key: cord-0896637-mvriep14
authors: Rizwan, Muhammad Suhail; Ahmad, Ghufran; Ashraf, Dawood
title: Systemic risk, Islamic banks, and the COVID-19 pandemic: An empirical investigation
date: 2022-02-18
journal: Emerging Markets Review
DOI: 10.1016/j.ememar.2022.100890
sha: 8698c4c051ce6a899a56e0364a7307958149413b
doc_id: 896637
cord_uid: mvriep14

While operating side-by-side with conventional banks, in a dual-banking system, the systemic risk profile of Islamic banks can be different due to their unique business model. The objective of this study is to understand the evolution of systemic risk in dual-banking systems and determine whether there are any differences in the systemic risk profiles of conventional and Islamic banks during the COVID-19 pandemic. This study also identifies the determinants of systemic importance (measured using spillover indices) of financial institutions. The sample includes ten countries where the Islamic banking sector is considered systemically important and covers the period from November 2015 to November 2020. The empirical results indicate a significant increase in systemic risk, in the sample countries, during the first half which is followed by a recovery in the second half of 2020. Comparative analysis shows that Islamic banks have similar systemic vulnerabilities to systematic and idiosyncratic factors during the exogenously induced real economic shock of the COVID-19. However, Islamic banks pose significantly less spillover to others relative to conventional banks while earning abnormal returns. The results are robust to the inclusion of macroeconomic factors and alternate estimation methodologies. The findings of this study provide valuable insights for the regulators of dual-banking systems.

Since the onset of the COVID-19 pandemic, a series of lock-downs have been imposed to curb its impact on healthcare infrastructure globally. The economic consequences of lock-downs present challenges of enormous complexity and magnitude for governments, multilateral development institutions, and non-government organizations. COVID-19-related measures, both lock-downs and social distancing, have affected all sectors, especially those relying on social interactions such as tourism (Škare et al., 2021) , hotel and lodging (Alonso et al., 2020) , agriculture (Boughton et al., 2021) , aviation and air travel (Suau-Sanchez et al., 2020; Sun et al., 2020; Iacus et al., 2020) , and small and medium enterprises (Shafi et al., 2020) . At the same time, volatility spillover has increased significantly in financial and commodity markets (Corbet et al., 2021; Salisu et al., 2020) .

The World Bank estimates that due to the pandemic the global economy shrunk by 4.3% in 2020 alone. 1 Such a contraction in the global economy has raised several flags for the financial and economic sectors (Jena et al., 2021) . In most countries, governments were quick to devise crisis management plans to tackle the immediate challenges of mounting unemployment and corporate bankruptcies.

For the corporate sector, government responses included a debt moratorium for a short period and easy and quick access to credit through government-guaranteed short-term to medium-term loan schemes. 2 Both of these measures alleviate the immediate risk of loan impairment albeit by compromising intermediation standards, and loss of income for the banking sector. 3 Many central banks announced several measures for liquidity support to the banking sector, including (i) lowering reserve requirements, (ii) lowering regulatory capital buffers, (iii) bond/S˙ukuk buying programs, and (iv) availability of central bank credit lines (reverse repo). But, these measures were not uniformly offered to all banks in countries with both conventional and Islamic banks. For example, the first two measures help both conventional and Islamic banks however, in some countries, Islamic banks were deprived of the last two measures which are crucial for liquidity management support either due to an absence of credit lines for Islamic banks or legal impediments (IsDB, 2020) . This further adds to the challenges for the sustainability of Islamic banks.

Recent literature on systemic risk provides some initial evidence in support of heightened systemic risk vulnerabilities during the pandemic. For example, Rizwan et al. (2020) study the evolution of systemic risk in eight countries that were most affected by the COVID-19 during the earlier phase of the pandemic. Their results show a significant increase in systemic risk during the first quarter of 2020 which remained at an elevated level for the second quarter of 2020 among all countries except for China which showed some recovery. Similarly, Borri and di Giorgio (2021) report a significant increase in systemic risk during the pandemic in European countries while Lai and Hu (2021) show higher risk in the financial networks of twenty countries during the period August 2019 to March 2020. However, in all these studies, the focus remained on countries with conventional banking systems that follow debt-based business models.

The business model of banks has serious consequences for the stability of banks (Ashraf et al., 2016) . 4 The business model for Islamic banks is quite different from that of conventional banks in terms of their asset-liability structure and product offering (Ashraf et al., 2016; Olson and Zoubi, 2017) . The equity-based and risk-sharing nature of Islamic banks helps reduce the maturity mismatch of assets and liabilities and enhances financial stability (Hasan and Dridi, 2011; Beck et al., 2013) . Furthermore, there is empirical evidence suggesting that Islamic banks were relatively more stable during the global financial crisis (GFC) of 2007-09 5 albeit they were affected in the later stages of the crisis due to the deterioration of the real economy (Hussien et al., 2019) .

Since the deterioration of the real economy is observed with the onset of the COVID-19 pandemic, an exogenous shock, the impact of the crisis on Islamic banks is expected to be comparable to that of conventional banks (IsDB, 2020) . However, we do not find any empirical studies analyzing the systemic risk contribution of Islamic banks in dual-banking systems where both conventional and Islamic banking services are offered side-by-side during the COVID-19 pandemic. This study aims to fill this gap by investigating the evolution of systemic risk in a sample of countries with dual-banking systems, and identify and compare the determinants of spillovers through their connectedness, before and during the pandemic, for conventional vis-à-vis Islamic banks.

To evaluate the evolution of systemic risk during the sample period, we use CATFIN (Catastrophic Risk in the Financial Sector), proposed by Allen et al. (2012) , which captures the aggregate risk-taking by the entire financial sector and complements bank-specific systemic risk measures by forecasting macroeconomic downturns. The systemic risk contributions are estimated using the connectedness measures: "spillover to others" and "spillover from others" as proposed by Diebold and Yılmaz (2014) . Spillover to (from) others measures the systemic vulnerabilities posed to (by) the financial system by (to) an individual bank.

We use market-data based measures in our analysis because of two reasons. First, from a systemic risk perspective, the market mechanism can directly or indirectly discipline financial institutions. Market prices of debt and equity may increase funding costs for banks and, therefore, prompt a direct market discipline. Supervisory authorities may benefit from very high frequency market data to complement traditional financial statement data for assessing bank fragility to identify systemically important banks (indirect market discipline) using price signals as screening devices or inputs into early warning models (Huang et al., 2009) . Similarly, Gropp et al. (2006) argue that asset market data should be valuable in assessing systemic risk as it arises from inter-temporal decisions while prices in the capital markets provide information about inter-temporal marginal rates of substitution and transformation.

