Mean Reversion: How to Build the System

Mean reversion assumes price oscillates around a level and that excessive deviations contract. The approach formalises better than most and tests more honestly — which is why its limits are also clearer.

Three decisions that define the system

First, what counts as the mean: a moving average, yesterday's close, VWAP, the centre of a range. Second, how to measure deviation: in pips, in standard deviations, in units of ATR or as position within a range. Third, what deviation is sufficient to enter. Our Bollinger measurement shows how sensitive this is: about 10 percent of bars exit two sigma and about 3 percent exit 2.5, so the threshold changes trade count threefold (more on settings).

What our test showed

The simplest formulation — entry after three one-directional bars — produced a median of 0.27%, profitable on 6 of 12 pairs, Sharpe 0.07, a result indistinguishable from chance at a very low drawdown. A more structured variant, entering the middle portion of a range in the direction of the trend, produced the best result of the ten rules tested: a median of 2.44%, profitable on 8 of 12 pairs, Sharpe 0.59. The difference between them is the trend filter, and that is probably the main conclusion: pure mean reversion gave no edge, mean reversion with a direction filter did.

The regime filter is a condition, not an improvement

Mean reversion assumes the market is oscillating. In a trend that assumption is false, and every trade against the move loses. A filter identifying market state is therefore not an add-on: without it the system trades in conditions it was not designed for. Practical filter options include position relative to a long-term average, channel width, or a trend strength measure.

Stops in a mean reversion system

The stop does not go where a trend system would place it. The entry logic assumes price has travelled too far, so the stop must sit beyond what you consider a reasonable deviation — otherwise it triggers on normal fluctuation. The distance is therefore substantial, and it must be compensated by reducing size rather than by tightening the stop. Attempting a tight stop in such a system contradicts its own premise.

Where the approach breaks

At the moment the regime changes, and it happens without warning. A system that earned steadily in a range begins to lose when a trend starts, and it will lose on every consecutive trade, because it enters against the move. A limit on consecutive losses after which trading halts is therefore especially important here: it is the only mechanism that notices a regime change before it becomes too expensive.

This material is educational and is not individual investment advice. Trading forex carries the risk of losing capital.