Finance Research Letters xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Finance Research Letters

journal homepage: www.elsevier.com/locate/frl

Asymmetric dependence between economic policy uncertainty and

stock market returns in G7 and BRIC: A quantile regression
approach
Guo Penga,b,*, Zhu Huimingb, You Wanhaic
a
School of Economics and Trade, Henan University of Technology, Zhengzhou 450001, P.R. China
b
College of Business Administration, Hunan University, Changsha 410082, PR China
c
School of Economics and Management, Fuzhou University, Fuzhou 350116, PR China

AR TI CLE I NF O AB S T R A CT

Keywords: This paper employs the quantile regression techniques to examine the dependence structure
Economic policy uncertainty between economic policy uncertainty (EPU) and stock market returns in G7 and BRIC. We find
Stock market new evidence to support the view that EPU will reduce stock market returns, with the exception
Asymmetry dependence of France and the UK. Our results show that eight out of ten stock markets reveal asymmetric
Quantile regression
dependence with EPU. Moreover, there is no dependence between EPU and France/the UK stock
JEL classification: market.
C32
E44
G15
G18

1. Introduction

The effect of economic policy uncertainty constructed by Baker et al. (2013) on economic activities has received increasing
attention among researchers and policy-makers (Colombo, 2013; Bloom, 2014; Li et al., 2015). Unlike previously research papers
which focus on the relationship between EPU and the US economic activity (Jones and Olson, 2013), the aim of this paper is to
contribute towards the effects of economic policy uncertainty on stock market performance in G7 and BRIC. In particular, according
to Baur's (2013) approach, we explore the structure and degree of dependence between EPU and stock market returns in G7 and BRIC
using Koenker's (1978) quantile regression techniques that permit us to analyze potential asymmetric dependence. Thus, we are able
to investigate differences in dependence across stock returns of positive and negative sign. Moreover, understanding how the returns
react to economic policy uncertainty when the stock market is bullish or bearish could be very useful to investors and policy-makers
(Chevapatrakul, 2015).

2. Methodology

In this paper we firstly employ the panel quantile regression model with fixed effect (Koenker, 2004) that takes into account
unobserved country heterogeneity to investigate the effect of EPU on stock market returns of G7 and BRIC. In addition, we should
eliminate the influence of the lagged return (β(τ)) and effect of both extreme positive as well as extreme negative previous month's

*
Corresponding author.
E-mail addresses: guophnufe@163.com (P. Guo), zhuhuiminghnu@sina.com (H. Zhu), 524732473@qq.com (W. You).

https://doi.org/10.1016/j.frl.2017.11.001
Received 24 May 2016; Received in revised form 10 April 2017; Accepted 1 November 2017
1544-6123/ © 2017 Elsevier Inc. All rights reserved.

Please cite this article as: Guo, P., Finance Research Letters (2017), http://dx.doi.org/10.1016/j.frl.2017.11.001
P. Guo et al. Finance Research Letters xxx (xxxx) xxx–xxx

returns (γ(τ)) in the following regression model, which also covers the presence of asymmetry dependence and estimates the slope
parameter at different quantiles. Consider panel quantile regression model with fixed effects as

QREt, i (τ REt − 1, i, LEPUt , i ) = αi + β (τ ) REt − 1, i + γ (τ ) REt − 1, i I ( REt − 1, i > RE q) + θ (τ ) LEPUt , i (1)

where REt,i is the monthly return of country i at month t. LEPUt,i is the logarithm of economic policy uncertainty index of country i at
time t. Q(τ|•) denotes a conditional quantile function of stock market return, evaluated at τ th quantile, where τ ∈ (0, 1). β(τ) is the
coefficient of the lagged return. The θ(τ) measures the dependence degree of stock returns at τ th quantile to LEPUt,i, which is the
focus of this study. The indicator variable I(|REt − 1, i| > REq) is equal to one if the stock return i at month t − 1 exceeds a certain
threshold REq and zero otherwise. REq is selected to be the 95% quantile of the unconditional distribution of absolute returns of the
respective stock.
Using panel quantile regression model, we could obtain the overall effect of EPU on G7 and BRIC. To examine whether economic
policy uncertainty has any significant influence on the stock market returns in each country, quantile autoregression approaches
(Koenker and Xiao, 2006) are used. Similarly, we also eliminate the influence of the lagged return (βj,i(τ)) and extreme return (γi(τ))
in the following regression model.
p
Qτ (REt , i Ft − 1, LEPUt , i ) = αi (τ ) + ∑ βj,i (τ ) REt−j,i + γi (τ ) REt−1,i I ( REt−1,i > RE q) + θi (τ ) LEPUt , i
j=1 (2)

where Ft − 1 is the lagged return of the respective stock. βj,i(τ) is quantile specific autoregressive parameter. The lag order p is
determined by Bayesian Information Criterion. The θi(τ) measures the dependence degree of stock returns at τth quantile to LEPUt,i,
which is ours target parameter.

