ORIGINAL RESEARCH

published: 16 June 2020

doi: 10.3389/fphy.2020.00206

Effective Transfer Entropy Approach

to Information Flow Among EPU,
Investor Sentiment and Stock Market
Can-Zhong Yao* and Hong-Yu Li
School of Economics and Commerce, South China University of Technology, Guangzhou, China

In contrast to the traditional view that economic policy affects investor sentiment and
eventually causes stock price fluctuations, we reveal that investor sentiment is a reflection
of both economic policy and stock market information. This article first proposes an
improved ETE method with a sliding window. We verify that this new method can capture
the dynamic orders effectively by validating this method with the linear TE method.
Furthermore, using the improved method, we investigate the strength and direction
of information flow among economic policy uncertainty (EPU), investor sentiment and
stock market by the novel concept of dynamic effective transfer entropy. The EPU and
investor sentiment results show that EPU influenced investor sentiment mainly from
August 2015 to June 2016. Among different policies, China’s exchange rate reform policy
and “circuit-breaker” policy in the stock market played an important role. Moreover, the
analysis of sentiment and stock price returns shows that investor sentiment is more
Edited by: a reflection of changes in stock price returns with a 1-month lag order and that the
Wei-Xing Zhou, stock market has a significant bargainer effect and a weaker bandwagon effect. Finally,
East China University of Science and
Technology, China
there is no significant information flow transmission relationship between EPU and stock
Reviewed by:
market volatility, indicating that stock market fluctuations are essentially not affected
Feng Ma, by national policy fluctuations. Although investor sentiment is affected by changes,
Southwest Jiaotong University, China such as exchange rate reform and stock market policies, many investors do not form
Peng-Fei Dai,
Tianjin University, China consensus expectations.
*Correspondence: Keywords: EPU, investor sentiment, stock market, information flow, transfer entropy
Can-Zhong Yao
[email protected]
INTRODUCTION
Specialty section:
This article was submitted to Transfer entropy arises from the formulation of conditional mutual information. When
Social Physics, conditioning on past values of variables, it quantifies the reduction in uncertainty provided by
a section of the journal
these past values in predicting the dependent variable, which presents a natural way to model
Frontiers in Physics
statistical causality between variables in multivariate distributions. In the general formulation,
Received: 16 March 2020 transfer entropy is a model-free statistic that is able to measure the time-directed transfer of
Accepted: 13 May 2020
information between stochastic variables and therefore provides an asymmetric method to measure
Published: 16 June 2020
information transfer.
Citation:
The information transfer method has been widely used in the finance field. Kwon & Yang [1]
Yao C-Z and Li H-Y (2020) Effective
Transfer Entropy Approach to
employed it to measure the relationship between equities indices, showing that the information
Information Flow Among EPU, transfer was greatest from the US and toward the Asia Pacific region. In particular, the S&P 500
Investor Sentiment and Stock Market. was shown to be the strongest driver of other stock indices. In earlier and somewhat related
Front. Phys. 8:206. work, Marschinski and Kantz [2] defined and used effective transfer entropy to quantify contagion
doi: 10.3389/fphy.2020.00206 in financial markets. Kyrtsou et al. [3] proposed a Granger causality method based on partial

Frontiers in Physics | www.frontiersin.org 1 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

