Regime-Switching Behaviour in the Conditional Volatility of MENA Stock Market Returns

Abstract

The objective of this article is to investigate the behaviour of the time-varying volatility in 11 Middle East and North African (MENA) countries’ stock market using a three-state Markov regime switching model over the period from 30 October 2006 to 21 October 2011. We find that MENA stock market volatility can be characterised by three regimes: tranquil period with low volatility of volatility, turmoil regime with high volatility of volatility and crisis regime with extremely high volatility of volatility. Besides, the Granger causation effects from the MSCI World index to MENA stock markets are stronger and statistically significant especially in crisis regime.

JEL Classification: F30, G01, G15

Keywords

Time-varying volatility MENA countries’ stock market three-state Markov regime switching model Granger causality test

Introduction

Over the last two decades, most of the Middle East and North Africa (MENA) countries have experienced a number of economic reforms, financial liberalisation and global integration process. These countries experienced a noticeable growth in market capitalisation, the number of listed companies, the value and the shares traded (Zaher 2007). These new characteristics of these markets may lead to changes in their volatility generating process.

Understanding the behaviour of volatility is important for pricing financial assets, implementing hedging strategies and for evaluating regulatory proposals to restrict international capital flows. Hammoudeh and Li (2008) examine the sudden changes in volatility for five Gulf area Arab stock markets (Bahrain, Kuwait, Oman, Saudi Arabia and UAE) over the period 1994–2001 and find that most of these stock markets are more sensitive to major global events than to local regional factors. Neaime (2006) studies the dynamic relationships in the volatilities of the stock market return within the MENA region¹ and the more developed financial markets of the US and the UK over the period 1995–2002. He shows that the group of countries having a stronger causal relationships in variance include the US, UK, Egypt, Jordan, Morocco and Turkey. Maghyereh and Al-Zoubi (2004) examine the dynamic interdependence among four emerging MENA stock markets, namely, Egypt, Jordan, Morocco and Turkey, over the period 1998–2003 and show that there are strong linkages among these markets at the volatility level. Yu and Hassan (2008) investigate the Granger causality test between seven MENA markets of Bahrain, Oman, Saudi Arabia, Jordan, Egypt, Morocco and Turkey and three developed countries (the US, the UK and France) over the period from 1 January 1999 through 31 December 2005. Empirical results from tests for unidirectional Granger causality between developed and MENA equity markets show with minor exception, a nonsignificant evidence of causality. Alkulaib, Najand and Mashayekh (2009) investigate the lead–lag relationship between the MENA countries and regions and found that there is more interaction and linkage in the Gulf Cooperation Council (GCC) region than in the North Africa and Levant regions.²

This article studies the regime-switching behaviour in the conditional volatility of MENA stock market returns using a Markov regime switching volatility model with three distinct states: tranquil period with low volatility of volatility, turmoil regime with high volatility of volatility and crisis regime with extremely high volatility of volatility. The conditional volatility has been modelled by using an autoregressive GARCH (1, 1) model. Moreover, using the tests for Granger-causality, we investigate whether the strength of spillovers from the MSCI World index to MENA stock markets change scientifically as the World market move from one regime to another³. The study is conducted using daily data for 11 MENA stock market returns (Bahrain, Egypt, Jordan, Kuwait, Lebanon, Morocco, Oman, Qatar, Saudi Arabia, Tunisia and UAE) over the period from 30 October 2006 to 21 October 2011.

Results show that MENA stock market volatility can be characterised by three regimes: tranquil period with low volatility of volatility, turmoil regime with high volatility of volatility and crisis regime with extremely high volatility of volatility. For example, for Lebanon, the volatility in the crisis period is 20 times higher than that in the turmoil regime which is 78 times higher than that in the calm period. Granger causality test results show significant evidence of causality in variance between the World market index and MENA markets in the turmoil regime for six MENA markets (Bahrain, Egypt, Morocco, Oman, Qatar and UAE) and for seven MENA markets (Jordan, Kuwait, Lebanon, Oman, Qatar, Tunisia and UAE) during financial crisis. This causality effect varies from one MENA country to another according to its degree of financial integration with the more mature financial markets. In the calm period, no significant causal relationship has been proved for all MENA markets. This result suggests that information from the World market is transmitted to the MENA stock markets, albeit mostly in turbulent periods.

The article is organised as follows: the first section introduces and motivates the importance of the study of the behaviour of MENA stock market volatility. The second section describes the data and the methodology. The third section presents and analyses the results. The last section concludes and gives important recommendations to policy makers.