3. Data

In this paper we use monthly data of the economic policy uncertainty indices for G7 countries, that is, Canada, Germany, France,
Italy, the UK, Japan and the USA, as well as, for BRIC countries, namely, Russia, China and India. It is worth noting that the choice of
countries and sample period is directed by data availability provided by Baker et al. (2013). Despite the fact that our study is limited
to countries, it represents a tremendous portion of the global economy. To measure the dependence structure more accurately, we use
exchange rates, obtained from the Investing.com, to convert local stock prices to dollars. The stock prices are: FCHI for France, GDAXI
for Germany, N225 for Japan, S&P 500 for USA, FTSE 100 for UK, GSPTSE for Canada, MIB for Italy, MICEX for Russia, BSESN for
India and SSECI for China. Monthly stock returns are calculated as REi,t = 100 × log (Pi,t/Pi,t − 1), where Pi,t is the monthly stock
market indices that are extracted from Yahoo Finance (http://finance.yahoo.com/). The logarithm of economic policy uncertainty
(i.e. LEPU) indices are used and obtained from the website (http://www.policyuncertainty.com/). The descriptive statistics results
are presented in Table 1.
As evident in Table 1, EPU indices have a positive average value, with a low standard deviation. With regard to stock market
returns, we observe a positive mean values, with the exception of Italy and Japan. We also notice that Japan and India present the
lowest and highest mean values respectively. Stock market returns are fairly volatile, as shown by the standard deviation. Fur-
thermore, the J-B statistics indicate that none of the series is normally distributed, except for the EPU indices of Germany, Italy, Japan
and India. According to the ADF unit root test, all variables are stationary.
The unconditional correlation between each stock market returns is reported in Table 2. It shows that the correlation is not low,
which means that the integration among these stock markets is very high. Germany–France, Italy–France and the UK–France are the
high correlations between their stock markets, however, lower correlation in China and other countries. That is, compared to other
stock markets, the integration is not high between China and other countries. The highest/lowest correlation is Germany–France and
Italy–China, respectively.
The problem with simple correlation coefficients is that it can only capture linear relationships and does not exhibit time-varying
behavior. In this subsection, we provide a rolling correlation analysis of the stock returns. The rolling correlation analysis could
address the time-varying nature of the correlation coefficients. A different picture emerges when the correlations are calculated in a
rolling window. Take the USA as an example,1 Fig. 1 presents the rolling correlation coefficients between the USA and other
countries. It shows that the correlation coefficients are time-varying and include some periods where it is negative (China and the
USA), which is a desired feature for portfolio diversification. Table 3 provides further information on the descriptive statistics of the
rolling correlation coefficients. The range of the correlation coefficients is very wide. For example, the maximum correlation between
stock returns in France and Germany is as high as 0.972 but the minimum figure is −0.275 in China and India. According to the
Table 3, there are only the negative correlations between China and other countries. Nevertheless, the degree of integration among
these countries is gradually increasing.

1
The results of other countries can be available upon request to the corresponding author.

2
P. Guo et al. Finance Research Letters xxx (xxxx) xxx–xxx

Table 1
Descriptive statistics.

Country Variable Period Observation Mean S.D. J-B ADF

USA RE 1985.2–2015.8 367 0.653 4.420 277*** −17.867***

LEPU 4.637 0.285 9.878*** −5.366***
Canada RE 1996.7–2015.8 230 0.471 6.231 184.3*** −13.128***
LEPU 4.657 0.541 5.356* −4.223***
Japan RE 1988.7–2015.8 326 −0.093 6.504 8.78** −17.736***
LEPU 4.541 0.342 3.248 −9.147***
UK RE 1997.2–2015.8 223 0.140 4.872 43.43*** −13.193***
LEPU 4.695 0.624 7.485** −3.803***
France RE 1990.4–2015.8 305 0.243 6.130 27.68*** −16.840***
LEPU 4.671 0.629 7.708** −5.943***
Germany RE 1993.2–2015.8 271 0.651 6.780 92.42*** −15.898***
LEPU 4.623 0.435 0.169 −8.526***
Italy RE 1998.1–2015.8 212 −0.090 7.436 13.99*** −13.681***
LEPU 4.617 0.359 0.473 −7.460***
Russia RE 1997.11–2015.8 214 0.406 14.171 199.5*** −11.977***
LEPU 4.471 0.668 7.295** −8.366***
India RE 2003.2–2015.8 151 1.167 8.568 20.86*** −10.694***
LEPU 4.477 0.543 2.625 −4.635***
China RE 1995.2–2015.8 247 0.831 8.373 10.55*** −14.559***
LEPU 4.538 0.606 12.15*** −8.574***

