New tool for stock investment risk management. Trend forecasting based on individual investor behavior
| Published date | 01 November 2019 |
| Pages | 388-405 |
| Date | 01 November 2019 |
| DOI | https://doi.org/10.1108/IMDS-03-2019-0125 |
| Author | Yi Sun,Quan Jin,Qing Cheng,Kun Guo |
| Subject Matter | Information & knowledge management |
New tool for stock investment
risk management
Trend forecasting based on
individual investor behavior
Yi Sun, Quan Jin and Qing Cheng
School of Economics and Management,
University of Chinese Academy of Sciences, Beijing, China, and
Kun Guo
Research Center on Fictitious Economy & Data Science,
Chinese Academy of Sciences, Beijing, China
Abstract
Purpose –The purpose of this paper is to propose a new tool for stock investment risk management through
studying stocks with what kind of characteristics can be predicted by individual investor behavior.
Design/methodology/approach –Based on comment data of individual stock from the Snowball, a
thermal optimal path method is employed to analyze the lead–lag relationship between investor attention (IA)
and the stock price. And machine learning algorithms, including SVM and BP neural network, are used to
predict the prices of certain kind of stock.
Findings –It turns out that the lead–lag relationships between IA and the stock price change dynamically.
Forecasting based on investor behavior is more accurate only when the IA of the stock is stably leading its
price change most of the time.
Research limitations/implications –One limitation of this paper is that it studies China’s stock market
only; however, different conclusions could be drawn for other financial markets or mature stock markets.
Practical implications –As for the implications, the new tool could improve the prediction accuracy of the
model, thus have practical significance for stock selection and dynamic portfolio management.
Originality/value –This paper is one of the first few research works that introduce individual investor data
into portfolio risk management. The new tool put forward in this study can capture the dynamic interplay
between IA and stock price change, which help investors identify and control the risk of their portfolios.
Keywords Risk management, SVM, Thermal optimal path method, Investor attention
Paper type Research paper
1. Introduction
Portfolio management has always been a research hotspot in the field of modern finance
(Smith and Stulz, 1985; Nocco and Stulz, 2006; Galluccio and Roncoroni, 200 6; Berry-Stölzle
and Xu, 2018). A good portfolio can not only increase investors’income and enhance their
ability to resist inflation but also contribute to the sustainable development and efficiency of
the financialmarket. As Christoffersenand Diebold (2000) andRoh (2007) stated, forecastingis
an essential part of performing risk management, specifically in allocating assets to various
portfolios in order to efficiently hedge those portfolios’risk. Therefore, financial forecasting is
of enormous importance for both individual investors and the whole market.
The development of information technology has enabled scholars to observe the
behaviors of financial market participants from a micro perspective as never before (Ngai
et al., 2015). In recent years, investor attention (IA) has been widely used in financial market
forecasting (Chemmanur and Yan, 2019). In previous studies, scholars used extreme stock
price volatility (Kaniel et al., 2012), upper price limit events (Seasholes and Wu, 2007) and
trading volume (Mao et al., 2011) as proxy variables of IA. Currently, people’s searches,
comments and other behaviors can be observed, recorded and stored in the form of data.
A great deal of research shows that these data can be used as a direct measure of investor
Industrial Management & Data
Systems
Vol. 120 No. 2, 2020
pp. 388-405
© Emerald PublishingLimited
0263-5577
DOI 10.1108/IMDS-03-2019-0125
Received 7 March 2019
Revised 7 June 2019
17 August 2019
Accepted 8 September 2019
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0263-5577.htm
388
IMDS
120,2
sentiment and can be used to predict social and economic phenomena (Da et al., 2011; Zhang
et al., 2011; Chen et al., 2014; Ji et al., 2019).
However, some studies have found that internet data do not necessarily increase the
accuracy of prediction models. Google launched Google Flu Trends (GFT) in 2008; it
estimates the current global flu epidemic in near-real time based on aggregated Google
search data. Ginsberg et al. (2009) successfully used GFT to predict the spread of H1N1
across the USA, even to specific regions and states. In the following years, however, the
accuracy of GFT prediction declined sharply. Lazer et al. (2014) pointed out that GFT had
overestimated the incidence of influenza compared to the US Centers for Disease Control
and Prevention’s report, from the time period of October 2011 to August 2013. Why does the
GFT-based flu prediction fail? A possible reason is that Big Data does not produce data that
are valid and reliable for scientific analysis, so it is impossible to construct validity,
reliability and dependencies between data. Another viewpoint is that the keywords that
users searched are not what they intended to look up; that is, the data generation mechanism
has undergone a qualitative change. The GFT-based influenza prediction failure indicates
that the perspective of predictive research based on internet data needs to change.
This paper holds that these internet data, as complementary to financial data, can
conditionally improve forecasting accuracy. We find that there is a dynamic lead–lag
relationship, which differs from stock to stock, between IA and stock price change.
Forecasts based on individual investor data are more accurate only when IA for the stock
stably leads its price change most of the time.
Moreover, although investor sentiment has been proven to be relevant to the stock
market (Zhang et al., 2011) and even to be a predictor of stock price (Ranco et al., 2015;
Dimpfl and Jank, 2016; Malandri et al., 2018), few scholars have studied the portfolio trading
strategy based on it. This paper aims to fill this gap by proposing a novel tool, namely, TOP-
SVR, for portfolio management. First, we construct the IA index of all listed companies in
the Shanghai Exchanges and Shenzhen Exchanges based on the comment data collected
from Snowball. We then employ the thermal optimal path (TOP) method in the financial
physics field to analyze the individual stock’s dynamic lead–lag relationship between IA
and the price fluctuation. Our findings show that the lead–lag relationship between IA and
price fluctuations is dynamic and that the characteristics of the lead–lag relationship differ
from stock to stock. Specifically, of 2,776 sample stocks, 102 stocks’IA are ahead of the
stock price change over more than 75 percent of the trading days. Further, we establish
three typical stock price forecasting models, namely, SVR, BP neural network and ARIMA.
We integrate the IA of the leading 1–5 periods into the explanatory variables. The results
show that the SVR model has the best prediction accuracy for two kinds of stocks. Stocks
whose IA leads the fluctuation of the stock price have a better predicting accuracy after
adding IA to the models. For stocks with the opposite character, the prediction accuracy is
higher only using the basic market indexes. Thus, a new tool for portfolio management
based on TOP-SVR is put forward.
This study contributes to both financial forecasting theory and portfolio management
practice. First, this paper shows how to use the TOP method to identify useful predictors for
individual stock price indicators. The TOP-SVR tool put forward in this paper incorporates
individual investor data into prediction models and achieves a substantial improvement in
prediction accuracy. In other words, this paper contributes to the financial market prediction
theory. More importantly, this study provides a possible explanation for the long-standing
debate on the predictive ability of investor sentiment. That is, not all stocks can be predicted
by IA. Stocks are predictable when they have a stable lead–lag relationship with IA.
Second, although the predictive ability of investor sentiment for stock prices has been
widely recognized, few scholars have studied incorporating investor sentiment into asset
allocation strategy. The TOP-SVR tool proposed in this paper has a higher accuracy than
389
Stock
investment
risk
management
To continue reading
Request your trial