The Impact of Macroeconomic Uncertainty on Bank Lending Behaviour in Jamaica

Transcription

1 Research Paper 2010/12 The Impact of Macroeconomc Uncertanty on Bank Lendng Behavour n Jamaca Sashana Whyte 1 Research Servces Department Research and Economc Programmng Dvson Bank of Jamaca Abstract Ths paper nvestgates the role that macroeconomc uncertanty plays n bankng sector lendng behavour n Jamaca usng a portfolo model recently proposed n the lterature. The econometrc results of the bounds contegraton testng procedure proposed by Pesaran et al. (2001) show that there s no long-run relatonshp between bank lendng and the ndcators of macroeconomc uncertanty. However, macroeconomc uncertanty does affect bank lendng n the short-run. Specfcally, the volatlty of the benchmark nterest rate, whch s affected by fscal and monetary polcy, was found to be the most crtcal macroeconomc varable. Therefore, concerns about the sustanablty of the current macroeconomc economc envronment could partly explan the current weak levels of credt. The results mply that, n the drve to stmulate credt, polcy makers also need to focus on the factors that wll enhance confdence about long-term macroeconomc stablty n addton to the bank/market specfc characterstcs. JEL Classfcaton: C52, E44, G21, Keywords: Bank lendng, macroeconomc uncertanty, ARDL Model, bounds test 1 The vews expressed are those of the author and does not necessarly reflect those of the Bank of Jamaca. 1

2 1.0 Introducton Followng the global fnancal crss n 2009 and the resultng economc recesson, there has been ncreasng nterest n the analyses of the lnkages between the macroeconomc envronment and the behavour of the bankng system. Therefore, the goal of ths study s to assess the effect of macroeconomc uncertanty on banks lendng behavour. Typcally, the volume of loans granted by a bank s thought to be a functon of ts nternal characterstcs such as sze, depost base, lqudty, credt polcy and other nternal factors. These factors are, for the most part, wthn the control of the bank. However, these factors, to a large extent, are nfluenced by the general macroeconomc envronment. Therefore, the general loan behavour of most banks wll be a reflecton of the sgnals from the aggregate economy. It s expected that f banks perceve the macroeconomc envronment to be stable, they form expectatons that borrowers wll be better able to repay loans because of ther mproved ablty to accurately predct ncome stream over the lfe of the loan. In a world wth perfect nformaton, only the key ndcators of macroeconomc performance such as GDP growth, nterest rates and nflaton would be needed to evaluate the outcome of a stmulus to the supply of credt. However, gven that banks rarely exhaust ther lendng capacty, ths study seeks to ascertan whether the ssues stemmng from asymmetrc nformaton nduced by macroeconomc volatlty s a major determnant of the bankng sector s wllngness to lend these avalable funds. In the presence of uncertanty, t s lkely that not only the frst moments (such as the rate of GDP growth, the level of nterest rates, or the level of nflaton) but also the second moments (measures of uncertanty about those magntudes) wll matter. There s also the lkelhood that frms demand for loans may be responsve to varatons n macroeconomc uncertanty, as they affect the expected return on nvestment projects. Baum et al (2004), suggests that snce banks must acqure costly nformaton on borrowers before extendng loans to new or exstng customers, uncertanty about 2

3 economc condtons (and the lkelhood of loan default) would have clear effects on ther lendng behavour and affect the allocaton of avalable funds. Therefore, as uncertanty ncreases, the loan to asset ratos should declne as greater economc uncertanty hnders banks ablty to foresee the nvestment opportuntes (returns from lendng). Conversely, when uncertanty s lower, ncomes wll be more predctable leadng to a hgher loan-toassets rato as managers take advantage of more precse nformaton about dfferent lendng opportuntes. A study by Talavera et al (2006) concluded that banks make more loans durng perods of boom and reduced level of macroeconomc uncertanty and curtal lendng when the economy s n recesson. Thus, the economc envronment s a systematc rsk component that affects every partcpant wthn the economy. Typcally, the state of the economy s measured by macroeconomc aggregates, whch nclude the gross domestc product (GDP), employment level, ndustral capacty utlzaton, nflaton, money supply and changes n the exchange rate. Ths would suggest that banks should adjust ther lendng behavour n response to the sgnals from these factors. Addtonally, banks loan portfolo ncludng volume, tenor and structure may be generally nfluenced by ther expectatons of the performance of the economy both n terms of stablty and quantum/level of performance. Based on the mportance of ths ssue to polcy, ths paper seeks to analyze the response of credt to macroeconomc volatlty or uncertanty. The bounds testng contegraton procedure proposed by Pesaran et al. (2001) s utlzed n ths exercse. Ths approach allows for the smultaneous determnaton of both the long-run and short-run relatonshp between macroeconomc uncertanty and bank lendng. It extends the emprcal research on ths topc wth respect to Jamaca, and seeks to add to the evdence reported by Urquhart (2008). Urquhart (2008) employed the GMM approach proposed by Arellano and Bond (1991) to examne the mportance of the bank lendng channel to monetary polcy. However, Urquhart (2008) focused on monetary polcy and as such dd not ncorporate aggregate demand or supply varables. In lght of ths concern, we use the autoregressve dstrbuted lag (ARDL) framework (Pesaran et al., 2001) to test the mpact 3