transfer entropy to explore the complex relationships among DATA DESCRIPTION

the S&P 500, VIX and volume. Dimpfl and Peter [4] proposed
an appropriate bootstrap to derive confidence bounds and The economic uncertainty index used in this article is an
showed in a simulation study that standard linear approaches index compiled by Paul Luk’s research team [43] from
in economics and finance, such as vector autoregressions and Hong Kong Baptist University. These authors construct the
Granger causality tests, are not well-suited to detect information index using words from 10 Hong Kong newspapers, including
transfer. Garcia-Medina et al. [5] used random matrix theory Wen Wei Po, Sing Pao, etc. For each newspaper, the research
and information theory to analyze the correlations and flow of group counts articles containing Chinese words related to
information between 64,939 news items from The New York economics, uncertainty and policy to formulate the economic
Times and 40 world financial indices over 10 months during the policy uncertainty index. The latest data of the index and
2015–2016 period. Their results suggested a deep relationship related papers can be downloaded from the web: https://
between news and world indices and showed that the news economicpolicyuncertaintyinchina.weebly.com/.
drives world market movement. Sensoy et al. [6] explored the The investment sentiment index is constructed with the
strength of information flow between exchange rates and stock principal component method based on four underlying proxie,
prices by the effective transfer entropy with symbolic encoding i.e., the relative strength index, the psychological line index, the
methodology. Yang et al. [7] proposed an effective phase transfer trading volume and the adjusted turnover rate [40]. The Shanghai
entropy method based on the transfer entropy method. These Composite Index mostly reflects the performance of the Chinese
scholars also analyzed the relationship among 9 stock indices stock market. The data of this index are from the China Stock
from the U.S., Europe and China by using transfer entropy, Market & Accounting Research Database. The data period for all
effective transfer entropy, Rényi transfer entropy, and effective three time series is from February 2005 to May 2019 (Figure 1).
Rényi transfer entropy [8].
The vast influential literature focuses on the correlation
METHODOLOGY
between EPU and its impact of investment [9–21]. Many new
studies investigated the fluctuation characteristics of China’s We calculate the statistical causality between time series using
uncertainty index and its impact on the economy [22–24]. two different approaches. The first assumes linearity and employs
Additionally, many other studies focused on the impact of vector autoregressive techniques to estimate the extent to which
investor sentiment on stock prices. A study investigating the knowing the driving time series can help predict the dependent
correlation between investor sentiment and the equity market series. The second technique compares the difference in mutual
was conducted following De long’s research [25]. Baker and information between the independent case and the joint case to
Wurgler [26] investigated investor sentiment using a principal describe the success of predicting the dependent series. When
component method. Subsequently, a large number of studies predictability is increased by considering the past values of the
emerged, suggesting that investor sentiment has a significant driving variable, statistical causality is observed.
impact on stock returns [27–40]. In addition, some studies reveal
a complex correlation between online investor sentiment and
Linear Causality
stock market volatility [35, 36, 41, 42].
We model a time series as autoregressive by expressing its value
Although there are many studies on the correlation between
Yt at time t as a sum of the contributions over m distinct lagged
EPU and financial or economic indices, there are few studies on
series using the following linear equation:
the influence of EPU on investor sentiment. There is also a lack
of research on the transmission pathways of the EPU, investor m
sentiment and stock market. Yt =
X
βk(Y) Yt−k + εt (1)
It is widely considered that national economic policies will
k=1
pass relevant information to influence investor sentiment,
affect investment decisions, and ultimately cause stock
price fluctuations. However, the real situation may be more Where βk(Y) is a general coefficient term and εt is the residual.
complicated especially in China. Motivated by previous studies, Linear regression estimates coefficient parameters βk(Y) , which
we follow the theorizing on information flow to reveal the minimize the sum of squared residuals.
transmission path of information among these three variables To detect whether the values of some second time series
and to provide more references for macro policy makers. X anticipate the future values of Y, we can compare Equation
The rest of this paper is organized as follows. In section (1) with
Data Description, we describe the time series data used in
the models. In section Methodology, we introduce the effective m m
βk′(Y) Yt−k + βk′(X) Xt−k + εt′
X X
transfer entropy (ETE) method. In section An Improved Effective Yt = (2)
Transfer Entropy Method Based on a Sliding Window, we k=1 k=1
propose an improved ETE method based on sliding windows.
In section Empirical Results, we present the empirical results of We determine that distribution Y is Granger caused by X if the
EPU, trade and the exchange rate obtained using the improved residual in the second regression is significantly smaller than the
ETE model. Finally, we provide the conclusion of this paper. residual in the first regression. When this distribution holds, then

Frontiers in Physics | www.frontiersin.org 2 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

in the model residuals. Transfer entropy extends this concept by

considering the uncertainty instead of the variance. Adopting
Shannon’s measure of information [46], we can express the
uncertainty associated with random variable X by
X
H(X) = − p (x) . log p (x) (4)
x

where H(x) is the Shannon entropy of the distribution, and p(x)

represents the probability of X = x, which can be conditioned on
a second variable to give the conditional entropy:

H(Y|X) = H(X, Y) − H(X) (5)

When two random variables share information, the mutual

information is given by

I(X; Y) = H(Y) − H(Y|X) = H(X) − H(X|Y) (6)

The entropy of Y conditioned on two variables is

H(Y|X, Z) = H(X, Y, Z) − H(X, Z) (7)

and the conditional mutual information is therefore

I (X; Y|Z) = H (Y|Z) − H (Y|X, Z) (8)

Now, for each lag k, we can describe the information transfer

from Xt−k to Yt in terms of the following conditional
mutual information:
(k)


TEX→Y = I Yt ; Xt−k |Yt−k = H Yt |Yt−k


− H Yt |Xt−k , Yt−k (9)

This equation represents the resolution of uncertainty in

predicting Y when considering the past values of both Y and X
compared with considering the past values of Y alone.
Considering Equations (5) and (7), we can therefore represent
the transfer entropy for a single lag k, which is shown in
FIGURE 1 | Time series of the CN EPU index, Investor sentiment and Return Equation (9), in terms of four separate joint entropy terms.
of the Shanghai Composite Index’s closing price.
Following equation (4), these terms may be estimated from the
data using a non-parametric density estimation of the probability
distributions. For multivariate normal statistics, Equations (9)
there must be some information transfer from X to Y. Following and (3) coincide [45].
Geweke [44], we can represent the information transfer by
Effective Transfer Entropy (ETE)
It is a feature of the non-parametric estimation of entropy that
 
1 var(εt )
TEX→Y = log (3) the absolute scale of the transfer entropy measure has only
2 var(εt′ )
limited meaning; to detect causality, a relative position must be
where we adopt the transfer entropy notation (TE) following considered. A simple technique proposed by Marschinski and
Barnett et al. [45], whose result shows Granger causality Kantz [2] is the ETE, derived by subtracting from the observed
to be equivalent to transfer entropy for multivariate transfer entropy an average transfer entropy figure calculated
normal distributions. over independently shuffled time series, which destroys the
temporal order and hence any possible causality. We adopt
Non-linear Causality a shuffling approach producing 50 null-hypothesis transfer
To detect non-linear causality, we apply an information-theoretic entropy values from independently shuffled time series over the
approach. Equation (3) measures the extent to which the same domain containing no causality. By calculating the mean
additional information in the lagged variable reduces the variance and standard deviation of the shuffled transfer entropy figures,

