Search in book...
Toggle Font Controls
Create new playlist

Name your new playlist

Playlist description (optional)
Sign In

Email address

Password

Forgot Password?

or

Continue with Facebook

Continue with Google
Sign Up

Full Name

Email address

Confirm Email Address

Password

or

Continue with Facebook

Continue with Google

Chapter 10

Are There Herding Patterns in the European Frontier Markets?

N. Blasco^*

P. Corredor^**

S. Ferreruela^*
^* University of Zaragoza, Zaragoza, Spain
^** Public University of Navarre, Pamplona, Spain

Abstract

This chapter attempts to determine whether there was herding behavior in the nine frontier markets included in the MSCI EFM Central and East Europe and CIS Index during the period 2011–14. Although the characteristics of these markets could suggest a strong mimetic behavior, the results do not seem to confirm this. On the one hand, using different methodologies, we find clear evidence of herding (at an individual market level) only within the Estonian market, although some other markets also show evidence of herd behavior in some extreme or fairly extreme quantiles of the return dispersion distribution. On the other hand, we find that the Bulgaria, Kazakhstan, Ukraine, and Croatia markets are more likely to herd around a global area consensus (the MSCI EFM Index) during both turbulent and calm periods. These results reveal the difficulty of characterizing herding behavior.

Keywords

herding

frontier markets

behavioral finance

Gephi

JEL classification

G15

1. Introduction

The efficient markets hypothesis assumes that investors have homogeneous expectations based on the information available and that they therefore try to maximize utility in a rational way. However, in reality it has been observed that investors may exhibit apparently irrational and predictable biases mainly attributable to psychological factors (Odean, 1998; Barber and Odean, 2000). The close link between rationality and emotion in decision making (Elster, 1998; Lo, 1999; Loewenstein, 2000; Peters and Slovic, 2000, among others) may explain this phenomenon. Rationality and emotions are not antithetical but are in fact complementary in decision making. Behavioral finance includes the emotional component within standard models of financial markets to explain the aggregate effects of decisions made by individual investors who may deviate from full neoclassical rationality (Thaler, 1991; Shefrin, 2000). In this context, the herd behavior of investors has been proposed as an alternative or complementary explanation of their decision-making process.

Herding behavior is defined as the apparent attempt by investors to copy the behavior of other investors (Bikhchandani and Sharma, 2000). Rational herding occurs when investors think that others are better informed, so they decide to imitate the decisions observed in other market participants, ignoring their own beliefs or information.

There are various explanations for this phenomenon. In terms of psychology, it has been suggested that the investor may prefer to conform to the market consensus (Devenow and Welch, 1996). Agency theory proposes that the reputation–compensation scheme rewards imitation, as compensation for an investor depends on how his or her performance compares to the performance of other investors, and whether deviations from consensus are costly (Scharfstein and Stein, 1990; Roll, 1992; Brennan, 1993; Rajan, 1994; Trueman, 1994). The type of market which is analyzed also seems to be decisive, as in the case of emerging markets where mimic behavior may be due to imperfect information (Chari and Kehoe, 2004; Calvo and Mendoza, 2000; Bikhchandani et al., 1998). Other explanations arise from differences in factors such as the relative importance of institutional versus individual investors (Lakonishok et al., 1992; Grinblatt et al., 1995; Wermers, 1999) or the level of sophistication of derivatives markets, aspects which could affect the decision-making process of investors. The work of Hirshleifer and Teoh (2003) provides a thorough review of the various explanations that have been offered for this phenomenon in the literature.

However, rational herding behavior is a relative concept that can be hard to verify. It is known that financial markets tend to work with moderate levels of intrinsic herding due to the unconscious impulses of investors to achieve positive returns and avoid negative ones. This makes the characterization and detection of rational herding behavior difficult.

Intuitively, intentional herding (which implies a follow-the-leader type of relationship) could be statistically described as a deviation of the returns of individual assets from market returns smaller than expected in the absence of herding behavior. This is due to the fact that the return of individual assets would not diverge substantially from the overall market return. Based on this idea, the works of Christie and Huang (1995) and Chang et al. (2000) (hereafter referred to as CH and CCK, respectively) are the leaders when it comes to analyzing market herd behavior. They assume that investors will probably ignore their beliefs in favor of market consensus in periods of large price changes, so herd behavior is more likely to appear during such periods.

While previous studies in the literature have examined herd behavior in various markets using this approach, this chapter tries to find further evidence of the presence of such behavior and its patterns in the nine European frontier markets. Frontier markets refer to stock markets in small nations that are at an earlier stage of economic and political development than larger and more mature emerging markets, according to various parameters such as their growth potential, market accessibility, liquidity, and foreign investment restrictions that are usually analyzed by market information providers.

Our study goes some steps further than previous studies. First, we analyze the presence of herding in each individual market using the methodologies developed by CH and CCK in order to provide results comparable with those previously reported in the literature (Chen, 2013). This approach has been widely applied (Demirer and Kutan, 2006; Tan et al., 2008; Chiang et al., 2010; Lao and Singh, 2011; Prosad et al., 2012). It tests for herding by examining whether the cross-sectional return dispersion decreases or increases at a decreasing rate as the market return increases. Authors have traditionally used the ordinary least squares (OLS) regression; however, we have chosen to further include quantile regression (QR) when applying the CCK measure (Koenker and Bassett, 1978), as this seems more suitable for our analysis. Herding is intuitively related to the lower quantiles of the return dispersion distribution, when the returns of most stocks in a market are similar to that of the market itself. However, we cannot ignore the high quantiles of the dispersion distribution, as even with high dispersion herding might be found if the expected positive relationship between return and return dispersion is broken. If we use only OLS, we could overlook herding if it exists only in certain quantiles, as this approach focuses mainly on mean values.

Second, we analyze a period after the onset of the 2008–09 global financial crisis and the ensuing European sovereign debt crisis in which the number of listed stocks in these markets significantly increased. Since the publication of previous results in Chen (2013), Bulgaria and Romania have tripled the number of listed shares; Slovenia and Croatia have doubled; and Estonia, Serbia, and Ukraine have grown by 50%.

Finally, in addition to the analysis of herd behavior in each market, we have used graph analysis tools such as Gephi to try to visualize under which circumstances or at which moments these stock markets or some of them herd around a more global consensus, given their similarity in terms of certain macroeconomic conditions, the economic environment, and market factors.

The chapter is organized as follows. The following section describes the database and the context of the markets under study. The subsequent section is devoted to describing the methodology used. The penultimate section presents the results obtained. The final section presents the main conclusions that can be drawn from the study.

