Credit Crunch, Bank Lending, and Monetary Policy: A Model of Financial Intermediation with Heterogeneous Projects

Transcription

1 Credt Crunch, Bank Lendng, and Monetary Polcy: A Model of Fnancal Intermedaton wth Heterogeneous Projects Mngwe Yuan and Chrstan Zmmermann Introducton and Motvaton A credt crunch s generally defned as a declne n the supply of credt because, although banks are less wllng to lend, lendng rates do not rse. Accordng to Green and Oh (1991), a credt crunch s an neffcent stuaton n whch credt-worthy borrowers cannot obtan credt at all, or cannot get t at reasonable terms, and lenders show excessve cauton, whch may or may not be traceable to regulatory dstorton, leavng would-be borrowers unable to fund ther nvestment projects. A credt crunch can have several causes, such as regulatory pressures and over-reacton to deteroratng bank asset values and proftablty. If regulatory pressure s the obstacle to credt growth, t should be removed, and credt growth can be restored. But f the crunch s caused by neffcent conservatve lendng by banks, t s an open queston whether easng monetary polcy can help. Ths paper attempts to develop a quanttatve model to address the ssue. * We would lke to thank John Bryant, Walter Engert, Scott Hendry, Davd Ladler, Wenl L, Kevn Moran, Stéphane Pallage, Jack Selody, Shouyong Sh, Robert Townsend, Nel Wallace, Cheng Wang, and the semnar partcpants at the Bank of Canada and the Mdwest Macroeconomcs 1999 Sprng Conference for ther helpful comments and suggestons. 323

2 324 Yuan and Zmmermann A number of papers suggest that a credt crunch exsted n several countres n the early 1990s. From survey data n Germany, Harholf and Kortng (1998) fnd that frms n fnancal dstress face comparatvely hgh lne-of-credt nterest rates and reduced credt avalablty. Usng Fnnsh data from the 1990s, Vhrala (1996) fnds that tghtenng captal regulatons, substantally depletng bank captal, and changng rsk atttudes may explan banks conservatve lendng. Usng a dsequlbrum econometrc model, Pazarbasoglu (1996) also suggests that banks become less wllng to supply credt durng perods of deteroratng asset qualty and reduced profts caused by declnng regulatory protecton from competton and a need to ncrease captal-adequacy levels. In , U.S. banks curtaled ther lendng. Sharpe (1995) clams ths occurred because of losses of bank captal, strngent bank regulatory standards, and heghtened market scrutny of bank captal. The Bank for Internatonal Settlements rsk-based captal standards were phased n begnnng n late 1990 and took full effect n The reduced credt occurred at a tme when banks had dffculty meetng ther mnmum-captal adequacy requrements. Accordng to Bernanke and Lown (1991), the regulatory pressures on bank captal postons would shft the credt supply curve to the left, gven constant real nterest rates and the same qualty of borrowers. Usng New England bank data, Peek and Rosengren (1995) fnd emprcal evdence supportng the hypothess that banks have experenced a captal crunch caused by large captal losses and bndng captal regulatons. An nterestng, recent paper by Wagster (1999) examnes the credt crunch n the Unted States, Canada, and the Unted Kngdom. Four supply-sde credt crunch hypotheses are proposed: () voluntary rsk reducton by bank managers, () tghter scrutny by bank regulators, () rsk-based captal requrements followng the 1988 Basle Accord, and (v) the unweghted captal rato mposed by U.S. bank regulators. The results show that none of the four hypotheses can be elmnated as explanatons for the U.S. credt crunch. For Canada, the results conform to the rsk-based captal requrement and tghter regulatory scrutny hypotheses. For the Unted Kngdom, tghter regulatory scrutny s the explanaton for the crunch. The recent credt crunch n Japan has drawn a lot of attenton from researchers and polcy-makers. The banks captal postons have been eroded by deteroratng loan qualty and losses of banks securty holdngs. Furthermore, banks were reluctant to ssue new equtes because of unfavourable market condtons. To meet the mnmum-captal adequacy requrement, banks had to reduce ther loans. Ths pushed margnal borrowers nto bankruptcy, puttng more pressure on the banks captal poston. Woo s (1999) emprcal results based on the data suggest a credt

3 Credt Crunch, Bank Lendng, and Monetary Polcy 325 crunch n 1997, but not durng most of the rest of the 1990s. The Bank of Japan and the Japanese government have taken measures to deal wth the crunch, such as lowerng nomnal short-term nterest rates to near zero, njectng captal nto several commercal banks, and relaxng captal adequacy requrements by way of accountng changes. What can monetary polcy do to help durng a credt crunch when reduced bank lendng leads to a shrnkng economy? The queston s related to the transmsson mechansm of monetary polcy. One transmsson channel of monetary polcy s credt. Expansons and contractons of credt affect both aggregate demand and aggregate supply, thus nfluencng aggregate actvtes and prces. The credt channel of monetary polcy has been drawng much attenton from researchers recently. In the credt-channel stores, fnancal market mperfecton s an mportant factor of the transmsson mechansm a fnancal accelerator effect. Because of frctons n credt markets, dfferent fnancng sources are mperfectly substtutable, especally for bank credt. Gven asymmetrc nformaton n fnancal markets, the mpact of monetary polcy wll be amplfed and dstrbuted among borrowers through two channels: the lendng channel and the balance sheet channel. The lendng channel (Bernanke and Blnder 1988) works through the supply sde of credt. A tght polcy reduces the supply of bank loans, whch n turn reduces spendng by borrowers that depend on banks. The balance sheet channel (Bernanke and Gertler 1995) works through the demand sde of credt. A tght polcy weakens borrowers balance sheets, whch reduces ther ablty to borrow. A shrnkng loan supply leads to a rsng lendng rate, whch would dscourage bank-dependent borrowers actvty. Instead of rasng the nterest rate on loans, banks can also raton credt. Banks raton credt rather than rase nterest rates because of asymmetrc nformaton between borrowers and lenders. The problems of moral hazard and adverse selecton can lead to credt ratonng (Keeton 1979, Stgltz and Wess 1981). In addton, Wllamson (1986) develops a model n whch credt ratonng arses from montorng cost. Several papers provde emprcal evdence on credt channels of monetary polcy. Banks lend more to large frms and less to small ones after a tghtenng polcy (Gertler and Glchrst 1994, L 1997). Lang and Nakamura (1995) also fnd that banks make proportonally more safe loans durng fnancal dstress. Morgan (1998) uses a contractual dfference across commercal bank loans to test for credt effects. Most commercal banks ssue loans under a commtment contract, for example, revolvng lnes of credt. Ths knd of contract sets the terms of future lendng n advance, whch would prevent a sudden contracton n the supply of loans under commtment. However, banks have a choce to reduce loans not under