Secondly, given the impact of the COVID-19 pandemic on the banking sector and a lag in receiving accounting data, this empirical study uses several variables derived from the market prices of sample banks. We use the abnormal returns of a bank (Jensen's alpha), market-specific risk exposures (systematic risk), and bank-specific risk exposures (idiosyncratic risk) estimated from a standard capital asset pricing model (CAPM), extended CAPM (Fama and French, 1993) along with bank-specific and macroeconomic variables as a possible set of determinants. For the inter-temporal (before and during the COVID-19 pandemic) and banking model (conventional or Islamic) differential analysis, we used intercept and slope dummies in the regression analysis.

Our sample consists of ten countries with dual-banking systems where the Islamic banking sector is categorized as systemically important by the Islamic Financial Services Board IFS, (2020). The sample countries are Bahrain (BHR), Bangladesh (BNG), Jordan (JRD), Saudi Arabia (KSA), Kuwait (KWT), Malaysia (MLY), Oman (OMN), Pakistan (PAK), Qatar (QTR), and United Arab Emirates (UAE). Using monthly data from November 2015 till November 2020, we find that systemic risk in the sample countries increased significantly during the first quarter of 2020 suggesting an adverse effect of the pandemic. However, recovery has been observed in all sample countries during the second half of 2020. We also observed a positive association between systematic risk and spillover indices suggesting that higher exposure to market risk may increase systemic vulnerabilities. The negative association of idiosyncratic risk with spillover from others, for both conventional and Islamic banks, validates the implementation of macro-prudential regulations in 2 IMF Policy Tracker: Policy Responses to COVID-19 3 S&P Global: COVID-19 Credit Update 4 International Monetary Fund, (2011). Global Financial Stability Report. Online: https://www.imf.org/external/pubs/ft/ar/2011/eng/ 5 See for example Č ihák and Hesse (2010). response to the GFC of 2007-09 when micro-prudential regulations were unable to control systemic vulnerabilities (Claessens, 2014) . This is in line with the literature that suggests that managing the idiosyncratic risk of financial institutions does not ensure the stability of the entire financial system (Suh, 2019) .

Comparative analysis reveals that Islamic banks trigger similar systemic concerns as conventional banks with their exposures towards market factors. Furthermore, Islamic banks are as vulnerable as their conventional counterparts during exogenous shocks, such as the COVID-19 pandemic, where the real economy is adversely affected. However, positive abnormal returns of Islamic banks provide systemic stability as compared to their conventional counterparts during such times.

We estimated the CAPM using Fama and French's three-factor model and International CAPM. Our main findings remained robust with alternate models. The empirical findings from the Fama-French three-factor model suggest that banks smaller in size, which are required to pay higher premiums under a CAPM framework, contribute less spillover to others, while banks exhibiting higher momentum in their stock prices contribute more to systemic risk. We also find a negative association of market capitalization with the spillover of banks suggesting that a decline in the market value of a bank may increase the appetite for risk-taking that, potentially, has adverse consequences for the entire financial system.

Regarding the pre-COVID and the COVID-19 period differences, we find evidence suggesting that during the COVID-19 pandemic, spillovers to and from other institutions increased significantly. The differential effect of COVID-19 on systematic and idiosyncratic risk shows positive but statistically insignificant results. These results suggest that exposure to market risk, such as interest rate risk, foreign exchange rate risk, commodity price risk, and corporate and sovereign credit exposures, not only increases systemic vulnerabilities during normal periods but also have a similar adverse impact during periods of exogenous shocks to the financial system. Interestingly, individual banks' high abnormal returns are usually considered spillover contributors both to and from others, however, it acted as a buffer against such vulnerabilities during the current pandemic as shown by the statistically significant negative coefficient of the interaction term of COVID-19 and Jensen's Alpha.

This study contributes to the literature by providing a comparative analysis of conventional and Islamic banks in terms of their systemic risk implications before and during the COVID-19 pandemic. The empirical analysis focuses on the evolution of systemic risk during the pandemic in a sample of 10 countries with dual-banking systems. Overall, we found insignificant differences in exposure to systemic risk vulnerabilities of conventional and Islamic banks, irrespective of the differences in business models, during an exogenous shock.

The rest of the paper is organized as follows; Section 2 discusses policy responses to the pandemic in the sample countries, Section 3 provides literature on systemic risk for the sample countries, Section 4 presents the empirical methodology used in this study, Section 5 gives information about the data, presents results, and provides discussion, and Section 6 provides conclusions and comments.

In mid-March 2020, most central banks around the globe were in crisis management mode to control the economic consequences of the COVID-19 pandemic. During any crisis, including the COVID-19 pandemic, the sustainability of the financial services sector boils down to three main themes: liquidity, capital, and profitability. Central banks announced several measures for liquidity support to the banking sector, including (i) lowering reserve requirements, (ii) lowering regulatory capital buffers, (iii) bond/S˙ukuk buying programs, and (iv) availability of central bank credit lines (reverse repo). The first two measures help both conventional and Islamic banks however, in some countries, Islamic banks were deprived of the last two measures which are crucial for liquidity management support. For some jurisdictions, central banks do not have credit lines for Islamic banks either due to non-availability or legal impediments. This further adds to the challenges of liquidity management for Islamic banks especially in the wake of loan deferment programs in place for various jurisdictions that offer Islamic financial services and has systemically important Islamic banks. Furthermore, there are mounting fears that bank borrowers may not be able to service their debts after the end of moratoriums/loan deferment programs. The challenges of liquidity management could potentially affect the sustainability of Islamic banks. 6 Table 1 provides a summary of the regulatory and policy interventions introduced during the pandemic in the sample countries. Qatar introduced the most interventions while Kuwait, Malaysia, and Oman considered only a smaller number of policy actions. Bahrain, Bangladesh, Malaysia, UAE, and Pakistan implemented more targeted policy interventions. While only Bahrain, Bangladesh, Pakistan, and Qatar announced special measures for Islamic banks that highlight the missing attention of regulators, in more than half of the sample countries, to the special needs of Islamic banks. Table 1 highlights that central banks in the sample countries focused on liquidity support including guarantees and loan deferral programs. However, it seems that there was much less focus on long-term structural measures to stabilize the financial sector. IsDB (2020) highlights that in the GCC region, the financial sector in general and the banking sector in particular, suffered liquidity constraints as withdrawals increased considerably from deposits maintained by various governments or government-related institutions, while the liquidity lines from central banks filled up quickly. The report further notes that countries where the banking sector is already struggling with high non-performing loans, low profitability, and weak capital buffers will be the worst affected as the guarantees and other support measures wind down. Given this scenario, the likelihood of a rise in systemic risk is likely and there is a need to see how systemic risk is evolving in countries with dual-banking systems.

