WPS6634 Policy Research Working Paper 6634 From Pawn Shops to Banks The Impact of Formal Credit on Informal Households Claudia Ruiz The World Bank Development Research Group Finance and Private Sector Development Team October 2013 Policy Research Working Paper 6634 Abstract This paper examines the effects of expanding access and accumulate more durable goods even though the to credit on the decisions and welfare of households. overall proportion of households that save went down by It focuses on the entry of Banco Azteca, the first bank 6.6 percent. These results suggest that the use of savings in Mexico targeting households from the informal as a buffer on income fluctuations declines once formal sector. Panel data suggest that informal households in credit is available. What is more, these effects vary across municipalities with Banco Azteca branches experienced households. Among informal households, those who several changes in their saving, credit and consumption never receive formal job offers have the highest decline patterns. In order to estimate the impact of Azteca’s in saving rates. The model is also used to evaluate a entry, the paper develops a dynamic model of household legislation to cap interest rates levied by formal credit choices in which the bank is endogenously selecting institutions. Simulations suggest that if the Mexican the municipalities for branch openings. The analysis government were to cap the interest rate of Azteca at the finds that in municipalities in which the bank entered, rate for traditional banks, Azteca would stop operating in households were better able to smooth their consumption the poorest and least populated municipalities. This paper is a product of the Finance and Private Sector Development Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at cruizortega@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team From Pawn Shops to Banks: The Impact of Formal Credit on Informal Households Claudia Ruiz JEL codes: O12, O16, G21 Keywords: Informal sector, formal credit suppliers, precautionary savings Sector Board: Financial Sector (FSE) Development Research Group, The World Bank, cruizortega@worldbank.org - I would like to thank Sandra Black, Miriam Bruhn, Moshe Buchinsky, Maria Casanova, Dora Costa, Pascaline Dupas, Fred Finan, Ahu Gemici, Xavier Gine, Hanan Jacoby, Adriana Lleras-Muney, Kathleen McGarry, Alvaro Mezza, Bernardo Morais, Leah Platt Boustan and Hernan Winkler for very helpful discussions and comments. I am very grateful to seminar participants at Banco de Mexico, Banco de Portugal, CNBV, Cornell University, Federal Reserve of Boston, Federal Reserve of Richmond, Federal Reserve of Saint Louis, ITAM, Maryland University, MIT/Harvard joint seminar of Development Economics, Ohio State University, RAND, University of Rochester, University of Virginia, Vanderbilt University, Washington University of Saint Louis and the DECFP group of the World Bank for very helpful comments and suggestions. Finally, I am extremely grateful to my advisor Maurizio Mazzocco for his invaluable suggestions and guidance. All views expressed in this paper are my own and should not be attributed to the World Bank. 1 Introduction Formal credit institutions are reluctant to lend to households without credit history and verifiable steady employment. These two characteristics are inherent to people employed in the informal sector, which includes any economic activity that is not taxed or monitored by the government. This is a concern in developing countries, where the fraction of people working in an informal occupation is considerably large. In 2002, for example, 55% of Mexicans belonged to the informal sector. This does not imply that usage of credit among these people is low. On the contrary, households that cannot obtain credit from banks are very active borrowers with alternative suppliers. They rely heavily on loans from relatives and friends, and on more expensive credit suppliers such as pawn shops and moneylenders. According to the Mexican Family Life Survey 2002, of all informal households in 2002, 3.7% of them used pawn shops or moneylenders’ credit and 13.8% obtained loans from friends or relatives. The goal of this paper is to analyze the effects of expanding access to credit on households’ deci- sions and welfare. To do this, I examine the opening of Banco Azteca, the first bank in Mexico that targeted households employed in the informal sector. Using panel data at the household level, I first analyze households’ saving and consumption patterns before and after the entrance of this bank. I then develop and estimate a dynamic model of household choices in which Banco Azteca selects the municipalities for branch openings. This is, to my knowledge, the first structural attempt at jointly modeling household choices and the location decisions of banks to measure the impact of access to credit. There are several advantages of using this structural approach. First, since the entry of Azteca in a given municipality is unlikely to be exogenous, estimating its impact on households’ welfare is difficult. The model deals with this issue by endogenizing the location decisions of the bank. Second, the estimated model allows to quantify the impact of the expansion of credit caused by the entrance of Banco Azteca. Third, the model can be used to evaluate alternative policies aimed at extending access to credit. In this paper, I evaluate the proposed regulation of capping interest rates of formal credit institutions. In October 2002, Banco Azteca opened more than 800 branches inside all Elektra stores that existed in the country. Some months later, various Azteca branches began relocating independently of the Elektra stores. At the time, these branches accounted for 15% of the supply of bank branches in Mexico. This new bank eliminated the proof of income requirement and hence allowed all Mexicans 2 from the informal sector to obtain bank credit for the first time. In terms of costs for borrowers, Azteca’s annual percentage rate (APR) has been significantly higher than traditional banks’: its APR in 2005 was 130% compared to 40% for regular banks. Nevertheless, it is lower than its competitors; in this same year, pawn shops charged on average an APR of 220%1 . Panel data suggest that households whose members were employed in the informal sector had significant changes in their saving and consumption patterns after an Azteca branch appeared in their municipalities. The household data consist of two waves that were collected in 2002 and 2005. By the time of the first wave, Banco Azteca had not opened its branches, but by 2005, Banco Azteca’s presence varied across municipalities. I exploit this variation over time and across municipalities of Azteca’s branches to compare household outcomes before and after the entrance of Banco Azteca. I find evidence that relative to households from municipalities where this bank did not open, informal households in Azteca municipalities were more likely to borrow from banks, less likely to obtain loans from pawnshops and less likely to hold savings. In addition, the fraction of informal households owning durable goods increased as did the value of these goods. Moreover, in munici- palities where Azteca opened, the data suggest that informal households were more likely to increase their consumption during bad economic times, such as illness of family members, unemployment or failure of the family business. Importantly, I find no substantial change in the saving or consump- tion behavior of households from municipalities where Azteca entered until 2006, suggesting that changes in these outcomes are associated to the opening of Azteca branches. The model includes important features from the Mexican economy. To capture heterogeneity among municipalities in Mexico, the economy in this model consists of M municipalities that differ in their composition of households, population size, percapita income and presence of credit suppliers. As in the Mexican credit market, households can borrow from traditional banks, pawn shops, Banco Azteca and friends or relatives. In the model, Banco Azteca is a for-profit institution that selects the municipalities for branch openings. The bank locates in municipalities where its expected profits are high enough to cover its operating costs. In this economy, credit suppliers differ in their costs, availability and requirements for clients. The APR of each institution is set to the average rate observed from the data, with pawn shops being the most expensive and friends the least costly. When borrowing from Azteca and pawn shops, households need to own collateral to back up their loans. Durable goods serve as households’ 1 Information on the APR of different institutions was obtained from the National Committee for the Defense of Users of Financial Institutions (CONDUSEF) and the Federal Bureau of Consumer Interests (PROFECO). 3 collateral and in case of default, the credit institution retains it. Traditional banks have the additional constraint of requiring their clients to belong to the formal sector. Given all these constraints, households maximize their expected lifetime utility each period by making several choices. First, households who receive job offers from the formal and the informal sectors must select their sector. Households then decide how many durable goods to own and how much savings or debt to acquire. After choosing their savings and expenditure on durable goods, the remaining resources at each period determine the consumption of the non-durable good. Consistent with the data, the quantitative results indicate that population size and income of municipalities are important determinants of the location choices of Azteca. However, access to other forms of financing matters, particularly among municipalities where Azteca just breaks even. Everything else constant, Banco Azteca is less likely to locate in municipalities where a large fraction of households have access to credit from friends and relatives. Likewise, the impact of Azteca is more modest as municipalities rely more in friends and relatives loans. According to the model results, having an Azteca branch allows households to smooth their con- sumption and to accumulate more durable goods. Providing them with access to bank credit trans- lates in lower usage of pawn shop loans and lower levels of savings. Once access to formal credit is available, the fraction of households that save declines by 6:6%: This suggests that savings were being used as a buffer on income fluctuations in the absence of credit. Consistent with the targeted pop- ulation of Banco Azteca, households who remain in the informal sector experience higher changes. But even among informal households, the effects of Banco Azteca are larger for households in the informal sector who never received a formal job offer. Since the model includes the entry and exit decisions of this bank, it is well suited to evaluate a regulation that would cap the interest rate that formal credit institutions charge. Several policy makers in Mexico have suggested this measure in order to make loans more affordable to people. However, as capping the interest rate would alter Azteca’s expected profits, this policy could have the unintended consequence of forcing Azteca out of municipalities where it is no longer profitable to maintain a branch. My results suggest that if the government were to cap Azteca’s APR to 40%, more households would obtain bank loans and the average size of these loans would be larger. However, in some municipalities households’ response would not compensate for the decrease in the APR. Simulations of this model indicate that the bank would exit from half of the currently covered mu- nicipalities. The likelihood of losing a branch is higher in municipalities with lower percapita income and smaller population. 4 Broadly, my results fit into the literature that analyzes the impacts of expanding financial ser- vices to poor households. There have been several studies using randomized settings to examine the impact of expanding access to credit. Banerjee et al. (2009) exploit the random expansion of a microfinance institution in urban India, and find that in slums where the MFI entered, borrowing from MFIs increased as did expenditure on durable goods. They also found heterogeneous effects on business creation and non-durable consumption, which depended on households’ propensity to become entrepreneurs. Crépon et al. (2011) find similar results for a randomized microcredit inter- vention in rural Morocco. On the other hand, Karlan and Zinman (2009) evaluated an individual lending program for micro entrepreneurs in urban Philippines, finding almost no effects on profits and business expansion. In addition to these papers, there is a growing literature studying the effects of expanding access to credit to low-income households in more quasi-experimental settings. Burgess and Pande (2005) exploit the implementation of a bank branch licensing rule in India to estimate the effect of rural bank branches on poverty. They find that the expansion of branches significantly decreased poverty in rural areas. Aportela (1998) examines the impact of an expansion of a savings institute across Mex- ican cities. He finds strong positive effects in the average saving rates of low-income households. In the same context as this paper, Bruhn and Love (2013) analyze the effects that the entrance of Banco Azteca had on business and employment outcomes in urban municipalities in Mexico. Their evidence suggests that relative to municipalities without Azteca, in municipalities where Azteca ini- tially entered, workers were more likely to open an informal business and benefited from higher income levels. Kaboski and Townsend (2009) and Kaboski and Townsend (2011) analyze the impact of the Thai Million Baht Village Fund program, a major microcredit initiative that was rapidly implemented in one year across all villages in Thailand. Exploiting that the same amount of funds was injected to all villages regardless of their size, they first use a reduced form approach to examine the data patterns. They then develop and estimate a structural model to predict and evaluate the impact of the program. They find sizeable effects on consumption as a result of the credit program. My paper contributes to this literature by providing one of the first attempts to estimate a model of household decisions that considers the endogenous location of credit providers. This paper is organized as follows. The next section describes Azteca’s background and outlines its business model. Section 3 describes the data set used and the patterns of households over time in municipalities with and without Azteca branches. Sections 4, 5 and 6 present the model, the model 5 solution and its estimation. Section 7 presents the estimation results and the model fit assessment. The robustness of the model and the quantification of Azteca’s impact are examined in Sections 8 and 9. The policy experiment is discussed and evaluated in Section 10. Finally, Section 11 concludes. 