Financial liberalization, credit constraints, and collateral: investment in the Mexican manufacturing sector

Journal of Development Economics Vol. 67 Ž2002. 1–27 www.elsevier.comrlocatereconbase

Financial liberalization, credit constraints, and collateral: investment in the Mexican manufacturing sector R. Gaston Gelos a,) , Alejandro M. Werner b b

a International Monetary Fund, 700 19th St., N.W., Washington, DC 20431, USA Banco de Mexico, Direccion 5 de Mayo No. 18, Col Centro, ´ ´ de Estudios Economicos, ´ Mexico D.F. 06059, Mexico ´

Received 1 March 1999; accepted 1 May 2001

Abstract We examine the impact of financial liberalization on fixed investment in Mexico using establishment-level data from the manufacturing sector. In addition to analyzing changes in cash-flow sensitivities, an innovative approach explores the role of real estate as collateral and addresses a potential censoring problem. The results suggest that financial constraints were eased for the smallest firms, but not for larger ones. However, the importance of possessing real estate increased, given banks’ reliance on collateral in their lending. The results also provide microeconomic evidence consistent with the role attributed to Afinancial acceleratorB mechanisms during lending booms and during post-crisis recessions. q 2002 Elsevier Science B.V. All rights reserved. JEL classification: E44; E22; G14; O16 Keywords: Investment; Financial constraints; Collateral; Real estate; Fixed-effects Tobit

1. Introduction Recent financial crises have spurred a renewed interest in the effects of financial deregulation in emerging markets. While some theoretical research and empirical analysis at the macroeconomic level has been undertaken in this area, )

Corresponding author. Tel.: q1-202-623-9427; fax: q1-202-623-4352. E-mail address: [email protected] ŽR.G. Gelos..

0304-3878r02r$ - see front matter q 2002 Elsevier Science B.V. All rights reserved. PII: S 0 3 0 4 - 3 8 7 8 Ž 0 1 . 0 0 1 7 5 - 4

2

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

empirical work using microeconomic data is still scarce.1 This paper attempts to make a contribution to this literature by examining the Mexican case. Prior to 1989, Mexico’s financial system was highly regulated. In particular, between 1982 and 1988, the government financed its deficits through increased reserve requirements on the domestic banking sector, and bank credit to the private sector plummeted. This situation changed in late 1988, when a comprehensive liberalization of the financial sector was initiated. Government deficits, which had been reduced significantly, were now financed mainly through domestic short-term debt, and the volume of bank loans extended to the private sector increased dramatically. This paper examines how these developments affected fixed investment using a unique, and largely novel plant-level data set covering nearly 80% of value added in the manufacturing sector in the period 1984–1994.2 The questions that we seek to answer are: how was investment by manufacturing firms affected by financial reform? To what extent were firms financially constrained before and after liberalization? Which firm types and sectors benefited most from the increased availability of credit after 1989? 3 Is there evidence for the importance often attributed to the role of real estate as collateral during lending booms after financial liberalization? The answers to these questions are important from a policy perspective. Not only is it crucial to understand the precise way in which financial liberalization in a developing economy affects bank lending behavior and firms’ access to external finance: 4 learning more about the significance of these credit constraints is also relevant for the understanding of the dynamics of boom-and-bust cycles. For example, if firms are financially constrained, being able to obtain credit only against collateral, shocks to the net worth of firms may be propagated through Afinancial acceleratorB mechanisms as described in Bernanke et al. Ž1996.. Such propagation mechanisms may be particularly relevant in the context of lending booms preceding and the severe macroeconomic downturns often following financial crises. We first follow the standard methodology adopted in empirical work on the importance of liquidity constraints for firm-level investment. We examine the effect of the availability of internal funds on capital expenditures and its change

1 Among the exceptions are Atiyas Ž1992., Harris et al. Ž1994., and Jaramillo et al. Ž1997.. For a collection of studies analyzing the impact of financial reforms, see Caprio et al. Ž1996.. 2 While other authors have used the 1984–1990 sample, to our knowledge, Gaston Gelos was the first person to have had access to the whole database. 3 The only other microdata-based study analyzing related issues in Mexico of which the authors are aware, Babatz and Conesa Ž1997., does not cover the years prior to 1988 Žthe period of the most marked financial repression. and only inspects the behavior of 71 stock-listed companies. 4 AExternalB financing as used here and in the following refers to funds external to the firm, not to access to foreign capital markets.