Data and Methodology

Data

Our data set concerns daily stock market price index in local currency of 11 MENA countries, namely, Bahrain, Egypt, Jordan, Kuwait, Lebanon, Morocco, Oman, Qatar, Saudi Arabia, Tunisia and UAE, and the MSCI World index in US dollar. The sample period is from 30 October 2006 through 21 October 2011 (29 September 2010 for Saudi Arabia). Data are from the Morgan Stanley Capital International.⁴ Daily returns in each market are represented as the natural logarithmic differences in prices as follows:

R_{it} Ln (\frac{P_{it}}{P_{it 1}})

(1)

where P_it is the closing price for each country’s index at time t.

Descriptive statistics for each series’ daily returns include mean, standard deviation, maximum, minimum skewness, Kurtosis, Jarque–Bera test, Augmented Dickey–Fuller (ADF) test and Ljung–Box test statistics applied to the return and squared return series are reported in Table 1. Except, Morocco, Oman, Qatar and Tunisia, all other markets present a negative mean return. Tunisian market appears to have, on average, the highest return over the sample period (0.029 per cent). The UAE has the highest risk as approximated by standard deviation of 2 per cent. Skewness values are negative for all series excluding Jordan, Lebanon and Tunisia; which indicates that data are skewed left. Furthermore Kurtosis values are larger than 3 for all indices showing that these series have fat tails compared to the normal distribution. The Jarque–Bera test shows that the null hypothesis of normality is rejected for all markets. The ADF test with drift and trend was conducted to check for unit root in the return series. All indices returns are stationary and the null hypothesis of unit root is rejected. The Ljung–Box Q-statistic, up to the eighth order in level and squares of returns, clearly indicates that there is serial correlation in levels and squared returns for all indices, suggesting the existence of the volatility clustering phenomenon.

Table 1.

Descriptive Statistics for Return Series

	Bahrain	Egypt	Jordan	Kuwait	Lebanon	Morocco	Oman	Qatar	S. Arabia	Tunisia	UAE	W. Index
Mean	−0.00120	−0.00017	−0.00043	−0.00024	−0.00006	0.00013	0.00009	0.00019	−0.00010	0.00029	−0.00093	−0.00019
Std. Dev.	0.01291	0.01762	0.01155	0.01623	0.01455	0.01066	0.01391	0.01677	0.01797	0.01025	0.02002	0.01352
Maximum	0.07169	0.08295	0.05577	0.09200	0.10328	0.05011	0.09648	0.09180	0.09068	0.09468	0.10981	0.06740
Minimum	−0.11602	−0.10453	−0.05321	−0.10518	−0.10180	−0.05884	−0.09109	−0.11764	−0.10286	−0.06997	−0.13245	−0.07325
Skewness	−1.522	−0.820	0.040	−0.899	0.963	−0.237	−0.362	−0.731	−0.524	0.132	−0.895	−0.648
Kurtosis	16.989	8.296	7.149	11.220	13.802	6.520	14.909	12.954	10.108	13.458	10.389	8.304
Jarque– Bera	11101**(0.000)	1665**(0.000)	933**(0.000)	3835**(0.000)	6522**(0.000)	684**(0.000)	7711**(0.000)	5483**(0.000)	2201**(0.000)	5928**(0.000)	3131**(0.000)	1615**(0.000)
ADF test statistic	−32.43**(0.000)	−31.93**(0.000)	−32.51**(0.000)	−32.99**(0.000)	−26.92**(0.000)	−29.50**(0.000)	−32.04**(0.000)	−31.61**(0.000)	−29.20**(0.000)	−30.96**(0.000)	−30.86**(0.000)	−32.11**(0.000)
Q(8)	19.644**(0.012)	29.527**(0.000)	26.938**(0.001)	21.906**(0.005)	32.967**(0.000)	69.384**(0.000)	34.248**(0.000)	42.297**(0.000)	20.379**(0.009)	44.282**(0.000)	40.461**(0.000)	27.392**(0.001)
Q²(8)	144.88**(0.000)	117.36**(0.000)	232.47**(0.000)	236.88**(0.000)	331.72**(0.000)	178.20**(0.000)	436.90**(0.000)	196.88**(0.000)	152.42**(0.000)	95.676**(0.000)	213.39**(0.000)	844.00**(0.000)

Source: Authors’ calculations.

