> PelvicOarfish: I have Claude Opus running on a server in a loop to data mine 10 years of EOD options data for trading strategies.

Explosive Ordinance Disposal?

PelvicOarfish wrote:

Leykis is closer than he knows. EOD is end of day bars, which is what most of these loops run on, and it's why three weeks found nothing.

Here's the thing nobody tells the guy with the Python loop. If you test a thousand strategies at the usual 5 percent cutoff, about fifty of them are going to come back "statistically significant" by pure luck. So if yours has been grinding for three weeks and hasn't found a single one, either it's correcting for all the strategies it tried, which is the honest way to do it and almost nobody does, or your data is so thin that luck can't even show up. The day it does finally find an edge, you should trust it less, not more. Then add commissions, slippage, and the fact that anything that worked on 2015 to 2024 daily bars got arbitraged out by guys with servers in New Jersey, and you're left with buy the index and go to bed.

I know because I did this by hand with a spreadsheet in a recliner. I found a perfect strategy. It worked on every year I tested except the year I put money in it. The machine is just doing my old mistake faster, and it has to stop every week when the usage limit runs out.

My advice is let it keep running. Three weeks with no edge means it's the most honest quant on the internet. It's still on daily bars.