THE IMPACT OF BANK CORRESPONDENTS ON FINANCIAL INCLUSION: A FIRST EVALUATION

Transcription

1 THE IMPACT OF BANK CORRESPONDENTS ON FINANCIAL INCLUSION: A FIRST EVALUATION Pablo Peña and Alejandro Vázquez I. Introduction Although there is yet no consensus on the meaning of financial inclusion, in the last couple of years this term has become part of the jargon used by economists and financial authorities, and there are international efforts aimed at promoting it. 1 A first approach to its meaning is the definition adopted by Mexican authorities, according to which financial inclusion is the access to and use of financial services under an appropriate regulation, that guarantees consumer protection schemes and promotes financial education to improve the financial capacities of all population segments. 2 The concern of authorities for financial inclusion arises out of the idea that, by being part of the formal financial system, people may achieve a higher level of wellbeing. Formal channels allow for people to transfer resources more efficiently between periods (by saving or by taking loans), between uncertain situations (through micro-insurances, precautionary savings and credit for emergencies) and between different geographic locations (sending and receiving money). The greatest efficiency when transferring resources must translate into greater wellbeing. Some studies find evidence in this sense: financial inclusion is positively related to the growth of economic activity. 3 To understand financial inclusion in the Mexican context, first it would be convenient to paint a panorama of the financial system. Institutions that offer financial services may be classified into three types: formal regulated, formal unregulated and informal. Formal regulated institutions comprise banks, savings institutions, popular savings associations, development banks, credit unions, limited-purpose 1 An example is the network Alliance for Financial Inclusion, known as AFI. 2 This definition was adopted in the agreement created by the National Financial Inclusion Council (October 3, 2011, Official Gazette of the Federation). The reader interested in a discussion about the aspects to be considered by financial inclusion may consult the Financial Inclusion Reports of the National Banking and Securities Commission (CNBV 2010, 2011a and 2011b). 3 See Levine (1997), and Beck, Demirgüç-Kunt and Levine (2007). 193

2 financial institutions and multiple-purpose financial institutions related to a financial group. Formal unregulated institutions comprise multiplepurpose financial institutions not related to a financial group and pawn shops. Informal institutions include group savings pools and usurers. The three types of intermediaries are not present in all places. The absence of some formal intermediaries such as banks is evident in rural areas. 4 In those places, the volume of transactions may not compensate the cost of opening and maintaining a branch office. Other formal intermediaries such as savings institutions have attended to some extent to the demand of financial services in those areas. It is however unclear whether they are sufficient. 5 Financial inclusion is not only an objective in rural areas. Even in urban areas where formal intermediaries such as banks are present many people do not use formal financial services. According to the 2010 National Survey on the Income and Expenses of Homes, only 30% of homes in places with 100,000 inhabitants or more possess a credit card. The inhabitants of cities have access to formal financial services, but probably face obstacles to use them. The lack of documentation is one of such obstacles. Until not so long ago, banks required documents as proof of identity and address practically for any proceeding whatsoever. Many people lack these documents and were therefore excluded. Even having the possibility to use the financial system, some people do not have the capability to compare and evaluate financial products. According to a survey by the World Bank, around 55% of adults in Mexico feel that they are not capable at all or little capable to compute interest rates. 6 For this reason, the definition of financial inclusion goes beyond access and use, and includes the financial education and protection of consumers. Mexican authorities have implemented public policies to foster the access to and use of financial services. The best example may be the object of study of this chapter: the creation of the legal concept of bank correspondents. Through a regulatory change, the government allowed banks to establish contact points with their clients through third parties. With this measure, the number of contact points between clients and banks has multiplied in recent years. This expansion may already have an effect on financial inclusion. 4 See the Financial Inclusion Reports of CNBV (2010, 2011a and 2011b). 5 The results of the National Survey on Financial Inclusion (ENIF) to be published soon will allow for having a better idea of the satisfaction of the financial service demand. See CNBV (2012). 6 Financial Capability Survey, to be published soon. 194

