After experimenting with the most popular backtesting libraries, I wanted something that handled 5-second (the lowest interval data I have access to) and bid/ask fill behavior better. So I ended up building my own prediction and simulation stack - Python, Pandas, SQLite, DuckDB, Torch, etc.
The simulated fills line up closely with empirical results from actual trades (the sim is deliberately a bit pessimistic), so I'm feeling confident in its accuracy so far. The live trader itself is another python codebase :)
I've just wrapped it in a small web frontend. It's running on my home machine, so performance might be rough. I'd love to hear what you think!
vain•2h ago