Notes: The sample period is 30 October 2006 through 21 October 2011 (29 September 2010 for Saudi Arabia). The ADF test for unit roots is based on the regression with constant and trend. Q(8) and Q²(8) are the eighth-lag Ljung–Box test statistics applied to the return and squared return series, respectively. p-values are in parentheses.

** and * indicate significance at the 5% and 10% levels, respectively.

Methodology

Regime-switching behaviour in the volatility generating processes

To investigate the regime-switching behaviour in the volatility generating process we need to determinate the conditional volatility. The most popular approach for modelling conditional volatility is the GARCH family models as introduced by Engle (1982) and generalized by Bollerslev (1986) and Nelson (1991). For capturing volatility of stock market returns, an AR (P) GARCH (1, 1) model is specified as follows:

r_{t} a_{0}_{s 1}^{p} a_{s} r_{t s}_{t};

(2)

\frac{_{t}}{_{t 1}} ~ N (0, h_{t});

(3)

h_{t}_{0}_{t 1}^{2} h_{t 1} .

(4)

where r_t is the daily stock market return at time (t), h_t is the conditional variance of the residuals from the mean equation and ε_t is the error term that follows a normal distribution with mean zero and time-varying variance.

Once the conditional volatility series have been determined, we use a Markov regime switching model. This model has been introduced by Hamilton (1989) and largely applied to different developed and emerging stock market returns (Abid and Bahloul 2011; Moore and Wang 2007; Wang and Theobald 2008). The selected model allows the variance to switch across different states, and the regime at any given date is supposed to be the outcome of a Markov chain whose realisations are unobservable. Three regimes of volatility have been defined: tranquil period with low volatility of volatility, turmoil regime with high volatility of volatility and crisis regime with extremely high volatility of volatility. Baba and Sakurai (2011) find that there are three distinct regimes in the VIX index during the period 1990–2010: tranquil regime, turmoil regime and crisis regime. Moore and Wang (2007) and Wang and Theobald (2008) show the existence of three volatility regimes (low, medium and high) for two new European Union states, namely, Poland and Slovakia, and for three East Asian emerging stock markets (Indonesia, Korea and Thailand), respectively.

The proposed model is given as follows:

h_{t}_{t}

_{t} ~ N (0,_{S_{t}}^{2})

(5)

where h_t represents daily conditional volatility, ε_t follows a normal distribution with zero mean and variance given by $_{S_{t}}^{2}$ . S_t is assumed to be a three-state first order Markov process with transition probability matrix represented as:

P [\begin{matrix} p_{11} & p_{21} & p_{31} \\ p_{12} & p_{22} & p_{32} \\ p_{13} & p_{23} & p_{33} \end{matrix}]

(6)

where $p_{ij} \Pr (S_{t} j |S_{t 1} i)$ and $_{j 1}^{3} p_{ij} 1$ , for j = 1,2,3, and for all i. These probabilities are specified as constant coefficients that are independent of time t.

From equation (6), the expected duration d of regime j is given by:

E (d) \frac{1}{1 p_{jj}}

(7)

Equation (5) and the transition probability matrix can be estimated by maximum likelihood using Hamilton’s filter and iterative algorithms (Hamilton 1994; Kim and Nelson 1999).

Granger causality test within regimes

In addition to the study of the regime-switching behaviour in the volatility of MENA stock market returns, we perform Granger causality tests (Granger, 1969) to investigate the effects of unidirectional causality between the World market index and MENA stock market volatilities across regimes. The sample has been divided into three sub-sample periods (calm, turmoil and crisis) using the smoothed states probabilities for the world market index. According to Hamilton (1989) a stock market is in regime i if the associated smoothed probability is higher than 0.5.

The pairwise Granger causality tests are represented empirically as follows:

{VMENA}_{t} c_{1}_{j 1}^{m} a_{j} {VMENA}_{t j}_{j 1}^{m} b_{j} {VW}_{t j}_{t}

(8)

where VW, VMENA and ε_t represent stock market volatility of the MSCI world index and 11 MENA countries, and vectors of the random error term, respectively. M is the order of the respective lag variable. The VW is said to Granger cause VMENA if lagged coefficients of VW are significantly different from zero. Since results of causality test are sensitive to the lag imposed, we use the Bayesian information criterion to select the optimal lag length.

Results and Analysis

The main results of this article are summarised in three propositions and are followed by discussions:

Proposition I: MENA stock market volatility is time-varying and best described by an autoregressive GARCH (1, 1) specification.

