Clicking on my books below takes you to Amazon.com. At no cost to you, they pay for the referral and that helps support this site.

Bulkowski’s Overlap Breakout

Written and copyright © 2008 by Thomas N. Bulkowski. All rights reserved.

Summary

This page discusses new research I conducted concerning the price trend immediately before the breakout from a chart pattern. Does price forming a congestion zone up to five days before the breakout result in better performance than price trending in a straight-line run? The answer I found comes in three parts. First, the average rise after the breakout tends to decrease the more overlap (congestion) forming before the breakout. The gain drops from 43% for straight-line runs (0% to 4.9% overlap) to a low of 32% when price overlaps between 65% and 69.9%. Failures are essentially flat, bouncing between 5% and 6% for much of the time when samples are high (above 30) but does tend to favor higher failures for straight-line runs. Finally, the occurrence of throwbacks decreases from 54% for straight-line runs (starting with overlap of 10% to 14.9%) to 41% for overlaps of 70% to 74.9%, both ranges keep the samples above 30.

In short, a straight-line run leading to the breakout gives a higher gain at slightly higher risk of failure with a higher chance of throwback.

Introduction

The chart shows two pictures on the daily scale. The top one is a descending triangle in Applied Magnetics, a stock which no longer trades. A descending triangle forms when price bounces between a down-sloping top trendline and a horizontal -- or nearly horizontal -- bottom trendline. Multiple touches of each trendline is best.

This one is interesting because of the horizontal price movement just before price closes above the top trendline. Such tight price movement (circled) appearing several times in the data series suggests that the stock is thinly traded, and price bounces between the bid-ask spread. The chart shows price before decimalization, so the minimum spread was often 1/8 of a point or even higher. For a stock trading under $3, that is a large move (nearly 5%) just to cover the spread.

What is important about this chart is the tight consolidation region just before the breakout. Price showed 100% overlap, meaning the high-low range from day to day had complete overlap. Such overlap is rare, occurring just 15 times in over 19,000 samples.

The second chart shows a bump-and-run reversal bottom chart pattern, one many of you are probably not familiar with, in the stock of Advanced Micro Devices. Barr bottoms show the effect of downward momentum. At the start of the pattern, price moves lower following a trendline in the lead-in phase of the pattern. Then comes the bump phase where a bad earnings report or other news is usually the cause of price making a swift decline. Then price begins to recover and pushes through the down-sloping trendline, shown in red.

When price closes above the trendline, the stock stages a breakout. This example shows a throwback that takes price back down to the trendline before it climbs away again. Notice the straight-line run (circled) leading to the breakout. Looking at the five days before the breakout shows that price has just 20% overlap between the price bars. This is a low value and it usually represents price trending. That is what we see here.

In the top chart we see a congestion region before the breakout, and in the lower chart, price is already trending. Which represents a better trading setup? This page answers that question.

Methodology

To solve the riddle of which is better, a straight-line run or congestion before the breakout, I mined my database of chart patterns starting from July 1991 to the current day, May 7, 2008. Although the period covers both bull and bear markets (a bear market occurred in the S&P 500 from March 24, 2000, to October 10, 2002 and a bull market is everything else), I used only the bull market data and upward breakouts from chart patterns.

I included 36 chart patterns in the test. Here is the complete list: broadening patterns of all types: tops/bottoms, right-angled and ascending/descending, rising and falling broadening wedges, bump-and-run reversal tops/bottoms, diamond tops/bottoms, all combinations of Adam and Eve double tops/bottoms, rising and falling wedges, head-and-shoulders tops/bottoms simple and complex, rectangle tops/bottoms, rounding tops/bottoms, ascending/descending/inverted scallops, ascending/descending/symmetrical triangles, and triple tops/bottoms.

These patterns found 19,067 chart patterns, with no duplicates allowed, using three databases and 867 unique stocks. Not all stocks covered the entire 1991 to 2008 range. Limiting this test to the bull market with upward breakouts gave 7,721 chart patterns that qualified.

For the test, I used the 6 days before the breakout to measure the overlap between prices. Let’s say that the breakout occurs on Friday. I computed the highest high (C in the picture) and lowest low (B) for Wednesday and Thursday. That gave the high-low range between the two days. Then I measured the price difference lowest high (A) and highest low (D) between the two days. This (AD) is the region of overlap between the two days. I divided the two (overlap/HL range) to get a percentage of overlap for the entire range. If the numbers were negative, then I used 0 as the overlap percentage (such as in the case of a price gap). Then I moved to the prior day and computed again, Tuesday and Wednesday, and so on until I had five values (using 6 days). I averaged the numbers to get a final overlap percentage which I tracked for each chart pattern.

The chart patterns and databases I used included only known good chart patterns, ones I used for my books on chart patterns. They measure the move from the breakout to the ultimate high. On the day of breakout, the lowest price is used because it is often closer to the trendline than other methods (such as the closing price) and helps compensate when price gaps up from the prior day. The ultimate high is the highest high before price drops at least 20%, measured from highest high to close. This drop represents a trend change or end of trend. It is the best (maximum) performance that one would hope to achieve. It is also unrealistic, but I am not using it as a gauge of what you can expect to make, but rather as a comparison between overlap values.