2. Database

Our study analyzes nine frontier markets included in the MSCI EFM Central and East Europe and CIS Index (Bulgaria, Croatia, Estonia, Kazakhstan, Lithuania, Romania, Serbia, Slovenia, and Ukraine) during the period 2011–14. The database, obtained from Thomson Reuters Datastream, comprises the daily prices of the shares which have been listed on these nine markets at least sometime in the sample period.

These nine countries may be classified into two groups according to their membership in the European Union (EU). Slovenia, Estonia, and Lithuania joined the EU in 2004, Bulgaria and Romania in 2007, and Croatia in 2013. The three remaining countries are not yet members of the EU. Such a difference may be a key element for describing the institutional characteristics of these markets.

According to data provided by the World Bank in 2011, the gross domestic product (GDP) of Kazakhstan, Romania, and Ukraine is between 3 and 4 times bigger than that of the other countries analyzed. However, if we look at the ratio of stock market capitalization to GDP, the ranking is as follows: Croatia (38.73), Kazakhstan (28.49), Serbia (22.12), Ukraine (20.09), Slovenia (16.36), Romania (14.99) Bulgaria (14.94), Lithuania (11.86), and Estonia (9.09). The stock market volatility also differs among countries: Ukraine (36%), Romania (26%), and Kazakhstan (24%) have high levels of volatility, whereas Slovenia (13%) and Croatia (15%) generate the lowest levels.

The markets under analysis have institutional and cultural differences compared to other more developed markets. For this reason it is of interest to compare our European frontier markets (EFMs) with other emerging stock markets belonging to their area of influence (Poland, Czech Republic, Hungary, and Russia), as well as with large-cap markets in Europe such as Germany, France, and Great Britain.

The Hofstede Index (Hofstede, 2001) comprises six cultural dimensions that allow different cultures to be compared. The results of a comparison of the mean values of the Hofstede dimensions calculated for the EFMs with the mean values for other more developed markets and for other emerging markets within their area of influence lend support to our analysis of EFMs.

The mean values of power distance and uncertainty avoidance in the Hofstede Index are higher for EFMs (67 and 80, respectively) than the values of developed countries (46 and 62) and very similar to those of other emerging countries (66 and 86). According to these dimensions of the Hofstede Index, European frontier countries are more tolerant of the established power hierarchy and less tolerant of unorthodox ideas and different codes of belief and behavior.

Scores in relation to individualism (38), masculinity (33), and indulgence (24) in these frontier countries are lower than those of developed markets (76, 58, and 53) and also with regard to emerging countries used as a reference (59, 61, and 28). Such low values suggest that in European frontier countries individuals tend to act predominantly as members of a lifelong and cohesive group, family, or organization, within a feminine culture emphasizing modesty and caring and having the perception that their actions are constrained by social rules.

The valuations of these countries for the long-term orientation dimension are close to those of the benchmark countries. Although some differences are found in the individual scores, all the European frontier countries are long-term-oriented societies, and hence the effort to change society and the capacity for adaptation are seen as good qualities for the future.

Institutional factors for EFMs are also different from those of other countries analyzed in the comparative study. We use references provided by the Worldwide Governance Indicators that summarize the quality of governance. These indicators are voice and accountability, political stability, government effectiveness, regulatory quality, rule of law, and corruption control. In general, it can be said that the nine frontier countries obtain much lower scores than those of emerging countries and very considerably lower than the values shown by the most developed economies in Europe. There are also clear differences among frontier countries. The highest scores belong to Estonia, Slovenia, and Lithuania, and the lowest to Kazakhstan, Ukraine, and Serbia, suggesting the lower institutional development of these markets. As mentioned previously, these results may be closely related to membership or otherwise in the EU.

An additional measure is the Chinn–Ito index (Kaopen) (Chinn and Ito, 2006), which measures a country’s degree of capital account openness. The last available index values indicate that while Kazakhstan and Ukraine have negative scores, Estonia, Bulgaria, and Romania show values indicating a degree of openness similar to that of most developed countries.

All the cultural and institutional indicators for the EFMs presented here show a lower degree of development for their stock markets, given their lower degrees of information transparency, economic and social openness, and governance indicators that may influence market trading and liquidity. The herding effect within this framework is not easy to predict since there are elements that encourage gregarious behavior, either individually or toward a global consensus (more collectivism, lack of transparency related to lower institutional development and lower quality of governance, cultural indicators such as uncertainty avoidance, etc.), and others which favor the opposite, particularly against a general consensus (six out of the nine are members of the EU, and EFMs have different capitalization volumes and GDPs and/or different degrees of capital account openness, among others). In these circumstances, the new and updated empirical evidence will be of great interest in order to gain a deeper knowledge of these markets.

3. Methodology

3.1. CH and CCK Methods

The idea behind the methodologies proposed by CH and CCK is that, in the presence of intentional herding behavior, individuals are likely to leave aside their own beliefs about market behavior in favor of market consensus. Empirically, this would imply that stock returns would be more closely grouped around the overall market return. CH and CCK suggest that this evidence should be particularly intense during periods of large price changes given that, at those moments, investors discriminate individual assets less and treat all stocks in a similar way.

CH’s methodological proposal consists of calculating the cross-sectional standard deviation (CSSD) of stock returns relative to the market return for each period t and observing its behavior in times of extreme market movements. The verification of the existence of herding is performed through the following regression model:

$C S S D_{t} = α + β_{D} D_{t}^{L} + β_{U} D_{t}^{U} + ɛ_{t}$ $C S S D_{t} = α + β_{D} D_{t}^{L} + β_{U} D_{t}^{U} + ɛ_{t}$

(10.1)

where

D_{t}^{L} (D_{t}^{U}) = 1

$D_{t}^{L} (D_{t}^{U}) = 1$

if the aggregate market return on day t is located in the lower (upper) tail of the distribution, and zero otherwise. The β_D (β_U) coefficients, associated with the dummy variables, allow the existence of herd behavior among market participants to be identified.