Frontiers in Physics | www.frontiersin.org 3 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

we estimate the significance of a causal result as the distance of the structural change, which cannot be achieved using the
between the result and the average shuffled result standardized traditional linear TE method. We next verify the validity of
by the shuffled standard deviation: the algorithm.
First, we generate a time series X following the geometric
ETE = TE − TEshuffle (10) Brownian motion according to Equation (11) as follows:
TE − TEshuffle
Z: = Xt+1 = (1 + µ) Xt + σ Xt ηt (11)
σshuffle
where ηt is a noise obeying the standard normal
where TEshuffle is the mean of the shuffled values, and σshuffle is
distribution, ηt ∼ N(0, 1), and µ and σ represent the
the standard deviation. The shuffling of the time series destroys
drift coefficient and the diffusion coefficient, respectively. Y
temporality and should ensure that the mean is approximately
depends on X, and the equation is constructed as follows
zero; therefore, the spread of the data dictates the significance of
(Equation 12):
the result. Assuming that the distribution is close to Gaussian,
we can say that a result with Z > 3 is roughly in the top 1% of ′
Yt = (1 − α) Xt−k + αXt−k (12)
results and hence is comparable to a p-value of 0.01. The nature of
the method typically enables clearer significance to be observed where Xt−k ′ is another time series generated according to
with fewer shuffles, even without a strict Gaussian distribution; Equation (11). k is the given lag order, and α ∈ (0, 1) determines
thus, this method is computationally more attractive than the dependence strength between the series Y and X, i.e., the
the p-value. values of the transfer entropy.
This expression corresponds to the degree to which the Assuming k = 2, α = 0.5; k = 4, α = 0.5; and k = 5, α =
result lies in the right tail of the distribution of the zero- 0.5, we can obtain three time series with a length of 200, i.e.,
causality shuffled samples and hence how unlikely the result Xt , Ytk = 2 ; Xt , Ytk = 4 and Xt , Ytk = 5 according to Equations (11)
is due to chance. Therefore, the Z-score figure represents the and (12).
significance of the excess transfer entropy in the unshuffled case. As shown in Figure 2, for a correlation series with a single
We compute the Z-score in Equation (10) for both linear and lag structure, both the traditional transfer entropy, i.e., the linear
non-linear results. TE, and the improved TE method can capture the lag order
accurately. However, according to the Z-score significance test,
AN IMPROVED EFFECTIVE TRANSFER we can observe that when the temporal order is destroyed, the
ENTROPY METHOD BASED ON A SLIDING linear TE does not show significance in the relevant order; thus,
the linear TE method depends on time evolution. As shown
WINDOW
in Figure 3, the linear TE could only identify the order k =
Improved Method Based on a Sliding 4, which is the highest corresponding transfer entropy value
Window and Comparison With a Traditional (Z-score indicates that the value is above a significant level).
However, the improved TE could detect both k = 4 and k = 5.
Linear Method Moreover, as shown in Figure 4, we can also track the specific
Keskin and Aste [47] validated that the non-linear TE method
time period during which the lead-lag order fluctuates with the
would be useful for detecting a non-linear process. However,
improved method.
the lag order they found was global and unique and thus
We reshape Xt , Ytk = 2 ; Xt , Ytk = 4 and Xt , Ytk = 5 into
was unsuitable for capturing the accurate order between two
two new time series Xt′ , Yt′ , where i Xt′ = [Xt , Xt , Xt ]
non-stationary series. For non-stationary time series, the data h
structure changes over time, which means that the causal and Yt′ = Ytk = 2 , Ytk = 4 , Ytk = 5 . These new series
relationships also evolve dynamically. In addition, due to policy show obvious structure fluctuations, and the features
or unexpected events, the causal structure of real financial are more consistent with the characteristics of real
sequences tends to change over time. Therefore, it would be financial data.
inaccurate to use a single k to measure global causality. Due to the shortcomings of traditional linear methods in
Considering that the locality of non-stationary data may revealing dynamic orders, in the empirical analysis in section
be stationary or approximately stationary, this paper proposes Empirical Results, we apply the improved transfer entropy to
an improved transfer entropy method based on a sliding explore the information flow between all sequences. The sliding
window to solve the influence of a non-stationary data window length of all structures is 36 months with a forward step
structure on traditional transfer entropy. The improved method size of 1 month.
calculates the transfer entropy as described in section Non-
linear Causality but is limited to a certain time segment. Comparison With the Granger Causality
Through forward scrolling, the transfer entropy at each time Test
point is obtained, and the causal relationship between the The Granger causality test is essentially a test used to determine
two times series can be revealed. In addition to its ability to whether a lagging variable can be introduced into an equation
capture the structural changes between two time series, the containing other variables. If a variable is affected by the
improved method can help us trace the specific time period lag of other variables, the variables are considered to have

Frontiers in Physics | www.frontiersin.org 4 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

FIGURE 2 | Demonstration that both methods identify the true lag values with maximal transfer entropy. Non-linear transfer entropy is calculated using a
quantile-binned histogram, of 6 classes per dimension over 2,500 points. The Z-score for each result is also plotted for both methods. According to the z > 3
principle, it can be concluded that for two time series with a single lag order, the two methods can both identify the lag orders accurately. (A) k = 2, α = 0.5. (B)
k = 4, α = 0.5. (C) k = 5, α = 0.5.

