The Portfolio Advantages of Sukuk: Dynamic Correlations Between Bonds and Sukuk

Abstract

The growth of the Islamic finance sector has been well-documented. One of the most booming sectors has been Sukuk. According to several past studies, non-Islamic investors' interest in Sukuk is due, at least in part, to the diversification benefits that Sukuk provides in the context of a fixed-income portfolio. This paper compares a pair between Sukuk and Bonds in the Malaysian market issued by the same issuer to have an unbiased comparison. Using unconditional correlation methodology provides an initial examination of the relationship between the matched pairs. In addition, this paper adopts the standard GARCH-DCC approach of Engle (2002). This is a generalization of the Bollserslev (1990) GARCH model, allowing for the conditional correlation matrices to be time-varying. The findings reveal that the correlation between bonds and Sukuk is similar to that of bonds, making Sukuk a less appealing type of bond from a diversification standpoint. There are no significant differences between Sukuk and bonds. These finding questions the previously considered differences among different types of Sukuk and supports the argument that some Sukuk might not be compliant with Islamic rules and their structure, as contracts have the same risks for Sukuk holders regardless of the type of Sukuk.

1. Introduction

The last four decades have seen the development of a considerable market in financial vehicles and products consistent with Islamic Law, “Share’ah”. Over the last decade alone, there has been considerable growth, from an estimated $400 billion in 2006 to $2 trillion in 2014 (Hussain et al., 2015) and $4 trillion by 2018. To be compliant with Share’ah, trading contracts should be free of three primary restrictions. Firstly, contracts must be free of “Gharar”, uncertainty or ambiguity, for example, anything that could cause dispute between parties in terms of payments, quality of products, or date of delivery. Secondly, the contract must be free of “Maysir” gambling, where gains for one party create losses for another and do not add value to the economy or community. Thirdly, contracts must be free of “Riba”, interest on loans or usury.

One of the most famous Islamic financial products is Sukuk. This is partly because these products are structured in such a way that overcomes the restriction on “Riba”. In addition, such products are ideally suited to projects relating to real estate development, infrastructure, and development and therefore to countries witnessing rapid economic growth. However, from an investor’s perspective, an interesting characteristic of Sukuk is the frequently reported diversification benefits that they can provide in the context of a broader fixed income portfolio (e.g. Cakir & Raei, 2007). The arguments about diversification advantages of Islamic products were not limited to fixed income products but stocks. Aziz et al. (2021) tested the influence of US financial uncertainty; a comprehensive index accounts for the common variance of 147 time-series financial items (Jurado et al., 2015) on Islamic stocks in Saudi Arabia, Turkey, Indonesia, and Malaysia. They found volatilities in the Islamic stocks are independent of the US index for financial uncertainly, meaning they can provide further diversification advantage for investors in the US. This has helped the development of a Share’ah compliant marketplace. While such products were initially designed to primarily appeal to Muslim investors, what was perhaps initially under-appreciated was how much they would appeal to non-Muslim investors.

Interestingly, Wong and Bhatti (2019) found that many East Asian countries, including developed markets such as Hong Kong and Japan, are developing the Islamic capital market in their financial system regardless of Muslims being minorities. This attractiveness can be partly attributed to the perceived diversification advantages, although there continues to be debate surrounding the exact nature of the relationship between Sukuk and bonds. On the one hand, while there are strong similarities with bonds, a number of papers (e.g. Ariff & Safari, 2012; Ariff et al., 2013; Ashhari et al., 2009; Fathurahman & Fitriati, 2013; Godlewski et al., 2013) have illustrated that Sukuk vehicles display distinct characteristics compared to conventional fixed income securities with respect to market reaction and yield-to-maturity. However, there are also well-reported challenges facing Sukuk with respect to liquidity. Thomson Reuters Zawya (2014) surveyed lead arrangers and investments banks, with 70% saying that they felt that liquidity was a major factor concerning the return and relationship with bonds as Sukuk holders have to hold the security until maturity. In addition, Thomson Reuters Zawya (2014) refers to a shortfall in Sukuk supply in the first half of 2014, to the tune of an estimated $230 billion. These issues, alongside issues relating to differences in the sample period and the data and methodology used, could be factors in the differences observed across empirical findings.