4 326 Yuan and Zmmermann commtment. Thus, by comparng the supples of these two knds of loans, one can examne credt effects of monetary polcy. Morgan uses two monetary polcy ndcators, Romer and Romer dates, and the federal fund rate. The results show that a tght monetary polcy slows the growth of loans not under commtment, whle the growth of loans under commtment s unaffected or accelerates. By further examnng the responses of bank loan offcers and small frms to survey questons about the avalablty of credt, Morgan fnds that the dvergence reflects a shrnkng loan supply due to credt effects. Frctons n fnancal markets are thought to be an mportant channel for propagatng monetary polcy shocks. Studes of the credt channel by Fuerst (1995) and Fsher (1996) suggest that t s quanttatvely unmportant. However, Bernanke, Gertler, and Glchrst (1998) develop a model by consderng external fnance premums and the net worth of borrowers. The model shows that credt market frctons may sgnfcantly amplfy both real and nomnal shocks to the economy. Cooley and Quadrn (1998) consder a model n whch frms fnancal decsons dffer systematcally wth the amount of equty they ssue. The captal structure of frms changes endogenously over tme as a result of ther fnancal decsons. The model provdes nsght on the mportance of frm heterogenety as a channel of transmsson for monetary polcy. Nevertheless, Cooley and Quadrn fnd the aggregate real effects of monetary shocks to be very small, smlar to the emprcal result of Sms (1992) and Leeper, Sms, and Zha (1996). Several papers dscuss credt-crunch models. Peek and Rosengren (1995) provde a smple one-perod model of banks. Banks face bndng captal constrants as a result of large loan losses and low or no earnngs. Ths causes banks to behave dfferently than they would f the requrements were not bndng. Chan-Lau and Chen (1998) consder a model of prvatedebt fnancng under neffcent fnancal ntermedaton, where neffcency s characterzed by costly loan montorng. They use the model to study large captal outflow and credt crunches n Asan countres n recent years. In ths paper, we develop a dynamc heterogeneous-agent model to study credt crunches and the effectveness of monetary polcy. In ths model, a credt crunch s caused by banks conservatve lendng behavour durng perods of reduced proftablty and fnancal dstress. Banks collect deposts from households, and provde loans to frms accordng to ther sze. Banks equate expenses and revenues. The depost rates are controlled by the central bank, and the lendng rates are chosen by the banks. To accommodate a credt crunch, the banks can allocate ther assets to loans and rsk-free fxed-ncome securtes. We assume that banks have a rskmanagement practce that s lnked to a benchmark loss/depost rato. Banks n ths model have no captal, so mnmum-captal adequacy requrements

5 Credt Crunch, Bank Lendng, and Monetary Polcy 327 cannot be appled here. The benchmark loss/depost rato s derved from a benchmark case the normal state of the economy. Ths rato s bndng when the economy s n a bad state n terms of returns. The model s able to generate a credt crunch such that banks wll reallocate balance-sheet assets from loans to government securtes and cause a reducton n lendng actvty. The model s then used to analyze monetary polcy. In our frst experment, we examne the effectveness of monetary polcy durng a credt crunch. Gven a half per cent decrease n depost rates, bank lendng rates are lowered by 40 bass ponts. Banks lend to some smaller frms, but they stll hold a large proporton of ther assets n government securtes. In a second experment, we allow banks to relax the loss/depost rato so that banks have the same bond/depost rato as n the benchmark case. We fnd that the loss/depost rato must be doubled. These results may mply that f a credt crunch s caused by regulatons or by a too-conservatve rskmanagement practce, monetary polcy has very lmted effects compared wth relaxng the regulatons or banks rsk-management practces. In Secton 1, we descrbe the model. In Secton 2, the model s calbrated to the Canadan economy. In Secton 3, we use ths model to generate a credt crunch and analyze the mplcatons for monetary polcy. In the last secton, we offer conclusons. 1 The Model In ths secton, we develop a general-equlbrum model of fnancal ntermedaton. 1 There are three agents n ths economy: households, banks, and a central bank. Households are endowed wth a number of nvestment projects n each perod. Because they are short of nternal funds, households must apply to banks for external fnancng. The banks screen loan applcatons accordng to the applcants net worth. Successful applcants become entrepreneurs, and others become workers. Each worker also faces an dosyncratc shock for unemployment. Employed workers get labour ncome, and unemployed workers receve unemployment benefts. Banks collect deposts and provde loans to entrepreneurs or purchase rsk-free government bonds. Banks have a rsk-management lendng polcy that may cause conservatve lendng n perods of reduced proftablty and fnancal dstress. The central bank has control over depost nterest rates, and adjustng depost rates wll nfluence banks lendng behavour. 1. The current model does not have aggregate shocks. Further extensons that nclude ths type of shock may allow us to examne credt crunches over the busness cycle.