The preliminary analysis of MENA stock market returns was conducted on AR (P) specifications. For all our indices, we obtain that a first-order autoregressive process is sufficient to describe the expected fluctuation in mean return. The estimation results of the AR (1)-GARCH (1, 1) are reported in Table 2. The autoregressive coefficient in the conditional mean (a₁) is positive and significant for all MENA markets and the MSCI World index. This coefficient varies from 0.0602 for Jordan to 0.1679 for Morocco.

Table 2.

Estimation Results of the AR (1)-GARCH (1, 1) Model

	Bahrain	Egypt	Jordan	Kuwait	Lebanon	Morocco	Oman	Qatar	S. Arabia	Tunisia	UAE	W. Index
Panel A: Mean equation
a ₀	−0.0005*(0.087)	0.0003(0.531)	−0.0002(0.424)	0.0002(0.607)	−0.0004(0.154)	0.0002(0.457)	0.0005**(0.036)	0.0005(0.133)	0.0007(0.107)	0.0005**(0.043	−0.0001(0.761)	0.0004(0.144)
a ₁	0.0821**(0.030)	0.0818**(0.002)	0.0602**(0.042)	0.0661**(0.038)	0.0628**(0.014)	0.1679**(0.000)	0.1249**(0.000)	0.1423**(0.000)	0.0659*(0.054)	0.1014**(0.000)	0.1305**(0.000)	0.1169**(0.000)
Panel B: Conditional variance equation
ω ₀	0.000004**(0.000)	0.000004**(0.000)	0.000002**(0.000)	0.000004**(0.000)	0.000005**(0.000)	0.000025**(0.000)	0.000002**(0.000)	0.000005**(0.000)	0.000002**(0.000)	0.000004**(0.000)	0.000005**(0.000)	0.000002**(0.000)
α	0.1224**(0.000)	0.0352**(0.000)	0.0489**(0.000)	0.0648**(0.000)	0.1234**(0.0000)	0.1881**(0.000)	0.1340**(0.000)	0.0869**(0.000)	0.0619**(0.000)	0.1685**(0.000)	0.0578**(0.000)	0.0900**(0.000)
β	0.8591**(0.000)	0.9534**(0.000)	0.9332**(0.000)	0.9206**(0.000)	0.8597**(0.000)	0.5805**(0.000)	0.8650**(0.000)	0.8945**(0.000)	0.9334**(0.000)	0.8100**(0.000)	0.9284**(0.000)	0.9032**(0.000)
α + β	0.9815	0.9886	0.9821	0.9854	0.9831	0.7686	0.9990	0.9814	0.9953	0.9785	0.9862	0.9932
Panel C: Diagnostic tests
Log(L)	4063	3507	4077	3724	3890	4158	4091	3745	2800	4269	3446	4052
Q(8)	6.834(0.555)	6.896(0.548)	4.654(0.794)	9.110(0.333)	10.382(0.239)	8.843(0.356)	11.710(0.165)	10.912(0.207)	7.527(0.481)	6.152(0.630)	9.517(0.301)	5.931(0.655)
ARCH-LM (4)	1.406(0.229)	3.353(0.010)	1.932(0.103)	0.979(0.418)	0.533(0.712)	0.389(0.816)	0.552(0.697)	0.686(0.601)	4.251(0.002)	0.992(0.410)	1.612(0.168)	3.037(0.017)

Source: Authors’ calculations.

Notes: Q(8) is the eighth-lag Ljung–Box test statistic applied to the standardised residual. ARCH-LM (4) is the Lagrange Multiplier test of Engle (1982) at four lags for heteroscedasticity on standardized residual. p-values are in parentheses.

** and * indicate significance at the 5% and 10% levels, respectively.

The time-varying pattern of the market index price variability was confirmed for all series. In fact the coefficients of the GARCH effects (α and β) are significant at the 1 per cent level in all cases. The sum of α and β was close but less than one, implying persistent volatility effects. These values vary from 0.7686 for Morocco to 0.9990 for Oman. Lamoureux and Lastrapes (1990) stipulate that the high persistence may reflect regime switch in the variance process. Figure 1, which displays time series of the conditional variance, shows different states of volatility for all markets. This volatility is low, medium or high during several periods.

The Ljung–Box Q statistic tests show that autocorrelation of standardised residuals are statistically insignificant at 1 per cent level for all indices. The Lagrange Multiplier test of Engle (1982) at four lags for heteroscedasticity on standardised residual are insignificant at 1 per cent level for all series except Egypt, Saudi Arabia and the world market index showing that standardised residual does not exhibit additional ARCH effect.

