A Survival Analysis of Islamic and Conventional Banks

Transcription

1. A Survival Analysis of Islamic and Conventional Banks Vasileios Pappas • Steven Ongena • Marwan Izzeldin • Ana-Maria Fuertes September 2015 Abstract Are Islamic banks inherently more stable than conventional banks? We address this question by applying a survival analysis based on the Cox proportional hazard model to a comprehensive sample of 421 banks in 20 Middle and Far Eastern countries from 1995 to 2010. By comparing the failure risk for both bank types, we find that Islamic banks have a significantly lower risk of failure than that of their conventional peers. This lower risk is based both unconditionally and conditionally on bank-specific (microeconomic) variables as well as macroeconomic and market structure variables. Our findings indicate that the design and implementation of early warning systems for bank failure should recognize the distinct risk profiles of the two bank types. Keywords Islamic banks • Failure risk • Survival analysis • Financial intermediation JEL Classification C41 • G21 • G20 Vasileios Pappas (Corresponding author) University of Bath, Bath, BA2 7AY, UK Tel: + 44 744 61 65 114; E-mail: v.pappas@bath.ac.uk Steven Ongena University of Zurich, SFI and CEPR, Plattenstrasse 32, CH-8032 Zurich, Switzerland Marwan Izzeldin Lancaster University Management School, Lancaster, LA1 4YW, UK Ana-Maria Fuertes Cass Business School, City University London, 106 Bunhill Row, London EC1Y 8TZ, UK
2. 1 Introduction The management and operation of Western commercial banks has been the subject of much debate and research . A sound banking system that maintains credit flows to the private sector has become a primary objective for both policymakers and bank regulators around the globe (Levine and Zervos 1998; Reinhart and Rogoff 2009). However, mismanagement, poor performance, and bank failures were prominent during the recent financial crisis. One sector of the global financial system that has attracted attention because of its resilience during the recent crisis is Islamic banking. With total bank assets around $1.8 trillion and a strong annual growth rate, Islamic banking is quickly becoming an important part of the global financial system (Ernst and Young 2015; IMF 2015). Some of the factors that have contributed to the growth of the Islamic banking industry have been the increased demand for products that are compliant with the Islamic Law (Shariah), persistent oil revenues that fuel the core Islamic finance markets, safer and more stable returns on investments, and the operational transformation that has led to enhanced and differentiated products. Islamic banking is built on a set of principles emanating from the Shariah. These principles differentiate it substantially from conventional banking in terms of financial products and objectives. In particular, Islamic banking is characterized by the prohibition of interest, complex derivatives, and short-selling. Shariah also bars investments that involve dealings in alcohol, gambling, and tobacco. Although key observable features distinguish the Islamic banking sector from the rest of the financial system, skeptics still argue that these differences have no practical significance; see, e.g. Khan (2010). To shed light on this debate, recent studies have compared Islamic and conventional banks on the basis of business models, efficiency, asset quality, credit risk, and stability.1 To mention a few, Boumedeine and Caby (2009) show that Islamic banks have low volatility in their returns and greater resilience to the impact of crises. Čihák and Hesse (2010) find that the smaller Islamic banks tend to be financially stronger than commercial banks overall. Yet the larger Islamic banks are not as financially stable, which may be due to the challenges in managing credit risk. Beck et al. (2013) conclude that Islamic banks were better capitalized and had higher asset quality that made them relatively less vulnerable to the recent financial crisis. Similar findings are reported in the studies of Moazzam and Zaheer (2013) and Van Wijnbergen and Zaheer (2013). However, Abedifar et al. (2013) compare the relation between risk and stability for both bank types, including “mixed” banks with both Islamic and 1 Although most studies tend to look at efficiency and stability in isolation, there are a few that examine the link between these concepts (see, e.g., Koutsomanoli-Filippaki and Mamatzakis (2009) and Koetter and Porath (2007) for applications using conventional banks, and Saeed and Izzeldin (2014) for a comparative study between Islamic and conventional banks). However, the exact link between efficiency and stability is not clear cut. For the Islamic versus conventional bank exposition, Kuran (2004) suggests that higher efficiency leads to higher stability, while Beck et al. (2013) finds the inefficient Islamic banks are more stable. Saeed and Izzeldin (2014) offer a nice exposition of the theories linking efficiency and stability in conventional banking. We opt not to address the issue of efficiency and stability in this paper due to data limitations arising from the necessity to use listed banks; we leave it instead for future research. 2