CCK observe that the model presented by CH is a very restrictive test that requires a high level of nonlinearity to detect the presence of herding, and they therefore propose an alternative. Their proposal uses the cross-sectional absolute deviation (CSAD) of stock returns over the market return as a measure of the dispersion to detect the existence of herding. If market participants imitate each other, there would be a nonlinear relationship between the absolute deviation of returns and market returns. To capture the relationship, they include an additional parameter in the regression. CCK’s model takes the following form:

$C S A D_{t} = α + γ_{1} | R_{m, t} | + γ_{2} {[R_{m, t}]}^{2} + ɛ_{t}$ $C S A D_{t} = α + γ_{1} | R_{m, t} | + γ_{2} {[R_{m, t}]}^{2} + ɛ_{t}$

(10.2)

Eq. 10.2 introduces a nonlinear item [R_m,t]², as a linearly increasing relationship no longer holds in the presence of herding; more precisely, the dispersion would be lower if herding occurs. Therefore, herding is said to occur if the coefficient γ₂ of the nonlinear item is found to be negative and significant.

Note that the model specification in Eq. 10.2 restricts γ₂ to be the same for both up and down markets, while recent empirical research (Bekaert and Wu, 2000; Hong et al., 2007; Tan et al., 2008) has highlighted the asymmetric characteristics of asset returns. In our context, it is interesting to examine whether the imitative behavior shows an asymmetric response on bullish versus bearish days. Therefore, following Zhou and Anderson (2013), Eq. 10.2 is generalized as follows:

$C S A D_{t} = α + γ_{1} (1 - D) | R_{m, t} | + γ_{2} D | R_{m, t} | + γ_{3} (1 - D) {[R_{m, t}]}^{2} + γ_{4} D {[R_{m, t}]}^{2} + ɛ_{t}$ $C S A D_{t} = α + γ_{1} (1 - D) | R_{m, t} | + γ_{2} D | R_{m, t} | + γ_{3} (1 - D) {[R_{m, t}]}^{2} + γ_{4} D {[R_{m, t}]}^{2} + ɛ_{t}$

(10.3)

In Eq. 10.3, we consider asymmetry in both linear and nonlinear terms by setting D = 1 if R_m,t < 0 and D = 0 otherwise.

It is worth noting that CCK, and most researchers following them, estimate Eq. 10.2 using OLS. However, there are good reasons to opt for QR as proposed by Koenker and Bassett (1978) when attempting to detect herding in equity markets. Least squares estimators focus on the mean of the distribution of return dispersion, something which is not optimal for detecting stress-related behaviors, as the information contained in the tails of the distribution is lost. QR is a more versatile tool in analyzing extreme quantiles of the return deviation distribution, given that it provides a method to estimate the effects of market return on the dependent variable over its entire distribution. An additional benefit of using QR is that some statistical problems such as the effect of outliers or nonnormality of the errors can be alleviated (Barnes and Hughes, 2002).

Therefore Eqs. 10.2 and 10.3 would now take the following form:

$C S A D_{t} (τ | x) = α + η_{1} | R_{m, t} (τ) | + η_{2} {[R_{m, t} (τ)]}^{2} + ɛ_{t}$ $C S A D_{t} (τ | x) = α + η_{1} | R_{m, t} (τ) | + η_{2} {[R_{m, t} (τ)]}^{2} + ɛ_{t}$

(10.4)

$C S A D_{t} (τ | x) = α + η_{1} (1 - D) | R_{m, t} (τ) | + η_{2} D | R_{m, t} (τ) | + η_{3} (1 - D) {[R_{m, t} (τ)]}^{2} + η_{4} D {[R_{m, t} (τ)]}^{2} + ɛ_{t}$ $C S A D_{t} (τ | x) = α + η_{1} (1 - D) | R_{m, t} (τ) | + η_{2} D | R_{m, t} (τ) | + η_{3} (1 - D) {[R_{m, t} (τ)]}^{2} + η_{4} D {[R_{m, t} (τ)]}^{2} + ɛ_{t}$

(10.5)

where the τth quantile is that value of the target variable distribution below which the proportion of the population is τ.

3.2. Gephi Proposal

In addition to analyzing each individual market, we have used Gephi software (Bastian et al., 2009) in order to find similarities among different frontier markets in terms of their herding behavior around a global consensus represented by the MSCI EFM Central and East Europe and CIS Index. Gephi is open-source software for graph and network analysis. Its flexible and multitask architecture brings new possibilities for working with complex data sets and producing valuable visual results. The graph consists of a set of nodes and a set of pairs of nodes called edges. The usefulness of a network analysis derives from the data associated to specific nodes and edges, which can be ordered and clustered according to specified criteria.

In our case, we build a network of relationships between markets where each node corresponds to a frontier market and an edge between two nodes corresponds to the frequency with which two stocks listed in two different frontier markets follow the global frontier markets’ consensus. That is, a stock in frontier market m₁ is connected to a stock in market m₂ if both stock returns are very close to the global return. The probability of finding such connections between stocks belonging to m₁ and m₂ gives the weight for the edge between m₁ and m₂. Specifically, we carry out the following steps:

1. We calculate the return deviation with respect to the MSCI EFM Central and East Europe and CIS Index (R_MSCIEFM) for every stock, in every market, and every day.

$R D_{i m, t} = {(R_{i m, t} - R_{MSCIEFM, t})}^{2}$ $R D_{i m, t} = {(R_{i m, t} - R_{MSCIEFM, t})}^{2}$

(10.6)

R_im,t is the daily return on day t of stock i belonging to market m. m = Bulgaria, Croatia, … Ukraine.

2. We set the criteria for identifying the set of minimum deviations, MD. Specifically, we select those daily deviations observed below the critical value corresponding to the first decile of the deviations distribution. The average return deviation is AvRD = 0.00335506, and the cutoff value, leaving aside 10% of the smallest deviations, is DecileRD = 1.28755E-06.

3. Individual stocks included in MD_t are identified, with MD_t being the set of stocks following the global (MSCI EFM Index) market consensus on day t.

$M D_{t} = {i m / R D_{i m, t} < DecileRD}$ $M D_{t} = {i m / R D_{i m, t} < DecileRD}$

(10.7)

4. We select the set of days when the frontier market consensus (minimum deviations) is followed by more than 1000 individual stocks, the aim being to analyze the days with more intense herd behavior.

5. For every day with intense herd behavior (larger number of individual stocks following the global consensus), we compute the number of pairs of stocks (i, j) included in MD_t belonging to two different markets (for every pair of markets m₁ and m₂).