Frontiers in Physics | www.frontiersin.org 5 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

FIGURE 3 | Demonstration that both methods identify the true lag values with maximal transfer entropy. The linear TE could only capture k = 2, corresponding to the
highest transfer entropy value (Z-score indicates that the value is above a significant level), while the improved TE method could detect k = 2, 4, and 5.

Granger causality is a regression-based interpretation of

Wiener’s causality definition [48]. In this section, the Granger
causality test is employed as a comparison with the improved
TE to detect the true lag orders. Following Granger’s work [49],
we model the Granger causality test with the following two
regression equations:

p
X
Xt = αi Xt−i + ui (13)
i=1
p p
X X
Xt = bi Xt−i + ci Yt−i + vt (14)
FIGURE 4 | Order identification by the two transfer entropy methods. The i=1 i=1
dashed line k = 2 corresponds to the lag order when the transfer entropy
value is the largest in the linear TE method in Figure 3. where X denotes the object needed to find the Granger cause, Y
denotes the object needed to determine whether it can Granger
cause X, and residuals ut and vt are assumed to be mutually
Granger causality. For the sequences X and Y, using different lag independent and individually distributed with a zero mean
orders, we obtain the causality test results of the two sequences and constant variance. These equations were tested using the
(Table 1). following hypothesis:

Frontiers in Physics | www.frontiersin.org 6 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

TABLE 1 | Granger causality test results.

Lag 1 2 3 4 5 6 7 8 9 10

F_test 0.5664 4.0797 2.9568 2.2895 6.6669 5.8331 4.9915 5.0310 4.4082 3.9662
P_val 0.4520 0.0174 0.0319 0.0586 4.76E-06* 6.48E-06* 1.68E-05* 4.75E-06* 1.38E-05* 3E-05*

*p_val < 0.01 indicates that the test result significantly rejects the null hypothesis and that at least one lag variable X Granger causes Y . Therefore, the orders in which X Granger
causes Y are 6, 7, 8, 9, and 10.

FIGURE 5 | Comparison of the Granger causality method based on a sliding window and the improved TE method. The widow length in both methods is W = 36,
and the significant level is 1%. The Granger method can clearly identify k = 2 but cannot identify k = 4, and there is considerable noise interference when identifying
k = 5. The improved TE method can clearly identify the three orders 2, 4, and 5. (A) Granger causality test based on the sliding window method. The gray part
indicates that the p-value of the F statistic is <1%, indicating that the causal relationship is significant in this area. (B) The result based on the improved TE method.
The gray part indicates that the Z-score is higher than 3, which is equivalent to a significance level of p_val < 0.01 [section Effective Transfer Entropy (ETE)].

H0 : Y does not Granger cause X(c1 = c2 = ... = cp = 0). (a) The maximum of the lag value p is set to a fixed number, such
The F − test can be expressed as follows: as 10.
(b) By calculating the total AIC of Equations (13) and (14)
(RSS0 − RSS1 ) /p
by traversing the p value from 1 to 10, we obtain the
F=
∼ F p, n − 2p − 1 (15) corresponding p of the minimum AIC. The experimental
RSS1 / n − 2p − 1
results show that the optimal p is 5.
(c) Equations (13) and (14) are estimated by OLS with p = 5.
where RSS0 is the residual sum of squares of Equation (13),
(d) F and F(p, n − 2p − 1) are calculated according to Equation
RSS1 is the residual sum of squares of Equation (14), n is the
(15). The results show that F = 4.0635 and F(p, n−2p−1) =
number of observations, and p is a lag value. We reject the
3.8549 (at the 99% confidence level).
hypothesis H0 and accept that Y is a Granger cause of X if
(e) If F>F(p,n–2p−1), we conclude that Y can significantly
and only if F > F(p, n − 2p − 1). The model order p can
Granger cause X.
be determined by minimizing the AIC [50], which is defined
(f) The window is moved forward by a 1-month step, and steps
as follows:
(a–e) are repeated.
2m2 p Using the process described above, we obtained the Granger
AIC(p) = 2 log (|σ |) + (16)
n̂ causality test results based on a window length W = 36
(Figure 5A). As shown in Figure 5A, although the Granger
where σ is the estimated noise covariance, m is the dimension causality test can identify k = 2, it cannot effectively capture the
of the stochastic process and n̂ is the length of the data two orders of 4 and 5. Using k = 4 cannot pass the significance
window used to estimate the model. For example, to detect test; although using k = 5 can pass the significance test, there
the causal relationship from exports to US EPU, Y should be may be other orders, such as k = 8. The improved TE method
set to the exports sequence, while X should be set to the US can accurately identify three different orders (Figure 5B). In
EPU sequence. In contrast, Y should be set to the US EPU addition, the stage during which the order jumps cannot pass the
before detecting the causal relationship between US EPU and significance test.
the exports.
The Granger causality test based on the sliding window
method can also obtain the order and significance of two series’ EMPIRICAL RESULTS
correlation. Using Y and X as an example, we elaborate upon the
processes of the Granger model estimation within a fixed window Since traditional linear methods cannot identify dynamic orders
as follows: between time series or track specific lead-lag orders when

Frontiers in Physics | www.frontiersin.org 7 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

structural fluctuations occur, we apply the improved transfer EPU and Investor Sentiment
entropy to explore the information flow among EPU, investor Based on the dynamic TE method, we analyze the causal
sentiment and the stock market. relationship between EPU and investor sentiment. As shown in

FIGURE 6 | Dynamic Entropy results between EPU and investor sentiment. (A) Lag structure. (B) TE fluctuation. (C) Z-score.

