1.1 World economies

Much attention is devoted to the general topic of financial constraints in several different countries; what I call the “world economies”. To begin with, many contributions have documented that, in developing economies in particular, the distribution of firms usually exhibits a “missing-middle”: there are many small and medium-size firms (SMEs) alongside a few large ones, and the former are usually credit constrained; Jaramillo et al. (1996), Hericourt and Poncet (2009), Khwaja and Mian (2008), Bigsten et al. (2003), Harrison and McMillan (2003). The key concern of this missing-middle distribution finding is that, as Beck and Demirguc-Kunt write (2006, p. 2932), many scholars take as a “premise” that small- to middle-size firms “are the engine of economic development.”

In the world economies literature many papers have examined limited access to credit at the firm level. While some of them have inquired about their impact on key economic variables, like employment and Investment, others have focused more on establishing the existence of such Credit Constraints and/or in examining their underlying reasons. Banerjee and Duflo (2014), for example, offer empirical evidence suggesting that many relatively large Indian firms have been severely credit constrained. Hericourt and Poncet (2009) finds that Chinese private firms have been credit constrained while public ones have not. Khwaja and Mian (2008) examines if Pakistani firms’ financing saw a reduction after the 1998 nuclear tests, which in turn negatively affected the supply of loans, finding that small firms were unable to smooth the loans crunch through the use of alternative sources of finance. Harrison and McMillan (2003) finds that Ivory Coast enterprises faced more acute Credit Constraints than foreign ones. Bigsten et al. (2003) delves into a wider range of African countries, namely, Cameroon, Ghana, Kenya, Zimbabwe, Burundi, and the Ivory Coast, finding that the larger and/or the more profitable an enterprise is the less likely is it is credit constrained. Jaramillo et al. (1996) finds that in Ecuador smaller/younger firms faced more acute Credit Constraints than larger/older ones.

In the world economies literature there are also contributions which, while maintaining the firm-level approach, investigate the impact of credit restrictions on key economic variables, like Investment and employment, Fazzari et al. (1988) is a seminal paper, which sparked large influence not only because of its findings but also because of its empirical methodology (to be discussed more in section 2). Its findings are that United States (U.S.) manufacturing firms were credit constrainedand that those constraints negatively impacted Investment. Hu and Schiantarelly (1998) also finds that U.S. manufacturing firms were credit constrained. Chodorow-Reich (2014) examines the role of U.S. firms’ access to credit on their employment decisions, finding that the credit crunch which happened after the 2008 crisis reduced employment but only among SMEs. With respect to other than U.S. enterprises, Galindo et al. (2007) examine the effect of financial liberalization on the allocation of Investment in several developing economies which underwent financial liberalization measures over the last several years: Argentina, Brazil, Chile, India, Indonesia, Korea, Malaysia, Mexico, Pakistan, Philippines, Taiwan, and Thailand. Itfinds that “financial liberalization in the majority of cases leads to an improvement in resource allocation, although there are interesting exceptions” (p. 564).

To finish this “world economies” selected literature review it is worth mentioning two very influential papers which, unlike the contributions so far mentioned, address the role of Credit Constraints on growth with a “macro”, cross country regressions perspective, and not with a “micro” or firm-level one. First, King and Levine (1993), which finds that financial development is “strongly and robustly correlated with” growth (p. 734). Second, Aghion et al. (2005), which finds “strong evidence … for the general proposition that whether or not a country converges to the frontier growth rate depends on its level of financial development” (p. 189). Interestingly, the estimates of these papers also suggest a “step” positive relationship between financial development and growth.

1.2 Mexico’s literature

When passing to the literature on Credit Constraints in contemporary Mexico, we find no dearth of papers either. What we found, however, is that instead of examining their possible impact on key economic variables, in this literature there are many more contributions trying to explain the reasons of the credit crunch. The central question is why lenders, especially the banks, lend so little for productive purposes, especially since the second half of the 1990s; even while credit for consumption and housing has in fact increased since then. Bergoeing et al. (2002) compares Chile’s and Mexico’s bankruptcy laws. Haber (2005), and Haber et al. (2008) focuses on the banks’ privatization process. Mántey de Anguiano (2007) points out that banks make good enough profits by buying government bonds and engaging in the derivatives market. Haber (2009) finds evidence suggesting that banks lend too little because of the oligopolistic nature of their market and because of the difficulties the lenders havein seizing assets from borrowers if they default. Haber and Mussachio (2012) presents a detailed study of the impact of foreign banks’ entry into Mexico. Tinoco-Zermeño et al. (2014) examines the effect of inflation on the low bank credit rates. Chavarín (2015) finds no evidence in favor of the thesis that the low lending comes from borrowers not paying back on time.