4 of macroeconomc uncertanty, whch s proxed by the mpact of changes n key macroeconomc ndcators on bank lendng. The advantage of usng the ARDL approach s that t allows testng for contegraton rrespectve of whether the regressors are purely I(0), purely I(1) or fractonally contegrated. Addtonally, ths method s attractve for modelng because of ts small sample propertes. The subsequent sectons of ths study are organzed as follows: secton 2 presents a bref revew of the lterature; secton 3 presents the emprcal and data specfcaton, secton 4 dscusses the econometrc technque; the penultmate secton presents the dscusson and fndngs and the fnal secton presents the concluson. 2.0 Lterature Revew Assessng bank lendng behavour and ts nteracton wth macroeconomc uncertanty wll nform polcy makers of the extent to whch developments n the macroeconomy affects banks performance. Consequently, ths topc has receved sgnfcant attenton n recent years partcularly n lght of the present recesson whch began n 2008 and the resultng credt crunch. Baum, Caglayan and Ozkan (2005) nvestgated the relatonshp between macro-economc uncertanty and bank lendng behavour of US banks usng quarterly data from They found that bank loans consttuted about 55% of bank total assets. The study measured bank lendng behavour as the dsperson of banks loans to total assets rato around ther mean values usng standard devaton as a measure of cross-sectonal dsperson of bank loans. The condtonal varance n quarterly ndustral producton and the change n the consumer prce ndex (CPI) nflaton were used as measures of macroeconomc nstablty. Usng a GARCH model the study found that one-year cumulatve effect of a 100 per cent ncrease n uncertanty, captured by the condtonal varance of ndustral producton (IP) and nflaton leads to somewhere between a 9-11 per cent (5-7 %) reducton n the dsperson of bank loans-to-asset rato for total loans, real estate loans and household loans. Ths fndng supports the vew that macroeconomc uncertanty dstorts the effcent allocaton of funds across potental borrowers. 4

5 Talavera, Tsapn and Zholud (2006) studed the behavour of bank lendng and macroeconomc uncertanty n Ukrane. Usng a proxy of the condtonal varance of consumer or producer nflaton or volatlty n money supply (M1 and M2) and ts component (demand and tme depost) for macroeconomc uncertanty, they found a negatve relatonshp between bank loan to captal rato and macroeconomc uncertanty. They found that banks ncreased ther lendng ratos when macroeconomc uncertanty decreases. However, the study found that the reacton of banks to changes n uncertanty s not unform and depends on bank-specfc characterstcs, n partcular, bank sze and proftablty. For the bank-specfc factors, changes n monetary aggregates whch can be related to macroeconomc polces are relatvely more mportant for large banks than for small banks. Ths fndng suggests that small banks are less able to change ther behavour over tme n response to changes n monetary polcy and ther lendng depends to a much greater extent on captal. Also, uncertanty emanatng from monetary polcy s sgnfcant for bank lendng behavour n the case of more proftable banks but less sgnfcant for the less proftable. The relatonshp between bank lendng behavour and economc uncertanty was also examned by Eckmeerwe et al (2006) for Germany and the Euro Area. Utlzng a vector-autoregressve (VAR) model and mposng aggregate demand, supply and monetary polcy shocks through short-run sgn restrctons on mpulse responses, the authors estmated the jont dynamc behavour of real GDP, the prce level, the short-term nomnal nterest rate and the stock of outstandng bank loans. The results suggest that the dynamc responses n the two areas are broadly smlar. However, there are some dfferences n the relatve contrbuton of the three shocks to output, prces, nterest rates and bank loans over tme. To assess the role of bank lendng n the transmsson of macroeconomc shocks, specfcally the dstrbutonal mplcatons of potental credt market frctons, the authors performed counterfactual smulatons and analyzed the dynamc responses of German loan sub-aggregates. The results suggest that there s no evdence that loans amplfy the transmsson of macroeconomc fluctuatons or that a fnancal accelerator va bank lendng exsts. 5

6 Quaglarello (2007) studed the role that macroeconomc uncertanty plays n banks decsons regardng optmal asset allocaton. Usng a portfolo model, the paper nvestgated the determnants of Italan banks choce between loans and rsk-free assets when macroeconomc uncertanty ncreases. The results confrmed that macroeconomc uncertanty s a sgnfcant determnant of banks nvestment decsons, after controllng for bank specfc factors such as nonperformng loans. In perods of ncreasng turmol, banks ablty to accurately forecast future returns s hndered and herdng behavour tends to emerge, as wtnessed by the reducton of the cross-sectonal varance of the share of loans held n portfolo. Somoye and Ilo (2009) nvestgated the mpact of macroeconomc nstablty on bankng sector lendng behavour n Ngera usng data on commercal banks and macroeconomc nstablty from 1986 to The study employed a contegraton and VECM framework to show that bank lendng has a long-run relatonshp wth macroeconomc nstablty. Usng the money supply, exchange rate of the Nara to the US dollar, and the nflaton rate as well as bank specfc control varables, the authors set out to explore the dynamcs of ths relatonshp for the Ngeran economy. Ths study showed that whle ncreases n broad money supply and nflaton nduced banks to curtal lendng, exchange rate deprecaton nduced the ndustry to ncrease lendng n the long-run. Addtonally, the depost moblzaton capacty of banks and bank sze were the most mportant bank characterstcs that explaned ther lendng behavour gven the vagares of the macroeconomc envronment. As t relates to the Jamacan economy, Urquhart (2008) examned the relevance of the bank lendng channel to the conduct of monetary polcy n Jamaca. Usng a GMM approach as proposed by Arellano and Bond (1991), the bank lendng channel of monetary transmsson was estmated. The fndngs showed that the bank lendng channel s mpacted by nformatonal asymmetres that exst between nsttutons. Specfcally, asset sze, captalzaton and lqudty nfluence the magntude of monetary polcy mpact on loans ssued by bankng nsttutons. Addng to the work of Urquhart (2008) whch focused on the relevance of the bank lendng channel to the conduct of monetary polcy, 6