Proposition II: MENA stock markets can be characterised by three regimes with different degrees in the variance of conditional volatility of stock market returns: tranquil period with low volatility of volatility, turmoil regime with high volatility of volatility and crisis regime with extremely high volatility of volatility.

Table 3 reports parameter estimates of the regime switching model in which stock market conditional volatility are assumed to be drawn from three distributions which differ in the variance of the stock market volatility.⁵ We apply the MATLAB package for Markov regime switching provided by Perlin (2011). Conditional volatility appears to be characterised by three regimes: tranquil period with low volatility of volatility, turmoil regime with high volatility of volatility and crisis regime with extremely high volatility of volatility. In fact, the volatility of volatility values are significant at 1 per cent level for all markets.

In the first regime, the variance of the volatility varies from 0.0001 per cent for Jordan and Morocco to 0.03 per cent for Egypt. In the second regime, the volatility is very important than that in the first regime. This importance varies from eleven times for the World market index to 78 times for Lebanon. In the third regime, the variance of the volatility is extremely important. It varies from 0.04 per cent for Morocco to 1.2 per cent for Lebanon. For example, for Oman, the volatility during crisis period is 69 times higher than that in the turmoil regime.

The probability of being in the same regime the following period is greater than 0.5 for all series; which indicates that regimes are persistent. Except for Bahrain and Egypt, the transition probabilities p₁₂, p₂₁, p₂₃ and p₃₂ are all significant showing that the transition between tranquil, turmoil and crisis regimes is easy and justify the use of the three-state Markov regime switching model.

Table 3.

Estimation Results for the Markov Regime Switching Model

	Bahrain	Egypt	Jordan	Kuwait	Lebanon	Morocco	Oman	Qatar	S. Arabia	Tunisia	UAE	W. Index
μ	0.0065**(0.00)	0.0210**(0.00)	0.0088**(0.00)	0.0121**(0.00)	0.0091**(0.00)	0.0078**(0.00)	0.0053**(0.00)	0.0103**(0.00)	0.0002**(0.00)	0.0048**(0.00)	0.0169**(0.00)	0.0085**(0.00)
$_{1}^{2}$	0.000003**(0.00)	0.000030**(0.00)	0.000001**(0.00)	0.000007**(0.00)	0.000008**(0.00)	0.000001**(0.00)	0.000003**(0.00)	0.000006**(0.00)	0.000018**(0.00)	0.000001**(0.00)	0.000012**(0.00)	0.000003**(0.00)
$_{2}^{2}$	0.00013**(0.00)	0.00049**(0.00)	0.00002**(0.00)	0.00026**(0.00)	0.00062**(0.00)	0.00003**(0.00)	0.00012**(0.00)	0.00021**(0.00)	0.00032**(0.00)	0.00004**(0.00)	0.00051**(0.00)	0.00003**(0.00)
$_{3}^{2}$	0.0039**(0.00)	0.0035**(0.00)	0.0006**(0.00)	0.0058**(0.00)	0.0122**(0.00)	0.0004**(0.00)	0.0082**(0.00)	0.0072**(0.00)	0.0042**(0.00)	0.0014**(0.00)	0.0100**(0.00)	0.0033**(0.00)
p ₁₁	0.9750**(0.00)	0.9904**(0.00)	0.9652(1.00)	0.9862**(0.00)	0.9850**(0.00)	0.9246**(0.00)	0.9842**(0.00)	0.9852**(0.00)	0.9929**(0.00)	0.9548**(0.00)	0.9868**(0.00)	0.9698**(0.00)
p ₁₂	0.0227**(0.00)	0.0096**(0.01)	0.0348**(0.00)	0.0138**(0.00)	0.0150**(0.00)	0.0589**(0.00)	0.0141**(0.00)	0.0133**(0.00)	0.0071*(0.08)	0.0452**(0.00)	0.0117**(0.01)	0.0302**(0.00)
p ₁₃	0.0023(0.28)	0.0000(1.00)	0.0000(1.00)	0.0000**(0.00)	0.000(1.00)	0.0165**(0.01)	0.0017(0.37)	0.0015(0.33)	0.000(1.00)	0.000(1.00)	0.0015(0.34)	0.000(1.00)
p ₂₁	0.0444**(0.00)	0.0239**(0.00)	0.0263**(0.00)	0.0287**(0.00)	0.0303**(0.00)	0.1775**(0.00)	0.0262**(0.00)	0.0269**(0.00)	0.0130**(0.05)	0.0510**(0.00)	0.0193**(0.00)	0.0279**(0.00)
p ₂₂	0.9488**(0.00)	0.9714**(0.00)	0.9686**(0.00)	0.9626**(0.00)	0.9564**(0.00)	0.7894**(0.00)	0.9575**(0.00)	0.9651**(0.00)	0.9772**(0.00)	0.9254**(0.00)	0.9715**(0.00)	0.9646**(0.00)
p ₂₃	0.0068(0.15)	0.0047(0.18)	0.0051*(0.09)	0.0087**(0.00)	0.0133*(0.06)	0.0331*(0.06)	0.0163**(0.01)	0.0080*(0.09)	0.0098*(0.09)	0.0236**(0.00)	0.0092**(0.04)	0.0075**(0.04)
p ₃₁	0.000(1.00)	0.000(1.00)	0.0000(1.00)	0.0000**(0.00)	0.000(1.00)	0.0000(1.00)	0.0000(1.00)	0.000(1.00)	0.000(1.00)	0.000(1.00)	0.000(1.00)	0.000(1.00)
p ₃₂	0.0163*(0.08)	0.0165(0.18)	0.0159*(0.09)	0.0161**(0.00)	0.0590*(0.06)	0.1687**(0.02)	0.0363**(0.01)	0.0166**(0.05)	0.0147**(0.05)	0.0596**(0.00)	0.0236**(0.02)	0.0112*(0.08)
p ₃₃	0.9837**(0.00)	0.9835(1.00)	0.9841**(0.00)	0.9839**(0.00)	0.9410**(0.00)	0.8313**(0.00)	0.9637**(0.00)	0.9834**(0.00)	0.9853**(0.00)	0.9404**(0.00)	0.9764(1.00)	0.9888**(0.00)
Log(L)	4745.15	40550.6	5332.76	4487.79	4476.45	5780.67	4646.09	4249.92	2880.74	5214.10	3800.51	4711.62
E(d₁)	40	104	29	72	67	13	63	68	141	22	76	33
E(d₂)	20	35	32	27	23	5	24	29	44	13	35	28
E(d₃)	61	61	63	62	17	6	28	60	68	17	42	89

