Herd Behavior in Frontier Markets: Evidence from Nigeria and Morocco
F. Economou Centre of Planning and Economic Research, Athens, Greece
Abstract
Herd behavior has attracted research interest during the last decades due to its important implications for stock market efficiency and portfolio diversification. However, herding has not been widely studied in the context of frontier markets. In this chapter we examine herd behavior in two African frontier markets, namely Nigeria and Morocco, employing the cross-sectional dispersion approach of Chang et al. (Chang, E.C., Cheng, J.W., Khorana, A., 2000. An examination of herd behavior in equity markets: an international perspective. J. Bank. Financ. 24, 1651–1679), also testing for asymmetries in herd behavior under different market states and the impact of the US stock market. Even though there is no evidence of herding employing the benchmark model for the whole sample period, there is evidence of herding during down market volatility days for Nigeria. Moreover, there is evidence of herding in Morocco during the global financial crisis. Testing for structural breaks reveals significant herding in Morocco for the subperiod from Dec. 2005 to Dec. 2014 with asymmetric effects. Finally, there is herding toward the US market in Nigeria.
Keywords
herding
cross-sectional dispersion of returns
frontier markets
JEL classification
G10
G14
G15
1. Introduction
Financial crises and stock market crashes have clearly demonstrated the impact of investors’ sentiment on asset pricing and stock markets’ efficiency. Herd behavior, which is behavioral similarity based on individuals’ interaction that leads to convergence of action and correlated trading (Hirshleifer and Teoh, 2003), is one of the most important behavioral biases that is more likely to occur during periods of market stress when individual investors prefer to follow the market consensus, being reluctant to follow their own knowledge or beliefs (Christie and Huang, 1995). Herding has been widely studied in financial markets (including the stock market, bond market, foreign exchange market, exchange-traded funds market, etc.), and it is evident in both retail and institutional investors’ behavior.a
Herding can be rational when it relates to payoff externalities, informational learning, principal–agent, and reputation-based problems, or it can derive from behavioral factors (Devenow and Welch, 1996). In any case, this behavior has important implications for market efficiency and portfolio diversification. However, there is also spurious herding (ie, correlated decision making based on the same set of fundamental information rather than imitation), which does not cause market inefficiency (Bikhchandani and Sharma, 2000).
Even though empirical evidence of herding in developed and emerging markets is mixed based on the period and the market under examination, herding is expected to be more pronounced in emerging markets since their special characteristics (thin trading, incomplete regulatory framework and corporate information disclosure, low transparency, information asymmetries, etc.) may facilitate herding behavior (Kallinterakis and Kratunova, 2007).
In the same spirit, frontier markets are also expected to display herding behavior. Frontier markets are less developed and less liquid markets that are too small to be considered as emerging (Balcilar et al., 2015; De Groot et al., 2012) and are characterized by low trading volume, high concentration, difficult access, inexperienced market participants, incomplete institutional framework, and limited information disclosure (Economou et al., 2015b; Speidell and Krohne, 2007). Quisenberry (2010) employs the definition of Merrill Lynch, according to which a frontier market is characterized as an “emerging emerging market” (ie, a market that is expected to become emerging). There is growing interest in these markets since their low correlations both among them and with developed markets offer market diversification benefits for international portfolios (Berger et al., 2011; Cheng et al., 2009; Jayasuriya and Shambora, 2009; Speidell and Krohne, 2007). As a result, their stock market behavior should be further analyzed in order to enhance the understanding of frontier markets.
While there is a growing strand of literature about herd behavior in developed and emerging stock markets, the existing literature dealing with frontier markets is limited. Balcilar et al. (2014) examine the cash- and oil-rich Gulf Cooperation Council (GCC) markets, indicating strong and persistent evidence of herding in Dubai, Kuwait, Qatar, and Saudi Arabia, while there is less frequent herding in Abu Dhabi. The authors also establish a direct link between herding and market volatility, and document the impact of shocks due to global factors on herding. A previous study of the GCC frontier markets indicates the presence of three market regimes regarding volatility as well as crossmarket herding effects driven by common factors in the GCC, especially during periods of extreme volatility (Balcilar et al., 2013). The retail investors–dominated Saudi Arabia stock market has also been studied by Rahman et al. (2015). The authors find evidence of herding irrespective of market conditions, which is more pronounced during periods of positive market returns and higher trading activity. Apart from retail investors, herding is also evident in institutional investors in frontier markets. Economou et al. (2015b) indicate that fund managers in Bulgaria and Montenegro herd significantly and intentionally, with herding being stronger during periods of positive market returns and high volume as well as during low-volatility periods for Montenegro. Moreover, Bulgarian and Montenegrin fund managers herded significantly both before and after the outbreak of the global financial crisis.