2 Background of Banco Azteca Banco Azteca is owned by Grupo Elektra, Latin America’s leading retail company. Elektra stores have operated in Mexico since 1957 and specialize in selling consumer electronic products, household appliances, and furniture through sales on credit schemes. Its in-store credit schemes have consisted of weekly payments at high interest rates that appeal to customers lacking alternative credit options (mainly people from the informal sector). By 2001, Grupo Elektra applied for a license to open a bank, which was approved in March 2002. In October 2002, Banco Azteca started operations inside all Elektra branches that existed in the country. After some months of operation, several Azteca branches began relocating independently of the Elektra stores. Banco Azteca’s business model closely replicates Elektra’s. Its loans are collateralized, they range from 2,000 to 20,000 Mexican pesos and their average maturity term is one year. Different from Elektra’s credit, Azteca’s loans are in the form of cash, and are not tied to the purchase of any item in the store. To obtain a personal loan, applicants visit a branch where, among other things, they state a list of belongings to be pledged as collateral for their loan. Household appliances such as refrigerators or televisions are common collateral accepted by the bank2 . To assess their value and to obtain more information about the applicant, a loan agent mounted on a motorcycle visits the applicant’s house shortly after the application is processed. After this visit, the bank decides whether or not to approve the loan. Once a loan is approved, Azteca’s clients begin making weekly payments, and those who delay receive weekly visits from loan agents requesting their payment3 . As stated by Azteca, its default rate is as low as that of traditional banks, plausibly due to its crude collection and repossession mechanisms4 . 2 In case of default, collateral is repossessed and resold at Elektra’s Bodega de Remates outlet stores. 3 Annual Report of Grupo Elektra, 2002. 4 According to Rhyne (2009), "Azteca has reportedly fired agents for crossing the line between peer pressure and public humiliation". 6 3 Data and Empirical Findings The data used to examine Banco Azteca’s entry combine three sources of information. First, the data for households come from the Mexican Family Life Survey (MxFLS), waves 2002 and 2005. At the time of the 2002 wave, Banco Azteca had not opened its branches, but by 2005, its presence varied across municipalities. MxFLS is representative at the national level and importantly, it is the only survey providing detailed information about Mexicans’ credit and savings habits during these years. I focus the analysis on households that were surveyed in both waves and in which the household head is between 18 and 65 years old. The final sample from the baseline survey consists of 5,639 households in 136 Mexican municipalities. Second, municipalities from the household data were merged with a panel dataset from the National Banking and Securities Commission (CNBV), which contains the location and year of opening of all bank branches across Mexican municipalities. With this dataset it is possible to identify the location decisions of Banco Azteca across municipalities over the period examined. Finally, aggregate variables at the municipality level such as population size and percapita income were added to examine the characteristics of municipalities attracting Banco Azteca’s branches. These variables were obtained from the 2000 Mexican Census and the Human Development Index of Mexican Municipalities5 . To first explore which factors determine the location decisions of Banco Azteca, I classify munici- palities in four groups according to the presence of Banco Azteca: i) municipalities that had an Azteca branch from 2002 to 2005 (63 of them); ii) municipalities where Azteca entered in 2002 but by 2005 the branches were no longer there (2 of them); iii) municipalities where Azteca did not enter at the be- ginning but by 2005 the bank was already there (4 of them); iv) and finally, municipalities that never had an Azteca branch in these years (67 of them). In table 1, I examine the distribution of population size and percapita income for each group. The table also includes the fraction of municipalities that had traditional bank branches by presence of Azteca. There are three patterns in the location of Azteca to highlight from the table. First, population size, per capita income and presence of traditional bank branches are strongly correlated with the presence of Azteca’s branches. Municipalities with an Azteca branch are more populated, have higher percapita income and are also more likely to have traditional bank branches than the average municipality in Mexico. On the contrary, municipalities where Azteca never entered have smaller 5 The Human Development Index imputes the percapita income for each municipality in Mexico using information from the National Income and Expenditure Survey (ENIGH) and the Census (see López Calva and de la Torre García (2003) and López Calva and de la Torre García (2005)). The index can be downloaded from: http://www.undp.org.mx/desarrollohumano/disco/index.html. 7 populations, lower per capita income and less penetration of traditional banks than the average mu- nicipality. Second, the entry of Azteca is very persistent. Most municipalities from this sample either had an Azteca branch from 2002 to 2005 or never had one. Only in six municipalities Azteca decided to exit before 2005 or enter after 2002. Third, although highly correlated, these variables do not ex- plain completely the location decisions of the bank. There are some municipalities with sufficiently high income percapita and population size where Azteca did not enter and some smaller munici- palities where the bank did locate. A plausible explanation could be that the degree of competition that Azteca faces varies across municipalities. For instance, in a municipality where Azteca might be indifferent about entering, if the fraction of households having access to transfers from friends or rel- atives suddenly increases, Azteca may find it unprofitable to open a branch since fewer households may want to commit to an Azteca loan. This however, is information that remains unobserved in the data, but that is considered in the model. 3.1 Empirical Strategy I now examine whether households’ saving and consumption patterns changed over time in munic- ipalities that received an Azteca branch. To do so, I exploit variation over time and across munici- palities of Azteca’s branches to compute difference-in-difference (DID) estimates. The DID estimates compare the difference in households’ mean outcomes before and after Azteca’s opening between Azteca and non-Azteca municipalities. Azteca municipalities are classified as those that had an Azteca branch by 2005 while non-Azteca are all other municipalities6 . The econometric specifica- tion to compute the DID estimates consists of: yh;m;t = 0 + 1 Aztecam + 2 yeart + 3 Aztyearm;t + 4 Xh;m;t + 5 Zh + "h;m;t (1) where h; m; t denote households, municipalities and whether the year is 2002 or 2005. Xh;m;t is a vector of time-varying controls for household demographics and municipality characteristics that are: if household is in a rural village, size of the municipality and presence of traditional bank branches or other government credit institutions. Zh is a set of household fixed effects that control for all the unobserved variation of households that is fixed over time, such as household preferences 6 Different treatment groups were tested, such as municipalities where Azteca first appeared in 2002, or municipalities with an Azteca branch from 2002 to 2005. While the magnitudes vary, the empirical results hold for the different classifica- tions. 8 towards risk aversion. The DID estimator is captured by 3, the coefficient of the indicator variable Aztyearm;t , which corresponds to the interaction of Aztecam and yeart . The former is a dummy that equals one if the municipality had an Azteca by the time the second wave of MxFLS was collected. The latter equals 0 if the year is 2002 and 1 otherwise. Table 2 presents descriptive statistics on the households data classified according to the presence of Azteca across municipalities. The first column reports information of households from municipal- ities that had an Azteca branch in 2005. Information of all other households is shown in column 2. The baseline year consists of 3,483 households in municipalities with presence of this bank and 2,156 households in municipalities where Azteca did not open. Since Banco Azteca targets clients em- ployed in the informal sector, classifying households into formal and informal is needed. Following Levy (2008), I consider a household as formal if any of its nuclear members- household head, spouse and sons or daughters- has a job that provides Social Security benefits. Otherwise, the household is considered informal. The share of households that belong to the formal sector was higher in Azteca municipalities than in non-Azteca ones in the baseline year, 0:238 and 0:131, and stayed higher in 2005, 0:229 and 0:125 respectively. 3.2 Saving patterns Table 3 presents the saving patterns of households over time. The first two columns report the 2002 and 2005 means for households from Azteca and non-Azteca municipalities. The last column presents the DID coefficient ( 3) of equation (1) on the different outcomes. The first outcome mea- sures households’ awareness about bank loans. This outcome is important since it proxies for house- holds’ potential ability to borrow in the future from banks, which could be altering their consumption and saving decisions in the present. This indicator variable equals one if a household knows it can obtain loans from banks and zero otherwise. As seen from the table, the fraction of households that knew they can obtain bank loans increased from 2002 to 2005 in both Azteca and non-Azteca munic- ipalities. Nevertheless, the DID estimator indicates that in municipalities with Azteca branches, the probability that households knew they can obtain bank loans increased substantially more than in municipalities with no Azteca. The DID estimator (0:0887) implies that after Banco Azteca opened, relative to municipalities with no branch of this bank, the probability that households in Azteca mu- nicipalities knew they can borrow from banks was 60% higher once this bank opened. The next variable examined is the probability that households obtain bank loans. While take up of bank loans is very low, the fraction of households with bank credit substantially increased in 9 municipalities where Azteca entered7 . According to the DID estimator, households from municipal- ities where Azteca opened were twice more likely to obtain bank loans than in other municipalities8 ( 3 = 0:0108). These results suggest that once Banco Azteca opened its branches in a municipality, households’ borrowing from banks increased. To explore if households from municipalities where Banco Azteca opened reduced their usage of more expensive credit suppliers, the next rows present the means over time and the DID estimates of the probability that households borrowed from pawn shops. The results suggest that the likelihood that a household borrowed from pawn shops dropped significantly by 0:0087 points in municipalities in which Azteca located its branches relative to mu- nicipalities where Azteca did not enter, which is a sizeable decline of 39%. This change in borrowing from pawn shops to banks is consistent with Karlan and Zinman (2009)’s findings for consumption loans in South Africa, where they find suggestive evidence that expanded access to formal credit changed households’ borrowing sources from informal to formal ones. If the opening of Azteca branches increased households’ ability to borrow from formal providers, households’ decisions to save may also be changing. According to the precautionary savings model (Deaton (1991), Carroll (1997)), in the presence of uncertainty about future income, households will use savings as buffer stocks to counter the effects of future income shocks. Maintaining savings is even more important for credit constrained households, who are unable to borrow when times are bad. An implication of this model is that when households’ ability to borrow in the future increases, their need to hold liquid assets declines. To examine whether there is evidence of this behavior in the data, I estimate DID regressions on the proportion of households holding savings. Similar to the find- ings in Kaboski and Townsend (2009) for the Thai Baht microcredit program evaluation in Thailand and Lee and Sawada (2010) for credit constrained rural households in Pakistan, my results suggest that in Azteca municipalities, the fraction of households holding savings significantly decreased in 0:0403 by 2005. This decline represents a drop of 11% from the 2002 mean. While the average house- hold did not decrease its saving levels, households below the 40th income percentiles were holding significantly less savings once Azteca entered in their municipalities. The results outlined in table 3 suggest that the saving patterns of households from Azteca mu- nicipalities significantly changed after Banco Azteca opened, relative to non-Azteca municipalities. However, given that Banco Azteca targets clients employed in the informal sector, these changes 7 In different settings, take up of credit products has also been found to be very low. In rural Kenya for instance, after randomly providing information on credit options and lowering the eligibility criteria for loan applicants, only 3% of people initiated the loan process (Dupas et al. (2012)). 