7 ths paper proposes to explore the mpact of macroeconomc uncertanty on bank lendng behavour through the utlzaton of demand varables as well as monetary varables. Addtonally, rather than focusng on the bank lendng channel ths paper wll assess the lendng behavour of banks, that s, how do banks react to macroeconomc uncertanty through loan allocaton. 3.0 Model Selecton Ths paper utlzes the framework developed by Beaudry et al. (2001) and employed by Baum et al. (2005), n whch a model was desgned to descrbe how banks set the optmal composton of ther portfolos. In ths model, bank managers operate n a rsky envronment and, n each perod, can nvest deposts nto ether loans or securtes. In the model, loans to customers ental the exposure to two dfferent sources of rsks: market rsk and default rsk. Market rsk s the rsk that the value of an nvestment wll decrease due to moves n market factors, these market factors may emanate from rsk assocated wth the overall status of the economy, whle default rsk s due to the probablty that the specfc customer wll default n the future wthout repayng the debt. In contrast, securtes are assumed to be free of default rsk, but nvolve some market rsk snce the value of the securtes may change as a result of varyng market condtons. Market rsk s assumed to be more predctable and can be managed and hedged aganst fnancal market and macroeconomc shocks. The return of such an nvestment s therefore assumed to be the rsk free rate (r f ). In any gven perod t, an ndvdual bank that nvest n rsky loans wll earn a rsk free return (r f ) and a rsk premum (r p ). Ths return provded by the loan s known as the stochastc return (r ), whch s equal for all loans assumed to be homogeneous and does not depend on the rskness of the sngle borrower: r r rp (1) f 7

8 The rsk premum s assumed to have an expected value E(rp )=ρ and a varance Var(rp )=σ 2 ε. Thus, the return on rsky loans can be expressed as: r (2) r f where ε s a random component dstrbuted as N(0, σ 2 ε). It s also assumed that each bank has a specfc portfolo wth dfferent rsk structures and, hence, the random components of return across dfferent ntermedares are not correlated (E[ε ε j ] =0). Wthn ths model, banks managers deal wth a portfolo optmzaton problem n whch the composton of ther assets are rearranged n an effort to obtan the preferred combnaton of rsk and expected return. Accordng to ther utlty functons, they have to choose the shares α and (1-α ) of ther assets to nvest n loans and securtes, respectvely. However, before takng the decson, banks observe nether the actual rsk premum nor the random component ε, but only a nosy sgnal of them: S v (3) where ν s a random varable ndependent of ε wth a normal dstrbuton N(0, σ 2 ν). Also, t s assumed that the nose component (ν) of the observed sgnal banks receve s dentcal, whle the overall sgnals reman dfferent across ntermedares because of ε. Addtonally, the cross-sectonal dfferences n the banks prvate nformaton set reman, although all banks are beleved to have the ablty to overcome asymmetrc nformaton problems. In theory, ν may be observed and uncertanty elmnated f all banks would share ther prvate nformaton. However, nformaton sharng s unlkely to hold n the credt market. The nose ν can be nterpreted as the degree of uncertanty on future macroeconomc condtons. Its mpact on all banks s homogeneous, regardless of the managers ablty to predct the random component of loan return ε. In fact, n tmes of greater macroeconomc uncertanty, a hgher varance of ν makes the estmates of the true return 8

9 of loans less accurate. On the contrary, when the macroeconomy s more tranqul, the return from bank lendng wll be more predctable. To determne the expected return on loans (r ), bank managers have to predct the value of ε. Wthout observng the nosy sgnal, a bank s (uncondtonal) forecast of ε would be the mean of ts dstrbuton,.e. zero. However, banks do observe the sgnal and can extract addtonal nformaton from t. The expected value of the return from loans condtonal upon S, E[ S ], s assumed to be a constant proporton (λ) of the sgnal, where λ represents a lnear regresson coeffcent of ε on S : S S (4) where The condtonal expected return of the th bank s portfolo R S s therefore gven by the followng expresson: 1 r r R S rf S 1 r f f f (5) And the condtonal varance Var[ R S ] s: 2 2 Var R S (6) Banks that are rsk averse are assumed to have the followng utlty functon: 9

10 U S R S varr S (7) 2 whch s ncreasng n expected return and decreasng n return volatlty (and s the coeffcent of rsk averson). Employng the portfolo s mean/varance equatons, the optmal loan-to-asset rato for bank and the assocated cross-sectonal dsperson can be derved as: S (8) 2 Var ( ) (9) The varance of the cross-sectonal dstrbuton of the loan-to-asset rato s negatvely correlated to the level of macroeconomc uncertanty 2. Takng the frst dervatve of 2 the varance of wth respect to yelds: Var ( ) < 0 (10) Equaton (10) provdes a testable mplcaton of the hypothess that the cross-sectonal varance of the loan-to-asset rato narrows as macroeconomc uncertanty ncreases. Quaglarello (2006) extended ths model by ncludng a component for bank specfc varables. He assumed that the varance of would wden when the varance of the bank specfc component ncreases. Ths s expressed as: Var > 0 (11) Therefore t s essental to control for ths component when testng for the mpact of macroeconomc uncertanty. 4.0 Data and Emprcal Specfcaton 4.1 Model Specfcaton To nvestgate the relatonshp between macroeconomc uncertanty and bank lendng as outlned n the prevous secton, the followng model wll be tested: 10