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

3

over time across different types of firms that are likely to differ in their access to external finance. Then, going one step further, we address two commonly neglected issues. Motivated by the fact that zeroes represent a large fraction of the investment observations, we deal with the censoring problem that may arise in the presence of credit rationing. We also explicitly investigate the importance of collateral. Building on a simple model that stresses the role of minimum project sizes and collateral, we explore the function played by real estate as a collateral before and after 1989. To our knowledge, this is the first study providing microeconomic evidence for the frequently mentioned role of real estate as collateral during lending booms. The results can be summarized as follows. First, the estimations show that cash flow is significantly correlated with investment before and after financial liberalization, particularly in the case of smaller firms. Second, financial constraints appear to have been eased for very small firms after financial liberalization. Third, the value of a firm’s real estate Ža proxy for collateral. is shown to be strongly correlated with investment throughout the period studied. Fourth, counting on real estate as collateral seems to have become more important after 1989. These results can be interpreted as saying that financial liberalization did not translate so much into a reduction in the premium of the cost of external funds over internal funds, but rather into an increase in the number of firms that were potentially eligible for credit. However, the poor state of the banks’ evaluating and monitoring capacities, together with prevailing legal and enforcement problems, led banks to rely heavily on collateral in their lending decisions. This collateralbased lending probably increased the vulnerability of the financial sector.

2. Financial repression and financial reform in Mexico During the years 1982–1988, Mexico was a textbook case of financial repression. As a reaction to the financial crisis in 1982, Mexico’s banking sector had been nationalized. Since the government had lost access to international capital markets, it was forced to finance its deficits domestically. For many years after the crisis, it did so largely through a crowding out of private borrowing.5 Banks were forced to lend to the public sector through a mixture of measures: high reserve requirements, mandated lending to preferential sectors Žamong them the public., and the imposition of regulations forcing banks to hold a certain fraction of their assets in the form of government liabilities. In 1986, for example, 72% of commercial bank credit flowed to the government. Interest rate ceilings on deposit and loans put further restrictions on banking activities. Together with high and variable inflation, they led to volatile and mostly negative real interest rates. 5

See Gruben and McComb Ž1997..

4

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

In late 1988, a rapid process of financial deregulation was initiated. This liberalization was part of a broader reform package adopted in that year: the success in reducing government deficits and in bringing down inflation contributed significantly to facilitating these changes. For example, in the more stable environment, the government was now able to cover its reduced financing needs through the issuance of short-term debt ŽCETES., and, after the implementation of the Brady plan, again through access to international capital markets. The reforms included the abolition of interest rate ceilings, a phase-out of reserve requirements, and the elimination of priority lending quotas. A new law allowed banks to move into universal banking. These changes resulted in a marked increase in the amount of savings intermediated through the banking system. As a result, the outstanding stock of private-sector loans increased at an inflation-adjusted annual rate of 30% between 1987 and 1994, and the share of these loans in GDP rose from 8.7% in 1987 to 41% by the end of 1994 ŽFig. 1.. To a large extent, this loan growth was financed through interbank borrowing via lines of credit from foreign banks. These capital inflows, which also took the quantitatively more important form of portfolio investments, increasingly substituted for domestic savings. A look at aggregate lending volumes, however, is not sufficient to understand the effects of financial liberalization. In a developing country like Mexico, the transition from a situation of financial repression to one with freer financial markets is unlikely to be sufficient for ensuring that the financial system fulfills its function optimally. Problems stemming from asymmetric information, poor regula-

Fig. 1. Credit of the consolidated banking system to the private and public sectors Žstocks in percent of GDP..

