Since the 1980s the Mexican economy has gone through a vast overhaul: it has been opened to foreign capital, State participation in economic affairs has been significantly diminished, protectionism has given way to trade liberalization, and, overall, the earlier import-substitution growth strategy has been displaced by an export-led one. Two pivotal moments in this vast shift were the implementation of NAFTA in 1994 and the last round of reforms which ended the state monopoly over the exploitation of oil and energy resources in 2012-2014.2 Naturally, each new round of reforms has been introduced with promises of high and sustained growth. Unfortunately, they have not materialized. Exports have increased but have not translated in growth for the economy as a whole; Moreno-Brid and Ros (2009), de Souza and Gómez-Ramírez (2018). Macroeconomic stability in the form of low inflation and reductions in the fiscal deficit may have been achieved, but even these achievements should be seen in the context of severe crises in 1982-1983, 1986, 1995, and 2008-2009. Most telling, World Bank and OECD data show a strikingly low 0.6% average per capita growth rate during 1982-2014.3
Scholars and policy makers consensually accept that one of the reasons for the disappointing growth results is limited access to credit. The standard view is that firms, especially small and medium-size firms, have found it very difficult to obtain financing for productive projects since the second half of the 1990s. Indeed, Kehoe and Ruhl (2010, p. 1011) state that “The most popular set of theories for Mexico’s stagnation focuses on its inefficient financial system and lack of contract enforcement.” And in fact, in 2013-2014 the executive and legislative powers revised more than 30 laws governing the financial system with the stated purpose of improving access to finance; their results are still to be seen. However, although in the world economies literature the negative impact of Credit Constraints on Investment is an established point, it has not been solidly verified for the more specific contemporary Mexico case. The lacuna is certainly worrisome given the central role all sides on the larger debate on the reasons for Mexico’s slow growth, summarized shortly, assign to the limited access to credit. This paper tries to start filling this lacuna. Using recent microeconomic, firm-level data which is reasonably nationally representative, it tests the hypothesis that Credit Constraints have reduced Investment in contemporary Mexico.
There are two broad variants of the explanation for Mexico’s slow growth. The dominant point of view, that is, the point of view held by policy makers and scholars with influence on policies, believes that the “heart of Mexico’s growth challenge” lies in the low total factor productivity; quote in Bolio et al. (2014, p. 6). The low productivity, in turn, is partially explained by the Credit Constraints faced in an inefficient financial system; Tinoco-Zermeño et al. (2014), Hanson (2010), and Kehoe and Ruhl (2010). Other scholars, however, have argued in favor of an alternative point of view. According to it, the total factor productivity slowdown is more the consequence than the cause of the slow growth, and the main reason behind the sluggishness lies in a poor Investment rate, which is in turn partially explained by the lack of bank finance for productive projects; other reasons include the low public Investment, the elimination of industrial policies, and the real exchange rate appreciation; Moreno-Brid et al. (2005), Moreno-Brid and Ros (2009) and (2010), Ros (2013) and (2015).4
Thus, whatever their larger disagreements, both sides agree Credit Constraints have played an important role in Mexican sluggishness. However, and rather surprisingly, the contemporary Mexico economic literature includes very few detailed studiesoffering empirical evidence underpinning the claim that financial constraints have reduced Investment and/or have reduced productivity (and/or have reduced growth directly for that matter). To the best of my knowledge, there are only two such contributions: Cotler and Woodruff (2008) and Love and Sánchez (2009). Cotler and Woodruff’s evidence is, however, localized and not nationally representative. Love and Sánchez’s analysis is similarly unrepresentative because it is circumscribed to the rural sector while Mexico is a highly urbanized country; furthermore, the time period examined by these scholars is not very recent. This contribution intends to start filling out this literature lacuna. Using microeconomic, establishment-level data which is reasonably nationally representative and covers more recent years, this paper tests the commonly stated claim that Credit Constraints affected firm’s Investment decisions. The data used are the World Bank Enterprise Surveys (hereafter WBES), carried out in Mexico for 2005 and 2009-2010 (hereafter 2010 as shorthand). The analysis is carried out separately for each period because the establishments covered in each wave of the survey are not the same.5 The main finding of this contribution is that, consistent with the general thrust of the literature, financial restrictions have indeed negatively affected Mexican firms’ capital accumulation decisions.
No one paper can conclusively prove that financial constraints have had a negative effect on capital accumulation in contemporary Mexico. In other words, this contribution has limitations. Although this paper is part of a large, firmly established world economies literature, it is a more novel contribution within the more specific literature on contemporary Mexico. It is the first test of a commonly held (and certainly intuitive) statement. This novelty makes the contribution important but it also implies it is isolated. Until further research on the impact of Credit Constraints on Investment in contemporary Mexico appears, thus, its findings have to be taken with basic caution.