3 The purpose of this chapter is to analyze the impact of the creation of the legal concept of bank correspondents on the use of financial services. Before presenting the evaluation, we will first briefly mention different types of financial inclusion policies, and then describe bank correspondents and the evolution of their presence. II. Financial inclusion policy Formal financial services expand the possibilities of production and consumption by people, homes, and communities. By using the formal financial system, it is possible to safeguard money in a more secure manner, to have a record of inflows and outflows, take loans in better conditions for emergencies or to finance projects such as a home remodeling. The formal financial system also allows for remittances to be sent and received in a cheaper and safer manner. On different fronts, Mexican authorities have implemented several policies to promote financial inclusion. 7 These policies may be grouped into two categories. The first category includes active interventions, in which the government meddles with the market. Some examples of programs are Cetes Directo, Fondo de Microfinanciamiento a Mujeres Rurales and the legal obligation of banks to offer basic products without commissions. The second category comprises passive interventions, in which the government stays out of the market. In these policies, the government relaxes the regulatory framework so that intermediaries may provide more services. In those cases, the authority does not make intermediaries offer more services, nor does it offer them directly. It simply opens a legal space for, if they decide so, intermediaries to do this for their own interest. In addition to the bank correspondents described below, this category also includes the introduction of simplified account opening and of the legal framework for bank transactions via mobile phone service. For the purposes of simplified account opening, the authority defined four account levels distinguished by their maximum transaction volume and by the documents required to open them. For instance, to open the simplest account (level 1) no data of the account 7 Since 2011 attempts have been made to form a common front with the creation of the National Financial Inclusion Council (CONAIF), with the participation of the Bank of Mexico, the Ministry of Finance, CNBV and others authorities. The intent is for CONAIF to consolidate the efforts of authorities to create a unique financial inclusion strategy. 195

4 holder are required. 8 As a result of this change, people who lack legal documents are expected to have better access to basic formal financial services. In the case of mobile phone service, regulatory changes allow the use of mobile phones to make payments associating a phone number to the client s bank account. This channel reduces transaction costs for both parties: the bank saves resources and the client saves time and effort. These savings could translate into greater access and use. Bank correspondents Basically, bank correspondents are companies that allow banks to have more contact points to attend their customers. The opening of correspondents in Mexico may be seen as part of an international trend. More and more countries are adopting the concept of correspondents in one way or another. 9 As we stated before, the creation of the legal framework for correspondents was a passive policy. The government did not open correspondents. It did not make banks do it either. No public resources were used to subsidize their creation. Simply, authorities relaxed the regulatory framework. Its effects, whichever they may be, are a consequence of voluntary decisions by banks and by those who want to become correspondents, and taxpayers have no charges implied in them. In synthesis, the legal change that created the concept of correspondents allows a third party to act on behalf of the bank, offering financial products and services to customers. 10 The regulation allows correspondents to offer the following services: cash deposits and withdrawals, the payment of services such as electricity and telephone, credit payment, account opening, the consultation of balances, and remittances. 11 Correspondents may be supermarkets, pharmacies, 8 Level 1 does not require data of the account holder, but only allows deposits up to 750 investment units (UDIS) per month and a maximum of 1,000 UDIS as monthly balance. Level 2 only requires name, date of birth and address, and allows deposits up of 3,000 UDIS per month. Accounts level 1 and 2 do not require the presence of the client to be opened. Levels 3 and 4 do require the presence of the client and more documents, but allow for transactions of higher amount. In October 2012 one UDI was equal to 4.81 pesos. 9 For more details on international experience see CNBV (2011b). 10 When the reform was proposed, CNBV suggested that other financial institutions such as savings institutions may also have correspondents. 11 Cash may be withdrawn for up to 1,500 investment units (UDIS) per day, and deposits, up to 4,000 UDIS per account. 196

5 convenience stores, post offices, lottery stands, service stations, department stores, etc. Banks do not need to recruit correspondents store by store. The law provides for the use of store networks, such as commercial chains. Furthermore, interested third parties may be correspondents of several banks at the same time. All correspondent transactions must be undertaken through pointof-sale terminals managed by the bank. Transactions must be executed on line and in real time, so that the accounts of the customer and of the correspondent are affected at the same time. As a safety measure, it is not possible to store sensitive customer information on the correspondent s electronic media, and two customer authentication factors are used for withdrawals and to consult balance. Correspondents must issue receipts with the specific data of every transaction. They must also indicate contact data to attend to doubts or complaints by customers. In theory, correspondents generate a benefit for the three parties involved. Customers win in terms of convenience. Banks win because the highest convenience may translate into a higher use of services. Correspondents are benefited in two ways. First, by the commissions per transaction agreed upon with the bank. For every service payment or for every deposit through the correspondent, the correspondent receives a commission. Second, by the higher number of people approaching the store to execute banking transactions. More people mean more potential customers for the correspondent s main line of business. Readers interested in learning more about the details of correspondent operation may consult the report called Business Models for Financial Inclusion (CNBV, 2011c). The report also contains an annex prepared by the Association of Banks of Mexico that collects the experience of banks with their correspondents. Evolution of operating correspondents The authorization of correspondents started in the first quarter of 2010 and by the end of the fourth quarter of 2011, 21,071 correspondents have been authorized. Among the most widely known commercial chains that act as correspondents are Walmart, Comercial Mexicana, Soriana, Chedraui, Suburbia, Coppel, Oxxo and 7-Eleven. 197