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

5

tion, and weak law enforcement are likely to constitute at least as important an obstacle to the access to external finance by firms as state interventions in the credit markets. With little transparency regarding the operations of banks and firms, deregulation may even have adverse effects on some firms’ liquidity constraints. Increases in real interest rates may weaken borrowers’ balance sheets and worsen the pool of applicants, so that access to credit may become more difficult for some firms. Ultimately, the actual effects of financial reform can only be assessed on an empirical basis, most likely by the use of disaggregated data. Since the early 1970s, a strand of the development economics literature has focused on the impact of financial deregulation in developing countries.6 Recently, another body of research has been stressing the importance of financial constraints for firms’ investment decisions in countries with highly developed financial systems.7 Finally, motivated in particular by the recent Asian crisis episodes, researchers have begun to pay increasing attention to bank lending behavior and its links to the macroeconomy. Building partly on the aforementioned Afinancial acceleratorB mechanisms, various models stress the role of asset prices, particularly real estate, in the development of lending booms.8 This paper intends to provide some microeoconomic evidence that is relevant for these three strands of research. While we borrow from the methodology used in recent work on the wedge between the cost of internal and external funds,9 we also propose methodological improvements to deal with some aspects neglected in the literature.

3. Data issues and summary statistics We use data from the Annual Industrial Survey conducted by Mexico’s National Institute of Statistics, Geography, and Information ŽINEGI..10 The survey covers 3199 manufacturing establishments from 1984 to 1994. The completion of the questionnaire is compulsory, and the purpose of the survey is merely statistical and not linked to tax collection. The database is a balanced panel: exiting plants were discarded from the sample by the collecting agency. However, according to INEGI, the number of exiting plants was very small. This can partly be explained by the fact that the survey attempts to cover roughly 80% of value added in manufacturing, having therefore a bias towards larger and more successful firms. 6

See McKinnon Ž1973. and Shaw Ž1973.. For surveys, see Hubbard Ž1998. and Schiantarelli Ž1995.. 8 See Chan-Lau and Chen Ž1998., Edison et al. Ž1998., Krugman Ž1998., McKinnon and Pill Ž1997. and Schneider and Tornell Ž2000.. 9 For a similar approach, see Atiyas Ž1992., Harris et al. Ž1994. and Jaramillo et al. Ž1997.. 10 Instituto Nacional de Estadıstica, Geografıa ´ ´ e Informatica. ´ 7

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

6

Nevertheless, a substantial number of small establishments is included in the database. We will discuss below how this bias may affect our results. The unusually rich database comprises a large number of variables, covering mainly production, input use, labor force, sales, inventories, investment expenditures and capital stocks. Capital expenditures are grouped into five categories: machinery, transport equipment, land, buildings, and other. Investment is defined as purchases minus sales of assets plus improvements. After the elimination of extreme outliers and plants with incomplete and inconsistent data, the balanced panel contains 1046 establishments. Details of the construction of capital stocks and investment rates as well as the criteria used for the elimination of outliers are given in Appendix A. A disadvantage for our purposes is the fact that most of the information is given at the establishment level only. To some extent, this limitation can be overcome. First, the data do contain information about profits at the firm level, which can be used to construct a measure of the firm’s cash flow. Secondly, it is possible to identify plants within the sample that pertain to a common firm.11 There is no indication of interconnections of establishments within the sample. Obviously, this does not preclude the possibility that there be other plants or firms outside the sample linked to establishments in the data set. Since the coverage of the sample is quite comprehensive, however, the working hypothesis maintained in the following is that all plants are single-establishment firms.12 As will be discussed later, if this hypothesis is violated in reality, it will be more difficult to find links between financial factors and investment. After eliminating establishments with less than three employees, incomplete or inconsistent data and extreme outliers, the sample used for all further purposes contains 1046 plants. A detailed description of the methods used in constructing the variables and eliminating outliers is given in Appendix A. The establishments were divided into four size categories, according to the total number of employees. Plants with less than 40 employees were classified as Avery small,B establishments between 40 and 100 employees were called AsmallB establishments with between 100 and 500 employees were categorized as AmediumB and those with more than 500 employees were considered Alarge.B Firms were classified as exporting if export sales represented at least 10% of their total sales. The main characteristics of the establishments are presented in Table 1. The table shows that, despite the bias towards larger firms, the database contains a significant number of smaller plants. Most establishments in the sample fall into the medium-size category. The most notable difference concerns the capital stocks and the number of employees of large firms: exporting establishments generally seem to be larger