Contributions focusing on the distributional dimension of the Mexican credit crunch could be found too. These papers offer evidence of and/or explanations for why SMEs are more credit constrained than larger ones; Garrido and Prior (2007), Lecuona (2009),Clavellina (2013). Other contributions have focused on the impact of Credit Constraints on poverty and on informality; Niño-Zarazua (2013), Carreón et al. (2007). Interestingly, in this literature Bruhn and Love (2014) not only examines the impact of easing Credit Constraints on poverty but also delves into the gender dimension of the issue. It is possible to even find contributions on the possible effects of limited access to credit on migration; Aroca and Maloney (2005), Angelucci (2015).

Some few contributions have examined the possible effects of the credit crunch on other economic variables. McKenzie and Woodruff (2008) investigates the effects of easing capital constraints on firms’ profits, finding a positive and significant effect.Inspired in the Mexican economy, Skott and Gómez-Ramírez (2018) examines from a theoretical perspective the implications for employment and distribution of a relaxation of the financial constraints in the context of a dual economy (with a large informal sector with hidden underemployment). It finds that, in the absence of supplementary policies, the relaxation of the financial restrictions may reduce the share of the formal sector, increase inequality and underemployment.

Now, given it would sustain the view customarily held among policy makers, one would expect a solid body of empirical literature confirming that limited access to credit has negatively affected total factor productivity. Nevertheless, such a body of literature could not be found; Villalpando (2014) is an exception, as it does try to offer evidence supporting the claim that there has been a negative effect of financial restrictions on productivity. Perhaps the reason of this worrying gap found in the literature is related with the difficulty of obtaining good quality data on Investment, especially at the firm level, which could reasonably be considered nationally representative too. In any case it is still rather disappointing to find that policy makers do not seem to have solid empirical evidence underpinning their standard point of view.

Among scholars adhering to the alternative point of view on the “Mexican morass”, empirical evidence confirming that Credit Constraints have deterred capital accumulation are also difficult to find. Again, data limitations may explain this lacuna but it is still worrying. Scattered papers offering evidence in favor of this negative relationship but covering time periods prior to the second half of the 1990s are found. Gelos and Werner (2002) examines manufacturing firms from 1984 to 1994. Sánchez (2001) also investigates manufacturing establishments for 1984-1999. Castañeda (2003) goes over the time period before and after the 1995 crisis. Ramírez (1994) and Warman and Thirwall (1994) both address the role of financial constraints on Investment only incidentally, that is, including financial restrictions only as control variables of their regressions; furthermore, they cover a period before the mid-1990s, and the Credit Constraints for productive projects in Mexico have become harsher after it. In any case it may be worth mentioning that both papers obtained negative and significant coefficients on the financial restrictions variable.

There are only a couple of contributions which cover more recent years and in which the effect of financial constraints on Investment is of primary interest. The first is Cotler and Woodruff (2008). It examines the impact of micro-lending on physical capital Investment, among several “measures of the firm’s performance.” Overall, the paper finding is a positive and significant relationship between obtaining a loan and Investment in fixed assets. The data for this contribution is quasi-experimental, that is, it consists of both a “treated” and a “control” group of enterprises similar in many respects. The use of this type of quasi-experimental data has clear strengths. They come at a cost, tough, which is, quoting Beck and Demirgunt-Kunt (2008 p. 390) “whether the results found in one specific geographic or socioeconomic environment can easily be applied to a different environment.” Cotler and Woodruff investigation is circumscribed to small retailers located in two neighborhoods of the suburban area of Mexico City. Mexico is a very large, highly uneven, and complex country. Results obtained from small retailers of two neighborhoods of Mexico City are hardly nationally representative. Cotler and Woodruff are aware of the scope of their findings (p. 848).

The second is Love and Sánchez (2009). These scholars find evidence suggesting that easing Credit Constraints would increase both the number of agents investing and their levels of Investment. Their findings, however, are not about a very recent time period, as they used surveys from 1999 and 2001. Most importantly, this contribution is circumscribed to the rural Mexican sector. Contemporary Mexico is, however, a highly urbanized country; at least since the 1980s some 75% of its inhabitants have been living in urban areas.

