Thread by @AgustinLebron3

Had a long think about things @KrisAbdelmessih and @VitruviusCurve wrote this week. About ML in trading.

The levels of quant trading:

1/n

0. Fumbling around in the dark.

Most people stay here forever, and even those who progress past this level still revert to it with annoying frequency. :)

1. Does +EV trades.

Has figured some things out, can do good trades and know they're good in expectation. This is a hard level to get to!

2. Has found +EV signals.

This is a generalization of finding good trades, because you've now started to more deeply understand _why_ they're good.

3. Has a reliable process for finding +EV signals.

No, I don't mean machine learning yet. Or at least, this isn't the most important part. The most important part by far is sociological. How have you organized the interactions between humans so that finding signals is reliable?

4. Has a reliable process for combining +EV signals in even more +EV ways.

AFAICT this is the highest level, but there may be higher ones I don't know. RenTech lives here, and maybe a half dozen others do too. Not many though.

What does this have to do with ML? Simple:

Most people who come from AI/ML into trading fail because they're trying to solve 3 and 4 without ever having tried to solve 1 and 2.

You end up building castles made of sand, which fall into the sea. Eventually. (HT Jimi)