6 Graph 1. Number of operating correspondents and percentage of municipalities covered in the country 20,000 55% 15,000 10,000 50% 51% 51% 52% 5,000 46% 47% 48% Trimestre Note: the total of municipalities in the country (including the delegations of the Federal District as municipalities) is 2,456. Source: file A-R10, CNBV Information Portfolio, December Graph 1 shows the number of operating correspondents. 12 It also presents the percentage of municipalities with at least one correspondent, i.e. the municipal coverage. 13 The information is obtained from the regulatory reports submitted by banks to the National Banking and Securities Commission (CNBV). 14 In the fourth quarter of 2011 there were 21,040 correspondents operating in 55% of the country s municipalities. The end of 2011 saw the most relevant increase in the number of correspondents, which almost doubled. This jump was due to the incorporation of the Oxxo store chain, with more than 8,000 establishments. In the period shown in graph 1, the number of correspondents increased five times and the municipal coverage increased from 46 to 55%. This means that most of the almost 17,000 correspondents that started operations between the second quarter of 2010 and the end of 12 The number of authorized correspondents and the number of operating correspondents do not match because of a shift. Operation starts after the authorization. 13 For statistical purposes, in this chapter the 16 political delegations of the Federal District are considered municipalities. 14 Part of this information is found in the Information Portfolio on the CNBV web page at 198

7 2011 are located in municipalities where other correspondents were already operating. Not all banks have developed a correspondent network with the same momentum. Chart 1 presents the share of each bank in the total of correspondents by December Only banks with correspondents are presented. Interestingly, among the five banks with the highest share are two relatively small banks: American Express Bank and Invex. By their operation scale, it could be that these two banks use their correspondents in a different way than BBVA Bancomer, Banamex or Banorte, which are the largest banks. Chart 1. Correspondent share by banking institution (%) BBVA Bancomer 25 HSBC 8 Bancoppel 3 American Express 12 Banco Walmart 7 Compartamos 1 Banamex 10 Inbursa 6 Afirme 0 Banorte 9 Santander 5 Invex 9 Scotiabank 5 Source: file A-R10, CNBV Information Portfolio, December The information of regulatory reports about the type of transaction undertaken at correspondents suggests that banks use correspondents under different business models. Chart 2 presents the percentage composition of transactions by the correspondents of each bank. The correspondents of American Express, Compartamos and BBVA Bancomer are used proportionally more for credit payment. The correspondents of Bancoppel, Inbursa and Scotiabank are used proportionally more for receiving deposits, and those of Santander, for cash withdrawal. The information in Chart 2 is consistent with a variety of different bank strategies. In addition to the number and location of correspondents, banks are distinguished by the services provided to their clients through correspondents. In brief, since the creation of their legal concept, correspondents have been used more and more. The next section will present a first evaluation of their impact. 199

8 Chart 2. Participation of services used at correspondents by banking institution (%) Credit Service Cash Deposits payment payment withdrawal Others Afirme 69*** 27*** 3*** 1*** 1*** American Express 100*** 0*** 0*** 0*** 0*** Banamex 24*** 26*** 42*** 6*** 1*** Banco Walmart 42*** 0*** 28*** 30*** 0*** Bancoppel 10*** 0*** 90*** 0*** 0*** Banorte 5*** 6*** 45*** 43*** 0*** BBVA Bancomer 98*** 0*** 2*** 0*** 0*** Compartamos 100*** 0*** 0*** 0*** 0*** HSBC 39*** 5*** 41*** 16*** 0*** Inbursa 0*** 0*** 84*** 16*** 0*** Invex 100*** 0*** 0*** 0*** 0*** Santander 12*** 0*** 30*** 58*** 0*** Scotiabank 0*** 0*** 80*** 20*** 0*** * For instance, electricity of telephone. Source: file B-R2, CNBV Information Portfolio, December III. The impact of correspondents This section presents a first evaluation of the impact of bank correspondents. First, it would be appropriate to clarify what we mean by impact evaluation. According to Gertler et al. (2011, p. 7): Impact evaluations are a specific type of assessments that attempt to answer questions on cause and effect. Unlike general evaluations, that may answer many types of questions, impact evaluations address the impact (or causal effect) of a program on a result of interest. It is only concerned with the program s impact: its direct effect on results. An impact evaluation analyzes the changes in the result that are directly attributable to the program. The focus on causality and attribution is the distinctive characteristic of impact evaluations and determines the methodologies to be used. To estimate the causal effect or impact of a program on the results, an eligible method must estimate the so-called counterfactual, i.e. what would have been the result for the participants of the program had they not participated in it. In practice, impact evaluation requires the evaluator to find a comparison group to estimate what would have happened with the participants without the program. Usually, impact evaluations are done for programs that have a concrete treatment. For instance, a concrete treatment would be giving pills to deworm children in order to increase their performance at school. For the impact evaluation of this program, the result (performance at school) of treated subjects (dewormed children) would 200