8 Moreover, the median size of a bank loan in municipalities where Azteca entered decreased from 11,000 Mexican pesos in 2002 to 10,000 Mexican pesos in 2005, suggesting that the composition of clients borrowing from banks was different. 10 should be driven by informal households. Tables 4 and 5 explore whether this is the case by sepa- rately focusing on the sample of households that are informal or formal in both periods. According to table 4, compared to informal households from non-Azteca municipalities, informal households in Azteca municipalities were 76% more likely to know they can borrow from banks ( 3 = 0:1009) after the opening of Azteca. Their likelihood of obtaining loans from banks more than doubled ( 3 = 0:0139)9 . Moreover, after the entrance of Azteca in their municipalities, the probability that informal households obtained pawnshop loans decreased in 45% ( 3 = 0:01). In addition, the frac- tion of informal households saving substantially decreased in 13% ( 3 = 0:043). While the average savings of households below the 40th income percentile declined by 46% (Mx $2,264), savings held by households below the 60th income percentile did not change substantially, suggesting that it was poorer households those reducing their savings. Consistent with the clients that Banco Azteca targets, table 5 suggests that formal households from municipalities with Azteca branches did not experience significant changes on their saving patterns after the opening of this bank. Altogether, this exercise suggests that informal households from municipalities where Azteca opened concentrate the changes in saving patterns observed from the data. 3.3 Consumption patterns In the MxFLS questionnaire, each household that obtained a loan is asked about its stated reasons for borrowing. The two most reported reasons were: i) to finance consumption expenditures during a bad economic situation such as unemployment, sickness of family member, etc. (50% of responses); and ii) to purchase/repair durable goods (35% of responses). Only 8.5% of the borrowers stated that they obtained their loan to start a new business or to invest in one. Table 6 examines whether consumption patterns of households in municipalities where Azteca opened a branch experienced any change. In particular, since borrowers report using loans towards durable goods purchases/repairs, this exercise first explores if households from Azteca municipal- ities own more durable goods and of higher value. Panel A of the table presents the means over time and DID estimates on: the proportion of households owning electronic appliances (radio, TV set, VCR, computer, etc.), furniture and large appliances (washing and dryer machine, stove, refrig- erator), and other appliances (blender, iron, microwave, etc.), and the value of these goods. The 9 Conditional on borrowing from banks, the average loan size also declined: while in 2002 the median bank loan size obtained by informal households in Azteca municipalities was 23,141 Mexican pesos, by 2005 it was 19,494. 11 results suggest that in Azteca municipalities, households were more likely to own electronic appli- ances ( 3 =0.0223) and large appliances and furniture ( 3 =0.0228). Moreover the value of these goods significantly increased in $2,305 Mexican pesos for electronic appliances and $2,342 Mexican pesos for furniture and large appliances. These patterns are consistent with Banerjee et al. (2009) who find that on average, Indian households were more likely to purchase durable goods such as televisions and refrigerators if a microfinance institution opened an office in their slums. As with changes in the saving patterns, table 7 (panel A) suggests that these changes are coming from the sample of households whose members are employed in the informal sector. In Azteca- municipalities after the opening of Banco Azteca, informal households were more likely to own elec- tronic appliances (0.0244) . The value of their electronic appliances increased by 22% (Mx $1,456), and of their furniture and large appliances by 30% ($2,756 Mexican pesos). On the other hand, households with members belonging to the formal sector do not report significant differences in their likelihood of owning durable goods or their average value (See table 8, panel A). From the responses of borrowers, there is reason to believe that credit is also being used by house- holds as an instrument to protect against bad economic shocks or to smooth consumption. Testing for consumption smoothing would require following households over several points in time and comparing their consumption to their income patterns. Since the first wave of MxFLS was collected in 2002, households are observed only one period before Azteca’s entry and one period afterwards, making this test infeasible. However, MxFLS collects information about certain events that caused economic losses to households and the date when these events occurred. Using this information, I explore whether households are better able to deal with economic shocks in the presence of formal credit. To do so, I examine the percapita expenditure of households that experienced a bad economic shock in the last 12 months they were surveyed, defined as: i) death of a family member; ii) serious illness of a family member; and iii) unemployment or failure of business of a family member10 . Panels B of tables 6, 7 and 8 present the DID results for the sample of all households, informal households and formal households experiencing bad economic shocks. The results however, must be taken as weak suggestive evidence of consumption smoothing for two reasons. First, the DID 10 1% and 1:4% of households in Azteca and non-Azteca municipalities experienced the death of a family member in 2002. This share increased to 1:3% and 2:4% in 2005. Serious illness of a family member was reported by 2:4% and 2:1% of households in Azteca and non-Azteca municipalities in 2002; by 2005 this share was 2:8% and 2:2%. Unemployment or failure of family business was reported by 5% and 2:8% of households in Azteca and non-Azteca municipalities in 2002; this share was 4:4% and 3:6% in 2005. 12 regressions for this exercise are estimated without household fixed effects11 . Second, since the num- ber of households in the sample experiencing shocks is low, the 2002 and 2005 means of per capita expenditures in Azteca and non-Azteca municipalities are noisy, especially for the control group. With these drawbacks noted, there is some evidence suggesting that households from Azteca mu- nicipalities were better able to increase their consumption during bad economic times after Azteca opened branches. The OLS DID estimates suggest that informal households in Azteca municipalities increased their percapita expenditure by Mx $9,339 and Mx $8,463 if experiencing illness of a house- hold member or failure of the family business (panel B of table 7). As seen from panel B of table 8, formal households from Azteca municipalities experienced no significant change relative to formal households from other municipalities. 3.4 Robustness Check According to the data, informal households from municipalities with an Azteca branch experienced various changes on their saving and consumption patterns once the branch opened. However, a con- cern with these findings is that informal households from municipalities where Azteca entered were in a different trend from households in other municipalities, and would have actually experienced these results even in the absence of this bank. One possible way to rule out this possibility would be to examine whether the trends in saving and consumption patterns between treated and control households were similar in the periods before Azteca opened. As the first wave of MxFLS was in 2002, this test is not feasible. However, an alternative robustness check is to analyze the change in saving and consumption patterns from 2002 to 2005 of households from municipalities that would get an Azteca until 2006 or later. If informal households from municipalities selected by Azteca have different trends to other households, and this is driving the results, then we should expect the sav- ing and consumption patterns of these households to be changing even in the absence of Azteca. If the entrance of Azteca is altering households’ behavior, then households from municipalities where Azteca entered in later periods must not have experienced substantial changes. Tables 9 and 10 report the DID coefficients of equation (1) on the saving and consumption out- comes of informal households using two different treatments. The first treatment corresponds to the standard group of informal households from municipalities with an Azteca by 2005 (Column 1). The alternative treatment is composed of informal households from municipalities where Azteca 11 Since less than 5% of all households experienced shocks both before and after Azteca opened its branches, the house- hold fixed effects DID estimates are positive but not significant, due to the lack of within-household variation. 13 only entered by 2006 or later (Column 2). If Azteca changed the behavior of informal households in municipalities where it opened, one might then observe that the DID coefficients from column 2 (municipalities where Azteca have not entered yet) are not substantially different from zero. Overall, the results confirm that in municipalities that have not yet received an Azteca branch, the saving and consumption patterns of informal households are not different from other households in municipal- ities without a branch.12 4 Model I now present the dynamic model in which households interact, among other credit suppliers, with Banco Azteca. The economy in this model consists of M municipalities populated by households. Consistent with the data, municipalities differ from each other in their composition of households, m ). their population size (P m ), percapita income (Y m ) and presence of credit suppliers (crt As in related literature modeling the coexistence of several lenders (Giné (2011), Jain (1999)), this model allows for four credit suppliers that are traditional banks, Banco Azteca, pawn shops and friends or relatives, cr = fB; A; P S; F Rg. The model can be divided in two parts: the problem of the households and the problem that Banco Azteca solves. 4.1 Households’ problem Households in the model have preferences over consumption goods (ch t ) and the service flow of e h ). While consumption goods only last one period, durable goods yield utility over durable goods (D t time, but depreciate periodically at a rate of . Households’ preferences are summarized by the utility function u[ch eh t ; Dt ]: The model of households is a standard precautionary savings model in which future income is uncertain (as in Deaton (1991), Aiyagari (1994) and Kaboski and Townsend (2011)), that includes non-durable as well as durable goods, in which borrowing constraints are mainly determined by the availability of credit suppliers in the municipality, the stock of durable goods held by households (i.e., Bertola et al. (2006), Alessie et al. (1997)), and the employment sector of households. Every period, households make three decisions in order to maximize their expected lifetime util- 12 The fact that more households are aware that they can borrow from banks even in municipalities where Azteca did not open is not surprising given the intense TV advertisements that Banco Azteca products have throughout the country. 14 ity. Households decide whether to belong to the formal sector or not (Fth = 0; 1), how many durable goods to purchase (ih h t ), and how much to save or borrow of a liquid asset st . h consist of two components: their labor income At every period t, households’ liquid resources Rt h h and their savings from last period sh , which include both the principal and the interest, r crt 1 . yt t 1 The model assumes that all households have access to the same saving technology that pays a pos- itive interest rate, while for loans, the interest rate rcr varies across credit suppliers13 . If households borrowed last period, sh t 1 < 0, default can occur. Let deftcr;h be an indicator function that equals one if h defaults on its t 1 loan to credit supplier cr. Liquid resources are then given by: h Rt h = yt + (1 deftcr;h ) (1 + rcr )sh t 1 (2) h from one of the two sectors of the economy, a formal sector and Households obtain their income yt F;h 1;h 0;h an informal one. Let yt be the income offered from the formal (yt ) and the informal sector (yt ) to household h at period t. For both sectors, the realized income depends on the previous income of the h household yt and the education and age of the household head eh ; ah 1 t . These variables proved to explain accurately the two income processes in the data. In the model, the magnitude in which these variables determine income is allowed to differ between sectors. In addition, every period each F;h F ): income offer is subject to an idiosyncratic shock, t N (0; F;h h h h F;h yt = fF (yt 1 ; e ; at ; t ) (3) At every period, all households receive an income offer from the informal sector, but they only re- ceive a formal-sector income offer with probability fth .14 Therefore, the income offers that household h observes each period are given by: 8 < y 1;h ; y 0;h with prob fth h t t yt = (4) : y 0;h with prob 1 fth t 13 As in the Mexican credit market, the model assumes that rF r < rB < rA < rP S 14 According to the data, more educated households are more likely to be in a formal occupation, and once a household h belongs to the formal sector, the probability of staying in this sector is high. Hence, the model allows ft to depend on whether the household belonged to the formal sector in the last period and on the education of the household head, eh . 