6 328 Yuan and Zmmermann 1.1 Households Households have a fnte lfe expectancy. We denote the net worth of household at perod t as m t. In each perod, a household s endowed wth j j n nvestment projects, x t, j = 12,,, n. Total nvestment s x t ( =Σ j x t ), and the nvestment for each project s x t n. The sze of total nvestment s a fxed multple φ of the current net worth of the household, m t. If φ s above 1, then the household needs external fnancng. To smplfy the problem, we assume that projects always need external fnancng: x t = wth φ > 1. φm t The returns on nvestment projects,, are assumed to follow a j normal dstrbuton wth mean r and standard devaton σ r. r t s drawn ndependently across all projects and ndvduals. Also, the returns face a postve probablty of beng suffcently negatve such that the bank faces losses from lendng to some projects. Households apply to banks for the extra funds needed to fnance ther nvestment projects. If loans applcatons are accepted by the bank, the household becomes an entrepreneur. The entrepreneur undertakes projects and receves the project returns. If the loan applcaton s rejected, the household may become employed or unemployed, wth gven probabltes. Assume u to be the probablty of beng unemployed. The employed worker provdes labour and receves labour ncome, y. The unemployed worker receves unemployment nsurance benefts, θy, where θ s the replacement rato. The remanng net worth s carred over to the next perod. In addton, beng ether an entrepreneur or a worker, each household faces a probablty τ of forced retrement. After retrement, the household can earn ncome only from deposts and pensons. Fnally, retrees face a probablty δ of dyng. They are then replaced by new agents wth no assets. Remanng assets are lost (there are no bequests n the model). The momentary utlty functon for workers, entrepreneurs, and retrees s U oc ( c) = ( ξ oc c 1 ρ ) 1 ρ , 1 ρ where oc s occupatonal choces and the set of oc s {W, U, E, R}, n whch W, U, E, and R are employed workers, unemployed workers, entrepreneurs, and retrees, respectvely. c s consumpton, and σ and ρ are postve parameters. ξ W, ξ U, ξ E, and ξ R > 0. Now we descrbe the value functons for each agent. We denote V W, V U, V E, and V R as the value functons of workers, the unemployed, r t j

7 Credt Crunch, Bank Lendng, and Monetary Polcy 329 entrepreneurs, and retrees. m* s the mnmum net worth elgble for external fnancng. A worker wth net worth m ( < m ) n the current perod has probablty ( 1 u) of beng employed. 2 Once employed, the worker gets labour ncome y, deposts hs or her net worth n banks, and receves nterest ncome, R d m. The total ncome s then allocated to consumpton, c, and savngs, m'. If unemployed, the worker receves unemployment benefts, θy, and also deposts hs or her net worth n banks. The total resources are also allocated to consumpton, c, and savngs, m'. In the followng perod, the worker faces a probablty τ, of retrng. Otherwse, the worker may become an entrepreneur or reman a worker, dependng on hs or her net worth, m'. For a worker, subject to + uv U ( m' ) + E r, V E ( m', r' )] + τv R ( m' )]}, (1) + m' = ( 1 + R d )m+ y, V W ( m) = 0 f m m. For an unemployed worker, subject to V W ( m) = max { c W, m' } {UW ( c W ) + β[1 ( τ)[1 ( u)v W ( m' ) c W V U ( m) = max U {U U ( c U ) + β[1 ( τ)[1 ( u)v W ( m' ) { c, m' } c U + uv U ( m' ) + E r, V E ( m', r' )] + τv R ( m' )]}, (2) + m' = ( 1 + R d )m + θy, V U ( m) = 0 f m m. An entrepreneur nvests n n projects and gets return r j for project x j for j = 12,,, n. He or she also receves labour ncome, y, and pays the borrowng cost, R l ( x m) ; what s left can be consumed or saved. The net wealth s constraned to be non-negatve (ths s a lqudty constrant). Project losses may, however, drve net wealth down to negatve numbers, n whch case the household consumes a mnmal consumpton allowance, 2. Snce we use a yearly model and the duraton of unemployment s below one year, u s not state-dependent.

8 330 Yuan and Zmmermann (funded by the banks), declares bankruptcy, and defaults on ts debt. 3 In that case, the household starts next perod wth no assets and no labltes. In the followng perod, the entrepreneur has a probablty τ of retrng. c mn subject to V E ( mr, ) = max { cm', } {U E ( c) + β[1 ( τ)[1 ( u)v W ( m' ) + uv U ( m' ) + E r, V E ( m', r' )] + τv R ( m' )]}, (3) c max c mn m y ( 1 + r j )x j R l =, + + ( m) m', n x j j = 1 = φm, V E ( mr, ) = 0 f m < m. j = 1 Also, V E, V W, and V U have to satsfy a partcpaton constrant, n E r V E ( m, r) ( 1 u)v W ( m ) + uv U ( m ), (4) whch mples that, at the margn, the household would prefer to be an entrepreneur (applyng for a loan) rather than a worker. Once a household retres, t receves penson ncome, y R. The net worth and penson ncome are allocated to consumpton and savngs. In the followng perod, the household has a probablty δ of dyng. V R ( m) = max { cm', } { U R ( c) + β[ ( 1 δ)v R ( m' )]}, (5) subject to c+ m' = ( 1 + R d )m+ y R. One aspect of credt crunches that has been largely gnored n the lterature s heterogenety among both nvestment projects submtted and the methods used by banks to screen them. In ths model, we ntroduce heterogenety n several ways: ndvdual unemployment rsk; rsk of project returns; and rsk of the lfe cycle (retrement, death, and brth). Households are then heterogeneous along several dmensons: actve or retred; 3. There s no moral hazard problem here, snce banks constantly montor the returns.