The contribution is organized as it follows. Section 1 presents a selected literature review. Section 2 presents the data in more detail, discusses the endogeneity issue of Credit Constraints with respect to Investment and the strategiesused to address it. Section 3 presents the econometric analysis. The concluding comments section summarizes the paper and opens the discussion of policy implications.
1. Selected Literature Review
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. Data, Measures of Investment and Credit Constraints, and Endogeneity Issues
This section is divided into three subsections. The first describes in more detail the WBES data. The second describes the measures used for the key variables of our analysis. The third discusses the thorny issue of the highly likely endogeneity of Credit Constraints with respect to Investment decisions.
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).
& | 2005 | & | 2010 | |
---|---|---|---|---|
Region (state): | & | & | ||
Mexico city | & | 482 (32.57%) | & | 311 (21.01%) |
East Center (state of México, Puebla,Veracruz) | & | 478 (32.30%) | & | 417 (28.18%) |
West Center (Guanajuato, Jalisco) | & | 201 (13.58%) | & | 449 (30.34%) |
North (Nuevo León, Coahuila, Chihuahua) | & | 319 (21.55) | & | 303 (20.47%) |
Sector | & | & | ||
Manufacturing | & | 1122 (75.8%) | & | 1152 (77.84%) |
Services | & | 237 (16%) | & | 125 (8.45%) |
Other | & | 121 (8.18%) | & | 203 (13.72%) |
Size (measured with the number of full- time permanent workers) | & | & | ||
Less than 5 workers | & | 12 (0.81%) | & | 34 (2.30%) |
Between 5 and 19 workers | & | 736 (49.73%) | & | 478 (32.29%) |
Between 20 and 99 workers | & | 439 (29.66%) | & | 429 (28.98%) |
100 or more workers | & | 293 (19.8%) | & | 475 (32.09%) |
Source: Author’s calculations based on Enterprise Surveys (http://www.enterprisesurveys.org), The World Bank
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.
& | 2005 | & | 2010 | |
---|---|---|---|---|
Invested | & | 422 (28.5%) | & | 792 (53.5%) |
Not Invested | & | 1054 (71.2 %) | & | 686 (46.3%) |
Credit Constrained | & | 438 (29.5%) | & | 399 (26.9%) |
Not Credit Constrained | & | 1019 (68.8%) | & | 1021 (68.9%) |
Author’s calculations based on Enterprise Surveys (http://www.enterprisesurveys.org), The World Bank
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.”
3. Econometrics
Our first estimation is of the bivariate probit of the following equations:
in which
The “unexpected shocks” to the firm’s cash flows included as IVs in
Table 3 presents the key results of the estimations of the bivariate probit system (1)-(2). To highlight that addressing the endogeneity issue does make a difference in the analysis, the simple probit estimations are also shown. The bivariate probit column for 2005 presents the shortest specification with only Lostsales as IV, and the column for 2010 presents this year’s shortest with only Infpayments as IV. As I just mentioned, however, the results with all the other longer specifications were quantitatively very similar and exhibited the same very high levels of statistical significance. Now, as it is well known, in any probit model (univariate or bivariate) we are usually interested not only in the directly estimated coefficient of the binary independent variable of interest but also in its implied “marginal effect”. Intuitively, the marginal effect of X on Y is an estimation of how much more likely is the outcome of Y occurs if the outcome X occurs, as compared with the case in which the outcome of X didn’t occur; and all else equal, of course. In concrete, in this paper the central interest is in the marginal effect of Credit Constraints on Investment, which intuitively is the estimation of how much more likely is some firm does not invest if it is credit constrained, as compared with the case in which it were not credit constrained, all else equal. As it is well known, in order to estimate this marginal effect of the independent variable of primary interest it is necessary to posit values of all the other explanatory variables. The bivariate probit literature usually uses their means; Greene (1996), (1998) and (2012), Christofides et al. (1997). Thus, the marginal effects of Credit Constraints on Investment shown in Table 3 were calculated following this custom.
