Encyclopedia of Chart Patterns. Thomas N. Bulkowski

Characteristic Discussion Appearance A zigzag pattern that has four turns, two of which are governed by Fibonacci ratios. CB/AB retrace The retrace should be one of the Fibonacci numbers: .382, .5, .618, .707, .786, or .886. CD/CB extension The extension should be one of the Fibonacci numbers: 1.13, 1.27, 1.41, 1.618, 2, 2.24, 2.618, or 3.14. Hills and valleys From A to B, there should be no peak higher than A and no valley lower than B. From B to C, there should be no peak higher than C and no valley lower than B. From C to D, there should be no valley lower than D and no peak higher than C. Volume The volume trend provides a slight performance improvement depending on the bull or bear market condition. See Table 3.2. Duration Limited to 6 months, but that's an arbitrary value.

      Figure 3.2 Another example of a bullish AB=CD pattern with an upward breakout (when price closes above the top of the pattern).

CB/AB retrace. The BC retrace of AB should be one of the Fibonacci numbers listed in Table 3.1. Let's use the pattern shown in Figure 3.2 as an example of how this works (rather, how I wrote my software to find these patterns).

      The high price at A is 228.89, the low at B is 213.12, and the high at C is 225.45. The CB/AB ratio is (225.45 – 213.12)/(228.89 – 213.12) or .782, which is darn close to the .786 number in the table. It qualifies as a proper turn. My software looks for a number within a window .5% of the target numbers listed in the table.

      Once you know the ABC turn, you can determine what D should be priced at. In this example D = C – (A – B) or 209.68, which is close to the actual low at D of 209.62.

      CD/CB extension. Point D can also be found by the formula C – (C – B)/(Fibonacci number) or 225.45 – (225.45 – 213.12)/.786 in this example. The result is 209.76.

      Hills and valleys. The table explains the requirements of peaks and valleys between the various points in the pattern.

      Volume. Volume trends upward from points A to D most often, but the trend is almost random.

      Duration. I limited patterns to 6 months duration or less. This is an arbitrary limit.

Focus on Failures

      Figure 3.3 shows one example of how this pattern fails. The four points in the pattern are labeled as ABCD, and they comprise the bullish AB=CD pattern. Volume slopes upward in this example, shown by the diagonal line at E.

      Figure 3.3 This bullish AB=CD has price falling through the predicted turning at D.

Point D is where the stock is supposed to bottom and turn upward, but it doesn't. It's obvious that the height of CF is much taller than AB, but the height can vary depending on the Fibonacci numbers used in construction of the pattern. The turn at D should be at 40.05, and yet the stock continues lower.

      This type of failure happens often, 62% of the time in bull markets and 67% of the time in bear markets. Let me also say that this high failure rate may be due to the model I used. Your software may find patterns that perform better than the ones I found.

      Another way the pattern fails is if the stock doesn't make it down to D. It turns before the predicted target, leaving you waiting to buy the stock with a fist full of money. Fortunately, this type of failure is exceedingly rare.

      Finally, the pattern can also fail if it does make it down to D and it does turn at D, but the rise isn't high enough to make money. These types of failures are what I call 5% failures and describe them in Table 3.2 as the breakeven failure rate. Let's check the statistics to see what we can learn about this pattern.

Statistics

      Table 3.2 shows the first batch of statistics for the bullish AB=CD.

      Number found. I found over 2,300 patterns, sorted by market condition, in 1,069 stocks from July 1991 to February 2020. I told my program to limit the number found so it didn't overload my spreadsheet. Not all stocks covered the entire period, and some no longer trade.

      Breakeven failure rate. The failure rate (a measure of how many patterns failed to see price climb more than 5%) is quite good, averaging between 3% and 12% of those patterns that reached D and reversed there. To put it another way, if you were to trade a lot of these patterns perfectly, you'd have an 88% chance of seeing price rise more than 5% above the low at D. That's terrific.

Average rise after D. I measured the average rise from the low at D for those patterns reaching D and turning upward. The average rise is disappointing. Non‐Fibonacci‐based patterns in bull markets perform better, averaging 42.4%. However, bear markets beat non‐Fib patterns with the rise averaging 30.5% compared to 28.1%, respectively.

Table 3.2 General Statistics

Description	Bull Market	Bear Market
Number found	1,741	565
Breakeven failure rate	11.6%	3.7%
Average rise after D	38.4%	30.5%
Volume trend	53% Upward	57% Upward
Performance Up/Down volume	40%U, 37%D	33%U, 27%D

      Volume trend, performance. Volume trends upward slightly more often than downward, as the table shows. You can improve performance by looking at the volume trend. I used linear regression to determine if volume was trending upward or downward from the start to end of the pattern (A to D). Often you can just look at volume to see the trend.

      Both markets see an improvement in performance if volume trends higher. To put it another way, a downward volume trend hurts performance.

Trading Tactics

      Imagine if you found a pattern that could predict when the stock is going to turn. That's what I had hoped for this pattern. Once you know that the stock had bottomed, you could buy