The GFC of 2007-09 highlighted the weaknesses of the regulations surrounding individual banks. Since then, systemic risk has become a central theme in macro-prudential policy-making and extensive research has been conducted both for developed and developing economies. This section provides a review of the systemic risk studies covering the countries included in the sample. 7 Literature shows that, in dual-banking systems, different regulatory policies have a varying impact on conventional and Islamic banks (Rizwan et al., 2017) . Due to the engagement of Islamic banks in asset-backed financing in risk-sharing modes, studies found mixed evidence on the effect of the GFC on Islamic banks relative to conventional banks. Ashraf et al. (2016) report a statistically insignificant effect of the GFC on the stability of Islamic banks and Kabir et al. (2015) found no significant difference in the effect of GFC on conventional and Islamic banks. Alqahtani and Mayes (2018) report that GFC did not negatively impact Islamic banks initially but once the financial shock reached the real economic level, Islamic banks suffered more than conventional banks highlighting their connection with the real economy. Furthermore, Islamic banks are generally better capitalized as compared to conventional banks (Beck et al., 2013) . Arguably, asset-backed business models and better capitalization may enable Islamic banks to cope better with the negative consequences of economic shocks.

Regarding systemic risk implications, Abedifar et al. (2017) provide a comparative analysis on the systemic resilience of fully conventional, fully Islamic, and conventional banks with Islamic windows in GCC member countries with dual-banking systems. The authors used Marginal Expected Shortfall (MES) (Acharya et al., 2017) , SRISK (Acharya et al., 2012; Brownlees and Engle, 2016) , and Delta Conditional Value-at-Risk (DCoVaR) (Adrian and Brunnermeier, 2009 ) on a sample for the period from 2005 to 2014. Their results show that conventional banks with Islamic windows are most prone to systemic events and, during the GFC, these banks were the most interconnected. Manap (2019) studies the systemic risk contribution of six banks in Malaysia (five conventional and one Islamic) using DCoVaR during the 2000-2017 period and concludes that the Islamic bank has the highest systemic risk contribution. However, this study includes only one Islamic bank and, therefore, the results cannot be generalized. Hashem and Abdeljawad (2018) compare the systemic risk of conventional and Islamic banks in Bangladesh from 2005 to 2014 using DCoVaR of 27 listed banks. Their findings suggest that conventional banks are more susceptible to systemic events with higher systemic risk spillover during the GFC. Overall, the existing literature suggests that conventional banks are comparably more prone to systemic risk and have higher systemic risk spillovers.

A gap in the literature exists regarding systemic vulnerabilities of dual-banking systems, and the systemic response of conventional and Islamic banks during exogenously-induced macroeconomic shock such as the COVID-19 pandemic. Such an analysis may have valuable insights for the policymakers of dual-banking systems.

In this section, we provide details for the estimation of systemic risk measures (country-and bank-level), the estimation of bankspecific factors that are included in the regression analysis, and the econometric modeling. Sources: IsDB staff compilation from the websites of central banks of our sample countries and the IMF. a FX fee refund.

For the country-level measure of systemic risk, we rely on CATFIN proposed by Allen et al. (2012) . CATFIN is defined as the average of three measures of value-at-risk (VaR): the generalized Pareto Distribution (VaR GPD ) (Pickands III et al., 1975) , the skewed generalized error distribution (VaR SGED ), and a non-parametric estimation (VaR NP ). CATFIN is calculated as follows:

Using forecasting evaluation of macroeconomic shocks, Giglio et al. (2012) show that CATFIN, based on the non-parametric component (VaR NP ), is a suitable measure of systemic risk among individual measures. Therefore, following Giglio et al. (2012) , we estimate CATFIN as:

where VaR NP is estimated using the cutoff point for the lower α percentile of the excess return. 8

For bank-level systemic risk measures, we use the connectedness measures proposed by Diebold and Yılmaz (2014) . These measures use the M-step variance decomposition, based on the generalized variance decomposition of Pesaran and Shin (1998) , of stock market returns. The ij-th component of the generalized variance decomposition (d M ij ) is given as:

In Eq. (3), σ jj is the j-th diagonal element of Σ and e i is a vector which is 0 except for its i-th element which is 1. Moreover, A m represents the matrix of coefficients of the moving average part for the m-th lagged shock vector and Σ is the covariance matrix of the error term in the vector autoregressive model. The M-step variance decomposition (d ij M ) is calculated as follows: (2014) construct bank-level measures, "spillover to others" (STO) and "spillover from others" (SFO), 9 as follows:

Hence, STO i measures the impact of shocks from bank i to all other banks while SFO i measures the effect on bank i from the shocks to other banks. In other words, STO i measures the impact of bank i on the system, i.e., the contribution of bank i to systemic risk, whereas SFO i represents the effect of systemic risk on bank i. 10

Market prices, due to their ability to reflect informational contents immediately, are inherently more forward-looking than accounting data and are considered as a leading indicator that systematically reflects all available information efficiently. From the systemic risk perspective, the securities issued by banks are interesting for two main reasons. One, market prices of debt and equity may increase banks' funding cost and, therefore, induce (direct) market discipline. Second, supervisors may benefit from market data to complement traditional accounting data for assessing bank fragility (indirect market discipline). Huang et al. (2009) suggest that due to easy access to market information at a very high frequency, relative to financial statement data, supervisors may utilize price signals as screening devices or inputs into early warning models geared at identifying systemically important banks. Similarly, Gropp et al. (2006) argue that asset market data should be valuable in assessing systemic risk as systematic risk ascends from inter-temporal decisions while capital market prices provide information about inter-temporal marginal rates of substitution and transformation. Bessler et al. (2015) argue that systematic risk measures the sensitivity of banks towards economic system-level risk factors, such as corporate credit risk, interest rate risk, foreign exchange risk, and sovereign risk, while idiosyncratic risk captures bank-level risks, such as, loan portfolio and non-interest income-related risks.