11 12

This is feasible since the data include the registered capital of the company. Therefore, the words AfirmB and AplantB will be used interchangeably in what follows.

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

7

Table 1 Summary statistics for exporting and non-exporting firms Firm class

Number of plant-year observation

Total personnel 1990

Capital stock K t 1990

Investment It r K ty1

Cash flow CFt r K ty1

Very small Exporting Non-exporting

45 799

25 Ž17. 26 Ž8.

33,500 Ž530. 5230 Ž15,100.

0.01 Ž0.01. 0.06 Ž0.11.

0.08 Ž0.13. 0.10 Ž0.12.

Small Exporting Non-exporting

145 1578

59 Ž25. 52 Ž24.

12,780 Ž23,421. 7031 Ž12,583.

0.07 Ž0.17. 0.06 Ž0.10.

0.12 Ž0.14. 0.19 Ž0.27.

Medium Exporting Non-exporting

1436 4460

274 Ž114. 243 Ž109.

49,779 Ž84,165. 29,755 Ž52,927.

0.05 Ž0.06. 0.09 Ž0.12.

0.17 Ž0.21. 0.18 Ž0.26.

Large Exporting Non-exporting

1234 1809

1525 Ž1936. 1041 Ž757.

270,104 Ž355,163. 146,864 Ž207,784.

0.07 Ž0.09. 0.06 Ž0.08.

0.19 Ž0.28. 0.20 Ž0.27.

The figures represent means and standard deviations Žin parentheses.. The capital stock figures are given in thousands of pesos of 1994. Cash flows were derived based on distributed profits and reported depreciation. ŽSee Appendix A.. A firm was classified as exporting if exports represented at least 10% of total sales in any year. Investment refers to gross investment ŽSee Appendix A..

than non-exporting plants and their capital intensity is higher.13 In the Avery smallB category, plants were classified as AexportingB in only 5.3% of the cases, while they constitute about 41% of the large establishments. The greater capital intensity also explains why cash flows relative to capital stocks are lower for the export-oriented firms. 4. The role of internal funds 4.1. Main issues In recent years, a number of studies have been examining the effects of financial constraints on investment.14 The usual methodology is to test whether adding cash flow to standard investment equations helps explaining capital expenditure. The reasoning is the following: in a Modigliani–Miller world, measures of firm’s liquidity should not enter significantly in a correctly specified investment 13 Although the mean capital-labor ratios are not given here, they are always lower for nonexporting firms. 14 See, for example, Fazzari et al. Ž1988., Schaller Ž1993. or Bond et al. Ž1997.. For developing countries, an example using industry level data from Columbia is Tybout Ž1983.. See also Nabi Ž1989., Harris et al. Ž1994. and Jaramillo et al. Ž1997..

