Electronic Thesis and Dissertation Repository

Thesis Format

Monograph

Degree

Master of Science

Program

Computer Science

Collaborative Specialization

Artificial Intelligence

Supervisor

Mercer, Robert E.

Abstract

Investors seek to take advantage of computer technology to gain an edge on their investments. This can be done through quantitative (historical number-based) analysis or qualitative (natural language-based) analysis. Subject matter experts have been known to make predictions between 70 and 79% accuracy at best and less than 50% accuracy on average. Sophisticated algorithms through qualitative analysis are known to demonstrate more successful market predictions for specific stocks. It stands to reason that the same technique could work just as well or better for attempting to predict entire sectors of the stock market. By using indices and exchange traded funds, it is possible to track entire sectors of the stock market and make evaluations. The S&P 500 Energy index demonstrated itself to be especially bullish over the three-month period from January 22, 2008 to April 22, 2008 and especially bearish over the three-month period from July 14, 2008 to October 10, 2008. The three-month period from January 30, 2012 to April 30, 2012 was the least volatile period and can serve as a base-line measure for the other two. By extracting news stories over these three time periods, and analyzing specific parts of speech (verb, adverb, and adjective) in those stories over those periods, it is possible to make predictions about expected changes in the S&P 500 Energy index between 55% and 85%. The bullish three-month period was especially predictable at a rate of 85.22% on average. A bigram analysis over the same three threemonth periods is also attempted with an accuracy between 50% and 55% on average. Applying an artificial neural network over the same three three-month periods can achieve an accuracy of between 55% and 65%.

Summary for Lay Audience

Investors seek to take advantage of computer technology to gain an edge on their investments. This can be done through analysis of number-based data such as sales figures or stock prices, or through analysis of word-based data such as press releases or news stories. Subject matter experts have been known to make predictions between 70 and 79% accuracy at best and less than 50% accuracy on average. Sophisticated algorithms are known to demonstrate more successful market predictions for specific stocks than subject matter experts. It stands to reason that the same technique could work just as well or better for attempting to predict entire sectors of the stock market. By using indices and exchange traded funds, it is possible to track entire sectors of the stock market and make evaluations. The S&P 500 Energy index demonstrated itself to be especially bullish over the three-month period from January 22, 2008 to April 22, 2008 and especially bearish over the three-month period from July 14, 2008 to October 10, 2008. The three-month period from January 30, 2012 to April 30, 2012 was the least volatile period and can serve as a base-line measure for the other two. An analysis of news stories over these three time periods was done using Artificial Intelligence techniques to predict stock market sentiment for the energy sector.

Download

Share

COinS