Harmonic Patterns: Gartley, Bat and Butterfly, Counted
Harmonic patterns are five-point price structures, labeled X, A, B, C and D, whose swings must match set Fibonacci ratios, such as the Gartley, Bat and Butterfly. The ratios fix where point D should form, and that level is read as a place where the move may reverse.
Covered on this page: TradingView.
Harmonic patterns are five-point price structures whose swings must match fixed Fibonacci ratios, and the best known are the Gartley, the Bat and the Butterfly. They are drawn by eye on most charts. This page writes the ratios down as code, runs them over every swing on the S&P 500 since 1970, and counts how often the textbook shapes appear at all.
How it forms
Five swing points, four legs. X is the start. XA is the first and largest leg. AB pulls back part of XA, BC pulls back part of AB, and CD runs to the point the ratios predict. A bullish pattern starts at a low X and ends at a low D; a bearish one is the mirror.
The ratios separate the patterns. These come from Scott Carney’s HarmonicTrader.com pattern pages, read on 3 October 2026 in the Internet Archive’s copies of August 2025, because the live site refused automated reading. They are his ratios as his own pages stated them then:
| pattern | B, as a share of XA | D, as a share of XA from A |
|---|---|---|
| Gartley | 0.618 | 0.786 (not beyond X) |
| Bat | under 0.618, preferably 0.50 or 0.382 | 0.886 |
| Butterfly | 0.786 | 1.27 (beyond X) |
The Gartley’s name is older than its ratios. Carney’s page says the shape appears in H.M. Gartley’s 1935 book Profits in the Stock Market, but that the book gave no Fibonacci ratios; Carney assigned the 0.618 and 0.786 in The Harmonic Trader (1998). He describes the Bat, which he dates to 2001, as “probably the most accurate pattern” in his arsenal. With one Bat found below, this page cannot test that claim either way.
His pages add conditions this page does not test. Each pattern should contain an AB=CD, the ABCD pattern in its last four points, and a set projection of the BC leg, at least 1.618 for the Bat. Carney calls his ratios Fibonacci ratios. His AB=CD page puts C between 0.382 and 0.886 of AB, and that is the C range used below.
The rule used here
The swing points come from a fixed rule, the same one the ABCD page uses. A swing high is a day whose high is above the highs of the five sessions before and after it; a swing low is the mirror with lows. Swings then alternate: when two highs come in a row, the higher one is kept, and the same for lows. That gives 1,349 alternating swing highs and lows, the same idea as a zigzag indicator, on the S&P 500 from 30 January 1970 to 16 September 2026, and 1,345 runs of five in a row to test.
A run of five counts as a pattern when all three ratios fit:
- B: AB divided by XA is within 0.03 of the pattern’s B ratio (for the Bat, between 0.352 and 0.53).
- C: BC divided by AB is between 0.382 and 0.886.
- D: the distance from A to D divided by XA is within 0.03 of the pattern’s D ratio.
For the Bat, the 0.382 to 0.50 band is also this page’s reading of Carney’s “preferably a 0.50 or 0.382”. Because the rule leaves out his AB=CD and BC-projection conditions, it is looser than his, and a stricter version could only find fewer patterns.
What counts as a reversal. The next swing after D is E. The move from D to E is measured as a share of AD, and the page counts how often it reached 38.2% and 61.8% of AD, two common retracement targets after D. The comparison is every run of five whose D sits on the A-to-X side of A, as it does in all three patterns: 1,188 of the 1,344 runs with a next swing. In the other 156, D lies beyond A, so the move to E cannot be a reversal of AD.
A worked example
August 2026 drew a Bat setup on the S&P 500, by this page’s rule. The swing points, all from daily highs and lows:
- X: a swing low of 7,313.92 on 29 July 2026.
- A: a swing high of 7,816.70 on 13 August. XA is 502.78 points.
- B: a swing low of 7,638.17 on 24 August. AB is 178.53 points, 0.355 of XA: just under the Bat’s preferred 0.382, and inside this page’s tolerance.
- C: a swing high of 7,771.48 on 28 August. BC is 133.31 points, 0.747 of AB.