Warning! Let me discuss the flaws in this study. Even though I used thousands of samples, the bins into which I placed the results sometimes showed few samples (as the coming table shows). Thus, conclusions on data having few samples should be avoided. Also, what is meant by price having a 50% overlap in 5 days? Price might not have any overlap for a few days followed by 100% overlap, giving an average of 50%. Thus, determining what is meant by a straight-line run or congestion in percentage terms can be misleading. An overlap of 5% or less does mean a straight-line run and 100% overlap means a solid block of congestion, but between those extremes the definitions become fuzzy. Keep that in mind when I discuss the results (next).

Results

The following table shows a frequency distribution of price overlap from 0% up to 99.9%. Instead of including the x.9% values, I just rounded up after the first row. Thus, the 5% row really includes 5% to 9.9% and so on.

Let’s take an example using the third row. This covers all chart patterns with price overlapping between 10% to 14.9%. The average percentage gain of the 145 chart patterns was 33%. Six percent, or 8 patterns, failed to climb at least 5% after the breakout. Just over half, 54% (78 patterns) saw price throwback to the pattern boundary or come close enough within a month after the breakout. Read the other rows in a similar manner.

The top of the table shows price that has almost no overlap leading to the breakout (a straight-line run) whereas the bottom of the table shows full overlap (congestion or consolidation) in the 5 days leading to the breakout.

Percentage Rise

The adjacent chart shows the percentage rise after the breakout when mapped against the overlap. The first point on the left corresponds to the 43% rise after a straight-line run leading to the breakout. At the far right shows the spike rise to 52% (at 75% overlap) and the drop back to 46% for the 80% overlap. As the table shows, sample counts less than 30 are unreliable. Ignoring the low-sample count points, we can see that the trend is downward, from a start of 43% (5% overlap) to 38% average rise for overlaps of 70%. In other words, as the overlap increases, the average gain decreases. A congestion area appearing before the breakout tends to impede the average rise.

5% Failure Rate

The adjacent chart shows the failure rate by the percentage of overlap. For the first point, the average rise is 43% but the failure rate is 12%. That means 12% of the chart patterns had price climb less than 5% after the breakout before either closing below the opposite side of the chart pattern or declining by at least 20%. The 12% number only uses 7 samples, so do not trust the result.

If you ignore the low sample counts at each end of the chart, the failure rate bounces between 5% and 6%, but shows a slight tendency to drop as congestion (overlap) increases.

Throwback Percentage

Before I discuss the adjacent chart, using the numbers from the table for which samples are above 30, throwbacks occur more often after a straight-line run than after a congestion region.

The chart show the percentage of throwbacks that occur as the overlap increases. This time, I split the samples into four chunks, 0% to 24.9%, 25% to 49.9%, 50% to 74.9% and 75% to 99.9%. A throwback has price return to the breakout price or near the trendline boundary within a month after the breakout. The chart shows that as the overlap increases, meaning more congestion before the breakout, the number of throwbacks decreases until the low sample count shoots the numbers up (the rightmost column). Now look at the left scale. It varies between 54.7% and 51.7%. The difference between the two is small.

Why the focus on throwbacks? Two reasons. First, chart patterns that have throwbacks do not perform as well as those without throwbacks. Read the study that discusses this result. I think of it as a loss of momentum that occurs when price turns lower after the breakout before resuming the upward move. The second reason is that throwbacks increase the risk of failure. If you are late getting into the trade or even on time, a throwback could drop price well into the body of the chart pattern or even below the bottom of it. Such a decline, in percentage terms, could be huge. A properly placed stop loss would minimize the risk, but how will you feel when price resumes climbing again? In short, throwbacks are a bad thing unless you want to buy into a setup after the throwback completes.

More Results

Instead of the table showing bin sizes of 5%, I used the 25% size. The first bin ranged from overlaps of 0% to 24.9%. The second ranged from 25% to 49.9%, and so on. I found that the average rise held steady at 36% for overlaps from 0% to 74.9% (covering 3 bins) and the failure rate held constant at 6% but the throwback percentage dropped from 54.2% to 51.7%, all using the 0% to 74.9% overlap range. Only the highest overlap (95 samples versus 1,102 to 4,862 samples for the other bins) had the average gain rise to 50%, failures drop to 3% (3 samples) and the throwback rate climb to 54.7% (52 samples). Since sample counts are few for the highest bin, I think this is anomalous. For the other three bins, more congestion before the breakout reduces the throwback frequency, which is good.

Finally, I used the median overlap percentage of 40% (which is also the average overlap percentage) and split the information into two pieces: Those patterns with overlap less than or equal to the median and those above the median. I found that the average rise was 37% for patterns at or below the median overlap and 36% for those above the median. The failure rate was 51% to 49% and the throwback rate was also 51% to 49%, respectively, for the two groups. A near tie, in other words with straight-line runs showing post-breakout performance slightly worse than congestion areas.

Trading

What does all of this mean? Given that the overlap percentage method may not correctly separate a congestion region from a straight-line run, I would err on the side of common sense. It makes sense to me that a congestion area before the breakout is the smarter play, if nothing else than to reduce the throwback rate. I hate suffering a loss after a throwback takes price down too far. Even though the percentage differences between the straight-line run and congestion groups is not large, it will probably pay to look for a congestion area just below the breakout when shopping for chart patterns. If there is no congestion area, then consider waiting for a throwback to complete before taking a position.