_____________________________________________________ 1. Introduction The global financial crisis has resulted from the U.S. subprime crisis, which arose from excessive growth of subprime lending as financial institutions offered nonstandard mortgages to individuals with dubious credit profiles. The crisis has been described as a “credit crunch” and a “credit squeeze”, two terms that differ only to the extent of the curtailment in credit supply. Whether it is a “crunch” or a “squeeze”, the present crisis is characterised by a decline in the terms and availability of credit for consumers and businesses, which has adversely affected consumption and investment spending, thus putting the economies of the U.S. and other countries in recession. Another characteristic of the crisis is disorder in financial markets, as financial institutions find it difficult to determine the true (or fair) value of assets that are no longer traded in sufficient volumes. This crisis is far more complex than earlier crises because financial innovation has (through securitisation) allowed the creation of complex assets whose risk profiles are difficult to assess and which are not easy to price. Another characteristic of the crisis is the speed with which it spread from the US to other countries, through a contagious effect that was transmitted via several channels. The financial markets and economies of emerging countries have been adversely affected by the crisis. The objective of this paper is to examine and evaluate the effects of the crisis on the six oil-exporting countries of the Gulf Cooperation Council (GCC): Bahrain, Kuwait, Oman, Qatar, Saudi Arabia and the United Arab Emirates. Specifically, this paper deals with stock market contagion during the early stages of the crisis (2007-2008). For the purpose of empirical analysis, contagion is proxied by the effect of changes in US stock prices on stock prices in these countries. The empirical work was based on a sample of daily data, covering the period January 2007 to December 2008, on stock prices in seven stock markets, including two UAE markets: Abu Dhabi and Dubai. The reason for choosing two markets from the same country is that they are located in two different states (emirates) with different oil endowments and different development policies (more explosive in Dubai than in Abu Dhabi). Since we are also going to examine the effect of oil prices, a sample of daily oil prices covering the same period has been collected. Data on the GCC markets were obtained from the Gulf Investment Corporation, whereas the Dow Jones data were obtained from the Yahoo Finance website. Data on oil prices were obtained from Bloomberg. 2. Effect of the Crisis on Financial Institutions and Markets in GCC Countries

Before we examine the effect of the crisis on the stock markets of the GCC countries, it may be useful to consider its effect on financial institutions and other financial markets in these countries. As is the case anywhere else in the world, financial institutions in the GCC countries have been affected by the crisis via several channels: (i) direct exposure to the US housing market, real estate products or CDOs; (ii) exposure to institutions (both as shareholders and creditors) that have direct exposure as in (i); and (iii) the effect of the crisis on the liquidity needs of the GCC financial institutions. We start with the direct and indirect exposure to subprime losses. By the end of 2008, the GCC countries held foreign assets valued at $1.8 to 2 trillion, about 60% of which were dollar denominated. Only a limited number of GCC banks have publicly admitted exposure to subprime losses, with estimated losses amounting to $3 billion, although some estimates put the figure at $10 billion

(Hussain, 2008). It is, however, believed that most of the losses resulting from direct and indirect exposure to the subprime market have been incurred by sovereign wealth funds such as the Abu Dhabi Investment Authority (ADIA) and the Kuwait Investment Authority (KIA). According to Woertz (2008), these funds represent sophisticated investors with a considerable equity component, which means that their exposure to the current market turmoil must be considerable. They probably invested in risky assets, such as CDOs, more than banks. The KIA injected massive amounts into Merrill Lynch, a US investment bank that has experienced a substantial loss of value as a result of direct exposure to “toxic assets”. The unavailability of exact and detailed figures prevents us from saying anything definitive about the direct and indirect losses incurred by the GCC financial institutions. As far as liquidity needs are concerned, Woertz (2008) argued that “while the direct subprime exposure of GCC banks has been limited thus far, the real problem for the GCC banks lies in the direct exposure to increased costs of funding amidst maturity mismatches and credit exposure to local consumers, project and real estate financing”. What makes things worse is that even before the onset of the crisis, there was already a liquidity squeeze resulting from negative real interest rates. Foreign funds used for speculating on what was thought to be an imminent revaluation of local currencies against the US dollar were withdrawn as speculators realised that it would not take place. Foreign investors also pulled out because of the need for short-term capital to rebalance their portfolios and cover losses, leading to the removal of liquidity from the banking system, particularly from the subsidiaries of foreign banks. As a result of the crisis, financial institutions in the GCC faced (and still do) constrained financing conditions while enduring losses resulting from exposure to the local stock and real estate markets, both of which had experienced bursting bubbles. It seems, therefore, that the status quo has not resulted entirely from, but it has been exacerbated by, the global financial crisis. According to the Business Monitor International (2008), the current liquidity crunch in GCC countries is not particularly linked to the global financial crisis, and it is the result of strong demand by a rapidly growing private sector. The crisis has also affected GCC financial markets other than stock markets. The crisis has to a certain extent disrupted the syndicated loan markets. In the second half of 2008, interbank rates across the GCC soared. In the UAE, the usually liquid one-year market has almost dried up and the three month rates rose 230 basis points in three months. The rise materialised despite cuts in policy rates and the fall in LIBOR, which LIBOR has historically followed. The other liquid regional interbank markets saw similar trends, with Gulf spreads on LIBOR moving far above the medium-term norm. Also, spreads (over LIBOR) for corporate bonds in the GCC rose from around 145 at the end of 2007 to 500 basis points at the end of 2008. GCC countries were affected through exposure to foreign financial markets that have brought in large private capital flows to the region during the recent boom years.

