I spent a lot of time learning from traders, and learning statistics. Most folks in trading use misleading profit and loss metrics to see if something is worth trading. I used the same kind of backtests, but I layered Bayesian inferencing on top of it.

I studied machine learning with Andrew Ng’s courses, studied deep learning with Ian Goodfellow’s book. Most importantly I took a course run by university professor and researcher in anthropology, Richard McElreath. I did my best to faithfully apply what I learned, though I am sure I strayed from academic standards.

At that point I had been doing this for years, for countless hours. It was my only hobby, and I dive hard into hobbies.

I tried my damnedest to be predictive every which way. I kept meticulous records to avoid fooling myself. Sometimes my models fooled me, and sometimes they combined with luck for my records to fool me. Long term, it’s pretty clear. No evidence of any edge, ever, for any approach taken.

At the end of all of this toil and labour, I have the skills I learned along the way: statistical skepticism, a hands-on understanding of fat tails, an appreciation for the experience of randomness and the highs and lows of gambling. I think that’s worth a lot - but I also think you can learn that a lot easier some other way.

I have done very well with buy and hold, it’s fantastic. There’s some bullshit in how you assign your portfolio - what proportions of what exposures - but its very profitable and exceptionally low stress compared to trading. It definitely has a better Sharpe/sortino/ulcer metric.