It turns out, thus, that this paper presents the first test of the effect of Credit Constraints on Investment using good quality firm-level data that are both recent and reasonably nationally representative. Given the importance of this purported link in scholarly literature and policy discussions, this empirical test makes an important contribution.

2.1 Data: The World Bank Enterprise Surveys for Mexico

The World Bank Enterprise Surveys offer a rich source of good quality information on Mexican establishments.⁶ Each survey interviewed 1480 establishments, gathering information on fifteen broad topics. First, the general characteristics of the establishments, including ownership, legal status, and size. Second, the infrastructure and services they have (or do not have) access to, like electricity, water, internet, etc. Third, the quantity and nature of their sales, including national/international and main/secondary products/services. Fourth, the quantity and nature of their demand for inputs. Fifth, the degree of competition they face. Sixth, the innovation measures they have (or not) introduced. Seventh, their capacity utilization. Eighth, their land tenure. Ninth, their interest in being and/or pressures to be environmentally responsible. Tenth, their relationship with the courts. Eleventh, the ways crime has affected their operations. Twelfth, their relationship with the government, including if they have bribed officials. Thirteenth, their financial situation. Fourteenth, their labor force: number, permanent or temporary, and their education levels. And, fifteenth, the self-perceived main constraints on their operation. It should be said that the World Bank WBES team has been doing a high quality job in collecting these surveys worldwide, and Mexico is not the exception.

An important strength of this data set, and therefore of the analysis in this paper, is that it is reasonable to consider it nationally representative. Table 1 presents key indicators with evidence supporting this assertion. It shows the number of establishments which were surveyed according to their geographic location, sector, and size; the latter is measured with the number of full time permanent workers, following what the Quarterly Survey on Enterprises Financing Sources carried out by the Mexican Central Bank does.⁷ The enterprises surveyed are distributed all across the country (with the exception of the Southeast, which we’ll discuss shortly), they belong to different economic sectors (many concentrate in manufactures, but this sector is itself subdivided into more than a dozen different subsectors), and they are of several different sizes (in terms of how many full-time permanent workers they hire).

It should be mentioned, however, that there are two important features of the Mexican economy which are missed in the WBES. First, with very few exceptions these surveys gathered information on formally registered establishments with more than 5 workers while it is well known that a large proportion of businesses in Mexico are informal; INEGI (2004) and (2011), Bolio et al. (2014). However, the absence of informal establishments in the WBES data set could actually strengthen the thesis that CreditConstraints have negatively affected firms’ real Investment decisions. If it is shown that Investment by large, formal firms has been negatively affected by Credit Constraints then it does not seem unreasonable at all to also believe that the negative effect on small, informal firms has actually been stronger, as the latter almost certainly have less access to credit than the former; indeed, the lack of access to credit faced by small informal firms is one of the reasons for which they remain small and informal. Second, the WBES did not interview firms located in the Southeast of the country, which is also well known to be poorer and more backwards than the rest of the regions; Ros (2015). But for analogous reasons to the onesrelated with missing data on small informal firms, missing information on Mexico’s Southeast could actually strengthen the findings of this paper. In short, the selection bias in the data set - the overweighting of large formal sector firms, not located in the most backwards region of the country- very likely reinforces the conclusion that Credit Constraints have negatively affected firms’ real Investment decisions in contemporary Mexico.

2.2 Investment and Credit Constraints

The measure of Investment used in this paper is a binary variable indicating whether in the last year the firm bought fixed assets such as machinery, vehicles, equipment, land or buildings (Investment).⁸ The independent variable of primary interest, our measure of Credit Constraints, is based on the categorical measure of financial restrictions proposed by Kuntchev et al. (2013) (Credit Constraints). It is appropriate for our analysis for two reasons. First, and very importantly, because it is built upon precisely the information obtained by the World Bank Enterprise Surveys. Second, because, quoting Kuntchev et al. (2013, p.3), it provides an innovative way of measuring credit-constrained firms based both ontheir usage of and ability to obtain new credit. This is an important contribution of the literature since most papers … either look only at usage of credit, as opposed to access, or focus on self-reported obstacles based on perceptions instead of objective on the experience of the firm.