Stylised Facts on the Effect of the Crisis on GCC Stock Markets

Figure 1 displays daily stock prices over the period 2 January 2007-31 December 2008 (or close to the year end) along with the Dow Jones (numbers on the horizontal axis refer to daily observations). A careful examination of the data reveals that the GCC stock markets have endured losses not only because of the global financial crisis but also because these markets were in a bubble before the recognition of the crisis, a bubble that was created by purely domestic factors, as the previous discussion indicates. The locations of turning

points in GCC stock prices relative to the Dow Jones indicate decoupling of the GCC markets from the US market. Consider the figures in Table 1, which reports changes in stock prices in the seven GCC markets normalised on the value of 100 on 2 January 2007. The column “last observation” refers to the price at the end of the sample period. The peak value refers to the highest value of the price and when that value occurred (month/year). We can see that most markets peaked in June 2008, whereas the Dow reached its peak value in October 2007. Thus, even as the US stock market was declining because of the crisis, the GCC stock markets kept on rising.

The column “first to peak” refers to the percentage change in stock prices between the first observation and the peak. Stock markets of the GCC countries rose much faster than the U.S. market, which registered a rise of 13.54%, compared with the Omani market that rose by 115.52%. The last column “peak to last” shows the percentage change in stock prices between the peak and the last observation. Apart from Bahrain, GCC markets lost more of their values than the U.S. market as represented by the Dow. These figures provide prima facie evidence for the proposition that the GCC markets experienced their own bubbles and crashes, which were more vibrant (in the case of bubbles) and more severe (in the case of crashes) than the corresponding bubbles and crashes exhibited by the Dow. Table 2 reports correlations of the GCC markets with the U.S. market. Correlations are not particularly high and they differ significantly from one market to another, ranging between zero in the case of Oman and 0.52 for Saudi Arabia and Dubai. Volatility, as measured by the standard deviation of the daily percentage changes in stock prices, also shows significant variation, with some markets being almost three times as volatile as the U.S. market. Again, these figures indicate that domestic factors that had no relation to the global financial crisis propelled the GCC markets, causing the bubble and crash behaviour that we have seen.

The methodology used to test the contagion hypothesis in this paper was based on the structural time series model of Harvey (1989, 1997), which is used to represent an observed time series in terms of its unobserved components. This model may be written as:

pt

P t It H t where pT is the logarithm of stock prices, t is the trend component, tis the cyclical component, and t is the random component, which is assumed to be white noise. The trend component, which represents the long-term movement in a series, is assumed to be stochastic and linear. This component can be represented by:

Pt

Et where K t ~ NID(0, V K2 ) , and ] t ~ NID(0, V ]2 ) . t is a random walk with a drift factor,t, which follows a first order autoregressive process as represented by equation (3). This is a general representation of the trend, which encompasses all other possibilities. For example, this process collapses to a simple random walk with drift if V ]2 0 ,, and to a deterministic linear trend if V K2 0 as well. If, on the other hand, V K2 0 while V ]2 z 0 then the process will have a trend which changes relatively smoothly.1 The cyclical component, which is assumed to be a stationary linear process, may be represented by:

It a cosTt b sin Tt where t is time and the amplitude of the cycle is given by (ab2) 2 . In order to make the cycle stochastic, the parameters a and b are allowed to evolve over time, while preserving continuity is achieved by writing down a recursion for constructing before introducing the stochastic components. If disturbances and a damping factor are also introduced, we obtain:

ª cosT sin T º ªI t 1 º ªZ t º »« * » « * » ¬ sin T cosT ¼ ¬I t 1 ¼ ¬Z t ¼ such that and 0 = a and I0* b . t are Z *t uncorrelated white noise disturbances with variances V Z2 and V Z2 , respectively. The parameters 0 ≤ ≤ and 0≤≤ are the frequency of the cycle and the damping factor on the amplitude, respectively. Once it has been written in this form, the model can be estimated by maximum likelihood, using the Kalman filter to update the state vector (whose elements are the time series components), as new observations become available. Related smoothing algorithms can be used to obtain the best estimate of the state vector at any point in time within the sample period.2 *

Harvey (1989, pp. 510-511) lists all of the possibilities. Koopman, Harvey, Doornik and Shephard (1995) identify several models including: a constant term model, local level model, random walk with and without drift, local level with fixed slope, a smooth trend model, etc. For details of the estimation method, see Harvey (1989, chapters 4 and 7) and Koopman et al. (1995, chapter 14). For applications of this technique in economics and finance, see Moosa (2006) and Moosa and Gazos (2009).

