University of Warwick institutional repository: A Thesis Submitted for the Degree of PhD at the University of Warwick

Transcription

1 Unversty of Warwck nsttutonal repostory: A Thess Submtted for the Degree of PhD at the Unversty of Warwck Ths thess s made avalable onlne and s protected by orgnal copyrght. Please scroll down to vew the document tself. Please refer to the repostory record for ths tem for nformaton to help you to cte t. Our polcy nformaton s avalable from the repostory home page.

2 Essays on Bankng and Monetary Polcy n the Presence of Islamc Banks Jardne Husman A thess submtted n partal fulflment of the requrements for the degree of Doctor of Phlosophy n Economcs Department of Economcs Unversty of Warwck March 2015

3 Table of Contents Lst of Tables... v Lst of Fgures... v Acknowledgement... v Declaraton.. v Abstract..v Introducton... x Chapter 1 Bank Proftablty and Busness Cycles: Are Islamc Banks more Stable? Evdence from Indonesa 1.1 Introducton A Bref Overvew of the Indonesan Dual Bankng Industry Methodology and Data Methodology Data Emprcal Results Dynamc Estmatons Statc Estmatons Dscusson Concluson Appendx Chapter 2 How Do Islamc Banks Affect the Transmsson of Monetary Polcy? The Case of Indonesa 2.1 Introducton Monetary Polcy Framework and Islamc Bankng n Indonesa The Theoretcal Framework The Envronment The Equlbrum... 50

4 2.3.3 Transmsson Mechansm of Monetary Polcy Comparatve Statcs Data and Testable Predctons Data Testable Predctons Emprcal Works Emprcal Model Emprcal Results Dscusson Conclusons Appendx Appendx 2.1 Islamc Contracts Appendx 2.2 The Government Budget Constrant Appendx 2.3 Untroot and Contegraton tests Bblography... 95

5 Lst of Tables Table Summary statstcs.17 Table Results of dynamc baselne estmaton usng growth of: Industral Producton Index (IPI) and real domestc cement consumpton (CMT).20 Table Results of dynamc estmaton usng growth of Industral Producton Index 24 Table Results of statc estmaton usng average proftablty of bankng ndustry...26 Table A1.1 Seral test.30 Table A1.2 Pooled Crosssectonal on Standard devaton of ROA..30 Table A1.3 OLS estmaton on Zscore.31 Table A1.4 Dynamc baselne estmaton on the numerator (Profts) and the denomnator (Assets) of ROA Table A1.5 Dynamc baselne estmaton on rato of nterest expendtures over depost (IED).32 Table Events on each perod.43 Table Balance sheet of a type bank 46 Table Summary statstcs..58 Table Estmaton results of rate of return on loans (Rloans) 70 Table Estmaton results of rate of return on deposts (Rdep)..73 Table Estmaton results of total loans (n log values)..76 Table Estmaton results of netborrowng from the central bank (NCB)...78 Table A2.1 ImPesaranShn Untroot tests...84 Table A2.2 Contegraton tests...84 v

6 Lst of Fgures Fgure Rato of total balances at the central bank to total assets of bankng ndustry (%).39 Fgure Bank Indonesa polcy rate (BI rate n %).56 Fgure Groupaverage LoantoAsset rato (%) 60 Fgure Groupaverage rate of return on loans (Rloans n %)...66 Fgure Groupaverage weghted rate of return on deposts (Rdep n %)..66 Fgure Log of total loans (Lloans).66 Fgure Groupaverage netborrowng from the central bank (NCB n %) 66 v

7 Acknowledgement I am truthfully grateful to my supervsors, Thjs van Rens and Herakles Polemarchaks, for ther nvaluable supports and nsghtful comments throughout my studes. They have spared ther patence n gudng me. I also acknowledge many helpful and crtcal comments from Mchael McMahon, Marcus Mller, Juan Carlos Gozz and all the staffs and fellow students at the Warwck Macro and Internatonal Group workshops. I would also lke to thank Bank Indonesa for the fnancal support and the opportunty they have gven me n pursung my graduate studes. My specal thanks go to my frends and colleagues at Bank Indonesa and Fnancal Servce Authorty (OJK) for ther encouragements and assstances that n many ways have helped me a lot n completng my research. I extend my deepest grattude to my parents who have relentlessly supported me wth ther prayers, and to my sster and brother for ther close attentveness. Fnally, to my lovely chldren Selma and Mkal, and my sncere husband, words could never express my devout gratefulness of your beng. v

8 Declaraton I declare that the thess s my own work and has not been submtted for a degree at another unversty. Jardne Husman October 2014 v

9 Abstract Ths thess conssts of two chapters and ams to nvestgate the presence Islamc banks n Indonesa n the context of stablty and monetary polcy transmsson mechansm. The frst chapter compares bank stablty, n partcular proft stablty, n Islamc versus conventonal banks amd busness cycle fluctuatons. The unque characterstcs of Islamc fnance prncples hypothetcally nvolve dfferent fnancal structures and provde stablty for banks that comply wth them. Usng monthly banklevel data wth comparable banks across the two types, I nvestgate the dynamc responses of ndvdual banks to busness cycle fluctuatons. The dynamc estmaton results show that the profts of Islamc banks are more stable than those of conventonal banks n the short run, yet generally ndcate no sgnfcant dfference n the long run. However, the ncluson of the loantoasset rato removes the remanng shortrun dfferences. I check for robustness by estmatng the statc relatonshp between ndvdual bank s profts and the average proftablty of the total bankng ndustry, and the outcomes support the nodfference results. The second chapter compares the monetary transmsson through Islamc and conventonal banks by nvestgatng how a partcular bank asset portfolo, whch corresponds to ther type from beng conventonal or Islamc, determnes the equlbrum rate of return on loans and on deposts whch n turn affects ther loans and netborrowng from the central bank. Certan applcaton of Islamc fnance prncples leads to a marked dfference n Islamc banks assets portfolo whch corresponds to a consstently hgher loantoasset rato n comparson to conventonal banks. I test a set of predctons conveyed by the theoretcal model usng a panel of ndvdual bank data. The results turn to be hghly dependent on how well the two types of banks are segregated from each other, n whch slack segregaton may dsspate the potental dfferences. In partcular whle ntally the results show no sgnfcant dfference n responses of the two types of banks to the central bank polcy rate, excludng Islamc wndows from the sample allow the potentally dfference to be more apparent and sgnfcant. Overall, the possblty that both types of banks may after the same pool of consumers exposes them to compete wth banks from the other type, generatng arbtrage opportuntes that drve prces toward equalty across types and mpede the potental dfference across banks. v