Frontiers in Physics | www.frontiersin.org 8 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

Figure 6A, there is an obvious dynamic order in the correlation bandwagon effect indicates that higher investor sentiment could
between EPU and investor sentiment. increase the stock price, which is reflected in the positive
As shown in Figure 6C, a Z-score > 3 is mainly located correlation between stock prices and sentiment during the same
in August 2015–June 2016. This means that during this time period; in contrast, the bargain shopper effect indicates that
period, EPU had a significant impact on investor sentiment, investors optimistically believe the shares at a relatively low
and uncertain information about national economic policies price represent a purchase opportunity; therefore, their sentiment
significantly affected investor sentiment. From Figure 6B, it can negatively changes the returns.
also be seen that in this stage, the EPU’s transfer entropy to The bargain shopper and bandwagon effect make it difficult
investor sentiment was significantly higher than the impact of to explore the causality between investor sentiment and stock
investor sentiment on EPU. returns. In our analysis results, the bandwagon effect is
The impact on investor sentiment is related to the nature of weaker, and the bargain shopper effect is more significant. The
the policy, i.e., whether the policy is a domestic policy or a foreign bandwagon effect reflects the herd effect of investors. This effect
policy. During this period, China’s economic policy involved the makes the stock market prone to sudden rises and falls in the
following two important measures: a change in the CNY fixing short term; it cannot reflect the true value of a company and
mechanism and the launch of the “circuit-breaker” mechanism. is not conducive to the healthy and stable development of the
On August 11, 2015, the central bank made more reference stock market.
to the closing price of the previous day in the daily CNY-
USD mid-price quotation formation mechanism. This change EPU and Stock Market
makes the method of forming the middle price more market- If investor sentiment has a significant impact on the stock price,
oriented, which more closely reflects the actual supply-demand then according to our expectations, national policy information
relationship of the market compared to the previous method. will be transmitted to the stock price through investors’
The circuit-breaker benchmark index is the CSI 300 Index, expectations and eventually cause stock price fluctuations;
which uses two thresholds of 5 and 7%. When the CSI 300 Index in other words, EPU also has some kind of information
triggers a 5% breaking threshold, the three exchanges suspend transmission relationship with the stock price. However, the
trading for 15 min, and if the 5% is triggered late in the day or 7% results now show that both stock price fluctuations and EPU have
is triggered at any time throughout the day, trading is suspended an effect on investor sentiment and are not affected by investor
until the market closes. From January 4th to January 7th, the sentiment. Therefore, either the stock price fluctuations and EPU
breaking mechanism was implemented for only 3 days, and it have a weak information transmission effect or there is a mutually
became the shortest-lived stock market policy in the history of offsetting effect.
Chinese securities. This policy uncertainty had a great impact on To further verify our assumptions, we explore the information
investor sentiment. transfer relationship between EPU and stock price returns
Before August 2015, there were incidents such as the (Figure 8). As shown from the results of Figure 8C, there are only
bankruptcy of Lehman Brothers (September 2008), the a few discontinuous time points with a Z-score > 3 in the entire
downgrade of the US sovereign credit rating (August 2011), and event period. Overall, the information transmitted by the EPU
the European debt crisis (January 2011–January 2014). However, to the stock market is non-significant; in other words, the EPU
probably because these events did not occur in China, their has no obvious information transmission relationship with the
impact on consumer sentiment was not significant. stock market.
A considerable number of related studies showed that the
stock market and EPU are significantly negatively correlated
Investment Sentiment Index and the Stock [11, 12, 21, 52]. Regarding the relationship between China’s EPU
Market and the stock market, Chen and Chiang [21] also verified that
The correlation between sentiment and stock price returns is the stock returns in China are negatively correlated with EPU.
illustrated in Figure 7. As shown in Figure 7C, the impact of Notably, the main correlation revealed by Chen and Chiang
sentiment on stock returns is non-significant; in contrast, the based on the GARCH method is the overall correlation between
fluctuation in stock price returns has a significant impact on sequences. However, we reveal a time-varying relationship
investor sentiment throughout the time period. This shows that between sequences based on non-linear methods. As shown in
in the Chinese stock market, using emotions to predict changes Figure 8C, it can be concluded that in the short term, China’s
in stock prices is useless, and investor sentiment is more a EPU also significantly impacts the stock market during the period
lagging reflection of stock price returns. Figure 7A shows that from 2011 to 2012 and in 2016, but in the long run, this effect is
the lag time is approximately 1 month. Our results further verify generally not significant.
the long-term correlation characteristics suggesting that investor
sentiment is mainly affected by fluctuation in the market, which DISCUSSION
may be related to the existence of cyclical fluctuations in the
market and futures arbitrage [40]. According to the efficient market hypothesis theory, an efficient
The study conducted by Brown and Cliff [29, 51] revealed market (Figure 9A) should reflect all changes in information,
that the bandwagon effect and bargain shopper effect can offset including regular investor sentiment changes and shocking policy
each other, reducing the predictability of stock returns. The fluctuations. Therefore, the information flow should flow from