Notes: *, ** and *** indicate statistical significance at the 10%, 5% and 1% significance levels respectively.

Table 2
Correlation coefficients of stock market returns.

Canada China France Germany India Italy Japan Russia UK USA

Canada 1
China 0.386 1
France 0.701 0.293 1
Germany 0.652 0.308 0.904 1
India 0.629 0.348 0.607 0.618 1
Italy 0.634 0.287 0.901 0.803 0.578 1
Japan 0.595 0.324 0.654 0.637 0.584 0.587 1
Russia 0.658 0.310 0.552 0.506 0.567 0.506 0.521 1
UK 0.761 0.311 0.880 0.806 0.619 0.794 0.627 0.556 1
USA 0.782 0.352 0.825 0.797 0.624 0.743 0.672 0.540 0.846 1

4. Results

4.1. Panel quantile regression results

We begin our analysis by concentrating on the graph of panel quantile regression coefficients that are depicted in Fig. 2. It reports
estimated coefficients of EPU in the Total, G7 and BRIC. For the total sample, the EPU coefficients are generally negative, large and
statistically significant at the lower and intermediate quantiles. It provides strong evidence that the EPU has a negative impact on the
bearish markets of all countries. The slope equality test is applied to examine whether the estimated parameters in selected quantiles
τ are different from those in the median and corresponding (1 − τ) quantile. The null of equality is rejected, confirming that the
coefficients are statistically different. It denotes that the dependence structure is asymmetric, with dependence in the lower quantiles
and independence in the upper quantiles. What's more, we find that the situation of G7 markets is consistent with the total sample.
That is, the EPU indices have an adverse effect on the bearish markets of G7. In addition, the dependence structure is also similar with
the total sample.
In the BRIC markets, we know that the coefficient of EPU is negative and generally significant for all quantiles. Applying the Wald
test for the coefficients equality, the null of equality is rejected, showing that the structure of dependence is asymmetric, with
dependence in the lower and upper quantiles. Another fact is that the degree of dependence becomes weakly as the quantiles
increasing. That is, with the increase in stock market returns, the impact of economic policies becomes increasingly weak. This
situation also occurs in the G7 and total markets. In general, the EPU has a negative impact on the bearish markets of the total, G7
and BRIC, and plays a negative role in the bullish market of BRIC.

4.2. Quantile autoregression results

Panel quantile regression analysis reflects the overall situation of G7 and BRIC stock markets. However, the conditions of each

3
P. Guo et al. Finance Research Letters xxx (xxxx) xxx–xxx

Fig. 1. Rolling correlation coefficients between the USA and other stock market returns.

Table 3
Descriptive statistics for rolling correlation coefficients.

