Background Research & Rationale

Risk reduction in stock market portfolios has become an increasingly popular topic in recent years, especially with advances in computing and machine learning. This project proposes a novel approach of utilizing recurrent neural networks (RNN) to forecast stock prices and increase buyer confidence. The question driving this research was: In comparison with traditional algorithms made for high frequency trading, is it possible to create a trained neural network with increased accuracy that will successfully forecast prices and risk factors for stocks? The proposed method includes utilizing RNN’s and probabilistic statistical risk analysis to mitigate risk. Among the statistics provided are the mean absolute error value (MAE), the huber loss value, the standard error of estimate value (SEE), the Sharpe ratio, and various profit scores. The specifications of the network remained constant and included 5 layers, bidirectionality, and training over 20 epochs, giving a forecasted stock price 15 days into the future. The specific stocks that were tested using the program included MSFT, TSLA, AAPL, NVAX, SPY, AMZN, GOOGL, SBUX, HSY and DIS. The model accurately predicted prices, and as testing samples increased, the overall accuracy for both prediction and risk similarly increased. A notable demonstration of the program's accuracy was shown in the SPDR S&P 500 ETF Trust (SPY), the largest index fund in the world that tracks 500 of the top companies. The given SEE was only $6.45, compared with its price of $387.71. In phase two of experimentation, the aim was to automatically analyze a CEO’s twitter page, and determine if recent tweets were negative, positive, or neutral, then see how it affects the daily stock price. By using a Kaggle dataset, the team was able to analyze Elon Musk’s as a keyword of interest and analyze those tweets. The model successfully predicted the sentiments, with an accuracy score of 0.9916, a precision score of 0.9919, a recall score of 0.9913, and an F1 score of 0.9916. Extending the sentiment analysis and seeing its impact on stock price is the next step for future research.