11 LTA, t a t t ut (12) where LTA,t s the loan-to-asset rato at tme t; Γ t represents a vector of ndcators of macroeconomc uncertanty evaluated at tme t; X t s the vector of the bank specfc varables and u t s the error term. s the parameters of macroeconomc volatlty factors to be estmated, and β s the parameters of bank specfc factors to be estmated. The LTA ndcates the proporton of the bank s assets represented by loans whch should naturally consttute the major earnng asset of banks and therefore capture lendng behavour. However, ths rato s expected to vary from tme to tme for each bank and across the ndustry dependng on factors that are bank specfc and those that are systemc, especally the macroeconomc factors. The data used to estmate the model consst of seasonally adjusted monthly tme seres data from 1997:01 to 2010:09 for the commercal banks operatng n Jamaca as well macroeconomc varables. The source of the data s the Bank of Jamaca. 4.2 Descrpton of Varables In order to determne the senstvty of bank lendng to macroeconomc uncertanty, bank specfc varables and ndcators of macroeconomc uncertanty are constructed. Consstent wth the lterature, macroeconomc uncertanty s proxed by the standard devaton of the change n the exchange rate of the Jamaca Dollar to the US dollar and the monthly nflaton rate as well as the standard devaton of the 180-day Treasury bll rate. Followng Somoye and Ilo (2009) and Quaglarello (2007) the bank specfc varables that wll be used n ths study are depost to captal rato (D/K) of bank at tme t, nonperformng loans to total loans (NPL), and the Herfndahl (H) ndex. NPL s a measure of the credt/default rsk faced by banks. It assesses the wllngness and ablty of borrowers to repay ther loans. 11

12 D/K shows the extent to whch a bank reles on customer s depost for fundng. The hgher ths rato s the greater the capacty of the banks to offer loans. Banks would normally determne ther optmal loan to captal rato wthn the framework specfed by the central bank gudelne. Ths rato s a measure of rsk and ndcates the level of bank equty exposed to credt rsk. Generally, banks wth hgh equty captal have greater lattude to make huge amount of loans as they are not under serous pressure from captal constrant or regulaton. H s a measure of market concentraton and s calculated as the sum of the squares of market shares for each frm. Essentally, t gauges the degree to whch an ndustry s olgopolstc and the concentraton of market control held by the largest frms n the ndustry. H ranges from a low of 0, ndcatng perfect competton, to a hgh of , ndcatng complete monopoly. Greater values mean greater concentraton, less competton and more market control held by ndvdual frms. For example, a value of to s hgh and sgnfes a tendency towards monopoly, to s medum and 0 to s low. As the H ncreases t sgnfes a tendency towards monopolstc behavor and as such should ncrease the ablty of banks to lend, the reverse s also true Descrptve Statstcs of Bank Performance Measurement and Macroeconomc Uncertanty Indcators n Jamaca Table 1 below presents the descrptve statstcs of each varable used n the study. It also shows the correlaton of each varable wth the loan-to-assets rato. The mean for the loan-to-asset rato for the perod under revew was 32.5 per cent, ndcatng that on average loans comprsed less than half of the commercal banks asset base. The rato of depost-to-captal had a mean of per cent. Ths shows that on average deposts were over fve tmes greater than the captal base of the commercal banks. The table also shows that the mean for the Herfndahl ndex was Ths mean ponts to an ndustry that s hghly concentrated, snce a Herfndahl ndex of and above s 12

13 ndcatve of a hghly concentrated ndustry. Further, a hghly concentrated ndustry s one that exhbts characterstcs of a monopoly. The average monthly nflaton rate and Treasury bll rate were 0.84 per cent and per cent, respectvely. Addtonally, the exchange rate exhbted a very hgh level of varablty as shown by the relatvely hgh value of the standard devaton of per cent. The table also shows the correlaton between the varables and the loans-to-asset rato. It shows that all the varables were moderately correlated wth the loan-to-asset rato, wth the excepton of the Treasury bll rate, whch showed the lowest correlaton of Ths outcome may have emanated from the fact that the majorty of the asset base for the bankng sector s rsk free, or has a low level of rsk. It s mportant to note that the depost-to-captal rato and the Herfndahl ndex are both negatvely correlated to the loan-to-asset rato. However, theory suggests that these varables should have a postve relatonshp wth the loan-to-asset rato. For example, as the depost-to-captal rato ncreases t means that depost s ncreasng or captal s declnng. When deposts ncrease banks are capable of lendng (.e. allocatng more loans) and as such loan-toasset rato should ncrease. The negatve correlaton can be expected f banks have a low captal base. In ths context, ths negatve relatonshp could mply that lendng s beng negatvely affected by captal-constraned banks (See Beatty and Gron (2001)). Table 1: Average Monthly Bank Varables and Macroeconomc Uncertanty (1999: :09) Bank Varables Mean Maxmum Mnmum Standard Devaton Kurtoss Correlaton wth LTA Loans to Asset Depost to Captal Herfndahl Index NPL Inflaton Ex. Rate Tbll Source: Authors Computatons 5.0 Estmaton Technque The methodology utlzed n ths research follows the technque appled by Somoye and Ilo (2009). In ther study the authors utlzed the error-correcton model to capture the long-run relatonshp between the varables. The error-correcton term provdes an 13