Relationships Mean S.D. Min Max Relationships Mean S.D. Min Max

USA and Canada 0.769 0.084 0.478 0.881 France and Canada 0.697 0.081 0.371 0.815
USA and China 0.204 0.208 −0.232 0.624 France and China 0.163 0.201 −0.148 0.587
USA and France 0.704 0.181 0.319 0.928 France and Germany 0.871 0.095 0.591 0.972
USA and Germany 0.754 0.146 0.353 0.921 France and India 0.622 0.147 0.230 0.833
USA and India 0.637 0.128 0.328 0.836 France and Italy 0.881 0.066 0.728 0.971
USA and Italy 0.718 0.160 0.290 0.929 France and Russia 0.508 0.175 0.062 0.737
USA and Japan 0.504 0.173 0.077 0.853 France and UK 0.851 0.073 0.663 0.938
USA and Russia 0.534 0.193 0.113 0.801 China and Canada 0.291 0.179 −0.054 0.620
USA and UK 0.824 0.091 0.562 0.919 China and India 0.312 0.236 −0.275 0.626
Japan and Canada 0.540 0.188 −0.181 0.864 China and Italy 0.223 0.185 −0.097 0.666
Japan and China 0.213 0.176 −0.180 0.574 China and Russia 0.306 0.158 −0.128 0.590
Japan and France 0.496 0.188 0.100 0.856 China and UK 0.152 0.242 −0.238 0.612
Japan and Germany 0.533 0.167 0.175 0.865 Canada and India 0.586 0.146 0.204 0.811
Japan and India 0.546 0.159 0.138 0.788 Canada and Italy 0.645 0.083 0.380 0.785
Japan and Italy 0.531 0.160 0.121 0.863 Canada and Russia 0.603 0.155 0.166 0.827
Japan and Russia 0.500 0.133 0.086 0.758 Canada and UK 0.713 0.074 0.479 0.847
Japan and UK 0.546 0.172 0.176 0.858 UK and India 0.642 0.122 0.293 0.802
Germany and Canada 0.669 0.085 0.285 0.787 UK and Italy 0.782 0.109 0.517 0.908
Germany and China 0.198 0.197 −0.197 0.626 UK and Russia 0.546 0.149 0.093 0.789
Germany and India 0.656 0.118 0.329 0.828 Russia and India 0.581 0.118 0.228 0.784
Germany and Italy 0.820 0.088 0.540 0.935 Russia and Italy 0.480 0.167 0.069 0.734
Germany and Russia 0.503 0.202 −0.078 0.771 Italy and India 0.632 0.128 0.313 0.849
Germany and UK 0.801 0.079 0.532 0.901

4
P. Guo et al. Finance Research Letters xxx (xxxx) xxx–xxx

Fig. 2. Plot of panel quantile regression results (EPU). Notes: The red solid line with rhombus represents the quantile estimates, with the green dash line depicting its
95% point-wise confidence interval. The blue solid line with triangles is the p-value. (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.)

country cause its unique stock market. Thus, we will discuss in detail on the stock market in each country. Fig. 3 presents these
detailed results.
According to the information in Fig. 3, the effect of EPU is positive for upper quantile and negative at lower quantile in the USA
and Italy markets. These coefficients estimated are significant. That is, there is the tail dependence in the bullish and bearish markets.
Using the slope equality test at low and high quantiles, the null of equality is rejected, confirming that the coefficients for the lower
and upper quantiles are statistically different. Thus, EPU and the USA/Italy market display an asymmetric and tail dependence
structure in the bearish and bullish markets, which means that economic policy has a negative/positive impact on the bearish/bullish
market of the USA and Italy when it is extreme fluctuations.
The effect of EPU on India and China stock market returns is negative and significant but only for the lower and intermediate
quantiles, whereas no significant effect is observed in the upper quantiles, which denotes that the dependence structure is asym-
metric. In addition, we reject the null of Wald test for the lower and intermediate quantiles. Therefore, EPU and these two stock
markets present an asymmetric dependence structure, with dependence in the lower tail and independence in the upper tail. It
presents that EPU indices have an adverse effect on the bearish market of India and China. Moreover, in terms of the degree of
dependence, India and China present the highest and lowest negative correlations in these two countries respectively.
As for Germany and Japan, the effect of EPU is negative and significant for the lower quantiles. Applying the Wald test for the
coefficients equality at lower quantiles, the null of equality is rejected. Hence, the structure of dependence is asymmetric, with
dependence in the lower tail and independence in the upper tail. This evidence suggests that the economic policy uncertainty plays a
negative role in the bearish market of France, Germany and Japan. The results for Canada and Russia show that there is the in-
termediate dependence and tail independence, meaning that the extreme EPU movements have no impact on the stock returns for
these two countries. Finally, the estimation results for France and the UK markets present that the effect of EPU is negligible for
almost quantiles. That is, there is no apparent dependence between economic policy uncertainty and France/the UK stock market.
For the above phenomena, a possible explanation comes from behavioral finance. According to the behavioral finance literature
(Barberis et al., 1998; Lewellen, 2002), the empirical phenomena are attributed to under and overreaction of stock returns to good or
bad news that is caused by economic policy uncertainty. As the consequences of under- and overreaction for the stock price are
different, this may lead to differing dependence structures across the return distribution (Baur et al., 2012). Another possible ex-
planation is economic integration that leads to close economic ties, while finance risks easily transmit from one country to another.
Next, we analyze the robustness of the above results. The robustness test is based on great economic events in each stock market.
Fig. 4 presents the robust check results. Compared the Figs. 3 and 4, we find that dependence degree of the EPU has changed, but

5
P. Guo et al. Finance Research Letters xxx (xxxx) xxx–xxx

Fig. 3. Plot of quantile autoregression results. Notes: The red solid line with triangles represents the quantile estimates, with the green dash line depicting its 95%
point-wise confidence interval. The blue solid line with rings is the p-values and horizontal purple short dash line is 0.1 in p-value. (For interpretation of the references
to colour in this figure legend, the reader is referred to the web version of this article.)

dependent structure remains essentially unchanged. Changes in dependence degree indicate that great economic events indeed affect
stock market returns; the relatively unchanged dependence structures show that the stock markets have considerable stability to the
external or internal shocks. In fact, we used indicator variable I(|REt − 1, i| > REq) to represent extreme events and have removed the
impact of some great events in the regression model (2). It allows the estimation results to keep relative robustness.