9 Credt Crunch, Bank Lendng, and Monetary Polcy 331 entrepreneur; worker or unemployed; and, most mportantly for our queston, wealth. In ths model, households have several motvatons to accumulate assets: to save for retrement; to maxmze returns through nvestment projects nstead of deposts [partcpaton constrant, constrant (4)] and to protect themselves aganst unemployment shocks that are mperfectly nsured. Banks take the bankruptcy rsk of ther clents nto account when they make loan decsons, so we need bankruptcy n ths model. But bankruptcy s very tough on households: In one bad perod, they lose all ther accumulated assets and may, for example, jeopardze ther retrement assets. For ths reason, we have ntroduced the possblty of nvestng n several projects at a tme, whch spreads the rsk. 1.2 Banks The bank n our model accepts deposts from workers and retrees, and makes loans to entrepreneurs to fnance nvestment projects. The depost nterest rate, R d, s consdered to be gven by monetary polcy. For projects that lose money, the bank pays an audtng fee, µ, to recover the salvage value of the project. The bank chooses m, and the nterest rate for loans, R l, such that () dsbursements for deposts equal revenue from loans: d l R t mt = E r R t ( x t m t ) L t m t < m m t m m t m n whch losses are defned as: d + R t max 0, mt m t < m m t m ( x t m t ), (6) L t = l max{ 0, ( 1 + µ )[1 ( + R t )( x t m t ) x t ( 1 + r t )] + g t }, where represents addtonal losses ncurred f a household has to be granted ts mnmal consumpton; () expected losses ncurred from loans do not exceed a certan share, α, of the deposts, g t E r L t m t m α m t m t < m ; (7)

10 332 Yuan and Zmmermann () banks cannot lend more than the value of deposts they have accepted, m t m ( x t m t ) m t. (8) Through the households choces of m and banks choce of m, the total quantty of loans s determned endogenously. The left sde of the equaton represents the loans, and the rght sde of the equaton represents the deposts that can be used for loans or bonds. Bonds represent safe assets used by the banks to reduce the loss/depost rato. The loss/depost rato α can be nterpreted n two ways. Frst, t can be lnked to the nternal rsk management (a rule of thumb) and the value at rsk (VaR). Second, t can be seen as an external regulatory constrant. Such a constrant s often related to captal-adequacy requrements. However, n ths model banks have no captal. Nevertheless, the loss/depost rato can be converted to a loss/captal rato. Banks screen projects accordng to borrowers net worth. Because nvestment projects earn sgnfcantly hgher expected returns than do deposts, each household would lke to be approved for a loan, whch s necessary to run a project. Gven that projects, and therefore loans, are proportonal to the collateral provded by the household (ts wealth), banks use household wealth to screen projects: Only projects backed up by collateral above a certan threshold are approved. Other observatons can back up ths hypothess: Many households have lttle wealth, and do not nvest t n assets other than deposts. They may want to nvest n proftable projects, but they know they wll not be approved for credt. Also, when banks tghten credt rules, small busnesses are most lkely to suffer. 1.3 The central bank m t < m The central bank s only nfluence s through settng the nterest rate on deposts, R d, whch can be vewed as the stance of monetary polcy and/or economc condtons. How does the central bank affect bank lendng behavour? A prvate bank s lendng polcy nvolves two optons: the cutoff pont, m, for loans and the lendng rate, R l. If the central bank lowers the depost nterest rate, how would banks change ther lendng polcy? If the banks fx one of these two optons, then the answer s straghtforward. When the lendng cutoff pont, m, s held fxed, a declne n banks depost cost wll lower the lendng rate accordngly, to equalze costs and revenues. Smlarly, f the

11 Credt Crunch, Bank Lendng, and Monetary Polcy 333 bank holds lendng rates fxed, then profts wll ncrease, and banks could lower m. Moreover, f banks can adjust both varables, t would be dffcult to determne the analytcal consequences, because both R l and m could be lowered or ncreased. In the next two sectons, we attempt a quanttatve analyss based on a calbrated model to address ths queston. 1.4 Computng steady-state equlbrum In descrbng the computatonal procedure used to solve the model, we frst defne the equlbrum concept and then outlne a procedure for fndng the equlbrum. Let z be the set of parameters that descrbe the economy, a set that can be altered from the benchmark for computatonal experments. An equlbrum s a collecton of a loan nterest rate, R l ( z), a loan threshold, m ( z), a law of moton of the agents n the economy, λ' = g( λ; z), and decson rules for households, m W ( mz ; ), m U ( mz ; ), m E ( mrz, ; ), and m R ( mz ; ), such that: () households solve ther optmzaton problems shown n equatons (1), (2), and (3); () the banks break even, equaton (6). Ther loss/depost rato does not exceed the benchmark value α, constrant (7). And they cannot lend more than deposts accepted, constrant (8). The partcpaton constrant, constrant (4), s also satsfed. To solve for ths equlbrum, we defne a grd for all household types (worker, unemployed, entrepreneur, retred), asset levels, and project returns. For ntal values of R l and m, we computed utltes for all feasble decsons, and then calculated teratvely the value functons. Once optmal decson rules are determned, they are used, along wth the laws of moton, to fnd the nvarant dstrbutons of households across types, asset levels, and project returns. Gven ths dstrbuton, the banks equlbrum constrants are evaluated. If they are not wthn a reasonable range, R l and m are modfed and the whole procedure starts agan. Once everythng has converged, we draw aggregate statstcs. 2 Calbraton: The Benchmark State The model s calbrated to the Canadan economy accordng to annual data from 1988 to 1992, a perod for whch we have the dstrbuton data of the return on equty (Statstcs Canada 1994). Followng the lterature, ρ and β are set to 2.5 and 0.96, respectvely. σ s set to ξ E, ξ W, ξ U, and ξ R are set to be consstent wth standard models wth explct lesure specfcaton,

