All trading is rooted in discretionary human decisions weather it is programed or manually hand traded which explains why the performance over the last 5 years is narrow. Both discretionary & systematic have their pros & cons, the trend these days is a hybrid between the two which so far has had the best performance. Discretionary traders are using more automation & system regulated overlays to ensure stricter rule following while system traders are more likely now to adapt faster to changing market conditions. They are getting to be more similiar over time than different.
Would be interesting to see a chart over a longer timeframe. I guess most of the systems over the last few years could approach something like: Code: def is_buy_signal(time, price, volume): return True I agree that a discretionary mix with algorithmic enhancement is the way forward. Until we have general AI any system is likely to have alpha decay which costs effort to tweak hyperparameters. When we have GAI trading as a means is fucked anyway. You can of course stack specific ML on top of ML and auto-tweak your systems, but you end up with something where you have little idea where your system is going wrong, only that it seems to be breaking. Building a system as the above necessitates that you also build into that system appropriate logging, so that it can explain to you its actions, and over time you can stay aligned with the expectancy. Anything less ends up with a bunk system and no way to repair it due to market meta decay. The days of true speculation are not dead, and we should remember that just because a generalised machine can make money profitably in the current market, it doesn't mean that it's a better or more robust solution than taking a speculative stance. Once you have that, sure, let the machine work for you.