Source: Authors’ calculations.

Notes: E(d₁), E(d₂) and E(d₃) are the expected durations of regime 1, regime 2 and regime 3, respectively. p-values are in parentheses. ** and * indicate significance at the 5% and 10% levels, respectively.

Table 3 reports also the expected duration of the tranquil, turmoil and crisis regimes. This duration varies from 13 days for Morocco to 141 days for Saudi Arabia in the first regime. The expected duration of the second regime is ranged between five days for Morocco to 44 days for Saudi Arabia. The third regime shows an expected duration varying from six days for Morocco to 89 days for the world market index.

Figure 2 displays time series of smoothed states’ probabilities for the world market index and the 11 MENA countries. The same figure shows different degrees of clearly defined states. For example, for Saudi Arabia, regimes are obvious, while for Morocco, the estimated state probabilities do not provide clear separation between the states. Regimes are very unstable as we observe a frequent shift between tranquil, turmoil and crisis periods. Figure 2 shows some common patterns in the switching dates among all series especially around May 2008 when there is an increase in the probability of crisis regime. Also it shows individual patterns in the switching dates for countries that witnessed a revolution during the last period, such as Tunisia and Egypt.

Proposition III: The unidirectional causality in variance between the world market index and MENA countries varies in magnitude across regimes as the information tranfer between financial market seem to be more important during crisis period. It varies also from one MENA country to another according to its degree of financial integration with the more mature financial markets.

Empirical results of the unidirectional causality between the world market index and MENA countries across regimes based on volatility series are presented in panel A of Table 4. Note that for the state 3, we use the first difference of VW and VMENA while these series are not stationary in level. Cross-market volatility dependency varies in magnitude across regimes. Significant evidence of causality is not observed during the tranquil period. While, except for Saudi Arabia, significant evidence of Granger causality from the world index to MENA markets is found during turmoil and/or crisis regime, although with differences in lags. Thus, information transfer between financial markets seems to be more important during financial crisis. The shock in financial market volatility from world market index in financial crisis is transmitted to MENA market. The insignificant impact on Saudi Arabia can be explained by the fact that Saudi Arabia stock market has remained relatively closed to foreign investors and its linkages with the global equity markets are still insignificant despite recent efforts to further enhance its intra-regional integration with the remaining MENA countries’ stock markets. This insignificant impact of causality on Saudi Arabia confirms the result found by Neaime (2012) who studied the spillover effects of the recent financial crisis on MENA countries.