Frontiers in Physics | www.frontiersin.org 9 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

FIGURE 7 | Entropy between investor sentiment and stock market. (A) Lag structure. (B) TE fluctuation. (C) Z-score.

the EPU and investor sentiment to the stock market. In addition, The results show that the Shanghai Stock market is not
since policy shocks often affect sentiment in the short term, yet an efficient market (Figure 9B) and cannot reflect
information flow should flow from policy to sentiment, but this information from regular investment and low-frequency
is uncertain. policy shocks. Therefore, investors can reap potential excess

Frontiers in Physics | www.frontiersin.org 10 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

FIGURE 8 | Entropy between EPU and stock market. (A) Lag structure. (B) TE fluctuation. (C) Z-score.

profits through operations. Furthermore, the stock market Compared with the market and policy factors, investor
cannot form an effective path to reflect investor sentiment sentiment has a certain lag (Figure 9B), reflecting the volatility
information; thus, in the long run, EPU cannot affect the information of the two. Therefore, we should consider policy
stock market. factors when studying the construction of investor sentiment

Frontiers in Physics | www.frontiersin.org 11 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

FIGURE 9 | The correlation among CN EPU, investor sentiment and the stock market. (A) The efficient market hypothesis. The stock market may be able to
effectively reflect information regarding conventional investment and policy shocks. (B) China’s stock market is an inefficient market, and stock market volatility is an
important factor affecting emotional volatility.

indicators, which is rarely investigated in research concerning the changes in stock price returns with a 1-month lag order.
factors affecting investor sentiment. The results show that in the Chinese stock market, the
bargainer effect is more significant and the bandwagon effect
CONCLUSION is weaker.
There is no direct information flow from EPU to stock
A widely accepted fact is that economic policy affects investor market, and according to our previous analysis, there is
sentiment and will be ultimately reflected in the stock market no indirect information flow through which EPU transmits
through investment decisions, causing stock price volatility. information to the stock market through investor sentiment.
Therefore, is this really the case? Therefore, stock market fluctuations are basically not affected
Since traditional linear methods cannot identify the dynamic by national policy fluctuations. Although investor sentiment
orders between time series and are unable to track specific is affected by changes such as exchange rate reform and
lead-lag orders when structural fluctuations occur, we proposed stock market policies, this effect is reflected only at the
an improved transfer entropy method based on a sliding emotional level. Many investors can digest and neutralize
window. By comparing with the linear ETE method and Granger extreme emotions. Therefore, a final consensus is not easy
causality method, we verify the effectiveness of the improved to form.
method. The main advantages of this methodology are the
easy implementation-interpretation by non-parametricity to DATA AVAILABILITY STATEMENT
capture the non-linear dynamics and the point in time when
the structure changes. Therefore, this method is considered All datasets generated for this study are included in the
a nice and promising alternative to the standard measures. article/supplementary material.
We further employ this improved method to examine
the information flow among EPU, investor sentiment and AUTHOR CONTRIBUTIONS
stock market.
The results of the information flow analysis of EPU All authors analyzed and discussed the results, and contributed
and investment sentiment show that EPU influenced to the draft paper and revisions.
investor sentiment mainly from August 2015 to June
2016. Among different policies, China’s exchange rate FUNDING
reform policy and “circuit-breaker” policy have played
an important role. For other time periods, there are This work was supported by the National Social Science
also points in time when policies were highly uncertain, Foundation of China (Grant No. 19BJL086), Natural Science
such as the bankruptcy of Lehman Brothers (September Foundation of Guangdong Province of China (Grant Nos.
2008), the downgrade of the US sovereign credit rating 2019A1515010471, 2017A030313396), MOE (Ministry of
(August 2011), and the European debt crisis (January 2011– Education in China) Project of Humanities and Social Sciences
January 2014). However, likely because these events did (Project No. 17YJAZH109), Fundamental Research Funds for the
not occur in China, their impact on consumer sentiment Central Universities (Grant Nos. 2019MS082, 2020ZDPY19), and
was non-significant. Guangzhou National Innovation-oriented City Development
The analysis of the information flow between sentiment Research Center (Grant No. 2017IC02).
and stock price returns shows that the impact of sentiment
on returns is non-significant, while the fluctuation in stock ACKNOWLEDGMENTS
price returns has a significant impact on investor sentiment.
Therefore, using emotions to predict changes in stock prices We would like to acknowledge the reviewers for their
is valueless. Investor sentiment is more a reflection of constructive comments.

Frontiers in Physics | www.frontiersin.org 12 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

