With the quick development of the Internet, online platforms that provide financial news and opinions have attracted more and more attention from investors. The question whether investor sentiment expressed on the Internet platforms has an impact on asset return has not been fully addressed. To this end, this paper uses the Baidu Searching Index as the agent variable to detect the effect of online investor sentiment on the asset price movement in the Chinese stock market. The empirical study shows that although there is a cointegration relationship between online investor sentiment and asset return, the sentiment has a poor ability to predict the price, return, and volatility of asset price. Meanwhile, the structural break points of online investor sentiment do not lead to changes in the asset price movement. Based on the empirical mode decomposition of online investor sentiment, we find that high frequency components of online investor sentiment can be used to predict the asset price movement. Thus, the obtained results could be useful for risk supervision and asset portfolio management.
The Internet has become an important platform to acquire and exchange information, and the impact of the Internet has penetrated every field of modern society in recent years. According to the
The empirical evidence on the relationship between investor sentiment and asset price has been documented by previous studies. For example, De Long et al. [
With the quick development of the information technology, modern financial market relates to the Internet more closely. With the easier access to the Internet, investors receive more and more information from the Internet, and their sentiment and opinions can also be expressed on online platforms and spread widely in a short time. Some scholars pay attention to this phenomenon and investigate the impact of the Internet on financial markets. Campbell and Cecez-Kecmanovic [
Previous works have verified the impact of investor sentiment on asset price, but until now, the impact of online investor sentiment on the asset price movement has not been fully explored. The question whether online investor sentiment leads to changes in the asset price movement has not been empirically examined. Aiming to fill the gap in the current literature, we apply the autoregressive distributed lag model and the empirical mode decomposition (EMD) to study the impact of online investor sentiment on the asset price movement based on a structural break point investigation.
The remainder of this paper is organized as follows. Section
To investigate the impact of investor sentiment on the asset price movement, we first define the variables of online investor sentiment and the asset price movement and then analyze the basic characteristic of the asset price movement.
There are two methods of evaluating online investor sentiment. The first is the direct way which detects the number and content of posts on the Internet forums, such as text mining methods. The second method uses the search intensity index provided by Google, Baidu, and other search engines. The evaluation result obtained by the first method contains more investor sentiment information, but at the same time, more noise information could be embedded in the evaluation results; moreover, the accuracy of the evaluation results is easily affected by manual data collecting. Due to the strong searching function of Google and Baidu search engine, the second method is more objective than the first method. Thus, we adopt the Search Index of the Shanghai Stock Index, released by the website Baidu (
We use the Shanghai Stock Exchange Index to represent the asset price movement of the Chinese stock market. The sample period is consistent with the online investor sentiment. The original data of the stock index are obtained from WIND database. Similar to investor sentiment data, we also use the primary
By applying (
The variation of the
The variation of the
The variation of the
Table
ADF test results.
ADF | Critical value |
|
|||
---|---|---|---|---|---|
1% | 5% | 10% | |||
|
−6.8038 | −3.4626 | −2.8756 | −2.5743 | 0.0000 |
|
−5.7672 | −3.4612 | −2.8749 | −2.5740 | 0.0000 |
|
−7.5798 | −3.4609 | −2.8749 | −2.5740 | 0.0000 |
Note: 1%, 5%, and 10% are significance levels.
Then, we use the Granger causality test to further test whether there is a causality relationship between each pair of variables and present the results in Table
Granger causality test results.
Null hypothesis |
|
Prob. | |
---|---|---|---|
|
|
10.3966 | 0.0000 |
|
6.7410 | 0.0014 | |
|
|
1.52490 | 0.2201 |
|
8.48564 | 0.0003 | |
|
|
0.59415 | 0.5530 |
|
1.94952 | 0.1449 |
Note: there are 213 observations for each sample. “
We employ multiple structural change models provided by Bai and Perron [
For each possible segment (
By comparing the residual square of different segments, we can get the optimal segment with minimal residual square. The corresponding break dates are
Applying the method of Bai and Perron [
Test results of structural break points.
