TRAP@NCI

Social Media Sentiment Analysis for Stock Price Behavior Prediction

Stuefer, Marina (2014) Social Media Sentiment Analysis for Stock Price Behavior Prediction. Diploma thesis, Dublin, National College of Ireland.

[img]
Preview
PDF (Diploma)
Download (4MB) | Preview

Abstract

This research consists in the analysis of social media sentiment from Twitter and its correlation to changes in stock closing prices for a total of ten companies in relation to tweets from the company itself and by all users about the same company. Three types of sentiment indices have been created to explain dynamics in stock price behaviour. These are an overall measure of sentiment score for each of the 29 trading days between 4th April 2014 and 19th May 2014 as well as a count of total positive words and total negative words mentioned on each day. It has been found that three series of stock prices are not well behaved in their distribution and cannot be utilized for correlation analysis. Furthermore, a set of 43 models has been proven to not be useful for the purpose of forecasting through testing for Granger causality. The remaining eight models have shown to be useful for predicting stock price through the analysis of social media sentiment. The correlations found between the sentiment indices and the stock price are very week with percentages between 12% and 21% correlation in the cases of Amazon and its total sentiment as well as positive sentiment count of opinion from users (17%,21%) and from tweets published from American Airlines itself (12%, 18%). Tweets about Starbucks, from Starbucks and about Facebook seem to affect stock price changes, negatively. These correlations suggest the increase of total sentiment for tweets about Starbucks and from Starbucks decreases stock price as well as the negative words count decreases stock price for Facebook. These models have been further investigated and found to not be robust enough to explain variations in data by looking at the R-squared statistic. For this reason the author concludes that it is recommended to enlarge the scope of the analysis further to include a broader timeframe and increase significance of results.

Item Type: Thesis (Diploma)
Subjects: Q Science > QA Mathematics > Electronic computers. Computer science
T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science
Divisions: School of Computing > Higher Diploma in Science in Data Analytics
Depositing User: CAOIMHE NI MHAICIN
Date Deposited: 15 Dec 2014 15:33
Last Modified: 15 Dec 2014 15:33
URI: http://trap.ncirl.ie/id/eprint/1879

Actions (login required)

View Item View Item