This paper re-examines the relationship between Sukuk instruments and conventional bonds. The majority of previous studies have either used index or portfolio level data or been reliant on small sample sizes. This study bases the analysis on 21 pairs of Sukuk and bonds based on factors such as the market the securities are issued in, the type of issuer, and the maturity. The analysis considers both unconditional and conditional correlations, the latter estimated using a GARCH-DCC framework. The results do suggest that the diversification argument behind investing in Sukuk is more nuanced and complex than some previous studies have suggested. The correlations reported illustrating that Sukuk is often highly correlated with mainstream bonds and the coefficients fall within a similar range to those across bonds. The degree of diversification potentially available is quite specific to the individual case in question. The paper is structured as follows. Section 2 provides an overview of Sukuk products. Section 3 discusses some of the pertinent literature, while Section 4 details the methodological framework adopted. Section 5 presents the empirical results, and the final section provides concluding comments.

2. Sukuk Financial Products

The origins of Sukuk, meaning “financial certificates” in Arabic, arose as the Islamic finance sector developed, and there was a resulting demand for a Share’ah compliant equivalent to fixed income securities. The primary difference is that the certificates must represent undivided shares in the ownership of tangible assets and services (Vishwanath & Azmi, 2009). The Sukuk structure was developed in Malaysia in 1990, and although it took some time to develop, issuance subsequently grew considerably, from $5 billion in 2002 to $477 billion in 2019 (Fahy, 2019). Franklin Templeton expects exponential growth in the Sukuk market to reach $2.7 trillion by 2030 (Fahy, 2019). However, it is worth mentioning the finding of Basyariah et al. (2021), which linked the growth in the Sukuk issuance to the macroeconomic performance of the issuance markets. Interestingly, sovereign issuers represent 80% of issuance (Standard & Poor, 2013). The dominant role of sovereign issuers is related to the use of Sukuk to finance infrastructure, real estate, and pipeline projects across the Middle East and Asia. The usefulness of Sukuk in such contexts was clearly illustrated in some of the early sovereign issues. In 2003 the Qatar Government raised $700m to finance the development of Hamad Medical City (MEED, 2003a). In the same year, Emaar Properties, a United Arab Emirates (UAE) government-owned company raised $50m for real estate development projects (MEED, 2003b). In 2004 a $120m issue financed the Durrat al-Bahrain resort (MEED, 2004). After the success of such issues, Sukuk has become a major source of funding for sovereign and quasi-sovereign issuers. Dubai raised $1bn to part-finance the $4.1bn Dubai International Airport (Euroweek, 2004), while Nakheel, a UAE government-related company, raised $3.5bn to finance landmark developments in Dubai such as the Palm, Jumeirah, and the World (Euroweek, 2006). In 2007, Saudi Arabia started issuing Sukuk across a range of sectors¹. In addition, since 2002, the major credit rating agencies have also been active in the Sukuk market, thereby aiding in investor awareness.

When examining Sukuk it is not only important to note the differences with conventional bonds but also those present across the alternative forms of Sukuk. There are four primary forms of sukuk, Murabaha, Musharaka, Mudaraba and Ijara. However, even within each category, the specifics of an individual deal may lead to significant effects when it comes to pricing and risk. As Sukuk is considered an alternative to conventional bonds, it is perhaps best to initially start with a clear definition of how Sukuk differs from bonds. Whereas bondholders are creditors, receiving interest payments from the borrower, Sukuk represents ownership of assets and services to Sukuk holders. In addition, because Sukuk is based on one of the nominal share’ah commercial contracts, it is expected to have some differences in expected risk and return as well as differences in issuance rights and responsibilities. However, Sukuk structures are more complex than plain vanilla bonds due to the need to fix returns and face value until maturity, whilst maintaining their share’ah compliance.

3. Literature Review

Over the last 10 to 15 years there has been a growing literature to have examined the relationship between Sukuk and the bond market. Much of this literature has considered the possible diversification benefits that may arise from including Sukuk in a fixed-income portfolio. Cakir and Raei (2007) examined a sample of securities from Bahrain, Malaysia, Pakistan, and Qatar between 2000 and 2007, evaluating whether the inclusion of Sukuk has a beneficial impact on the maximum potential loss that may occur. While the authors do report that the inclusion of Sukuk did reduce risk, as measured by Value-at-Risk (VaR), there are important caveats in this study. Specifically, only one Sukuk security per country was included in the analysis, while the bond sample was limited to, at most, three securities. This naturally increases the idiosyncratic nature of the findings and also makes the results depend upon the Sukuk type analyzed. Additionally, all of the Sukuk were floating rate, while the bonds were all fixed-rate, thereby creating an additional point of difference between the two groups. There were also significant differences in the range of maturities used, from 5 to 30 years. All of the above issues could influence the outcome of the empirical tests and result in the low correlations reported in the paper.

