Fibonacci Retracement 38/61: The Best of Ten Rules

A retracement into the zone between 38 and 61 percent of a prior move is treated as an entry point with the trend. We formalise this as the position of price within the range of the last 20 bars, filtered by direction against EMA(200).

Fibonacci pullback: how it was formalised

In an uptrend defined by EMA(200), go long when price sits in the middle portion of the last 20 bars' range; in a downtrend, the mirror image in short. Outside that zone there is no position, so the rule spends roughly an eighth of its time in the market.

The best result of the ten rules tested

Run with the portal's own engine over 518 daily bars for each of 12 pairs, costs of 1 pip round turn, canonical parameters. The method and its limitations are set out in how we run backtests.

• Median return — 2.44%, profitable on 8 of 12 pairs
• Sharpe — 0.59, drawdown — 2.4%
• Trades — 35, win rate — 25%, time at risk — 13%
• Extremes: GBPJPY +6.5%, NZDUSD -6.2%

What worked was staying out, not the Fibonacci numbers

This is the strongest result among the ten rules: a positive median, eight profitable pairs out of twelve, the highest Sharpe ratio and — most importantly — the smallest drawdown. The reason for the last of these is simple and has nothing to do with the magic of Fibonacci numbers: the rule is out of the market almost all the time. Less exposure means less risk, and the rare entries occur when price is not at an extreme.

What was actually tested here

Do not file this as confirmation of Fibonacci levels. We did not test the coefficients themselves but entry in the middle portion of a range in the direction of the trend — that is, a refusal to buy at the high. That is most likely what works, rather than the specific numbers: boundaries of 40 and 60 percent would probably produce a similar outcome. And crucially, even the best of the ten rules did not beat simply holding the position. For comparison: simply holding the position on the same data returned 4.79%, and this rule did not beat it — why a benchmark is not optional.

A caveat that matters more for the best result

The best result needs a stricter caveat than the worst ones. This rule was selected as the best of ten tested, and the best of a set is biased upward by construction. Treating this figure as an expected return would be a mistake — the mechanism is explained in our piece on data mining bias. The general limitations — a short sample, correlated pairs, a signal tested without risk management — are listed in the methodology. The data is open and the run reproduces from a script in the repository.

This material is educational and is not individual investment advice. Backtested results do not guarantee similar results in the future. Trading forex carries the risk of losing capital.