10 Introducton Ths thess ams to nvestgate the presence of Islamc banks n a country that has a dual bankng system,.e. a system where conventonal and Islamc banks operate sde by sde wthn the economy, n the context of stablty and the monetary polcy transmsson mechansm. Islamc banks are banks that comply wth the prescrpton n Shara law concernng ts operaton, whch s based on the Quran and the Hadth the authentc tradton. The general prncples of Islamc fnance could be summarzed, but not lmted, to the followng: prohbton of nterest, thus encouragng the proft and loss sharng mechansm; prohbton of contractual uncertanty; and prohbton of gamblng, speculaton or excessve rsktakng n carryng out transactons (Askar et al., 2010). Islamc prncples also promote a materalty aspect, requrng the drect lnkng of fnancng wth the underlyng asset so that any fnancng actvty s clearly and closely dentfed wth the real economy (Iqbal and Mrakhor, 2007). In ts operatons, accordng to Askar et al. (2010), the concept of Islamc bankng ncludes two types of bankng actvtes: () safe keepng and payment actvtes and () equtybased nvestment actvtes. The frst s bascally 100% reserve bankng, where deposts reman hghly lqud and where checkng servces are perfectly avalable. In the second, the depostor s, n essence, purchasng equty n the bank, and the bank tself has an equty poston n the borrower s busness n whch Islamc banks drectly partcpate n a rsktakng process through trade, leasng and other productve nvestments n the real sectors. In a more general context, the concept of Islamc bankng actvtes resembles proposals for monetary reform known as the Chcago Plan, ntated n Smons (1933, 1948) and revved n Benes and Kumhof (2012); and, more recently, the Lmted Purpose Bankng x

11 (LPB) proposed n Kotlkoff (2011) and Chamley et al. (2012). Both proposals emphasse the necessty of 100% reserve requrement and the mportance of elmnatng leverage from the fnancal ntermedares n achevng a stable and wellfunctonng fnancal system. 1 Whle those two proposals are not yet mplemented, many countres have partally adopted the Islamc bankng concept wthn ther dual bankng system. Nevertheless, the concept of Sharacomplant product and servces of Islamc bankng vares a great deal across countres (Beck et al., 2013). Many also argue that Islamc banks daytoday operatons are dvergng from the ntal concept above, dmnshng the gap between Islamc and conventonal banks (see among others: Chong and Lu (2009); and Cevk and Charap (2011)). Ths thess examnes Islamc bankng practces n the context of stablty and the monetary polcy transmsson mechansm n the case of Indonesa n two chapters. Whle the frst chapter drectly assesses stablty n terms of proft stablty, the second chapter does ths by evaluatng banks responsveness to monetary polcy. The frst chapter focuses on proft stablty of Islamc and conventonal banks by measurng the senstvty of bank profts to fluctuatons n the busness cycle usng dynamc panel estmaton of Blundell and Bond (1998). Based on the above concept, Islamc banks proftablty s more stable than conventonal banks, snce Islamc banks have the ablty to adjust ther lablty n accordance wth ther assets performance. The results show that Islamc banks have slghtly better, yet statstcally sgnfcant, stablty n the shortrun n comparson to conventonal banks. Nevertheless, ths dfference s not large enough to delver any dfference n profts varablty across the two types of banks n general. The second chapter provdes theoretcal and emprcal studes on how the specfc transmsson mechansm of monetary polcy works n Islamc and conventonal banks wthn a dual bankng system. In partcular, how a certan bank characterstc, whch corresponds to ther type as ether conventonal or Islamc, determnes the equlbrum rate of return on loans and on deposts, whch, n turn, affects ther loans and netborrowng from the central 1 LPB s actually a proposal that desgnated to reform not just the bankng system but also all fnancal ntermedares such as nsurances, hedge funds, etc. (Kotlkoff, 2011). x

12 bank n response to the polcy rate. Ths characterstc s the loantoassets rato, whch, due to the materalty aspect of Islamc fnance explaned above and the lmted sharacomplant fnancal products that are avalable n the market, s consstently hgher n Islamc banks n comparson to conventonal banks. The emprcal work tests a set of predctons usng dfferencendfference method and ntally fnds that there s nodfference n responses of the two types of banks to the central bank polcy rate. However, when Islamc wndows are excluded from the sample, the results show sgnfcant dfferences n the response of rate of return on loans and netborrowng from the central bank n response to the polcy rate. 2 In partcular, n support for the theoretcal framework, monetary polcy that works va conventonal banks s stronger than Islamc banks. Overall, these studes show that the dual bankng system may nevtably drves both types of banks to somewhat share ther customer base, generatng arbtrage opportuntes that may drve prces toward equalty across types. Ths s partcularly stronger for Islamc wndows where the two types of banks are closely ntercorrelated to each other. Whle emprcal results of the second chapter show that Islamc fullfledged may relatvely have more lmted exposure to conventonal banks, t does not necessarly mply that they are not n competton wth conventonal banks. Ths s reflected n the frst chapter, that only deals wth Islamc fullfledge banks, whch suggests that there s a possblty that some Islamc banks may partly depart from the ntal concept of an equtybased system by smoothng out the payment to ther depostors n order to sustan ther compettveness wth conventonal banks. Nevertheless, future research s needed to further nvestgate ths ncome smoothng possbltes. Further, the emprcal results n the second chapter confrm the smlar evoluton of rate of return on deposts across the two types of banks, whch may also ndcate that the two types of banks are n competton wth each other. As one of the consequences, though Islamc banks tend to be more stable n term of ther responsveness to monetary polcy, yet they generally could not provde better stablty n term of ther proftablty n comparson to ther conventonal counterpart. 2 Islamc wndow are unts of conventonal banks that offer Islamc bankng servces. x