12 334 Yuan and Zmmermann.e., ξ E = ξ W = ξ U = , and ξ R = 1, whch mples that the workng hours of employed workers and entrepreneurs and the search effort of the unemployed are set to The labour ncome of workers and entrepreneurs, y, and the retrement ncome, y R, are set at 1 and 0.3, respectvely. The probablty of retrement, τ, and the mortalty rate of retrees, δ, are set at 5 per cent and 10 per cent, respectvely. Calculated from the real Guaranteed Investment Certfcate rates and the real savngs depost rates, the depost rate, R d, s set at 1.00 per cent. The unemployment nsurance beneft s per cent of wage rates. 4 Mnmum consumpton s assumed to be the same as the unemployment beneft. The nvestment/net worth rato, φ, s calculated from the debt/equty (D/E) rato. The average D/E rato durng the reference perod s 1.2, whch mples φ to be 2.2. The audtng fee, µ, s set at 6 per cent. Statstcs Canada (1994) reports the dstrbuton of return on equty for non-fnancal enterprses from 1988Q4 to 1992Q4. For each quarter, the average returns on equty are reported for the top, mddle, and bottom tertles. For our model, we also use dscrete-return dstrbuton to smplfy computaton. The returns of projects are calculated from the return-onequty data of frms by assumng normal dstrbutons. Durng the perod, ths dstrbuton mples dscrete returns of r = { 50 per cent, 4.52 per cent, 60 per cent} wth probabltes of {0.0098, , }, respectvely. The lowest return s chosen to generate occasonal bankruptces, the mddle one to match the average return. The lendng rate, R l, and the asset for the mnmum qualfed frm, m, are chosen such that banks equlbrate ther ncome and dsbursements and the bond/depost rato s close to the average of the bond-holdng ratos over the reference perod. The lendng rate, R l, s calculated to be 1.18 per cent, and m to be In the benchmark steady state, the number of workers s 0.58, the number of entrepreneurs s 0.09, and the number of retrees s Wth respect to the wealth dstrbuton, the benchmark economy has a Gn coeffcent of In ths benchmark model, there are some loan losses for banks because of the possble negatve returns. The loss/depost rato s found to be per cent. We use ths rato as a benchmark for banks lendng polcy. It turns out that, durng perods of reduced return, ths rato becomes bndng, 4. The measure s based on the replacement rate constructed n Hornsten and Yuan (1999). The followng elements of unemployment nsurance (UI) benefts are consdered: the legslated replacement rate, the percentage of the labour force covered by UI (the coverage rate), the maxmum number of beneft weeks for a mnmally qualfed clamant, and the mnmum number of workng weeks needed to qualfy for UI.

13 Credt Crunch, Bank Lendng, and Monetary Polcy 335 whch leads to banks conservatve lendng behavour and causes a credt crunch. 3 Credt Crunch, Bank Lendng Behavour, and Monetary Polcy We now use ths model to generate a credt crunch. The return-on-equty data show that frms returns n 1992Q4 are less than the average of the reference perod. We calculate the annualzed return dstrbuton accordng to 1992Q4. Our calculaton suggests the returns of projects to be r ={ 50 per cent, 2.57 per cent, 60 per cent} wth probabltes {0.0179, , }. The unemployment rate n the reference perod s per cent, and φ s 2.3. The depost rate s set to be dentcal to the benchmark case, 1.00 per cent, so that we can examne the banks lendng behavour durng perods when banks see reduced proftablty, but the central bank does not change ts monetary polcy stance. To generate a possble credt crunch, we choose m and R l, such that the bank loss/depost rato matches the benchmark case, per cent, and banks equlbrate revenues and expenses. In equlbrum, we fnd that the last qualfed frm has asset m of 6.84; that sze s larger than that of the benchmark case, 6.76 (see Table 1). Ths suggests that the banks have reduced loans to smaller frms. The total loans are reduced dramatcally, by more than 60 per cent n the credt-crunch state, compared wth the benchmark state. The lendng rate s calculated to be 1.30 per cent, whch s hgher than the benchmark lendng rate, 1.18 per cent. In the dstrbuton of bank assets, we see a bg ncrease n the rato of bank deposts allocated to bonds. The rato ncreases to 58 per cent compared wth the benchmark result of 13 per cent. The restrcton of the loss/depost rato effectvely mmcs the bank s conservatve lendng behavour. Ths avods explctly modellng changes n bank rsk-averson n lendng under dfferent economc states. We also fnd that more people become workers (0.64 compared wth 0.58) and fewer become entrepreneurs (0.03 compared wth 0.09). The Gn coeffcent decreases to 0.62, compared wth the benchmark case, 0.65, and the average utlty decreases from 0.23 to 0.26 durng ths credt crunch. In ths state, bankruptces occur more often because projects post lower returns and lendng rates are hgher. Banks are ted by ther loan strategy and, as losses ncrease as a share of loans, banks must reduce the quantty of loans. As m s ncreased, some entrepreneurs become workers, whch drves down the quantty of loans. The decrease n the number of loans s partally offset by the ncrease n the average sze of the remanng

14 336 Yuan and Zmmermann Table 1 Steady-state analyss Benchmark Credt crunch Monetary polcy Lendng polcy Depost rate, R d Return dstrbuton Cut-off pont, m* Lendng rate, R l Bonds/depost (%) Loss/depost, α (%) Total loans Total deposts Number of workers Number of entrepreneurs Number of retrees Gn coeffcent Average utlty loans. Ths process contnues untl the banks fnd the m that yelds the orgnal loss/depost rato. Along the way, the banks must also modfy the loan rate to recuperate the hgher costs per loan due to losses. 3.1 Steady-state polcy experments We now carry out a steady-state monetary polcy experment, n whch the central bank lowers the depost rate n the credt-crunch state. In ths experment, R d s reduced by one-half per cent. We examne the changes n the banks lendng behavour,.e., how banks change R l and m. Banks choose R l and m, such that revenues equal expenses and the loss/depost rato matches the benchmark case (0.166 per cent). Gven a one-half per cent decrease n the depost rate, bank lendng rates are lowered by 40 bass ponts to 0.90 per cent. Banks ncrease ther lendng to small frms by reducng the cut-off pont for lendng, m, to 6.55 from the credt crunch level of The bond-holdng rato remans the same as n the crunch case. The total depost and total loans are reduced slghtly. Overall, the effects of reduced depost rates are qute lmted. Reducng nterest rates s largely neffectve, because the banks are mostly constraned by ther loss/depost rato, α. However, the nterest rate change has varous effects. Lower depost rates gve workers less ncentve to save deposts. However, the banks have fewer nterest payments to make, so they can reduce the loan rate to balance ther books, thereby nctng some workers to opt for entrepreneurshp. But all these effects are small.

