I started this project because I wanted a hard AI problem to chase. Recognizing written digits isn't nearly as cool as making cash. The MoneyMonster is an unsupervised learning AI which creates a list of trades based off historical price data, TA indicators and an original, scriptable trading strategy. It runs on a Jetson Nano dev kit (which is also hosting this site) for GPU accelerated Keras inferencing in a Linux environment.

Pretty good! I lost nearly $10,000 immediately, but have recovered and doubled my money! This particular strategy seems to benefit from high volatility. This chart shows performance as of 11/15/2019. Unfortunately, I was never able to find an API for papertrading, which would have allowed me to fully automate the process. After months of manually opening and closing positions every Monday and Friday, I've retired the project. I would have liked to have seen how the volatility of COVID19 affected my position. Half of my positions were OTM puts, and I probably would have seen another meteoric rise.

The MoneyMonster generates its own tagged dataset by looking back at daily price data. The output of the Neural Net is a price prediction 1 week in the future, so if we look back farther than 1 week, we know the actual result. I scheduled a script to pull new data from the S&P500 every Friday at close to update and train the neural net.