The conceptual essence of this measure of financial restrictions is that an establishment is credit constrained if it wanted to have external finance but for some reason it could not get it (in which case Credit Constraints equals 1), and it is not credit constrained if it either did not want to obtain external funding or did obtain it (in which case Credit Constraints equals 0).⁹

Two features of our variable Credit Constraints are important to be highlighted. The first is it directly allows us to know whether a firm was credit constrained or not. This is important because, in turn, it allows us to depart from the widely followed but challenged methodology Fazzari et al. (1988) sparked. In it, the existence of Credit Constraints is indirectly discovered, so to speak, in the following way. The whole sample of firms is divided, using a priori criterions (like retained earnings), into subsamples according to the degree of financial restrictions each group is supposed to face. Then regressions of Investments on cash flows and other control variables are carried out for each such subsample. Reasoning that in aworld with perfect financial markets (a-lá Modigliani and Miller (1958)) their liquidity levels should have no effect on firms’ Investment decisions, if the coefficients of the cash flows variables are significant and increasing with the degree of financial exclusion, then it could be established that there are Credit Constraints. This methodology has in essence been followed, with extensions and minor modifications, by many in the world economies literature; Jaramillo et al. (1996),Hu and Schiantarelly (1998), Harrison and McMillan (2003), Almeida and Campello (2007), Hericourt and Poncet (2009); and for the Mexican case by Gelos and Werner (2002), Sánchez (2001), Castañeda (2003). However, it has been criticized precisely for a priori partitioning the sample into groups of firms according to their degree of Credit Constraints; Kaplan and Zingales (1997). It is beyond the scope of this paper to evaluate this debate but it suffices to mention that here we don’t need to deal with it because our Credit Constraints variable directly captures if a firm was financially restricted or not; in this respect, this paper is similar to Bigsten et al. (2003).

The second is that Credit Constraints allows us to confidently say our findings are about “real” effects of Credit Constraints. In other words, it allows us to establish that firms financially excluded for some reason were not able to find alternative sources of financing. Indeed, the literature has often stressed that, in order to establish that some form of credit crunch transmits into firms’ “real” outcomes, it should be established that the firm could not find alternative financing sources; Kashyap and Stein (2000), Khwaja and Mian (2008), Chodorow-Reich (2014). In this paper, we can be sure of it because, when asking if the establishment received credit or not, the WBES referred to any financing source: private commercial banks, state-owned banks and/or government agencies, non-bank financial institutions, credit from suppliers and advances from customers, and “other” (“moneylenders, friends, relatives, bonds, etc.”).

Table 2 presents descriptive statistics of Investment and Credit Constraints. In both rounds of the surveys, around 30% of establishments were credit constrained. That this number is not as large as other surveys show (like the already mentioned above Quarterly Survey on Enterprises Financing Sources carried out by Mexico’s Central Bank) is not surprising, given that the firms surveyed in the WBES are not informal, small ones, and actually many of them made handsome profits. With respect to capital accumulation decisions, in the 2005 survey 28.5% of establishments reported that they invested. In 2010 53.5% accumulated physical capital. This increase is not surprising since larger firms were surveyed in 2010 than in 2005 (see Table 1).More detailed examination (not shown) of Investment behavior by enterprise size showed that larger ones invested more than smaller ones in both rounds of the survey, not surprisingly.

2.3 Endogeneity Issues

It is highly likely that Investment decisions affect the credit situation of the establishment, that is, it is highly likely that Credit Constraints are not exogenous but are endogenous with respect to physical capital accumulation decisions. Literature examining the effects of financial restrictions on innovation does assume endogeneity of the financial situation with respect to the innovation decisions; Savignac (2008), Gorodnichenko and Schnitzer (2013). Since innovation is actually one form of Investment, it is straightforward to believe the issue extends for physical capital Investment too. Furthermore, the intuitive argument explaining this endogeneity, which is the following, is compelling. An establishment decides to invest. For financing such Investment, it could use internal or external funds. Due to the difficulties of enforcing loan contracts and due to asymmetric information issues, using external funds would very likely be more expensive than using internal ones. Thus, it is likely the firm would rely on internal funds to the extent possible. However, using up internal funds on Investment increases the likelihood of facing Credit Constraints because the firm would then lack collateral.