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

8

regression given that internal and external funds are perfect substitutes for the firm.15 By contrast, in an environment with informational asymmetries, bankruptcy costs and law enforcement problems, external funds will be more costly than internal funds. This wedge arises from the need to compensate lenders for adverse selection and moral hazard problems on the borrower’s side.16 Generally, theory predicts that the premium on external funds will decrease with the firm’s net worth.17 Higher cash flows today improve the financial position of the firm and increase the internal funds available for investment. Therefore, investment should respond positively to increases in cash flow. Empirically, the main problem with this approach stems from the possibility that cash flow may be correlated with investment for other reasons. For example, even without financial constraints, firms will respond to increases in cash flow if current cash flow is a good predictor of future profitability, which is likely to be the case. Possible solutions to this problem are including BrainardrTobin’s marginal q in the estimated equation or estimating Euler equations directly. Both of these approaches have their own drawbacks.18 Fortunately, the problem is much less severe in our case since we are primarily interested in assessing the effect of financial liberalization, thereby focusing on changes in the cash flow sensitivities of investment. A priori, there is no reason to believe that the correlation of current cash flow with future profit opportunities decreased after financial liberalization in 1989. If one does indeed observe a decline in the coefficients on cash flow with financial deregulation, this is indicative of a loosening of financial constraints. In the spirit of Fazzari et al. Ž1988., we also look at differences across firms that are likely to be related to their relative access to external financing. For example, small firms are more likely to be liquidity constrained.19 The argument is 15

See Modigliani and Miller Ž1958.. There is a large theoretical literature deriving these general results in a variety of set-ups. See, for example, Townsend Ž1979., Stiglitz and Weiss Ž1981., or Gale and Hellwig Ž1985.. 17 See, for example, the model in Bernanke and Gertler Ž1989. and the discussion in Bernanke et al. Ž1996.. 18 See Caballero and Leahy Ž1996., Chirinko Ž1997., and Gelos Ž1998. for a more detailed discussion of problems associated with the q approach. Our data do not include a measure of average Q. However, we made an attempt to follow Gilchrist and Himmelberg Ž1998. in using VAR’s to estimate the expected value of future marginal products of capital and the expected present value of future financial state variables of the firm, conditional on observed fundamentals. This attempt was not successful, probably because it is unlikely that a linear projection can appropriately capture expectations in a period characterized by substantial regime shifts and discrete events. Estimating Euler equations directly in principle circumvents the problem, since the impact of future profitability on current decisions is controlled for. However, this method is very susceptible to misspecification problems and its small sample properties are poor. In Gelos and Werner Ž1999. we nevertheless show Euler equation estimates which broadly confirm the results presented below. 19 There is an ongoing debate regarding the validity of this argument. See Kaplan and Zingales Ž2000. and Fazzari et al. Ž2000.. 16

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

9

that there are economies of scale in the collecting and processing of information about their situation that help to overcome the problems associated with asymmetric information between borrowers and lenders.20 Smaller firms are also more likely to have lower collateral and to be exposed to higher idiosyncratic risks.21 This distinction may be important in an environment where political connections —at least before financial deregulation—were important for the obtention of credit. In this regard, larger firms most likely had an advantage. Meaningful distinctions across firm types can also be made according to ownership structure and export orientation. A detailed discussion is given below. Which empirical investment model should be adopted as the baseline specification? Apart from models of the BrainardrTobin’s q variety, simple accelerator specifications are the most widely used in the literature. We adopt such a model, with the change in output as the accelerator variable. Although not an entirely satisfactory proxy, the change in output in the accelerator model should capture short-term changes in expected profitability fairly well.22 In the next section, the specification is modified to take into account the role of irreversibilities andror fixed costs of investment. To control for unobserved heterogeneity across firms, changes in the cost of capital, and other aggregate effects not explicitly modeled here, the model is estimated with firm-specific and time effects. All variables are scaled by the lagged capital stock. In order to test for the effect of internal liquidity, cash flow is included in the regression equation, giving: Ii t K i ty1

sb

D yi t K i ty1

qf

CFi t K i ty1

q lt q n i q ´i t

Ž 1.

