If you wanna run a stock's intraday data through an algorithm to identify a few "well-known" technical patterns such as H&S or rounding top, then apparently you need to define the mathematical/statistical properties of them. Given they are rather abstract or loosely defined and scale-invariant, how does (or can) one approach such an exercise? Any pointer is much appreciated.
they are visually "discovered" after they happen. maybe you can compute local maxima/minima over a period and see if peaks/troughs are within a certain threshold level of one another
the scale-invariant nature puzzles me when brainstorming . it is not constrained to a specific period of data
These patterns come from higher time frames. H&S on the daily/weekly is a monthly pivot Daily flags are inside bars from weeklies or monthlies. It goes on and on like that.
HS pattern is extremely popular. Newbies are attracted to it. But oldies have different thinking. All oldies were newbies before. One day, after you have thrown away the Head and Shoulder book into the burning Mount of Doom, you might start to make progress in your trading. Meanwhile, Good Luck exploring the SHS thing.
It doesn't matter, man. The old TA patterns mean nothing in this latest round of Fed+Prez shit. Time for you to embrace FA.
You can transform the data using regression. If you difference the data against a trend estimate, and then difference that against the integral of the transform, you can pinpoint your so called "rounding top" formation. That one is related to the second derivative of price. The h&s one is essentially just trend analysis of local boundaries applied to price.