$E_{t} (m_{1} m_{2}) = # [(i m_{1}, j m_{2}) / i m_{1}, j m_{2} \in M D_{t} with m_{1} \neq m_{2}]$ $E_{t} (m_{1} m_{2}) = # [(i m_{1}, j m_{2}) / i m_{1}, j m_{2} \in M D_{t} with m_{1} \neq m_{2}]$

(10.8)

6. We also classify herding days depending on the global return R_MSCIEFM, so that we can build networks for extreme positive returns (daily R_MSCIEFM higher than 1%), extreme negative returns (daily R_MSCIEFM lower than −1%) and nonextreme returns (calm days). We aggregate E_t(m₁m₂) for extreme positive return days [denoted as E_EPR(m₁m₂)], extreme negative return days [E_ENR(m₁m₂)], and calm days [E_NER(m₁m₂)], respectively. Using this calculation, in order to avoid market size bias, we scale using the total number of possible combinations of two stocks belonging to each pair of different markets and taking into account the number of intense herding days in every state E (E = extreme positive return, extreme negative return, and nonextreme return). Therefore, we compute the edge between two markets depending on the state as:

$Edg e_{E (m_{1} m_{2})} = \frac{E_{E} (m_{1} m_{2})}{# [(i m_{1}, j m_{2}) / m_{1} \neq m_{2}] * # (herding days in state E)}$ $Edg e_{E (m_{1} m_{2})} = \frac{E_{E} (m_{1} m_{2})}{# [(i m_{1}, j m_{2}) / m_{1} \neq m_{2}] * # (herding days in state E)}$

(10.9)

Using the Gephi software, the graph analysis leads to the visualization of two properties: First, the size of a node gives a measure of the intensity of herd behavior for the stocks of a given market; second, strong edges highlight strong links between markets.

4. Empirical Results

4.1. Descriptive Statistics

Table 10.1 presents the summary statistics for CSSD, CSAD, and market returns for each market in the sample. As shown in the table, the mean of the aggregate return R_m remains negative for all countries. Both CSSD and CSAD exhibit significant skewness and kurtosis, and are hence not normally distributed.

Table 10.1

Summary Statistics of CSSD, CSAD, and R_m

	Bulgaria	Croatia	Estonia	Kazakhstan	Lithuania	Romania	Serbia	Slovenia	Ukraine
CSSD
Mean	0.0023	0.0025	0.0056	0.0033	0.0075	0.0048	0.0016	0.0067	0.0017
Median	0.0017	0.0022	0.0048	0.0017	0.0047	0.0036	0.0013	0.0043	0.0009
Maximum	0.0172	0.0114	0.0303	0.1168	0.0854	0.0510	0.0377	0.0835	0.0287
Minimum	0.0005	0.0006	0.0014	0.0000	0.0005	0.0009	0.0002	0.0007	0.0001
Std. dev.	0.0019	0.0011	0.0032	0.0069	0.0091	0.0042	0.0017	0.0084	0.0027
Skewness	3.1123	2.3775	2.3370	9.0544	3.9530	4.2934	15.3184	5.0181	5.3687
Kurtosis	13.9817	9.1296	8.9698	112.2516	20.4928	28.8945	300.1422	33.5178	37.8097
Jarque-Bera	6,626.0667	2,527.7456	2,423.9732	502,817.4035	15,431.0011	31,475.9674	3,777,493.3800	43,088.4159	54,849.7749
CSAD
Mean	0.0091	0.0119	0.0140	0.0075	0.0189	0.0185	0.0050	0.0176	0.0058
Median	0.0076	0.0112	0.0125	0.0043	0.0139	0.0163	0.0045	0.0133	0.0037
Maximum	0.0711	0.0476	0.0538	0.2244	0.1654	0.0987	0.0756	0.1662	0.0954
Minimum	0.0021	0.0028	0.0034	0.0000	0.0012	0.0043	0.0005	0.0018	0.0005
Std. dev.	0.0059	0.0041	0.0068	0.0137	0.0181	0.0105	0.0038	0.0171	0.0072
Skewness	3.4436	1.6104	1.5376	8.4192	3.4652	2.8407	10.2390	4.3340	5.9589
Kurtosis	25.6231	10.2072	6.3061	101.7209	19.6010	16.0824	173.0361	29.1456	55.5625
Jarque-Bera	23,045.2989	2,593.9468	851.1448	407,860.0256	13,416.8844	8,526.9941	1,231,928.9924	31,360.6599	119,220.1335
Observations	989	999	1,002	976	995	1,006	1,008	992	985
R_m
Mean	−0.0002	−0.0004	−0.0003	−0.0004	−0.0002	−0.0004	−0.0001	−0.0008	−0.0003
Median	−0.0001	−0.0002	0.0001	−0.0002	0.0002	−0.0003	−0.0002	−0.0004	−0.0001
Maximum	0.0372	0.0130	0.0486	0.0851	0.0636	0.0398	0.0335	0.0885	0.0253
Minimum	−0.0221	−0.0335	−0.0663	−0.1160	−0.0855	−0.0448	−0.0080	−0.0831	−0.0496
Std. dev.	0.0039	0.0037	0.0101	0.0078	0.0133	0.0073	0.0023	0.0110	0.0043
Skewness	1.0501	−0.7918	−0.6104	−2.5140	−0.8223	−0.7367	5.6585	−0.4390	−2.8686
Kurtosis	19.6241	9.7598	9.5020	80.4481	11.6307	9.2950	82.0737	19.2671	40.7614
Jarque-Bera	11,570.0986	2,006.4260	1,827.2330	244,955.1000	3,200.2800	1,752.0090	267,990.7000	10,969.3900	59,873.1392
Observations	989	999	1,002	976	995	1,006	1,008	992	985

This table reports the summary statistics of the cross-sectional standard deviations (CSSD), cross-sectional absolute deviations (CSAD), and cross-sectional equally weighted average returns (R_m) for the markets belonging to the MSCI EFM Central and East Europe and CIS Index. The data are obtained from Thomson Reuters Datastream. They range from the start of 2011 to the start of 2015.

4.2. Evidence on Herding: CH and CCK Methods^a

Table 10.2 presents the estimation results of herding based on Eq. 10.1, which is in the spirit of CH’s specification. For clarity reasons only, the results for the extreme 1% are shown on the table. However, extreme 2, 5, and 10% upper and lower tails have also been set to represent market stress, and the results are available from the authors upon request. The results indicate that herding, as identified by significant and negative β_D and β_U coefficients, is detected only in extreme up markets in Croatia, Romania, and Ukraine, regardless of the extreme percentiles of the returns distribution which have been considered as market stress, although these negative β_U coefficients indicate a stronger herding behavior in the most extreme percentiles. These results are largely in line with Chen (2013), as he does not find any sign of herding in the eight EFMs considered in his sample (Croatia is not included), but differ in reference to Romania and Ukraine.

