TEAM 2: ALGOTRADING WITH MACHINE LEARNING
TEAM 2: ALGOTRADING WITH MACHINE LEARNING
Data-Centric Comp Capstone | Spring 2021
Jash Lal, Jason Li, Jeff Peng, Colin Peppler
ABSTRACT
This Algotrading program is designed to maximize profit of buying and selling stocks by predicting future stock prices using historical trends. The program architecture is shown on the left side. A LSTM model is trained and tested using data given by data processor component. After the model is properly trained and tested the stock ranker would pass historical data to the model to let it predict. Then the stock ranker would rank all stocks based on highest return percentage. Algotrading strategy would then use the ranked stocks to choose the best stocks to buy/sell. Afterwards the results will be sent to QuantConnect Brokerage to display in a user-friendly UI.
MODEL RESULTS
DFNN performed the best. Surprisingly, LSTM performed just as well as a linear regression model.
FINAL RESULTS
DFNN performed the best, but the performance difference between TA (trades all companies) and DFNN + TA (ranks and trades a subset of companies) shows that a model can successfully select profitable companies to trade.
