Do the technical analysis features that day-trading course sellers actually work? Maybe. Here's what I've found:
"Technical Analysis" indicators and patterns like the head and shoulders and others typically actually reproducibly emerge from the market from pure randomness (I saw this in a youtube video) which gave me an idea.
The thing is it seems like these patterns emerge as a symptom or accidental result of underlying trade flow rather than constituting edge but I ran a study on this anyway. The factors I studied here are called "Orderbook Memory" factors. The idea is that you can actually draw the typical trading patterns that the gurus teach simply from supports and resistances. And these support and resistance lines are simply derived from past lows or highs, hence this concept of the orderbook having "memory"
I picked and computed some of these factors in order to "spot check" whether they can visibly track any rallies or market movements. I then superimposed each individual factor with the price of BTC back to 2017 on hourly bars. Minute bars and lower timeframes are too noisy to gather anything worthwhile.
I computed the following factors:
("obm_vap_hvn_dist_72h", "VAP HVN distance / ATR (close − nearest high-volume node)"),
("obm_vap_pctile_168h", "Volume-CDF percentile of current price (1 = above all 168h volume mass)"),
("obm_level_revisit_72h", "Count of bars within ±0.5 ATR of close (high-touch level)"),
("obm_support_hold_72h", "Support-hold flag: wick below low, close above"),
("obm_vwap_disp_24h", "VWAP dispersion: (close − VWAP) / σ")
I feel like sometimes there are some resemblances but it's so noisy it's almost misleading.