15 Credt Crunch, Bank Lendng, and Monetary Polcy 337 To evaluate the relatve effectveness of ths polcy change, we conducted another experment, n whch banks relax ther loss-control rule,.e., the loss/depost rato, α, s ncreased. Banks choose m and R l, such that revenues equal expenses and the bond/depost rato matches the benchmark case (13 per cent) as closely as possble. The last column of Table 1 shows that once the loss/depost rato α s ncreased to per cent, about double the rato n the benchmark case, the bond/depost rato s reduced to 14 per cent, whch s close to the value n the benchmark case. Compared wth the credt-crunch state, more people become entrepreneurs and fewer become workers. The average utlty s restored to the same level as n the benchmark case. Ths polcy s much more effectve, because t relaxes the most bndng constrant on the banks. Indeed, although the loss/depost rato, α, ncreases, the banks are stll able to balance ther books wth only a slght ncrease n loan rates. The lowerng of m nctes some workers to become entrepreneurs, and the mpact of lower returns largely dsappears: Whle projects have a lower average return than n the benchmark wthout a credt crunch they stll have a hgher return than do deposts. Wth more projects, the returns realzed n the economy as a whole ncrease. Conclusons In ths paper we have developed a dynamc general-equlbrum model of fnancal ntermedaton that s nnovatve n that t takes account of the heterogenety of nvestment projects. Furthermore, the decson to become a loan-contractng entrepreneur s endogenous, and workers decde on deposts that the bank can use for loans and bonds. Gven that projects and even ndvduals can go bankrupt n ths model economy, banks take explctly nto account possble losses from loans. Ths self-regulaton, whch has recently been mposed on the banks by the Basle Accord, appears to be the determnng factor n a credt crunch. Calbrated for the Canadan economy, our model fnds that monetary polcy can do lttle to ease a credt crunch that arses because of ncreasng loan rsk. However, more flexble loans regulaton, n partcular, rules that allow the banks to take more losses that are compensated by hgher loan rates, s very effectve. References Bernanke, B.S. and A. Blnder Credt, Money, and Aggregate Demand. Amercan Economc Revew 78 (2): The Federal Funds Rate and the Channels of Monetary Transmsson. Amercan Economc Revew 82 (4): Bernanke, B.S. and M. Gertler Insde the Black Box: The Credt Channel of Monetary Polcy Transmsson. Journal of Economc Perspectves 9 (4):

16 338 Yuan and Zmmermann Bernanke, B.S., M. Gertler, and S. Glchrst The Fnancal Accelerator n a Quanttatve Busness Cycle Framework. NBER Workng Paper No Bernanke, B.S. and C.S. Lown The Credt Crunch. Brookngs Papers on Economc Actvty 2: Canada. Statstcs Canada Quarterly Fnancal Statstcs for Enterprses, Thrd Quarter Catalogue No Chan-Lau, J.A. and Z. Chen Fnancal Crss and Credt Crunch as a Result of Ineffcent Fnancal Intermedaton wth Reference to the Asan Fnancal Crss. Internatonal Monetary Fund Workng Paper No. 98/127. Cooley, T.F. and V. Quadrn Monetary Polcy and the Fnancal Decsons of Frms. Photocopy. Fsher, J.D.M Credt Market Imperfectons and the Heterogeneous Responses of Frms to Monetary Shocks. Federal Reserve Bank of Chcago Workng Paper No Fuerst, T.S Monetary and Fnancal Interactons n the Busness Cycle. Journal of Money, Credt and Bankng 27 (4) Part 2: Gertler, M. and S. Glchrst Monetary Polcy, Busness Cycles, and the Behavor of Small Manufacturng Frms. Quarterly Journal of Economcs 109 (2): Green, E. and S.N. Oh Can a Credt Crunch Be Effcent? Federal Reserve Bank of Mnneapols Quarterly Revew 15 (4): Harholf, D. and T. Kortng Lendng Relatonshps n Germany: Emprcal Results from Survey Data. CEPR Dscusson Paper No Hornsten, A. and M. Yuan Can a Matchng Model Explan the Long-Run Increase n Canada s Unemployment Rate? Canadan Journal of Economcs 32 (4): Keeton, W.R Equlbrum Credt Ratonng. New York: Garland Publshng. Lang, W.W. and L. Nakamura Flght to Qualty n Bank Lendng and Economc Actvty. Journal of Monetary Economcs 36 (1): L, W Captal Constrants and Frm Heterogenety over Busness Cycles. Photocopy. Leeper, E.M., C.A. Sms, and T. Zha What Does Monetary Polcy Do? Brookngs Papers on Economc Actvty 2: Morgan, D.P The Credt Effects of Monetary Polcy: Evdence Usng Loan Commtments. Journal of Money, Credt and Bankng 30 (1): Pazarbasoglu, C A Credt Crunch? A Case Study of Fnland n the Aftermath of the Bankng Crss. Internatonal Monetary Fund Workng Paper No. 96/135. Peek, J. and E. Rosengren The Captal Crunch: Nether a Borrower Nor a Lender Be. Journal of Money, Credt and Bankng 27 (3): Sharpe, S.A Bank Captalzaton, Regulatons, and the Credt Crunch: A Crtcal Revew of the Research Fndngs. Fnance and Economcs Dscusson Seres, No Washngton: Federal Reserve Board. Sms, C.A Interpretng the Macroeconomcs Tme Seres Facts: The Effects of Monetary Polcy. European Economc Revew 36 (5): Stgltz, J. and A. Wess Credt Ratonng n Markets wth Imperfect Informaton. Amercan Economc Revew 71 (3): Vhrala, V Credt Crunch and Collateral Squeeze? An Emprcal Study of the Credt Supply of the Fnnsh Local Banks n Bank of Fnland Workng Paper No. 11/96. Wagster, J.D The Basle Accord of 1988 and the Internatonal Credt Crunch of Journal of Fnancal Servces Research 15 (2): Wllamson, S.D Costly Montorng, Fnancal Intermedaton, and Equlbrum Credt Ratonng. Journal of Monetary Economcs 18 (1): Woo, D In Search of Captal Crunch : Supply Factors Behnd the Credt Slowdown n Japan. Internatonal Monetary Fund Workng Paper No. 99/3.