Mosaid and Boutti (2014) examine whether Malaysian bond and Sukuk portfolios display a significant difference in their means over the 2007 to 2012 period. The paper uses indices rather than individual issues, and from these, five Sukuk and five bond portfolios are constructed based on varying maturities. In four of the five cases, an insignificant difference in means was reported, while in each case, a strong positive correlation was also noted. These findings would appear to bring into question some of the other literature and the common perception within the industry that Sukuk can provide diversification benefits. However, there is a major caveat with this study and that is that the sample period focused on 2007 to 2012 and hence is heavily influenced by the Global financial crisis and its immediate aftermath. It is quite likely that this alone would have contributed to a heightened positive relationship. It is therefore an important component in the current study to re-examine these issues during a period without the major global shocks that were observed a decade ago. Additionally, as index level data was used the paper didn’t distinguish between types of Sukuk or differences that may have arisen due to factors such as issuer or capital size.

Another approach in examining the relationship between bonds and Sukuk is to compare Yield-to-Maturity. Ariff and Safari (2012) use this approach on a Malaysian sample from 2005 to 2011². Examining monthly index data, they test the relationship between bonds and Sukuk by running paired samples, each with the same duration and the same category of the issuer. The sample includes a variety of issuers, including; sovereign and quasi-sovereign issuers, AAA- rated financial institutions, and AAA-rated corporate, and for various maturities ranging from 3 months to 20 years. Across the 64 pairs 46 report statistically significant differences in the yield-to-maturities. In the majority of cases, a lower yield is reported for bonds. The paper also tested for causal relationships, although a limited number of significant findings were reported. In tests examining whether bond yields granger caused Sukuk yields only 13 significant cases were noted, while 10 significant cases were reported in the reverse direction³. Fathurahman and Fitriati (2013) adopted a similar approach, in this case examining the Indonesian market. The Sukuk sample comprises 31 issues. These issues are then compared to ten randomly selected groups of bonds from an overall sample of 234. A lower proportion of significant differences are observed in this case, but again, Sukuk issues are noted to have higher yields. A major caveat with this paper, and also Ariff and Safari (2012) and Ariff et al. (2013), is that all three papers are heavily reliant on a sample period focused on the financial crisis and its immediate aftermath.

An alternative methodological approach that has been used, is to examine the market reaction to bond and Sukuk issues and whether any abnormal returns are observed, and in turn, if those responses differ. Ashhari et al. (2009), Godlewski et al. (2013), and Alam et al (2013) have all used this approach and have broadly found similar results in that significant abnormal stock returns were observed for Sukuk issuances, in comparison to no significant reaction to bond issues. What is of interest is that the direction of the response differs. Ashhari et al. (2009) examined the Malaysian market from 2001 to 2006 and reported significant positive abnormal returns the day before the announcement of a Sukuk issuance. They also note that there was evidence of differences across company size, with larger companies tending to offer issue bonds and smaller firms Sukuk. Godlewski et al. (2013) examined a Malaysian sample of 77 Sukuk and 93 bonds, finding that the market reacted negatively to Sukuk announcements. As with Ashhari et al. (2009), they noted that the companies issuing Sukuk were smaller and that additionally, they had higher existing debt levels. Godlewski et al. (2014) also found that the type of Sukuk influenced the market reaction to the announcement. et al (2013) extended this methodological approach to consider six countries, Malaysia, Indonesia, Singapore, Pakistan, UAE, Bahrain, and Qatar. Their sample contains 79 Sukuk and 87 bonds over the period 2004 to 2012. An important caveat with the interpretation of their findings is that none of their sample companies issued both Sukuk or bonds; in every case, the firms only issued either Sukuk or bonds. As with the prior studies, the Sukuk issuing companies tended to be smaller. Their results indicate that before the financial crisis, there was an adverse market reaction to Sukuk issues; however, post-crisis, the market response was positive.

