Developed by:
- Elayne Blancas -
New York University - Gregory Perez -
Cornell University - Jonathan Zamudio -
University of Arkansas
Our Stock Market Predictor dictates whether an individual stock will either over-perform or under-perform, calculate the long and short gains of the portfolios stocks, and obtain the portfolios stock average return and Sharpe Ratio. We used yfinance API to obtain a decade of daily historical stock data from Yahoo Finance for the top 30 Standard and Poor (S&P) 500 companies. Our data consisted of non-negative stock values with no upper bound for weekdays and non holidays. Our Stock Market Predictor uses Random Forest as a baseline for our LSTM model. To prevent over-fitting, we dropped 20 percent of our data, a common practice used in machine learning. Additionally, in an effort to obtain a higher accurate score, we use a decade of the stockˆas history to predict its future trends. In making this model pipeline, we would use the 3 prior years as our training data. This pipeline would take the years of data and go over it in a series of sequences, each sequence containing 240 days. This sequence of data would then give us the value of the stock. Our model has 60% to 70% training accuracy. Our predictor would advise for a long investment on either Comcast Company or Adobe, and a short investment on either Protector Gamble Company or Salesforce. Although this type of predictor is difficult to generalize, our future work would be to include sentimental analysis, web scraping, and Beautiful soup for real time data to mitigate for its volatility. Our model does not account for anything external and should, therefore, not be used as a main investment tool.
Run notebook in Google Colab.