Table 10.2

Analysis of Herding Behavior in Up and Down Extreme Markets (1–99%) (CH)

Extreme return percentiles 1 and 99%	α	β_D	β_U	R² adj.
Bulgaria	0.0022***	0.0076***	0.0077***	0.3150
Bulgaria	(43.09)	(15.15)	(15.34)	0.3150
Croatia	0.0039***	0.0024***	−0.0014***	0.0578
Croatia	(11.42)	(6.89)	(−4.11)	0.0578
Estonia	0.0056***	0.0057***	0.0044***	0.0515
Estonia	(55.54)	(5.98)	(4.65)	0.0515
Kazakhstan	0.0026***	0.0422***	0.0315***	0.5840
Kazakhstan	(18.05)	(29.92)	(22.36)	0.5840
Lithuania	0.0075***	0.0000	0.0016	−0.0017
Lithuania	(25.58)	(0.01)	(0.57)	−0.0017
Romania	0.0187***	0.0157***	−0.0142***	0.2659
Romania	(17.17)	(14.30)	(−12.98)	0.2659
Serbia	0.0014***	0.0024***	0.0089***	0.3202
Serbia	(32.67)	(5.76)	(21.20)	0.3202
Slovenia	0.0058***	0.0451***	0.0332***	0.4759
Slovenia	(29.82)	(24.44)	(17.97)	0.4759
Ukraine	0.0139***	0.0151***	−0.0124***	0.5169
Ukraine	(23.37)	(25.26)	(−20.84)	0.5169

This table reports the estimation results of herding in the markets belonging to the MSCI EFM Central and East Europe and CIS Index according to Eq. 10.1,

$C S S D_{t} = α + β_{D} D_{t}^{L} + β_{U} D_{t}^{U} + ɛ_{t}$ $C S S D_{t} = α + β_{D} D_{t}^{L} + β_{U} D_{t}^{U} + ɛ_{t}$

where CSSD_t is the equally weighted cross-sectional standard deviation of returns and R_m,t is the equally weighted market portfolio return at time t. A significant negative value of β_D and/or β_U suggests the existence of herding. Numbers in parentheses are t-statistics. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively.

The results of estimating the herding regression represented by Eq. 10.2, which follows the specification given by CCK, are shown in Table 10.3. As suggested in the literature, a negative value on the coefficient γ₂ is consistent with herding. The coefficient on the market return square is negative and statistically significant only for the Estonian market, suggesting that herding behavior exists only in that market. This holds for the asymmetric herding behavior under market ups and downs as represented in Eq. 10.3, the results of which are shown in Table 10.4. Only the Estonian market exhibits negative γ₃ and γ₄ coefficients for both up and down markets. Chen (2013) also detects herding in the Estonian market when using this approach. However, unlike us, he also observes negative γ₂ coefficients in Kazakhstan, Lithuania, and Romania. As stated earlier, Estonia is the country with the smallest ratio of stock market capitalization to GDP among all the countries in the sample. Moreover, it is the smallest market, with only 15 stocks considered during this period. Both characteristics may affect the behavior of investors in this market and its measurement.

Table 10.3

Analysis of Herding Behavior in European Frontier Markets (CCK)

	α	γ₁	γ₂	R² adj.
Bulgaria	0.0045***	1.8149***	0.1461	0.8312
Bulgaria	(37.59)	(39.79)	(0.07)	0.8312
Croatia	0.0085***	1.2732***	−0.9709	0.5742
Croatia	(59.32)	(23.64)	(−0.29)	0.5742
Estonia	0.0090***	0.8492***	−8.7632***	0.3749
Estonia	(31.66)	(17.80)	(−7.41)	0.3749
Kazakhstan	0.0012***	1.8945***	0.3967*	0.9818
Kazakhstan	(15.84)	(114.85)	(1.85)	0.9818
Lithuania	0.0074***	1.3081***	4.9070***	0.8256
Lithuania	(20.33)	(25.73)	(5.30)	0.8256
Romania	0.0111***	1.3858***	9.9957***	0.6987
Romania	(36.09)	(18.78)	(3.88)	0.6987
Serbia	0.0027***	1.6078***	14.4822***	0.8479
Serbia	(36.24)	(34.44)	(8.01)	0.8479
Slovenia	0.0067***	1.6165***	3.5631***	0.8798
Slovenia	(23.84)	(36.97)	(4.74)	0.8798
Ukraine	0.0017***	1.8576***	1.5202*	0.9514
Ukraine	(24.62)	(71.31)	(1.84)	0.9514

This table reports the estimation results of herding in the markets belonging to the MSCI EFM Central and East Europe and CIS Index according to Eq. 10.2,

$C S A D_{t} = α + γ_{1} | R_{m, t} | + γ_{2} {[R_{m, t}]}^{2} + ɛ_{t}$ $C S A D_{t} = α + γ_{1} | R_{m, t} | + γ_{2} {[R_{m, t}]}^{2} + ɛ_{t}$

where CSAD_t is the equally weighted cross-sectional absolute deviation of returns and R_m,t is the equally weighted market portfolio return at time t. A significant negative value of γ₂ suggests the existence of herding. Numbers in parentheses are t-statistics. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively.

Table 10.4

Analysis of Herding Behavior in Up and Down European Frontier Markets

	α	γ₁	γ₂	γ₃	γ₄	R² adj.	Wald γ₃ = γ₄
Bulgaria	0.0045***	1.8530***	1.7201***	−1.5204	7.7165	0.8316	0.0750
Bulgaria	(36.50)	(32.69)	(25.53)	(−0.66)	(1.50)	0.8316	0.0750
Croatia	0.0087***	1.0623***	1.1874***	40.1281***	0.5708	0.5788	0.0082
Croatia	(56.00)	(9.02)	(19.88)	(2.63)	(0.16)	0.5788	0.0082
Estonia	0.0089***	0.9400***	0.8455***	−13.0256***	−8.0106***	0.3768	0.0310
Estonia	(30.48)	(13.99)	(15.88)	(−5.78)	(−6.28)	0.3768	0.0310
Kazakhstan	0.0011***	1.9104***	1.9064***	−0.3836	0.4622*	0.9819	0.0610
Kazakhstan	(15.34)	(77.27)	(96.76)	(−0.92)	(1.93)	0.9819	0.0610
Lithuania	0.0078***	1.1406***	1.2746***	11.3823***	4.3265***	0.8292	0.0001
Lithuania	(21.05)	(15.61)	(21.84)	(6.39)	(4.28)	0.8292	0.0001
Romania	0.0112***	1.3230***	1.3927***	15.0826***	8.8067***	0.6988	0.1809
Romania	(36.02)	(13.94)	(16.58)	(3.44)	(3.04)	0.6988	0.1809
Serbia	0.0027***	1.5930***	1.4670***	14.8317***	48.2214**	0.8480	0.1309
Serbia	(33.13)	(26.13)	(14.19)	(6.86)	(2.13)	0.8480	0.1309
Slovenia	0.0067***	1.5754***	1.6356***	2.4277**	4.5284***	0.8820	0.0894
Slovenia	(24.18)	(28.72)	(32.31)	(2.41)	(4.86)	0.8820	0.0894
Ukraine	0.0017***	1.7999***	1.8057***	8.4986***	2.3577***	0.9519	0.0335
Ukraine	(24.56)	(38.35)	(56.40)	(2.92)	(2.59)	0.9519	0.0335

