One of the things that drives me crazy about most indicators is that they’re based on averages which as many of you know are incredibly laggy, cloak patterns and generally don’t have much nuance. See https://blog.revolutionanalytics.com/2017/05/the-datasaurus-dozen.html if you’re doubtful.
So I was wondering if people had experimented with plugging in histograms as a feature for their model. For instance it might divide the prices from the last N periods into M buckets, and increment each bucket once every time the close was in that range. That might help the model be more accurate.
Anyone try this or similar ideas to get away from the peril of aggregates?
Submitted October 07, 2020 at 01:11PM by silverbugoutbag