REFERENCES 23. Jiang YH, Zhu ZX, Tian GY, Nie H. Determinants of within and cross-country
economic policy uncertainty spillovers: evidence from US and China. Finance
1. Kwon O, Yang J-S. Information flow between stock indices. EPL. (2008) Res Lett. (2019) 31:195–206. doi: 10.1016/j.frl.2019.08.004
82:68003. doi: 10.1209/0295-5075/82/68003 24. Chen L, Du Z, Hu Z. Impact of economic policy uncertainty on
2. Marschinski R, Kantz H. Analysing the information flow between financial exchange rate volatility of China. Finance Res Lett. (2020) 32:101266.
time series. Eur Phys J B. (2002) 30:275–81. doi: 10.1140/epjb/e2002-00379-2 doi: 10.1016/j.frl.2019.08.014
3. Kyrtsou C, Kugiumtzis D, Papana A. Further insights on the relationship 25. De Long JB, Andrei S, Lawrence HS, Robert JW. Noise trader risk in financial
between SP500, VIX and volume: a new asymmetric causality test. Europ J markets. J Polit Econ. (1990) 98:703–38. doi: 10.1086/261703
Finance. (2019) 25:1402–19. doi: 10.1080/1351847X.2019.1599406 26. Baker M, Wurgler J. Investor sentiment and the cross-section of stock returns.
4. Dimpfl T, Peter FJ. Group transfer entropy with an application J Finance. (2006) 61:1645–80. doi: 10.1111/j.1540-6261.2006.00885.x
to cryptocurrencies. Phys A Stat Mech Appl. (2019) 516:543–51. 27. Fisher KL, Statman M. Investor sentiment and stock returns. Financ Anal J.
doi: 10.1016/j.physa.2018.10.048 (2000) 56:16–23. doi: 10.2469/faj.v56.n2.2340
5. García-Medina A, Sandoval L, Bañuelos EU, Martínez-Argüello AM. 28. Baker M, Stein JC. Market liquidity as a sentiment indicator. J Financial Mark.
Correlations and flow of information between the New York times (2004) 7:271–99. doi: 10.1016/j.finmar.2003.11.005
and stock markets. Phys A Stat Mech Appl. (2018) 502:403–15. 29. Brown GW, Cliff MT. Investor sentiment and asset valuation. J Bus. (2005)
doi: 10.1016/j.physa.2018.02.154 78:405–40. doi: 10.1086/427633
6. Sensoy A, Sobaci C, Sensoy S, Alali F. Effective transfer entropy approach 30. Baker M, Wurgler J. Investor sentiment in the stock market. J Econ Perspect.
to information flow between exchange rates and stock markets. Chaos (2007) 21:129–52. doi: 10.3386/w13189
Soliton Fract. (2014) 68:180–5. doi: 10.1016/j.chaos.2014.08.007 31. Kaplanski G, Levy H. Sentiment and stock prices: the case of aviation disasters.
7. Yang PB, Shang PJ, Lin AJ. Financial time series analysis based on effective J Finan Econ. (2010) 95:174–201. doi: 10.1016/j.jfineco.2009.10.002
phase transfer entropy. Phys A Stat Mech Appl. (2017) 468:398–408. 32. Corredor P, Ferrer E, Santamaria R. Investor sentiment effect in stock markets:
doi: 10.1016/j.physa.2016.10.085 stock characteristics or country-specific factors? Int Rev Econ Finance. (2013)
8. He JY, Shang PJ. Comparison of transfer entropy methods for 27:572–91. doi: 10.1016/j.iref.2013.02.001
financial time series. Phys A Stat Mech Appl. (2017) 482:772–85. 33. Yang C, Gao B. The term structure of sentiment effect in stock
doi: 10.1016/j.physa.2017.04.089 index futures market. N Amer J Econ Finance. (2014) 30:171–82.
9. Chung SL, Hung CH, Yeh CY. When does investor sentiment doi: 10.1016/j.najef.2014.09.001
predict stock returns? J Empir Finance. (2012) 19:217–40. 34. Yang C, Zhou L. Investor trading behavior, investor sentiment and asset prices.
doi: 10.1016/j.jempfin.2012.01.002 N Amer J Econ Finance. (2015) 34:42–62. doi: 10.1016/j.najef.2015.08.003
10. Dakhlaoui I, Aloui C. The interactive relationship between the US economic 35. Yao CZ, Sun BY, Lin JN. A study of correlation between investor sentiment
policy uncertainty and BRIC stock markets. Int Econ. (2016) 146:141–57. and stock market based on copula model. Kybernetes. (2017) 46:550–71.
doi: 10.1016/j.inteco.2015.12.002 doi: 10.1108/K-10-2016-0297
11. Arouri M, Estay C, Rault C, Roubaud D. Economic policy uncertainty and 36. Renault T. Intraday online investor sentiment and return patterns
stock markets: long-run evidence from the US. Finance Res Lett. (2016) in the U.S. stock market. J Bank Finance. (2017) 84:25–40.
18:136–−41. doi: 10.1016/j.frl.2016.04.011 doi: 10.1016/j.jbankfin.2017.07.002
12. Antonakakis N, Chatziantoniou I, Filis G. Dynamic co-movements of stock 37. Xu Q, Wang L, Jiang C, Zhang X. A novel UMIDAS-SVQR model
market returns, implied volatility and policy uncertainty. Econ Lett. (2013) with mixed frequency investor sentiment for predicting stock market
120:87–92. doi: 10.1016/j.econlet.2013.04.004 volatility. Expert Syst Appl. (2019) 132:12–27. doi: 10.1016/j.eswa.2019.
13. Antonakakis N, André C, Gupta R. Dynamic co-movements between 04.066
economic policy uncertainty and housing market returns. J Real Estate Portfol 38. Mascio DA, Fabozzi F. Sentiment indices and their forecasting
Manage. (2015) 21:53–60. doi: 10.5555/1083-5547-21.1.53 ability. Int J Forecast. (2019) 38:257–76. doi: 10.1002/fo
14. Wisniewski TP, Lambe BJ. Does economic policy uncertainty drive CDS r.2571
spreads? Int Rev Finan Anal. (2015) 42:447–58. doi: 10.1016/j.irfa.2015.09.009 39. David EA, Michael M, Abhay KS. Daily market news sentiment and
15. Li XM, Peng L. US economic policy uncertainty and co-movements stock prices. Appl Econ. (2019) 51:3212–35. doi: 10.1080/00036846.2018.15
between Chinese and US stock markets. Econ Model. (2017) 61:27–39. 64115
doi: 10.1016/j.econmod.2016.11.019 40. Yao CZ, Li HY. Time-varying lead–lag structure between investor
16. Hammoudeh S, Kim WJ, Sarafrazi S. Sources of fluctuations in islamic sentiment and stock market. N Amer J Econ Finance. (2020) 52:101148.
US, EU, and Asia equity markets: the roles of economic uncertainty, doi: 10.1016/j.najef.2020.101148
interest rates, and stock indexes. Emerg Mark Financ Tr. (2015) 52:1–15. 41. Corea F. Can twitter proxy the investors’ sentiment? The
doi: 10.1080/1540496X.2014.998561 case for the technology sector. Big Data Res. (2016) 4:70–74.
17. Sarwar G, Khan W. The effect of US stock market uncertainty on doi: 10.1016/j.bdr.2016.05.001
emerging market returns. Emerg Mark Financ Tr. (2017) 53:1796–811. 42. Ballinari D, Behrendt S. Structural breaks in online investor sentiment: a
doi: 10.1080/1540496X.2016.1180592 note on the nonstationarity of financial chatter. Finance Res Lett. (2020).
18. Tsai IC. The source of global stock market risk: a viewpoint doi: 10.1016/j.frl.2020.101479. [Epub ahead of print].
of economic policy uncertainty. Econ Model. (2017) 60:122–31. 43. Huang Y, Luk P. Measuring economic policy uncertainty in China. China Econ
doi: 10.1016/j.econmod.2016.09.002 Rev. (2020) 59:101367. doi: 10.1016/j.chieco.2019.101367
19. Li XL, Balcilar M, Gupta R, Chang T. The causal relationship between 44. Geweke JF. Measures of conditional linear dependence and feedback
economic policy uncertainty and stock returns in China and India: evidence between time series. J Amer Statistical Assoc. (1984) 79:907–15.
from a bootstrap rolling window approach. Emerg Mark Financ Tr. (2016) doi: 10.1080/01621459.1984.10477110
52:674–89. doi: 10.1080/1540496X.2014.998564 45. Barnett L, Barrett A, Seth A. Granger causality and transfer entropy
20. Yao CZ, Liu C, Ju WJ. Multifractal analysis of the WTI crude oil market, are equivalent for gaussian variables. Phys Rev Lett. (2009) 103:238701.
US stock market and EPU. Phys A Stat Mech Appl. (2020) 550:124096. doi: 10.1103/PhysRevLett.103.238701
doi: 10.1016/j.physa.2019.124096 46. Shannon C. A mathematical theory of communication. Bell Syst Tech J. (1948)
21. Chen X, Chiang TC. Empirical investigation of changes in policy uncertainty 27:379–423. doi: 10.1002/j.1538-7305.1948.tb01338.x
on stock returns—evidence from China’s market. Res Int Bus Finance. (2020) 47. Keskin Z, Aste T. Information-theoretic measures for non-linear causality
53:101183. doi: 10.1016/j.ribaf.2020.101183 detection: application to social media sentiment and cryptocurrency prices.
22. Dai PF, Xiong X, Zhou WX. Visibility graph analysis of economy arXiv:1906.05740v2. (2019).
policy uncertainty indices. Phys A Stat Mech Appl. (2019) 531:121748. 48. Wiener N. The Theory of Prediction. Modern Mathematics for Engineers. New
doi: 10.1016/j.physa.2019.121748 York, NY: McGraw-Hill Book Company Inc (1956).

