Using social media mining technology to improve stock price forecast accuracy

• Author: Jia‐Yen Huang,Jin‐Hao Liu
• DOI: http://doi.org/10.1002/for.2616
• Date: 01 January 2020
• Published date: 01 January 2020

RESEARCH ARTICLE

Using social media mining technology to improve stock

price forecast accuracy

Jia‐Yen Huang | Jin‐Hao Liu

Department of Information Management,

National Chin‐Yi University of

Technology, Taichung City, Taiwan, ROC

Correspondence

Jia‐Yen Huang, Department of

Information Management, National

Chin‐Yi University of Technology, No. 57,

Sec. 2, Zhongshan Rd., Taiping Dist.,

Taichung City 41170, Taiwan, ROC.

Email: jygiant@ncut.edu.tw

Abstract

Many stock investors make investment decisions based on stock‐price‐related

chip indicators. However, in addition to quantified data, financial news often

has a nonnegligible impact on stock price. Nowadays, as new reviews are

posted daily on social media, there may be value in using web opinions to

improve the performance of stock price prediction. To this end, we use logistic

regression to screen the chip indicators and establish a basic stock price predic-

tion model. Then, we employ text mining technology to quantify the unstruc-

tured data of social media opinions on stock‐related news into sentiment

scores, which are found to correlate significantly with the change extent of

the stock price. Based on the findings that the higher the sentiment scores,

the lower the prediction accuracy of the logistic regression model, we propose

an improved prediction approach that integrates sentiment scores into the

logistic regression model. Our results show that the proposed model can

improve the prediction accuracy for stock prices, and can thus provide a new

reference for investment strategies for stock investors.

KEYWORDS

chip indicators, logistic regression model, prediction accuracy, sentiment scores, text mining

1|INTRODUCTION

In order to make the best investment decisions, stock

investors generally collect related information from

sources such as TV media, newspapers, magazines and

the Internet. However, faced with various information,

investors often cannot identify which information is most

important. How to build up high‐yield investment strate-

gies with the help of quantitative indicators and stock

price forecasting tools is obviously a topic of concern to

investors.

Stock price fluctuations are dynamic, nonlinear, non-

stationary, and carry a lot of noise, which makes the

stock price difficult to forecast. Stock price forecasting

has been a research topic of wide concern in academia

and financial domains. Early research has mainly been

based on random walk theory (RWT) and efficient mar-

ket hypothesis (EMH). RWT argues that stock price fluc-

tuations are random and, hence, the next step of the stock

price is as irregular as a person walking on a square.

EMH suggests that designing a system based on any

information to predict stock price changes is impossible

because all information has already been reflected in

the existing stock prices.

According to previous EMH research, stock prices are

mainly driven by new information rather than current

and past prices. Since the news is unpredictable, the stock

price will follow the random walk model and cannot

exceed 50% accuracy prediction (Qian & Rasheed, 2007).

However, many studies have shown that stock prices

are not random, but can indeed be predicted to a certain

extent. News may be unpredictable, but early metrics can

Received: 9 November 2018 Revised: 10 May 2019 Accepted: 16 May 2019

DOI: 10.1002/for.2616

Journal of Forecasting. 2020;39:104–116.wileyonlinelibrary.com/journal/for© 2019 John Wiley & Sons, Ltd.104