13 Chapter 1 Bank Proftablty and Busness Cycles: Are Islamc Banks more Stable? Evdence from Indonesa 1.1 Introducton Macroprudental analyses requre thorough knowledge of how bankng sector proftablty s related to busness cycles (Albertazz and Gambacorta, 2009); ths s especally true for a bankbased economy. The bankng sector s soundness and stablty are closely determned by ts proftablty and the structural factors affectng t (DemrgucKunt & Detragache, 1998). In partcular, bank balance sheets can deterorate when the rate of return on the asset sde, whch s closely related to overall busness performance through the economc performance of bank borrowers, falls short of the rate that must be pad on the lablty sde. Hence, from ths pont of vew, f banks could adjust ther lablty accordng to ther assets performance, hypothetcally the system would be nherently stable and less senstve to busness cycle fluctuatons. Theoretcal studes n Islamc bankng, such as those of Khan and Mrakhor (1987, 1991) and Askar et al. (2010), assert such flexblty. 3 In theory, there are several marked dfferences n Islamc fnance compared to conventonal fnance. What follows are the basc 3 The theoretcal Islamc bankng s not the only concept who offers flexble as opposed to fxed predetermned return n bankng products. Smons (1948) wth hs proposal the Chcago Plan, revved recently by Benes and Kumhof (2012); and most recently the Lmted Purpose Bankng n Kotlkoff (2011) and Chamley et al. (2012) are among the proponent of equtybased bankng system. 1

14 prncples of Islamc fnance, accordng to Iqbal and Mrakhor (2007). Frst, Islamc fnance prncples prohbt nterestbased payments. Any predetermned rate ted to the maturty and the prncpal amount, so that t s guaranteed regardless of the performance of the nvestment, s prohbted. Second, Islamc law encourages proft makng, for t represents successful entrepreneurshp and the creaton of addtonal wealth. Islamc fnance requres borrowers and lenders to share rewards and losses n an equtable fashon, and ths process of accumulatng and dstrbutng wealth n the economy represents true productvty. Thrd, Islamc fnance prncples promote a materalty aspect that requres fnancng to be assocated drectly wth the underlyng asset, n whch the fnancng actvty s clearly and closely dentfed wth the real sector actvty. Lastly, hoardng s dscouraged and transactons featurng extreme uncertanty or gamblng wth excessve rsks and the fnancng of specfc llct actvtes are prohbted. Wth reference to stablty, Khan and Mrakhor (1987) and Askar et al. (2010) emphasse the usage of the proft and loss sharng mechansm on both sdes of the bank balance sheets, especally wth respect to the lablty sde, as a devce to ensure banks fnancal stablty n general or proft stablty n partcular, albet departng from any potental problems of moral hazard. In practce, however, many argue that, nowadays, Islamc banks daytoday operatons are dvergng from these ntal prncples, albet mplctly. Some of the emergng fnancal products of the Islamc bankng ndustry closely resemble products of conventonal bankng, dmnshng the gap between the two systems. Emprcal studes by Chong and Lu (2009) and Cevk and Charap (2011), among others, on Malaysan Islamc banks fnd that Islamc depost or equvalent rates are not nterestfree, but closely pegged to conventonal depost rates. On the asset sde, Khan (2010) fnds that, for a sample of large Islamc banks across several countres, only a neglgble porton of Islamc banks fnancng s actually based on a proft and loss sharng mechansm. These sorts of fndngs may ndcate an mpedment to the nherent stablty that theory predcts. Whether there are 2

15 dfferences n proft stablty across busness cycle fluctuatons between Islamc and non Islamc banks s an emprcal queston. 4 In a small but growng lterature on Islamc bankng, some studes have already sought to analyse the fnancal stablty, n partcular nsolvency rsk, of Islamc banks and tease out any dfferences n comparson to ts conventonal counterparts. 5 Chak and Hesse (2010), Beck et al. (2013), and Abedfar et al. (2013) use crosscountres banklevel data of OIC members to assess and compare the fnancal stablty of Islamc banks versus non Islamc banks. Whle Chak and Hesse focuses ts work solely on assessng nsolvency rsk, Beck et al. and Abedfar et al. do ths as a part of another research focus, such as effcency and busness models for the second study, and credt rsk, nterest rate rsk and relgous rent for the thrd study. All these three studes employ the zscore as a measure of bank stablty or nsolvency rsk. 6 In general, they fnd that, controllng each country s characterstcs, there s no sgnfcant dfference n terms of stablty between Islamc and nonislamc banks. If the analyss takes nto account bank sze, Chak and Hesse (2010) fnds that small Islamc banks tend to be more stable than ther conventonal counterparts, whle the contrary apples for large Islamc banks. Smlar results are also found by Abedfar et al. (2013) for small banks, but they do not dscover any dfference for large banks. As for Beck et al. (2013), they do not encounter any sgnfcant dfference n stablty across all specfcatons. In ths paper, we follow Albertazz and Gambacorta (2009) n usng one of the components of the zscore, namely proftablty, as the dependent varable n measurng bank stablty n facng fluctuatons of the busness cycle. There are other factors whch can serve as fnancal stablty ndcator such as for example lqudty and loan qualty. However, we focus on proftablty because t s one of the man ndcators of fnancal soundness n bankng (see Fnancal Soundness Indcator, IMF (2006)) and some studes fnd that bank 4 I use the term nonislamc and conventonal banks nterchangeably; however the emprcal sectons merely use the term nonislamc banks. 5 Abedfar et al. (2013) provdes a summary of emprcal lterature on Islamc bankng. 6 The zscore, theoretcally developed by Boyd and Runkle (1993), measures the probablty or dstance to default, combnng proftablty, leverage and volatlty. Hgher value of zscore mples a lower probablty of nsolvency. It has been used quet extensvely n the recent lterature; see e.g. Laeven and Levne (2005), DemrgucKunt and Huznga (2010) and Houston et al. (2010). 3

