Building AI Trading Systems
Lessons learned building a profitable algorithmic trading system using Reinforcement Learning techniques.
trading finance machine-learning algorithmic-trading reinforcement-learning article

If you come from a tech or startup background, transitioning to trading may require a change in thinking. Products and engineering are often about absolutes. If your website takes 100ms to load that's pretty good. Making it load in 99ms provides negligible benefit to the end-user. It'd be a waste of engineering resources. The same is true for startups. Paul Graham likes to say that startups are rarely killed by competitors. Rather, they commit suicide. They fail because they cannot find customers, don't have a solid business model, or because of founder issues. Being killed by competition with a slightly better product is quite rare.

Trading is different. It's about relatives. It's fine if your website takes a horrible 10 seconds to load if your competitor needs 11 seconds. Having crappy data is fine if everyone else's data is even crappier. Paying high trading fees is fine if everyone is paying the same. This is pretty obvious if you look at the market as a multiplayer game. Even if you're a bad player on an absolute scale, you can win if your competition is worse. This has direct effects on how you build software to trade in the markets.

Don't forget to tag @dennybritz in your comment, otherwise they may not be notified.

Authors community post
High-school dropout. Ex Google Brain, Stanford, Berkeley. Into Startups, Deep Learning. Writing at and Lived in 日本 and 한국
Share this project
Similar projects
Probabilistic Machine Learning 4 Trading + Business Understanding
You will learn how to structure a trading problem in a probabilistic way using real Binance data, also improve your feature engineering process.