During turmoil period, the stock markets of Egypt and UAE, were the most affected given their strong linkages with global stock markets (Neaime 2012). In fact, the null hypothesis that the world index does not Granger causes these two markets has been rejected at 1 per cent level. Hatemi-J (2012) suggested by using an asymmetric causality test that the UAE market is integrated with the USA market and that the degree of integration is stronger when the markets are falling than rising. For Bahrain, Morocco, Oman and Qatar, the Granger causality is accepted at 5 per cent significance level showing that the spillover effect is less important. Non-resident foreigners are allowed to buy only 25 per cent of the shares of a company in Qatar, 49 per cent in Bahrain and 70 per cent in Oman, while there is unrestricted access to foreign investors in Egypt.

During the crisis period, the spillover effect was different at varying degrees, as some countries responded stronger than others in the region. The Granger causality hypothesis is accepted for Tunisia and six other MENA markets namely Jordan, Kuwait, Lebanon, Oman, Qatar and UAE at the 5 per cent and 1 per cent significance levels, respectively. Stock markets in GCC, mainly, Kuwait, UAE, Oman and Qatar reacted strongly to the breakdown of stock markets in developed economies, as these countries held part of their financial wealth in foreign assets particularly in US bonds and other securities. For Jordan and Lebanon, these countries, offer unrestricted access to foreign investors. The Tunisian market was only marginally affected by the global financial crisis owing to the built up of adequate foreign exchange reserves, low level of economic integration and openness.

Table 4.

Granger Causality Test Results and Correlation Coefficients

	Bahrain	Egypt	Jordan	Kuwait	Lebanon	Morocco	Oman	Qatar	S. Arabia	Tunisia	UAE
Panel A: Granger causality tests
The World index does not cause MENA in calm regime	0.0597	0.3755	0.2107	0.0975	0.0810	2.4131	0.5378	0.2105	2.5405	2.4347	1.7296
The World index does not cause MENA in turmoil regime	5.7837*	8.0580**	3.2370	3.0129	1.2531	3.8681*	5.4135*	6.1550*	1.0241	0.3249	11.577**
The World index does not cause MENA in crisis regime	0.6776	0.4482	9.5292**	6.4736**	6.5729**	3.4084	6.2009**	9.2715**	0.4994	3.6380*	6.9437**
Panel B: Correlation coefficients between the World market index and MENA stock market volatilities
Calm regime	0.06	−0.11	−0.065	0.097	−0.16	−0.17	0.08	0.08	0.30	−0.23	0.057
Turmoil regime	0.40	0.63	0.43	0.24	0.22	−0.166	0.36	0.360	0.02	−0.01	0.55
Crisis regime	0.74	0.83	0.87	0.45	0.75	0.21	0.42	0.79	0.76	0.27	0.81

Source: Authors’ calculations.

Notes: ** and * denote rejection of the null hypothesis at the 1% and 5% levels, respectively. The lag length selection is determined according to the Bayesian information criterion.

Panel B of Table 4 reports correlation coefficients between the world market index and MENA stock market volatilities in different world market regimes: calm, turmoil and crisis. In the first regime, correlation coefficients are very low or negative for some markets such as Egypt, Jordan and Lebanon. The most important correlation coefficient is for Saudi Arabia with a value of 0.30. In the second regime, except for Tunisia, and Saudi Arabia, correlation coefficients are more important and are higher than 0.5 for Egypt and UAE. In the third regime, correlation coefficients are extremely important for all markets except those of Kuwait, Morocco, Oman and Tunisia. The most important coefficient is for Jordan with a value of 0.87. Thus, during periods of financial crisis correlations between the volatilities of various stock markets tend to increase significantly, implying limited benefits from international portfolio diversification in highly volatile market regime.

Conclusion and Policy Implications

This article investigates the time-varying volatility and the volatility of volatility behaviour in MENA region with three distinct states of nature: tranquil period with low volatility of volatility, turmoil regime with high volatility of volatility and crisis regime with extremely high volatility of volatility. The conditional volatility is modelled by using an autoregressive GARCH (1, 1) specification and the spillovers from the MSCI world index to MENA stock markets across different regimes are studied basing on the Granger causality test.

Results over the period from 30 October 2006 to 21 October 2011 show that MENA stock market volatility can be characterised by three regimes: tranquil, turmoil and crisis. Also, significant evidence of causality in variance from the world market index to MENA markets has been proved in the turmoil regime and especially during financial crisis. This causality effect varies from one MENA country to another according to its degree of financial integration with the more mature financial markets.

