Why Do Guaranteed SBA Loans Cost Borrowers So Much? Flavio de Andrade* Deborah Lucas**

Transcription

1 December 31, 2008 Prelmnary and Incomplete, please do not quote wthout permsson Why Do Guaranteed SBA Loans Cost Borrowers So Much? Flavo de Andrade* Deborah Lucas** *Northwestern Unversty **Northwestern Unversty and NBER We thank Jennfer Man, Damen Moore, Sylva Merola and especally Wendy Kska for ther comments and help wth obtanng and understandng the data used n ths project. All errors reman our own. 1

2 1. Introducton The federal government ncreasngly reles on subsdzed credt guarantees n leu of grants or other forms of assstance to targeted groups. In provdng credt assstance, the government has a choce between drect lendng where t funds loans drectly va the Treasury, and guaranteed lendng where funds are rased ndrectly va prvate fnancal nsttutons n the captal markets. Whether guaranteed lendng s an effcent way of subsdzng credt s an mportant queston that we explore here n the context of the Small Busness Admnstraton s (SBA) 7(a) program. The goal of the SBA 7(a) program s to lower the cost and mprove the avalablty of credt to U.S. small busnesses. In 2006, t guaranteed 82,000 loans totalng $12 bllon. The loans generally have an orgnal maturty of 7 to 10 years, and nterest s based on the prme rate plus a fxed spread. Qualfed borrowers are able to obtan loans from SBA-certfed prvate fnancal nsttutons, usually commercal banks, wth the backng from a federal guarantee that nsures the lender aganst 50 to 85 percent of credt losses. Despte the beneft of a substantal federal guarantee and default experence that s no worse than that on seemngly comparable loans and bonds, 7(a) borrowers are charged rates that are no lower than on the comparable securtes. In ths paper we revew the evdence that supports ths concluson, and evaluate several possble explanatons for why borrowers are charged such seemngly hgh rates. One hypothess s that hgh borrowng rates are a consequence of mperfect competton n the 7(a) lendng market, makng the loans a proftable lne of busness for partcpatng lendng nsttutons. In fact some banks orgnate large numbers of 7(a) loans, whle others choose not to partcpate n the program at all, and n some regons there are very few SBA lenders. To examne the hypothess that hgh costs are due to lmted competton, we frst calculate a measure of SBA local lender concentraton based on a Herfndahl ndex and then run a regresson of nterest rate spreads on Herfndahl ndex values, controllng for observable borrower and loan characterstcs. We fnd no evdence that rate spreads ncrease wth lender concentraton, and n fact 2

3 consstently fnd the opposte. We then consder an alternatve to the Herfndahl ndex, whch s based on the number of loans orgnated by a partcular lender. Based on ths measure, we fnd that the largest lenders tend to charge hgher spreads than smaller ones. Snce large lenders are also lkely to be admnstratvely more effcent, ths may reflect the greater market power of large lenders. Hgh borrowng rates for 7(a) borrowers may also be attrbutable to a hgher cost of captal for 7(a) lenders than what one would expect wth a full fath and credt federal guarantee. Past studes have shown that federally guaranteed oblgatons such as RTC bonds and student loans bear hgher rates than standard Treasury securtes despte full fath and credt federal guarantees, and that the rate dfferences can be substantal (Longstaff (2004), Lucas and Moore (2008)). These authors pont to lower lqudty as a lkely reason for the hgher requred rates of return than on comparable Treasury s. In evaluatng the rate charged on 7(a) loans, one has to take nto account that 7(a) loan cash flows are rsky, despte beng guaranteed aganst default rsk. 1 There s consderable uncertanty as to the tmng of payments because of the possblty of default and prepayment, and due to movements n the prme rate. Ths can ntroduce systematc rsk nto cash flows even though drect losses from default are precluded. The addtonal layers of ntermedaton ntroduced by relyng on banks and securtzaton markets also create may create addtonal costs that can elevate spreads. To evaluate the cost of captal on the guaranteed porton of SBA loans, we develop a Monte Carlo model of the cash flows on such securtes, takng nto account hstorcal default and prepayment behavor, and modelng nterest rates usng a two-factor Cox, Ingersoll and Ross (CIR) model. Dscountng these cash flows at the rsk-free rates along each path, we fnd the theoretcal value of the cash flows at Treasury rates. By comparng these theoretcal prces to actual secondary market prces of the securtzed pooled and guaranteed porton of SBA loans, we can then nfer the spread over Treasury rates necessary to equate the theoretcal prce wth the market prce. We fnd evdence that nvestors typcally demand a spread between 100 and 200 bps over Treasury rates n the perod for whch we have 1 Ths s also true of other federally guaranteed loans such as FHA loans and student loans. 3