In this chapter we extend the limited herding literature in frontier markets by examining herd behavior in two African frontier markets, namely Nigeria and Morocco from 2004 to 2014 employing the cross-sectional dispersion of returns approach of Chang et al. (2000). Moreover, we test for possible asymmetries in herding estimations under different market states (up/down market returns, market volatility, and volume) as well as for the impact of additional explanatory variables, such as the oil price return, the US stock market return, and the US sentiment captured by the Chicago Board Options Exchange (CBOE) VIX index. The impact of the global financial crisis on the two markets under examination also provides interesting insight into the international stock markets dynamics.
The rest of the chapter is structured as follows: Section 2 presents the methodology and the data set employed in order to examine herding in frontier markets, Section 3 reports the empirical results and finally Section 4 offers conclusions.
2. Methodology and Data
2.1. Methodology
Herding has been examined in many different contexts in the financial markets employing different methodological approaches based either on portfolio holdings or on stock market returns. In this chapter we employ the cross-sectional dispersion approach, which is based on the seminal work of Christie and Huang (1995) and Chang et al. (2000). The authors propose an intuitive measure to capture herding and argue that in the presence of herding the cross-sectional dispersion of individual asset returns tends to decrease. The cross-sectional dispersion of returns is calculated as follows:
CSADt=∑Nk=1|Ri,t−Rm,t∣∣N
(4.1)
where Ri,t is the return of stock i on day t, Rm,t is the stock market return on day t, and N is the number of all listed stocks in the market on day t. The stock market return is defined as the equally weighted average return of all the individual stocks on day t.
Rational asset pricing models predict an increasing linear relationship between the cross-sectional absolute deviation (CSAD) and stock market return since the individual stocks differ in their sensitivity to the market return. On the other hand, Christie and Huang (1995) suggest that investors are more likely to act in a correlated manner during periods of extreme market returns and market stress resulting in reduced cross-sectional dispersion of returns.
The nonlinear model of Chang et al. (2000) (CCK model) is used in order to capture the relationship between the CSAD and the market return as follows:
CSADm,t=a+γ1∣∣Rm,t∣∣+γ2R2m,t+ut
(4.2)
where all variables are already defined. Rational pricing models predict a linear relationship with a positive and statistically significant coefficient γ1. In the presence of herding, the relationship is nonlinear and coefficient γ2 is expected to be negative and statistically significant; that is, the CSAD increases but at a decreasing rate.
We further examine whether herding displays an asymmetric behavior under different market states as it is usually reported by previous literature (Economou et al., 2011; Chiang and Zheng, 2010; Chiang et al., 2010). To this end we employ a single equation with a dummy variable that enables us to reestimate the model for up/down market days, high/low market volatility days, and high/low trading volume days as follows:
where Dup is a dummy variable that takes the value 1 on days with positive market returns/high market volatility/high volume and the value 0 otherwise. In order to identify high market volatility and high volume days, we compare market volatility/volume on day t with the 30-day moving average, and if it is higher Dup takes the value 1 and the value 0 otherwise.
Empirical results on asymmetric herd behavior are mixed, depending on the period and market under examination. However, herding is expected to be more pronounced during periods of market stress that are usually characterized by negative market returns (Chang et al., 2000; Chiang and Zheng, 2010; Mobarek et al., 2014), increased market volatility, and increased trading volume (Tan et al., 2008; Economou et al., 2011). As a result, we examine whether coefficient γ2 is greater than coefficient γ4 in Eq. 4.3 and whether the difference is statistically significant.
Moreover, since the sample period is quite long, ranging from 2004 to 2014, the global financial crisis that started from the US market may considerably affect herding estimations. Samarakoon (2011) provides empirical evidence of interdependence and contagion in frontier markets to US shocks which is more pronounced during crises, also indicating that the global financial crisis was more contagious for frontier markets (especially for markets with nonoverlapping trading activity with the United States, which respond with 1-day lag) than for emerging markets. In the same spirit, Chen et al. (2014) documented that leading stock markets (US and regional leading markets) Granger-cause frontier equity markets, a relationship that is largely influenced by the global financial crisis. The results are less significant after the crisis, and each individual frontier market differs in its relationship with the leading market. These findings indicate reduced diversification benefits during the global financial crisis period even when investing in frontier markets. In this case we would also expect herding to be more pronounced under extreme US market conditions. In order to test this hypothesis, we reestimate the benchmark model as follows:
where Dcrisis is a dummy variable that takes the value 1 during the global financial crisis of 2007–09 and the value 0 otherwise.