15 Households that have job offers from both sectors, decide in which sector to be employed by comparing their value functions of belonging to each sector. Households who only observe an offer from the informal sector, stay in the informal sector. Households in this model own durable goods that are used as collateral if borrowing from Azteca or pawn shops. Let q be the market price of one unit of the durable good relative to one of consump- h ; must equal the value of the depreciated durable tion, then the value of durable goods at each t; qDt goods from period t 1 plus any purchase made at t. In case of default to a loan from Azteca or pawn shops- i.e. when deftA;h = 1 or deftP S;h = 1, h loses the durable goods that were pledged as collateral of the loan (qxh 15 t 1 ): h qDt = (1 h )qDt 1 + ih t deftcr;h qxh t 1 (5) Note from equations (2) and (5) that when the default indicator function equals 1, h’s debt disap- pears but as a penalty, h loses the durable goods xh t 1 that were pledged as collateral. Households face four constraints in the credit market that are summarized next. 1) Geographic constraints. The first constraint refers to the inability of households to borrow from suppliers that are not located in their municipalities. The set of lenders in m at period t is given by fB m ; Am m m m m t ; P S ; F Rt g, where B and At are indicator variables that equal 0 or 1, depending on whether there are branches of these suppliers in m or not. Since all municipalities have pawn shops, P S m equals 1 for any m. In the model, credit from friends and relatives exists everywhere, but the fraction of households with access to these loans varies over time and across municipalities. This m = F R; F R . Therefore, municipalities can have a low fraction is assumed to take two values: F Rt m with or a high fraction of households with access to these lenders. Every period, each m draws F Rt a probability that depends on the percapita income of the municipality and the fraction of households m ). with access to loans from friends and relatives in the last period, g = g (Y m ; F Rt 1 2) Negative credit history. This constraint only applies to clients of Azteca and traditional banks. In Mexico, these suppliers rely on households’ payment histories to approve loans, and do not lend to households with previous history of default. Interestingly, banks and Azteca do not share informa- tion with each other, therefore if a household defaulted in the past to an Azteca loan, traditional banks 15 Different from Azteca, pawn shops retain the collateral while the loan is paid. To capture this in the model, during periods when households borrow from pawn shops, households do not derive utility from the durable goods pledged as collateral. 16 are not aware of it. Thus in the model, if borrowing from Azteca or traditional banks, households need to have no default incidents in the past with them. 3) Official proof of income. This restriction only applies to clients requesting loans to traditional banks, who are required to be employed in the formal sector at the time of the loan request. 4) Collateral. Azteca and pawnshops require households to pledge durable goods as collateral to secure the repayment of their loans. The amount that a household can borrow is bounded by a limit that depends on the value of durable goods that households own. In sum, the set of lenders that households have access to is determined by the following munici- pality characteristics and decisions that households make: 8 > > Bth;m if B m = 1; deftB;h h > > 1 = 0; Ft = 1 > > < Ah;m if Am A;h h h t t = 1; deft 1 = 0; Dt > 0 crt = (6) > P S h;m > h >0 if Dt > > t > > : fr h;m m =f with prob F Rt m m t F R (F Rt 1 ; Y ) Besides these four constraints that limit households’ access to suppliers, loan size is also bounded differently by lenders. The bounds on loans from Azteca and pawn shops are determined by the value of durable goods pledged as collateral.16 The bounds on loans from traditional banks and friends and relatives are determined outside the model: 8 > > A ! sh h qDt > > t 1 > > < PS ! sh h qDt h t 2 crt = (7) > > B ! sh > > t 3 > > : FR ! sh t 4 Under this setting, households’ problem consists of selecting the optimal set of decisions (Fth ; ih h t ; st ) that maximizes their expected lifetime utility, defined as: P T t eh E u[ch t ; Dt ] ; t=1 16 The collateral requirements are considered exogenous in the model. Their values were obtained from CONDUSEF, Azteca’s Financial Reports and PROFECO. According to these sources, the average collateral required by a pawn shop in Mexico is three times the value of the loan . Thus 2 is set to 1=3. According to CONDUSEF and Banco Azteca’s reports, Banco Azteca requires a collateral equivalent to the value of the loan, hence 1 is set to 1. 17 subject to (2) (7), and the budget constraint equation: ch h h h t + it + st = Rt (8) h fall short of a subsistence consumption level (c), it is assumed If households’ liquid resources Rt that households have a form of insurance or consumption floor that allows them to achieve the sub- sistence level. This consumption floor is a parameter estimated in the model. In the model, default occurs when a household with sh t 1 < 0 receives an income shock such that the maximum amount of debt it can borrow is not enough to pay off sh t 1 and cover its minimum consumption (c). In this situation, deftcr;h = 1 and if the supplier was Azteca or a pawn shop, h loses the durable goods pledged as collateral. The optimal policy of the defaulting household is then given by: ch t =c sh t =0 ih t =0 In this model, the state space at period t consists of the following variables at the household and municipality level: cr m m m m wt = fst 1 ; yt 1 ; Ft 1 ; crt 1 ; Dt 1 ; deft 1 ; e; at 1 ; B ; P ; Y ; F Rt 1 g (9) 4.2 Problem of Banco Azteca Banco Azteca’s total expected profits are given by the sum of expected profits from each of the M municipalities: M X m E[ t] = E[ t ] (10) m=1 The problem of this bank is to maximize its expected profits by deciding in which municipalities to open its branches. Therefore, at every period t and for each municipality m, Azteca selects Am t = f0; 1g such that: m m m E[ t ] = max fE [ t jAt = 1] ; 0g (11) 18 Two components determine the expected profits of opening an Azteca branch in municipality m. The first one corresponds to the gains that Azteca expects to receive at t + 1 from lending at t. The second component refers to the cost associated to the operation of the branch, m, which is a cost at the municipality level that must be paid every period that Azteca operates in a municipality. To compute the expected gains from lending at t, it is assumed that Azteca knows and solves the problem of the households, and observes the state space of each household (wt ), as well as the processes and parameters determining: i) the income draws; ii) the job offers of households; iii) m ). Like households themselves, Azteca is unable to and the friends and relatives credit draws (F Rt observe the realization of shocks at t + 1, mainly: i) whether households will have a formal job offer 1;h 0;h at t + 1, fth +1 ; ii) the income shocks at t + 1 of the formal and informal sectors, t+1 ; t+1 ; and iii) h . whether households will have access to loans from friends and relatives at t + 1, f rt +1 With the information that Azteca has, at the beginning of each period the bank is able to determine h . Azteca the pool of loan applicants that in each municipality will visit its branch at t to obtain a loan lt computes each applicant’s expected profits by estimating the probability that the loan is paid back at t + 1. Let this probability be ph 0 t , which is simply the likelihood that h s liquid resources at t + 1 are h . Then, with probability ph , h will pay off its loan and Azteca high enough to cover c and pay off lt t h . With probability 1 will receive a net gain of rA lt ph t , h will default on its debt and Azteca will lose h but will recover the collateral, qxh . Therefore, the expected profits of lending to household h at lt t period t are given by: h i h;m h E t+1 j lt > 0; ! h h A h t = pt r (lt ) + (1 ph h t ) (qxt h lt ) (12) As equation (13) indicates, Azteca’s expected profits from households who do not borrow at all, or borrow from a different lender, are zero. h i h;m h E t+1 j lt = 0; ! h t =0 (13) Let H be the number of households in m. From these H households, every period some will borrow from Azteca while others will not borrow at all or will borrow from different lenders. Let J be the number of households borrowing from Azteca. Azteca’s expected gains from lending at t in municipality m are then given by summing up the expected gains of its J clients: 19 H X h i XJ h i h;m h;m E t+1 j !h t = E t+1 h j lt > 0; ! h t (14) h=1 j =1 The second component of Azteca’s expected profits at the municipality level is the cost of oper- ating a branch, m. As equation (15) shows, there are two variables that determine m, which are population size and whether the municipality has an Elektra store. Let IE be an indicator variable that equals 1 if m has an Elektra store and 0 otherwise. To account that having an Elektra store sub- stantially facilitates Azteca’s operations (Conger (2003)), m consists of a fixed cost that differs for municipalities with an Elektra store. Population size enters Azteca’s cost function through a linear spline approximation to reflect potential economies of scale of having a branch in large enough mu- nicipalities, where IP is an indicator variable that equals 1 if m’s population size exceeds a threshold 4 estimated in the model. m = 0 + 1 IE + 2 Pm + 3 (P m 4) IP (15) Therefore, the expected profits of opening an Azteca branch in municipality m are given by: H X h i m m h;m E[ t jAt = 1] = E t+1 j !h t m (16) h=1 As this model focuses on the short-term effects of the entry of Banco Azteca, Azteca is the only credit supplier in the model that decides where to locate its branches. The location of the other credit suppliers is considered exogenous. A potential concern is the response of pawn shops to the entrance of Banco Azteca, not only in location of their branches but also in pricing of their loans. Since there is no reliable information at the national level on pawn shops’ pricing or location decisions before 2005, the reaction of pawn shops to the entrance of Azteca is unknown. The model assumes that in the short-run, pawn shops did not change their behavior once Azteca entered the market. Additionally, the model abstracts from other decisions that credit suppliers make such as the interest rate to charge or the collateral to require. Data on these decisions began to be collected by the Mexican authorities after 2005. Hence, in the model these decisions are fixed to the choices observed from the 2005 data, which are assumed to be the optimal decisions that credit suppliers made in the past under the assumption that these decisions did not change by the entrance of Azteca. 20 5 Model Solution The interaction between households and credit suppliers is as follows. At the beginning of each t, the idiosyncratic shocks are drawn and observed by all households and credit suppliers. Concretely: m = F R; F R ; based on F Rm , each household draws its – Each municipality draws F Rt t h = f0; 1g; realization of credit from friends and relatives f rt – Households find out whether they receive an offer from the formal sector or not, and 1;h 0;h observe their formal and informal labor incomes, yt ; yt : Once the shocks are realized and observed, households who brought a debt to the period and have not enough resources to pay it back, default. Collateral pledged to pawn shops and Azteca is collected by the lenders. Azteca observes all information from households and municipalities, and decides in which municipalities to open its branches. To decide the location of its branches, Azteca computes the profits it expects to receive from every h i household, E h;mt+1 j ! h ; and sums up the expected profits by municipality. Based on them, the t bank opens its branches in municipalities where its expected gains cover the operation cost, 8 h i > PH h;m > < 0 if E t+1 j !h t Am t = h=1 h i f or m = 1; :::; M > > PH h;m : 1 if E t+1 j !h t > h=1 Once Azteca opened its branches across municipalities, households make their decisions. The recursive problem of each household at every period t can be written in the following form: V h (! h uh [ch eh h !h ; t + 1 t ; t) = max t ; Dt ] + E V t+1 sh t ;ih t h ;Ft s:t:(2) (7) and ch h h h t + it + st = Rt ; I solve the model by backwards recursion, starting from the assumed last period of life of the household T = 75, to the assumed initial period of its formation t0 = 18. As it is a finite horizon 21 problem, it is assumed that the terminal value is equal to zero- i.e. in their terminal period of life, the value functions of the households equal the utility at T . At periods t < T , the value functions of the households equal the utility at t plus the expected value function of t + 1. Keane and Wolpin (1994) show how to recover these expected value functions, which they call the Emax function. This function is calculated for every point of the state space, any period t and every possible choice set. In this model, the size of the state space was discretized and, following Keane and Wolpin (1994), the Emax functions were approximated by a parametric function of the current state variables. 