16 proftablty s an mportant predctor of bankng sector vulnerablty (DemrgucKunt and Detragache (1999), Chak and Schaeck (2007) and Navajas and Thegeya (2013)). Adverse economc condton leads to deteroraton of loans qualty, ncreasng nonperformng loans and credt losses, whch ultmately worsen banks profts. We further compare the stablty of Islamc versus nonislamc banks by measurng the senstvty of bank profts to the busness cycle. The more senstve bank profts are to busness cycle fluctuatons, the more lkely t s for proftablty to be unstable. Ths mode of analyss stems from the predcton that banks ablty to adjust ther lablty n accordance wth ther assets performance (through proft and loss sharng mechansm n the case of Islamc banks) nfluences the stablty of ther profts. Based on ths, as suggested by Khan and Mrakhor (1987) and Askar et al. (2010), Islamc banks proftablty s more stable than nonislamc banks. The resultng emprcal framework allows predcton of factors affectng bank proftablty, as well as comparson of senstvty to the busness cycle across the two bank groups. By so dong, the emprcal framework also partally serves as a test for the mplementaton of proft and loss sharng mechansm of Indonesan Islamc banks. There are qute a range of studes n modellng bank proftablty (for cross countres studes, see Bourke, 1989; Molyneux and Thornton, 1992; and, more recently, Goddard et al., 2011; Athanasoglou et al., 2008, for the Greek case; Detrch and Wanzenred, 2011, for the Swtzerland case), whereas only a dearth of studes exst on Islamc banks; to name a very vew s Bashr and Hassan (2003) for a crosscountres analyss. Most of these studes use macroeconomc and busness cycle varables as control varables. There are also studes that explctly nvestgate the relatonshp between bank proftablty and the busness cycle: Bkker and Hu (2002), and, more recently, Albertazz and Gambacorta (2009) emprcally assess the cyclcalty of bank proftablty by nvestgatng the way components of bank proftablty relate to the busness cycle. Ths paper attempts to compare the fluctuatons or stablty of Islamc and conventonal banks proftablty by nvestgatng ts senstvty to the busness cycle fluctuatons. To our knowledge, t s the frst paper to provde such a comparson from ths 4

17 angle. The novelty of ths paper les n ts ablty to smultaneously model Islamc and conventonal banks proftablty, as well as comparng ts stablty amd fluctuatons n the busness cycle. We utlze Indonesan banklevel data for monthly perods durng 2007:1 2012:10. To nvestgate the dynamc response of ndvdual bank proftablty to busness cycle fluctuatons, we employ a dynamc panel technque whch allows estmaton on shortrun and longrun responses smultaneously. We check for robustness by runnng the same regresson on two sample sets, one ncludes all banks and the other only ncludes banks wth comparable n terms of ther ownershp structure and assets sze across types, Islamc and conventonal. The dynamc results show that the proftablty of Islamc banks s more stable than conventonal banks n the shortrun and ths result s robust across sample sets. Though the shortrun dfference n proft stablty between Islamc and conventonal banks s statstcally sgnfcant, yet the magntude s farly small. The nclusons of bankspecfc characterstcs, such as sze, credt rsks and captal, do not alter the results. However, the ncluson of the loantoasset rato removes the remanng shortrun dfferences. As there s evdence that a bank wth a hgher value of ths partcular characterstc has more stable proftablty, whch Islamc banks nherently possess, ths mples that the loantoasset rato may well explan the perceved shortrun dfference between the two groups. These dynamc results are even stronger for the sample set of banks wth smlar ownershp structure and assets sze. For the fnal robustness check and to grasp an overvew on fluctuatons of proftablty across the two bankng groups, we estmate the statc relatonshp between ndvdual bank proftablty and the average proftablty of the total bankng ndustry. We fnd no evdence of dfferng fluctuatons between the two bank groups across sample sets, whch supports the nodfference results of the frst stage s longrun estmatons. Thus, n general, we could not observe sgnfcant dfference n fluctuatons of Islamc and non Islamc banks proftablty. The paper s organzed n the followng manner. Secton 2 brefly descrbes Indonesan dual bankng ndustry. Secton 3 explans the methodology and outlnes the data. Secton 4 presents the emprcal results. Secton 5 concludes the paper. 5

