Can Low Returns to Capital Explain Low Formal Credit Use? Evidence from Microentrepreneurs in Ecuador.

Transcription

1 Can Low Returns to Captal Explan Low Formal Credt Use? Evdence from Mcroentrepreneurs n Ecuador. Sarah Pearlman 1 Department of Economcs, Vassar College, Poughkeepse, NY 12604, Unted States December 2008 Abstract: Increasng evdence shows the use of formal credt, ncludng mcrofnance, s very low amongst poor entrepreneurs. One explanaton for ths s the exstence of producton nonconvextes; poor entrepreneurs wth low levels of captal generate returns to captal that are nsuffcent to cover borrowng costs. I test ths theory usng new, cross-sectonal data on urban mcroentrepreneurs n Ecuador whch asks specfc questons about demand for credt at prevalng nterest rates. Usng semparametrc technques I estmate returns to captal and fnd that entrepreneurs wth low levels of captal generate monthly returns between 6.56% and 13.78%, whle entrepreneurs at hgher levels of captal generate returns between 1.03% to 4.25%. Whle I cannot rule out producton noncovextes altogether, these results provde lttle evdence of ther exstence at low levels of captal. I also fnd that of the large number of entrepreneurs who say they have no demand for a loan wth a 20%, a majorty generate returns hgher than the borrowng rate. Ths suggests the nablty to afford prevalng nterest rates s lkely not a prmary drver of low credt demand. JEL Classfcatons: O12, O16, O54 1 I am grateful to Masahro Shoj, Sarah Bohn, Davd McKenze, brownbag semnar partcpants at Vassar, and semnar partcpants at Bard for comments. Tel.: ; fax E-mal contact: sapearlman@vassar.edu. 1

2 Mcroenteprses, non-crop enterprses wth 10 or fewer employees, are ncreasngly recognzed as major generators of ncome and employment n the developng world. In Latn Amerca, the regon of focus of ths paper, mcroenteprses are estmated to account for 20% of GDP and anywhere from 30%-50% of total urban employment 2. Recent surveys also reveal that the use of formal credt by mcroenterprses s very low. For example, n a 2004 survey of mcroentrepreneurs n Ecuador, less than 5.6% report usng formal credt to start ther enterprses and less than 2% report usng formal credt for on-gong operatons. In many cases low formal credt use exsts despte efforts to ncrease access, partcularly n the form of mcrofnance. Indeed, many mcroentrepreneurs reman unserved by mcrofnance nsttutons despte rapd expanson of the ndustry. For example, n Kenya and Tanzana t s estmated that only 1% of all mcroentrepreneurs have any type of mcrocredt (CGAP, 2000). In Peru, the estmate s 5% (Berger, 2003), n Brazl the estmate s 2% (BNDES, 2002), and n Bolva, one of most penetrated mcrofnance markets n Latn Amerca, the estmate s only 28% (IDB, 2004). One explanaton for low credt use s that poor entrepreneurs generate returns to captal that are lower than the cost of borrowng. They smply can t afford formal credt at prevalng nterest rates. Ths explanaton corresponds wth the exstence of nonconvex producton technologes, n whch returns to captal are low at low levels of captal and ncrease only after captal rses above a certan threshold. Entry costs for certan types of enterprses, for example, are one cause of nonconvextes 3. In the presence of nonconvextes poor entrepreneurs can be shut out of credt markets as they lack the captal to meet collateral requrements and the returns to cover borrowng costs. Several theoretcal papers have shown that n the presence of producton nonconvextes credt constrants prevent households from engagng n hgh yeld enterprses (Banerjee and Newman 1993, Lloyd-Ells & Bernhardt 2000). The result s a poverty trap as poor entrepreneurs lack the means to ether borrow or save ther way nto hgher levels of captal and thus hgher earnngs. Whle there s some emprcal evdence that fnancal constrants prevent entry nto entrepreneurshp and some enterprses from reachng an effcent scale (Paulson and Townsend, 2005), recent work on returns to captal for poor entrepreneurs fnds lttle evdence of producton nonconvextes at low levels of captal. For example, McKenze and Woodruff (2006), usng survey data on mcroentrepreneurs n Mexco, fnd returns to captal n the order of 15% per month for 2 IADB 2003, Fajnzylber, Maloney, Rojas 2006, Maloney For example, nonconvextes would arse f t were necessary to acqure certan types of machnes, vehcles, or bulky nventory to generate hgh returns. 2

3 captal levels below $200. Udry and Sanagol (2006) fnd returns n the nformal sector n Ghana that range from 30% to 250% a year, dependng on the technology. Fnally, Woodruff, McKenze and del Mel (2007), usng expermental data from Sr Lanka fnd real returns n the order of 5.7% a month for low levels of nvested captal. Ths suggests that many poor entrepreneurs generate returns to captal that are well above formal borrowng costs and would be able to ether borrow or save ther way nto hgher levels of wealth and ncome. Ths paper contrbutes to the debate over the exstence of producton nonconvextes and ther role n formal credt use usng new cross-sectonal survey data on mcroentrepreneurs n Ecuador. The data, a product of the SALTO Ecuador project, can contrbute unquely to the debate for two reasons. The frst s that they are representatve of all urban mcroentrepreneurs n Ecuador, reducng some concerns about external valdty that arse wth expermental data and allowng for more general statements about credt behavor. The second s that because the SALTO Ecuador project s man goals were to understand mcroentrepreneurs access to and use of formal fnancal servces as well as constrants facng the mcrofnance ndustry, the survey ncludes extensve questons on the use of and demand for formal credt. In partcular, the survey asks entrepreneurs f they would be nterested n a hypothetcal loan at prevalng nterest rates, a queston that helps gauge demand. These responses, when lnked wth nformaton about formal credt use and returns to captal, can shed more lght on the role producton nonconvextes play n determnng formal credt use. I estmate returns to captal for Ecuadoran mcroentrepreneurs usng semparametrc technques and fnd results are very smlar to the works cted above. For entrepreneurs wth captal between $0 and $1000, I fnd monthly returns between 6.6% and 13.8%. Ths translates nto annual returns, on a uncompounded bass, between 78.7% and 165.4%. For the poorest entrepreneurs, those wth captal of $100 of less, I fnd average monthly returns of 13.5%, approxmately 162.0% annually. I also fnd more moderate returns at hgher levels of captal. Estmated, monthly returns for entrepreneurs wth captal between $1000 and $6000 range from 1.0% to 4.3%. Whle I cannot rule out producton nonconvextes altogether, these results provde lttle evdence of them at low levels of captal. The results also show that many entrepreneurs generate returns that are well above standard mcrofnance borrowng rates. I compare the estmated returns to captal to a threshold amount (1.667%) needed to generate a 20% annual return, close to the medan nterest rate charged by mcrofnance nsttutons n the sample, and fnd that all of the entrepreneurs wth captal between 3

