Harmonic Patterns, An Honest Look

[Fairman]

Harmonic Patterns, An Honest Look

Harmonic patterns, like the Gartley, Bat, Butterfly, and Crab, use specific Fibonacci ratios to define potential reversal zones. They look precise, which is part of their appeal.

diagram.png (35.02 KiB) Viewed 51 times

How they work: traders identify a sequence of swings (X, A, B, C, D) and check whether the retracements and extensions match set ratios. The D point is the potential reversal zone.

Points in favor:

- They give clear entry, stop, and target levels
- They encourage patience until a pattern completes
- They can combine with other confluence tools

Points of caution:

- Identifying patterns is subjective, and many charts contain several possible patterns
- Traders may force patterns that nearly fit
- Many apparent patterns fail
- Backtesting is difficult unless you apply strict rules

If you want to explore them:

1. Learn one pattern deeply, rather than all of them
2. Define exact tolerance for ratios
3. Require confluence: structure, liquidity sweep, higher-timeframe bias
4. Use strict stop losses beyond the D point
5. Track your results across at least 30 trades

Remember that the market doesn't know about your ratios. Any pattern is only useful to the extent that other participants also react to those levels or that there's another reason behind them.

Trade evidence, not appearances.

[LondonNewsTrader]

Your last paragraph is the one I'd underline. Fibonacci ratios only matter to the extent that enough people act on them, and on currency pairs that is rarely the main driver of a reversal. On news days it is even less so. If a D point happens to sit at the level price reaches on a CPI surprise, the reversal from there is the market digesting the number, and the harmonic label gets attached afterwards.

The tolerance point is where most people go wrong. If a Gartley's B point can be anywhere from 0.55 to 0.70, nearly any chart produces a pattern, and the backtest results become meaningless. Writing down an exact tolerance before looking at the data is the only way to know whether it works.

I'd also check whether any of the patterns in your 30-trade sample completed on a release. If they did, I'd log them separately.

[FXS Prop Desk]

Harmonic patterns look precise, which is exactly why their subjectivity is easy to miss. Two ways to reduce it: use a tolerance range for each ratio (for example ±5% around the textbook value) and decide that range in advance, and require the D point to coincide with an independent level, such as a previous swing high or a session extreme.

Patterns that only "work" when ratios are stretched to fit are usually curve fitting. Requiring the extra level confluence also cuts the number of trades, which makes each one easier to review properly.

[Shadow Trader]

The plus or minus 5 percent FXS Prop Desk suggests needs one more decision before it means anything. Five percent of the ratio itself puts a Gartley B between 0.587 and 0.649, while five points either side gives 0.568 to 0.668, which is about a third wider.

I tried both on a year of EURUSD H1 with a simple scanner. The wider version found a little over twice as many Gartleys, the same effect LondonNewsTrader warned about with the 0.55 to 0.70 example.

The more useful test was running the scanner on random walk data with the same volatility. It still found patterns, about 60 percent as many as on real prices, and the reaction rate at D was not far from the real one.

That pushed me toward the independent level as the actual filter. Once D had to sit within a few pips of a prior swing or a session extreme, the count fell to around one a fortnight, and I could no longer tell whether the pattern or the level was doing the work.

At that rate Fairman's 30 trade sample takes most of a year on one pair. Is anyone running identical rules across several pairs to get there faster, or does that bring its own problems?