5.1 Empirical specification Functional form assumptions were made for the following processes of the model: household utility function; labor income process from the formal and informal sectors; and transition of credit from friends and relatives across municipalities and over time. Utility: Households’ preferences u[ch eh t ; Dt ] are assumed to have the following functional form with respect to the nondurable goods and the service flow of durable goods: 1 ch t eh Uth = + 0 Dt (17) 1 e h ) is produced by a linear household production where the flow of services from durable goods (D t h is transformed by the productivity parameter function, in which the stock of durable goods Dt >0 1 into the flow of services enjoyed by the household at each period t: et D h = h 1 Dt (18) The parameter captures the intertemporal substitution of the nondurable good ch t . This para- metrization implies that household h0s intertemporal elasticity of substitution is 1= . Since the flow of services of durable goods is produced by a linear household production function, it is not possible to separately identify the preference for the service flow of durable goods ( 0 ) from the productivity parameter 1. Only their product 0 1 is identified. 22 Labor income from the informal and formal sectors: Labor income offers from the formal and informal sectors are drawn from the following processes: 2 1 h yt;h = 1;1 yt 1 + 2;1 eh + 3;1 ah t + 4;1 ah t + h 1;t (19) 2 0 h yt;h = 1;0 yt 1 + 2;0 eh + 3;0 ah t + 4;0 ah t + h 0;t (20) Equations (19) and (20) relate the labor income offer of household h at t to h’s labor income at t 1 plus a linear return to education and a quadratic return to age of the household head. The labor income parameters are allowed to differ between the two sectors. Municipalities’ access to credit from friends and relatives: It is assumed that each period mu- nicipalities can draw either a high (F R) or low (F R) access to credit from friends with probability g , which depends on: i) whether municipalities’ percapita income is above or below the national percapita income; ii) and whether access to credit from friends and relatives at t 1 was low or high. These conditions give more flexibility to the modelling of credit from friends and relatives by: allowing wealthier municipalities to have different informal credit arrangements than poorer mu- nicipalities such as family networks, "tandas" or cooperatives; and also by allowing for persistence within a municipality over time on the fraction of households with access to loans from friends and relatives. Therefore, the probability that a municipality draws F R at t is given by: 8 > > g1 if m = F R; Y m > Y ) (F Rt > > 1 > > < g2 if m = F R; Y m (F Rt Y) 1 g= > > g3 if m = F R; Y m > Y ) (F Rt > > 1 > > : g if m = F R; Y m (F Rt Y) 4 1 where the probabilities g1 g4 are parameters estimated in the model. 6 Model Estimation There are 28 structural parameters of the model which are estimated by the method of simulated method of moments. The goal of this method is to estimate a vector of structural parameters , by matching a set of simulated statistics, denoted as , with the corresponding set of actual data 23 statistics, denoted as m. The estimated structural parameters are those that minimize the weighted average distance between the set of simulated statistics and the set of data statistics. Because the simulated statistics depend on the underlying structural parameters, minimizing this distance will provide consistent estimates of the structural parameters under certain conditions. The estimator of is defined as the solution to the minimization of b = arg min mn 1 s 0c 1 s s n( ) W N mn s n( ) where the subscript n refers to the number of households in the sample and s denotes the num- cN is a positive definite matrix that converges in probability to a deterministic ber of simulations. W positive definite matrix W . I use the inverse of the covariance matrix of the data moments as the cN . The covariance matrix is computed using a standard bootstrap method with weighting matrix W 1000 bootstraps. The statistics to be matched are listed in table 11. The first four moments correspond to the proportion of households that belonged to the formal sector in 2005, conditional on the education of their head and on their employment sector in 2004. Statistics from 5 to 14 are related to the labor income process from the formal and informal sectors17 . Statistic 5 captures the persistence of labor income in the formal sector, by regressing the 2005 labor income of formal households on their 2004 labor income. The next moment captures the returns to education on the formal labor income by computing the mean difference between labor income of household heads with less and more than nine years of schooling that belonged to the formal sector in 2005. Moments 7 and 8 describe the returns to age in the formal sector, by comparing the 2005 labor income of household heads older and younger than 35 years and household heads older and younger than 50, conditional on being employed in the formal sector. Moment 9 captures the variance of the income shocks received by formal households in 2005. These shocks are the residuals from the OLS regression of formal labor income in 2005 on 2004 labor income, education, age and age squared for households observed in the formal sector at 2005. The same statistics are used to capture the income process of the informal sector (moments 10 to 14). The next four moments relate to households’ consumption behavior. These moments correspond to the 5th percentile of the distribution of households’ percapita expenditure in 2005; the proportion of households that owned radio, TV sets, VCRs or computers in 2005; the 2005 ratio of percapita 17 The households’ data (MxFLS) includes retrospective information for the years 2004 and 2001 regarding labor deci- sions, which is used to compute these moments. 24 expenditure to total income across households; and the log of the ratio of percapita expenditure in 2005 to the 2002 percapita expenditure across households. Moments 19 to 24 describe the patterns of access to friends and relatives credit over time and across municipalities. Moments 19 and 20 capture the 2005 fraction of households with loans from friends and relatives in municipalities below and above the mean access to friends and relatives credit. To compute them, I first obtained the 2005 average fraction of households with loans from friends and relatives across municipalities. Moment 19 uses the sample of municipalities below this average, while moment 20 corresponds to the sample of municipalities above the average fraction of households with credit from friends or relatives in 2005. The next four moments capture the persistence of credit from friends and relatives over time. I first computed the average fraction of households with credit from friends and relatives in 2002 and 2005, respectively. I then classified municipalities into two groups: municipalities below the average fraction in 2002 and municipalities above. Moments 21 to 24 report the proportion of households that in 2005 obtained a loan from friends or relatives conditional on whether municipalities are low/high income and below/above the 2002 mean of credit from friends and relatives. The last five moments describe the entry patterns of Azteca branches into MxFLS municipalities. Moment 25 corresponds to the fraction of municipalities with an Azteca branch in 2005. Moments 26 to 28 report the fraction of municipalities with an Azteca branch in 2005 conditional on their population size being: between the 10th and 30th percentiles; 30th and 50th percentiles; and 50th and 70th percentiles. The last moment refers to the fraction of municipalities that had an Azteca branch from 2002 to 2005 conditional on having an Elektra store in 2002. 7 Estimation results and model fit The estimation of the model requires some choices regarding the size of the state space. I discretize households’ savings and labor income, the value of the durable goods, the household head’s years of schooling, the fraction of households covered by friends and relatives, and the income of the munic- ipalities, which are the continuous state variables of the model. The grids were selected so that they reflect the distributions of these variables in the data. The grid of savings consists of 10 points, the first one equals the 5th percentile of the empirical savings distribution and the consecutive points refer to the 15th, 25th,..., 85th and 95th percentiles. I tested the robustness of the simulations using fewer grid points and I found that it is important to include at least 10 points. The value of the durable goods 25 was discretized to 4 point grids. The first point corresponds to a value of $0 and the last 3 points reflect the empirical distribution of durable goods value: these points correspond to the 30th, 60th and 90th percentiles of the data distribution. Household heads’ years of schooling were discretized into two grid points: the first point corresponds to all household heads with less than 9 years of schooling and the second point includes all households in which the head had 9 or more years of education. At the aggregate level, municipalities were classified by their fraction of households with access to credit from friends and relatives into two categories: below and above the mean fraction of households who borrowed from friends and relatives. Municipalities were also classified in two groups according to their percapita income, using the median income as the cut-off. Households’ labor income was also discretized using a 3-point grid, whose points correspond to the mean values between the 1st and the 33th percentiles, the 33th and 66th percentiles and above the 66th percentile of the distribution of total labor income. Finally, I approximate the discrete distributions of formal and informal labor income shocks following Kennan (2006). I specify a continuous distribution for each sector shock, and given the parameters of this distribution, I specify a discrete approximation to them. I allow for 3 support points for these discrete approximations. 7.1 Estimation results Table 12 presents the estimated parameters and their asymptotic standard errors. I compute the asymptotic standard errors following Berndt, Hall, and Hall (1974) and Nash (1990). Probability parameters of receiving a formal offer: fth j eh ; Fth 1 In the model, the probability that a household receives a formal-sector job offer depends on whether the household belonged to the formal sector in the last period, Fth 1 = f0; 1g, and on the education of the household head, eh = f0; 1g. According to the estimation results, the probability of receiving an offer from the formal sector increases substantially if households were formal in the previous period. For low and high educated households, the estimated probabilities are 0.55 and 0.72, respectively. Households employed in the informal sector have a lower probability of receiving a formal job offer in the next period. These probabilities are 0.07 for low educated households and 0.132 for high educated ones. Parameters for formal labor income process: 1;1 , 2;1 , 3;1 , 4;1 , 1 The labor income process is allowed to depend on previous labor income, education and age. The estimated persistence for the formal income is 0.492, which means that each period, households income consists of 49% of their lagged labor income. The returns to education in the formal sec- 26 tor are estimated to 1.1. This coefficient implies that in the formal sector, income of high-educated households is 42% higher than low educated households. Regarding returns to age, the estimated coefficients are 0.04 and -0.0004, reflecting a concave pattern of income with respect to age. The estimated standard deviation of the income shocks from the formal sector is 2.03. Parameters for informal labor income process: 1;0 , 2;0 , 3;0 , 4;0 , 0. According to the estimated model, the persistence of lagged labor income on current income in the informal sector is 0.51, higher than in the formal sector. The returns to education (1.0) imply that relative to households with low education, the income of high-educated households in the informal sector is 60% higher. The estimated coefficients on age are 0.02 and -0.0002, respectively. Compared to the formal sector, the peak income age is reached earlier in the informal sector, when household heads are 41 years old. The standard deviation of the informal income shocks is 2.1, slightly higher than the variance from the formal sector. m Parameters for access to credit from friends and relatives: F Rt In municipalities with low access, only 5% of households can borrow from friends or relatives if they need to. In municipalities with high coverage (F R), each period 32.1% of households have a friend or relative from whom they can borrow. Parameters for the transition of fraction of households with access to credit from friends and relatives: g1 ; g2 ; g3 ; g4 Low-income municipalities who had low access to credit from these suppliers in the previous pe- riod, stay with low access to credit from friends and relatives with probability 0.89, and experience high access with probability 0.11. If low-income municipalities previously had high access to these lenders, they stay with high access with probability 0.87. For high-income municipalities, the transi- tion probabilities are the following. Municipalities who had low access to friends and relatives credit in the previous period, stay with low access with probability 0.81. Municipalities that experienced high access to these lenders in the past, remain with high access at t with probability 0.93. Households’ preferences parameters: , 0 1. The estimated intertemporal substitution of nondurable goods, , is 2.14. The estimated joint product of 0 1 is 2.1 but given the parametrization, it is not possible to identify separately the pref- erence for the service flow of durable goods ( 0 ) from the household productivity 1. Banco Azteca’s operating cost parameters: 0; 1; 2; 3; 4. According to the estimation, the fixed cost that Azteca faces in municipalities with no Elektra branches is 0 = 432, while for municipalities with an Elektra store 0 + 1 equals 221:7, suggesting 27 that the operating costs are substantially lower if Azteca serves a municipality previously covered by Elektra. Consumption floor: c This parameter determines the minimum consumption that a household receives. Its estimated value is 0.365, which corresponds to a daily percapita consumption of $0:83U SD. Exogenous parameters. The parameters that are not estimated inside the model and that are considered exogenous are the following: information about interest rates and collateral requirements was obtained from CONDUSEF, Azteca’s Financial Reports and the households’ data. According to this information, the average APRs of pawnshops, Banco Azteca, traditional banks and friends or relatives in 2005 were 220%, 130%, 40% and 0% respectively. Regarding collateral, 90% of the households in the dataset that obtained credit from friends and relatives reported they were not required to own any collateral, therefore in the model, the collateral that friends and relatives require is set to zero. According to reports from PROFECO, the average collateral required by a pawn shop in Mexico is three times the value of the loan. According to CONDUSEF and Banco Azteca’s reports, Banco Azteca requires that the value of collateral is equivalent to the value of the loan. The maximum loan that households can borrow from traditional banks and friends and relatives is also determined outside the model as a function of the head’s education. The maximum loan size from friends and relatives is $3; 000U SD, while bank loans can reach $5; 500U SD. These values were obtained from the distribution of loans from banks and friends and relatives from the households data. The depreciation of the durable goods, , was fixed to 0.10. The relative price of the durable goods with respect to the consumption goods, p, was obtained from the Consumer Price Index in 2002, and was set to $10. The discount factor, , was fixed to 0.99. 