We use the capital asset pricing model (Black, 1972; Jensen et al., 1972; French, 1992, 2004) to determine the factors for the association of a bank's performance on its systemic risk contribution and fragility with the following specification for bank i:

In Eq. (7), R it and R it m are the monthly log-returns of bank i and the market index of the country of bank i in period t, respectively.

The estimated values of α i (Jensen's alpha) and β i measure the excess monthly log returns (abnormal returns) and relative riskiness (systematic risk) of bank i relative to the benchmark index, respectively. In terms of interpretation, α i > 0 implies that the stock of bank i has "abnormal returns", i.e., the returns are higher than market-based risk-adjusted returns, whereas β i > 1 (β i < 1) suggests that bank i is riskier (safer) than the market index. Aside from Eq. (7), we also estimate a three-factor model (Fama and French, 1993) to consider risk factors other than the market. The specification for the three-factor model of bank i is as follows:

The three-factor model presented in Eq. (8) includes the monthly growth rates of SMB and HML which are size (small minus big) and momentum (high minus low) factors 10 augmented in the standard capital asset pricing model.

In this paper, we also consider an international CAPM based on local and international sensitivities. The specification for bank i is as follows:

The international model further includes the monthly growth rate of the MSCI world index (WI). Eqs. (7)-(9) are estimated over a 9-month rolling window for each bank in the sample. Following Huang et al. (2021) , the idiosyncratic risk for bank i is estimated as the 9-month rolling standard deviation of the estimated residuals (ε it ) for each of the Eqs. (7)-(9). The idiosyncratic risk of bank i in period t is denoted as σ it . In the next step, we used the estimated values of Jensen's Alpha (α i ),

, and idiosyncratic risk (σ it ) in our systemic risk spillover analysis.

Aside from the bank-specific factors based on the market model, we also use the natural logarithm of total assets as a measure of a bank's size (Size) and growth of market capitalization to account for changes in the capitalization of banks (%ΔMarketCap) to see the systemic risk implications of market valuation.

Theoretically, systematic risk (beta) captures the sensitivity of a bank's performance towards systematic factors. However, literature such as Engle et al. (2015) , studies the nexus between systematic risk and macroeconomic factors using specific factors. To capture these aspects, we include three macroeconomic factors in our model, namely Industrial Production (IP it g ) to capture economic activity, Consumer Price Index to capture inflation (π it ), and Policy Rates (PR it ) to capture the monetary policy stance of regulatory authorities. All variables are standardized to have a zero mean and a standard deviation of one to take care of the size-driven bias in the systemic risk analysis (Varotto and Zhao, 2018) and to identify the relative importance of the coefficients (Siegel, 2016) . The standardized variables indicate the number of standard deviations the bank (for bank-level variables) or the country (for country-level variables) is away from its mean. The coefficients of these variables show the change in the dependent variable (in number of standard deviations) if the independent variable changes by one standard deviation from its mean. Therefore, the interpretation is in terms of deviations of the variable from its mean instead of its increments. 11

In this subsection, we provide the econometric methodology for ascertaining the determinants of the spillover measures. Our first regression equation is specified as:

In terms of the variables in Eq. (10), Y it denotes the dependent variable which we take as the spillover measures, i.e., Y it is either STO it or SFO it , the bank-specific factors α it , β it , and σ it are estimated using Eq. (7), 12 Country i are dummy variables for the sample countries, and ε it is the error term. The coefficients in Eq. (10) are as follows: γ 0 is the intercept term, λ's depict the effect of α it , β it , and σ it on Y it , and δ is a vector of coefficients that control for country-level fixed-effects.

The second regression equation is defined as:

10 These factors have been calculated for every sample country except for Qatar which has very few stocks available at Clarivate Thomson Reuters database. Therefore, we used emerging market SMB and HML factors, available at the Fama-French data repository, for Qatar. 11 All variables are winsorized at the 1st and 99th percentiles to remove outliers. 12 The hatted variables represent the estimated variables.

Compared to Eq. (10), Eq. (11) includes responsiveness of banks to SMB and HML. Note that α it , β it , β SMB it , β HML it , and σ it are estimated from Eq. (8).

The third regression equation includes each bank's responsiveness to changes in the world index, growth rate of market capitalization, and size. It also includes different intercept terms for conventional and Islamic banks for the periods before and during the COVID-19 pandemic. The estimated equation is:

In Eq. (12), Islamic i is a dummy variable that is 1 if the bank is Islamic and 0 otherwise whereas COVID − 19 t is a dummy variable that is 1 for the pandemic period (starting from 27th December 2019) and 0 otherwise. Therefore, γ 1 and γ 2 allow the intercept term to be different for Islamic banks and during the pandemic period, respectively. Moreover, Eq. (12) includes all bank-specific factors where

, β WI it , and σ it are estimated from Eq. (9). Note that Eq. (12) does not control macroeconomic factors that could play a role on the spillover measures. Therefore, Eq. (12) is extended as follows:

In Eq. (13), IP it g is the monthly growth rate of industrial production, π it is the monthly inflation rate calculated from CPI, PR it is the policy rate, and ϑ's show the effect of these macroeconomic factors on the spillover measures.

Since we want to explore how the effect of bank-specific factors varies across conventional and Islamic banks, we estimate the following equation:

In Eq. (14), the coefficients λ 1 , λ 2 , and λ 3 allow α it , β it , and σ it to differently impact spillover measures for Islamic banks. In addition to varying effects of conventional and Islamic banks in Eq. (14), we extend the estimation to account for the COVID-19 period as follows:

In Eq. (15), the coefficient of Islamic i ⋅ COVID − 19 t (γ 3 ) allows different intercept terms for conventional and Islamic banks during the pandemic whereas λ 's allow bank-specific factors of conventional and Islamic banks to affect the spillover measures differently before and during the COVID-19 period.

The varying coefficients of α it , β it , and σ it for the conventional and Islamic banks, before and during the pandemic period, in regression Eqs. (14) and (15) are summarized in Table 2 .