4. Data and Methodology

4.1. Data

Daily bond and Sukuk data were sourced from Thomson Reuters. To provide as direct a comparison as possible, we adopt a different approach than that adopted in many other papers that have considered the dynamics present between Sukuk and conventional bonds. The majority of those papers have relied upon either index or combined portfolio data. We, however, examine pairs of bonds and Sukuk that are as closely matched as possible. This approach has been commonly adopted in the general fixed income literature as a robust means of comparing performance (e.g. Bali & Skinner, 2006).

A number of criteria were used to select the most appropriate pairing. Firstly, each pair is issued by the same issuer, eliminating the uncertainty of risks associated with parties within one pair. Secondly, the size of the issue should be broadly comparable. Of the pairs examined only one, Pair 10, displays a non-insignificant difference in issuance size. Thirdly, the maturity date of the two issues should be closed. Fourthly, we impose the restriction that the number of matched daily observations is no less than 250 for an entire year to ensure an adequate sample size. Finally, all issues must be categorized as plain vanilla issues, have fixed semi-annual coupon payments, and no options be associated with them. Finally, none of all the issues can be guaranteed by a third party. The imposition of these restrictions means that the Malaysian market is the only viable to examine. A total of 21 pairs were identified and are detailed in Table 1. The issuer, category, type of Sukuk, capital size, maturity size, and sample size are detailed. Pair 1 was however removed from the analysis as both the bond and Sukuk matured during the sample and as will be shown, both securities displayed low volatility as they converged on the redemption price. Table 2 displays descriptive statistics for all of the series after converting them to log returns. A few issues warrant noting. Firstly, all of the series are stationary at a 1% significance level. Second, in the majority of pairs, the Sukuk standard deviation is higher than that for the corresponding bond.

Table 1: Sample Details

Notes: * Bond and Sukuk in pair 1 matured while conducting this research. ** and *** both types of Sukuk fall under the Sukuk Murabaha type

Table 1: (Continued)Notes: * Bond and Sukuk in pair 1 matured while conducting this research. ** and *** both types of Sukuk fall under the Sukuk Murabaha type.

Table 2: Summary Statistics

4.2. Methodological Framework

Unconditional correlations provide an initial examination of the relationship between the matched pairs. The unconditional tests also split each paired sample into two equal parts to consider the consistency of the findings. Naturally, given that each pairing covers a different specific period, this does mean that broad statements across all of the pairs are limited. We, therefore, also construct a common period to estimate the correlation coefficients. This is done as follows. We take the pair with the shortest period observed for any pair. After ignoring pair 1 this is pair 16 which has 335 observations. We then match all of the other pairs to that sample period, producing correlation tests for the last 335 observations. This sample is referred to as L-335 (Table 1).

While the unconditional correlations do provide an opportunity to initially examine the time-varying nature of the relationship, the simple analysis of sub-periods is limited as it just considers the relationship across discrete periods. Therefore, the primary empirical analysis considers time varying dynamic correlations (Table 2).

There are a number of options available, including a simple rolling window for the estimation of unconditional correlations. As noted by Forbes and Rigobon (2002), heteroscedasticity causes simple rolling correlation coefficients to be biased upward in periods where one of the assets increases in volatility. The issue of volatility and trading volume is especially important in the context of emerging markets. Although Sukuk has been growing very fast, the majority of Sukuk holders tend to hold till maturity (Thomson Reuters Zawya, 2014). This will be in turn, lead to a dampened level of trading volume and potentially lead to bias entering the results. A large number of studies have examined the relationship between trading volume and volatility, including in an emerging market context (e.g. Pyun et al., 2000). In addition, the rolling correlation method has no theoretical or empirical basis for selecting the size of the window or the number of observations to be included in the window. The other simple method, exponential weighted moving average, avoids some issues, such as all observations within the window have the same weight, but it is still not based on theoretical or empirical information in terms of choosing the value of the smoothing parameter λ.