14 addtonal channel through whch the mpact of macroeconomc uncertanty on the loanto-asset rato may be assessed. Ths s so because the error-correcton term tells how fast lendng behavour n the bankng system, measured by the loan-to asset-rato, adjusts to equlbrum followng a shock caused by macroeconomc uncertanty. However, gven the order of ntegraton of the varables n ths research, the autoregressve dstrbuted lag (ARDL) approach to contegraton was appled. The ARDL approach deals wth sngle contegraton and s ntroduced orgnally by Pesaran and Shn (1999) and further extended by Pesaran et al. (2001) who showed that the exstence of a level relatonshp between a dependent varable and a set of regressors can be tested when t s not known wth certanty whether the regressors are trend or frst-dfference statonary. They proved that once the order of the ARDL has been determned, OLS may be used for the purpose of estmaton and dentfcaton. The presence of a unque long-run relatonshp s crucal for vald estmaton and nference. Such nferences on long and short-run parameters may be made, provded that the ARDL model s correctly augmented to account for contemporaneous correlatons between the stochastc terms of the data generatng process ncluded n the ARDL estmaton. Hence, ARDL estmaton s possble even where explanatory varables are endogenous. Other econometrc advantages of the ARDL method nclude: () the smultaneous estmaton of long- and short-run parameters of the model; () the nablty to test hypotheses on the estmated coeffcents n the long-run assocated wth the Engle-Granger method are avoded; () all varables are assumed to be endogenous. Whereas other methods of estmaton requre that the varables n a tmeseres regresson equaton are ntegrated of order one,.e., the varables are I(1), only that of Pesaran et al. could be mplemented regardless of whether the underlyng varables are I(0), I(1), or fractonally ntegrated. The ARDL framework s mplemented by modelng equaton 12 as follows: lltasa m 0 a 6 t lhsa a t a m 1 a lstd nf 11 0 a lltasa 12 t 1 m a 0 lstdxr t 7 t m a 0 lstdtbll 2 t a lstd nf 13 ln plsa t a t a lltasa 8 14 t lstdtbll m a ldtksa 0 t 3 a ldtksa 9 t t t a m a ln plsa lhsa t t m a lstdxr 0 5 t (13) 14

15 where a 1 to a 7 represents the short-run coeffcents related to bank lendng behavor, bank specfc varables and macroeconomc uncertanty varables and a 8 to a 14 are the level effects. The long-run coeffcents are computed as 9, a10, a11, a12, a13, a14/ a8 a and represent the speed of adjustment to the long-run relatonshp. The term t s the classcal dsturbance term wth the usual assumptons of zero mean and ndependent, dstrbuton. To nvestgate the presence of a long-run relatonshp amongst the varables of Eq. (13) the bounds testng procedure of Pesaran et al s utlzed. The bounds testng procedure s based on the F or Wald-statstcs, whch has a non-standard dstrbuton. The bounds testng procedure nvolves applyng a jont sgnfcance test that mples no contegraton, that s, H a a a a a a a a 0. 0 : Two sets of crtcal values are computed by Pesaran et al for a gven sgnfcance level. One set assumes that all varables are I(0) and the other set assumes they are all I(1). If the computed F-statstc exceeds the upper crtcal bounds value, then H 0 s rejected. If the F-statstc falls nto the bounds then the test becomes nconclusve. Lastly, f the F- statstc s below the lower crtcal bounds value, t mples no contegraton. 6.0 Results The emprcal analyss begns by examnng the tme seres propertes of the data. The standard Augmented Dckey-Fuller (ADF) test for unt roots (Dckey and Fuller, 1979, 1982) s used. However, the power of the ADF can be sgnfcantly reduced snce t corrects for seral correlaton n the error term by addng lagged values of the frst dfference of the dependent varable. Ths reduced power can be more of an ssue n small samples. As such, the paper also uses the Phllps-Perron, PP, (1988) whch, nstead of addng dfferenced terms as explanatory varables to correct for hgher order seral correlaton, makes the correcton on the t-statstc of the coeffcent of the lagged dependent varable. 15

16 The results from the unt root analyss are presented n Table 2 below. The analyss ndcates that four of the varables can be consdered to be ntegrated of order one, that s, I(1), whle four are statonary I(0). Thus, havng establshed the order of the varables as well as the fact that the dependent varable s I(1), the ARDL method was carred out. Table 2: Unt Root Analyss Varables ADF PP Llta Δllta *** *** ldtk Δldtk *** *** lnpl Δlnpl *** *** lh Δlh *** *** lstdxr *** *** lstdnf *** *** lstdtbll *** *** Notes: *, ** and *** denotes rejecton of the null hypothess at the 10%, 5% and 1% level, respectvely. Δ s the frst dfference operator and L represents the natural logarthm. Havng establshed that the dependent varable as well as the bank specfc varables are I(1) and the macroeconomc uncertanty varables I(0), the ARDL technque s appled to equaton (12). The model was estmated wth thrteen lags and the general-to-specfc approach (Hendry, 1995) utlzed to reduce the model to a parsmonous representaton. Thrteen lags are consdered to be suffcent snce we are workng wth monthly data. Several dagnostc tests are conducted on the fnal model ncludng tests for normalty, seral correlaton, model msspecfcaton, and heteroskedastcty. The results of the ARDL are shown n Table 3, and the results of bounds test s reported n Table 4. The calculated F-statstcs for the model as shown n Table 2 are greater than the upper bound crtcal value at 5% level. Thus, the null hypothess of no contegraton s rejected. Thus, there s a contegraton relatonshp among the varables as presented n Equaton (13). 16