17 Dscusson Emanuela Carda Yuan and Zmmermann construct a model that can generate credt crunches,.e., neffcent conservatve lendng, and examne whether monetary polcy, loan regulaton, or both, can reduce the mpact of credt crunches. They fnd that, whle loan regulaton can elmnate the effects of a credt crunch, monetary polcy can do lttle to allevate the credt crunch tself. The Yuan and Zmmermann paper s an nterestng contrbuton to the lterature on credt crunches. Whle several emprcal papers have suggested that n the early 1990s Germany, Fnland, the Unted States, and Japan were constraned by neffcent conservatve lendng, few have tred to develop a theoretcal framework n whch to examne ths ssue. The authors use a general-equlbrum model wth several types of households and heterogeneous projects. The most mportant type of heterogenety comes from assumng that households can be workers f ther net assets, m, are smaller than a certan crtcal level, m*, a choce varable of the bankng system, and that households can be entrepreneurs f m m*. A household wll ask for lendng f the desred project mples outlays greater than ts net assets. Only f m m* wll the household see the project approved and become an entrepreneur. Ths set-up s essental for creatng the possblty of credt crunches. The authors also assume that households retre wth a fxed probablty, τ, can be unemployed wth a fxed probablty, 1 µ, and have a fnte lfe expectancy. However, Yuan and Zmmermann do not explan how these addtonal sources of heterogenety are essental to the study of credt crunches. From Table 1, I see that the number of retrees does not change across all experments. Perhaps we would lose nothng by omttng ths feature. If ths s the case, I thnk that t would be best to omt 339

18 340 Dscusson: Carda retrement n the model and add a footnote to explan that nothng would change f we were to assume that households can retre. The authors should dscuss the channel n ther model through whch credt crunches may affect unemployment, and they should also report how unemployment changes across the dfferent experments. Agan, I beleve that f unemployment does not add much to the result t should be omtted. Fnally, I don t understand how fnte horzons can make a dfference. Although these elements do generate dfferent reasons for why people may save, they are not the focus of the paper. Unless they are shown to affect the results, they should not be ncluded, as they dlute the contrbuton of the paper and render the model and results dffcult to understand. The bankng system chooses the crtcal level, m*, and the lendng rate, R l, and takes as gven the rate on the depost, R d, and the loss/depost rato, α. Ths s where one mportant element of the lendng neffcency s ntroduced: α may not be optmal. The central bank sets the nterest rate on deposts, R d. There s another constrant,.e., that m has to be greater than m*. Ths s also a source of neffcency, and s probably used to descrbe the dea that small projects have less chance of gettng fnanced. The authors should dscuss the sources of lendng neffcency n the model. My man dffculty wth the paper concerns the results. The results presented n Table 1 present four scenaros. The frst s the benchmark case. The second s the credt-crunch case where returns are (exogenously) lower. The other two cases examne how monetary and lendng polces could allow us to go from the credt-crunch stuaton to the benchmark one. The credt-crunch stuaton occurs because lendng has decreased (from 0.76 to 0.27). However, even wthout neffcent lendng, lendng would have decreased n response to a drop n the projects returns. The queston s to know how much more lendng has decreased because of neffcent lendng. To know that, one would have to compare the stuaton wth one n whch there s no neffcent lendng. I don t know whether ths s feasble n the current set-up; however, t seems crucal to me to know the contrbuton of the lendng neffcency versus the effects on lendng due to lower returns. When the polcy experments are examned, the authors compare the results of the polces wth the benchmark case. However, returns were hgher n the benchmark case. The paper s objectve was to examne the role of monetary and lendng polces n allevatng the effects of credt crunches, not the effects of reduced proftablty. The objectve should be redefned or the polcy experment clarfed. I would also lke to see the effects of reduced proftablty on m and y.

19 Dscusson Shouyong Sh Yuan and Zmmermann pose the followng queston: Can monetary polces help the economy recover from a credt crunch? Ther answer s: Not much, f banks manage rsks conservatvely by mposng an upper bound on the loan loss/depost rato. In fact, f the central bank cuts the depost rate durng a credt crunch, the total deposts and total loans fall. Therefore, easng monetary polces durng a credt crunch s counterproductve. The authors reached ths concluson by calbratng a dynamc general-equlbrum model wth credt market mperfectons. They took the year 1992 as one of a credt crunch relatve to the sample perod, , and assumed that a negatve shock to borrowers project returns generated the crunch. The authors computed how the equlbrum would change f the central bank cut the depost rate by one-half per cent durng the crunch. The effects are as follows. Bank lendng rates would fall, and banks would ncrease ther lendng standards. However, the total deposts and total loans would fall, both by small magntudes. To demonstrate how the model generates these results, let me brefly descrbe the model. It has one prvate bank and many agents. Agents are grouped nto employed workers, unemployed workers, entrepreneurs, and retrees. At the begnnng of each perod, all agents, except retrees, can apply for loans, but only those who have enough wealth get loans. If an agent obtans a loan, he/she becomes an entrepreneur and nvests the funds n projects. Otherwse, the agent becomes employed, unemployed, or retred, and deposts some of the wealth at the bank. The bank pools all of the deposts and makes loans, takng the depost rate as gven. The loan decsons are the lendng rate and the lendng standard, a cut-off level of borrowers wealth above whch a loan s provded. The bank balances ts 341