We, therefore, adopt the now-standard GARCH- DCC approach of Engle (2002). This is a generalization of the Bollserslev (1990) GARCH model, allowing for the conditional correlation matrices to be time-varying. The variance of each return series is modeled using the univariate GARCH process, and the conditional correlation between the return series is directly parameterized. The model follows the same steps as in the univariate GARCH and initially estimates the GARCH (1, 1) and then uses the resulting standardized residuals to estimate the varying correlation matrix. This requires transforming the residuals by their estimated standard deviations as below:

\(h_{i i, t}=\gamma_{i}+\alpha_{i} \varepsilon_{i, t-1}^{2}+\beta_{i} h_{i i, t-1}\)       (1)

\(z_{t}=\varepsilon_{t} / \sqrt{h_{t}}\)       (2)

\(\varepsilon_{i, t}=\sqrt{h_{i i, t}} z_{t} \text { and } \varepsilon_{i, t} \sim N\left(0, h_{i i, t}\right)\)       (3)

\(q_{i j, t}=\bar{\rho}_{i j}+\alpha\left(z_{i, t-1} z_{j, t-1}-\bar{\rho}_{i j}\right)+\beta\left(q_{i j, t-1}-\bar{\rho}_{i j}\right)\)       (4)

\(\rho_{i j, t}=\frac{q_{i j, t}}{\sqrt{q_{i i, t} q_{i j, t}}}\)       (5)

\(\bar{\rho}_{i j}\) is the unconditional correlation between z_i,t and z_j,t .z_t represents the standardized residuals. The covariance matrix is H_t ≡ D_tR_tD_t, where D_t is a diagonal matrix of univariate GARCH volatilities. R_t^t = Q_t^*-1Q_tQ_t^*-1 is the time-varying correlation matrix, and Q_t as below:

\(Q_{t}=(1-a-b) \bar{Q}+a\left(z_{t-1} z_{t-1}^{\prime}\right)+b Q_{t-1}\)       (6)

Where \(\bar{Q}\) is the unconditional covariance of standardized residuals, Q* is a diagonal matrix composed of the square root of the diagonal elements of Q_t. To generate the coefficients of Univariate GARCH as well as GARCH- DCC, the models are estimated by the maximum likelihood procedure using the algorithm of Broyden-Fletcher Goldfarb- Shanno (BFGS).

5. Empirical Results

5.1. Unconditional Correlations

Table 3 shows the unconditional correlation results across the 21 pairs. It is evident that Sukuk is highly correlated to bonds, with most of the pairs having a correlation above 0.50. The average for all pairs across the full sample is 0.6118, with no major differences observed for sub-sample 1, sub-sample 2, and L-335. The fact that daily data is being analyzed makes these findings especially noticeable. None of the pairs report a negative correlation, and only two pairs (5 and 7) display a low positive correlation in the full sample test; however, these two do display an increased correlation in sub-sample 2 and the L-335 tests. If, however, one compares the unconditional correlations in sub-sample1 to sub-sample 2, there is noticeable inconsistency. Pairs 3, 4, 12, 13, 17, 20, and 21 all show reductions in the correlation over time, while pairs 1, 5, 7, 10, 14, 15, and 19 report increases. The remaining pairs show minor changes in either direction. These findings would imply that the type of issuer, or Sukuk, does not influence the relationship in any one specific direction; instead, there is a degree of independence. For example, pairs 4 and 5 are both issued by the government, but they display opposite changes in the correlation coefficient over time. Another example is pair 14 and pair 17, both of which are Musharaka Sukuk and are issued by financial institutions, yet they show opposite changes in correlation over time. This could be associated with issues related to the issuers rather than the market, in general, and this could be the reason for changes in the correlation in some of the pairs, such as pair 3 or pair 12.⁴ We also report evidence of a high correlation between bonds and Murabaha sukuk, Musharaka sukuk, Bai’Inah sukuk, Bai Bithaman Ajil Sukuk and Ijara Sukuk. In all cases, high correlations with bonds are reported. In addition, with the exception of Ijara Sukuk, which saw a decline from a strong positive correlation, the findings are consistent.

Table 3: Unconditional Correlations

Notes: This table reports the unconditional correlation coefficients reported across the entire sample period of each pair, and three subperiods, including the first half, second half, and the last 335 observations for each pair. The final row in the table reports the average correlation coefficient reported across each pairing for each period.

These correlation results between bonds and Sukuk are actually within the boundaries of relationships within bonds themselves. Indeed, adding Sukuk to a bond portfolio might improve its performance; however, the improvement is not the same as adding an independent instrument, as Sukuk was assumed in earlier studies, but most probably like adding another bond to a bond portfolio with a different issuer or a different rating.