This table reports the estimation results of herding in the markets belonging to the MSCI EFM Central and East Europe and CIS Index according to Eq. 10.3

where CSAD_t is the equally weighted cross-sectional absolute deviation of returns, and R_m,t is the equally weighted market portfolio return at time t. D is a dummy variable which takes a value = 1 if R_m,t < 0, and 0 otherwise. A significant negative value of γ₃ suggests the existence of herding during bullish markets, whereas a significant negative value of γ₄ suggests the existence of herding during bearish markets. Numbers in parentheses are t-statistics. Also reported are the p-values for the hypothesis test γ₃ = γ₄. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively.

Table 10.5 presents the estimated results for the nine markets using the quantile regression method as represented in Eq. 10.4. We observe that the Estonian market shows a negative coefficient across all quantiles, which is in line with the results of the OLS estimation. Some other interesting results arise in the QR estimation. Croatia and Ukraine, as well as Bulgaria and Kazakhstan, where herding was not detected using previous methodologies, display negative coefficients for some of the quantiles above the 70% level. The phenomenon of herding behavior being more significant for upper quantiles shows that herding activity is more likely to occur under volatile market conditions—that is, when individual asset returns clearly deviate from the market consensus because the market return comprises both very high and very low stock returns. Under these circumstances, in relative terms, additional changes in market return tend to promote the market consensus. On the other hand, Lithuania shows a negative γ₂ coefficient in quantiles up to 10%, indicating a difference between this market and the others and revealing the traditional concept of herding underlying the usual methodologies.

Table 10.5

Quantile Regression Analysis of Herding Behavior in European Frontier Markets

	Bulgaria	Croatia	Estonia	Kazakhstan	Lithuania	Romania	Serbia	Slovenia	Ukraine
τ = 1%	4.4305***	−0.8992	−6.5998***	2.5567***	−8.8415***	−1.6493	8.1685***	5.8397***	4.1456***
τ = 1%	(3.79)	(−0.61)	(−4.36)	(11.43)	(−2.82)	(−0.25)	(9.51)	(13.87)	(21.56)
τ = 5%	4.7286***	0.0223	−11.2364***	1.9324***	−9.7771**	1.8252	8.3324***	4.9871***	4.5994***
τ = 5%	(3.19)	(0.01)	(−6.89)	(11.73)	(−2.56)	(0.26)	(6.89)	(9.71)	(15.62)
τ = 10%	2.7879**	0.7967	−8.7295***	1.5557***	−7.1221***	4.6722***	8.4334***	4.0497***	4.7169***
τ = 10%	(2.19)	(0.45)	(−4.83)	(7.81)	(−4.48)	(3.14)	(6.22)	(8.34)	(14.76)
τ = 30%	1.6174	0.2042	−7.0393**	1.0854***	6.6152***	6.7666	11.0633***	4.9403***	6.0421***
τ = 30%	(1.22)	(0.12)	(−2.37)	(4.61)	(8.34)	(0.23)	(6.15)	(4.91)	(8.84)
τ = 50%	−1.1407	−1.5703	−8.4163***	0.6230***	5.9164***	12.4063***	18.7040***	5.4901***	3.5212***
τ = 50%	(−0.80)	(−0.82)	(−8.60)	(2.81)	(6.91)	(5.98)	(9.36)	(6.33)	(3.91)
τ = 70%	−1.6238	−4.1170**	−7.3115***	0.6974	6.7458	16.5371***	18.0042***	3.7743***	1.7418
τ = 70%	(−1.01)	(−2.09)	(−7.49)	(0.05)	(1.01)	(5.05)	(11.03)	(4.33)	(0.63)
τ = 90%	−5.9347**	51.1121	−11.8469***	0.2263	6.5731	24.5026***	19.0583***	15.9079***	−3.6614**
τ = 90%	(−2.30)	(0.72)	(−8.62)	(0.15)	(1.14)	(2.73)	(9.81)	(5.51)	(−1.97)
τ = 95%	−5.9842*	27.2995	−12.8287***	−2.0698***	7.8789	31.6356	12.4509**	11.8136***	−8.4822***
τ = 95%	(−1.74)	(0.41)	(−5.28)	(−2.61)	(0.32)	(1.52)	(2.32)	(4.04)	(−5.92)
τ = 99%	−13.4720	8.7867	−18.9427***	−0.7746	21.3858	24.1117	4.5009	10.4323	−11.2948***
τ = 99%	(−1.55)	(0.13)	(−2.83)	(−1.31)	(1.22)	(0.84)	(0.66)	(0.11)	(−5.29)

This table reports the quantile regression estimates for the markets belonging to the MSCI EFM Central and East Europe and CIS Index by different quantile groups according to Eq. 10.4.

$C S A D_{t} (τ | x) = α + γ_{1} | R_{m, t} (τ) | + γ_{2} {[R_{m, t} (τ)]}^{2} + ɛ_{t}$ $C S A D_{t} (τ | x) = α + γ_{1} | R_{m, t} (τ) | + γ_{2} {[R_{m, t} (τ)]}^{2} + ɛ_{t}$

where CSAD_t is the equally weighted cross-sectional absolute deviation of returns, and R_m,t is the equally weighted market portfolio return at time t. γ_k,τ refers to the kth coefficient conditional on the τth quantile distribution in the estimated equation. For reasons of brevity only the results for γ₂ are shown. α and γ₁ are positive and significant for all countries and quantiles. A significant negative value of γ₂ suggests the existence of herding. Numbers in parentheses are t-statistics. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively.