Variables | Break test |
|
Scaled |
Critical value | Break date |
---|---|---|---|---|---|
|
0 versus |
207.71 | 207.71 | 8.58 | 10/10/2014 |
1 versus |
96.26 | 96.26 | 10.13 | 09/02/2011 | |
2 versus |
14.64 | 14.641 | 11.14 | 06/29/2012 | |
3 versus 4 | 1.87 | 1.87 | 11.83 | / | |
|
|||||
|
0 versus |
254.29 | 254.29 | 8.58 | 10/09/2014 |
1 versus 2 | 0.19 | 0.19 | 10.13 | / | |
|
|||||
|
0 versus |
12.45 | 24.91 | 11.47 | 07/29/2014 |
1 versus 2 | 0.66 | 1.31 | 12.95 | / | |
|
|||||
|
0 versus 1 | 3.25 | 3.25 | 8.58 | / |
|
|||||
|
0 versus |
226.32 | 226.32 | 8.58 | 10/10/2014 |
1 versus |
12.580 | 12.58 | 10.13 | 08/26/2011 | |
2 versus |
19.17 | 19.17 | 11.14 | 02/14/2014 | |
3 versus 4 | 2.23 | 2.23 | 11.83 | / | |
|
|||||
|
0 versus 1 | 6.58 | 6.58 | 8.58 | / |
Note:
On October 26, 2011,
On June 8, 2012, the People’s Bank of China (PBC) cut the benchmark interest rate for deposits and loans, and after the interest rate adjustment, the
Later, on July 29, 2014, China’s stock market had a new round of sharp increases, and the bull market lasted until June 2015. The structural break point occurred around the date of July 29, 2014. On November 17, 2014, the Shanghai-Hong Kong Stock Connect was implemented in China’s stock market. This is a remarkable event that increased the openness of the Chinese stock market. Thus, the
As shown in above results, we find that there is only one structural break point in the data of investor sentiment, and investor sentiment does not lead to the changes of the stock price movement. We consider that online investor sentiment is probably driven by the asset price movement. Investors maybe overreact to the changes of the market price movement, while the market price movement is less influenced by investor sentiment through judging the structural break points.
According to the structural break points, we divide the original sample into subsamples (see Tables
Autoregressive distributed lag model with the dependent variable of
Variable | Coefficient | Std. |
|
|
Adj. |
---|---|---|---|---|---|
|
|||||
|
241.9021 | 222.0302 | 1.0895 | 0.2846 | 0.854 |
|
0.9595 | 0.0729 | 13.1648 | 0.0000 | |
|
−0.0069 | 0.0032 | −2.1524 | 0.0395 | |
|
|||||
|
|||||
|
583.7682 | 272.4279 | 2.1428 | 0.0394 | 0.5208 |
|
0.7448 | 0.1236 | 6.0246 | 0.0000 | |
|
0.0001 | 0.0033 | 0.0418 | 0.9669 | |
|
0.0004 | 0.0031 | 0.1444 | 0.8860 | |
|
|||||
|
|||||
|
0.9124 | 0.0364 | 25.0576 | 0.0000 | 0.7645 |
|
0.0081 | 0.0038 | 2.1256 | 0.0425 | |
|
|||||
|
|||||
|
0.9537 | 0.0893 | 10.6845 | 0.0000 | 0.9672 |
Autoregressive distributed lag model with the dependent variable of
Variable | Coefficient | Std. |
|
|
Adj. |
---|---|---|---|---|---|
|
|||||
|
−0.0015 | 0.0018 | −0.8235 | 0.4114 | 0.0150 |
|
−0.0419 | 0.0777 | −0.5391 | 0.5906 | |
|
−0.0103 | 0.0078 | −1.3096 | 0.1921 | |
|
−0.0100 | 0.0087 | −1.1530 | 0.2505 | |
|
−0.0019 | 0.0078 | −0.2444 | 0.8072 | |
|
|||||
|
|||||
|
0.0169 | 0.0069 | 2.4546 | 0.0194 | 0.0084 |
|
0.0425 | 0.1818 | 0.2335 | 0.8168 | |
|
0.0002 | 0.0195 | 0.0116 | 0.9908 | |
|
0.0287 | 0.0197 | 1.4538 | 0.1552 | |
|
0.0235 | 0.0189 | 1.2450 | 0.2217 |
Autoregressive distributed lag model with the dependent variable of
Variable | Coefficient | Std. |
|
|
Adj. |
---|---|---|---|---|---|
|
|||||
|
0.0002 | 0.0001 | 2.0082 | 0.0551 | 0.7548 |
|
0.7065 | 0.1392 | 5.0771 | 0.0000 | |
|
0.0000 | 0.0002 | 0.1553 | 0.8778 | |
|
−0.0002 | 0.0002 | −0.9777 | 0.3372 | |
|
0.0000 | 0.0002 | 0.1396 | 0.8900 | |
|
|||||
|
|||||
|
0.0001 | 0.0000 | 3.0601 | 0.0028 | 0.7072 |
|
0.8431 | 0.0514 | 16.3937 | 0.0000 | |
|
0.0000 | 0.0001 | −0.2758 | 0.7832 | |
|
0.0000 | 0.0001 | −0.2424 | 0.8089 | |
|
0.0000 | 0.0001 | 0.0818 | 0.9350 | |
|
|||||
|
|||||
|
0.0001 | 0.0001 | 1.9930 | 0.0573 | 0.6295 |
|
0.7024 | 0.1333 | 5.2686 | 0.0000 | |
|
0.0003 | 0.0005 | 0.5438 | 0.5914 | |
|
0.0001 | 0.0003 | 0.1911 | 0.8500 | |
|
0.0000 | 0.0002 | 0.1142 | 0.9100 | |
|
|||||
|
|||||
|
0.0002 | 0.0001 | 1.9263 | 0.0660 | 0.7878 |
|
0.8325 | 0.0890 | 9.3494 | 0.0000 | |
|
0.0002 | 0.0003 | 0.6155 | 0.5440 | |
|
−0.0005 | 0.0003 | −1.5050 | 0.1454 | |
|
0.0004 | 0.0003 | 1.3353 | 0.1943 |
In (
According to the estimation results of autoregressive distributed lag models, the impact of online investor sentiment on the asset price movement is insignificant except for the subsamples 1 and 3. These results suggest that investor sentiment has more significant impact on the market during the calm period and the beginning period of the bear market, and the relationship between investor sentiment and market return will be weaken during the high-sentiment period. The results could be explained from the psychological and economic views. Firstly, the investors’ neural structure can be divided by different neural quantum when they react to the external stimulus. The unit of a neural quantum is likely to be formed when the stimulus is strong enough (see Stevens et al. [