4 data (1/06 through 6/08), whch explans part of why 7(a) loans cost borrowers so much despte the beneft of a federal guarantee. The rest of the paper s organzed as follows: Secton 2 gves a bref overvew of the 7(a) program. Secton 3 presents evdence that 7(a) loans appear relatvely expensve for borrowers, and also very proftable for lenders. Secton 4 reports on how borrower characterstcs, loan characterstcs, and the concentraton among SBA lenders are related to the cross-secton of credt spreads. In Secton 5 we develop the model to prce SBA guaranteed loan pools, and compare model prces wth market prces to nfer the premum over Treasury rates demanded by nvestors. The lkely proftablty of SBA loans s reevaluated gven ths nformaton. We conclude n Secton 6 wth a dscusson of the polcy mplcatons of the fndngs. 2. SBA 7(a) Program Overvew In ths secton we brefly descrbe the most salent features of the 7(a) lendng program to ths analyss, snce more complete descrptons are avalable elsewhere (e.g., CBO (2000) and Glennon and Ngro (2005)). In terms of total sze, the program that has grown rapdly snce 2001, wth $12 bllon n prncpal dsbursed on more than 82,000 loans n Stll, 7(a) loans represent less than 5 percent of total U.S. small busness borrowng. What we refer to as the 7(a) program actually has two parts, a regular program and an Express program. The Express program promses borrowers and lenders quck elgblty decsons, but provdes a federal guarantee on only 50% of loan prncpal and has a strcter restrcton on loan sze. The Express program has grown to represent near 75 percent of loans guaranteed, but only 20 percent of dollar loan volume. 2.1 Borrowers Busnesses seekng 7(a) loans must satsfy a number of crtera. The most subjectve s that they must establsh that they have not been able to obtan credt elsewhere on affordable terms. They also 4

5 must demonstrate an ablty to repay the loans, for nstance based on collateral, the amount of owners equty, and the general qualty of management. 2 More concretely, the busness must be for-proft, small (the defnton of whch s based vares by ndustry), and ndependently owned and operated. Borrowers come from a wde varety of ndustres, wth the largest three beng retal trade, accommodaton and food servces, and manufacturng, whch together account for about 50 percent of loans dsbursed. 2.2 Loan Terms and Characterstcs Over 96 percent of 7(a) borrowers pay an nterest rate set to a fxed spread over the prme rate. The prme rate s a floatng rate that snce 2000 has hovered around 3 percent above the overnght Fed Funds rate. The spread s capped by SBA regulaton, but on most loans the spread cap does not appear to bnd. There are also a varety of fees assessed by the SBA, some pad by the borrower drectly and others by the lender. Borrowers pay a graduated one-tme guarantee fee that ranges from 2 percent of prncpal for loans of less than $150,000 to 3.5% for loans of $1 mllon. Borrowers wth long-term loans also face hgh prepayment penaltes n the frst three years of the loan s lfe, after whch prepayment s a free opton. Lenders pay the SBA an annual servcng fee of.545 percent on the guaranteed balance, although they bear most servcng costs themselves. Loan maturty typcally s 7 to 10 years, but vares wth the purpose of the loan. Loans for real estate and equpment may extend to 25 years. The effectve duraton of loans s consderably shorter, both because the loans are amortzng and because of voluntary prepayments and defaults. Loans n the regular program can be for up to $2 mllon, and n the Express program up to $350,000. The maxmum percentage guaranteed vares wth the loan sze and program. In the regular program, for loans under $150,000, SBA guarantees up to 85% and for loans above $150,000 the maxmum s 75%. On Express loans the maxmum guarantee percentage s 50%. Loans do not always bear the maxmum allowable guarantee, presumably to reduce the amount of fees pad to SBA. Upon 2 The SBA assgns a rsk ratng to ndvdual loans, but t s not a standard credt score and that data s propretary. 5

6 default, a lender can recover from SBA the face value of the outstandng prncpal balance, and subsequently wll receve a pro rata share of any recoveres. However, lenders sometmes choose not to exercse the guarantee when they expect to recover more from retanng the entre loan. Thus the SBA default statstcs that we rely on are somewhat downward-based estmates of the default rate experenced by the borrower Lender Characterstcs SBA-backed loans are typcally made by commercal banks, but some thrfts and fnance companes also partcpate. Lenders classfed by SBA as preferred lenders have the authorty to determne elgblty n the Express program. For commercal banks, an advantage of partcpaton s that t helps to satsfy Communty Renvestment Act requrements. As descrbed n more detal n secton 4, relatvely few banks 4 orgnate large numbers of 7(a) loans. Of 2017 dstnct lendng organzatons dentfed for 2006, only 14 orgnated more than 1,000 loans each, and 880 orgnated only 1 or 2 loans. The denttes of the 20 largest orgnators and ther lendng volume are shown n Table The Guaranteed Loan Pool Certfcate Program (GLPCP) To ncrease access to captal for 7(a) lenders, snce 1995 the SBA has sponsored the Guaranteed Loan Pool Certfcate Program (GLPCP). Under ths program 7(a) lenders can sell the guaranteed porton of 7(a) loans to a Pool Assembler. The Assembler puts together pools of loans that are smlar n terms of maturty, rate bass (e.g., monthly and prme-based), and rate spread. Each pool has a mnmum of four underlyng loans that total at least $1 mllon n aggregate. The strctest restrctons on 3 For the purposes of calculatng returns to lenders, ths problem s mtgated by the fact that non-surrendered loans are those wth relatvely hgh recovery rates. 4 A bank for ths tabulaton consoldates over nsttutons governed by the same holdng company, even f they operate n far flung geographc regons. 6

