first, then the net result.
  5. Only then change one setting at a time.

Bottom line: mean reversion isn't a free edge, it's a bet with a known way to lose.

If you use TradingView, the Revert strategy is a ready version of the above with the settings exposed, and the risk presets show three sets of risk values to start from. It's a tool. Results depend on the hands.

Questions, and the odd failed backtest, go to the Telegram channel.

Not financial advice. Trading carries risk of loss.