To test the contagion hypothesis, the basic structural time series model represented by equation (1) must be modified by introducing an explicit explanatory variable. Thus, the modified model is:

P i ,t Ii ,t GpUS ,t H i ,t where Pi,t is the logarithm of the stock price index in market i, such that i = 1,2,...,7 and Pus,t is the logarithm of US stock prices measured by the Dow Jones index.3 The hypothesis of contagion is confirmed if the coefficient on Pus,t is positive and statistically significant. The beauty of this model is that it allows us to examine the effect of factors other than US stock prices even without identifying them explicitly. The effect of these factors is reflected in the behaviour of the components μ and . If these components are statistically significant while  is also significant, then this means that factors other than US stock prices affect stock prices in the GCC countries. If, on the other hand, the components are insignificant while  is significantly positive, this means that the only factor affecting stock prices in the GCC countries is US stock prices, thus providing very strong evidence for the contagion hypothesis. If, however,  turns out to be insignificant while the components are significant, this means that only other factors affect stock prices in the GCC countries, which provides evidence against the contagion hypothesis. Table 3 reports the results of estimating equation (6) over the whole sample period. The results include the final state vector (with the t statistics given in parentheses), the coefficient of determination (R2), the modified coefficient of R determination ( d ) and three diagnostic test statistics: the Durbin Watson (DW) statistic, the Ljung-Box (1978) test statistic for serial correlation (Q), and a test statistic for heteroscedasticity (H).4 We can see that the model is reasonably well-determined2in terms of goodness of fit and the diagnostic tests (all equations have positive Rd ). In all cases the level of the trend (which is equivalent to the constant term in a conventional regression) is highly significant, whereas the slope (which is equivalent to the coefficient on a deterministic time trend in a conventional regression) is significantly negative in five cases.

No seasonal component is included in equation (6) because daily data are used in this study. The modified coefficient of determination, ( R d2 ) is calculated on the basis of the first differences rather than the levels of the variables, and it is more appropriate for trended data. A model with a positive R d is better than a random walk with a drift model ( R d can be negative). The Q statistic is calculated as Q (n, k ) ) where n is the number of au2 tocorrelation coefficients and k is the number of estimated parameters. It has a  (k ) distribution. The test statistic for heteroscedasticity, H(h), is calculated as the ratio of the squares of the last h residuals to the squares of the first h residuals, where h is the closest integer to one third of the sample size. It is distributed as F(h,h).

* Q is distributed as 2 (18), which gives a 5 % critical value of 28.86. H is distributed as F(165,165), which gives a 5% critical value of 1.29.

* Q is distributed as 2 (18), which gives a 5% critical value of 16.91. H is distributed as F(63,63), which gives a 5% critical value of 1.52.

* Q is distributed as χ2 (18), which gives a 5% critical value of 22.36. H is distributed as F(99,99), which gives a

5% critical value of 1.39.

The cyclical component is also significant in some cases. The coefficient on the explanatory variable is significantly positive in three cases only. There is, therefore, more evidence for the role played by other factors than for the role played by US stock prices, providing only weak evidence for contagion. It has been established that stock markets tend to be more strongly correlated when they are falling (particularly when they are falling rapidly). For this reason, equation (6) was estimated by dividing the sample period into two sub-samples, the first extending between the beginning of the sample and observation number 194, which marks the peak of the Dow Jones, whereas the second sub-sample covers the subsequent period.5 The results are reported in Tables 4 and 5, for the two sub-samples. Again, the model seems to be well determined in terms of the goodness of fit and diagnostics, but again the results show only limited evidence for contagion. There is no effect of US stock prices on GCC stock prices in the first sub-sample, but a significant effect on these markets in the second sub-sample, when the markets were going up.6 For nearly eight months stock prices in GCC countries were rising while the Dow was falling, providing some indication that other factors must have been influencing stock prices in GCC markets.

The first sub-sample extends between 2 January 2007 and 4 October 2007. The second sub-sample covers the rest of the period until the end of 2008. It seems, therefore, that the positive effect of US stock prices on the markets of Dubai, Oman and Saudi Arabia over the whole sample period is a reflection of the effect arising in the second sub-sample, when the Dow was going down. These markets were affected by the decline, but not by the rise, of the Dow, which means that the effect is asymmetric.