Because the Baidu Search Index is constructed on a weekly basis, while online investor sentiment could change during a week, thus, it is difficult to capture the short term characteristics based on the original investor sentiment data. To detect the impact of online investor sentiment on the stock market from a shorter term view, we use the empirical mode decomposition (EMD) to obtain the low, medium, and high frequency components of online investor sentiment. EMD is a data-driven analysis method for both nonlinear and nonstationary data. Because it is intuitive, direct, posterior, and adaptive, EMD is widely used in different fields [
(1) Find the local maximum value
(2) Calculating the mean value of the envelope line, we can obtain the new time series
(3) Calculate the standard deviation (SD):
(4) Separate
The original investor sentiment can be expressed as
Finally, we acquire
EMD results of the original data. The
EMD of SHINDEX
EMD of SENINDEX
EMD of RSH
EMD of VSEM
EMD of GARCHRSH
EMD of GARCHVSEM
Estimation results of the autoregressive distributed lag model with high frequency components.
Independent variable | Coefficient | Std. error |
|
|
Adj. |
---|---|---|---|---|---|
|
|||||
|
−0.2577 | 0.0678 | −3.8003 | 0.0002 | 0.0771 |
|
−0.0006 | 0.0003 | −1.8411 | 0.0670 | |
|
|||||
|
|||||
|
−0.4562 | 0.0618 | −7.3789 | 0.0000 | 0.2239 |
|
−0.0121 | 0.0063 | −1.9263 | 0.0554 | |
|
|||||
|
|||||
|
−0.0001 | 0.0000 | −2.4204 | 0.0164 | 0.0429 |
|
0.0001 | 0.0000 | 2.1079 | 0.0362 |
From Table
The Shanghai Stock Index and the daily trading volume. Note: in this figure, the full line is the Shanghai Stock Index, and the dotted line is the daily trading volume in the Shanghai stock market. The unit of trading volume is 100 billion Chinese Yuan. The left
Meanwhile, investors can be influenced by online news and opinions. During the period from April 2015 to June 2015, some news and articles released on online platforms strengthened the optimistic sentiment, and some investors believed that the 4000 points of the Shanghai stock market was the start of the bull market and there was no bubble in the stock market. Because investors are more likely to be influenced by the Internet, supervisors should pay more attention to online investor sentiment.
In the era of the Internet, investor sentiment can be expressed on online platforms. In this paper, we systematically analyze the impact of online investor sentiment on the asset price movement by using the Baidu Search Index and the Shanghai Stock Exchange Index from January 1, 2011, to June 1, 2015, as the sample. The empirical results show that there is a cointegration relationship between the weekly investor sentiment and the asset price movement, but the autoregressive distributed lag model with the independent variable of sentiment has poor ability to predict the index, return, and volatility of asset price, and the structural break points do not lead to changes in the asset price movement. By using the empirical mode decomposition, we find that the high frequency component of online investor sentiment can be used to predict the asset price movement. Additionally, higher online investor sentiment is associated with a high possibility of market risk. Our empirical results also have policy implications. Because online investor sentiment has significant impact on asset price movement, the supervisors should pay more attention to the Internet news, online forum, instant messages, and other online platforms. For maintaining the stability of financial markets, supervisors could enact the laws or regulations to punish the rumor contagion which may lead to investor panic or overoptimism. Meanwhile, the supervisors can collect web information based on the big data technology and construct the online investor sentiment index, especially the high frequency sentiment index to alert the risk of the stock market.
The authors declare that there is no conflict of interests regarding the publication of this paper.
This work was supported by the National Natural Science Foundation of China under Grant nos. 71373072 and 71501066, the Specialized Research Fund for the Doctoral Program of Higher Education under Grant no. 20130161110031, and the Foundation for Innovative Research Groups of the National Natural Science Foundation of China under Grant no. 71521061.