20 342 Dscusson: Sh books,.e., to equate the expected revenue from loans and the nterest payment on deposts. In ths model, agents are ratoned f ther wealth levels are below the lendng standard. Such credt ratonng s generated by three mperfectons of the loan market, hghlghted n the followng assumptons: (A1) Borrowers have lmted lablty ther consumpton cannot be negatve and so the bank cannot recover the loan when a project has a suffcently low return. (A2) For each unt of a loan loss, the bank must ncur an addtonal cost and may also provde a mnmum consumpton level for the bankrupt agent. (A3) The bank s conservatve n lendng t mposes an upper bound on the loan loss/depost rato, α. These mperfectons compel the bank to set a lendng standard above agents lowest wealth. Whle the frst two mperfectons are standard n credt-ratonng models, the last s relatvely new, and s crtcal for a credt crunch. A credt crunch occurs n the model as follows. When there s a negatve shock to entrepreneurs project returns, the expected revenue from loans falls. More borrowers declare bankruptcy and, by (A1) and (A2), the bank faces a hgher loan loss. To balance the books, the bank ncreases the lendng rate and the lendng standard. The loan market becomes tghter and, n partcular, more small (poor) borrowers are squeezed out. The loan market becomes tghter n ths way even wthout the conservatve lendng constrant, (A3), but ths tghtenng mght not be large enough to consttute a credt crunch. A credt crunch occurs when the bank further ncreases the lendng standard and greatly reduces the loan supply to meet the conservatve lendng constrant. I lke the followng two aspects of the model. Frst, t has a generalequlbrum flavour and s dynamc each agent lves for qute a long tme. The long lfetme and the general-equlbrum feature allow the authors to fully ncorporate how loan market mperfectons affect consumpton and savngs and how these effects n turn affect the loan market through deposts. In contrast, most other models allow agents to lve for only two perods, and thus the loan market mperfectons do not have a strong feedback effect through savngs (e.g., Wllamson 1987, and Bernanke and Gertler 1989). Second, loan applcants are endogenously dvded nto entrepreneurs and depostors, accordng to ther wealth: Rch applcants are entrepreneurs and poor applcants are depostors. There s even a wealth dstrbuton among entrepreneurs. Ths modellng approach not only makes loan supply

21 Dscusson: Sh 343 and demand endogenous, but also allows a credt crunch to generate dstrbutonal effects. For example, poor and small borrowers suffer the most n a credt crunch, snce they are the frst ones to be ratoned when the lendng standard ncreases. Agents are also heterogeneous n other ways. Some are employed and some are not; some are n the labour force and others are retred. Thus, the current model s useful for addressng both the aggregate effects and the redstrbutonal effects of credt market polces. Now, let me offer some crtcsms on the paper. The authors should explan more clearly why conservatve lendng can generate a credt crunch. The model appears to have pcked up a strong negatve wealth effect on depostors, whch reduces loans through the conservatve lendng constrant. In the model, total loans fall from 0:76 n the sample to 0:27 n 1992, accompaned by only a small ncrease n the lendng standard (from 6:76 to 6:84), but by a large drop n deposts (from 0:87 to 0:65). Ths drop s responsble for the sharp ncreases n the bank s bond holdngs/depost rato (from 13 to 58) and n the lendng rate (from 1:18 per cent to 1:30 per cent). In ths sense, the credt crunch n the model s more precsely a depost crunch. Ths s how the conservatve lendng constrant s mportant for the credt crunch. When entrepreneurs wealth falls after a negatve shock to project returns, the bank s severely restrcted by the conservatve lendng constrant, for two reasons. Frst, those entrepreneurs who are lucky enough to be entrepreneurs agan n the future wll requre more external fnancng, whch ncreases the expected loan loss. Second, and more mportant, those who cease to be entrepreneurs n the future save less and depost less. The two effects both ncrease the loss/depost rato and, to mantan the conservatve lendng constrant, the bank greatly reduces loans. My guess s that such a credt crunch wll occur n the model even when the bank s lendng standard does not change. The behavour of deposts mght also explan why a lower depost rate does not help the economy to recover. In fact, t further reduces depostors wealth and makes loans even more dffcult to obtan. As the numercal exercses ndcate, when the central bank reduces the depost rate by one-half per cent, total deposts and total loans both fall n spte of the fact that the lendng rate and the lendng standard both decrease. Snce a negatve shock to project returns generates a credt crunch by greatly reducng deposts, the monetary authorty should ncrease rather than decrease the depost rate durng a credt crunch. The authors dd not explctly outlne why banks are conservatve n lendng, although they offered several reasonable explanatons. Snce the man results of the paper depend so much on ths conservatve lendng

22 344 Dscusson: Sh constrant, one cannot help but wonder about such an exogenous restrcton on the bank s behavour. How can we be sure that the upper bound on the loan loss/depost rato does not change durng a credt crunch? More mportant, how can we be certan that the bound does not react to monetary polces? By mantanng an exogenous lendng constrant, the authors cannot be confdent n the polcy evaluaton. To llustrate ths pont, consder the explanatons that the authors provded for the conservatve lendng constrant. One concerns the bank s assessment of the value at rsk; the other concerns captal-adequacy requrements. Both explanatons suggest opposte adjustments n the upper bound on the loss/depost rato durng a credt crunch. If the bank s conservatve n lendng because of the value at rsk, the negatve shock to project returns ncreases the value at rsk per unt of loan and nduces the bank to become more conservatve. That s, the bank should reduce the upper bound on the loss/depost rato. In ths case, the shock generates a more severe credt crunch than n the model. On the other hand, f the bank s conservatve because of the captal requrement, the reducton n loans durng a credt crunch should make the bank less conservatve. Ths s because the bank holds government bonds the value of whch does not change durng the credt crunch n the current model. Wth fewer loans, the bank can stll meet the captal requrement f t ncreases the upper bound on the loss/depost rato. The conservatve lendng constrant becomes less bndng n ths case, and the credt crunch less severe. Monetary polces have dfferent effectveness n these two cases. If the value at rsk s the reason for the upper bound on the loan loss/depost rato, the central bank can acheve lttle by reducng the depost rate. But f the captal requrement s the reason for the upper bound, an expansonary monetary polcy can reduce the nterest rate, ncrease the value of government bonds, and make the captal-adequacy requrement less bndng. In ths case, the polcy nduces the bank to ncrease the upper bound on the loss/depost value and greatly mtgates the credt crunch. There s a wde range of heterogenety n the model, but the authors dd not fully explore t. As mentoned, ths wealth of heterogenety allows a credt crunch to have redstrbutonal effects. The authors mentoned only one that was already n the lterature,.e., a credt crunch hurts small entrepreneurs more than large entrepreneurs. But the authors can explan much more wth ths model. How does a credt crunch redstrbute wealth between entrepreneurs and depostors? Are employed and unemployed workers affected dfferently? How do monetary polces redstrbute wealth among agents?