7 rate floors or celngs on the ndvdual loans are reported for the pool. The pools are sold va an aucton to nvestors, and there s also a secondary market n the Certfcates. It appears that the securtzaton process does not ntroduce any sgnfcant counterparty rsk nto the lendng process. Payments on the securtes are centralzed and made by Fscal and Transfer Agent (FTA-Colson Assocates), who s apponted by the government and acts as ther agent. Payments are made on tme even when default and prepayments occur. These payments are made by the FTA to nvestors n the form of advances. Lenders also do not bear counterparty rsk ether snce they deal drectly wth the FTA. The program has become an ncreasngly mportant source of fnancng, wth a $4.0 bllon of guaranteed securtes sold nto pools n 2006, representng 42% of the total guaranteed amount of loans approved n that year. We have no evdence on whether securtzed loans devate systematcally from a typcal SBA loan n terms of default or prepayment behavor. We assume that there s no sgnfcant selecton bas snce they represent a large fracton of total orgnatons. The market prces of these certfcates s of nterest n ths analyss because they can be used to nfer the cost of captal for the default-rsk-free porton of 7(a) loans, as dscussed n Secton 5 below. 3. The Cost and Proftablty of 7(a) Loans Our nferences about the cost of SBA loans rely on comprehensve data obtaned from SBA on all dsbursed loans from 1988 to Assocated wth each loan s statc nformaton about the borrower (e.g., type and sze of busness), the lender (e.g., name, state), and loan terms (e.g., rate spread, bass, maturty, f collateral). There s also a record of prepayments and defaults. The SBA loan data s descrbed n more detal n Appendx Cost to Borrowers Average rate spreads and guaranteed shares on 7(a) loans are summarzed n Table 2. The average spread over prme s 2.36%, based on all loans dsbursed from 1988 and Over the sample 7

8 perod the prme rate averaged 2.77% over the overnght Fed Funds rate 5, and the level of the Fed Funds rate was roughly equvalent to the 1-year Treasury rates (see Fgure 1). Thus on average borrowers pad a spread of about 5.13% over the 1-year Treasury rate, even though the average loan bore a federal guarantee on 70.6% of prncpal. The ntal guarantee fee added about 40 bps to the effectve rate pad by borrowers. 6 To assess whether these spreads are n lne wth other credt oblgatons wth comparable default rsk, we draw on the analyss n CBO (2007), whch uses the same SBA loan data. Fgure 2 (reproduced from CBO (2007), Fgure 4) shows that the spread charged SBA borrowers has been consstently hgher than that for all commercal and ndustral (C&I) loans by a margn of about two percentage ponts for regular 7(a) loans, and over three percentage ponts for Express loans. Hgher rates on SBA loans mght be justfed, even wth a partal guarantee, f the probablty of default s sgnfcantly hgher than on other C&I loans. The most drect evdence that t s not hgher comes from, Glennon and Ngro (2005), who fnd that, although medum-maturty loans orgnated under the SBA 7(a) loan guarantee program are targeted to small frms that fal to obtan credt through conventonal channels, the default experence s comparable to that of a large percentage of loans held by larger commercal banks. Further evdence on default rates s found by comparng the cumulatve default rates on 7(a) loans and publc debt. CBO (2007) fnds that the default experence on 7(a) loans falls between that on BB and BBB-rated bonds (see Fgure 3, reproduced from Fgure 9 n CBO (2007)). Fgure 4 shows that overall default rates on 7(a) loans have fallen over tme, and have stablzed at just less than 2 percent of prncpal outstandng snce Nettng the average recovery rate of between 40 and 50 percent suggests an annual loss rate to SBA of about 1 percent. 5 Fed Funds s used as a reference pont because snce 2000 the prme rate has been fxed at 3% over Fed Funds. 6 Ths assumes a 2.5% guarantee fee, a 7-year ntal maturty, and a dscount rate of 7%. 8

9 3.2 Returns to Lenders The value of 7(a) loans from a lender s perspectve depends on the margn between revenues and costs. Revenues arse from borrower payments and proceeds from loan sales. Costs nclude the lender s weghted average cost of funds, servcng and other ongong admnstratve costs, and fees pad to the SBA. Takng these costs nto account, we do a smple back-of-the-envelope calculaton n ths secton suggestng that 7(a) loans on average are qute proftable for banks. For bank lenders, holdng a 7(a) loan s lke ownng a portfolo of a default-rsk-free government securty and a rsky small busness loan. Banks weghted average cost of captal for 7(a) loans should reflect these two components, and n a compettve market the advantage of the guarantee should be passed through to the borrower. The cost of funds s clearly dfferent for the nsured and unnsured portons of a loan. The nsured porton s vrtually default-rsk free, suggestng that nvestors would requre a return close to a short-term rsk-free Treasury rate on a floatng rate loan. 7 present a more careful analyss of the cost of funds for the guaranteed porton.) (In Secton 5 we On the unnsured porton, default experence suggests fundng costs smlar to those on BB to BBB-rated loans. Based on hstorcal BB rate spreads, the gross fundng rate on the rsky porton s taken to be Treasury percent. Ths s also consstent wth the spread between C&I loans and SBA loans n Fgure 2. Based on an average 70.2% of guaranteed prncpal, the WACC s at a spread of 1.03% above Treasury. The net spread between revenues and expenses s calculated by subtractng the weghted average cost of captal and other costs from nterest payments receved net of the expected default loss rate. Other expenses are assumed to be.75% annually for servcng and other admnstratve costs, and.545% annually for the SBA servcng fee on the guaranteed porton. Default losses are assumed to be 1.25% on the non-guaranteed porton, slghtly hgher than those suggested by SBA experence. 8 Together, these 7 Consstent wth ths vew, banks nvestng n GLPCP certfcates have a 0% captal requrement aganst these holdngs. 8 Ths s based on about a 1% default rate for the SBA, and a.25% adjustment for the loans that default but are entrely retaned by the lender. 9