9 be compared to subjects that did not receive treatment (identical children who were not dewormed). The case of bank correspondents is very different from a program with a concrete treatment. The creation of the concept of correspondents was nothing like giving a pill. Rather, it was a regulatory change to allow banks to provide services through third parties, only if they decided to do so. It would be like having the government to issue a legal framework for pharmaceutical companies to sell deworming pills for children. The government did not provide the treatment nor did it subsidize it: it only provided a legal framework for private companies to offer it to people interested in it. Since it is an indirect intervention, it would be convenient to split the impact of the creation of the legal concept of bank correspondents into two: an intermediate impact and a final impact. The intermediate impact is the answer in terms of the opening of correspondents. We may say that there is an intermediate impact if banks use correspondents and if clients execute transactions through them. At first, banks may not find the concept of correspondents profitable, commercial establishments may not find becoming correspondents attractive, or consumers may not use correspondents. The final impact is the result of the opening of correspondents on the financial inclusion metrics used. Even having bank correspondents operating, it is thinkable that there may not be a huge effect on financial inclusion. For instance, it could be that both banks and clients only deviated transactions from branches and ATMs to correspondents. In that case, there would not be additional transactions or accounts. To measure intermediate impact it would suffice to compare situations before and after the creation of the legal concept of bank correspondent. The questions that await an answer are: were correspondents opened where there were no contact points before? and are there correspondents being used? In this sense, it is relatively simple to determine whether there has been an intermediate impact. The measurement of final impact is rather complicated. First, an inclusion metric must be defined. We use three: the balance of deposits and the number of debit and credit cards. The question is: has the opening of correspondents translated into a higher savings balance or into more credit and debit cards? To try to answer this question we use two techniques: multiple differences and instrumental variables. 201

10 Intermediate impact Chart 3 shows the 2,456 municipalities in Mexico grouped into five segments according to the number of bank branch offices they had in December 2009 before correspondents started operating. A total of 1,655 municipalities did not have bank branch offices. This is equal to two out of every three municipalities. Chart 3 also shows that municipalities without branch offices are small in terms of population, with an average of 11,200 inhabitants. Municipalities with one branch office have, in average, more than twice this number of inhabitants, and municipalities with 2 to 4 branch offices have more than four times this number. Category of municipalities according to branch office no. in Dec Chart 3. Intermediate impact: bank response No. of municip. Average population (thou.) 202 Average no. of branch offices Dec Apr Average no. of correspondents Dec Apr , a a or more Total 2, Population in Source: files 040-4A-R1 and A-R10, CNBV Information Portfolio, and INEGI. Chart 3 shows the growth of branch office numbers between 2009 and The average in each municipality increased from 4.4 to 4.8. This growth however was concentrated in municipalities that already had branch offices. In those that did not, the average changed from 0.0 to 0.1. The two last columns of chart 3 present the change in the average number of correspondents. In December 2009 there were no correspondents. In April 2012 there were 8.8 correspondents in average in each municipality. The opening of correspondents in absolute terms was concentrated in municipalities with more branch offices. Nonetheless, in relative terms, more were opened in municipalities with only a few branch offices or none. In municipalities that started off without branch offices the number of correspondents is eight times the number of branch offices, while in municipalities that started off with more branch offices the number of correspondents is less than two times

