Abstract
This project is an attempt to exploit the news articles and predict short term stock price movements. It shows how a combination of stock market history, and the relevance of the news articles to the financial context can be used to label the news articles. Then, these labels are used to perform adjustments on the trained time-series predictions of the future stock price. The time-series model is trained by using the historical stock prices as an input to a recurrent neural network.
The main idea of the project is to show that the difference between the time-series predicted stock price and the actual stock price can be explained by the news articles. Therefore, the objective is to analyze and extract such information, and derive numerical indicators from the news text. For prediction of stock price for a day, the news articles of that day are analyzed and the combined effect of them is realized in the time-series prediction to compute the final prediction. It is also shown how the model can be used in a financial market setting to generate profitable results.
Description
Prediction of stock market trend is a very difficult problem in general because of its volatile nature. Another reason is that there are many reasons that are responsible to determine the price of a stock. News always accounts for the significant amount of importance when investors and stock analysts evaluate and trade stocks. Actually, news contains information which may also influence the confidence and expectations of markets. Namely, the news is an obvious and easily accessible resource for realizing and predicting market condition in the present and future. In addition, some of the news may not seem to be related to the market; however, they may somehow affect the market through a series of butterfly effects. Because it is very easy to obtain huge amount of unstructured news online, exploiting the data and analyzing it can be beneficial for the prediction of stock prices. Therefore, all the articles should be taken into consideration for analysis.
There has been a considerable amount of work done to predict stock prices in the past. Nonetheless, they predict merely based on financial news articles. In this project, we trained two independent models: one to analyze the importance of news articles and their impact on the stock market, and second to perform time-series prediction of the stock price. We adopt Long Short-Term Memory (LSTM) approach to analyzing textual statements in news articles and combine with time series to predict stock price movements. Finally, it is shown how these two models can be combined and be used in a trading simulation setting to earn profits.
Member
Sean Chung
sean.chung@wisc.edu Akshata Bhat
akshatabhat@cs.wisc.edu Aditya Rungta
arungta@wisc.edu Rohit K Sharma
rsharma54@wisc.edu
