Stock forecasting using transformers, an emerging machine learning technique

Abstract

The stock market, being a major form of investment, has been given increased importance and attention in recent years. Many investors, analysts have, therefore, shown forecasting the direction of the stocks with significant interest. Furthermore, Deep Learning models and Artificial intelligence have time and time again prove to have high accuracy in predication of stock prices. Until recently, investors and analysts have sole rely on technical indicator for technical analysis of stock data, however sentimental analysis – study of investors’ emotion and wiliness to invest, may be used to determine the movement of stock prices.

This project studies the comparison of efficiency of the Transformer model to that of the Long Short Term Memory (LSTM) model in both sentimental and technical analysis of stocks, as well as to study the effects of sentimental analysis to stock price forecasting.

Firstly, sentimental analysis of news headline for the companies Alphabet Inc (Google), Meta Platforms Inc (Formerly known as Facebook) and Apple Inc, all listed on the NASDAQ, is done using both LSTM and Transformer model. Secondly, sentimental scores will be concatenated together with stock historical indicators before predicting stock price movement using the 2 models. Lastly, comparison of efficiency is done by studying the varying results gotten by the various combination of the 2 model.