I see, thanks for the insight. Great to see an self-sufficient independent trader executing like that. I've got my signals and approach etc. but am finding it too complicated and mentally exhausting to manage manually, so a ML approach would solve a lot of issues.
I am sitting on over $30k in unrealized gains from the swing trades opened earlier this week. The hardest part of having an automated trading system is sitting on your hands and NOT intervening. Its so tempting to make it close the position and lock in those gains... especially when the discretionary trader inside you looks at the market and thinks a pull back may come... Every time I get that urge, I look at a sheet of paper I put on my monitor a year ago that said DO NOT TOUCH. The market could very well pull back and some of that 30k can evaporate... or it can gap up another 1% tomorrow for all I know. I just have to trust the process.
I find the temptation to override an automated system is very strong during a long losing or drawdown period. The temptation is to skip signals, take profits early, take losses early. I find it is less tempting to override a system when it has been winning recently and is at or near high water mark. Even if I make or save money by overriding the system on one occasion, I inevitably lose more the next time I attempt to do so.
For me its the opposite. I get the strongest urge when I have had a string of wins. I know that a draw down is due and is around the corner. I start wanting to "protect" profits. Especially large unrealized ones like I am sitting on right now. For me, when I am in a draw down, I don't want to touch it. In-fact, I find I stop looking at the results altogether.
It all depends on how you modeled your algorithm and fit the data. I run multiple ai strategies that take advantage of short term "mispricing" dislocations (caused by fear and greed in the market). So, the occasional large moves in my favor are usually not captured and I suffer from the models not taking advantage of those large moves (not locking in such unrealized pnl). I ran backrests and the overall performance improved when I manually take profits on those few outsized moves that the model does not capture properly. But it really depends on how you trained your models. Some models are specifically trained to aim to capture such 3 or 4 sigma events.
I've added it to the list I follow (though I don't follow directly, just a search I have for twitter people I want to follow)
Absolutely zero. Did you check out the book on Simons yet? As a quant/system trader you'll find it amusing. Funny thing is that Simons also had issues with wanting to override his systems - especially during drawdowns (I saw you had it the other way around).