18 1.2 A Bref Overvew of the Indonesan Dual Bankng Industry Ths secton frstly outlnes overall performance, detals some facts about the Indonesan bankng sector, and then brefly descrbes the development of the country s Islamc bankng ndustry. The bankng ndustry domnates the entre fnancal sector of Indonesa, whose assets account for 80% of the sector s total assets. Nevertheless, the fnancal sector assets account for around only 85% of total GDP, whch reflect relatvely shallow fnancal sector nclusons. In terms of contrbuton to the economy, the total loans of the ndustry only account for around 35% of GDP, whle nternal funds reman the man source of funds for frms nvestment actvtes. Ths low contrbuton s due to relatvely underdeveloped and shallow fnancal markets n Indonesa. In regard to the recent wdespread fnancal crss of 2008, though the mpact to Indonesa was farly lmted, the bankng sector endured a swft perod of lqudty shortage n August, when the rato of lqud nstruments held by banks to noncore deposts reached ts lowest pont of around 85%, 7 and the fnancal stablty ndex (FSI) reached ts hghest pont n November of In spte of ths, the total bank loans by that tme were stll growng qute hgh for annual growth more than 30%. In terms of the structure of the ndustry, the bankng sector s domnated by governmentowned banks, n whch fve of the top banks account for around 45% of market share. Followng deregulaton n 1998, trggered by the bankng crss, foregn ownershp ncreased dramatcally for up to 30% changes n total market share. The Indonesan bankng sector adopts a dual system, n whch conventonal and Islamc banks operate sde by sde wthn the economy. The frst Islamc bank was establshed n In order to encourage the development and network expanson of the Islamc bankng ndustry, the government amended the Bankng Act n 1998, whch allowed conventonal banks to open an Islamc wndow, a unt that provdes Islamc bankng servces. It further strengthened the busness by amendng the Central Bank Act n 1999, 7 Bascally, a rato less than 100% ndcates that a bank does not have adequate lqudty to meet depost wthdrawals. 8 The FSI n Indonesa represents fnancal sector stablty, whch ncludes bankng, bonds and stock markets. The maxmum ndcatve lmt s level of In the economc crss of 1997/98, the ndex topped at

19 allowng monetary control usng nstruments that are based on Islamc prncples. To further boost the ndustry, a dedcated act for Islamc bankng was ssued n 2008, n whch the exstence of Islamc wndows s regarded as temporary, and shall be completely spunoff to fullyfledged Islamc banks by addng more captal up to a certan level. In addton, besdes establshng new Islamc banks, one can convert a conventonal bank nto an Islamc bank, but not the other way around. Durng the sample perod of ths paper, there are cases of such conversons, whch are commonly carred out by the acquston of smaller banks by conventonal banks then converted nto Islamc banks. The nclnaton of people to employ Islamc bankng servces s stmulated by the ssuance of a verdct on the prohbton of nterest by the councl of the country s Islamc scholars n Nonetheless, though the Indonesan populaton s predomnantly Muslm, the Islamc bankng sector stll accounts for just a very small porton of the bankng ndustry, wth total assets accountng for around 5% of the total ndustry s assets, though there s rapd annual growth of more than 50%60% on average. One of the condtons that commonly cted as an explanaton for low market share s the dearth of product nnovaton n the Islamc bank portfolo. Complance wth Shara, or Islamc, prncples may hnder fnancal product nnovaton and, n turn, hamper Islamc bank compettveness. In addton, there are dfferences n the nterpretaton of Shara across Muslm countres. 1.3 Methodology and Data Methodology To compare Islamc and nonislamc banks profts senstvty wth the busness cycle fluctuatons, the followng emprcal model s appled to a panel of ndvdual Islamc and nonislamc banks n Indonesa: jt NI j X jt Z t NI j X jt jt c, jt ujt (1.1) where jt s the proftablty of bank from group j,.e. Islamc or nonislamc, at tme t, c s a constant term, NI j s a nonislamc dummy whch s equal to 1 for non 7

20 Islamc banks and 0 for Islamc banks, X jt s a matrx of bank characterstcs as control varables, Z t s the busness cycle varable and jt the dsturbance, wth the unobserved bankspecfc effect and ujt the dosyncratc error. Equaton (1) s a oneway error 2 component regresson model, where IIN, and ndependent ofu jt, whch may 0 a have bankspecfc patterns of heteroskedastcty and seral correlaton that are uncorrelated across banks. The effect of busness cycles on proftablty s captured by the sum of parameters and for Islamc banks and by the sum of parameters, and for nonislamc banks. In general, we expect procyclcalty n bank proftablty. The nteracton terms of the components n the bracket wth the busness cycle varable are ntended to capture how banks from the two groups, Islamc and nonislamc, and wth dfferent characterstcs, may respond dfferently to busness cycle fluctuatons. In partcular, the nteracton wth NI j s the man varable of nterest n whch any dfference between varablty and or stablty n proftablty of the two groups s captured by parameter. The nterpretaton of whch group s the one wth more stable proftablty depends on the sgn and magntude of these estmated parameters. NonIslamc banks would have more unstable profts f both parameters have the same sgn, or f the absolute sum of the two s greater than the absolute value of. 9 The straghtforward ntuton s that proftablty of a bank wth a larger response to the busness cycles s more volatle than a bank wth a smaller response. Accordng to the theoretcal analyss of Khan and Mrakhor (1987) and Askar et al. (2010), a nonislamc bank s expected to be more unstable n comparson to ts Islamc counterpart based on a condton that the latter treats deposts smlarly wth shares or equty, n whch depostors would not be guaranteed a predetermned nomnal return; they would receve a share of the bank proft nstead. In the Khan and Mrakhor model, a shock to real ncome n both types of bank may delver shortrun dfferences between the two. In partcular, Islamc banks do not necessarly need to resort to the labltymanagement technque of rasng nterest rates to bd on deposts, a shortrun process that they clam 9 The emprcal results n the next part report the mpled response of nonislamc bank profts separately. 8