The regression specifications in Eqs. (10)-(15) use estimated coefficients, from Eqs. (7)- (9), as explanatory variables. Therefore, the entire estimation was bootstrapped with 1000 replications (stratified by countries and clustered over banks). 13

This study uses data from all listed banks 14 operating in the sample countries covering the period November 2015 to November 13 The country stratification maintains each country's representation in the replicated sample while clustering over banks guarantees that each bank included in the replicated sample has all observations for an appropriate estimation of Eqs. (7)-(9). 14 Subject to the data availability of daily stock prices on Refinitiv's DataStream. 2020. The sample includes 147 banks 15 from 10 countries, of which 33 are Islamic and 114 are conventional. Table 3 shows the list of sample countries, the market index for each country, and country-wise sample distribution of the number of conventional and Islamic banks. Fig. 1 shows the evolution of systemic risk for the sample countries during the study period. The gray-shaded area represents the period of COVID-19. All countries exhibit a sharp peak during the pandemic which is followed by a recovery. Except for Oman, all countries show their highest peaks during the pandemic. Although Oman shows elevated levels of systemic risk during the COVID-19 period, its highest peak occurs during the first quarter of 2016 which corresponds with Oman's stock market crash when the market dropped to its 7-year lowest value. 16 Table 4 reports the average values of CATFIN during 2019 and 2020 along with the K-Wallis test statistic. Results show a significant increase in CATFIN during 2020, as compared to 2019, except for Bahrain which has a statistically insignificant increase.

In summary, all countries have experienced a rapid increase in their systemic risk during the pandemic which was followed by a quick recovery. This rapid increase is in line with the results from the most affected COVID-19 countries reported by Rizwan et al. (2020) . It is plausible that regulatory actions played a role in the quick recovery experienced during the second half of 2020.

Using the methodology explained in Section 4.2, we identified the Systemically Important Financial Institutions (SIFI) based on their ability to affect others (STO) and their vulnerability of being affected by others (SFO) during the pandemic. Such an analysis enables regulators to devise appropriate policies for managing systemic risk. Figs. 2 and 3 provide heat maps based on STO and SFO, respectively. 17 In Fig. 2 , year-wise heat maps are shown for each country. The first letter of the bank code shows if the bank is conventional (C) or Islamic (I). Heat maps show that, for almost all sample countries, Islamic banks show lower STO (blue colors) while conventional banks are major contributors of systemic risk. Only relative exceptions are Jordan, Kuwait, and Qatar where Islamic banks show elevated STO. Fig. 3 provides heat maps based on SFO which shows that Bangladesh, Kuwait, Malaysia, Pakistan, and Qatar have high levels of interconnectedness in the banking sector. Overall, based on the STO and SFO heat maps, Islamic banks are recipients of risk from the system while conventional banks are net contributors. This observation is in line with Mensi et al. (2019) who reports similar results from dual-banking systems in the GCC region. Table 5 reports the descriptive statistics of the variables used in this study for the sample as a whole, as well as for conventional and Islamic banks. The last column reports the K-Wallis statistics for the test of differences in the two sub-samples (conventional and Islamic). On average, conventional banks show significantly higher mean values of STO and SFO as compared to Islamic banks. Both Table 2 The coefficients of α it , β it , and σ it for conventional and Islamic banks, before and during the pandemic period, in the regression specifications of Eqs. (14) and (15).

Bank type COVID-19 Eq. (14) Eq. (15) *αit Conventional 0 conventional and Islamic banks show a negative mean value of Jensen's Alpha without any significant difference. Comparatively, significant differences exist between conventional and Islamic banks for all other variables. Conventional banks show higher systematic risk and idiosyncratic risk, while Islamic banks show positive average growth in market capitalization as compared to negative average growth for conventional banks during the sample period. Overall, the descriptive statistics show that conventional banks are systemically more important than Islamic banks and also have higher systematic and idiosyncratic risks without any significant difference in abnormal returns performance. Table 6 reports the correlation matrix among variables used in this study. A high correlation exists between STO and SFO which suggests that banks with high spillover to the financial system are also more prone to spillover from the system. There is a positive correlation between Jensen's alpha and systematic risk, while idiosyncratic risk and market capitalization have negative correlations with spillover measures. Overall, the correlation table suggests a higher systemic risk of banks with higher abnormal returns and systematic risk exposure while lower for larger banks with higher idiosyncratic exposure.

To identify the appropriate estimation methodology for Eqs. (10)-(15), we conducted the Modified Wald statistic for panel-wise heteroskedasticity and the Wooldridge test for serial correlation in the errors of a linear panel data model (Wooldridge, 2002; Drukker, 2003; Greene, 2003) . The test statistics are presented in Table 7 . The tests indicate the presence of heteroskedasticity for Eqs. (12)-(15) and serial correlation in all regression specifications. To account for the heteroskedasticity and serial correlation, we fit a linear panel data model using feasible generalized least squares (FGLS) estimation. 18 The serial correlation is controlled by using a bank-specific first-order autoregressive model, i.e., AR(1) model is fit on the error term for each bank. 19 Table 8 reports the regression results with STO as the dependent variable. Model (1) reports the results of Jensen's alpha, systematic risk (beta), and idiosyncratic risk which are estimated using the market model. The coefficient of abnormal returns (Jensen's alpha) is statistically insignificant suggesting that the abnormal market return performance of the bank has no significant role in the systemic risk spillover to other financial institutions. The coefficient of beta, which is a measure of a bank's exposure to systematic factors, is positive and statistically significant suggesting that banks with higher exposure to market fluctuations have higher spillovers to other financial institutions.

The coefficient of beta suggests that one SD increase in systematic risk is associated with a 0.0637 SD increase in STO. This is in line with the literature that discusses the overlapping portfolio problem (Poledna et al., 2021) suggesting that financial institutions may have overlapping portfolios and, therefore, higher exposure to systematic factors may result in the systemic vulnerability of interconnected financial institutions.

Idiosyncratic risk, which captures a bank's idiosyncratic risks related to loan portfolio and non-interest earning activities, has a negative and statistically significant coefficient. The coefficient shows that one SD increase in idiosyncratic risk is associated with a − 0.0291 SD change in STO. This is in line with the literature that suggests that managing the idiosyncratic risk of a financial institution does not ensure the stability of the entire financial system (Suh, 2019) . These findings are in line with the literature that investigated the causes and effects of the GFC and concluded that micro-prudential regulatory frameworks failed to address systemic vulnerabilities (Claessens, 2014 ) and, therefore, as a response to the GFC, macro-prudential regulations were introduced.