Finally, following Balcilar et al. (2014), we augment the benchmark model with several global market factors that may have an impact on the stock markets under examination such as the oil price return and the US stock market return captured by the S&P 500 index. Additionally, we test for the impact of the US sentiment captured by the CBOE VIX index in the same spirit with Economou et al. (2015a) and Philippas et al. (2013). In order to test these hypotheses, the benchmark model is restated as follows:
CSADm,t=a+γ1∣∣Rm,t∣∣+γ2R2m,t+γ3R2i,t+ut
(4.5)
where Ri,t is the return of the crude oil, S&P 500, and CBOE VIX indices, respectively. Eq. 4.5 enables us to test whether herding is more pronounced in the stock markets under examination under extreme changes in the crude oil prices, the US stock market, and the US sentiment (ie, whether coefficient γ3 is negative and statistically significant).
2.2. Data
Several African countries with developing financial markets have attracted research interest, being also likely to attract investors’ interest as part of a second generation of emerging markets (Nellor, 2008). Fig. 4.1 depicts the considerable increase of the MSCI Frontier Markets Africa Index from May 2002 to the first quarter of 2008 and from 2011 to 2014. The index captures mid-cap and large-cap representations of five African frontier markets (Nigeria, Morocco, Kenya, Mauritius, and Tunisia). Nigeria and Morocco represented 74.17% of the index country weights (52.73% Nigeria and 21.44% Morocco) in Dec. 2014.
In this chapter we constructed a survivor-bias-free data set including both active and dead stocks at any time during the period 2004–14. We employ percentage log-differenced returns calculated as follows for all listed equities in Nigeria and Morocco:
Ri,t=100×[log(Pi,t)−log(Pi,t−1)]
(4.6)
where Pi,t is the daily closing price of every stock i on day t, derived from the Thomson–Reuters Datastream database.
Trying to reduce the impact of thin trading on our estimations, we employ only stocks that displayed trading activityb on day t, and their number ranged between 8 and 161 for Nigeria and between 12 and 63 for Morocco. Table 4.1 reports the descriptive statistics for the markets under examination as well as for the additional explanatory variables employed in Eq. 4.5. The sample consists of 2655 daily observations for Nigeria and 2748 daily observations for Morocco.
Table 4.1
Descriptive Statistics
Nigeria
Morocco
Other factors
CSAD
Market
CSAD
Market
S&P 500
CBOE VIX
Crude oil Brent
Mean
1.0212
−0.0009
0.6823
0.0093
0.0099
−0.0149
0.0123
Median
0.9676
−0.0114
0.6556
0.0188
0.0319
−0.2341
0.0286
Maximum
11.1977
5.3372
4.7429
2.4099
4.7587
21.5414
5.8649
Minimum
0.0990
−6.5567
0.1515
−1.9719
−4.1126
−15.2259
−4.8326
Std. dev.
0.5606
0.4795
0.2302
0.3561
0.5386
2.9455
0.8618
Observations
2655
2655
2748
2748
2655
2655
2655
3. Empirical Results
Table 4.2 reports the empirical results of the benchmark CCK model estimation (Eq. 4.1). Coefficient γ1 is positive and statistically significant for both Nigeria and Morocco, whereas coefficient γ2 is positive and statistically significant only for Nigeria. As a result, there is no evidence of herding employing the benchmark model for the entire sample period. The estimated relationship for Morocco is linear and increasing, consistent with the predictions of the rational pricing models. On the other hand, the relationship for Nigeria is positive and nonlinear, providing evidence of antiherding.
Table 4.2
Estimates of the Standard CCK (2000) Model
Constant
|Rm,t|
R2m,t
R2 adj.
Nigeria
0.8134 (38.70)***
0.5242 (6.39)***
0.2126 (7.38)***
71.47%
Morocco
0.5669 (34.53)***
0.4085 (3.14)***
0.0747 (0.46)
27.08%
Notes: This table reports the estimated coefficients of the CCK (2000) model.
CSADm,t=a+γ1|Rm,t|+γ2R2m,t+ut, where CSADm,t is the cross-sectional absolute deviation of the individual stock returns and Rm,t is the return for each market m The sample consists of daily data from 2004 to 2014. t-statistics are given in parentheses using Newey–West (1987) heteroscedasticity- and autocorrelation-consistent standard errors. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively.