4 $0 and $1000 and 94% of entrepreneurs wth captal between $1000 and $8000 generate returns above the threshold. Although we don t know how returns vary month to month and thus cannot construct rsk adjusted returns, the estmates suggest that a large number of entrepreneurs n the sample lkely can afford avalable mcrofnance credt. Despte the hgh estmated returns to captal, however, almost half of the sample has no demand for a hypothetcal loan at a 20% annual nterest rate. The majorty say that the man reason they don t want the loan s that the 20% nterest rate s too hgh. Whle lack of demand could reflect avalablty of other external and nternal fundng sources, responses to other survey questons suggest ths s not always the case. For example, close to one-ffth of entrepreneurs wth no demand for the hypothetcal loan cte lack of workng captal fnance as the major problem facng ther frm, and about 45% of these lst no other sources of credt. 34% of entrepreneurs wth no demand for the loan have never had a formal loan, have no suppler credt and have no savngs. Whle ths s not suffcent to rule out credt constrants, the responses suggest that for many of the entrepreneurs a need for short-term credt exsts, even f the demand for the hypothetcal loan s not there. In terms of the ablty to afford mcrofnance credt, I fnd that a majorty of entrepreneurs wth no demand for the 20% loan and a majorty of those who say they don t want t because the nterest rate s too hgh, lkely generate returns above the borrowng rate. Although these returns are not adjusted for rsk and whle I cannot rule out cases where entrepreneurs ext formal credt markets because they feel they wll be rejected due to unobservable trats, lke entrepreneural skll, the lack of demand for mcroloans by many mcroentrepreneurs who seemngly can afford them ndcates a degree of voluntary excluson. Indeed, one of the conclusons of the authors of the SALTO survey s that perhaps the most mportant challenge to MFIs n Ecuador s to overcome the mcroentrepreneur s resstance to usng credt. (Magll and Meyer 2005) The results ndcate that low returns to captal probably are not a sgnfcant cause of low formal credt use and that efforts to ncrease use amongst poor entrepreneurs may requre more than makng loanable funds avalable. The paper proceeds as follows. Secton II descrbes the data. Secton III descrbes the semparametrc and parametrc technques used to estmate returns to captal and presents estmates for the entre sample. Secton IV further nvestgates the reasons for low demand for formal credt. Secton V ncludes robustness checks for the general estmates. Secton VI concludes. 4

5 Secton II: The Data The data used n ths paper come from the SALTO Ecuador project 4, a cross-sectonal survey of Ecuadoran urban mcroentrepreneurs conducted from March to August The survey s natonally representatve of all urban mcroentrepreneurs n Ecuador and contans over 17,000 ndvduals, an estmated 3.8% of urban mcroentrepreneurs n the country. The purpose of the study was to provde nformaton about the characterstcs of the mcroenterprse sector, partcularly use of formal fnancal servces. The man goals were to understand the extent to whch mcroentrepreneurs have access to and use fnancal servces, dentfy constrants mcroentrepreneurs face n accessng formal sector fnancal servces, and provde a framework through whch donors and mcrofnance nsttutons could plan more effectve programs and expand outreach. (Magll and Meyer 2005) As such the survey contans extensve questons about knowledge of formal lenders, use of these nsttutons, and demand for loans at prevalng nterest rates. Ths makes the dataset useful for drawng general conclusons about formal credt use amongst poor entrepreneurs and how use mght be lnked to returns to captal. In the analyss I restrct attenton to mcroentrepreneurs between the ages of 18 and 65. I also trm the data to remove frms whose profts and/or captal stock were more than 2 standard devatons away from the mean were removed. Ths left a sample of 12,036 mcroenterprses 5. Of these, 49.4% were owned by women and 66.0% were nformal. 21.0% have full tme employees and 10.9% part tme employees 6. Approxmately 68% of these employees are famly members. The majorty of enterprses are n the retal sector and lkely have hgh workng captal needs n the form of nventory. The medan busness duraton s 5.8 years and roughly one-ffth of the enterprses are new, wth duratons of 2 years or less. Of entrepreneurs wth new busnesses, only 23% say they operated another enterprse pror to the current one. The majorty moved nto entrepreneurshp from wage labor or unremunerated household work, reflectng a hgh level of entry nto the mcroenterprse sector. 4 See for the data, documentaton, and detals of the survey. SALTO stands for Strengthenng Access to Mcrofnance and Economc Lberalzaton. 5 Consderng only entrepreneurs age removes 1,172 observatons. Removng enterprses wth monthly profts more than 2 standard devatons away from the mean removes 2,774 observatons. Removng enterprses wth enterprse assets more than 2 standard devatons away from the mean removes another 521 observatons. Another 763 are removed because they have enterprse assets greater than $8,000. Mssng observatons on the regressors removes an addtonal 486 observatons. 6 Ths ncludes pad and unpad part-tme and full-tme employees. Unfortunately the SALTO survey does not ask respondents to outlne whch employees are pad and whch are unpad, nor how many hours employees spend at the enterprse each week. 5