21 would lead to nstablty. In sum, theoretcally shocks n the assets value would be passed through to the lablty sde, leavng the system stable. Based on ths argument, the expected sgn and magntude of the estmated parameters ˆ and ˆ would be those whch mply more unstable proftablty of a nonislamc bank. Nevertheless, the stablty of bank proftablty may well be nfluenced by ther own ndvdual characterstcs, whch may or may not be related to them beng Islamc or nonislamc. For ths reason, model (1.1) also ncludes bank characterstcs as possble explanatory and control varables. The same nterpretaton of parameter also apples to on the nteracton of bank characterstcs wth the busness cycle varable, whch explans how dfferent bank characterstcs may respond dfferently to the busness cycle. Though our man objectve s to capture the stablty and volatlty of bank proftablty through fluctuatons n the busness cycle, model (1.1) provdes ths by estmaton of the level of proftablty rather than drectly of ts volatlty. 10 By so dong, not only does t preserve the avalable tme observaton, t also allows estmaton of the effect of bank characterstcs on volatlty of bank proftablty through and on level of proftablty through, based on the assumpton that the busness cycle varable Z could capture most of the varablty of bank proftablty. Whle other studes n the lterature do not generally focus explctly on the effect of the busness cycle on bank proftablty due to ts crosscountry and short tme dmenson nature, the avalable observaton and frequency of ths study allows more elaboraton, n partcular the ablty to measure the shortrun effect alongsde the longrun effect. The bank characterstcs ncluded n the estmaton are () captal, () sze, () credt rsk, and (v) loantoasset rato. The latter may represent bank product dversfcaton and also sgnals ts lqudty condton. The frst characterstc, captal, s measured by rato of equty to total assets: the effect of captal on bank proftablty s ambguous. Whle captal can postvely affect bank proftablty through reducng the cost of external fundng as captal may reflect lower rsks, t may also negatvely affect bank proftablty through conventonal rskreturn hypothess where banks wth a lower captal rato, and therefore 10 For robustness check, OLS estmaton on standard devaton of ndvdual bank proftablty s also undertaken. 9

22 more rsky, delver hgher returns n comparson to bettercaptalzed banks. Thus, the net effect of captal on bank proftablty cannot be antcpated theoretcally. The effect of captal on volatlty of proftablty s drectly assocated wth ts capacty as a proxy for rsk, n whch hgher rsk mples greater volatlty. The second characterstc, sze, s measured by rato of a bank s total assets to the average of ndustry s assets. Generally, most studes suggest that bgger banks are able to generate hgher profts, though ths may not be n a lnear manner as n Athanasoglou et al. (2008), and some relate sze to captal, then, n turn, to proftablty, as seen n Bkker and Hu (2002) and Goddard et al. (2004). The effect of sze on the volatlty of proftablty s not straghtforward, snce there s no drect relaton between bank sze and ts rskness. However, we could follow prevous studes by relatng bank sze to ts captal and, n turn, to ts rsk poston. Thrd, credt rsk, measured by rato of loanloss provsons to total loans. Ths theory expects negatve effects of credt rsk on proftablty, as lower credt qualty reduces a bank s ablty to generate greater proft. The effect of credt rsk on volatlty s straghtforward, as more rsk s assocated wth hgher volatlty. Last, the loantoasset rato s one of the measures for lqudty rsk and product dversfcaton. 11 Agan, a conventonal rskreturn hypothess expects a hgher loanstoasset rato, hence hgher lqudtyrsk, delverng a hgher return and hgher volatlty. On the other hand, a hgher loantoasset rato reflects lower product dversfcaton, and suggests that a hgher product mx enables more stable revenue. Thus, accordng to ths vew, we expect a postve relaton between the loantoasset rato and bank proftablty, and a negatve relatonshp wth ts volatlty. Overall, the effect of the loantoasset rato s ambguous. Whle the lterature manly explores the effect of bank characterstcs on bank proftablty, t rarely drectly examnes the effect of these characterstc on stablty or senstvty of bank proft on the busness cycle. 12 The emprcal model specfcaton n model 11 Another common measure of lqudty rsk s rato of lqud assets such as cash, bank deposts and securtes to total assets, as n among others Bourke (1989) and Molyneux and Thornton (1992). 12 Some of these studes look at the effect of product dversfcaton on proft stablty n a smlar way wth analyss n nonfnancal frms. Others examne bank stablty based on probablty of default or zscore mentoned prevously. 10

23 (1.1) allows estmaton of the effect of these characterstcs on both bank proftablty and ts volatlty across tme n a sngle equaton. In order to observe the partal effect of each partcular characterstc on the estmated parameter ˆ, and as a parsmony model s preferred to a more general one, the ncluson of bank characterstcs s done by ntroducng one characterstc at a tme. 13 Due to the monthly data frequency and relatvely long perod of observaton, the tme seres component of the estmaton needs to be treated carefully for the model to be correctly specfed. Wth ths n mnd, and supported by the lterature of persstence n bank profts (see, among others, Berger et al. (2000), Goddard et al. (2011) and Athanasoglou et al. (2008)), model (1.1) s augmented to dynamc form as the followng: jt c L l1 jtl L L l l NI j X jtl Z t l NI j X jtll jt l0 l0 (1.2) where jtl are the lags of bank proftablty, whle the explanatory varables also enter the equaton wth the necessary number of lags based on a statstcal test, except for the NI dummy. Proft persstent mples parameter n model (1.2) to be nonzero, thus the model exhbts dynamc panel whch ncludes lagged of dependent varable. Though gnorng seral correlaton does not lead to nconsstency, t would mpar effcency, bas the standard error, and thus nflate the resultng tstatstcs. On the other hand, the ncluson of the lagged dependent varable has ts own drawback through ncorrectly domnate the regresson by obscurng the actual relatonshp (Beck, 2005). For ths reason, t s strongly suggested that a model wth a lagged dependent varable s done based on theoretcal consderaton or vald checks of the resdual structure that leads to the necessty of dynamc estmaton. Based on ths, to check the valdty of usng the dynamc model (1.2), the statc model (1.1) s frst estmated to see whether the resultng resduals are serally correlated usng test for frst order seral correlaton n lnear fxed (or random) effects paneldata by Wooldrdge (2002). The test for seral correlaton shows that the null of no seral correlaton s strongly rejected wth Fstatstcs and pvalue equal to and 13 Havng parsmonous model s one of the ways n avodng problem of toomany nstruments explaned below. 11