However, herd behavior may exist under different market states. Table 4.3 reports the results of the CCK model during up and down market returns. Overall, there is no evidence of asymmetric herding behavior during up and down market days. Coefficients γ2 and γ4 are both positive and statistically significant for Nigeria and statistically insignificant for Morocco.
Table 4.3
Estimates of the CCK (2000) Model During Up and Down Periods of the Market
Constant
Dup|Rm,t|
DupR2m,t
(1 − Dup)|Rm,t|
(1 − Dup) R2m,t
R2 adj.
Nigeria
0.8200 (41.67)***
0.4910 (6.32)***
0.2572 (9.80)***
0.4961 (6.11)***
0.1916 (7.49)***
72.11%
Morocco
0.5665 (36.30)***
0.3240 (2.70)***
0.1404 (0.92)
0.5100 (3.65)***
−0.0022 (−0.01)
27.67%
Wald tests for equality of herding coefficients
Nigeria
Morocco
γ1 − γ3
t-statistic, H0: γ1 = γ3
−0.0051
(−0.07)
−0.1859
(−2.06)**
γ2 − γ4
t-statistic, H0: γ2 = γ4
0.0656
(1.87)*
0.1425
(0.98)
Notes: This table reports the estimated coefficients of the CCK model during up and down periods of the market: CSADm,t=a+γ1Dup|Rm,t|+γ2DupR2m,t+γ3(1−Dup)∣∣Rm,t∣∣+γ4(1−Dup)R2m,t+ut, where CSADm,t is the cross-sectional absolute deviation of the individual stock returns, Rm,t is the return for each market m, and Dup is a dummy variable that takes the value 1 on days with positive market returns and the value 0 otherwise. The sample consists of daily data from 2004 to 2014. t-statistics are given in parentheses using Newey–West (1987) heteroscedasticity- and autocorrelation-consistent standard errors. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively. t-statistics and the Wald tests for the null hypothesis γ1 = γ3 and γ2 = γ4 in the estimated model are reported in the lower panel.
Testing for possible asymmetries during up and down market volatility days indicates the existence of herding during down market volatility days for Nigeria (Table 4.4), while there is no evidence of asymmetric herding during up and down volume days (Table 4.5). Coefficients γ2 and γ4 remain statistically insignificant for Morocco in all the model specifications for possible herding asymmetries.
Table 4.4
Estimates of the CCK (2000) Model During Up and Down Periods of Market Volatility
Constant
Dup vol|Rm,t|
Dup volR2m,t
(1 − Dup vol) |Rm,t|
(1 − Dup vol) R2m,t
R2 adj.
Nigeria
0.7557 (32.57)***
0.5968 (6.69)***
0.2017 (6.92)***
1.2239 (8.72)***
−0.8252 (−4.85)***
72.14%
Morocco
0.5685 (37.42)***
0.3911 (2.96)***
0.0858 (0.52)
0.3705 (2.83)***
0.2661 (1.06)
27.22%
Wald tests for equality of herding coefficients
Nigeria
Morocco
γ1 − γ3
t-statistic, H0: γ1 = γ3
−0.6271
(−7.79)***
0.0205
(0.25)
γ2 − γ4
t-statistic, H0: γ2 = γ4
1.0269
(6.54)***
−0.1803
(−0.92)
Notes: This table reports the estimated coefficients of the CCK model during up and down periods of the market: CSADm,t=a+γ1Dupvol∣∣Rm,t∣∣+γ2DupvolR2m,t+γ3(1−Dupvol)∣∣Rm,t∣∣+γ4(1−Dupvol)R2m,t+ut, where CSADm,t is the cross-sectional absolute deviation of the individual stock returns, Rm,t is the return for each market m, and Dup vol is a dummy variable that takes the value 1 on days with high market volatility and the value 0 otherwise. The sample consists of daily data from 2004 to 2014. t-statistics are given in parentheses using Newey–West (1987) heteroscedasticity- and autocorrelation-consistent standard errors. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively. t-statistics and the Wald tests for the null hypothesis γ1 = γ3 and γ2 = γ4 in the estimated model are reported in the lower panel.
Table 4.5
Estimates of the CCK (2000) Model During Up and Down Periods of Market Volume
Constant
Dup vo|Rm,t|
Dup voR2m,t
(1 − Dup vo)|Rm,t|
(1 − Dup vo) R2m,t
R2 adj.