7.2 Model Fit As seen from table 13, the estimated model matches closely the simulated statistics to the real ones. The first panel presents the moments for the share of households employed in the formal sector in 2005, conditional on their education level and previous sector employment. The likelihood of being employed in the formal sector is strongly correlated with having participated in the formal sector in the previous period, for both low and high educated households. But also, regardless of whether households were formal or not in the previous period, education increases the probability of being formal. The simulated moments replicate these data patterns closely. Interestingly, while 7% and 13% of low educated households receive formal job offers each period, only 4% and 11% of these 28 households end up switching to the formal sector. A similar pattern occurs with high educated households, suggesting that for some households it is optimal to stay in the informal sector even when they have access to formal employment opportunities. The model fits well the labor income process of households employed in the formal and informal sectors, as the next panels show. The simulated statistics for the 5th percentile of the distribution of households’ percapita consumption, the fraction of households who own durable goods and the consumption moments are close to the data moments. The simulated moments governing the be- havior of credit from friends and relatives across municipalities accurately fit the data patterns, and the persistence of access to friends and relatives loans over time are also replicated by the simulated statistics. Regarding Azteca’s entry patterns, the model slightly overpredicts the fraction of munici- palities with an Azteca branch in 2005 (0.52 vs 0.51). Table 14 examines the performance of the model in predicting the type of municipalities where Azteca locates. The table compares the characteristics of municipalities with Azteca between the simulations and the real data. Information in the first column refers only to municipalities in which Azteca operated from 2002 to 2005. The second column contains statistics only for municipalities where Azteca located in 2002 but exited before 2005. The third column presents information of mu- nicipalities where Azteca entered after 2002 and stayed until at least 2005. The fourth column contains information from municipalities where Azteca opened after 2002 and exited before 2005. Finally, the fifth column presents statistics for municipalities that never had an Azteca branch. The table sum- marizes population size and percapita income distributions, as well as the fraction of municipalities with branches of traditional banks. The first panel of the table reports information from the real data and the second panel reports on the simulated data. At the estimated parameters, the model replicates closely Azteca’s location patterns. In the model, Azteca selects more populated municipalities with higher percapita income and higher presence of commercial banks. Also, as in the data, Azteca’s decision of location is very persistent. Out of 136 simulated municipalities, Azteca only relocates branches in 12, compared to 6 in the data. Finally, while population size and income are important drivers of the location decision, not all populated municipalities get an Azteca branch. Likewise, Azteca decides to enter in some smaller and poorer municipalities. In the model, besides population size and percapita income, the differential access of credit from friends and relatives across municipalities is a relevant factor explaining Azteca’s loca- tion. Figure 1 provides some insights of how credit from friends and relatives influences the decision 29 of location of Azteca. This figure holds all the parameters of the model constant and only varies the fraction of households with access to credit from friends and relatives. First, as the fraction of house- holds with access to credit from friends and relatives increases in a municipality, Azteca’s expected profits decrease. Second, in low populated and poorer municipalities (such as Jopala, Puebla), Azteca would not open a branch even if few households have access to credit from friends and relatives. In municipalities closer to the median income and population size (such as Amecameca or Linares), Azteca’s entry or exit decision is influenced by the access to credit from friends and relatives. In larger municipalities (such as Apodaca), Azteca would enter even if most households have access to loans from friends and relatives. 8 Robustness of the model Saving outcomes were not used as moments in the estimation of the structural parameters. A ro- bustness check to examine the performance of the model is to analyze if the model can reproduce the change in saving outcomes experienced by households in municipalities where Azteca located its branches. To do this, I estimate the difference-in-difference regressions using the simulated data and compare them with the real data results. As in equation (1), the regressions compare the saving patterns of simulated households in 2002 (before Azteca started) and 2005 (once Azteca had decided location for 3 periods). Table 15 presents the DID coefficients ( 3) for all households, and for households who were in- formal or formal in both periods, using real data (column 1) and simulated data (column 2). Overall, the model reproduces the change in saving patterns observed in the data. In municipalities where Azteca enters, simulated households are more likely to borrow from banks, and less likely to borrow from pawn shops and to hold liquid savings. As in the data, these results are concentrated among informal households, and formal households remain practically unaffected.18 8.1 Different counterfactuals in the model The reduced form results presented in tables 3 to 8 compare outcomes over time of households from municipalities where Azteca entered with municipalities where Azteca did not open branches. If households from control municipalities respond as households in treated municipalities in the pres- 18 While the magnitude of the coefficients in the model is larger than in the data, this occurs mainly because saving in the model is a continuous variable, so any loan or saving greater than zero is recorded as one in the dependent variable. In the data, households are not likely to report loans and savings lower than a certain threshold. 30 ence of Azteca, then the difference-in-difference specification would provide valid estimates of the impact of Azteca. However, if there are differences (observed or unobserved) between control and treated municipalities that affect the impact of Azteca, the effect of Azteca will then differ if the bank were to locate in control or treated municipalities. In this case, the difference-in-difference will no longer produce unbiased estimates. The estimated structural parameters governing municipalities’ access to credit from friends and relatives reflect substantial differences between low and high income municipalities. Municipalities that tend to be selected by Azteca (high income municipalities) also tend to have more access of credit from friends and relatives. In the model, as municipalities have more access to credit from friends and relatives, they demand less credit from Azteca, since households can borrow from these suppliers at lower interest rates and more flexible contract terms. Hence, everything else constant, the opening of an Azteca branch will produce more modest changes in municipalities with more access to credit from friends and relatives, which tend to be the treated municipalities. Comparing municipalities selected by Azteca with other municipalities therefore may produce biased estimates of the impact of the bank, since treated municipalities have more access to credit from friends and relatives than control ones. This bias can be measured in the model by examining what would have happened with households from municipalities selected by Azteca in the case that Azteca would not have entered. Column 1 of table 16 presents the biased DID regressions that use as a control group municipalities that were not selected by Azteca in 2005. Column 2 presents the DID regressions that use as control group households from the same municipalities where Azteca entered under the scenario that Azteca never entered. The size of the bias is then given by the difference in DID estimates. According to the model, comparing households from municipalities selected by Azteca with households from municipalities where Azteca decided not to enter overestimates the real impact that Banco Azteca had on the likelihood of saving. In the absence of this bank, the fraction of households saving in treated municipalities would not have decreased as much as what control municipalities suggest. 9 Quantification of the Impact of Banco Azteca 9.1 Estimating the Impact of Banco Azteca on households’ outcomes Saving and Consumption patterns Table 17 presents the means of saving and consumption out- comes of households from municipalities selected by Azteca under two scenarios: with and without 31 the presence of Azteca. The first two columns present the outcomes for the entire sample of house- holds. The next two columns restrict the sample to households who did not obtain offers from the formal sector in any period, and the last columns report outcomes for the sample of households with access to formal jobs. Once Azteca’s credit is available in municipalities, households accumulate more durable goods. Consumption of nondurable goods, however, remains mostly unchanged once Azteca operates its branches. In addition, the fraction of households saving declines, but only households with income below the 40th percentile save less. On average, the fraction of households saving declined by 6:6%. Households who did not obtain any formal-sector offer experience the largest changes from the en- trance of Azteca. Consumption Smoothing To analyze consumption smoothing of households over time, I adapt the index proposed in Mazzocco (2012). The consumption smoothing index is defined as follows: V ar (y h ) V ar (ch ) I= ; V ar (y h ) where V ar(y h ) and V ar(ch ) correspond to the labor income and consumption variances of house- hold h over time. This index (I ) takes values from 0 to 1. See for example the extreme case in which households’ consumption equals their labor income each period. In this case the numerator would equal zero, and hence, I = 0. If households smooth their consumption entirely, then each period households would consume the same amount, V ar(ch ) would equal zero and I = 1. Therefore, the higher the index, the better able are households smoothing their consumption. Figure 2 plots the density of this index for households from municipalities selected by Azteca with and without an operating branch of Banco Azteca. Households’ ability to smooth consumption increases when an Azteca branch operates in the municipality, as seen from the shift to the right of the distribution. This effect is larger for the set of households who are constrained to the informal sector (figure 3). 10 Policy Evaluation The model can be used to understand what would happen if the interest rate that Banco Azteca charges were to be capped by the central authority. Capping interest rates of formal credit institutions has been suggested by several Mexican policy makers who are opposed to excessive interest rates 32 charged to households. This policy however, could have the unintended effect of making Azteca re-locate its branches and exit from municipalities where it currently operates. As the model includes Azteca’s location choices, it is well suited to analyze this issue. I simulate the model at different APRs charged by Azteca to examine changes in household credit requests, loan sizes and ultimately Banco Azteca selection of municipalities. In the model, Azteca locates in municipalities based on its expected profits, which depend on households’ demand for Azteca’s loans. As the number of clients and the amount borrowed by them changes with the APR, Azteca’s selection of municipalities can also change. Figure 4 presents the average size of loans conditional on obtaining Azteca’s credit and the pro- portion of households obtaining Azteca’s credit at different APRs. As the APR declines, the fraction of households requesting Azteca’s credit increases. At an APR of 130%, 3:4% of households obtain credit from Azteca, while at an APR of 40%, this number increases to 5:8%. The average loan of households borrowing from Azteca at an APR of 130% is $462 U SD. This average increases to $885 U SD when the APR charged is 40%. At the current APR, Azteca locates in 0.52 of all municipalities. However, as the APR of Azteca decreases, the number of municipalities with Azteca’s branches declines (Figure 5). At an APR of 40%, Azteca would only open in 25% of the simulated municipalities. Figure 6 provides information regarding the average population size and percapita income of municipalities where Azteca would locate at different APRs. As Azteca’s APR decreases, the average population size of municipalities with Banco Azteca increases. This implies that at lower APRs, Azteca would exit from less populated municipalities and would locate in the more populated ones. The same pattern is observed when examining the percapita income of municipalities with Azteca’s branches. At lower APRs, Azteca’s branches would concentrate in wealthier municipalities. Altogether, these results suggest that if Banco Azteca were forced to charge lower rates and no other adjustment is done, households from poorer and smaller municipalities would lose their Azteca branches. 