Arguably, Jensen's alpha, beta, and idiosyncratic risks are highly dependent on the choice of the asset pricing model. Therefore, we extended the market model and include SMB and HML factors (Fama and French, 1993 ) (Model (2)) and cross-border spillover (world index), and bank-specific variables (Model (3)). Results are given in Model (2) and (3) of Table 8 respectively. Results of systematic and idiosyncratic risks from Model (1) remained consistent, therefore, the inferences drawn above remain valid. Among SMB and HML factors, SMB shows a statistically significant negative coefficient with a magnitude that suggests that one SD increase in the beta of SMB is associated with a 0.0187 SD decrease in STO. This suggests that smaller banks that are required to pay a higher premium under a CAPM framework, contribute less to STO. The beta of HML shows a positive and statistically significant coefficient. This shows that Table 3 List of sample countries, market index for each country, and country-wise sample distribution of the number of conventional and Islamic banks. (10) and (11), we specify a homoskedastic error structure. 19 In the estimation, the error structure is heteroskedastic but uncorrelated across banks because we have unbalanced panels.

financial institutions exhibiting higher momentum in their stock prices contribute more to systemic risk. Among bank-specific factors, growth in market capitalization shows a significantly negative coefficient suggesting that a decline in the market value of a bank may increase its contribution to systemic risk. This may suggest that banks with declining market capitalization may increase their risk appetite to earn higher returns and boost their market capitalization. Interestingly, results show that the size of the financial institution has a negative association with systemic risk. At first, this result may seem surprising as regulatory intuition suggests that larger banks have higher systemic importance. However, as explained in the methodology section, these results are based on standardized variables. Therefore, this result implies that when a bank's balance sheet shrinks (grows) by one standard deviation from its long-term mean, it asserts more (less) vulnerability to and from the system. This is in line with the literature that shows a negative association of asset growth with systemic risk (Varotto and Zhao, 2018) . This finding may indicate that regulatory efforts to monitor large banks all over the world to control systemic risk may have their merits, but a sharp decline in a bank's balance sheet, irrespective of the size, has severe systemic implications. We also observe a statistically insignificant coefficient for the dummy of Islamic banks suggesting that, on average, Islamic banks have similar systemic spillovers as conventional banks. The dummy of COVID-19 shows a significant and positive coefficient showing that banks have a significantly higher contribution to systemic risk during the pandemic. This is in line with Rizwan et al. (2020) who show significantly higher systemic risks during the COVID-19 pandemic in a sample of eight most affected countries.

Model (4) reports the estimation results after accounting for the impact of macroeconomic factors and shows a significant and positive (negative) coefficient of industrial production (policy rates). Results of industrial production suggest the procyclicality of financial institutions. During periods of high economic activity financial institutions tend to underestimate the risk implications of their loan portfolios which may lead to higher systemic risk. On the other hand, the results of policy rates testify to the counter-cyclical benefits of a monetary policy stance which is in line with the literature suggesting that persistent loosening of the monetary policy may face a trade-off with financial stability (Kabundi and Simone, 2020) . Overall, from Model (1) to (4), our inferences regarding bankspecific factors remained robust after using different asset pricing models and controlling for macroeconomic factors.

One of the motivations for studying countries with a dual-banking system is to ascertain the varying effects of bank-specific factors on the spillover measures across conventional and Islamic banks. To do this, we re-estimate the model after incorporating interactive dummies of Islamic banks and COVID-19 with Jensen's alpha, systemic risk, and idiosyncratic risk. For better interpretation, Table 2 explains the interactive variables. Results of the interactive analysis are provided as Models (5) and (6) in Table 8 .

The interactive term of Islamic with Jensen's alpha shows a significant and negative coefficient, and in terms of magnitude, the interactive term has a higher coefficient than conventional banks' alpha. This suggests that Islamic banks, while offering higher abnormal returns, have a lower contribution to the STO as compared to conventional banks. This can be attributed to the risk-sharing model of Islamic banks that can potentially pass a negative shock on the asset side (e.g., abnormal loan loss) to investment depositors. The risk-sharing arrangements on the deposit side provide another layer of protection to the bank, in addition to its book capital and, thus, making it less vulnerable to systemic shocks (Čihák and Hesse, 2010) . Regarding the coefficients of systematic risk and idiosyncratic risk for Islamic banks, both are insignificant suggesting that Islamic banks are not significantly different than conventional banks in terms of their systematic and idiosyncratic exposure and systemic risk.

In terms of the COVID-19 pandemic, we estimate the final model that allows for bank-specific factors to affect STO differently before and during COVID-19 as well as across conventional and Islamic banks. The regression results are provided in Model (6) of Table 8 . Results show that in the pre-COVID-19 period, the impact of Jensen's alpha on STO is the same as shown by the intermediate model. However, during the pandemic, results show a negative association between Jensen's alpha and STO and Islamic banks do not behave any differently than conventional banks. This suggests that during exogenous shocks to the financial system, such as the COVID-19 period, the abnormal return performance of individual banks plays a stabilizing role. Regarding systematic risk and idiosyncratic risk, results do not show any significant differences between conventional and Islamic banks or before and during the COVID-19 period. Table 9 reports the regression results with SFO as the dependent variable. In Model (1), the coefficient of Jensen's alpha shows a positive, albeit, statistically insignificant coefficient with SFO. This suggests that the abnormal return performance of banks in the market does not play any significant role in systemic vulnerability.

Beta (systematic risk) shows a statistically significant positive coefficient which shows that a one SD increase in Beta is associated with 0.0611 SD increase in SFO. This suggests that banks with higher exposure to market fluctuations are more vulnerable to spillover from other financial institutions.

Idiosyncratic risk has a negative and statistically significant coefficient which shows that a one SD increase in idiosyncratic risk is associated with a 0.0485 SD decrease in systemic vulnerability. However, idiosyncratic risk loses its statistical significance when bankspecific factors are included in the model.

In line with the results of STO, results of SMB (HML) factors (in Model 3) show significantly negative (positive) coefficients suggesting that smaller banks while paying higher return premium have less vulnerability to systemic factors while banks with higher The data covers the period November 2015 to November 2020. K-Wallis statistics report if conventional and Islamic banks are statistically different from one another. Alpha, IdioRisk and Beta are given with numbers in brackets to show the models these belong to in the regression results. ***, **, and * denote significance at the 1%, 5%, and 10% significance level, respectively. IdioRisk (3-7) 0.01 − 0.01 0.01 0.11* 0.08* 0.14* 0.87* 0.94* 1 (10) Beta-SMB (2) − 0.02 − 0.14* − 0.13* 0.01 − 0.15* − 0.12* 0.19* 0.14* 0.14* 1 (11) Beta-SMB ( Results from Model (5) regarding conventional and Islamic banks show that there are no statistically significant differences in the coefficients of Jensen's alpha, beta, and idiosyncratic risk across conventional and Islamic banks. This suggests similar behavior in terms of systemic vulnerability.

