Thomas Bulkowski’s statistics on Technical Patterns

[cassiopeia]

Has anyone bought Bulkowski's (expensive) book and conducted a back-test on his patterns? I'm sure with the advent of AI and pattern recognition systems it should be possible to do this. What conditions does he place on a defined pattern, does there need to be an uplift prior to the pattern?
He claims that for H&S patterns the average price drop was as much as 23 percent and out of a sample of 431, and only 30 gave false signals where the stock fell less than five percent below the neckline to then reverse upward. These figures are difficult to believe.

A search of the technical literature suggests that some patterns do exhibit excess returns, although they are not as reliable as Bulkowski suggests. For example Savin et al states

We find little or no support for the profitability of a stand-alone [Head and Shoulders] trading strategy. But we do find strong evidence that the pattern had power to predict excess returns. Risk-adjusted excess returns to a trading strategy conditioned on “head-and-shoulders” price patterns are 5–7% per year.

whilst Ahlawat, Samit claims

no pattern produces statistically and economically significant profits for a cross-section of stocks and indices analyzed, though a few patterns are more successful predictors.

Perhaps if anyone else can find any peer reviewed technical papers on this subject, they could post them here.

 

[tomorton]

Great idea, someone with an equivalent amount of discipline as Bulkowski should update on these patterns.

Having said that, I no longer put a lot of store in chart or candlestick patterns. I used to, and Bulkowski was the author of the bibles on these. What changed was I realised that regardless of entry pattern my trades went really well if price entered a trend in my direction. The odds of that happeneing from a reversal are low: but they're much higher if price was already moving in a pre-existing trend. Which of course took me towards continuation patterns. Which in time took me to the realisation that you don't need a continuation pattern, just any transient weakness in the trend.

 

[cassiopeia]

My limited analysis going through the last 7 years consisting of about a hundred H&S bottoms and tops suggests a success rate somewhere between what Bulkowski and Samit suggests. Bear in mind Bulkowski uses a 5% price change as a definition of success, which is a very low bar.

It would be far more useful to use a mathematical algorithm to select H&S patterns to avoid subjectivity and bias. Is anyone aware of an H&S algorithm or published H&S backtest data using an algorithm? I know one of the higher tradingview subscriptions provides identification of technical patterns, but does this only include recent patterns, or patterns in formation? To assess them we would need a full list of past H&Ss over the past 5 years or so. From that list, it may be be necessary to filter out certain characteristics, because indistinct H&S patterns are forming all the time.