6 The key varables of nterest for the analyss are profts and captal 7. For profts entrepreneurs are asked to report the monthly ncome the household receves from the enterprse n queston. Summary statstcs, cted n US dollars, are presented n table 1. The medan monthly proft s $160, approxmately $1,920 a year, whle the average monthly proft s $222.2, approxmately $2,666 a year. For reference, the World Bank estmated that average per capta ncome n Ecuador n 2003 was $1,790, slghtly below that of the medan urban mcroentrepreneur 8. For captal entrepreneurs are asked to gve the approxmate dollar value they have nvested n four categores: vehcles and machnery; equpment and tools; merchandse or nventory; and furnture, nstallatons, other adaptatons. Total captal s the sum of these four categores 9. Entrepreneurs are asked only to lst assets used n the enterprse, although gven that many entrepreneurs operate n ther homes, some mxng of household and enterprse assets lkely occurs. Whle entrepreneurs were also asked the value of land, gven the urban nature of the sample (only 5% of entrepreneurs report land values), the greater chance of msvaluaton gven shallow land markets, and greater potental for mxng of household and busness uses, I do not nclude land n the total values for captal. In general the SALTO data provde less detal on captal than other surveys, prncpally the Mexcan Natonal Survey of Mcroenterprses (ENAMIN). Dssmlar to the ENAMIN survey, entrepreneurs are not asked about the ownershp status of the assets, makng t mpossble to dstngush between owned and rented captal. They are also not asked to report the current resale value of each asset, leadng to greater concerns about measurement error n the captal values. The trade-off, however, s more nformaton about formal credt use. The concern about measurement error wth captal stock values s addressed n Secton V. Informaton on the dstrbuton of captal stock values and the values as a percentage of monthly earnngs s provded n table 1. The overall medan captal stock s $500. Ths s 3.12 tmes general medan monthly profts, mplyng that an entrepreneur generatng a medan level of monthly profts would be able to fnance a captal purchase of $500 n 3.12 months. Table 1 also llustrates large dfferences across ndustres. For example, the medan captal stock n the transportaton ndustry s more than ten tmes that n the hosptalty ndustry (restaurants, food stalls, lodgng) and 7 By the tme of the survey Ecuador had dollarzed, so all values are quoted n USD. 8 In 2001 the World Bank estmated that average monthly labor ncome n the nformal sector was $147.8 for ndependent workers and $ for employees (Magll and Meyer). If we account for nflaton of 10%, ths means that the average mcroentrepreneurs s lkely slghtly better off than the average wage laborer. 9 Whle the SALTO survey does not explctly separate owned and rented captal, gven the way the queston s asked, t s doubtful entrepreneurs ncluded rented captal n the total value. 6

7 almost fve tmes that n the retal ndustry 10. Captal for transportaton also takes much longer to fnance. It takes an entrepreneur earnng the medan amount n transportaton more than a year to self fnance the medan amount of captal. Ths compares to 1.10 months for an entrepreneur n the hosptalty ndustry and 2.5 months for an entrepreneur n retal. On the whole, however, captal stocks are not a large percentage of monthly earnngs. For most ndustres entrepreneurs n the 10 th earnngs percentle could self-fnance captal stock purchases n the 25 th percentle n less than half a year. Whle these calculatons do not account for household uses of mcroenterprse profts and potental dffcultes n savng, t mples that barrers to entry for most types of enterprses are low. Table 1: Captal Stock Dstrbuton 10 th Percentle Earnngs Medan Earnngs Percentles of Owned Captal Stock (US$) Captal Stock Percentle Captal Stock Percentle Industry N 10 th 25 th Medan 75 th 90 th 10 th 25 th 25 th 50 th Constructon ,500 4, Manufacturng ,500 3,000 5, Personal Servces ,750 3, Professonal Servces 1, ,100 3, Repar ,000 2,500 4, Hosptalty 1, , Retal and Wholesale 6, ,210 3, Transportaton ,510 5,000 7, All Industres 12, ,600 4, Ths ncludes small kosks, market stalls, and street vendors sellng thngs other than food, among others. 7

8 II.B: Formal Credt Markets and Formal Credt Use The SALTO project was desgned to gather nformaton about the use of mcrofnance servces, manly credt, and examne challenges facng the ndustry n Ecuador. At the tme of the survey there were 36 regulated nsttutons n Ecuador that offered mcroloans 11, ncludng publc sector development banks, prvate sector banks wth mcrofnance programs, fnanceras 12, credt unons, and non-governmental organzatons, whch nclude smaller MFIs. One of the nsttutons, Banco Soldaro, s one of the largest MFIs n Latn Amerca and, at the tme of the survey, had over 100,000 clents 13. Entrepreneurs n the sample who receved mcroloans n the past year from these nsttutons provde nformaton about the structure of typcal loans. Terms ranged from one month to 96 months, wth the majorty of loans havng a term of one year or less. 75% of the repayment schedules were monthly, whle 20% were ether weekly or bweekly. Thus smlar to most mcrofnance loans, the credt offered s largely short term and requres frequent repayment. Most of the mcroloans are ndvdual rather than group loans, reflectng a trend n the mcrofnance ndustry towards the latter. Annual nterest rates ranged from 10% to 70%, wth the medan nterest rate equal to 17%. Medan loan sze was $1000, whch s twce the sze of the medan captal stock for the entre sample, but only half the sze of the medan captal stock for entrepreneurs who have ever used mcrocredt 14. What quckly emerges from the data s that the use of formal credt by mcroentrepreneurs s very low. Only 5.6% of entrepreneurs n the trmmed sample report usng formal loans to start ther busness and only 1.9% lst formal credt as a major source of current enterprse fnancng. Mcroentrepreneurs report largely relyng on personal savngs and retaned earnngs to fnance ther busnesses. Overall 27.5% of entrepreneurs have ever had a formal loan. In the past 12 months only 15.5% of entrepreneurs appled for a loan, wth over 80% applyng to only one nsttuton 15. Of 11 Mcrocredt s defned by the Ecuadoran Bank Superntendence as a small loan not backed by regular ncome lke a salary. 12 Regulated fnancal ntermedares that have lower captal requrements and lack some of the powers of commercal banks (Magll and Meyer, pg. 87) 13 Clent nformaton for Banco Soldaro and other Latn Amercan MFIs as of year-end 2004 ( 14 Unsurprsngly those who have had a formal loan are wealther than those who never have. The medan captal stock for the 3,817 entrepreneurs who have ever had a formal loan s $1900. The medan amount for the 9,984 entrepreneurs who have never had a formal loan s $400; a sgnfcant dfference. 15 Of those who appled for a loan n the past year approxmately 21% dd not receve the full amount they requested. 0.01% were rejected outrght, recevng $0. The rejecton rate could be hgher, however, as 17 ndvduals who appled for a loan last year dd not answer questons about the amount requested and the amount receved. We also don t know f other entrepreneurs who appled and were rejected elected not to provde ths nformaton to ntervewers. 8