Frontiers in Physics | www.frontiersin.org 13 June 2020 | Volume 8 | Article 206

Yao and Li Dynamic Effective Transfer Entropy

49. Granger C. Investigating causal relations by econometric models and Conflict of Interest: The authors declare that the research was conducted in the
cross-spectral methods. Econometrica. (1969) 37:424–38. doi: 10.2307/19 absence of any commercial or financial relationships that could be construed as a
12791 potential conflict of interest.
50. Akaike H. Fitting autoregressive models for prediction. Ann Inst Stat Math.
(1969) 21:243–7 doi: 10.1007/BF02532251 Copyright © 2020 Yao and Li. This is an open-access article distributed under the
51. Brown GW, Cliff MT. Investor sentiment and the near-term stock market. J terms of the Creative Commons Attribution License (CC BY). The use, distribution
Empir Finance. (2004) 11:1–27. doi: 10.1016/j.jempfin.2002.12.001 or reproduction in other forums is permitted, provided the original author(s) and
52. Alexopoulos M, Cohen J. The power of print: uncertainty shocks, the copyright owner(s) are credited and that the original publication in this journal
markets, and the economy. Int Rev Econ Finance. (2015) 40:8–28. is cited, in accordance with accepted academic practice. No use, distribution or
doi: 10.1016/j.iref.2015.02.002 reproduction is permitted which does not comply with these terms.

Frontiers in Physics | www.frontiersin.org 14 June 2020 | Volume 8 | Article 206