11 the number of branch offices. We could say the access increased. Correspondents allowed consumers to have more contact points with the banks. Chart 4 presents figures that show how correspondents have been used for the execution of transactions. Municipalities are organized into the same categories as in the previous chart. The first column shows the number of municipalities with correspondents. In these municipalities, we can count the number of average transactions in a given period and their average amount. This is what is reported in the two columns to the right for December The average number of transactions is lower in the municipalities with less branch offices. But let us not forget that those municipalities have less people and probably also less economic activity. The average amount of each transaction is however lower in the municipalities with more branch offices. Category of municipalities according to branch office no. in Dec Chart 4. Intermediate impact: transactions at correspondents Municipalities with correspondents in Apr Transactions at correspondents in Dec Average no. per municipality Average amount per transaction (pesos) , ,601 2 a ,734 2,339 5 a ,016 1, or more ,452 1,554 Total 1,365 3,651 2,218 Includes only municipalities with correspondents. Source: files A-R10 and B-R2, CNBV Information Portfolio. Charts 3 and 4 show that banks responded to the creation of the legal concept of correspondents by starting to build a network of thousands of them. By April 2012, 611 municipalities that did not have branch offices in 2009 now have at least one correspondent. And clients have not been indifferent. In December 2011 they executed an average of 836 transactions at correspondents in those same municipalities. In summary, the response of the market to the creation of the concept of correspondent was a higher number of contact points with the banking system than what are actually used. The intermediate impact has been positive. However, by not having a reference point, it is difficult to judge whether this impact has been large or small. 203

12 Final impact The main challenge to measure final impact is endogeneity. 15 Correspondents were not put into randomly selected places to operate. In general, we might think that banks decided to open them where it made more sense. If a bank had foreseen an increase in the demand of services in a specific place, it is possible that it decided to open more correspondents in that place. In that case, a simple comparison of the use of banking services in places where correspondents were opened and where not could lead to an erroneous inference. The same could be said of an ordinary least squares regression. A greater use in places where more correspondents were opened does not imply the causality of correspondents toward use. It is possible for causality to go in the opposite direction because the decision of opening correspondents may be in response to a greater demand. To avoid the issue of endogeneity when measuring the final impact of correspondents, we should ideally use an experimental design with a randomly controlled test. In practice, there is no such design. Correspondents have started operations without having a control group. There are however elements to try to measure the final impact with quasi-experimental methods using the data of the regulatory reports submitted by the banks to CNBV. In this section, we apply two methods to assess final impact: multiple differences and instrumental variables. 16 Each bank reports geographically disaggregated operational information to CNBV. For each place, banks report the number of correspondents and branch offices, and three variables that we use as metrics of the impact: the number of debit cards, the number of credit cards and the balance of deposits. According to internal CNBV analyses, the quality of information at the level of places is less than perfect. For this reason, we use the aggregate information of municipalities, which is the way CNBV publishes it as well. 17 In the period object of the analysis, banks did not report information in a consistent manner. Some reported information from the client s place of residence, and others reported 15 In economic terms, there is endogeneity between two variables when there is causality in both directions. 16 For an introduction to experimental and quasi-experimental methods, see Gertler et al. (2011). 17 The number of correspondents is available in file A-R10. The balance of deposits is in file 040-4A-R0. The number of debit and credit cards is found in file 040-4A-R1. All of the above files are in the CNBV information portfolio, on its web page. 204

13 information from the place where the branch office with the account is located. As a consequence of this inconsistency, the data we use present a measurement error. We do not know how large this error is, or how it may bias our results. The three inclusion metrics were picked based on the data available. At first, we would like to use transactions and their amount in each municipality (both in branch offices and in correspondents) as a metric of impact. However, CNBV does not have information on the number or amount of transactions at bank branch offices we only have that information on correspondents. In the period analyzed, correspondents were not opening accounts. This means that if there is an impact of correspondents on the number of debit and credit cards, it is probably indirect. For instance, if there is a lower transaction cost or more convenience in using these cards thank to correspondents, it is possible that more people will obtain this type of cards at the branch offices. The metrics of impact are imperfect: it is possible that there is a greater use not reflected by these metrics. For instance, if the presence of correspondents results in more deposits and withdrawals, balancing out each other without affecting the balance, the balance of deposits would not reflect such increase in their use. In the case of debit and credit cards, their number can remain unchanged even if their use increases at point-of-sale terminals. This is why results must be interpreted with precaution. Chart 5 presents the description of the sample used to measure final impact. The period we used ranges from the fourth quarter of 2008 through the fourth quarter of Correspondents started operations in the middle of that period, in the first quarter of The observation unit is the combination of bank and branch office. 19 The sample includes 245 municipalities with the presence of one bank only and 525 with the presence of two or more banks. Due to the differences between the size of municipalities and banks, we decided to use logarithms for impact metrics. In this way, the differences between periods can be interpreted 18 Since April 2011, banks report information by municipality differently, which makes this information not comparable with data prior to that date. Banks currently only report the balance of accounts with activity by the client in the last six months, or without activity but with a balance higher than 100 pesos. Before, they reported these figures for all accounts. 19 This means that if n banks have presence in municipality x, then municipality x appears n times in the sample. And if bank y has presence in m municipalities, that bank appears m times in the sample. 205