9 those who have ever had a formal loan, 56% report applyng for one n the past 12 months, showng that formal credt s not an ongong part of operatons for many frms 16. The most common reasons gven for not havng appled for a loan n the past year are: a lack of need; desre to not become ndebted; and nterest rates that were too hgh. The low use of formal credt led the survey authors to conclude that n spte of the large expanson of the Ecuadoran mcrofnance ndustry n recent years, t has had lttle mpact on most mcroenterprses. (Magll and Meyer 2005) The survey also attempts to gauge demand for mcrocredt by askng entrepreneurs f they would be nterested n a loan at a 20% annual nterest rate, close to prevalng rates. 55% of the sample sad yes. 45% of the sample sad no. Over half of the entrepreneurs who have no demand for the loan say the man reason s that the 20% annual nterest rate s too hgh. Ths s followed by not wantng to ndebt oneself (30.3%) and, at a dstant thrd, no need (7.9%). The order of responses suggests that some entrepreneurs feel they cannot afford the loan. The goal of ths paper s to assess whether or not nablty to afford formal credt explans the low level of use. Secton III: Estmatng Returns to Captal III.A. Estmaton Strategy To estmate returns to captal t s necessary to model the relatonshp between profts and captal. Let the monthly earnngs of mcroenterprse, K the level of captal used by mcroenterprse, and X a vector of other factors that nfluence enterprse earnngs, ncludng characterstcs of the entrepreneur and the enterprse. To allow returns to vary across the captal stock I leave the functonal form of captal unspecfed, lettng profts take the followng partal lnear form 17 : X ' f ( K ) (1) Estmaton of the frst dervatve of the captal functon, f ˆ'( ), provdes estmated margnal return to captal. There are two ways to proceed. The frst s to leave the functonal form of f K ) unspecfed and estmate (1) usng semparametrc technques. The advantage of ths ( strategy s that we can reman agnostc about what returns to captal look lke. By allowng f K ) to assume any form (wth mnor restrctons), we can see f and how the margnal return to captal K ( 16 Ths s nterestng gven that surveyed entrepreneurs say the man use for formal loans s workng captal, whch cannot always be fnanced va nternal funds or nformal credt, requrng external fnance from formal sources. 17 Note that f captal enters the equaton lnearly, returns to captal wll be constant at all levels of captal snce f (K) s just the constant β K. 9

10 changes over the captal stock. The cost of ncreased flexblty, however, s precson as allowng the data to determne the functonal form generates estmates wth greater varance. As a result, t s useful to assume a functonal form for f K ), usually a hgher order polynomal, and compare these ( estmates to the semparametrc ones. Specfcaton tests can gve gudance as to the approprateness of one model relatve to the other. I start wth the semparametrc model and estmate t usng the two step dfferencng method of Yatchew (1997, 2003). The frst step s to estmate the parametrc component. Ths s done by sortng the data n ncreasng order by the varable that enters the model nonparametrcally (captal), dfferencng the data by order m, and weghtng the dfferences wth weghts d, d1,..., d. 18 Optmal 0 m dfferencng weghts satsfy condtons that ensure that the nonparametrc component of the model s dfferenced out as the sample sze ncreases and that the transformed resdual has varance Followng the dfferencng, equaton (1) becomes: m j 0 d j j m j 0 d j x 1 1, j m j 0 d j 2 x 2, j m m... d d (2) j W, j j 0 j 0 j j Where estmates yelds x 1,2,,, W are the W components of X. OLS estmaton of (2) yelds parameter ˆ dff. I use a dfferencng order of 4 and Yatchew s optmal dfferencng weghts. Ths ˆ dff that acheve 88.9% effcency relatve to non dfferenced OLS estmates (Yatchew 2003). Equaton (1) s now rewrtten usng dff ˆ dff and estmated usng nonparametrc technques 20. X ' ˆ f ( K ) u (3) I estmate (3) usng the locally weghted lnear regresson method outlned by Fan. Ths estmates the unknown functon s value at a gven K K0 by runnng a weghted lnear regresson n an area around K 0. The weghts are determned by a kernel whle the area, and thus the number of 18 There s also the restrcton that the frst dervatve of nonparametrc component be bounded by a constant, L. Ths ensures that for sequental Ks, the functonal values are suffcently close. 19 The dfferencng weghts satsfy the followng condtons: 0 j m 0 2 d ; d 1 j j m 0 j. The frst condton ensures that the nonparametrc term s dfferenced out, gven the restrcton that the frst dervatve s bounded. The second condton normalzes the dfferenced error terms such that the varance equals that of the error term tself. 20 The consstency of dfferenced parameter estmates allows for ths transformaton 10