10 assumptons mply a net spread of 2.6%. The net spread measures proft as a percentage of loan prncpal. Another way to assess the sgnfcance of ths margn s to ask what the breakeven cost of unnsured captal would be for the lender to break even accordng to ths calculaton. The break-even spread on unnsured captal s 12.4%, whch s 8.85% more than the cost of unnsured captal suggested by market prces. Clearly ths estmate s senstve to the many assumptons made, but t does suggest that 7(a) lendng s a proftable lne of busness. 4. Loan Spreads and Market Concentraton A possble explanaton for the hgh spreads charged to borrowers s that lendng nsttutons have market power that allows them to charge above market rates. As mentoned earler, a relatvely small number of banks are actve orgnators of 7(a) loans, but many others make few or no such loans. Fgure 5 shows the dstrbuton of the number of loans orgnated by nsttuton n The average number of loans orgnated s 39.6, but the medan s only 3 loans. The dstrbuton by dollar amounts of orgnatons s smlarly skewed to the left. To explore the effect of local lender concentraton on rates charged, we frst defne a measure of the regonal compettveness of SBA lendng markets, and then regress rate spreads (over the prme rate) on ths quantty and a set of control varables that could also affect the rate spread, ncludng the percentage guarantee, loan maturty, loan sze, and whether t s an Express loan. The cross-sectonal data s pooled over the years 1988 to In the second set of regressons we only use the 2006 cohort, and account for market power by assgnng lenders to 5 lender-dstrbuton buckets, constructed wth the number of SBA loans that lenders orgnated n For constructon of the Herfndahl ndex, a regon s taken to be a Core Based Statstcal Area (CBSA), whch defnes mcropoltan and metropoltan areas. In each of these areas we look at the level of competton n the lendng market for SBA loans. Usng the borrower s zp code we are able to assgn 10

11 each loan to a CBSA. Then for a gven year and a gven CBSA we determne the number and the dollar amount of loans orgnated by each lender. Wth ths nformaton we calculate a Herfndahl ndex for each CBSA and each year. The Herfndahl ndex for CBSA at tme t s gven by: N, t 2 H t s n t n1, (1) where s, s market share of lender n, n CBSA at tme t. The Herfndahl ndex gves a number n t between 0 and 1, wth 1 representng a monopolstc market and 0 a perfectly compettve market. The SBA has enacted nterest rate-spread lmts and maxmum loan guarantee whch are apples to loan type, loan sze and maturty. For loans wth rates close to the celng, ths could bas nferences about the effect of market power on rate spreads. The spread and guarantee lmts are presented n Table 3. In the regressons of spreads on Herfndahl ndex we splt the sample to ncorporate these spread and guarantee caps, and run a regresson for each subsample separately. In the regresson of spread on lender s dstrbuton we ncorporate a dummy varable f the spread of the loan s wthn 25 bass pont of the nterest rate cap. In table 4 we present the regresson results, for tests of spreads on regonal concentraton. The sample s subdvded by whether the loan s regular or Express, by loan sze, and by maturty. Control varables nclude the percentage guaranteed, the loan sze, and maturty. All regressons suggest that the Herfndahl ndex does not explan hgher nterest rate spread. In fact, the regressons suggest that regons wth more concentrated lendng markets are assocated wth lower nterest rate spreads, whch seems counterntutve. Perhaps borrowers cross cty boundares to look for affordable loans, and therefore regonal concentraton may not be the best way to capture market power. If that s the case, then hgher Herfndahl ndex may be assocated wth smaller markets, and perhaps smaller lenders whch have less market power. 11

12 For the 2006 cohort we use an alternatve measure of lender market power, based on a set of ndcator varables for the number of loans that a lender orgnated n that year. The categores are 0-9 loans, loans, loans, loans, and at least 1000 loans. In addton, we ntroduce two other dummy varables, one to flag f a loan s express, and another to ncorporate loans that are wthn 25 bass ponts of the lmts shown on table 1. The results n table 5 suggest that the largest lenders tend to charge hgher nterest rate spreads than smaller ones. In addton, controllng for other loan characterstcs, Express loans charge an extra 127 bass ponts above Regular loans. The adjusted R 2 of the current specfcaton shows a sgnfcantly hgher explanatory power than n the prevous specfcatons. Ths suggests that much of the unexplaned varaton n spreads that was captured by the ntercept n prevous specfcatons s accounted for by lenders characterstcs. In future drafts we wll look at whether default and prepayment rates dffer systematcally by lender sze, snce these are alternatve explanatons for the hgher rates charged. 5. Cost of Captal for Fully Guaranteed Loans The analyss thus far suggests that borrowers pay a relatvely hgh rate on SBA loans, and that the market power of larger lenders may be part of the explanaton. We also know, however, that many banks choose to make very few or no 7(a) loans, suggestng that our estmate of proftablty to lenders n Secton 3.2 s lkely too hgh. In ths secton we show that the apparent proftablty of 7(a) lendng s reduced by the fact that the cost of captal for fully guaranteed 7(a) loans s sgnfcantly above Treasury rates, as nferred from the prcng of securtes n the Guaranteed Loan Pool Certfcate Program (GLPCP). 5.1 Data We were able to obtan a hstory of monthly market prces on 166 of GLPCP pools that were ssued by Coastal Securtes between January 2006 and May Coastal Securtes s one of the leadng dealers n SBA-backed securtes, and actvely markets GLPCP pools. Other than secondary 12