5.2. Conditional Correlations

While unconditional correlation can provide an indication of the relationship between bonds and Sukuk, it does ignore the time-varying dynamics and how correlations may change over time. Therefore, it is important to consider the time-varying dynamic correlations to provide an alternative perspective upon the relationship between bonds and Sukuk. Table 4 reports the GARCH (1, 1) estimations for each instrument within each pair separately, and also the GARCH-DCC (1, 1) estimation for each pair. Overall, the GARCH-DCC seems to provide a good representation of the data’s conditional variance. In the GARCH model for individual instruments, most of the instruments display very high β_i, very low α_i, and close to unified α_i + β_i, indicating a strong persistence in volatility. In the GARCH- DCC model, the parameters a + b, which represent the conditional covariance, showed positive and significant findings for all pairs, with the exception of 7 and 16. This would indicate a strong interaction between bonds and Sukuk within each pair.

Table 4: Dynamic Conditional Correlations

Notes: This table reports the coefficients from the GARCH-DCC estimations. The γ, α, and β coefficients refer to the respective GARCH (1,1) model, with a subscript of 1 referring to the bond and a subscript of 2 referring to the Sukuk of each pair. The a and b coefficients refer to the GARCH-DCC (1,1) estimates. *, ** and *** indicate p-value at the 10, 5 and 1% levels, respectively

Another interesting finding is related to the assumption that types of Sukuk are different from each other because of the nature of their structure and underlying assets, if any. In the GARCH-DCC tests we find positive and significant parameters, a + b, close to unity for pairs containing Murabaha sukuk, Musharaka sukuk, Bai’Inah sukuk, Bai Bithaman Ajil sukuk and Ijara sukuk. This questions the existence of differences among Sukuk types; even if they take different approaches or have different structures, they may end up with the same result, in which case the market would treat them the same. There are some noticeable points when comparing the unconditional correlation findings in Table 3 to the average conditional correlation finding in Table 5. Except for pairs 5 and 7, in those cases where there was a noticeable change in the conditional correlation, the direction of the change was downwards, with a lower conditional correlation reported. Furthermore, the mean conditional correlations, excluding pair 1, were 0.5883 in comparison to 0.5991 with respect to the unconditional coefficients. This could be due to the unconditional estimates failing to capture the volatility and upward/downward trends over the sample period.

Figure 1 shows a graphical display of the conditional correlation over time for each of the pairs. Overall, it is hard to prove the upward trend of the relationship between bonds and Sukuk because of data availability and limited time to test them. Chong et al. (2012) included 18 years of daily observations to test Real Estate Investment Trust (REIT) behavior. Another interesting finding shows that the majority of the pairs displayed stable levels of correlation over time. Also, it can see, especially in the case of pairs 9, 11, 13, 17, and 18, that correlation levels recover quickly after aftershocks. In addition, very few of the pairs have a negative conditional correlation over an observed period. The exceptions are pairs 3, 12, and 21, which report a small number of negative observations, and pair 7, which displays a continuous negative conditional coefficient. Pair 7 is indeed the only pairing where considerable fluctuation in the conditional correlation was noted, thereby implying possible enhanced diversification benefits compared to the sample.

Figure 1: Pairs Time Series Plots for GARCH-DCC

Note: This graph graphically displays the conditional correlation coefficients, as estimated using the GARCH-DCC (1,1) procedure, for each of the pairs.

When considering the effect of Sukuk, it is interesting to observe different patterns over time. Although each pair showed its fluctuation between bond and Sukuk, there was not a clear pattern for any of the Sukuk types. Also, we could not find a type of Sukuk with a constantly high and stable correlation with its bond pair, nor could we find a type of Sukuk with constant high levels of fluctuation in its correlation with bonds. It should be explicitly noted that while pair 12 showed high levels of fluctuation between bond and Sukuk, it is the only pair containing Sukuk Ijara, and therefore it is not possible to compare it to another (Table 5).