In Model (6), we observe that the coefficient of Jensen's alpha is positive and significant during the pre-COVID-19 period for both conventional and Islamic banks. However, during the COVID-19 period, the association turns negative for both conventional and Islamic banks. There are no significant differences in terms of systematic risk and idiosyncratic risk for conventional and Islamic banks before or during the COVID-19 period.

Overall, results show that abnormal returns performance elevates systemic risk during normal periods and plays a stabilizing role during stress periods such as the COVID-19 period. The exposure of banks to market risk has a severe negative impact on systemic stability which suggests that regulators should devise regulations that limit market exposure by the implementation of macroprudential policy instruments such as limits on foreign currency exposure and foreign exchange counter-cyclical reserves (Cerutti et al., 2017) . Results show that the idiosyncratic risk of individual banks does not contribute as much when compared with the contribution of abnormal returns or systematic risk for systemic vulnerability. However, in the wake of exogenous shocks, such as the COVID-19 period, idiosyncratic risks can become pertinent. In terms of comparative analysis of conventional and Islamic banks, we observe that conventional and Islamic banks behave similarly except for Jensen's alpha in STO where Islamic banks, while earning abnormal returns, have significantly lower systemic risk spillover to others. In terms of systematic risk exposure and idiosyncratic risk exposure, Islamic banks do not show any significant differences as compared to conventional banks during the COVID-19 period. This might be due to the exogenous nature of the COVID-19 shock.

As a robustness check, we use the panel-corrected standard error (PCSE) estimation where coefficients are estimated using ordinary least squares (OLS) (Greene, 2003) . For the PCSE estimation, alongside heteroskedasticity and bank-specific serial correlation, we also allow contemporaneous correlation across banks in the error structure. 20 Tables 10 and 11 provide the estimation results for the regression specifications with STO and SFO as the dependent variables, respectively. Generally, inferences remain the same and the results are in line with the already reported FGLS estimation results.

As found in the overall results of STO and SFO, Jensen's alpha has significant systemic risk implications. Jensen's alpha represents the return performance that has not been explained by the risk factors considered in the capital asset pricing model. What then is the source of this abnormal return performance? In other words, what is the channel through which this abnormal return performance is effecting systemic risk? To dig deeper, we re-estimated our model with an interaction of Jensen's alpha and idiosyncratic risk. By doing so, we can know if abnormal return performance is arising from individual bank-specific risks. We add an interaction term for Jensen's alpha and idiosyncratic risk in Eq. (15). That is, we estimate the following equation: 20 We assume independent errors, across the panels, for Eqs. (10) and (11). Results given as Model (1) to Model (6) are for the regression Eqs. (10)-(15), respectively, with STO as the dependent variable. Sample period is from November 2015 to November 2020. ***, **, and * denote significance at the 1%, 5%, and 10% significance level, respectively. Results given as Model (1) to Model (6) are for the regression Eqs. (10)-(15), respectively, with SFO as the dependent variable. Sample period is from November 2015 to November 2020. ***, **, and * denote significance at the 1%, 5%, and 10% significance level, respectively. Results given as Model (1) to Model (6) are for the regression Eqs. (10)-(15), respectively, with STO as the dependent variable. Sample period is from November 2015 to November 2020. ***, **, and * denote significance at the 1%, 5%, and 10% significance level, respectively. Results given as Model (1) to Model (6) are for the regression Eqs. (10)-(15), respectively, with SFO as the dependent variable. Sample period is from November 2015 to November 2020. ***, **, and * denote significance at the 1%, 5%, and 10% significance level, respectively. Results correspond to Eq. (16). Sample period is from November 2015 to November 2020. ***, **, and * denote significance at the 1%, 5%, and 10% significance level, respectively.

Y it = γ 0 + γ 1 Islamic i + γ 2 COVID − 19 t + γ 3 Islamic i · COVID − 19 t +λ 1αit + λ 2βit + λ 3σ it + Islamic i ·(λ 1αit +λ 2βit +λ 3σ it ) +COVID − 19 t ·(λ 4αit +λ 5βit +λ 6σ it ) +Islamic i · COVID − 19 t ·(λ 7αit +λ 8βit +λ 9σ it ) +λ 4β SMB it + λ 5β HML it + λ 6β WI it + λ 7 Size it +λ 8 %ΔMarketCap it + λ 9αit ·σ it +ϑ 1 IP g it + ϑ 2 π it + ϑ 3 PR it + Country i δ + ε it (16) In Eq. (16), the coefficient λ 9 allows the effect of α it , on Y it , to be different for different values of the idiosyncratic risk (σ it ). For Eq. (16), the statistics for the Modified Wald and Wooldridge tests for STO are 966.64 and 298.39, and for SFO are 1500.37 and 41.52, respectively. All of these statistics are significant at the 1% level. Therefore, heteroskedasticity and serial correlation are controlled as discussed earlier.

Results are reported in Table 12 . Results show an insignificant interaction term of Jensen's alpha and idiosyncratic risk suggesting that abnormal return performance, which is considered to be an anomaly in the standard capital asset pricing model, can not be explained by idiosyncratic risk. As the results of the interaction term turned out to be insignificant, there is a need to do further analysis using the financial statement data, once it is available for a reasonable duration. We believe this might be an interesting future research avenue.

Since the onset of the COVID-19 pandemic, the banking sector has witnessed a decline in efficiency (Zheng and Zhang, 2021) , and a deterioration in loan portfolio quality (Ratnovski et al., 2020) . Simultaneously, access to capital is constrained, especially from capital markets, due to credit rating downgrades 21 and an increase in insolvency risk in the corporate sector, except for limited high-quality issuers (Mirza et al., 2020) . It is feared that "vulnerabilities in credit markets, emerging countries, and banks could even cause a new financial crisis". 22 COVID-19 has slowed the global economy. Consequently, financial institutions are facing issues with liquidity, loan performance, and inter-mediation revenues. Interconnectedness among financial institutions can spread vulnerability to the network of institutions resulting in an overall heating-up of the financial system. Islamic banks have shown resilience during the GFC owing to the risk-sharing nature of their business model. One of the major differences between the COVID-19 pandemic crisis and the GFC is that COVID-19 is an exogenous shock whereby the overall economy stands still due to extended lock-downs and the adoption of various social distancing measures. Due to the different nature of the shocks, it is expected that Islamic banks face similar challenges to their sustainability as conventional banks. This paper empirically investigates the determinants of individual banks' spillovers in the financial networks, through their interconnectedness in a dual-banking system before and during the pandemic. Systemic risk contributions are estimated using the connectedness measures "spillover to others" and "spillover from others".