The Bat put D at 0.886 of XA below A: 7,816.70 minus 0.886 times 502.78, which is 7,371.24, with the page’s tolerance spanning 7,356.15 to 7,386.32.
Price never got there. The index turned at a swing low of 7,507.77 on 16 September, 0.614 of XA, 121.45 points above the top of that range, and did not trade lower through 25 September, the last day in this data. It closed at 7,743.41 that day. A trader waiting for the Bat’s D never got the setup, and the turn that did come matched none of the three patterns’ D ratios.
The original data
The textbook shapes are rare. Of 1,345 runs of five swing points from 1970 to 2026, three met a pattern’s ratios within 0.03: a bearish Gartley from 27 April to 28 May 1981, a bullish Bat from 20 October to 4 December 1987, and a bearish Butterfly from 13 April to 1 June 1993. That is 0.22% of sequences, and none in the 33 years since.
Three is too few to measure anything. Two of the three reversed by 61.8% of AD, against 67.26% of the 1,188 comparable sequences (799 of them). By 38.2% of AD, all three did, against 83.92% (997).
Loosening the tolerance adds patterns, not an edge.
| tolerance | patterns found | share of all runs | reversed 61.8% of AD | p against 67.26% |
|---|---|---|---|---|
| 0.03 | 3 | 0.22% | 2 of 3, 66.67% | 0.98 |
| 0.05 | 8 | 0.59% | 5 of 8, 62.50% | 0.77 |
| 0.08 | 16 | 1.19% | 10 of 16, 62.50% | 0.69 |
| 0.10 | 21 | 1.56% | 13 of 21, 61.90% | 0.60 |
Every row is within chance, and all four rates sit below the comparison rate. At a tolerance of 0.10 the B bands of the Gartley and the Bat overlap, so a run that fits both is counted once.
Most setups never complete. A trader sees B and C first, and only learns later where D lands. Among runs whose B and C fit within 0.03:
- Gartley-shaped (15): D landed in the Gartley range once, stopped short 3 times and ran past 11 times.
- Bat-shaped (23): once in range, 20 short, 2 past.
- Butterfly-shaped (16): once in range, 12 short, 3 past.
So 51 of 54 setups that looked right at C did not finish where the pattern said. The finished patterns are what the textbooks show; the unfinished ones are what a trader watching live mostly gets. Every matched sequence at 0.03 is in the harmonic patterns file, with all five points and ratios.
Reading D as a level, not a promise
The D level is a calculation from three earlier swings, so it is known before price arrives. That is its appeal: a precise price, worked out in advance, with X as a natural line for being wrong.
But a precise level is not a precise outcome. In the counts above, the patterns that did complete reversed no more often than comparable swings did. The precision is in the drawing.
Timeframe and swing rule change everything. A tighter swing rule finds more, smaller swings and more candidate patterns; a looser one finds fewer. Results from one rule do not carry over to another, which is why this page states its rule before counting.
When it fails
D never arrives. Most setups turn before D or run through it, as 51 of 54 did here. A plan that only works if D is reached is a plan for a minority of cases.
The pattern is fitted afterwards. Any five swings can be relabeled until some ratio fits within a loose enough tolerance. The tolerance table shows the count rising sevenfold from 0.03 to 0.10 with no change in what followed.
The ratios move between sources. Carney’s own pages note that others assign different Fibonacci numbers to the Gartley’s B and D. Two traders can call the same chart a Gartley and a Bat.
Price runs through X. Carney’s Gartley page says D should not go beyond X, and on a fast market price can run through X in a single session, past any order resting at that level.
Related
The ABCD pattern is the four-point shape inside every harmonic, tested on the same swings. Fibonacci covers the retracement ratios the patterns are built from. And swing highs and lows explains the turning points that every letter here sits on.
If you trade a harmonic level, write down the ratio rule and the tolerance before the pattern finishes, and count the setups that never reached D as well as the ones that did. Treat the D level as a place to watch, sized as if it can be run straight through.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money. Some links on this page earn a commission if you buy through them. It costs you nothing and it does not decide what appears here or in what order — how these pages are made is set out in our methodology.