Both because this argument is intuitively compelling and because the literature usually accepts there is endogeneity, in this paper we, too, accept that Credit Constraints are endogenous with respect to Investment. Accepting it, however,leaves us with the need to address an econometric issue certainly difficult to address; indeed, it is fair to say that the core concern of a large amount of modern econometrics is about how to deal with endogeneity.

Addressing this issue when the dependent variable is binary has its particular difficulties, and in fact our dependent variable of primary interest (Investment) is binary. Lewbel, Dong and Yang (2012) discuss four types of convenient estimators which could be used in this context; first, two-stage linear probability models (LPMs); second, maximum likelihood estimators, like a bivariate probit system of equations; third, control function estimators, like Stata’s ivprobit, which according to the same Lewbel and coauthors (p. 818) “despite its name… is actually a control function estimator, not an instrumental variables estimator”; and fourth, the “special regressor” method proposed by the same scholars in different contributions; Lewbel (2000), Lewbel et al. (2012), Dong and Lewbel (2015). In this paper, the analysis is carried out, first, with bivariate probit estimation, and, second, with the special regressor method. But it is worth explaining in some more detail the reason for choosing these two.

Using two-stage LPM was discarded for the well-known problems of LPMs in general; to begin with, it may not yield estimates which actually fall in the range β ∈ [0,1].¹⁰ As a potential solution to this problem, ivprobit is commonly found in the literature. In fact, however, carrying out an ivprobit when the independent variable of key interest is not continuous is mistaken. That an ivprobit would not generally yield consistent estimators when the endogenous regressor is binary, discrete, or censored is a point established by Lewbel and coauthors, and indeed the Stata reference manual (version 14) clearly states that ivprobit is “not appropriate” when the endogenous regressor is binary.¹¹ Given that our variable Credit Constraints is binary, this estimation method was therefore discarded too. Now, highly regarded econometricians have established that estimating a bivariate probit in the context of both binary dependent and independent variables with the latter being endogenous is an appropriate strategy; Greene (1996) and (2012). Furthermore, several literature contributions facing this binary-variables-with-endogeneous-regressor problem have actually estimated a bivariate probit; Evans and Schwab (1995); Greene (1998), Christofides et al. (1997), Ploetscher and Rotmman (2002), Savignac (2008). And indeed Lewbel and coauthors do not simply discredit the use of bivariate probit but instead highlight the other requirements needed for its proper application. As one would expect, however, estimating a bivariate probit is not entirely a free lunch. The main difficulty is that, as Lewbel and colleagues highlight, to yield consistent estimators it requires that the equation for the endogenous regressor is not just any equation but the right one, one which does not omit any relevant instruments (IV).

This mentioning of instruments leads us to the very important question of what variables are appropriate IV for Credit Constraints. We know they have to affect the firm’s financial restrictions but without having an unmediated effect on its xtitInvestment decisions. In other words, they have to influence firms’ Investment decisions only through the channel of affecting its credit situation. It is certainly not that easy to find such variables. Fortunately, Gorodnichenko and Schnitzer’s (2013) contribution was very useful for us here. In their study of the effects of Credit Constraints on innovation decisions, these scholars propose to use, as IV for the firm’s Credit Constraints, “unexpected shocks” to its cash flows; like goods stolen or lost while in transit to their final destination, among others. The key reason for which these shocks are considered as proper IV is that, if they are truly unexpected, then they do affect the firm’s financial situation but cannot directly affect its Investment decisions. Fortunately for us, the WBES gathered information on variables that can be deemed to be unexpected shocks to the firm’s cash flows. These are the IV used in this paper, described in moredetail in section 3.

The equation for Credit Constraints in the bivariate probit estimation, thus, include these unexpected shocks as IV. As we just mentioned, however, Lewbel and coauthors have established that, for it to yield consistent estimators, no relevant instrument should be missed. For these and other reasons, these scholars propose the use of what they call the “special regressor method.” It only requires the existence of the special regressor V with certain properties to yield consistent estimates even if not all the relevant IV for the endogenous regressor are included. In other words, with this method all what we need is any proper instrument and the special regressor V. In turn, the properties of V which make it special are that it is continuously distributed, it has a large support, it is independent, and it is positively related with the dependent variable. Given that among our independent variables we do have some which satisfy these properties, it was rather natural to carry out the analysis with this special method too, and therefore we did it. Doing so is, quoting Lewbel et al. (2012, pp. 828) “useful, at least in providing robustness checks of results against alternative, more standard estimators.”