During the COVID-19 pandemic, both spillover measures show a significant increase suggesting the heating-up of the financial systems during this exogenous shock. Overall, systematic risk, which shows banks' exposures towards market risk factors, increased systemic vulnerability. However, idiosyncratic risk, that captures institutional-level risk, shows a negative association with systemic risk. This suggests that market risks, such as interest rate risk, foreign exchange rate risk, commodity price risk, and corporate and sovereign credit exposures, not only increase systemic vulnerabilities during normal periods but have even more devastating outcomes during periods of exogenous shocks to the financial system.

For individual banks, high abnormal return performance may increase spillover to and from others. However, during the current pandemic, it acted as a buffer against such vulnerabilities. Comparative analysis of conventional and Islamic banks reveals that Islamic banks show abnormal return performance, as compared to their conventional counterparts, with lower systemic risk. But during the current pandemic, systematic risk and idiosyncratic risk exposures of Islamic banks have similar systemic costs as conventional banks.

The close alignment of Islamic banks' systemic vulnerability with exogenous shock suggests that the business model of Islamic banks links their performance with the real economy. However, the limited attention of regulators in providing support measures to Islamic banks during the COVID-19 pandemic may exacerbate their systemic vulnerability.

There is a need for regulators and lawmakers to establish the necessary support mechanisms needed to help Islamic banks weather the pandemic in the same manner as their conventional counterparts. Furthermore, the negative impact of excessive systematic risk has a severe negative impact on systemic stability which suggests that regulators should devise regulations that limit market exposure of banks. Implementing macro-prudential policy instruments such as limits on foreign currency exposures and foreign exchange countercyclical reserves may prove to be helpful in this regard. Future research can be conducted on the role of macro-prudential regulations in dual-banking systems during the COVID-19 pandemic with a comparative analysis between conventional and Islamic banks. 

Heterogeneous market structure and systemic risk: evidence from dual banking systems

Capital shortfall: a new approach to ranking and regulating systemic risks

Measuring systemic risk

Covar: a method for macroprudential regulation

COVID-19 Worsens Pre-existing Financial Vulnerabilities

Does systemic risk in the financial sector predict future economic downturns?

COVID-19, aftermath, impacts, and hospitality firms: an international perspective

Financial stability of islamic banking and the global financial crisis: evidence from the gulf cooperation council

Political institutions and bank risk-taking behavior

A net stable funding ratio for islamic banks and its impact on financial stability: an international investigation

A review of micro-based systemic risk research from multiple perspectives

Islamic vs. conventional banking: business model, efficiency and stability

Systemic Risk

Time-varying systematic and idiosyncratic risk exposures of US bank holding companies

A survey of systemic risk analytics

Capital market equilibrium with restricted borrowing

Systemic risk and the COVID challenge in the european banking sector

Impacts of COVID-19 on agricultural production and food systems in late transforming southeast asia: the case of myanmar

SRISK: a conditional capital shortfall measure of systemic risk

The use and effectiveness of macroprudential policies: new evidence

Islamic banks and financial stability: an empirical analysis

The Regulatory Responses to the Global Financial Crisis: Some Uncomfortable Questions

Pandemic-related financial market volatility spillovers: evidence from the chinese COVID-19 epicentre

On the network topology of variance decompositions: measuring the connectedness of financial firms

Testing for serial correlation in linear panel-data models

Systemic risk in europe

The cross-section of expected stock returns

Common risk factors in the returns on stocks and bonds

The capital asset pricing model: theory and evidence

Systemic risk and the macroeconomy: an empirical evaluation. SSRN Electron

Econometric Analysis. Pearson Education India

Equity and bond market signals as leading indicators of bank fragility

The effects of the global crisis on islamic and conventional banks: a comparative study

Chapter 2 islamic banks' resilience to systemic risks: myth or reality-evidence from bangladesh

A framework for assessing the systemic risk of major financial institutions

Brand equity and the covid-19 stock market crash: evidence from us listed firms

Estimating and projecting air passenger traffic during the COVID-19 coronavirus outbreak and its socio-economic impact

Islamic Development Bank. Islamic Financial Services Industry Stability Report

Impact of COVID-19 on GDP of major economies: application of the artificial neural network forecaster

The capital asset pricing model: some empirical tests

Comparative credit risk in islamic and conventional bank

Monetary policy and systemic risk-taking in the euro area banking sector

A study of systemic risk of global stock markets under COVID-19 based on complex financial networks

Measuring systemic risk in dual banking system: the case of Malaysia

Risk spillovers and hedging effectiveness between major commodities, and islamic and conventional GCC banks

Impact of covid-19 on corporate solvency and possible policy responses in the EU

Convergence in bank performance for commercial and islamic banks during and after the global financial crisis

Generalized impulse response analysis in linear multivariate models

Statistical inference using extreme order statistics

Quantification of systemic risk from overlapping portfolios in the financial system

Covid-19 and non-performing loans: lessons from past crises

Does a one-size-fits-all approach to financial regulations alleviate default risk? The case of dual banking systems

Systemic risk: the impact of COVID-19

Revisiting oil-stock nexus during COVID-19 pandemic: some preliminary results

Impact of COVID-19 pandemic on micro, small, and medium-sized enterprises operating in Pakistan

Multiple regression

An analysis of the literature on systemic financial risk: a survey

Impact of COVID-19 on the travel and tourism industry

An early assessment of the impact of COVID-19 on air transport: just another crisis or the end of aviation as we know it?

Asset correlation and bank capital regulation: a macroprudential perspective

How did COVID-19 impact air transportation? a first peek through the lens of complex networks

Econometric Analysis of Cross Section and Panel Data

The impact of COVID-19 on the efficiency of microfinance institutions

We are thankful to Dr. Barbara L' Huillier for her helpful comments and suggestions that enabled us to improve the quality of this paper considerably. The views expressed in this paper are those of the authors and do not necessarily reflect the views of the Islamic Development Bank Institute or the Islamic Development Bank Group. All errors are the responsibility of the authors.