Dependent Variable: Investment | ||||||||
---|---|---|---|---|---|---|---|---|
& | 2005 | & | 2010 | |||||
& | Probit | & | Bivariate Probit | & | Probit | & | Bivariate Probit | |
Independent Variable: Credit Constraints | & |
|
& |
|
& |
|
& |
|
Marginal Effect of Credit Constraints on Investment | & |
|
& |
|
& |
|
& |
|
Observations | & | 840 | & | 840 | & | 957 | & | 884 |
Wald test chi statistic under the null of \rho=0 | & | & | 99.2632*** | & | & | 124.728*** |
Standard errors under parenthesis, *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively
Three pieces of information of Table 4 are important to be highlighted. First, and most importantly, in both 2005 and 2010 it is estimated that financial restrictions negatively affected Investment decisions at high levels of statistical significance. In 2005, it is estimated that, if credit constrained, the likelihood an establishment would invest is reduced by 47%, at the 1% level of statistical significance. In 2010, it is estimated that Credit Constraints reduce the likelihood the firm would invest by almost 48%, again at a 1% level of statistical significance. Second, addressing the endogeneity of financial restrictions with respect to Investment is important, as we had expected. The simple probit estimations show a much weaker marginal effect of Credit Constraints on Investment decisions. In 2005 and 2010, they tell us that Credit Constraints would have reduced the likelihood that a firm invested by only 8% and 9%, respectively. Third, the Wald-test of the bivariate probit model rejects the null of
As robustness check, the bivariate probit models were also estimated using the profits of the establishment (sales minus total costs) instead of its simple sales (see footnote 12) again with shorter and longer different specifications of equation (2); thesales specifications are preferred because using profits implied losing around 100 observations. The results of the estimations using profits (not shown) yield estimates that are quantitatively very similar and highly statistically significant too, as expected. It strengthens the validity or the results.
Now, we have argued our IV can be deemed as true unexpected shocks to the firm cash flows. Lewbel and coauthors have, however, raised the valid concern that it is still the case that maximum likelihood estimation usually yields inconsistent estimates if any relevant IV is omitted. It calls for carrying out the analysis using their proposed “special regressor” method. As these scholars say (2012, p. 823; italics are mine) “Special regressor estimators can use any valid set of instruments, given only the standard linear instrumental variables assumptions.” This standard assumption is that the IV affect the dependent variable only through the endogenous regressor, which in the context of our discussion is the case if our proposed instruments are in fact unexpected shocks to the firm’s cash flows. Therefore, the only other thing we need for properly carrying out the special regressor method estimation is to have a regressor V satisfying the conditions which make it “special.” These are that it is continuously distributed, it has large support, it is independent, and it is positively related with Investment. Fortunately for us, given the WBES information we could find at least one such regressor V: the variable Sales, which expresses the total amount of sales of the firm. That it is continuous and has large support can be seen with “eye-analysis,” and that it is positively related with Investment was verified with regression analysis.
Thus, we carried out the estimations with the special regressor method, having Sales as the special regressor, Lostsales as IV for Credit Constraints in 2005, and Infpayments and Obstcrime as IV in 2010. Kernel density estimation was used, and the standard errors of the marginal effects of Credit Constraints on Investment were computed by bootstrapping.14 Table 4 presents the estimates of the marginal effects estimated. Intuitively, they are estimates of how much more likely is some firm does not invest in the unfortunate case it is credit constrained, as compared with the case in which it were not credit constrained, all else equal.
Dependent Variable: Investment | ||||
---|---|---|---|---|
& | 2005 | & | 2010 | |
Marginal Effect of Credit Constraints on Investment | & |
|
& |
|
Observations | & | 840 | & | 882 |
Standard errors under parenthesis, *, **, and *** denote statistical significance at the 10%, 5%, and 1% level, respectively.
As it is shown in Table 5, with the special regressor method estimation it is again found that financial restrictions have negatively affected Mexican firms Investment decisions. Both in 2005 and 2010 the marginal effects of Credit Constraints on Investment obtained are negative and highly statistically significant. In particular, for year 2005 it is estimated that Credit Constraints reduce the likelihood a firm invests in 18%, with a 5% level of statistical significance. In turn, for year 2010 it is estimated that financial restrictions make less likely some firm invests by 21%, with a 10% level of statistical significance.
Thus, using special regressor methods we obtain further confirmation of the negative link between Credit Constraints and Investment decisions the bivariate probit had already suggested. In other words, the special regression method estimation in fact yields a “robustness check” to the “more standardly” used in the literature bivariate probit approach.
It is worth pointing out, however, that the special regressor method estimates are, from a quantitative point of view, smaller than the bivariate probit ones. Recalling, with the latter it was estimated that limited access to credit reduced the likelihoodof investing by 47% in 2005 and by almost 48% in 2010. In contrast, with the special regressor method, it is estimated that financial restrictions reduce such likelihood by 18% in 2005 and by 21% in 2010. These 18% and 21% effects are still certainly large, to be clear, but they are not as impressive as the 47% and 48% obtained earlier. This quantitative difference suggests the main strength of the findings of this paper is qualitative more than quantitative. In other words, it suggests that the strong finding of this contribution is that financial restrictions have been deterring Investment decisions, but we cannot establish by how much with that much confidence. It obviously opens a topic for future research.