17 Table 3: The Estmated ARDL Model of Bank Lendng Behavour llta = * llta t * llta t *ldtk * ldtk t6 (6.68) (-2.58) (2.97) (-4.41) (-5.23) * t11 ldtk * ln pl * ln pl t *lh * lh t 8 (-2.43) (5.95) (3.46) (5.52) (2.64) * t 10 lh * lh t * lstdxr t * lstd nf t 2 - (2.12) (2.59) (2.16) (2.16) * t4 lstdtbll * llta t * ln pl t 1 (-2.51) (-2.20) (-5.28) Dagnostcs _ 2 R = 0.57 F =12.81 Norm = AR =0.868 ARCH =1.83 [0.000] [0.781] [0.4253] [0.1776] RR = 2.29 HET = DW = 1.98 AIC= SBC= [0.081] [0.2597] Long Run Elastctes of Bank Lendng Behavour: ln pl = Notes: T-statstcs are shown n parentheses. R2 s the fracton of the varance of the dependent varable explaned by the model, F s the F-statstcs for the jont sgnfcance of the explanatory varables, DW s the Durbn Watson statstc, AR s the Lagrange multpler test for p-th order resdual autocorrelaton correlaton, RR = Ramsey test for functonal form ms-specfcaton (square terms only); Norm s the test for normalty of the resduals based on the Jarque-Bera test statstc (χ2 (2)). ARCH s the autoregressve condtonal heteroscedastcty for up to p-th order (see Engle, 1982). HET s the uncondtonal heteroscedastcty test based on the regresson of squared resduals on squared ftted value. Table 4: F-statstc for testng the exstence of a long-run relatonshp for bank lendng Order of lag F-statstc 13 F(2,135) = ** Notes: The relevant crtcal value bounds are obtaned from Table CI() (wth an unrestrcted ntercept and no trend; wth sx regressors) n Pesaran et al. (2001). They are at 90% and at 95%. **denotes that the F-statstc falls above the 95% upper bound. Havng passed all the relevant dagnostc tests, the fnal model of bank lendng behavor n Jamaca s presented n Table 3. Ths model can be taken as an adequate representaton 17

18 of bank lendng behavor n Jamaca, explanng approxmately 56 per cent of the bank lendng behavour over the perod. The fndngs ndcate that whle macroeconomc uncertanty affect bank lendng n the short-term, t has no effect on lendng behavour n the long-run. The presence of a long-run equlbrum relatonshp between bank lendng behavour and ts determnants s confrmed based on the result of the bounds test. The computed F-statstc on the excluson test of the two level varables s , whch exceeds the asymptotc crtcal upper bounds value of 3.61 n Pesaran et al. (2001), Table CI() for the exstence of a contegratng relatonshp. As such, the null of no contegraton relatonshp s rejected at the 5 per cent level of sgnfcance. The coeffcent on the lagged loan-to-asset term, representng the mplct speed of adjustment towards equlbrum, s negatve and hghly sgnfcant and ndcates that approxmately 2 per cent of any devaton from the long-run equlbrum lendng level s corrected each month, or 24 per cent n a year. Thus, t takes approxmately four years for equlbrum to be restored followng a shock to bank lendng. The coeffcent on the lagged change n the loan-to-asset rato mples that a one percentage pont ncrease n bank lendng n a gven month would translate nto a percentage pont ncrease n bank lendng n the followng month. The cumulatve negatve effect of the changes n the depost-to-captal rato mples that a one percentage pont ncrease n the short-run results n a 0.49 percentage pont declne n bank lendng. Ths fndng s contrary to expectaton and s puzzlng. Estmates from the model also suggest that n the short-run, a one per cent change n the rato of non-performng loans to total loans have a postve mpact on bank lendng n Jamaca. Ths most lkely reflects the fact that banks do not respond mmedately to an ncrease n non-performng loans by lowerng lendng. However, the results also ndcate that over tme the mpact of non-performng loans on bank lendng s negatve (-1.3), consstent wth the noton that as the non-performng loans rato rses banks reduce ther loan portfolo, gven that they use ths rato as a sgnal for the rsk of default. 18

19 The estmates also revealed that bank lendng s sgnfcantly nfluenced by the Herfndahl ndex n the short-run but has no effect n the long-run. A one percentage pont ncrease n the Herfndahl ndex (measure of market concentraton) brngs about a 0.24 per cent growth n bank lendng. We note that as the Herfndahl ndex ncreases t sgnfes a tendency towards monopolstc behavour and as such should ncrease the ablty of banks to lend, resultng n an ncrease n the loan-to-asset rato (see Bergstresser (2005)). As t relates to the effect of macroeconomc uncertanty; the results from the short-run models show that the response of bank lendng to short-run shocks due to macroeconomc uncertanty resultng from the exchange rate and nflaton rate stmulates a postve response, whle macroeconomc uncertanty resultng from volatlty n the 180 Treasury bll rate resulted n a negatve response. It s mportant, however, to note that no macroeconomc uncertanty varable had a long-run mpact on bank lendng. Ths suggests that uncertanty about the macro economy does affect bank managers decson regardng lendng n the long run. The results wth regards to the exchange rate and nflaton seem paradoxcal but may reflect the hypothess of Talavera et al (2006) f the bouts of hgh nflaton and hgh rates of deprecaton n Jamaca, concde wth perods of recessons. The estmates show that a one per cent ncrease n uncertanty assocated wth the exchange rate volatlty resulted n a per cent ncrease n the loan-to-asset rato. Ths ndcates that as uncertanty from the exchange rate ncreases n the short-run, banks lendng ncreases. Ths could be as a result of an ncrease n the demand for loan as frms requre addtonal Jamacan dollar to meet the payment for mport of raw materals, machneres, and fnshed goods. As t relates to uncertanty regardng nflaton, a one per cent ncrease resulted n a per cent ncrease n bank lendng. Ths result could be reflectng the fact that n a hgh nflaton envronment demand for credt could ncrease gven the ncentve to purchase real goods and the fact that borrowers tend to gan as aganst agents who save. Uncertanty n the Treasury bll rate has the expected mpact on bank lendng. The estmates also show that a one per cent ncrease n the uncertanty assocated wth the 19