6
P. Guo et al. Finance Research Letters xxx (xxxx) xxx–xxx

Fig. 4. Robust check results.

5. Conclusion

This study offers new and abundant evidence on the relationship between economic policy uncertainty and stock market returns
for G7 and BRIC countries. We use a quantile regression approach to measure the structure and degree of dependence between EPU
and stock market returns. Some interesting results are found for ten countries: there is asymmetric and tail dependence between EPU
and the USA/Italy stock market; EPU indices have an adverse effect on the bearish market of Germany, Japan, India and China; there
is intermediate dependence in Canada and Russia stock markets; no apparent linkage between EPU and the UK/ France stock market
is found.

7
P. Guo et al. Finance Research Letters xxx (xxxx) xxx–xxx

Several economic policy implications can be drawn from the empirical results of this study. The dependence in the lower or upper
tails of return distributions may be related to market participants’ risk appetite behavior. Independence in the lower tail means that
no matter how to adjust economic policies in the bearish market, market participants do not respond to the policy changes. Only
when the stock market is bullish (dependence in the upper tail), they are sensitive to the risk of policy uncertainty shocks. Thus, risk
aversion tends to occur in the USA and Italy. However, risk pursuit tends to occur in the stock markets of Germany, Japan, India and
China. In addition, an understanding of the asymmetry and extreme effect of EPU to stock market returns can help the market
participants to make important portfolio allocation decisions (Mensi et al., 2014). Recognition of dependence is crucial for policy-
makers because it would help them discern the directions of the co-movements and reduce the volatility of the G7 and BRIC stock
market resulting from contagion of economic policy uncertainty.

Acknowledgment

This research is supported by the National Natural Science Foundation of China under grants No. 71171075, No. 71221001 and
No. 71431008.

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.frl.2017.11.001.

References

Baker, S., Bloom, N., Davis, S., 2013. Measuring economic policy uncertainty. Discussion Paper 13-02, Chicago Booth Research Paper.
Barberis, N., Shleifer, A., Vishny, R., 1998. A model of investor sentiment. J. Financ. Econ. 49, 307–343.
Baur, D.G., 2013. The structure and degree of dependence: a quantile regression approach. J. Bank. Finance 37, 786–798.
Baur, D.G., Dimpfl, T., Jung, R.C, 2012. Stock return autocorrelations revisited: a quantile regression approach. J. Empir. Financ. 19, 254–265.
Bloom, N., 2014. Fluctuations in uncertainty. J. Econ. Perspect. 28, 153–176.
Chevapatrakul, T., 2015. Monetary environments and stock returns: international evidence based on the quantile regression technique. Inter. Rev. Financ. Anal. 38,
83–108.
Colombo, V., 2013. Economic policy uncertainty in the US: does it matter for the euro area? Econ. Lett. 121, 39–42.
Jones, P.M., Olson, E., 2013. The time-varying correlation between uncertainty, output, and inflation: evidence from a DCC-GARCH model. Econ. Lett. 118, 33–37.
Koenker, R., Bassett, G., 1978. Regression quantiles. Econometrica 46, 33–50.
Koenker, R., 2004. Quantile regression for longitudinal data. J. Multivariate Anal. 91, 74–89.
Koenker, R., Xiao, Z., 2006. Quantile autoregression. J. Am. Stat. Assoc. 101, 980–990.
Lewellen, J., 2002. Momentum and autocorrelation in stock returns. Rev. Financ. Stud. 15, 533–564.
Li, X.M., Zhang, B., Gao, R., 2015. Economic policy uncertainty shocks and stock-bond correlations: evidence from the US market. Econ. Lett. 132, 91–96.
Mensi, W., Hammoudeh, S., Reboredo, J.C., Nguyen, D.K., 2014. Do global factors impact BRICS stock markets? A quantile regression approach. Emerg. Mark. Rev. 19,
1–17.