Here, Ii t , K i t , D yi t , CFi t denote investment, the capital stock, the change in output and the cash flow of firm i at time t, respectively, and l t , n i and ´ i t stand for time effects, time-invariant firm effects and idiosyncratic error terms. If the firm-specific effects are not correlated with the explanatory variables, this equation can be estimated using random effects; otherwise, an estimation with fixed effects is appropriate. Hausman tests reject the hypothesis of no correlation at all usual significance levels; therefore, we only present the results of regressions with fixed effects. The explanatory variables may not be strictly exogenous. A possibility is to treat them as predetermined, i.e. assuming that EŽ x i t e i t . / 0 for s - t. To address 20

See Bernanke et al. Ž1996.. For the case of Mexico, Glaessner and Oks Ž1998. report that in late 1993, the nominal interest rates for large prime borrowers was 17–22%, while small and medium scale enterprise borrowers faced rates of around 27–36%. However, such differences in rates may also partly reflect economies of scale on the banks’ side. 22 Alternatively, one could include the change in sales as the activity variable. The results presented in the following are not essentially affected by this choice. 21

10

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

this issue, we estimated the investment equations using a Generalized Method of Moments ŽGMM. instrumental variables estimator developed by Arellano and Bond Ž1991.. There, the model is estimated in first differences, with lagged levels of the regressors as instruments. If the right-hand side variables are endogenous, twice-lagged levels are valid for a serially uncorrelated error in the levels equation; if the variables are predetermined, levels lagged one period or more are permissible. However, the drawback of this method is that measurement error problems may be exacerbated; in addition, lagged levels of the regressors are not always good instruments. In fact, cash flow and changes in output lagged two periods or more proved to be only weakly correlated with current differences. This is explainable by the large changes in economic conditions experienced during the period. Despite these problems, the GMM estimates using instruments lagged one period and more will be discussed in addition to the Ordinary Least Squares ŽOLS. results. We allow the coefficient on cash flow to vary across firm size categories by interacting cash flow with dummies for each size class and present results for separate samples depending on ownership structure and the importance of exports. Beyond size, access to credit is likely to depend on ownership structure Žprivate vs. public, and with vs. without foreign ownership. and on the firms’ share of exports in total sales. A priori, one would expect publicly owned firms to be less financially constrained than private companies, particularly in the earlier years of the sample. For example, public firms probably benefited more from selective credit policies before financial liberalization or faced soft budget constraints. Similarly, firms with foreign participation are more likely to have had access to foreign capital, suffered less from liquidity constraints, and to have benefited relatively less from financial liberalization. With a volatile domestic market, export-oriented firms are less vulnerable to demand shocks at home, generating a more predictable income stream, and should therefore be preferred by lenders. On the other hand, we would expect nonexporting and purely Mexican firms to be more financially constrained and to benefit relatively more from financial deregulation. During the time period covered in our sample, many state-owned enterprises were privatized in Mexico. If we were not able to distinguish between publicly and privately owned firms, our results might be affected by this privatization process: we might find an increased importance of liquidity constraints in the latter period simply because after they were privatized, firms which formerly enjoyed access to credit as state-owned enterprises lost this privilege.23 23 Data on ownership structure was only available until 1990. Although after that date, privatization of many enterprises continued, the classification of 1990 had to be retained for the following years. Any sample separation raises the question of a possible endogeneity of the selection criteria: for example, changes in firm size might be correlated with movements in investment. In principle, this problem leads to the same econometric approach as discussed above. See Schiantarelli Ž1995..