3. conventional “windows.” While they uncover no significant differences with respect to insolvency risk, their findings regarding credit risk indeed depend on the proxy that is used. Islamic and conventional banks differ conceptually in many aspects regarding their operations business model which, in turn, may lead to different survival rates. As regards their operations, the “no money for money” principle of Islamic banking implies that risk-sharing practices are embedded on both sides of the bank’s balance sheet. Islamic banks’ depositors are treated as investment account holders or preferred stockholders with a residual claim to profits but without explicit capital protection; hence, they share the risks of the banks’ investments. On the one hand, this is likely to exacerbate the bank's withdrawal risk (Ebrahim, 1999). On the other hand, religious beliefs may instil a certain loyalty in depositors which would allow banks to pass on realized losses in bad times, thereby achieving some pro-cyclical protection. Islamic banks use two types of funding sources: equity/participation type and fee-based services.2 The equity type is mainly used by large banks due to the relatively high costs involved (e.g., execution and Shariah compliance screening). By contrast, fee-based services offer a more certain source of income without exposure to adverse information and moral hazard issues. A common theoretical aspect of all Islamic banking financial products is that they are asset-backed which should enhance stability during market distress. Debt contracts are precluded (unless these are backed by tangible assets such as real estate or commodities, that is, Islamic bonds) and thus Islamic banks face important restrictions on how to obtain liquidity. Furthermore, the tradability of Islamic bonds is curtailed by underdeveloped secondary markets in the majority of the countries in question which imposes additional constraints. But Islamic banks are subject to the same market conditions as conventional banks including competitive pressure and having a similar need for protection against deposit withdrawal. Against these risks, Islamic banks withhold more liquidity, maintain profitequalisation reserves (profits in good periods as buffer to reduce volatility of payout in bad periods) and protect depositors/investment accountholders by shifting losses to shareholders (displaced commercial risk).3 The objectives of Islamic banks do not neatly square with the conventional profit maximization (cost minimisation) dogma. Islamic banks are more likely to be proponents of ethical banking with emphasis on charitable actions and interpersonal trust. Such practices are deeply rooted within the 2 Mudarabah and Musharakah are commonly used equity/participation type of contracts. In Mudarabah an investor (usually an Islamic bank) and an entrepreneur (individual or institutional) enter a joint venture where the bank provides the necessary funds and the entrepreneur provides the know-how. Fee-based services include the widely used contracts of Murabahah and Ijarah. Murabahah is in essence a cost-plus-profit sale. The bank arranges to sell a good to a customer at a premium which incorporates risks, costs and a profit margin. Ijarah is a lease contract where the bank leases an asset to an investor (or consumer) and the latter pays fees for utilising the asset. 3 Islamic banks operating alongside conventional banks are subject to displaced commercial risk which is the risk arising from managing assets on behalf of investment accountholders that is effectively transferred to the Islamic banks’ own capital. This risk unfolds as the bank may forgo its profit share on such investment when it considers this essential due to the commercial pressure in order to increase the rate of return payable to investment accountholders. 3
4. Islamic banking culture with a dedicated Shariah board overlooking the conformity of products and practices to the Islamic law . The role of Islamic banks as business partners in financing operations, in principle, ought to mitigate adverse selection and moral hazard. As such, suppliers of funds may be induced to greater diligence while borrowers’ opportunistic behaviour to avoid repayment would be discouraged in fear of social disapproval.4 Lastly, Islamic banks predominantly focus on large scale financing of infrastructure and real estate projects given that contracts are tailor-made and need to undergo screening for compliance which increases complexity and costs. However, country variations do exist in Islamic banks with regards to clientele profile, operations and availability of financial products. Following the differences outlined above, we anticipate the risk of failure to differ substantially between the two bank types. Failure risk is multifaceted; it involves credit risk, deposit withdrawal risk, insolvency risk, liquidity risk, and operational risk. For each type of risk, reasonable priors and theoretical arguments imply that Islamic banks are either less or more likely to experience failure than conventional banks. Given the differences in the business model of Islamic and conventional banks, we expect that distinct levels of exposure to all of these risks types might eventually lead to different failure risk profiles. But whether Islamic banks survive longer than conventional banks is an important yet still open empirical question. This paper aims to contribute to a rapidly growing empirical literature by singularly investigating whether differences in failure risk exist between the Islamic banks and the conventional banks. For this purpose, we use the survival analysis to compare the hazard rates