11 Conclusions This paper examined the impact of expanding access to credit on the decisions and welfare of house- holds. To do so, the paper focused on the opening of Banco Azteca, the first bank in Mexico that targeted households whose members belong to the informal sector. A comparison of household out- comes from municipalities where this bank opened with other municipalities suggests that house- 33 holds, especially those employed in the informal sector, experienced significant changes in their sav- ing, credit and consumption patterns. In order to address the impact of Banco Azteca and the issue of endogenous location of its branches, I developed a model of household choices in which the bank endogenously selects municipalities for branch openings. I used the model to quantify the impact of Banco Azteca. I find substantial effects on household saving and consumption decisions once access to credit is available. First, households increased their bank credit usage and decreased loans from other more expensive suppliers, such as pawn shops. Second, in municipalities with presence of Azteca branches, the fraction of households saving declined. These effects suggest that the use of savings as a buffer on income fluctuations declined once formal credit was available. Simulations of the model indicate that consumption smoothing improved once households had access to credit from Azteca. Consistent with the targeted population of Banco Azteca, informal households experienced most changes. I then use the estimated model to evaluate the effect of capping the APR that Azteca currently charges on household demand for credit and Azteca’s location choices. Several policy makers in Mexico have suggested imposing a ceiling on the interest rates of formal credit institutions in order to make loans more affordable to people. The model simulations indicate that if the APR that Azteca charges were capped to 40%, both the fraction of households obtaining Azteca loans and the aver- age loan size of those borrowing would increase. Nevertheless, this increase in demand would not compensate for the reduction in APR, and half of the municipalities that currently have branches of this bank would lose them. The likelihood of losing a branch is higher for poorer and less populated municipalities. 34 References Aiyagari, S. (1994). Uninsured idiosyncratic risk and aggregate saving. The Quarterly Journal of Eco- nomics 109(3), 659–684. Alessie, R., M. Devereux, and G. Weber (1997). Intertemporal consumption, durables and liquidity constraints: A cohort analysis. European Economic Review 41(1), 37–59. Aportela, F. (1998). Effects of financial access on savings by low-income people. manuscript, Massa- chusetts Institute of Technology, aportela@ mit. edu. Banerjee, A., E. Duflo, R. Glennerster, and C. Kinnan (2009). The miracle of microfinance. Evidence from a randomized evaluation. Bertola, G., R. Disney, and C. Grant (2006). The economics of consumer credit. The MIT Press. Bruhn, M. and I. Love (2013). The economic impact of banking the unbanked: evidence from Mexico. Journal of Finance, forthcoming. Burgess, R. and R. Pande (2005). Do rural banks matter? Evidence from the Indian social banking experiment. American Economic Review 95(3), 780–795. Carroll, C. (1997). Buffer-stock saving and the life cycle/permanent income hypothesis. The Quarterly Journal of Economics 112(1), 1–55. Conger, L. (2003). A bold experiment at banco azteca. Accenture Case Study, Outlook Journal. Crépon, B., F. Devoto, E. Duflo, and W. Parienté (2011). Impact of microcredit in rural areas of mo- rocco: Evidence from a randomized evaluation. Massachusetts Institute of Technology. Deaton, A. (1991). Saving and liquidity constraints. Econometrica 59, 1221–1248. Dupas, P., S. Green, A. Keats, and J. Robinson (2012). Challenges in banking the rural poor: Evidence from kenya’s western province. Technical report, National Bureau of Economic Research. Giné, X. (2011). Access to capital in rural thailand: An estimated model of formal vs. informal credit. Journal of Development Economics 96(1), 16–29. Jain, S. (1999). Symbiosis vs. crowding-out: the interaction of formal and informal credit markets in developing countries. Journal of Development Economics 59(2), 419–444. 35 Kaboski, J. and R. Townsend (2009). The impacts of credit on village economies. MIT Department of Economics Working Paper No. 09-13. Kaboski, J. and R. Townsend (2011). A structural evaluation of a large-scale quasi-experimental mi- crofinance initiative. Econometrica 79(5), 1357–1406. Karlan, D. and J. Zinman (2009). Expanding credit access: Using randomized supply decisions to estimate the impacts. Review of Financial Studies. Keane, M. and K. Wolpin (1994). The solution and estimation of discrete choice dynamic program- ming models by simulation and interpolation: Monte Carlo evidence. The Review of Economics and Statistics, 648–672. Kennan, J. (2006). A note on discrete approximations of continuous distributions. University of Wis- consin Madison. Lee, J. and Y. Sawada (2010). Precautionary saving under liquidity constraints: Evidence from rural pakistan. Journal of Development Economics 91(1), 77–86. Levy, S. (2008). Good intentions, bad outcomes: Social policy, informality, and economic growth in Mexico. Brookings Inst Press. López Calva, L. and R. de la Torre García (2005). Informe sobre desarrollo humano méxico 2004. México, Programa de las Naciones Unidas para el Desarrollo. López Calva, L. and T. de la Torre García (2003). Informe sobre desarrollo humano méxico 2002. México, Programa de las Naciones Unidas para el Desarrollo. Mazzocco, M. (2012). Testing efficient risk sharing with heterogeneous risk preferences. The American Economic Review 102(1), 428–468. Rhyne, E. (2009). Microfinance for Bankers and Investors: Understanding the Opportunities and Challenges of the Market at the Bottom of the Pyramid. McGraw-Hill. 36 Table 1: Characteristics of municipalities by presence of Banco Azteca All muns with muns with muns with muns without Variable muns Azteca Azteca from Azteca after Azteca’s from 02 to 05 02 before 05 02 until 05 branches Population size p10 9; 413 57; 375 62; 773 47; 106 4; 318 p25 19; 475 78; 512 62; 773 50; 168 10; 644 p50 59; 675 227; 026 63; 319 57; 602 19; 447 p75 226; 642 516; 255 63; 864 366; 068 41; 402 p90 609; 829 1; 110; 997 63; 864 670; 162 69; 381 Percapita Income p10 18; 789 31; 642 48; 436 25; 375 15; 100 p25 28; 687 38; 561 48; 436 26; 349 22; 908 p50 37; 136 50; 912 48; 876 31; 308 30; 434 p75 54; 551 66; 846 49; 317 57; 985 37; 044 p90 71; 967 76; 379 49; 317 80; 676 50; 423 Other banks 0:71 0:98 1:00 1:00 0:43 # muns 136 63 2 4 67 Table reports population size and percapita income percentiles (10th,25th,50th,75th,90th) of the 136 municipalities of the sample, and the fraction of municipalities with presence of branches from other banks. Data from Mexican Economic Census, Human Development Index and CNBV. Table 2: Households’ Descriptive Statistics Azteca No Azteca Number of households 2002 3,483 2,156 2005 3,470 2,146 Fraction of formal hhds 2002 0.238 0.131 (0.426) (0.337) 2005 0.229 0.125 (0.421) (0.331) Table reports means and standard deviations (in parenthesis) of households’ characteristics from muns with and without an Azteca branch in 2005. 37 Table 3: Saving patterns of all households Azteca No Azteca DID Probability hhd is aware it can obtain loans from bank 2002 0.1513 0.1026 0.0887*** 2005 0.3064 0.1740 [0.0312] Probability hhd obtained loan from bank 2002 0.0083 0.0042 0.0108** 2005 0.0242 0.0093 [0.0042] Probability hhd obtained loan from pawnshop 2002 0.0221 0.0088 -0.0087* 2005 0.0112 0.0065 [0.0050] Probability hhd saved 2002 0.3672 0.2500 -0.0464** 2005 0.2646 0.1943 [0.0208] Average savings of hhds 2002 9,588.4 5,774.6 447.3 2005 8,508.1 4,244.7 [1,294.2] Avg savings of hhds below the 40th income pctile 2002 4,978.7 3,431.4 -2,449.7* 2005 2,232.1 2,259.9 [1,486.1] Avg savings of hhds below the 60th income pctile 2002 5,417.2 4,910.6 418.8 2005 2,999.2 2,714.8 [1,335.0] Columns 1 and 2 report means of households’ characteristics from municipalities with and without Azteca in 2005. Column 3 reports the difference-in-difference (DID) estimates and the standard errors in parentheses. Standard errors are clustered at the municipality level. 38 Table 4: Saving patterns of informal households Azteca No Azteca DID Probability hhd is aware it can obtain loans from bank Azt No Azt DD 2002 0.1322 0.0966 0.1009*** 2005 0.2761 0.1537 [0.0305] Probability hhd obtained loan from bank 2002 0.0064 0.0037 0.0139*** 2005 0.0232 0.0075 [0.0045] Probability hhd obtained loan from pawnshop 2002 0.0218 0.0069 -0.0105* 2005 0.0105 0.0059 [0.0059] Probability hhd saved 2002 0.3337 0.2231 -0.0438* 2005 0.2290 0.1677 [0.0230] Average savings of hhds 2002 9,000.3 4,969.2 -926.1 2005 6,396.4 2,975.4 [1,452.5] Avg savings of hhds below the 40th income pctile 2002 4,855.5 3,164.7 -2,264.7* 2005 1,633.4 2,142.7 [1,239.7] Avg savings of hhds below the 60th income pctile 2002 5,142.7 4,173.7 257.3 2005 2,365.9 2,504.0 [1,526.7] Columns 1 and 2 report means of informal households’ characteristics from municipalities with and without Azteca in 2005. Column 3 reports the difference-in-difference (DID) estimates and the standard errors in parentheses. Standard errors are clustered at the municipality level. 39 Table 5: Saving patterns of formal households Azteca No Azteca DID Probability hhd is aware it can obtain loans from bank 2002 0.2121 0.1418 0.0147 2005 0.4056 0.3146 [0.0819] Probability hhd obtained loan from bank 2002 0.0145 0.0071 -0.0135 2005 0.0276 0.0224 [0.0203] Probability hhd obtained loan from pawnshop 2002 0.0229 0.0213 0.0163 2005 0.0138 0.0112 [0.0213] Probability hhd saved 2002 0.4746 0.4291 -0.0214 2005 0.3844 0.3806 [0.0685] Average savings of hhds 2002 11,474.2 11,126.7 2,490.5 2005 15,575.6 13,120.7 [7,218.5] Avg savings of hhds below the 40th income pctile 2002 5,552.4 6,182.7 2,636.1 2005 5,718.0 3,681.2 [3,760.6] Avg savings of hhds below the 60th income pctile 2002 6,551.7 11,308.0 5,609.1 2005 5,993.5 4,959.5 [5,486.2] Columns 1 and 2 report means of formal households’ characteristics from municipalities with and without Azteca in 2005. Column 3 reports the difference-in-difference (DID) estimates and the standard errors in parentheses. Standard errors are clustered at the municipality level. 40 Table 6: Consumption patterns of all households Azteca No Azteca DID PANEL A Probability hhd owns electronic appliances 2002 0.9730 0.9336 0.0223** 2005 0.9668 0.9057 [0.0104] Avg value of electronic appliances 2002 7,162.1 5,704.0 2,305.4*** 2005 8,349.2 4,919.2 [633.9] Probability hhd owns large appliances/furniture 2002 0.9641 0.8876 0.0228* 2005 0.9442 0.8466 [0.0135] Avg value of large appliances/furniture 2002 9,715.4 8,031.4 2,342.1** 2005 10,779.0 7,190.7 [1,173.1] Probability hhd owns other appliances 2002 0.9540 0.8862 0.0224 2005 0.9271 0.8386 [0.0153] Avg value of other appliances 2002 2,003.9 1,358.2 463.7 2005 2,465.7 1,363.5 [566.6] PANEL B Avg pce if death of a hhd member 2002 21,456.2 23,523.3 9,324.3* 2005 20,069.9 11,371.7 [5,192.3] Avg pce if serious illness of a hhd member 2002 19,557.2 19,517.7 6,973.9* 2005 20,791.5 12,526.2 [3,973.4] Avg pce if unemployment/failure of business 2002 18,674.0 17,455.5 7,244.6** 2005 20,787.2 10,985.6 [3,080.9] Columns 1 and 2 report means of households’ characteristics from municipalities with and without Azteca in 2005. Column 3 reports the difference-in-difference (DID) estimates and the standard errors in parentheses. Standard errors are clustered at the municipality level. 41 Table 7: Consumption patterns of informal households Azteca No Azteca DID PANEL A Probability hhd owns electronic appliances 2002 0.9691 0.9257 0.0244* 2005 0.9625 0.8949 [0.0129] Avg value of electronic appliances 2002 7,083.9 5,458.7 1,456.1** 2005 7,207.3 4,156.4 [589.6] Probability hhd owns large appliances/furniture 2002 0.9578 0.8739 0.0207 2005 0.9345 0.8290 [0.0178] Avg value of large appliances/furniture 2002 9,495.6 7,543.9 2,756.6* 2005 9,830.4 5,820.9 [1,623.8] Probability hhd owns other appliances 2002 0.9465 0.8771 0.0282 2005 0.9146 0.8225 [0.0191] Avg value of other appliances 2002 2,049.5 1,298.5 351.0 2005 2,527.7 1,166.3 [503.0] PANEL B Avg pce of hhds if death of a hhd member 2002 22,120.5 24,962.5 9,034.2 2005 18,425.2 10,760.8 [5,529.7] Avg pce of hhds if serious illness of a hhd member 2002 17,860.4 20,307.0 9,339.0** 2005 19,631.0 12,133.1 [4,160.4] Avg pce of hhds if unemployment/failure of business 2002 17,766.4 16,543.9 8,463.0** 2005 20,469.0 9,700.7 [3,421.0] Columns 1 and 2 report means of informal households’ characteristics from municipalities with and without Azteca in 2005. Column 3 reports the difference-in-difference (DID) estimates and the standard errors in parentheses. Standard errors are clustered at the municipality level. 