24 0.000, respectvely. Further, n order to fnd the approprate dynamc specfcaton, test for statonarty on each varable are undertaken by usng a untroot test for the unbalanced panel by Im, Pesaran and Shn (2003). The test results (Table A1.1 n the Appendx 1) show that ROA and all the control varables are statonary n level and that the maxmum lag s around 1. The busness cycle varables, explaned n the next secton, are dfferencedstatonary; thus we use the percentage of monthly growth n the estmaton. These results lead us to use an autoregressve dstrbuted lags (ADL) model of order (1, 1). Ths partcular model s very convenent as t provdes estmaton of shortrun and longrun effects on statonary data n a sngle equaton Error Correctons Model (ECM). Thus, model (2) can be jt rewrtten as the followng: c jt l l NI j X jtl Z t l NI j X jtll jt 1 (1.3) l0 l0 In estmatng model (1.3), the fxed effect method, commonly used n the lterature around ths topc, gves rse to dynamc panel bas or Nckell bas (see Nckell, 1981), especally for small T and large N panel, and nconsstent estmates (see Baltag, 2001). To obvate such problems, model (1.3) s estmated usng the dynamc panel estmaton method suggested by Blundell and Bond (1998), whch was ntally outlned n Arellano and Bover (1995). Ths method offers greater effcency than the earler method by Arellano and Bond (1991) (see Roodman, 2009), partcularly n the case of the dependent varable, whch s close to random walk. 15 The method uses lagged values of dependent varable n levels, as well as n dfferences n nstruments. It also allows possble forms of endogenety of the control varables, such as captal and sze, by applyng the same technque used n nstrumentng the lagged dependent varable. 16 In order to get the rght specfcaton, we apply Hausman test for each model wth a control varable. The test s based on the 14 As a smple example for the baselne estmaton whch does not nclude any control varable, the dfference n shortrun effect of busness cycle to bank proftablty s and the longrun effect s 0 ( 0 1) /(1 ). The same rules apply for other parameters. 15 The next secton on emprcal results shows that the frst lagged of the dependent varable n (2) s ndeed have hgh value wth ˆ up to GarcaHerrero et al. (2009) suggests that banks wth hgher proftablty may more easly ncrease ther captal and also may be more able to expand ther sze. 12

25 assumpton that the correct specfcaton for the control varable s fullexogenety. Treatng the control varable as predetermned s consstent, whether or not the assumpton s true, and treatng t as exogenous s only effcent and consstent f, and only f, the assumpton s true. Nevertheless, cauton should be exercsed when determnng the number of lags used as nstruments. The dfference and system GMM approaches are prone to too many nstrument problems (see Roodman, 2009), n whch a longer tme perod aggravates the problem of nstrument prolferaton. The large number of nstruments may fal to oblterate ther endogenous components, whch leads to a bas estmated coeffcent toward those of nonnstrumentng estmator. One of the symptoms of ths toomany nstrument problem s the mplausbly hgh p values of of the Hansen test (Bowsher, 2002). One of the ways to avod such a problem s that, as an arbtrary rule of thumb, the number of nstruments may not exceed the number of ndvduals n the panels. For ths reason, we restrct the lags range used and collapse the nstrument matrx. The number of nstruments used n estmaton s reported alongsde the Hansen test p value. The emprcal work ncludes two parts of analyss. The frst part s ntended to nvestgate the dynamc of ndvdual bank proftablty usng busness cycle varables whch come from outsde the bankng sector for the varable Z t ; thus, they may ental dfferent shortrun and longrun effects. We employ the dynamc specfcaton n (1.3), whch allows estmaton of shortrun and longrun responses smultaneously. The second part, as robustness check, estmates the statc model (1.1). Ths tme the average bankng sector proftablty s used for Z t varable. Snce the relatonshp between ndvdual bank proftablty and the average proftablty of the total bankng ndustry s mostly dentcal across tme, usng statc specfcatons as n (1.1) s approprate, though may not be effcent. Ths estmaton s ntended to nvestgate the general comparson of profts stablty across the two bankng groups durng the regresson perod. However, the small number of Islamc banks n the sample may weaken the sgnfcance or the statstcal power of ths statc estmaton. 13

26 We also do another robustness check of the results wth respect to the selected sample. Beck et al. (2009) fnds that, usng banks n Germany as the sample, government banks are more stable than prvately owned banks. On the other hand, Iannotta et.al (2007) on crosscountres European banks, suggests that government banks have poorer loan qualty and hgher nsolvency rsk than other types of banks. To account for these, we estmate the same regressons of each part for two sample sets wth dfferent bank ownershp or nsttutonal types. In partcular, for the frst sample (Sample1) we nclude all banks, whle for the second sample (Sample2) we exclude government banks and regonal development banks Data The data source s the monthly bank reports to the central bank of Indonesa (Bank Indonesa, b), whch s an oblgatory report for every bank. The perod covered s from January 2007 to October 2012,.e. seventy perods. The sample consders 138 commercal banks, of whch 11 are Islamc banks. 18 Ths data set comprehensvely ncludes all banks whch are operatng n the perod of observaton. For cases of mergers and acquston, the newly formed bank has the same dentty number wth the man bank. For cases of conversons, the same number of dentty s used but wth dfferent values n the NI dummy varable. Durng the observaton perods, 7 nonislamc banks converted to Islamc banks and 1 Islamc wndow spunoff to fullfledged Islamc bank. The nonislamc banks consst of 5 government banks, 26 regonal development banks and 101 prvate banks, whle the Islamc banks are all prvatelyowned banks. The excluson of government and regonal development banks from Sample1 leaves the nonislamc banks and Islamc banks to be more comparable n term of ownershp structure snce now, n Sample2, both types solely consst of prvatelyowned banks. The panel s unbalanced snce there are some bank 17 For government banks the ownershp s 100% central government whle for regonal development banks the majorty of shareholders s the regonal government. 18 For techncal reason, the Islamc banks n the sample only consder fullfledged Islamc bank and excludes Islamc wndows or busness unt of nonislamc banks. 14