13 market certfcate prces, the data set ncludes the orgnaton date, fnal maturty, loan orgnaton amount, and spread to prme. The average pool sze at orgnaton s $10.3 mllon, wth a standard devaton of $12.2 mllon. The smallest pool sze at orgnaton s $1.0 mllon, and the largest s $95.2 mllon. Our sample falls nto a perod wth hstorcally narrow credt spreads. Fgure 6 shows that pror to the sample perod and followng the end of our sample (based on Bloomberg estmates and not our model), the yelds on GLPCP securtes were much hgher. Ths suggests that usually there may be an even hgher rsk premum than what we nfer n the analyss that follows. 5.2 Prcng Model To estmate the rsk premum prced nto GLPCP securtes, we mplement a Monte Carlo Model to project the dstrbuton of pool cash flows over the maxmum lfe of the pool. The term structure of Treasury rates evolves stochastcally accordng to a 2-factor CIR model followng Kapln, Sun and Jagannathan (2001). The Prme rate set to a 3 percent spread over the mpled nstantaneous Treasury rate. The model s descrbed n more detal n Appendx 2. Cash flows are affected by three sources of uncertanty: prepayments, defaults, and changes n the prme rate. Prepayments and defaults both trgger a termnal payment of outstandng prncpal and nterest. Rate changes affect the nterest payment n the next perod, and also requre a re-amortzaton of the remanng prncpal. Annual prepayment and default rates n each year after orgnaton are based on average rates derved from the SBA loan data for each year n a loan s lfe. Both prepayment and default rates vary sgnfcantly over the lfe of a loan. Prepayment rates on medum- and long-term loans peak between 6% and 8% n years 3 to 6 and then steadly declne. Default rates peak n year 2 and 3 at between 2% and 3%, and drop off sharply thereafter. These patterns are reflected n the smulatons, whch employs draws from a unform random number generator to determne whether a default or prepayment has occurred n a gven month. 13

14 Uncertanty arsng from changes n the prme rate, default, and prepayment potentally ntroduces systematc rsk nto the cash flows. The spread between Prme and Treasury, whch we take to be fxed, n fact s expected to be countercyclcal, snce credt spreads ncrease n downturns. Ths mparts a slghtly negatve beta to the promsed cash flows, whch thereby get larger when tmes are bad. The sgn on the systematc rsk ntroduced by default and prepayment s not obvous. Both trgger an early return of prncpal, so the effect depends on the systematc rsk of the trggerng event, and whether the GLPCP securtes tend to systematcally prced at a dscount or premum to par n a downturn. We do not try to quantfy these potental effects of systematc rsk. Instead, we look for the fxed percentage pont spread over 1-month Treasury rates mpled by the CIR model that equates the model prce wth the market prce from the Coastal Securtes data. 5.2 Results and Senstvty Analyss The results of the analyss are shown n Fgure 7, whch show the dsperson of mpled rsk prema across loan pools n each month n the sample. Note that the amount of prce data ncreases over tme, snce new pools are enterng the sample and older pools are not agng out of t. The overall average premum s 137 bps, wth a standard devaton of 35 bps. An nterestng queston s how much of the varaton n the premum can be explaned by observable pool characterstcs such as the orgnal sze and fnal maturty of the pool, and the spread to prme. Ths queston, and addtonal senstvty analyss, wll be ncluded n the next draft. 5.3 Implcatons for Proftablty The smple proftablty calculaton n Secton 3.2 took the short-term Treasury rate as the cost of captal for the guaranteed porton of 7(a) loans. The analyss of ths secton suggests that the cost s hgher by an average of 137 bps. Takng ths nto account n the cost of captal calculaton decreases estmated profts as a percent of loan value from 2.6% to 1.6%, and decreases the breakeven cost of 14

15 unnsured captal from 17.6% to 14.3%. Hence, the sgnfcantly hgher than rsk-free cost of captal to fnance the guaranteed porton of SBA 7(a) loans explans part, but not all, of why loans cost borrowers so much. 6. Summary and Conclusons In ths paper we have looked at two possble and complementary explanatons for the hgh rates charged to borrowers n the 7(a) program. We fnd support n the data for both: (1) that large SBA lenders may be able to exert market power and thereby charge hgher rates, and (2) that the cost of fundng the guaranteed porton of SBA loans s sgnfcantly above Treasury s cost of funds. These fndngs suggest that there may be ways to modfy the 7(a) program to reduce costs to borrowers and fnance the program more effcently. The leadng alternatve to guaranteed lendng s drect lendng, where the government drectly funds and orgnates loans. A ratonale for nvolvng prvate ntermedares n the orgnaton process for government loan programs s when screenng for credt qualty s mportant. Although we dd not try to assess the value of ths functon for the 7(a) program, clearly small busness loans are rsky and requre judgment, suggestng an mportant screenng role for the prvate sector. Nevertheless, t seems that the government could reduce program costs by purchasng the guaranteed porton of loans drectly from bank lenders, and fundng those purchases through the Treasury rather than through sponsorshp of a prvate securtzed loan market. 9 If n fact large SBA lenders are able to exert market power n loan prcng, a further role for the federal government could be to try to ncrease compettveness n the market, perhaps by ntroducng a more centralzed loan applcaton process that would force lenders to compete more drectly for borrowers. Ths would be consstent wth the general trend toward the ncreased relance on credt 9 Ths was the ratonale n 1973 for consoldatng federal borrowng for dfferent programs through the Federal Fnancng Bank. 15

16 scorng and other types of hard nformaton n small busness lendng, although local lenders may stll have nformaton advantages that make them the most effcent orgnators (ctatons TBA). 16