11 observatons n any gven local regresson, s determned by the bandwdth. Fan shows that locally weghted lnear regressons are preferable to kernel smoothers due to boundary concerns. To understand how local lnear regresson yelds estmates of the functon and ts frst dervatve, consder some K n a neghborhood of K A frst order Taylor expanson of K 0 yelds 22 : f ( 0 K around K ) f ( K0) f '( K0) *( K K ) (4) Lettng f K ) and f ( K ) equal coeffcents allows (4) to be wrtten as: ( 0 ' 0 Let f ( 0 K ) 0 ( K K ) (5) ~ (the left hand sde of (3)). Regressng ~ on ) f ( K yelds estmates of the X' ˆ dff coeffcents 0 and 1. The local nature of the regresson comes from only usng observatons located wthn the bandwdth h. The kernel, KER, assgns each observaton K wthn bandwdth h a weght dependng on ts dstance from K 0. ˆ 0 fˆ( K 0 ) and ˆ 1 f ˆ'( K 0 ) are thus the solutons to a local mnmum problem, and the setup s weghted least squares. n 1 ~ [ (6) 2 0 1( K K0)] * KERh ( K K0) To estmate (6) I use an Epanechnkov kernel. Optmal bandwdth s chosen usng cross-valdaton (Yatchew 1998) 23. Overall the local polynomal smoother s convenent because t yelds drect estmates of f ˆ'( ) =the margnal returns to captal. Other smoothers that calculate local means K 0 necesstate fndng the frst dervatve numercally, whch generally produces noser estmates. For comparson I also estmate a parametrc model, assumng a fourth order polynomal as the functonal form for f K ). Results are presented wth those from the semparametrc ( estmaton. A fourth order polynomal was chosen over ffth and thrd order polynomals because t 21 Ths dscusson follows Guterrez et. al closely. The estmates are done usng the locpoly command n Stata. 22 Hgher order local polynomals lead to hgher order Taylor expansons. Ths also generates estmates of hgher order dervatves of the functon n queston, f(k). 23 Optmal bandwdth s crtcal n nonparametrc estmaton due to the nherent tradeoff between bas and varance n the estmators. Smaller bandwdths follow the data more closely, yeldng less based but more varable estmates. Larger bandwdths ncorporate more observatons n each estmate, provdng less varable but more based estmates. 11

12 performed better n specfcaton tests 24. Ths ndcates that the fourth order polynomal fts the data better than the other polynomals. III.B. Covarates Before dscussng estmaton results t s necessary to outlne the components of X, the varables that enter the model lnearly. These are factors other than captal that nfluence enterprse profts and nclude characterstcs of the entrepreneur and of the enterprse. For characterstcs of the entrepreneur the dffculty s that the most mportant characterstc, entrepreneural skll, s unobservable. Furthermore, gven the cross sectonal nature of the data and lack of good nstruments, the most vable strategy s to control for skll usng observable proxy measures. I start wth educaton and experence, whch are frequently used to measure skll (Paulson and Townsend 2005, Gne and Townsend 2004, McKenze and Woodruff 2006). Educaton s measured by dummy varables for four categores of educatonal attanment by the entrepreneur; less than prmary educaton, prmary educaton, secondary educaton, and college educaton. Experence s measured by the amount of tme the enterprse has been n operaton and ts square, along wth the age of the entrepreneur and age squared. Fnally, I nclude controls for martal status and for gender, as both may play a role n determnng profts 25. In addton to entrepreneur characterstcs, I nclude labor used by the frm, measured by the total number of famly and non-famly full-tme and part-tme employees. The labor varables serve multple purposes. The frst s the need to control for labor n estmatng profts, so as not to nflate returns to captal f there are complementartes between labor and captal. The second s that due to the hgh percentage of total employees who are famly members (around 68%) and the fact that many famly employees lkely are unpad, t s necessary to control for the returns to unpad labor that may accrue to the entrepreneur n the form of profts. The thrd, whch stems from lterature on the sze dstrbuton of frms, s that total labor may measure entrepreneural or manageral skll. For example, both Lucas (1978) and Jovanovc (1982) present models n whch resources are allocated across enterprses accordng to manageral ablty. In these models frms wth more sklled entrepreneurs grow n sze whle those wth less sklled entrepreneurs do not, drvng the predcton 24 The specfcaton tests the followng: under the null that the parametrc model s the true model, the test statstc V follows a standard normal dstrbuton. 1/ ( mn) ( sres sdff )/ sdff varance (parametrc estmaton). 2 sdff 2 sres restrcted estmator of the resdual dfferenced estmator of the resdual varance (semparametrc estmaton). 25 McKenze and Woodruff (2006) pont out that martal status may pck up entrepreneural skll, as research has shown that marred male workers earn more, after controllng for other characterstcs, than unmarred workers. 12

13 of a postve lnk between skll and frm sze 26. To measure returns to manageral ablty other papers use total hours worked by pad employees (McKenze and Woodruff (2006)). The SALTO data do not nclude nformaton on whether or not employees are pad. I assume that all part tme and full tme employees who are not famly members are pad and that these varables capture returns to manageral skll. Fnally, I control for the type of busness, based on 8 categores, and for the provnce n whch the enterprse s located. The baselne model (Model 1) ncludes all of the controls lsted. Due to potental concerns that the controls for skll are nsuffcent I also consder two addtonal skll varables. These varables are based on the reasons entrepreneurs gve for startng ther busnesses (McKenze and Woodruff (2006)). One of the varables equals one f an entrepreneur says he/she entered entrepreneurshp because of famly tradton or because he/she can earn more than n wage employment. These responses lkely ndcate greater entrepreneural skll. The other varable equals one f an entrepreneur says he/she entered entrepreneurshp because there was nothng else avalable. Ths response lkely ndcates lower levels of entrepreneural skll. The second model, Model 2, adds these addtonal skll measures to those n the baselne model. III.C. General Estmaton Results I estmate returns to captal semparametrcally and parametrcally, usng a fourth order polynomal. Gven that optmal bandwdth wll vary dependng on the sample sze, I splt the sample nto a low captal group (captal between $0 and $1000) and a hgh captal group (captal values between $500 and $8000) and estmate returns separately for each. Ths allows greater focus on poor entrepreneurs wth low levels of captal. The cutoff for the low captal group les slghtly below the average captal stock, whch s $1238 for the trmmed sample. The cutoff for the hgh captal group les at the medan, whch s $500. The parameter estmates and standard errors for the covarates n X are shown n table 2. The standard errors for the semparametrc estmates are adjusted to account for dfferencng (Yatchew 2003) 27 and are greater than those from the parametrc estmaton. The parameter estmates, however, are smlar across all of the models and generally follow ntuton. For example, women have lower profts, on average, than men. Older, marred and more educated entrepreneurs 26 Lucas and Jovanovc present unconstraned models. Relevant barrers to the allocaton of resources such as credt constrants or an urban equvalent to a harvest labor constrant do not apply. 27 Standard errors are multpled by (1+1/2m)^0.5 to account for the reduced effcency of the dfferenced estmates (Yatchew 2003). 13