20 nterest rate resulted n a per cent declne n the loan-to-asset rato. Ths s so because as the cost of borrowng ncreases the demand for loans declne and as such bank lendng declnes. In order to control for the possble concdence of recessons wth hgh nflaton and deprecaton rates, a smlar estmaton was conducted ncludng gross domestc product (GDP) (see Table 4). Snce monthly GDP s currently not avalable from the offcal statstcal source, the Statstcal Insttute of Jamaca, the study employed an nterpolaton of the quarterly GDP. The nterpolaton method employed was the quadratc matched average method n Evews, where the quarterly data avalable was ftted by a quadratc polynomal for each observaton and then used to fll n all observatons of the monthly seres. The quadratc polynomal s formed by takng sets of three adjacent ponts from the source seres (two for end-ponts) and fttng a quadratc so that the average of the monthly ponts matches the actual quarterly data. One advantage of the quadratc matched average method s that t mantans the trend of the source data, makng ths an acceptable approxmaton. Havng controlled for aggregate demand, the hypothess that the paradoxcal results for nflaton and exchange rate seems confrmed as the exchange rate and the nflaton rate no longer appears n the results. The standard devaton of the Treasury Bll rate remaned sgnfcant wth very lttle change n magntude. The growth rate n GDP s now the other sgnfcant macroeconomc varable, havng a negatve mpact on bank lendng n the short-run and a postve mpact n the long-run. The estmate shows that a one per cent ncrease n the growth rate n GDP reduces bank lendng by 2.95 per cent n the short-run. Ths result reflects the fact that as aggregate demand ncreases n the short- run frms and ndvduals wll be n a better poston to fnance ther expenses and as such wll not have to borrow as much. In the long-run, a one per cent ncrease n GDP ncreases bank lendng by 5.89 per cent. Ths reflects frms desre to borrow to expand producton stemmng from the ncreased demand due to the growth n economc actvty. More mportantly, an ncrease n demand could be due to sgnfcant mprovements n borrowers' balance sheets, whch n turn are the consequences of hgher collateral values, 20

21 and hgher earnngs. As such, frms can mprove ther balance sheets by ncreasng effectve demand for external fnance, partcularly bank credt n a bank-based fnancal system. The results for the bank specfc varables reman generally unchanged. Table 4:The Estmated ARDL Model of Bank Lendng Behavour Controlled for Demand llta = * llta t * llta t * ldtk t * ldtk t7 (-3.07) (-3.01) (4.63) (-2.98) (-2.31) * t11 ldtk * ldtk t * ldtk t * ln pl *lh (-3.54) (-2.59) (-3.55) (4.40) (2.62) * t 6 lh * lh t * lh t * lh t * lh t 12- (2.94) (2.48) (2.04) (3.41) (2.17) * t4 lstdtbll * lg dpt * lg dpt * lg dpt 13- (-2.29) (-2.42) -2.54) (-2.46) 0.059* t1 llta * ln pl t *lg dp t1 (-3.41) (-2.22) (3.44) Dagnostcs _ 2 R = 0.55 F =9.32 Norm = AR =1.272 ARCH =1.45 [0.000] [0.5038] [0.2837] [0.2298] RR = 2.18 HET = DW = 1.98 AIC= SBC= [0.081] [0.8794] Long Run Elastctes of Bank Lendng Behavour: ln pl = lgdp = 5.89 Notes: T-statstcs are shown n parentheses. R2 s the fracton of the varance of the dependent varable explaned by the model, F s the F-statstcs for the jont sgnfcance of the explanatory varables, DW s the Durbn Watson statstc, AR s the Lagrange multpler test for p-th order resdual autocorrelaton correlaton, RR = Ramsey test for functonal form ms-specfcaton (square terms only); Norm s the test for normalty of the resduals based on the Jarque-Bera test statstc (χ2 (2)). ARCH s the autoregressve condtonal heteroscedastcty for up to p-th order (see Engle, 1982). HET s the uncondtonal heteroscedastcty test based on the regresson of squared resduals on squared ftted value. 21