17 APPENDIX 1 7(a) Loan Data The SBA admnsters 7(a) Program data n ther Electronc Loan Informaton Processng System (ELIPS). Ths system ncorporates statc data nformaton that does not regularly change over the lfe of a loan as well as loan transactons related to balance, purchase, recovery, and other actvtes. The current project reles on data from ELIPS to dentfy basc loan, sze, maturty, lender nformaton, borrower zp code, as well as to dentfy transactons related to purchases and recoveres to dentfy defaults and prepayments. Statc Loans Fle: The fle conssts of over 1.2m of SBA loans snce the program has been n place, wth one statc record per loan. Each record contans a lst of relevant nformaton ntroduced at tme of orgnaton. It ncludes record of the followng: date loan was taken out, loan amount, loan sze, maturty date, SBA guarantee, nterest rate (ncludng the spread above prme rate), program type (Regular or Express), borrower and lender s zp code, lender s name, and characterstcs of the busness. Transactons Fle: The fle conssts of all transactons for each ndvdual loan n the statc fle. These transactons nclude dsbursements, monthly loan payments (ncludng fees), balance, SBA purchases to ndcate default, and recoveres (when there s default). Based on the nformaton contaned n the transacton fle t s possble to determne whether and when default and prepayment has occurred. The transactons fle also provdes full nformaton about recoveres and other nformaton whch are mportant n determnng the default and prepayment behavor of all SBA loans. 17

18 APPENDIX 2 VALUING GLPCP SECURITIES Overvew of Securtes and Underlyng Rsks The GLPCP securtes n our dataset amortze every month and pays nterest at PRIME rate plus a fxed spread. The pool cash flows are subject to early termnaton n the form of default or prepayment. Ether of these events trggers a fnal payment of outstandng prncpal and accrued nterest. The full Government guarantee on these securtes means that certfcate holders do not lose any prncpal when default occurs. However, snce these pools typcally trade above par, nvestors may lose the premum above par when these pools termnate earler than expected. Yet, t s unclear whch porton of these rsks s dversfable and whch porton s systematc. The approach used to address ths queston s to treat these rsks as ndependent from nterest rate rsk, and use the emprcal frequency to model these events. Furthermore, we perform a senstvty analyss to nfer that the magntude of an ncrease n default (prepayment) frequency does not have a large mpact on pool prces. Ths appendx descrbes detals of the Monte Carlo mplementaton used to model the dstrbuton of pool cash flows and the value of the certfcates, and s organzed as follows; we frst use a two-factor Cox-Ingersoll-Ross (CIR) model to smulate rsk-neutral paths of nterest rates used for credt-rsk-free dscountng and for determnng the pool floatng nterest payments. We then model prepayment and default usng the hstorc experence derved from 18 years of comprehensve data obtaned from the SBA. We combne these three components to compute the value of each pool n our sample, and then devse an algorthm to measure the spread above Treasures embedded n the market prce of these securtes. 18

19 A2.1 Interest Rates and Pool Amortzaton We employ a two-factor CIR model to smulate paths of future Treasury rates. In the Jagannathan Kapln Sun (JKS 2003) verson of the CIR model the nstantaneous nterest rate R(t) s the sum of a constant R and the two state varables y (t), for = 1, 2: R( t) R y1( t) y2( t) (1) Each state varable follows an ndependent, mean-revertng, square root process along any perod s, between the current tme t and the maturty date T: y ( s) ds y ( s) dw ( ) dy ( s) s, = 1, 2. (2) where W s are standard and ndependent Brownan motons. Under the prcng (rsk-neutral measure) the factors evolve accordng to: dy ( s) [ y ( s)] ds y ( s) dw * ( s), = 1, 2 (3) where (4) and (5) The market prce of rsk, λ for each state varable s assumed to be lnear. The prce at current-tme t of a zero coupon bond that pays $1 at maturty date T s R( T t) T e p t, T p t, T P t, (6)

20 20 where t y T t B T t A e T t p,,, t T t T T t A 2 1 exp ) ( 2 / ) ( exp 2 ln 2, 2 t T t T T t B 2 1 exp ) ( 1 exp 2, and The yeld to maturty, YTM of a zero coupon bond maturng at T s t T T P t T t YTM, ln, In our analyss we use the two-factor parameters estmated n JKS usng weekly LIBOR rates of varous maturtes: Table A2.1 Two factor CIR parameters, estmated by JKS Factor κ θ σ λ R = (7) (8) (9) (10) (11)

21 Factor 1 has stronger degree of mean reverson and drves the gap between long and short rates. Factor 2 has hgher volatlty parameter and determnes the long run rates. In every perod t n our sample we solve for the ntal state varables, y 1 (t) and y 2 (t), by fttng two ponts n the yeld curve from hstorc Treasury data, the three month T-Bll rate and the ten year Treasury bond rate. We perform ths n two steps, frst set the hstorc yelds on left hand sde of equaton (11) to solve for P(t,T). Then use P and system (6) (10) to solve for y 1 (t) and y 2 (t). We need ntal state varables to smulate the (monthly) dscrete verson of (3), wth Monte Carlo paths that start at tme t and end at T. Monthly dscretzaton means that tme step h For any tme s between t and T we have: y s h y s [ y ( s)] h max y s, hw 0 (12) where W s are drawn from standard and ndependent normal random generator. From (1) and the ntal state varables we compute the ntal short rate R(t). In turn cash flows arrvng one month from t are dscounted back to t wth the factor 1 d( t, t h) 1 R( t) h (13) Cash flows arrvng at s+h are dscounted back to t usng factor d t, s h d t, s 1 Rsh (14) We assume that the prme rate s 3% above the short rate. Suppose a SBA guaranteed pool pays a fxed spread above prme, Δ. Then for the same Monte Carlo run as above the pool s floatng nterest rate s set to 21 (15)