Table 5: Summary Statistics of Conditional Correlations

Notes: ‘Trend’ is the slope coefficient of regression of conditional correlations on a constant and a time trend. The sample covers the full period for each pair. *, ** and *** indicate p-value at the 10, 5 and 1% levels, respectively

5.3. Time Trends

One of the goals of this paper is to examine possible changes in Sukuk behavior over time as well as its homogenous movement with regard to the fixed-income sector. To consider this issue, we regress the conditional correlation against a time trend. Significant results with positive signs would indicate enhanced integration between bonds and Sukuk over time, and vice-versa with respect to negative results. Table 6 presents the finding for each pair. It is immediately noticeable that half of the pairs display negative signs, and therefore not supportive of the hypothesis of increased integration between bond and Sukuk. This could be a result of a number of factors. While it may be due to Sukuk not being integrated into the fixed income sector, it may also be related to factors specific to individual pairs. This is an issue that is especially important to consider given the challenges present due to the short sample periods available. For example, Chong et al. (2012) studied REITs and reported a reduction in conditional correlations in many of the tests over a short period. However, over the entire 18 year sample period used significant positive integration was found and reported. The short-sample period, therefore, does make it hard to make definite conclusions from this analysis. However, not having conclusive evidence to support the premise of increased integration does not mean that there are differences between bonds and Sukuk. Rather it may imply that there are either no changes over time or that there is a degree of dispersal over time, which might be large or small. Looking back to Figure 1 it is clear that conditional correlations reduced over time across the pairs.

Table 6: Conditional Correlations and Volatility

Noted: The results are obtained from estimating the regression     t B     index hB t index  . The conditional volatilities and covariances are calculated as the fitted values. The conditional correlations are measured as the ratio of the conditional covariances to the product of the conditional volatilities. R2 is the adjusted coefficient of determination statistic. The sample covers the full period for each pair. *, ** and *** indicate p-value at the 10, 5 and 1% levels, respectively

5.4. Conditional Correlations and Volatility

The last empirical issue considered in the empirical analysis is concerned with the relationship between bonds and Sukuk during periods of high volatility. To do so we test the relationship between the conditional correlation and conditional volatility by regressing the former on the latter as shown below:

\(\rho_{t}=\alpha+\beta_{B-\text { index }} \sqrt{h_{B-\text { index }}}+\varepsilon_{t}\)       (13)

We use the Thomson Reuter 5-year Government Bond Index as the measure of conditional volatility, obtained through the estimation of a standard GARCH (1, 1) model. Having positive beta coefficients suggests that conditional correlations rise with heightened bond volatility. Although we significant and positive betas in only 11 out of 20 cases, interestingly these pairings are those that have the conditional correlations that fluctuate the most, as clearly seen in Figure 1 with respect to pairs 3, 5, 12, 15, 17, and 21. Three pairs showed significant negative signs, and six were insig- nificant. Those insignificant pairs may be affected by the stability of their conditional correlation over the sample period.

6. Concluding Comments

This paper has attempted to find a robust answer to how bonds and Sukuk are related, taking into consideration the importance of using the most comparable Sukuk and bonds, then using appropriate methodologies to test them. The results, from both unconditional and conditional correlations, find that Sukuk is not different from bonds even with all the differences in their structure or type of contracts. The correlation between bonds and Sukuk falls within the correlation of bonds themselves, making Sukuk a type of bond that is less attractive from diversification perspectives. This finding is the most important point to consider as it questions the future attractiveness of Sukuk and whether they will have an advantage against bonds, except for complying with Islamic roles. Another major finding is that bonds and Sukuk are not only highly correlated, but there are no significant differences from Sukuk to bonds. This finding questions the previously considered differences among different types of Sukuk and also supports the argument that most Sukuk is not truly compliant with Islamic rules and their structure, as contracts have the same risks for Sukuk holders regardless of the type of Sukuk. This shows a further absence of significant differences in Sukuk performance compared to bonds. Also, although this paper could not provide definite evidence of increased integration between bonds and Sukuk the results do show a high correlation in almost all of the pairs in comparison to previous studies. These findings would support the argument that the relationship between bonds and Sukuk has altered over time.

Overall, the paper provides additional empirical evidence to help in the understanding of Sukuk’s place within the fixed income market. Although the expected demand for Sukuk is high, this might not be driven by the diversification advantage that is assumed to emerge after including Sukuk in a portfolio with other assets. Kuwait and other emerging countries with strong financial ratings can still consider this option as an alternative source of funding for their major development plans. Furthermore, those countries that are considering issuing Sukuk to fund their development projects should note that, as there is no significant difference in performance between different types of Sukuk, they might need to go with the type most preferable to them, or the easiest one, such as Sukuk Murabaha. It is not necessarily worth issuing the other, more complicated types.