Nigeria
0.8145 (38.56)***
0.4679 (6.31)***
0.2426 (8.30)***
0.5542 (5.74)***
0.2018 (6.34)***
71.66%
Morocco
0.5633 (46.75)***
0.4017 (2.97)***
0.1011 (0.54)
0.4746 (7.12)***
−0.0287 (−0.35)
27.37%
Wald tests for equality of herding coefficients
Nigeria
Morocco
γ1 − γ3
t-statistic, H0: γ1 = γ3
−0.0863
(−1.15)
−0.0729
(−0.68)
γ2 − γ4
t-statistic, H0: γ2 = γ4
0.0407
(1.03)
0.1298
(0.74)
Notes: This table reports the estimated coefficients of the CCK model during up and down periods of the market: CSADm,t=a+γ1Dupvo|Rm,t|+γ2DupvoR2m,t+γ3(1−Dupvo)∣∣Rm,t∣∣+γ4(1−Dupvo)R2m,t+ut, where CSADm,t is the cross-sectional absolute deviation of the individual stock returns, Rm,t is the return for each market m, and Dup vo is a dummy variable that takes the value 1 on days with high market volume and the value 0 otherwise. The sample consists of daily data from 2004 to 2014. t-statistics are given in parentheses using Newey–West (1987) heteroscedasticity- and autocorrelation-consistent standard errors. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively. t-statistics and the Wald tests for the null hypothesis γ1 = γ3 and γ2 = γ4 in the estimated model are reported in the lower panel.
Moreover, we examine the impact of the global financial crisis of 2007–09 on herding estimations (Table 4.6). The results for Nigeria do not indicate herding either before, during, or after the crisis. However, there is evidence of herding in Morocco during the global financial crisis, with coefficient γ2 being negative and statistically significant at the 10% significance level. This is the only evidence of herding for Morocco for the whole sample period.
Table 4.6
Estimates of the CCK (2000) Model During the Global Financial Crisis
Constant
Dcrisis|Rm,t|
DcrisisR2m,t
(1 − Dcrisis)|Rm,t|
(1 − Dcrisis) R2m,t
R2 adj.
Nigeria
0.8190 (40.69)***
0.5602 (6.43)***
0.1805 (7.30)***
0.4636 (5.65)***
0.2602 (10.06)***
72.05%
Morocco
0.5693 (42.78)***
0.4446 (6.42)***
−0.0854 (−1.76)*
0.3505 (2.81)***
0.2138 (1.19)
29.38%
Wald tests for equality of herding coefficients
Nigeria
Morocco
γ1 − γ3
t-statistic, H0: γ1 = γ3
0.0966
(1.11)
0.0941
(0.92)
γ2 − γ4
t-statistic, H0: γ2 = γ4
−0.0796
(−2.39)**
−0.2992
(−1.83)*
Notes: This table reports the estimated coefficients of the CCK model during the global financial crisis and before/after the crisis period: CSADm,t=a+γ1Dcrisis∣∣Rm,t∣∣+γ2DcrisisR2m,t+γ3(1−Dcrisis)∣∣Rm,t∣∣+γ4(1−Dcrisis)R2m,t+ut, where CSADm,t is the cross-sectional absolute deviation of the individual stock returns, Rm,t is the return for each market m, and Dcrisis is a dummy variable that takes the value 1 during the global financial crisis of 2007–09 and the value 0 otherwise. The sample consists of daily data from 2004 to 2014. t-statistics are given in parentheses using Newey–West (1987) heteroscedasticity- and autocorrelation-consistent standard errors. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively. t-statistics and the Wald tests for the null hypothesis γ1 = γ3 and γ2 = γ4 in the estimated model are reported in the lower panel.
Table 4.7 reports the results of augmenting the benchmark model with several explanatory variables that may relate to pronounced herding (ie, crude oil returns, the US stock market returns captured by the S&P 500, and the US “fear index,” the CBOE VIX index). Overall, it seems that there is herding toward the US market only in Nigeria, while we do not report any statistically significant result for the rest of the examined variables. This finding is in line with previous literature regarding Nigeria indicating that US investor sentiment is more important than local factors (Todorov and Bidarkota, 2013). There is no evidence of herding for Morocco, and the addition of several explanatory variables does not affect our initial conclusions.
Table 4.7
Estimates of the Standard CCK (2000) Model Augmented with Additional Variables
Panel A
Constant
|Rm,t|
R2m,t
R2oil,t
R2 adj.