27 closures durng the perod of observatons, consstng of 8674 observatons for Sample1 and 6472 observatons for Sample2. Followng the usual practce n the lterature the proftablty varable s represented by the rato of profts to assets,.e. the return on assets, ROA, (Goln, 2001), measured by the annualzed proft before tax dvded by average assets n percentage value. The possble varablty of the denomnator may obscure the actual varablty of profts n the numerator. Therefore, for addtonal robustness check, we also estmate the baselne dynamc model wth the annualzed profts as dependent varable for both Sample1 and Sample2 (Table A1.4 n the Appendx 1). There are some ssues regardng the monthly data whch caused the calculated ROA n January to exhbt far larger absolute values than n other perods for most of the banks. Ths phenomenon, known as the Januaryeffect (Bank Indonesa, 2006a), could be ntated by several aspects that affect the calculated ROA through bank total assets, or, more drectly, through bank profts whch are outlned n the followng. The effect that works through bank total assets s that, n general, banks experence a drop n ther deposts and total loans at the begnnng of a year. The drop n deposts s due to a bank s hgh dependency on the realsaton of government spendng, whch reaches ts lowest level n January, partcularly for government banks and n turn affects the whole ndustry snce these government banks are account up to around 40% of the whole ndustry. 19 On the other hand, the drop n loans s due to a bank s busness plan, where January s often reserved for nternal consoldaton causng lower loans decsons. Largely, small banks are more prone to these drops, whch may actually lead to a loss n that partcular month. Equally mportant, the drop n bank loans s also due to lower deposts as the man source of funds for the supply of loans. The effect that possbly works through bank profts s related to another Januaryeffect that occurs n the fnancal market and whch manly affects those banks wth hgh tradng actvtes. 20,21 Hence, the January data does not govern or predct data for the remanng 19 See the Outlook of Indonesan Islamc banks 2012 (Bank Indonesa, 2012) for January effect as an explanaton of a drop n Islamc banks deposts durng January A seasonal anomaly n whch stock s prces ncrease n the month of January more than n any other month ( 15

28 months n a year. Based on these condtons, the January data s corrected by takng the average of the months precedng and followng January. These correctons are dentcally appled to both groups, Islamc and nonislamc banks. The busness cycle varable s represented by the ndustral producton ndex, whch measures changes n real producton of large and medumscale nonol manufacturng frms n Indonesa. Ths ndex s avalable n monthly perods and based on the monthly survey by a nondepartmental government nsttuton (Statstcs Indonesa, 2012). The manufacturng or ndustral sector contrbutes about a quarter of total Indonesan GDP, and around 20% of the total loans of the Indonesan bankng ndustry are dedcated to ths partcular sector. Ths sector s hghly senstve to market condtons and consumer demand, especally prvate nvestment, whch makes the ndex not only reflectng the manufacturng sector but also a good proxy for overall economc performance (Bank Indonesa, 2010). The ndex s often used by the central bank as one of the ndcators for the real sector s actvtes n formulatng monetary polcy (Bank Indonesa, 2006b). In addton, another proxy for a busness cycle varable that s also consdered s the total domestc consumpton of cement n real value, whch roughly represents the constructon sector. However, ths sector only accounts for around 6% of total GDP and around 5% of the total bank loans. Nevertheless, cement sales monthly data s also commonly used as an ndcator for trackng Indonesan domestc demand. Table reports the descrptve statstcs for the varables used n the estmaton and test the dfference n mean values of nonislamc and Islamc banks n both sample sets, Sample1 and Sample2. On average, the ROA of nonislamc banks s sgnfcantly hgher than those of Islamc banks and wth bgger varatons n both sample sets. The coeffcentofvaratons,.e. the rato of mean to standard devaton, of nonislamc banks ROA are and 2.402, for Sample1 and Sample2 respectvely, both are larger than of Islamc banks. These coeffcentofvaratons mply that nonislamc government and regonal banks varaton of ROA s less than the prvatelyowned ones, and that n both 21 See Nagasastra and Utam (2012) who nvestgate the exstence of the Januaryeffect n the Indonesan captal market. 16

29 sample sets Islamc banks tend to have less varaton of ROA n comparson to nonislamc banks. In addton, a smple regresson on 12 months standard devaton of ROA also shows that nonislamc banks tend to have hgher volatlty of ROA, n partcular n Sample2 (Table A1.2 n the Appendx 1). Table Summary statstcs NonIslamc a) Islamc b) Sample1 Sample2 Varables Mean Std.dev Mean Std.dev Mean Std.dev π (ROA) 2.30*** ** CAR 17.38* LLP 3.55*** *** LAR 56.66*** *** SIZE 1.04*** *** Number of small banks / Total banks n group SMALL 70/128 57/96 7/11 Descrpton π (ROA) Rato of profts before tax to average assets (%) CAR Rato of total equty to total assets (%) LLP Rato of loanloss provsons to total loans (%) LAR Rato of total loans to total assets (%) SIZE Rato of assets to the average of ndustry's assets SMALL Small bank f total assets < IDR5000 bllons Note: a) NonIslamc banks consst of government, regonal and prvate banks; b) Islamc banks are all prvate banks. Sample1 conssts of all banks, whle Sample2 exclude government and regonal banks. The dfference between mean value of NonIslamc and Islamc banks that s sgnfcant at 1%, 5% and 10% are marked wth ***, **, and * respectvely. In terms of assets sze, Islamc banks on average are smaller than nonislamc banks. The dfference s more apparent n Sample1 snce most of the government banks are large banks. Though the mean tests of SIZE varable show sgnfcant dfferences between non Islamc and Islamc banks, the dfference s relatvely smaller n Sample2. If we categorze banks as small for assets sze less than IDR 5000 bllon, the rato of small banks to total banks of nonislamc banks n Sample2 s around 60% smlar wth the number of small banks n Islamc banks group whch s around 63%, whle n Sample1 ths rato s around 17