Quantitative investment research tooling.
Try it for yourself
Trying to beat the market Looking to use an ensemble to predict stock prices or wether to buy/sell and beat the market gen 1 Time horizon 1 day XGB predicting exact prices - 5/10. Got some results but ultimately it failed to generalize and struggled with large datasets due to outliers(even with different eval functions). It potentially read too much into the numbers rather than the change. gen 2 XGB predicting %price change - 8/10. Much better generalization, I think the main reason for this is that it doesn't worry about the current price at all and only focuses on finding the % change. Also began passing the month and day of the week(they have a relatively low F score). Also introduced trades based on conviction (if the absolute value of the % change is greater than a limit) this led to better results but still not profitable. No returns on modern markets: time to mix it up. begin gen 3 scraping my own data using yfinance. Starting with russell 3000. Exloring adding the following to the data - more lag collumns, label earnings dates/dividend dates, moving averages, PE ratio, sector, and/or any other indicator I can come up with. check out studies file