Although regimes were unstable, we observed some common patterns in the switching dates among all series especially around May 2008 when there was an increase in the probability of crisis regime. Individual patterns in the switching dates are observed for countries that went through a revolution during the last period, such as Tunisia and Egypt.

Policy makers can benefit from the results of this article: First, while transition from calm to turbulent markets is sudden and coincides with a higher volatility period, estimation of regime switching is crucial for policy makers as it allows them to predict financial crises and to estimate their duration in order to determine how they should be managed. Second, seeing the statistical significance of the presence of regime shifting in financial market volatilities, it is interesting for policy makers to predict risk on the light of regime switching models.

Footnotes

Acknowledgements

The authors would like to thank the editors, anonymous reviewers and participants at the ERF 18th Annual Conference for helpful comments. All remaining errors are our own.

Notes

References

Abid

Bahloul

(2011), ‘Regime Switching, Asymmetric Correlation and International Portfolio Choices’, International Journal of Monetary Economics and Finance, 4(2): 175–94.

Alkulaib

Y. A.

Najand

Mashayekh

(2009), ‘Dynamic Linkages among Equity Markets in the Middle East and North African Countries’, Journal of Multinational Financial management, 19(1): 43–53.

Baba

Sakurai

(2011), ‘Predicting Regime Switches in the VIX Index with Macroeconomic Variables’, Applied Economic Letters, 18(15): 1415–19.

Bollerslev

(1986), ‘Generalized Autoregressive Conditional Heteroscedasticity’, Journal of Econometrics, 31(3): 307–27.

Engle

R. F.

(1982), ‘Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation’, Econometrica, 50(4): 987–1007.

Granger

(1969), ‘Investigating Causal Relations by Econometrics Models and Cross Spectral Methods’, Econometrica, 37(3): 424–38.

Hamilton

(1989), ‘A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle’, Econometrica, 57(2): 357–84.

Hamilton

(1994), Time Series Analysis. New Jersey: Princeton University Press.

Hammoudeh

(2008), ‘Sudden Changes in Volatility in Emerging Markets: The Case of Gulf Arab Stock Markets’, International Review of Financial Analysis, 17(1): 47–63.

10.

Hatemi-J

(2012), ‘Is the UAE Stock Market Integrated with the USA Stock Market? New Evidence from Asymmetric Causality Testing’, Research in International Business and Finance, 26(2): 273–80.

11.

Kim

Nelson

(1999), State Space Model with Regime Switching: Classical and Gibbs-sampling Approaches with Applications. Cambridge, Massachusetts, London, England: MIT Press.

12.

Lamoureux

C. G.

Lastrapes

W. D.

(1990), ‘Persistence in Variance, Structural Change and the GARCH Model’, Journal of Business and Economic Statistics, 8(2): 225–34.

13.

Maghyereh

Al-Zoubi

(2004), ‘Modelling the Dynamic Interdependence of MENA Stock Markets: A Multivariate Analysis’, Journal of Economic Research, 9(1): 239–70.

14.

Moore

Wang

(2007), ‘Volatility in Stock Returns for New Member States: Markov Regime Switching Model’, International Review of Financial Analysis, 16(3): 282–92.

15.

Neaime

(2012), ‘The Global Financial Crisis Linkages and Correlations in Returns and Volatilities in Emerging MENA Stock Markets’, Emerging Market Review, 13(3): 268–82.

16.

Neaime

(2006), ‘Volatilities in Emerging MENA Stock Markets’, Thunderbird International Business Review, 48(4): 455–84.

17.

Nelson

D. B.

(1991), ‘Conditional Heteroscedasticity in Asset Pricing: A New Approach’, Econometrica, 59(2): 347–70.

18.

Perlin

(2011), ‘Ms_Regress—The MATLAB Package for Markov Regime Switching Models. Available at SSRN: http://ssrn.com/abstract=1714016

19.

Wang

Theobald

(2008), ‘Regime-switching Volatility of East Asian Emerging Markets’, Research in International Business and Finance, 22(3): 267–83.

20.

J. S.

Hassan

M. K.

(2008), ‘Global and Regional Integration of the Middle East and North African (MENA) Stock Markets’, The Quarterly Review of Economics and Finance, 43(3): 482–504.

21.

Zaher

T. S.

(2007), ‘Middle East and North Africa Markets: Investment Challenges and Market Structure’, SSRN Working Paper Number 1077134.