Evidence given in Tables 10.5 and 10.6 clearly indicates that the estimated coefficients vary with the quantile levels. Looking at panels A and B in Table 10.6, the estimated statistics suggest that the results presented in Table 10.5 hold when asymmetry is taken into account, although some differences can be observed between up and down markets. Bulgaria, Estonia, Kazakhstan, and Ukraine are again the markets where negative γ₃ and γ₄ coefficients can be found. Bulgaria shows herding in up markets only in some of the higher quantiles, whereas Lithuania shows significant negative coefficients only when the market is down, and again in the lower quantiles, as shown in Table 10.5, where up and down markets are not differentiated.

Table 10.6

Quantile Regression Analysis of Herding Behavior in Up and Down European Frontier Markets

Panel A	Bulgaria	Croatia	Estonia	Kazakhstan	Lithuania	Romania	Serbia	Slovenia	Ukraine
τ = 1%	5.0289***	29.7009**	−4.5098**	2.2772***	−5.5546	10.4678**	8.3398***	7.6931***	6.3577
τ = 1%	(4.21)	(2.18)	(−2.55)	(4.10)	(−1.38)	(2.04)	(7.34)	(10.59)	(0.90)
τ = 5%	4.8091***	23.2915	−10.7206***	1.8951***	−3.9308	8.9455	8.3777***	6.1993***	8.9578***
τ = 5%	(3.42)	(1.30)	(−9.92)	(11.24)	(−0.30)	(1.26)	(5.52)	(6.87)	(6.16)
τ = 10%	3.0449	31.0229*	−10.8802***	1.4601***	3.2064	15.8435	7.7447***	5.0202***	9.5469***
τ = 10%	(1.63)	(1.73)	(−9.48)	(3.24)	(0.24)	(1.22)	(4.75)	(6.48)	(5.51)
τ = 30%	−0.2713	16.4512	−8.2138***	0.2831	9.4790***	20.7352***	11.4217***	3.2269***	11.0198
τ = 30%	(−0.14)	(0.73)	(−4.76)	(0.49)	(7.08)	(9.76)	(4.60)	(5.20)	(1.59)
τ = 50%	−2.6439	23.6987	−10.9427***	0.4468	13.0215***	17.6642***	18.8716***	3.6113***	14.7470
τ = 50%	(−1.61)	(1.12)	(−5.20)	(1.23)	(10.60)	(8.10)	(7.18)	(2.77)	(0.55)
τ = 70%	−1.3974	53.0741	−11.1471	−0.2436	11.7339***	15.7558***	16.0865***	1.1808	14.2287***
τ = 70%	(−0.78)	(0.99)	(−0.90)	(−0.48)	(8.90)	(6.45)	(6.52)	(0.68)	(5.31)
τ = 90%	−6.4821***	67.3296***	−12.8224	−2.5799***	8.8249**	15.7451***	19.3868***	14.1938*	5.0623
τ = 90%	(−3.27)	(2.82)	(−1.61)	(−3.07)	(2.07)	(2.13)	(8.21)	(1.70)	(1.26)
τ = 95%	−8.4606	44.9005	−12.7934*	−2.9430***	10.1631	4.1344	16.0603***	9.9779	−7.6548
τ = 95%	(−1.41)	(1.19)	(−1.68)	(−3.00)	(0.88)	(0.65)	(2.72)	(0.98)	(−0.23)
τ = 99%	−13.4720*	133.0665	−32.0886**	−2.3508	3.6815	7.7983	21.0088***	28.0555	−20.7195***
τ = 99%	(−1.74)	(0.55)	(−2.14)	(−0.90)	(0.24)	(1.01)	(3.52)	(0.13)	(−4.19)

Panel B	Bulgaria	Croatia	Estonia	Kazakhstan	Lithuania	Romania	Serbia	Slovenia	Ukraine
τ = 1%	17.1877***	3.1244	−4.2976**	1.7125	−7.9777**	−1.1460	5.5630	8.6148***	4.2120***
τ = 1%	(4.50)	(1.56)	(−2.49)	(0.08)	(−2.33)	(−0.17)	(0.77)	(12.17)	(18.89)
τ = 5%	12.8880***	1.2608	−9.2504***	2.0667***	−9.3587**	−4.1883	−1.5119	7.8070***	4.7099***
τ = 5%	(4.44)	(0.68)	(−5.43)	(17.09)	(−2.07)	(−0.48)	(−0.15)	(7.86)	(16.30)
τ = 10%	10.0774***	3.4258*	−5.2716***	1.7972***	−5.9378***	1.7458	−4.3329	6.2557***	4.8076***
τ = 10%	(3.79)	(1.87)	(−6.88)	(7.51)	(−3.17)	(0.22)	(−0.37)	(5.33)	(15.18)
τ = 30%	9.3353***	2.3731	−6.6391***	1.1953***	6.3622***	2.0440	73.4671***	6.0456***	4.9816***
τ = 30%	(3.34)	(1.19)	(−7.34)	(4.33)	(3.16)	(1.07)	(2.71)	(8.58)	(11.31)
τ = 50%	4.2079	0.3095	−8.2993***	0.5640**	5.3047***	12.8872***	54.3073**	5.2546***	5.9790***
τ = 50%	(1.22)	(0.14)	(−7.18)	(2.09)	(6.46)	(6.06)	(2.12)	(6.33)	(5.63)
τ = 70%	18.3370	−1.7481	−7.0088***	1.8918	3.7300***	13.2907***	42.0015*	6.2546	2.6486**
τ = 70%	(1.41)	(−0.75)	(−5.62)	(0.78)	(3.78)	(2.79)	(1.84)	(0.47)	(2.24)
τ = 90%	8.1174	1.8146	−9.4331***	1.6738	3.2395	29.9976***	183.8882	14.5899***	−2.5027
τ = 90%	(0.71)	(0.01)	(−9.13)	(0.80)	(0.50)	(2.07) **	(0.73)	(4.71)	(−1.25)
τ = 95%	24.3789	−1.1197	−11.6174***	−2.2485***	7.8789	41.9129	212.1480**	8.1965	−7.7789
τ = 95%	(1.08)	(−0.00)	(−4.93)	(−2.87)	(0.29)	(2.33)	(2.18)	(1.31)	(−1.35)
τ = 99%	−37.6980***	−11.9793	−14.0714**	−2.9497***	28.1360	0.1315	220.4379	−8.5124	−9.3564***
τ = 99%	(−2.47)	(−0.10)	(−2.48)	(−18.89)	(0.56)	(0.00)	(0.39)	(−1.15)	(−2.01)

This table reports the quantile regression estimates for the markets belonging to the MSCI EFM Central and East Europe and CIS Index by different quantile groups according to Eq. 10.5.