14 have hgher profts than younger, unmarred and less educated entrepreneurs. Enterprses wth more tme n operaton and more full tme employees and part tme employees who are not famly members have hgher profts. The estmated coeffcents on employees and duraton support the story that more proftable busnesses are the ones that reman n operaton and are the ones that grow over tme. The addtonal skll controls have lttle mpact on the values of the estmates. 14

15 Table 2: Parameter Estmates for the Lnear Porton of the Model 28 Parametrc Estmates Sem-Parametrc Estmates Low Captal ($0-$1400) Hgh Captal($500-$8500) Low Captal ($0-$1000) Hgh Captal($500-$8000) Noncaptal varables 29 Model 1 Model 2 Model 1 Model 2 Model 1 Model 2 Model 1 Model 2 Owner a woman (3.680) (3.689) (5.886) (5.883) (3.903) (3.917) (6.256) (6.255) Owner has prmary educaton (8.400) (8.387) (20.731) (20.715) (8.789) (8.781) (22.142) (22.129) Owner has secondary educaton (8.679) (8.669) (20.862) (20.847) (9.110) (9.104) (22.294) (22.280) Owner has college educaton (10.058) (10.043) (21.632) (21.615) (10.610) (10.601) (23.085) (23.071) Duraton of busness (0.606) (0.605) (0.968) (0.968) (0.642) (0.641) (1.026) (1.026) Full tme employees, famly (3.089) (3.084) (3.888) (3.889) (3.362) (3.359) (4.124) (4.123) Full tme employees, nonfamly (4.374) (4.364) (3.975) (3.973) (4.839) (4.834) (4.194) (4.129) Part tme employees, famly (3.948) (3.941) (4.793) (4.800) (4.374) (4.370) (5.077) (5.074) Part tme employees, nonfamly (5.675) (5.649) (5.732) (5.730) (6.136) (6.133) (6.037) (6.036) Entered busness to No Yes No Yes No Yes No Yes ncrease ncome/famly Entered busness due to No Yes No Yes No Yes No Yes lack of better optons No. observatons Standard errors show n parenthess. Estmators from the sem-parametrc dfferenced equaton are scaled by (1+1/2m)^0.5, as the standard errors are larger due to the dfferencng. 29 Other covarates nclude martal status, age, age squared, age of busness squared, busness category fxed effects and provnce fxed effects. 30 In dfferenced equaton there s no constant, whch means the left out group, no educaton, has a separate dummy varable. Ths group s absorbed n the constant term n the parametrc, non-dfferenced estmaton. 15

16 Monthly return(%) 0 5 Monthly return(%) 0 5 The estmated returns to captal from the parametrc and semparametrc models are dsplayed n fgure 1. A sold horzontal lne s set at 1.667%, the monthly amount needed to generate an annual return of 20%, assumng no compoundng 31. Ths lne dstngushes entrepreneurs who probably can afford a 20% nterest rate and those who probably cannot. The nonparametrc estmates are less smooth than the parametrc ones, but the results generally le wthn the same range and follow smlar patterns. Ths mples that the fourth order polynomal s a decent approxmaton of the functonal form for captal. Fgure 1: Estmated Returns to Captal Parametrc returns, $0-$1400 Captal Sem-Parametrc returns, $0-$1000 Captal Captal Model 1 Model 2 Parametrc Returns, $500-$8400 Captal Captal Model 1 Model 2 Sem-Parametrc returns, $500-$8000 Captal Captal Model 1 Model Captal Model 1 Model 2 Several conclusons arse from the results. The frst s that returns to captal are qute hgh for entrepreneurs wth low levels of captal. For entrepreneurs wth captal between $0 and $1000, sem-parametrc estmates of model 2 fnd monthly returns that range from 6.56% to 13.78%. Ths mples approxmate annual returns between 78.7% and 165.4%, assumng no compoundng. Consderng only the poorest group, entrepreneurs wth captal of $100 or less, the average monthly return s 13.5%. These results are smlar to those found by McKenze & Woodruff (2006), Udry and Anagol (2006), and Woodruff, McKenze and del Mel (2007) and provde further evdence that some poor entrepreneurs generate very hgh returns to captal. The second concluson s that returns are more moderate for hgher levels of captal. For captal values between $500 and $8000, semparametrc estmates of model 2 fnd monthly returns 31 Compoundng would assume that entrepreneurs re-nvest all monthly profts nto the busness. However, over 70% of entrepreneurs say that the frst use of enterprse profts s for household expenses, and only 47% of entrepreneurs lst re-nvestment as one of the three man uses of profts. Thus an approprate benchmark for monthly nterest rates would be one not based on compounded. 16

17 that range from -0.93% to 4.25%. Ths translates nto approxmate annual returns of 51.0% for the upper bound. Consderng the group wth captal between $1000 and $5000, monthly returns range from 1.03% to 4.25%. Ths translates nto approxmate annual returns between 12.4% and 51.0%. It s mportant to note, however, that the results do not provde evdence of dmnshng returns to captal, as returns fall n certan ranges ($300-$500) and rse n others ($600-$1000). I can only say that returns are generally lower for the hgh captal group than the low captal one. The thrd concluson s that there s lttle evdence that producton nonconvextes exst at low levels of captal. The hghest estmated returns n the sample come from entrepreneurs who have captal of $100 or less, whle the lowest returns come from entrepreneurs who have captal over $7000. Nonconvextes may stll exst over certan ranges. For example, margnal returns are constant n the range of $500 to $900 and rse at $900, suggestng potental threshold effects. However, there s lttle evdence of ther exstence at low levels of captal. Many poor entrepreneurs generate returns that are well above borrowng rates, whch means they should be able to ether borrow or save ther way nto hgher levels of captal. Ths provdes lttle evdence that low margnal returns are a major cause of poverty traps for many poor entrepreneurs (although I cannot rule out the case of poverty traps caused by other factors). The fnal concluson s that a hgh porton of mcroentrepreneurs generate returns to captal above standard mcrofnance borrowng rates. All of the entrepreneurs n the low captal group and 94% of the entrepreneurs n the hgh captal group have estmated monthly returns above standard borrowng rates. Ths suggests that nablty to afford formal credt s not a major drver of low formal credt use. I explore ths ssue more deeply n secton IV. Gven the hgh varance assocated wth nonparametrc estmates, t s useful to show the estmated returns to captal wth confdence ntervals. To do ths I carry out 100 bootstrap replcatons of the semparametrc estmaton of Model 2 and plot the 95% pontwse confdence nterval wth the orgnal estmates. Ths s done separately for the low captal and the hgh captal samples and the results are shown n fgure 2. The range for the estmates usng the optmal bandwdth s wde, reflectng the hgh varance assocated wth nonparametrc estmates. For the low captal group, however, the lower bound of the confdence nterval les above 1.667% for all of the observatons. Ths provdes further evdence that low returns to captal are not a domnant explanaton for low formal credt use. 17