22 s Rs For a gven balance B(s) the pool payment n the followng month (when there s no termnaton) and new balance are set by the amortzaton schedule: Pmt s h 1 Bs ( s) h s( T t) 1 ( s) h (16) and B s h 1 shb s Pmts h (17) A2.2 Default and Prepayment A guaranteed pool s composed of several underlyng SBA guaranteed loans. In practce these certfcates experence partal termnaton when only a porton of the underlyng loans default or prepay. However, some nvestors acqure sngle guaranteed SBA loans n the secondary market nstead of pool certfcates, and are therefore subject to full termnaton when the loan defaults or prepays. Snce nvestors may dversfy ther holdngs n these sngle loans, for valuaton purposes we can model pools as f they were sngle loans by assumng that default or prepayment trggers full termnaton. In essence, the termnaton events n ths secton are reflected n smulatons, whch employ draws from a unform random number generator to determne whether a loan has defaulted or prepad. We estmate default and prepayment frequences as a functon of loan age, usng eghteen years of comprehensve data provded by the SBA. In perod s the default and prepayment probabltes for loan j are expressed by p s and s d j. j In tme step s of the Monte Carlo run where we draw W from eq. (12) we also draw two ndependent numbers from a unform number generator 22 d U s and p U s. Default on loan j s trggered f

23 U d s d p p s and prepayment s trggered f U s j s j. When ether event occurs the pool termnates and nvestors receve the remanng balance plus any accrued nterest. In those nodes Pmt n (16) becomes: Pmt s 1 s hhb s h (16 ) For valuaton purposes t s convenent to set u 0 Pmt for any u s, where s s a termnaton tme. A2.3 Valuaton Usng the nterest rate behavor and termnaton behavor above we obtan pool prces at any tme t. The only nputs for each pool are the orgnaton date t o, the maturty date T, and the pool s fxed spread above the prme rate Δ. For a gven Monte Carlo path ndexed by n the prce at tme t of the pool j s m n j t T st1 Pmt s d( t, s) (17) The model prce of pool j at tme t s computed as the average prce of all smulated paths, and represented by M j t 1 N N n1 m n j t (18) In our smulatons we use N=100,

24 A2.4 Market Prces, Spreads, and Senstvty In our sample, pool market prces are often lower than model prces M j (t), whch suggests that nvestors dscount pool cash flows at hgher than credt-rsk-free Treasury rates. Alternatvely, nvestors may demand a premum for bearng any systematc termnaton-rsk whch may not have been accounted n our model. In ths secton we ntroduce a premum above Treasury rates to equate model prces to market prces. We also perform a smlar exercse by ntroducng a multplcatve premum over default and prepayment probabltes to examne the prce mpact of changes n termnaton probabltes. Defne the dscount premum ψ, and modfy expresson (13) to dscount cash flows from t+h back to t and expresson (14) to dscount cash flows from s+h back to t: d( ; t, t h) 1 1 R( t) h (13 ) and d ; t, s h 1 dt, s Rs h (14 ) For each pool j and tme t we solve for the ψ whch equates M j (t) to market prces. The results are summarzed n Fgure 7, whch shows the monthly dsperson of premum along our sample. An alternatve way to measure the premum demanded by pool nvestors s to ntroduce a multplcatve constant φ to enhance (or depress) the default or prepayment probabltes. For nstance, f φ d s a default probablty premum then default on loan j s trggered at tme s f U d s s d d j and 24

25 prepayment s trggered f U p s p j s. Analogously, f φ p s a prepayment probablty premum then default on loan j s trggered at tme s f U d s d p p p s and prepayment s trggered f U s j s. j 25

26 References Colln-Dufresne, Perre and Bruno Solnk (2001), On the Term Structure of Default Prema n the Swap and LIBOR Markets, Journal of Fnance 56, Cox, J. C., J. E. Ingersoll and S. A. Ross (1985), A Theory of the Term Structure of Interest Rates, Econometrca, vol. 53, pp Gale, W.G. (1991), Economc Effects of Federal Credt Programs, The Amercan Economc Revew, vol. 81, no. 1, pp Glennon, D. and P. Ngro (2005), Measurng the Default Rsk of Small Busness Loans: A Survval Analyss Approach, Journal of Money, Credt, and Bankng, vol. 37, no. 5. Jagannathan, R., A. Kapln, and S.G. Sun, (2003), An Evaluaton of Mult-Factor CIR Models Usng LIBOR, Swap Rates, and Cap and Swapton Prces, Journal of Econometrcs, vol. 116(1-2), pp Longstaff, F. (2004), The Flght-to-Lqudty Premum n U.S. Treasury Bond Prces, The Journal of Busness, 2004, vol. 77, no. 3. Lucas, D. and Damen Moore (2009), Guaranteed vs. Drect Lendng: The Case of Student Loans, forthcomng n Measurng and Managng Federal Fnancal Rsk, D. Lucas edtor, NBER SBA (2007), The Small Busness Economy for Data Year 2006, US Government Prntng Offce, U.S. Congressonal Budget Offce (2005), Estmatng the Value of Subsdes for Federal Loans and Guarantees, Washngton, D.C. U.S. Congressonal Budget Offce (2007), Federal Fnancal Guarantees under the Small Busness Admnstraton s 7(a) Program, 26