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

11

By including an interaction term of cash flow with time dummies after 1989, we test whether the effect of liquidity changed after financial liberalization. However, care has to be taken not to attribute all changes after that year to domestic financial liberalization. Around 1988r1989, economic conditions also changed in other ways. As noted earlier, the severity of the debt problem diminished; after a successful debt renegotiation within the Brady plan, capital began to return to Mexico. The reduction in financial constraints stemming from these inflows cannot only be attributed to a liberalized internal financial system. Moreover, the stabilization program adopted in 1988 may by itself have led to a credit expansion through a remonetization of the economy and a decrease in risk associated with lower inflation levels.24 Lastly, significant trade liberalization measures were undertaken, reducing import barriers and resulting in an increased outward orientation. An increased ability by Mexican manufacturing firms to compete on international markets might have contributed to an easing of credit constraints, as argued in the earlier discussion of exporting vs. nonexporting firms. Although, as mentioned above, there are reasons to suspect that the obtained cash flow estimates will reflect more than the pure liquidityrnet worth effects, the coefficients on cash flow are on the other hand likely to be biased towards zero. If the working assumption of single-establishment firms is violated, the relation between the cash flow variable Žconstructed using data on profits at the firm level. and plant-level investment will be blurred. Moreover, remember that the sample is a balanced panel, with a selection bias toward established and successful firms. These firms are less likely to suffer from financial constraints than the general population of enterprises. Since we do not have any information on any potential exiting firms that were discarded from the sample Žas mentioned earlier, according to INEGI, these cases were very few. or on the specific selection criteria that differentiates this sample from the universe of establishments, our inferences cannot be generalized to the total population. However, we will provide some indicative tests on the direction of the bias. 4.2. Results The results for the whole sample show that according to both estimation methods ŽOLS and GMM., cash flow enters significantly for small and medium firms ŽTable 2.. While this is also true for large establishments according to the OLS results, the coefficient is not significant in the GMM estimation. Overall, the coefficients are lower than the ones in most other studies, but comparable to those reported by Harris et al. for a similar dataset from Indonesia.25 Interestingly, the 24

See Khamis Ž1996. for a discussion of these issues. When the data include many smaller firms, issues related to the life cycle of a firm may bias the coefficient on cash flow towards zero. As the data show, those firms that invest more initially tend to have higher variable costs at the beginning and therefore lower cash flows than the mean. Later on, the same firms have lower investment and higher cash flows. 25

12

Variable

All firms OLS

Cash flow, very small firms CF i t Ds K i ty 1 Cash flow, small CFi t firms Ds K i ty 1

Only private GMM

OLS

GMM

Only with public participationa

Only with foreign participation

OLS

GMM

OLS

GMM

Only purely Mexican

Only non-exporting

Only exporting

OLS

GMM

OLS

GMM

OLS

GMM a

0.10 Ž3.31 .

0.17 Ž4.01 .

0.11 Ž3.67 .

0.17 Ž4.23 .

y0.67 Žy1.89 .

0.51 Ž0.66 .

0.10 Ž3.78 .

0.17 Ž4.44 .

0.12 Ž4.16 .

0.19 Ž4.62 .

y0.07 Žy0.19 .

–

0.12 Ž1.45 .

0.21 Ž8.55 .

–

–

0.04

0.08

0.12

0.17

0.10

0.09

0.13

0.06

Ž1.09 .

Ž8.45 .

Ž6.27 .

Ž3.54 .

Ž5.77 .

Ž2.67 .

Ž2.12 .

Ž0.31 .

0.10

0.11

0.10

0.11

Ž6.44 .

Ž3.02 .

Ž6.33 .

Ž2.84 .

Cash flow, medium-sized CF i t Dm K i ty 1

0.04 Ž 4.75 .

0.05 Ž 2.41 .

0.04 Ž 4.68 .

0.05 Ž 2.19 .

y0.03 Žy0.44 .

y0.01 Žy0.96 .

0.05 Ž3.21 .

0.04 Ž1.69 .

0.04 Ž3.63 .

0.05 Ž2.31 .

0.03 Ž2.66 .

0.04 Ž1.96 .

0.10 Ž5.57 .

0.09 Ž2.84 .

Cash flow, large firms CF i t Dl K i ty 1

0.04 Ž2.25 .

y0.02 Žy0.67 .

0.04 Ž2.01 .

y0.05 Žy1.77 .

0.09 Ž1.11 .

0.10 Ž6.95 .

0.04 Ž2.00 .

0.03 Ž1.21 .

0.03 Ž1.16 .

y0.02 Žy0.80 .

0.02 Ž1.00 .

y0.06 Žy2.63 .

0.07 Ž2.87 .

0.09 Ž1.78 .