18 Fgure 2: Estmated Returns wth Confdence Intervals Sem-Parametrc returns, $0-$1000 Captal Optmal bandwdth Captal Model 2 2.5% confdence bound 97.5% confdence bound Sem-Parametrc returns, $500-$8000 Captal Optmal bandwdth Captal Model 2 2.5% confdence bound 97.5% confdence bound Secton IV: Demand for Formal Credt Ths secton examnes lnks between returns to captal and formal credt use n more detal. The SALTO data are apt for explorng the lnks for two reasons. The frst s that the data are representatve of all urban mcroentrepreneurs n Ecuador, reducng some concerns about external valdty that arse wth expermental data. The second s that the survey ncludes questons about demand for formal credt at prevalng nterest rates, allowng for some separaton of supply and demand as determnants of credt use. Ths s mportant because oftentmes the researcher only sees the ncdence of formal credt and does not know f low credt use s due to lack of supply (nvoluntary excluson from credt markets), lack of demand (voluntary excluson) or both (mutual excluson). Voluntary excluson s often gnored as a potental explanaton for low credt use, but recent evdence shows that non-trval numbers of potental mcrofnance borrowers sometmes have lttle nterest n the loans (Johnston and Morduch, 2007), suggestng t should be consdered. The nformaton on demand for a hypothetcal loan at prevalng nterest rates can shed some lght on voluntary excluson from formal credt markets amongst Ecuadoran mcroentrepreneurs. The SALTO survey attempts to gauge demand for formal credt by askng entrepreneurs f they are nterested n a formal loan wth a 20% annual nterest rate, close to prevalng nterest rates. If entrepreneurs say yes they are asked how much they would want to borrow. If they say no they are asked the reasons why. Almost half of the entrepreneurs (45%) say they are not nterested n the loan. 54% of these entrepreneurs say the man reason s because the 20% annual nterest rate s too hgh. Ths s followed by not wantng to ndebt oneself (30.3%) and, a dstant thrd, no need (7.9%). 18

19 The hgh percentage of entrepreneurs wth no demand for the hypothetcal loan s surprsng gven the assumpton that many are credt constraned. Perhaps, however, many entrepreneurs are not, n fact, credt constraned and have no need for addtonal credt. Responses to other questons regardng fnancng sources and busness needs, however, suggest ths s not a compellng explanaton. For example, 18% of those wth no nterest n the loan cte workng captal constrants as a major problem facng the frm. 45% of these, n turn, have never had a formal loan, have no suppler credt, and have no savngs. These entrepreneurs clearly appear credt constraned and we would expect greater demand for the hypothetcal loan. Another example s that 34% of those wth no nterest n the loan have never used formal credt, do not have suppler credt and do not have savngs. Whle ths s not a comprehensve pcture of credt avalable to these entrepreneurs (SALTO does not have nformaton on loans from moneylenders, famly and frends or ROSCAs), t s suffcent to suggest lmted access to external or nternal fnance. Table 3 provdes more nformaton on fundng sources and workng captal constrants for the group wth no nterest n the loan. I only consder the sample of entrepreneurs n the return to captal estmates. The SALTO survey does not have comprehensve questons of frm fnance, but we do know f an entrepreneur has ever used formal credt, currently uses suppler credt (a major source of nformal fnance) and has savngs, ether formal or nformal. These responses gve an dea of the avalablty of external and nternal fnance and hnt at whether or not credt constrants are bndng. Snce there may be dfferences across poorer and wealther entrepreneurs, I dvde the sample nto two groups; those wth captal below $500 and those wth captal above $500. I also show separate results for retal enterprses, whch need contnual supples of nventory and lkely have a greater need for workng captal fnance than other types of frms. 19

20 Table 3: Entrepreneurs who do not demand hypothetcal loan All Enterprses Retal Enterprses Captal Captal Below $500 Above $500 Below $500 Above $500 No Demand for 20% Interest Rate Loan Percent of total group (observatons) 44.2% (2,855) 46.5% (2,599) 42.5% (1,631) 45.7% (1,315) Of whch: Lst lack of workng captal 19.3% (552) 17.5% (454) 23.5% (384) 19.6% (258) fnance as man problem facng frm Have never had a formal loan 81.4% (2,138) 56.7% (1,432) 80.6% (1,195) 57.5% (727) Dd not apply for a formal loan 90.3% (2,579) 76.1% (1,978) 89.4% (1,459) 75.7% (996) last year Do not have suppler credt 70.3% (2,008) 56.8% (1,476) 60.9% (993) 41.5% (546) Do not have savngs, formal or 78.3% (2,235) 50.9% (1,324) 77.9% (1,270) 52.3% (688) nformal Have never had a formal loan, do not have suppler credt or savngs 46.1% (1,317) 21.5% (558) 38.5% (629) 16.3% (214) Of those who cte lack of workng captal as a major problem facng the frm % wth no cted external or nternal fnance 44.4% (245) 20.9% (95) 40.1% (154) 15.5% (40) The frst thng to note s that whle demand for the hypothetcal loan s hgher amongst poorer entrepreneurs than amongst wealther ones, the dfference s slght. 55.8% of the low captal group demands the hypothetcal loan, as compared wth 53.5% of the hgh captal group. Indeed, demand for the hypothetcal loan dffers very lttle across entrepreneurs at dfferent levels of captal. For example, 52.7% of entrepreneurs wth captal of $100 or less demand the loan, as compared wth 50.3% of entrepreneurs wth captal of $2000 or more, and 45% of entrepreneurs wth captal of $5000 or more. The second thng to note s that a surprsngly hgh percentage of entrepreneurs appear to have lmted sources of fundng. 46% of entrepreneurs wth captal less than $500 and 23% of entrepreneurs wth captal above $500 have nether formal credt, suppler credt nor savngs. Whle these entrepreneurs could have fnancng from other nformal sources, lke famly and frends, moneylenders, or retaned earnngs, the numbers suggest that ample fnancng sources do not explan low formal credt demand. Ths s partcularly true for the entrepreneurs who cte a lack of workng captal fnance as the major problem facng the frm. 19.3% of entrepreneurs wth low levels of captal and 17.5% of entrepreneurs wth hgh levels of captal say lack of workng captal fnance s a major problem. Of these, 44% and 21%, respectvely, do not have formal credt, suppler credt or savngs. Why then, do these entrepreneurs have no nterest n the hypothetcal loan? 20