42 Table 8: Consumption patterns of formal households Azteca No Azteca DID PANEL A Probability hhd owns electronic appliances 2002 0.9855 0.9858 -0.0174 2005 0.9811 0.9813 [0.0114] Avg value of electronic appliances 2002 7,405.5 7,254.7 1,671.9 2005 11,965.9 9,739.8 [2,984.8] Probability hhd owns large appliances/furniture 2002 0.9843 0.9787 0.0073 2005 0.9760 0.9700 [0.0148] Avg value of large appliances/furniture 2002 10,396.2 11,057.1 -6,370.2 2005 13,709.2 15,063.1 [9,272.7] Probability hhd owns other appliances 2002 0.9783 0.9468 -0.0018 2005 0.9684 0.9513 [0.0263] Avg value of other appliances 2002 1,863.3 1,732.7 282.4 2005 2,278.7 2,514.1 [428.4] PANEL B Avg pce of hhds if death of a hhd member 2002 18,989.0 14,528.3 10,187.7 2005 33,227.4 15,299.4 [7,254.4] Avg pce of hhds if serious illness of a hhd member 2002 25,272.6 16,755.4 -3,936.1 2005 23,915.7 15,907.1 [9,632.4] Avg pce of hhds if unemployment/failure of business 2002 21,235.4 21,922.2 1,240.1 2005 21,685.9 19,722.5 [7,375.3] Columns 1 and 2 report means of formal households’ characteristics from municipalities with and without Azteca in 2005. Column 3 reports the difference-in-difference (DID) estimates and the standard errors in parentheses. Standard errors are clustered at the municipality level. 43 Table 9: DIDs of informal households using alternative treatment group Azteca Azteca by in 05 06 or later Probability hhd is aware it can obtain loans from bank 2002 0.1009*** 0.0767* 2005 [0.0305] [0.0420] Probability hhd obtained loan from bank 2002 0.0139*** 0.0096 2005 [0.0045] [0.0078] Average size of bank loans 2002 485.7* 283.6 2005 [273.8] [272.1] Probability hhd obtained loan from pawnshop 2002 -0.0105* -0.0018 2005 [0.0059] [0.0047] Probability hhd saved 2002 -0.0438* -0.0129 2005 [0.0230] [0.0487] Average savings of hhds 2002 -926.1 -3,051.5 2005 [1,452.5] [3,698.3] Avg savings of hhds below the 40th income pctile 2002 -2,264.7* -2,508.6 2005 [1,239.7] [2,232.8] Avg savings of hhds below the 60th income pctile 2002 257.3 -1879.9 2005 [1,526.7] [3,766.5] Table reports DID estimates and their standard errors in parentheses. Standard errors are clustered at the municipality level. 44 Table 10: DIDs of informal households using alternative treatment group Azteca Azteca by in 05 06 or later Probability hhd owns electronic appliances 2002 0.0244* 0.0017 2005 [0.0129] [0.0225] Avg value of electronic appliances 2002 1,456.1** -120.9 2005 [589.6] [920.4] Probability hhd owns large appliances/furniture 2002 0.0207 0.0459 2005 [0.0178] [0.0293] Avg value of large appliances/furniture 2002 2,756.6* 1,553.6 2005 [1,623.8] [2,000.8] Probability hhd owns other appliances 2002 0.0282 -0.0301 2005 [0.0191] [0.0326] Avg value of other appliances 2002 351.0 -902.40 2005 [503.0] [584.1] Avg pce of hhds if death of a hhd member 2002 9,034.2 -6,970.10 2005 [5,529.7] [12,020.3] Avg pce of hhds if serious illness of a hhd member 2002 9,339.0** -12,700.1 2005 [4,160.4] [7,801.8] Avg pce of hhds if unemployment/failure of business 2002 8,463.0** 5,707.1 2005 [3,421.0] [3,608.6] Table reports DID estimates and their standard errors in parentheses. Standard errors are clustered at the municipality level. 45 Table 11: Moments used in the model estimation Fraction of households in the formal sector at 2005 that 1) belonged to the informal sector in 2004 whose head had less than high school 2) belonged to the informal sector in 2004 whose head had at least high school 3) belonged to the formal sector in 2004 whose head had less than high school 4) belonged to the formal sector in 2004 whose head had at least high school Labor income from the formal sector (sample of formal households at 2005) 5) OLS coefficient of labor income of 2004 on labor income of 2005 6) Difference between mean income of low- and high-educated hhds 7) Difference between mean income of hhds with head older/younger than 35 8) Difference between mean income of hhds with head older/younger than 50 9) Variance of the residuals from the OLS regression of current labor income Labor income from the informal sector (sample of informal households at 2005) 10) OLS coefficient of labor income of 2004 on labor income of 2005 11) Difference between mean income of low- and high-educated hhds 12) Difference between mean income of hhds with head older/younger than 35 13) Difference between mean income of hhds with head older/younger than 50 14) Variance of the residuals from the OLS regression of current labor income Percapita expenditure and durable goods consumption patterns 15) 5th percentile of the households’ percapita expenditure distribution 16) Fraction of households that own household and electronic appliances 17) Average ratio of percapita expenditure to total labor income in 2005 18) Average log ratio of percapita expenditure (pce) in 2005 to pce in 2002 Credit coverage of friends and relatives (F-R) in muns in 2005 19) % of hhds with F-R loans in muns below the mean fraction of F-R credit 20) % of hhds with F-R loans in muns above the mean fraction of F-R credit 21) % of hhds with F-R loans in low-income muns below mean fraction of F-R credit 22) % of hhds with F-R loans in low-income muns above mean fraction of F-R credit 23) % of hhds with F-R loans in high-income muns below mean fraction of F-R credit 24) % of hhds with F-R loans in high-income muns above mean fraction of F-R credit Entry patterns of Banco Azteca at the municipality level 25) % of muns with an Azteca branch in 2005 26) % of muns with pop. size >10th & <30th pctiles with an Azteca branch in 2005 27) % of muns with pop. size >30th & <50th pctiles with an Azteca branch in 2005 28) % of muns with pop. size >50th & <70th pctiles with an Azteca branch in 2005 29) % of muns with Elektra in 2002 and Azteca branch from 2002 to 2005 46 Table 12: Estimated parameters param ASE Probability of job offer from the formal sector Low educated households, informal sector at t-1 0:070 0:0020 Low educated households, formal sector at t-1 0:550 0:0172 High educated households, informal sector at t-1 0:132 0:0014 High educated households, formal sector at t-1 0:772 0:0035 Labor income from the formal sector Persistence of one-year lagged labor income 0:492 0:0659 Returns to education 1:102 0:0255 Returns to age 0:036 0:0071 Returns to age squared 0:0004 0:0002 St dev of residuals 2:037 0:2488 Labor income from the informal sector Persistence of lagged labor income 0:516 0:0190 Returns to education 1:022 0:0683 Returns to age 0:019 0:0047 Returns to age squared 0:0002 0:0000 St dev of residuals 2:095 0:1183 Credit coverage from friends and relatives (F-R) % hhds with access to F-R credit in muns with low coverage 0:050 0:0172 % hhds with access to F-R credit in muns with high coverage 0:321 0:1545 Probability that mun has high coverage from F-R if mun is low income and had low coverage of F-R credit in t-1 0:114 0:0625 if mun is low income and had high coverage of F-R credit in t-1 0:879 0:2348 if mun is high income and had low coverage of F-R credit in t-1 0:186 0:0025 if mun is high income and had high coverage of F-R credit in t-1 0:936 0:0731 Utility parameters Intertemporal substitution of the nondurable good 2:145 0:5593 Preference from durable goods service flow 2:100 0:3420 Subsistence consumption level 0:365 0:0000 Azteca’s operating cost parameters fixed cost to operate a branch 221:7 79:71 additional fixed cost in muns without Elektra stores 210:3 58:22 pop size shifter 0:00453 0:0006 pop size shifter for muns above pop size threshold 0:0039 0:0003 pop size threshold 55000 23571 Columns 1 and 2 report the structural parameters and their asymptotic standard errors. 47 Table 13: Model Fit- Households’ moments in 2005 model data Fraction of households in the formal sector at 2005 who were: Informal at 2004 and low educated 0:0416 0:0437 Informal at 2004 and high educated 0:4906 0:4931 Formal at 2004 and low educated 0:1114 0:1100 Formal at 2004 and high educated 0:7020 0:7096 Labor income from the formal sector h on y h OLS coefficient of yt 0:4097 0:4264 t 1 Returns on education 1:9516 1:9623 Returns on age (if age 35) 1:0374 1:0094 Returns on age (if age 50) 0:1385 0:1441 Variance of OLS residuals of labor income regression 1:8980 2:4480 Labor income from the informal sector h on y h OLS coefficient of yt 0:4304 0:4301 t 1 Returns on education 1:1606 0:9027 Returns on age (if age 35) 0:1295 0:0167 Returns on age (if age 50) 0:3838 0:2463 Variance of OLS residuals of labor income regression 1:7277 2:1096 Durable and non-durable goods consumption patterns 5th percentile of hhds’ consumption 0:3651 0:3658 Fraction of hhds that own durable goods 0:9970 0:9567 Average ratio of ch h 2005 to y2005 0:8214 0:7708 Average log ratio of ch h 2005 =y2005 0:0949 0:1030 Credit coverage of friends and relatives (F-R) at the mun level % hhds with F-R loans in muns below mean F-R credit 0:0157 0:0141 % hhds with F-R loans in muns above mean F-R credit 0:0699 0:0812 % hhds with F-R loans in low-income muns < mean F-R credit 0:0333 0:0384 % hhds with F-R loans in low-income muns > mean F-R credit 0:0580 0:0740 % hhds with F-R loans in high-income muns < mean F-R credit 0:0362 0:0376 % hhds with F-R loans in high-income muns > mean F-R credit 0:0521 0:0541 Entry patterns of Banco Azteca at the mun level % muns with an Azteca branch in 2005 0:522 0:510 % muns with pop. size >10th & <30th pctiles with Azteca in 2005 0:086 0:114 % muns with pop. size >30th & <50th pctiles with Azteca in 2005 0:594 0:514 % muns with pop. size >50th & <70th pctiles with Azteca in 2005 1:000 0:923 % muns with Elektra in 2002 and Azteca from 2002 to 2005 0:867 0:883 48 Table 14: Municipalities characteristics by presence of Banco Azteca Variable from 02 from 02 after 02 after 02 no Azteca’s to 05 before 05 until 05 before 05 branches DATA Population size p10 57; 375 62; 773 47; 106 4; 318 p25 78; 512 62; 773 50; 168 10; 644 p50 227; 026 63; 319 57; 602 19; 447 p75 516; 255 63; 864 366; 068 41; 402 p90 1; 110; 997 63; 864 670; 162 69; 381 Percapita Income p10 31; 642 48; 436 25; 375 15; 100 p25 38; 561 48; 436 26; 349 22; 908 p50 50; 912 48; 876 31; 308 30; 434 p75 66; 846 49; 317 57; 985 37; 044 p90 76; 379 49; 317 80; 676 50; 423 Other banks 0:98 1 1 0:43 Number of muns 63 2 4 0 67 MODEL Population size p10 61; 974 53; 229 47; 934 49; 145 2; 723 p25 88; 343 57; 375 48; 917 49; 145 9; 862 p50 227; 026 62; 773 93; 476 49; 145 15; 261 p75 516; 255 63; 864 145; 622 49; 145 30; 017 p90 1; 110; 997 68; 738 491; 436 49; 145 46; 053 Percapita Income p10 28; 612 35; 293 29; 606 26; 207 13; 762 p25 36; 705 39; 447 29; 885 26; 207 20; 134 p50 50; 200 45; 680 31; 587 26; 207 30; 842 p75 66; 846 48; 436 33; 916 26; 207 41; 212 p90 80; 676 49; 317 67; 025 26; 207 51; 731 Other banks 1 1 0:83 1 0:36 Number of muns 65 5 6 1 59 Panels compare the presence of Azteca branches between model simulations and real data, by population size, percapita income of municipalities (10th,25th, 50th,75th, 90th percentiles) and presence of other bank branches. 49 Figure 1: Expected profits of Azteca in different municipalities Puebla, Jopala Amecameca, Mexico P opulation size: 13,486. P ercapita Income: Mx $16,514 P opulation size: 45,255. P ercapita Income: Mx $36,704 100 0 -100 0 A verage expected profits A verage expected profits -200 -100 -300 -200 -400 -300 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 A ccess to Friends and Relatives Credit A ccess to Friends and Relatives Credit Linares, Nuevo Leon Apodaca, Nuevo Leon P opulation size: 69,205. P ercapita Income: Mx $44,495 P opulation size: 283,497. P ercapita Income: Mx $60,616 1,500 400 1,000 200 A verage expected profits A verage expected profits 500 0 -200 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 A ccess to Friends and Relatives Credit A ccess to Friends and Relatives Credit 50 Table 15: DID on saving outcomes for real and simulated data DID data DID model Probability of borrowing from banks All households 0.0108** 0.0315*** [0.0042] [0.0014] Informal households 0.0139*** 0.0421*** [0.0045] [0.0014] Formal households -0.0135 -0.009** [0.0203] [0.0041] Probability of borrowing from pawn shops All households -0.0087* -0.021*** [0.0050] [0.0022] Informal households -0.0105* -0.028*** [0.0059] [0.0024] Formal households 0.0163 0.0042** [0.0213] [0.0016] Probability of saving All households -0.0464** -0.052*** [0.0208] [0.0094] Informal households -0.0438* -0.057*** [0.0230] [0.0100] Formal households -0.0214 -0.016** [0.0685] [0.0077] 51 Table 16: DID coefficients with different control groups (simulated data) Muns where Azteca Muns where Azteca decided not to enter decided to enter prob borrowing from pawn shops All households 0:021 0:019 Informal households 0:028 0:025 Formal households 0:0042 0:000 prob saving All households 0:052 0:007 Informal households 0:057 0:009 Formal households 0:016 0:001 Figure 2: Density of the consumption smooting index 2.5 2 1.5 Density 1 .5 0 0 .2 .4 .6 .8 1 Consumption Smoothing Index with azteca without azteca 52 Figure 3: Density of the consumption smooting index by access to the formal sector 2.5 2 1 1.5 Density .5 0 0 .2 .4 .6 .8 1 households without access to the formal sector 2 2.5 1 1.5 Density .5 0 0 .2 .4 .6 .8 1 households with access to the formal sector with azteca without azteca Table 17: Saving and consumption patterns of households by presence of Azteca All households Hhds with no access Hhds with access to the formal sector to the formal sector No Azteca Azteca No Azteca Azteca No Azteca Azteca Average value of durable goods 27845:3 28907:2 22730:6 23899:0 30730:8 32014:8 Average consumption 24746:8 24849:6 23035:7 23173:2 25712:2 25795:4 Fraction of hhds saving 0:769 0:721 0:682 0:633 0:818 0:811 Average savings 8435:8 9767:8 3689:6 4522:9 11113:4 12726:8 Avg savings below 40th pctile 2042:5 2011:4 1275:8 1078:8 2882:3 2959:1 Avg savings below 60th pctile 3479:3 3886:7 2092:7 2439:3 4726:0 5236:2 53 Figure 4: Azteca loans by APR .065 900 .06 800 .055 .05 700 Loan size (USD) % borrowers .045 600 .04 .035 500 .03 .025 400 .4 .5 .6 .7 .8 .9 1 1.1 1.2 1.3 Azteca's APR Fraction of borrowers Av g loan size in USD Figure 5: municipalities with Banco Azteca branches by APR .6 .5 % municipalities .4 .3 .2 .4 .5 .6 .7 .8 .9 1 1.1 1.2 1.3 Azteca's APR 54 Figure 6: Size and income of Azteca municipalities by APR 700000 65000 600000 60000 Percapita income Population size 500000 55000 400000 300000 50000 .4 .5 .6 .7 .8 .9 1 1.1 1.2 1.3 Azteca's APR Population size Percapita income 55