27 Table 1: Top 20 SBA Lendng Insttutons n 2006 BANK - AMERICA NATL ASSOC JPMORGAN CHASE BANK NATL ASSOC RBS CITIZENS NATL ASSOC WELLS FARGO BANK NATL ASSOC CAPITAL ONE NATL ASSOC U.S. BANK NATIONAL ASSOCIATION WASHINGTON MUTUAL BANK PNC BANK, NATIONAL ASSOCIATION NATIONAL CITY BANK BANCO POPULAR NORTH AMERICA CIT SMALL BUS. LENDING CORP MANUFACTURERS & TRADERS TR CO HSBC BK USA NATL ASSOC Loans per lender $ amount per lender 9, ,449,130 6, ,544,516 4, ,151,862 4, ,003,115 4, ,557,962 4, ,909,009 2, ,455,477 2, ,056,517 1, ,624,308 1, ,218,724 1, ,124,943 1, ,180,273 1,025 71,770,931 CITIBANK, N.A ,509,780 THE HUNTINGTON ,736,523 NATIONAL BANK KEYBANK NATIONAL ASSOCIATION ,012,935 COMPASS BANK ,241,957 COMMERCE BANK ,338,573 NATL ASSOC ZIONS FIRST NATIONAL BANK CALIFORNIA BANK & TRUST ,877, ,047,012 27

28 Table 2: Rate Spreads (bps) and Guarantee Percentages Entre Sample Regular 7 (a) Express Average* Spread (over Prme) Average Guarantee * - smple average TABLE 3 MAXIMUM GUARANTEE AND INTEREST SPREAD CAPS 3a. MAXIMUM GUARANTEE Loan Type Loan Sze Maxmum SBA Guarantee Regular $0 - $150K 85% Regular $150K + 75% Express all 50% 3b. MAXIMUM INTEREST RATE SPREAD EXPRESS LOANS Loan Sze Maxmum Interest Rate Spread $0 - $50K 6.5% $50K + 4.5% 3c. MAXIMUM INTEREST RATE SPREAD REGULAR LOANS Loan Sze Maturty Maxmum Interest Spread $0 $25K 7yrs % $0 $25K 0 7yrs 4.25% $25K $50K 7yrs % $25K $50K 0 7yrs 3.25% $50K + 7yrs % $50K + 0 7yrs 2.25% 28

29 TABLE 4 The Effect of Regonal Concentraton on Interest Rate Spreads Loan Type Loan Sze Maturty Sample Sze Intercept Herfndhl Guarantee Loan Sze Maturty Adj Rsq Express_Small Express 0-$50K - 95, (20.05) (-14.33) (-7.49) (-50.4) (75.4) Express_Large Express $ , (24.19) (-15.43) (-4.35) (-65.25) (29.9) Regular_Small_Short Regular 0-$25K 0-7yr 11, (5.61) (-5.18) (7.03) (1.57) (2.95) Regular_Small_Long Regular 0-$25 7yrs + 1, (0.78) (-4.52) (4.67) (1.24) (-3.66) Regular_Avg_Short Regular $25-$50 0-7yr 29, (9.25) (-14.59) (15.59) (-2.3) (3.0) Regular_Avg_Long Regular $25-$50 7yrs + 8, (4.35) (-9.04) (9.81) (-1.49) (-5.45) Regular_Large_Short Regular $50-$ yr 76, (27.87) (-28.0) (15.23) (-5.67) (4.33) Regular_Large_Long Regular $50-$150 7yrs + 64, (23.79) (-30.27) (18.79) (4.53) (-13.38) Regular_Huge_Short Regular $ yr 41, (34.38) (-14.93) (4.16) (-5.11) (2.17) Regular_Huge_Long Regular $150+ 7yrs + 167, (70.33) (-8.79) (18.41) (-23.64) (-46.81) TABLE 5 The Effect of Lender Sze on Interest Rate Spreads Estmates t-stat Guarantee Loan Sze Maturty (months) Lender w/ 0 to 9 loans Lender w/ 10 to 99 loans Lender w/ 100 to 499 loans Lender w/ 500 to 999 loans Lender w/ loans Express Dummy Interest Celng_dummy Adjusted R-Squared Number of Observatons 37,728 29

30 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96 Jan-97 Jan-98 Jan-99 Jan-00 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Annualzed Percent 12.0 Fgure 1: Hstorc T-Bll, Fed Fund, and PRIME 1Y T-Bll Fed Funds (Effectve) PRIME Month 30

31 Fgure 2: 31

32 Fgure 3: 32

33 Fgure 4: Source: CBO tabulatons from SBA data. Default rate s clams over total outstandng loan balances. Recovery rate s recoveres dvded by total clams. 33

34 Frequency 1200 Fgure 5: Dstrbuton of Lenders (by # of Loans Orgnated) Frequency of Lenders (by # of Loans) Number of Loans Orgnated 34

35 Fgure 6: Estmated Yeld Spread to Prme on SBA Partcpaton Certfcatons Source: Coastal Securtes Note: Over ths perod Prme was 3% above overnght Fed Funds. Hence -2% on ths scale s a yeld of Fed Funds plus 1%. 35

36 Spread (bps) Fgure 7: Spread (bps above 1-month Treasury) of SBA Pools Spread (bps) Jan-06 Apr-06 Aug-06 Nov-06 Feb-07 Jun-07 Sep-07 Dec-07 Mar Date 36