22 Table 4: F-statstc for testng the exstence of a long-run relatonshp for bank lendng Order of lag F-statstc 13 F(2,128) = 9.359** Notes: The relevant crtcal value bounds are obtaned from Table CI() (wth an unrestrcted ntercept and no trend; wth sx regressors) n Pesaran et al. (2001). They are at 90% and at 95%. **denotes that the F-statstc falls above the 95% upper bound. 7.0 Concluson Ths paper examned the role that macroeconomc varables play n commercal banks decson to allocate loan. Adoptng the man dea of the portfolo model proposed by Baum et al, and applyng the method used by Ilo and Somoye (2004), the paper dscusses how Jamacan banks choose between loans and rsk-free assets when the uncertanty about macroeconomc condtons ncreases. Smlar to the work of Quaglarello (2007) and Ilo & Somoye (2004) the role of dosyncratc factors (.e. bank specfc varables) s taken nto account. The rato of non-performng loan to total loan was found to be the most mportant bank characterstcs that explan ther lendng behavour gven the macroeconomc envronment. The Herfndahl ndex s also mportant because t ndcates that as the bankng ndustry becomes more concentrated, banks ncrease lendng. Evdence from the ARDL contegraton analyss showed that the macroeconomc uncertanty does not have a long-run mpact on bank lendng behavour n Jamaca. Addtonally, the paper shows that uncertanty regardng the exchange rate and the nflaton rate has a postve effect on bank lendng n the short-run. However, uncertanty assocated about nterest rates has a negatve effect. The sgnfcance of concentraton and non-performng loans n the results ponts to the mportance of the cost of nformaton gatherng and rsk assessment n bank lendng behavour. Increase n concentraton mples fewer but larger banks, whch through economes of scale could acqure nformaton and undertake rsk assessment at lower unt costs. The polcy mplcaton of ths s that ntatves such as the establshment of a credt bureau whch would facltate rsk assessment wll boast lendng. 22

23 However n lght of the results for the volatlty of nterest rates, such ntatves that address bank/market specfc factors, have to be complemented by polcy reforms that wll enhance confdence about long-term macroeconomc stablty. Ths polcy strategy should engender less volatlty/uncertanty n nterest rates. 23

24 Bblography Baum, Chrstopher F, Mustafa Caglayan and Neslhan Ozkan, (2005), The Second Moment Matter: The Response of Bank Lendng Behavour to Macroeconomc Uncertanty. Downloaded 15/10/10 Beatty, Anne and Anne Gron (2001), Captal, Portfolo, and Growth: Bank Behavour Under Rsk-Based Captal Gudelnes. Journal of Fnancal Servces research 20:1, 5-3. Beaudry P., M. Caglayann, and F. Schantarell (2001), Monetary Instablty, the Predctablty of Prces, and the Allocaton of Investment: An Emprcal Investgaton Usng UK Panel Data, Amercan Economc Revew, Vol. 91, n.3. Bergstresser, Danel. (2005), Bankng Market Concentraton and Consumer Credt Constrants: Evdence from the 1983 Survey of Consumer Fnances, Harvard Busness School, Workng Paper Drver, C., L. Trapan L and G. Urga (2004), Cross-Secton vs. tme Seres Measures of Uncertanty usng UK Survey Data, Royal Economc Socety Annual Conference Eckmeer, Sandra, Bors Hofmann, Andreason Worm, (2006), Macroeconomc Fluctuatons and Bank Lendng: Evdence for Germany and the Euro Area. Dutche Bundesbank Eurosystem Dscusson Paper Seres: Economc Studes No 34/2006 Enders, Walter. (1995), Appled Econometrc Tme Seres. John Wley & Sons, New York Cty, New York. Granger, C. W. J and P. Newbold, (1974), Spurous n Econometrcs, Journal of Economcs, 2, pp: Granger, C. W. J and P. Newbold, (1974), Spurous n Econometrcs, Journal of Economcs, 2, pp: Green, Wllam H. (2006), Econometrc Analyss, 5th edton, Dorlng Kndersley (Inda) Pvt Ltd. Hamlton, James D., 1994, Tme Seres Analyss, Prnceton Unversty Press, Prnceton, New Jersey Johansen, S. and K. Juselus, (1992), Testng Structural Hypothess n a Multvarate Contegraton Analyss of the PPP and the UIP and for the UK, Journal of Econometrcs, vol.53, pp Pesaran, M. H., Shn, Y., and Smth, R. J. (1996), "Testng for the 'Exstence of a Longrun Relatonshp", Cambrdge Workng Papers n Economcs 9622, Cambrdge UK. Pesaran, M. H. and Y. Shn (1998), "An Autoregressve Dstrbuted Lag Modellng Approach to Contegraton Analyss," Econometrcs and Economc Theory n the 20th 24

25 Century: The Ragnar Frsch Centennal Symposum., S. Strom, ed., Cambrdge Unversty Press: Cambrdge, pp Pesaran, M. H., Shn, Y., and Smth, R. J. (2001), "Bounds Testng Approaches to the Analyss of Level Relatonshps", Journal of Appled Econometrcs, vol. 16, no. 3, pp Quaglarello, M. (2007), Macroeconomc Uncertanty and Banks Lendng Decsons: The case of Italy, Banca d Itala, Tem d Dscussone Satyanath Shanker and Arvnd Subramanan (2004), What Determnes Long-Run Macroeconomc Stablty? Democratc Insttutons. IMF Workng Papers WP/04/215, November pg Somoye, Russsell and Bamdele, Ilo (2009), The mpact of macroeconomc Instablty on the Bankng sector lendng behavour n Ngera, Journal of money Investment an bankng, Issue 7 Talavera Oleksandr; Andry Tsapn and Oleksandr, Zholud (2006), Macroeconomc uncertanty and Bank Lendng: The Case of Ukrane. German Insttute for Economc Research Dscusson Paper Seres 637. November pp.1-24 Urquart, M. (2008), The Reacton of Bank Lendng to Monetary Polcy: The Case of Jamaca, Journal of Busness, Fnance & Economcs n Emergng Economes Vol.3 N0.2 25