across the two bank types and to assess their determinants. While recent papers address the issues that pertain to the credit and insolvency risks of Islamic banks, none estimates their hazard rates directly by using the survival analysis. The survival models circumvent the need for proxies of the failure risk of banks that might lead to a distorted comparison (Abedifar et al. 2013). Indeed, one of the main advantages of the survival analysis in the present context is that it uses the actual time-to-failure as the main observable variable. The survival functions that identify the probability of survival beyond a certain number of years can also help to identify the determinants of the differential failure risk profiles associated with the two bank groups. We find that, both unconditionally and conditionally on bank-specific (microeconomic) variables as well as macroeconomic and market structure variables, the Islamic banks exhibit on average lower hazard rates than the conventional banks. In addition, the two bank types differ in their failure-risk sensitivities to several of the covariates that, in turn, confirm that their failure risk profiles are distinct. The z-score measure, widely-used as bank stability proxy, cannot differentiate between the two bank types in our sample. Bank characteristics explain at least one third of the total variation in the hazard 4 Relatedly, Ostergaard et al. (2015) find that banks in communities with high social capital, such as interpersonal trust, civic engagement and charitable work are more likely to survive compared to banks driven by profit-maximization motives. 4
5. rates , while the inclusion of macroeconomic and market structure variables adds another ten percentage points to the survival models' explanatory power. Moreover, the country affiliation has a significant impact on the failure risk for the conventional banks but not for the Islamic banks. This impact suggests that only the conventional banks are locally interconnected from the viewpoint of failure risk. The findings also indicate that the conceptual differences in the business model of the two bank types are ultimately manifested in their distinct failure risk profiles. Among the bank-level covariates for which the sensitivities of the hazard rates differ across the two bank types, higher leverage and higher margins imply that Islamic banks are less likely to fail, while the opposite effect applies to conventional banks; this finding vividly demonstrates how their modi operandi differ. Among the key macroeconomic factors, high inflation harms the Islamic banks the most, possibly because of their greater reliance on cash reserves and their widespread use of commodities as collateral. The greater banking sector concentration adversely affects the survival propensity of Islamic banks, while it reduces the failure risk of conventional banks; this may relate to the fact that in many markets, the Islamic banks are the new entrants while the conventional banks are the incumbents. Lastly, we show that the survival models that in reduced-form incorporate distinctive features of the Islamic banking system yield more reliable predictions of the risk of Islamic bank failure than general (one-size-fits-all) survival models. This paper aims to investigate whether differences in failure risk exist between the Islamic banks and the conventional banks. The paper contributes to the literature on Islamic banks in at least three ways. First, we show that the hazard function of Islamic banks is different from those of conventional banks and provide the first formal tests of the hypothesis that Islamic banks are as equally likely to fail as conventional banks by using a survival analysis that exploits actual bank failures. We also justify why it is more appropriate to adopt models relying on actual bank failures than panel models relying on default risk proxies. The survival framework can deal with censoring effects; hence inferences are based on surviving as well as failed banks, all of which could have started operating at different points in time, thereby eliminating any unaccounted for survivorship bias that earlier studies might suffer from. Second, the use of the semiparametric Cox proportional hazard model enable us to draw comparisons between both bank types regarding the sensitivity of their failure risk to a comprehensive array of bank characteristics, macroeconomic conditions and market structure variables that extend those previously featured in the literature (see, e.g., Čihák and Hesse 2010 and Beck et al. 2013). Moreover, it has the advantage of not invoking any distributional assumptions with respect to the baseline hazard function. Third, we show that failure-risk predictions generated through general and bank type-specific models lead to different conclusions. The results highlight the importance of taking into account the 5