Nigeria
0.8897
(49.48)***
0.2906
(5.20)***
0.2048
(13.28)***
−0.0006
(−0.12)
64.94%
Morocco
0.5667
(32.93)***
0.4126
(3.18)***
0.0736
(0.45)
−0.0013
(−0.43)
27.21%
Panel B
Constant
|Rm,t|
R2m,t
R2S&P500,t
R2 adj.
Nigeria
0.8188
(39.98)***
0.5398
(6.54)***
0.2099
(7.29)***
−0.0340
(−5.36)***
72.13%
Morocco
0.5659
(33.17)***
0.4058
(3.08)***
0.0762
(0.46)
0.0038
(0.68)
27.01%
Panel C
Constant
|Rm,t|
R2m,t
R2VIX,t
R2 adj.
Nigeria
0.8162
(39.31)***
0.5290
(6.36)***
0.2117
(7.33)***
−0.0005
(−1.34)
72.13%
Morocco
0.5642
(33.98)***
0.4057
(3.08)***
0.0775
(0.47)
0.0003
(1.53)
27.06%
Notes: This table reports the estimated coefficients of the CCK (2000) model: CSADm,t=a+γ1∣∣Rm,t∣∣+γ2R2m,t+γ3R2i,t+ut, where CSADm,t is the cross-sectional absolute deviation of the individual stock returns, Rm,t is the return for each market m, and Ri,t is the return of the crude oil, S&P 500, and VIX indices in panels A, B, and C, respectively. The sample consists of daily data from 2004 to 2014. t-statistics are given in parentheses using Newey–West (1987) heteroscedasticity- and autocorrelation-consistent standard errors. ***, **, and * represent statistical significance at the 1, 5, and 10% levels, respectively.
Finally, since the employed sample covers a long period, apart from the impact of the global financial crisis, we have also tested for structural breaks that would affect herding estimations. We have endogenously identified the structural breaks for the two markets under examination using the Quandt–Andrews break point test, and the results indicate Oct. 28, 2008, for Nigeria and Dec. 1, 2005, for Morocco as structural breaks. Even though the results remain qualitatively the same before and after the structural break for Nigeria,c there is evidence of herding in Morocco during the second subperiod from Dec. 1, 2005, to Dec. 31, 2014. The results of the estimated coefficients and t-statistics of the benchmark model for the period from Dec. 2005 to Dec. 2014 are the following:
Coefficient γ2 is negative and statistically significant, indicating the presence of herding behavior for the subperiod under examination. When we reestimate all model specifications for Morocco for possible asymmetric herding behavior, herding is evident on days with both up and down market returns as reported in Eq. 4.8, and on days with high market volatility Eq. 4.9) and trading volume (Eq. 4.10). Adjusted R2 is close to 24% for all model specifications.d
Even though investor psychology and its impact on asset allocation and pricing have been examined in developed markets, frontier markets have not been widely analyzed. The documented international portfolio diversification benefits that are related to investments in frontier markets necessitate an in-depth examination of herding behavior in these markets since correlated trading patterns that exist in the presence of herding may reduce diversification benefits.
In this chapter we examine herding in two African frontier markets that have not been analyzed earlier, using daily data for the period 2004–14. The empirical results employing the benchmark Chang et al. (2000) model do not indicate evidence of herding. However, when testing for asymmetries in herding estimations, we identified herding during down market volatility days for Nigeria. Moreover, there is evidence of herding during the global financial crisis only in Morocco. Finally, testing for the impact of additional variables on herding estimations indicates that there is herding toward the US market in Nigeria (ie, CSAD in Nigeria is reduced under extreme market returns in the US market). Finally, testing for structural breaks reveals significant evidence of herding in Morocco for the subperiod from Dec. 2005 to Dec. 2014, with herding being more pronounced during days of high market volatility and volume.
The empirical results offer useful insight for both retail and institutional investors who consider asset allocation in these stock markets to be particularly beneficial due to the international portfolio diversification benefits they bear. Correlated return patterns that exist in the presence of herding along with higher transaction costs charged in frontier markets (Marshall et al., 2015) may significantly reduce diversification benefits and investment performance.
Future research should further examine additional explanatory variables that may induce herding and that are not related to the market returns. For example, the impact of domestic market sentiment and crossmarket herding with neighboring (or other) stock markets should also be tested in order to better understand domestic market investors’ behavior as well as the magnitude of international diversification benefits.