21 The most common reason gven for lack of demand n the hypothetcal loan s that the nterest rates are too hgh; some entrepreneurs feel they cannot afford a loan wth a 20% annual nterest rate. Table 4 examnes the ssue of affordablty. It shows medan estmated returns to captal for entrepreneurs who demand the hypothetcal loan and for those who do not. The samples are agan dvded nto low and hgh captal groups and separate results for retal enterprses are shown. I also show estmated returns to captal for those who say that the 20% annual nterest rate s too hgh. The last row shows the percentage of ths group estmated to have monthly returns to captal above ths borrowng rate (1.667% monthly). Table 4: Affordablty All Enterprses Retal Enterprses Captal Captal Medan Values shown Below $500 Above $500 Below $500 Above $500 Demand for 20% Interest Rate Loan (n) (3,599) (2,983) (2,205) (1,565) Estmated monthly return to captal 13.11% 3.80% 13.11% 3.78% No Demand for 20% Interest Rate Loan Estmated monthly return to captal 13.16% 3.70% 13.11% 3.82% Entrepreneurs who cte nterest rate beng 46.3% (1,321) 63.5% (1,651) 46.7% (761) 60.4% (794) too hgh as reason for not wantng the loan Estmated monthly return to captal 13.11% 3.70% 13.06% 3.70% Percent who have estmated returns to captal above 20% a year 100% 90.7% 100% 99.1% Surprsngly, for the low captal group, the medan estmated return to captal s hgher for the group that does not demand the loan than for the group that does. For entrepreneurs who demand the hypothetcal loan and have captal below $500, the medan monthly return s 13.11%. For entrepreneurs who do not demand the hypothetcal loan, the medan monthly return s 13.16%. Furthermore, the medan estmated monthly return for entrepreneurs n the low captal group who say the nterest rate s too hgh s 13.11%, the same as the group that demands the loan. Indeed, 100% of the low captal group that says a 20% annual nterest rate s too hgh has estmated returns to captal above ths borrowng rate. Ths falls to 90.7% for the hgh captal group, reflectng a declne n estmated monthly returns at hgher levels of captal. Overall the results show that a majorty of entrepreneurs, even those who do not want the loan and say the nterest rate s too hgh, lkely can afford prevalng nterest rates. One caveat s that the estmates of annual returns assume constant monthly returns and do not account for fluctuatons across months. Some entrepreneurs may have varable returns, n 21

22 whch case one month of hgh returns may not mean that they could cover a 20% nterest rate over the year. Indeed, hgh returns could represent compensaton for hgher rsk, n whch case a monthly return does not tell us about ablty to make regular loan payments throughout the year. Rsk s lkely an mportant part of ths story, but gven the cross-sectonal nature of the SALTO data t s dffcult to learn much about the varance of monthly returns. Panel data would allow for greater exploraton of varance, as well as greater control of skll, but I am unaware of any natonally representatve panel data sets on mcroentrepreneurs. Another caveat s that I cannot rule out cases where entrepreneurs ext formal credt markets because they lack collateral or feel they wll be rejected due to unobservable trats, lke entrepreneural skll 32. However, the lack of demand for mcroloans by many mcroentrepreneurs who seemngly can afford them ndcates a degree of voluntary excluson. There are potentally a large number of poor entrepreneurs who actvely choose not to partcpate n formal credt markets. The authors of the SALTO survey conclude that perhaps the most mportant challenge to MFIs n Ecuador s to overcome the mcroentrepreneur s resstance to usng credt. (Magll and Meyer, 2005). They also conclude that whle t s often assumed that there s large unsatsfed demand for credt by mcroentrepreneurs, several fndngs n the survey-especally regardng the low frequency of loan applcatons and the hgh success rate n gettng loans- cast doubt on ths assumpton. (Magll and Meyer 2005). My fndngs are n lne wth these conclusons, as t seems that an nablty to afford mcrofnance loans probably s not the man explanaton for low formal credt use exhbted by many poor mcroentrepreneurs. It s unclear f ths s due to lack of understandng about how nterest rates work, a lack of good measurement of profts, varablty of profts and dfferences n rsk averson or a desre not to become ndebted, among other factors. The overall mplcatons, however, s that low formal credt use cannot be fully explaned by a lack of avalable credt for poor entrepreneurs. Secton V: Robustness Checks There are two man concerns about the valdty of estmates of returns to captal presented n secton III. The frst s that measurement error n the captal stock leads to based estmates of the returns. 32 For entrepreneurs who appled for formal credt last year, medan captal was $1300 whle the 25 th percentle value of captal was $400. The mnmum amount was $0. Whle the medan and 25 th percentle values are hgher than for entrepreneurs who dd not apply for a formal loan last year, there are many entrepreneurs wthout formal credt wth captal values n these ranges or hgher. It s unclear, however, f they collateral would be deemed suffcent by lenders. 22