14 as growth rates. 20 Column A shows the mean, the standard deviation and the number of observations of the differences between the fourth quarter of 2008 and the fourth quarter of Column B presents the same statistics for the differences of the first half of the period: from the fourth quarter of 2008 through the fourth quarter of Column C present the statistics for the differences observed in the second half of the period: from the fourth quarter of 2009 through the fourth quarter of Finally, column D presents the double difference: the subtraction of the differences from the two halves of the period of study. Only observations for which double differences can be computed are presented. 21 Chart 5. Descriptive statistics of the sample: mean, standard deviation and number of observations Difference between fourth quarters Double Variable difference A B C D=C-B Balance of the deposits in logarithm (1.079) (0.895) (0.630) (1.109) [2,999] [2,999] [2,999] [2,999] Number of debit cards in logarithm (0.897) (0.784) (0.438) (0.899) [2,912] [2,912] [2,912] [2,912] Number of credit cards in logarithm (1.029) (0.925) (0.591) (1.163) [2,779] [2,779] [2,779] [2,779] Number of correspondents (7.469) (0.000) (7.469) (7.469) [3,346] [3,346] [3,346] [3,346] Number of branch offices (5.603) (5.543) (1.089) (5.695) [3,346] [3,346] [3,346] [3,346] Standard deviation in round brackets and number of observations in square brackets. The sample includes 245 municipalities with the presence of one bank only, and 525 with the presence of two or more banks. Column A of chart 5 shows a positive growth in the period of study for the three impact metrics and for the number of correspondents, and a negative growth for the number of branch offices. Columns B and C show that this growth was not uniform before and after the introduction of correspondents. The balance of deposits increased faster at the beginning of the period. The number of debit cards grew basically at the same rate. The number of credit cards was constant in the first half of 20 Algebraically, ln(x1)-ln(x0) = ln(x1/x0) (X1/X0)-1 21 For correspondents and branch offices, only observations for which double differences can be computed in at least one of the impact metrics are included. 206

15 the period and grew in the second. Before 2010 there were no correspondents operating, this is why no variations were observed in the first half of the period. The number of branch offices decreased in the first half of the period and grew in the second half. Column D summarizes the changes in the growth rates between the first and the second half of the period object of the study. The double difference is positive (negative) when the growth in the second half was larger (smaller) than in the first half, and zero if the growth was the same in both halves. We will now present the two strategies we used in the attempt to try to evaluate the final impact of bank correspondents: multiple differences and instrumental variables. We used the same database on both cases. However, the two strategies start from different approaches and assume different sources of identification. Hence it is possible that they do not provide the same results. They are conceptually independent exercises. Multiple differences. The intuition behind the multiple differences method is that banks behave differently in the same municipalities as a result of exogenous idiosyncratic factors, and that those differences may be exploited as a quasi-experiment. It is a fact that, despite banks facing the same conditions in the same municipality, ones open more correspondents than others. This different behavior is not due to the characteristics of the municipality but to an idiosyncratic factor of each bank. This idiosyncratic factor could be their capacity to make strategic agreements with commercial chains or their quickness in deploying their technology network. In principle, this behavior allows for estimating the impact of the opening of correspondents. We will now estimate two equations. The first may be written as follows: (1) where Δ represents the difference of the values between the fourth quarter of 2008 and the fourth quarter of 2010, ln( ) is the natural logarithm, y is the financial inclusion metric, s is the number of correspondents, c is the number of branch offices, and sub-indices m and b indicate the municipality and the bank, respectively. Coefficients λ and μ are fixed effects for each municipality and for each bank, respectively. Coefficient β is interpreted as the increase in the impact metric associated with an additional correspondent of bank b in 207

16 municipality m. Coefficient γ is interpreted as the growth associated with an additional branch office of bank b in municipality m. The second equation to be estimated is very similar to the first one, except for the fact that it considers double differences: (2) Δ 2 represents the difference between two changes: the one observed between the fourth quarter of 2008 and fourth quarter of 2009, and the one observed between the fourth quarter of 2009 and the fourth quarter of Double differences allow for contrasting growth rates before and after the start of correspondent operations. In equation (2), coefficients β and γ are interpreted, respectively, as the additional growth of the impact metric associated with an additional correspondent or additional branch office of bank b in municipality m. Fixed effects of λm y μb in equations (1) and (2) imply that impact metrics have a trend in each municipality that is common to all banks, and a trend for each bank that is common to all municipalities. In other words, the model allows for the balance of deposits and the number of debit and credit cards to have different trends in each municipality and at each bank. 22 Chart 6 presents the results of applying the multiple differences method. Panel A presents the results of estimating equation (1) and panel B presents the results of estimating equation (2). The chart has three columns and each presents results using a different impact metric. The first column shows the results for deposit balance, the second shows results using the number of debit cards, and the third contains the results using the number of credit cards. As we mentioned before, each observation corresponds to a bank in a municipality. Coefficients estimated for changes in the number of correspondents are statistically significant only for the balance of deposits in panel A. In all other cases, changes in the number of correspondents and in the rhythm of growth seem not to have an effect that is statistically different from zero on deposits or on the number of debit and credit cards. 22 For this reason it would be redundant to include controls for the characteristics of municipalities or banks. 208