$C S A D_{t} (τ | x) = α + γ_{1} (1 - D) | R_{m, t} (τ) | + γ_{2} D | R_{m, t} (τ) | + γ_{3} (1 - D) {[R_{m, t} (τ)]}^{2} + γ_{4} D {[R_{m, t} (τ)]}^{2} + ɛ_{t}$ $C S A D_{t} (τ | x) = α + γ_{1} (1 - D) | R_{m, t} (τ) | + γ_{2} D | R_{m, t} (τ) | + γ_{3} (1 - D) {[R_{m, t} (τ)]}^{2} + γ_{4} D {[R_{m, t} (τ)]}^{2} + ɛ_{t}$

where CSAD_t is the equally weighted cross-sectional absolute deviation of returns and R_m,t is the equally weighted market portfolio return at time t. γ_k,τ refers to the kth coefficient conditional on the τth quantile distribution in the estimated equation. D is a dummy variable which takes a value = 1 if R_m,t < 0, and 0 otherwise. For reasons of brevity only the results for γ₃ (panel A) and γ₄ (panel B) are shown. α, γ₁, and γ₂ are positive and significant for all countries and quantiles. A significant negative value of γ₃ suggests the existence of herding during bullish markets, while a significant negative value of γ₄ suggests the existence of herding during bearish markets. Numbers in parentheses are t-statistics. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively.

4.3. Evidence of Herding: Gephi

Figs. 10.1–10.3 show Gephi visualizations of herding relationships among EFMs, both on extreme positive or negative return days and on calm days. The main results can be summarized as follows. Ukraine, Kazakhstan, and Bulgaria are the markets with the clearest tendency to herd toward the global MSCI EFM Index, followed by Croatia. Lithuania and Estonia are, in contrast, the markets with the least global herding behavior. Romania, Serbia, and Slovenia show an intermediate herding level toward the EFM index. It should be noted that Kazakhstan and Ukraine (the markets with the most intense herding relationships) are the countries outside the EU with higher indicators of economic activity in terms of GDP but a lower degree of economic openness and less institutional development, whereas Lithuania and Estonia (the markets with the least intense herding relationships) are members of the EU, small in terms of economic development, but more open and more developed institutionally. Nevertheless, Bulgaria, one of the markets with more intense global herding behavior, can not be so easily classified together with Kazakhstan and Ukraine except for the fact that these three markets also share individual herd behavior under volatile market conditions together with Croatia, which is the next (fourth) participant of the global consensus.

Figure 10.1 Herding relationships among European frontier markets on calm days (nonextreme R_MSCIEFM days).
BL, Bulgaria; CR, Croatia; EO, Estonia; KZ, Kazakhstan; LT, Lithuania; RM, Romania; SB, Serbia; SV, Slovenia; UA, Ukraine.

Figure 10.2 Herding relationships among European frontier markets on negative extreme global return days (negative extreme R_MSCIEFM days).
BL, Bulgaria; CR, Croatia; EO, Estonia; KZ, Kazakhstan; LT, Lithuania; RM, Romania; SB, Serbia; SV, Slovenia; UA, Ukraine.

Figure 10.3 Herding relationships among European frontier markets on positive extreme global return days (positive extreme R_MSCIEFM days).
BL, Bulgaria; CR, Croatia; EO, Estonia; KZ, Kazakhstan; LT, Lithuania; RM, Romania; SB, Serbia; SV, Slovenia; UA, Ukraine.

It is also interesting to note that calm herding days (nonextreme return days with more than 1000 individual stocks following the overall consensus) multiply by 10 the herding on extreme (positive or negative) return days. Thus Fig. 10.1 is, perhaps, the most representative of herd behavior toward the MSCI EFM Index, even though the relationships structure is largely reproduced in periods of extreme returns. Intense herding days represent 10% of the time period under analysis.

Estonia is the market that most clearly deviates from the global consensus and offers its own performance on both calm and extreme return days, suggesting that the market with the clearest evidence of individual herding is the one less prone to follow the area consensus described by the MSCI EFM Index.

In summary, although we have found some evidence of herding in EFMs both at an individual level and at a global level, our results indicate that special attention should be paid to the methodology applied in herding analysis, given that there are many types of imitative behavior and the models and procedures offered by the literature to date detect only a part of them.

5. Conclusions

This study has examined the herding behavior in nine EFMs at both individual and global levels using the models proposed by CH and CCK and a graphical approach which allows possible relationships among markets to be detected (using Gephi software). The results are not homogeneous among the methodologies: the CH model finds herding only in extreme up markets in Croatia, Romania, and Ukraine, whereas the results of applying the CCK measure are compatible with the existence of herding only in the Estonian market, for both the market as a whole and up and down markets. We further tested for herding by employing a quantile regression approach, and observe that estimated CCK coefficients vary with the quantile levels. Estonia still shows negative coefficients across all quantiles, whereas Bulgaria, Croatia, Kazakhstan, and Ukraine show significant herding only in the higher quantiles of the distribution. Lithuania, in contrast, displays negative coefficients in the lower quantiles.

With respect to the tendency to herd around a global consensus, the results using the Gephi tool indicate that the markets showing a more intense herding tendency are Bulgaria, Croatia, Kazakhstan, and Ukraine, the ones that show significant herding in the highest quantiles. Nevertheless, Estonia and Lithuania, two small markets that provide evidence of individual herding behavior along the entire return dispersion distribution or in the lowest quantiles, respectively, are the markets which do not follow the MSCI EFM Index consensus.

These nonhomogeneous results are difficult to explain solely on the basis of market characteristics or institutional and cultural factors. Besides the differences caused by analysis over different time horizons, as is the case of our chapter compared with Chen (2013), the results also depend on the methodology applied. The usual methodologies look for a specific type of herding and therefore may be useful for detecting such a herd behavior. It should be noted that there is no universally accepted methodology for detecting herding, however or whenever it is present in a market or a set of markets. This may be the reason why the results found in the empirical literature are not homogeneous or comparable.

Acknowledgment

This chapter has received financial support from the Spanish Ministry of Economy and Competitiveness (ECO2012-35946-C02-01 and ECO2013-45568-R) and the Government of Aragón/European Social Fund (S14/2).

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.

Table of Contents for Chapter 10: Are There Herding Patterns in the European Frontier Markets?

Create new playlist

Sign In

Sign Up