6. distinctiveness of Islamic banks in the context of failure-risk prediction . For instance, an Islamic bank specific model can identify the troubled Islamic banks better than a general model that does not cater for these banks’ distinct features. This finding is useful for regulators involved in the design of Early Warning Systems in a financial system where both bank types co-exist. The rest of the paper unfolds as follows. Section 2 presents the methodology and Section 3 describes the data. The empirical findings and implications are gathered in Section 4. Section 5 concludes. 2 Methodology We use the survival analysis that is better suited for our purposes than conventional classification techniques, such as the discriminant analysis or the binary logit model. The main reason for using the survival analysis is because it yields estimates of the expected time-to-failure. Second, the parameter estimation can be done with the partial maximum likelihood that requires no distributional assumptions on the time-to-failure. A third reason is that the analysis recognizes the continuous-time nature of the failure probability. Lastly, both censored and complete lifetime data are easily accommodated. The latter aspect is very appealing because it implies that the survival analysis of the banks’ failure risk naturally controls for the fact that the observation period might not represent a bank’s entire lifetime. Because the models exploit information on duration or survival time, defined as the actual number of years a bank has been in business, left-censoring is naturally avoided. However, a bank could remain in business beyond the sample, a problem that is known as right-censoring; the likelihood function of the survival models is formulated explicitly to account for the right-censored data.5 Moreover, the survival analysis is a more flexible method to analyze the risk of bank failure than an OLS analysis because an OLS requires a proxy for default or insolvency risk such as the z-score or Merton's distance-to-default. The z-score gives the number of standard deviations that the bank's return on assets (ROA) must drop below its mean in order to deplete equity as a percentage of assets, which leads to insolvency. One of the reasons for the popularity of the z-score is that it can be simply calculated from the ROA and the capital-asset ratio. But it has several drawbacks. One is that it assumes that the probability distribution of the bank's ROA is Gaussian that effectively implies ignoring higher-order moments such as skewness and kurtosis. But it is widely recognized that the Gaussian distribution is a crude (at best, first-order) approximation for financial returns because it presumes symmetry and underestimates downside tail risk. Another drawback to the z-score is that it merely acts as a proxy for the risk of insolvency but it does not convey information on the actual failure event. Because of the distinctive business model of the Islamic banks, it is not necessarily the case that the z-score is 5 For a deeper technical discussion on survival analysis, see Hosmer et al. (1999) and Kalbfleish and Prentice (2002). 6
7. applicable to them . Moreover, other measures such as Merton's distance-to-default require the use of listed banks that imposes an important constraint on the cross-sectional dimension of the sample. Banking studies that specifically address the issue of failure risk through the survival analysis follow two strands. The first one makes use of the semiparametric Cox proportional hazard model (Cox 1972; henceforth, the Cox model) that has the advantage of not requiring any distributional assumption on the hazard function and hence is a distribution-free approach. An early paper by Lane et al. (1986) adopts this framework to investigate the prediction of failure in the US banking sector. Whalen (1991) and Wheelock and Wilson (2000) extend Lane et al.’s study in terms of the sample. In a different setting, Dabos and Escudero (2004) examine failure in the Argentinean banking sector using the banks’ accounting information. More recently, Cole and Wu (2009), Gomez-Gonzalez and Kiefer (2009) and Molina (2002) also use the Cox model to assess conventional bank failure. The second strand of studies relies on parametric survival models that invoke distributional assumptions (see, e.g., Sales and Tannuri-Pianto (2007) for Brazil; Evrensel (2008) for a set of developed and developing countries; Männasoo and Mayes (2009) for Eastern Europe). Each of these studies assumes a different distribution for the baseline hazard (i.e., exponential, Weibull, and complementary log-log, respectively) that illustrates the potential problem of misspecification. We use the Cox model where T ∈ [0, ∞) denotes the time-to-failure that is a random variable with the probability density function f(t) and the cumulative density function F(t) defined as f(t) = −dF(t)/dt (1) F(t) = Pr(T ≤ t) (2) The survivor function S(t) gives the probability of surviving beyond year t, and the hazard function or hazard rate h(t) is defined as the instantaneous risk of the bank’s disappearance in year t conditional on its existence up to time t. These two crucial functions can be formalized, respectively, as S(t) = 1 − F(t) = Pr(T > t) h(t) = limdt→0 Pr(t ≤ T < t + dt) f(t) = dt × S(t) S(t) (3) (4) The object of primary interest in the survival analysis is the hazard rate that must be non-negative but not otherwise constrained, h(t) ≥ 0 , and that provides a time-varying risk of bank failure. We use the unconditional Kaplan and Meier (1958) estimator of the survivor function S(t) by using actual data on whether a bank failed over the observation window and the time when the failure occurred. The null hypothesis, the equality of the unconditional survival rates for the two bank types, is examined using a log-rank test statistic that is χ2(1) distributed. 7
8. Using the same data on bank failure together with the data on a vector of bank- and country-specific covariates, denoted