17 Chart 6. Estimation of the impact of correspondents by multiple differences Dependent variable (natural logarithm) Explanatory variable Balance of deposits Debit cards Credit cards A. Differences Correspondents * ** Branch offices * *** Fixed effects for banks Yes Yes Yes Fixed effects for municipalities Yes Yes Yes R squared Municipalities Observations 3,070 2,931 2,856 B. Double differences Correspondents * Branch offices Fixed effects for banks Yes Yes Yes Fixed effects for municipalities Yes Yes Yes R squared Municipalities Observations 2,999 2,912 2,779 * p<.05; ** p<.01; *** p<.001. Robust standard errors grouped by municipality. Every observation corresponds to a municipality-bank combination. Differences between the fourth quarter of 2008 and the fourth quarter of Differences between the changes in two periods: the change from the fourth quarter of 2008 through the fourth quarter of 2009, and the change from the fourth quarter of 2009 through the fourth quarter of Instrumental variables. In theory, instrumental variables can solve the problem of endogeneity mentioned before. If they are appropriate, instrumental variables induce an exogenous variation into the explanatory variable and allow for estimating its causal effect on the dependent variable. Their application however requires restrictive assumptions that are not verifiable. 23 For the final impact of correspondents, the idea behind the use of instrumental variables is that bank branch offices have very high fixed costs. This prevents banks from opening branch offices in places with certain geographic and socio-economic characteristics. In a place with few inhabitants and low mean income it is probable that the volume of transactions does not make having a branch office a viable option. However, that same volume could make one or several correspondents cost-effective. If that was the case, the characteristics of the municipality could be used as instrumental variables for the opening of correspondents. The (non-verifiable) assumption we must make to apply this strategy is that the characteristics of municipalities that determine in which municipalities it is more probable to open more correspondents 23 For a detailed explanation, see Wooldridge (2002), chapter

18 does not directly determine the growth rates of inclusion metrics. However, it is easy to imagine situations in which this assumption is not fulfilled. 24 We estimated two equations by instrumental variables for each inclusion metric. The first equation is: (3) Variables are defined as in equation (1). In this case there is only subindex m because the information is aggregated at municipality level. Since the characteristics we used as instrumental variables are at municipal level they are characteristics of the municipality, we implicitly suppose that the variation in incentives to open correspondents is at that level as well. Therefore, equation (3) does not include fixed effects for bank or for municipality. 25 Coefficient β is interpreted as the growth of the impact metric associated with an additional correspondent in the municipality. Coefficient γ is the growth associated with an additional branch office in the municipality. The second equation is analogue to (3) but with double differences: (4) In this case, coefficients β and γ are interpreted as the additional growth of the impact metric associated with an additional correspondent or an additional branch office in the municipality, respectively. In both equations, each observation corresponds to a municipality. Instrumented variables are simple and double differences in correspondents and branch offices. Instrumental variables are: the natural logarithm of population, the natural logarithm of surface, altitude over the sea level, the percentage of the population that is indigenous, the number of post offices, the number of telegraph offices, the number of Diconsa stores and the average educational level of the heads of homes. These variables come from a database collected by 24 For instance, if there is a secular trend of poorer or more remote municipalities to catch up with the rest in terms of debit cards, regardless whether there are correspondents or not. 25 Fixed effects for banks cannot be included because the information is aggregated at municipal level. Fixed effect for municipalities cannot be included because each observation corresponds to a different municipality. 210