Concluding comments
Mexico has gone through a transformative series of economic policy changes in the last three decades. The justification for each new policy was the pursuit of economic growth. The high expectations, however, have not been fulfilled. There are many factorsthat may have contributed to the slow- to no-growth outcome but among them limited access to credit has been pointed out as a very important one by scholars and policy-makers alike. Actually, there are two broad variants of the explanation for Mexico’s slow growth. One, which is dominant, highlights the low total factor productivity of the country; Bolio et al. (2014), Tinoco-Zermeño et al. (2014), Hanson (2010), Kehoe and Ruhl (2010). The other, which is alternative, argues such low total factor productivity is more the consequence than the cause of the slow growth, and states the main reason behind Mexico’s slow growth lies in a poor Investment rate; Moreno-Brid et al. (2005), Moreno-Brid and Ros (2009) and (2010), Ros (2013) and (2015). Whatever their larger disagreements, however, both sides of the debate agree Credit Constraints have played an important role: in the low productivity literature it is in turn partially explained by the financial restrictions faced by Mexican firms, and in the poor Investment rate one it is also partially explained by limited access to credit.
In this paper we focused on the role of Credit Constraints in limiting Investment. It turns out, however, that the negative influence of financial restrictions on firm’s Investment decisions has not been firmly empirically established in the scholarly literature on Mexico. This paper presents the first attempt to empirically test this consensus (and rather intuitive) explanation of the low growth using firm-level, good quality data which is nationally representative and coversrecent years. Two sorts of estimation analyses were carried out. First, the commonly found in the literature bivariate probit estimation. Second, the “special regressor” estimation method proposed by Lewbel and coauthors. It is less well-known but it addresses the problem of possible omitted IVs. In each analysis it was estimated that in fact Credit Constraints have been reducing physical capital accumulation of Mexican firms. Thus, I find it fair to say that the econometric analysis of this paper offers good evidence supporting the general thrust of scholars and policy-makers.
Policy implications may seem straightforward. It seems clear that improving access to finance for productive project is necessary in order to finally put Mexico on the path of high and sustained growth. In fact, in 2013-2014 reforms in more than 30 laws governing the credit system were introduced with the intended purpose of improving access to credit. Its results are still to be seen.
However, a proper discussion of the policy implications to be derived from the findings of this paper is very complex and calls for future research. The following three general comments highlight the care needed to translate policy implications gleaned from this research into genuinely helpful policy.
First, an insight offered in Moreno-Brid (2013) is worth keeping in mind. This scholar rightly points out that the simple conclusion “more should be lent” could be misleading. If carried out indiscriminately, more lending could be of little use. Lending must be carefully targeted and carefully administered. Furthermore, as shown in Skott and Gómez-Ramírez (2018), in the context of an economy with a significant informal sector like Mexico, easing access to finance for formal firms although would verylikely increase their capital accumulation does not guarantee formal employment would grow too; supplementary policies may be needed to achieve the latter.
Second, the role of the State in improving the access to finance is a disputed topic. The dominant view is that the role of the State is to promote more competition -for example, in the highly concentrated Mexican banking system- and to improve the ruleof law; indeed, the 2012-2014 reforms had this perspective in mind. However, other scholars have favored more state intervention, especially through national development banks; Levy (2007); Moreno-Brid (2013). In more general terms, we need to address not only the question of how much credit will be available and to which borrowers, but also on what terms credit will be available and from which lenders.
Third, easing Mexico’s credit crunch will require working through multiple channels at once, not only those directly related to the financial system. Undergraduate level textbooks on economic development rightly stress that, in developing economies, solving one pressing issue is hard to be achieved if other issues are not addressed too; Todaro and Smith (2015). Indeed, the need to address several problems at once in order to act effectively on any one of them is one of the most important obstacles to getting out of “underdevelopment traps.” In this respect, it should not be forgotten that appropriate measures to address poverty and inequality in contemporary Mexico should be an integral part of any sound policy package intending to ease thecountry’s financial crunch. At the same time, easing Credit Constraints may make an important contribution to easing Mexico’s distressing poverty and inequality burdens. In fact, influential international literature has argued that the presence of inefficient financial markets together with large inequality/poverty burdens is probably a very undesirable mix; Aghion et al. (1999), Benabou (1996), Bardhan et al. (2000), Galor and Zeira (1993). There is also more specific literature on contemporary Mexico which suggests that the credit crunch there has been positively related with its poverty burden; Niño-Zarazua (2013), Carreón et al. (2007), Bruhn and Love (2014).