19 CNBV in collaboration with the Consortium on Financial Systems and Poverty. 26 Chart 7 presents the results of estimating the impact of correspondents using ordinary least squares. Panel A presents the results of estimating equation (3) and panel B presents the results of estimating equation (4). As mentioned before, each observation corresponds to a municipality this is why there are fewer observations in chart 6. Only in the case of credit card number with double differences is the coefficient in the change of the number of correspondents significant. It is probable that estimations in chart 7 are biased by endogeneity. The strategy of instrumental variables could help us correct that bias. Chart 7. Estimation of the impact of correspondents by ordinary least squares Dependent variable (natural logarithm) Explanatory variable Balance of deposits Debit cards Credit cards A. Differences Correspondents Branch offices * ** ** R squared Observations B. Double differences Correspondents * *** Branch offices * * R squared Observations * p<.05; ** p<.01; *** p<.001. Robust standard errors. Every observation corresponds to a municipality. Differences between the fourth quarter of 2008 and the fourth quarter of Differences between the changes in two periods: the change from the fourth quarter of 2008 through the fourth quarter of 2009, and the change from the fourth quarter of 2009 through the fourth quarter of Chart 8 presents the results of estimating the impact of correspondents using instrumental variables. The results of regressions show a positive and significant impact of correspondents only for the case of credit cards in panel A. In the other cases, the changes in the number of correspondents, and in the rhythm at which they grew do not seem to have an effect that is statistically different from zero on deposits or on the number of debit and credit cards. In summary, if we take the estimations from both approaches (multiple differences and instrumental variables), the evidence does not 26 The database is Mexico s Municipalities Savings and Intermediation Dataset, forms part of the project called A Regional Approach to Financial Savings and Intermediation, to be published soon. 211

20 support the hypothesis that correspondents have had a positive effect on impact metrics. We find a positive effect when considering simple differences in the balance of deposits (chart 6) and in the number of credit cards (chart 8). However, when using double differences, these results are not sustained. Chart 8. Estimation of the impact of correspondents by instrumental variables Dependent variable (natural logarithm) Explanatory variable Balance of deposits Debit cards Credit cards A. Differences Correspondents ** Branch offices R squared... Observations B. Double differences Correspondents * * Branch offices R squared... Observations * p<.05; ** p<.01; *** p<.001. Robust standard errors. Every observation corresponds to a municipality. Differences between the fourth quarter of 2008 and the fourth quarter of Differences between the changes in two periods: the change from the fourth quarter of 2008 through the fourth quarter of 2009, and the change from the fourth quarter of 2009 through the fourth quarter of The strategies employed are not exempt from criticism. To apply them, non-verifiable restrictive assumptions must be made. In addition to the criticism of strategies, there are at least three reasons why the results we are presenting herein are not conclusive. In the first place, the period of analysis is relatively short (one year after the first correspondents started operations) and their results should probably be interpreted as short-term. In the medium and long terms, results could be different. In the second place, correspondents have kept on offering more and more products. This makes the estimated impact be the result of a more modest and less attractive version of what has been observed later. In the third place, in the two methods we applied, the variation in each municipality has been omitted. Each municipality has several towns, some with branch offices, others without. The effects of correspondents in each town could be different depending on whether the town has branch offices or not. However, the analysis presented herein ignores the heterogeneity among towns in each municipality, possibly biasing the results. 212

21 IV. Conclusions Since it is a passive policy that does not involve spending taxpayers money or intervening directly in the market, the creation of the concept of correspondents has had very encouraging results. Banks have adopted it voluntarily and have multiplied their contact points with clients. Clients, in turn, have been using those correspondents. In terms of the impact evaluation of the opening of correspondents, we can say that we did not find an impact on the number of debit and credit cards, or on the balance of deposits. However, we do believe that the evaluation may be premature. Correspondents have been operating for a short time only. It is natural to think that it will take time and maybe more aggressive information campaigns for clients to become familiarized with them and with the possibilities they represent. We hope that the approach presented herein will serve as the base for future evaluations. The technical difficulties to estimate final impact should not opaque the evidence on intermediate impact. Correspondents are yet one more channel that the authorities have provided for the market. A proof that this channel is valuable for banks and clients is that banks have kept on opening correspondents and expanding the services offered through them. 213

22 References Beck, Thorsten, Asli Demirgüç-Kunt y Ross Levine Finance, inequality and the poor. Journal of Economic Growth 12(1): CNBV Reporte de Inclusión Financiera 1. Junio a. Reporte de Inclusión Financiera 2. Abril b. Reporte de Inclusión Financiera 3. Septiembre c. Modelos de Negocio para la Inclusión Financiera 3. Septiembre Encuesta Nacional de Inclusión Financiera. El desarrollo de una encuesta de demanda. La experiencia de México. Gertler, Paul J., Sebastian Martinez, Patrick Premand, Laura B. Rawlings y Christel M. J. Vermeersch La evaluación de impacto en la práctica. World Bank. Levine, Ross Financial Development and Economic Growth: Views and Agenda. Journal of Economic Literature 35(2): Wooldridge, Jeffrey M Econometric Analysis of Cross Section and Panel Data. The MIT Press. 214