CF, very small firms after liberalization CF i t D s D after K i ty 1

y0.10 Žy2.99 .

y0.13 Žy2.58 .

y0.13 Žy3.77 .

y0.12 Žy2.24 .

0.70 Ž1.82 .

–

y0.10 Žy1.16 .

y0.29 Žy7.95 .

y0.09 Žy2.57 .

y0.13 Žy2.39 .

y0.10 Žy2.73 .

y0.12 Žy2.40 .

0.15 Žy0.40 .

y0.68 Žy0.79 .

Cash flow, small after lib. CF i t D s D after K i ty 1

y0.03 Žy1.66 .

y0.05 Žy1.37 .

y0.03 Žy1.53 .

y0.05 Žy1.20 .

–

–

0.05 Ž1.32 .

y0.07 Žy4.06 .

y0.04 Žy2.18 .

y0.11 Žy2.41 .

y0.02 Žy1.17 .

y0.03 Žy0.77 .

y0.09 Žy0.94 .

y0.13 Žy0.71 .

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

Table 2 Accelerator model ŽOLS and GMM .

0.02 Ž1.99 .

y0.04 Žy1.39 .

0.02 Ž2.13 .

y0.03 Žy1.19 .

0.02 Ž0.20 .

y0.04 Žy1.77 .

y0.02 Žy1.02 .

0.00 Ž0.21 .

0.04 Ž3.14 .

y0.05 Žy1.77 .

0.03 Ž2.94 .

y0.04 Žy1.40 .

y0.03 Žy1.45 .

y0.05 Žy0.97 .

CF, large after liberalization CF i t D l D after K i ty 1

0.03 Ž2.06 .

0.00 Ž0.04 .

0.04 Ž2.13 .

0.04 Ž1.07 .

y0.08 Žy0.85 .

y0.13 Žy6.65 .

y0.01 Žy0.29 .

y0.01 Ž0.39 .

0.07 Ž2.87 .

y0.01 Žy0.21 .

0.05 Ž2.49 .

0.05 Ž1.63 .

y0.01 Žy0.86 .

y0.08 Žy1.26 .

0.01

0.00

0.01

0.00

0.00

0.01

0.01

0.00

0.01

0.00

0.01

0.00

0.01

0.00

m1 m2 Sargan test

Ž4.65 . 205.7 Ž9 . – – –

Ž4.65 . 210.8 Ž9 . – – –

Ž3.28 . 308.1 Ž7. y3.87 y0.60 33.9 Ž38 . 478

Ž2.55 . 41.7 Ž9. – – –

Ž1.09 . 339.6 Ž9. y7.35 y1.30 53.6 Ž60 . 2763

Ž4.10 . 186.2 Ž9. – – –

Ž0.67 . 48.1 Ž9. y12.43 y1.23 85.9 Ž68 . 6651

Ž3.92 . 154.9 Ž9. – – –

Ž0.47 . 59.2 Ž9. y12.8 y1.42 92.5 Ž68 . 7074

Ž2.59 . 74.6 Ž9. – – –

10,460

Ž0.73 . 61.0 Ž9 . y13.93 y1.50 87.3 Ž68 . 8807

Ž0.89 . 22.4 Ž7 . – – –

No. of observation

Ž0.54 . 55.1 Ž9 . y14.40 y1.67 92.5 Ž68 . 9414

Ž0.86 . 19.6 Ž9. y17.00 y1.06 100.3 Ž68 . 2340

D yit K i ty 1 Wald test

9841

619

3070

7390

7860

2600

Dependent variable: Ž IrK . i t . T statistics in parentheses. Time dummies were included in all regressions Žcoefficients omitted .. In the GMM estimation, the model was estimated in first differences, with levels of the regressors lagged one or more periods as instruments. The reported Wald test is a significance test for all the included variables Žexcept dummies .; the test statistic is distributed as x Ž p ., where p is the difference between the number of instruments and the number of regressors. m1 and m 2 are tests of first- and second order autocorrelation with a N Ž0,1 . distribution. The Sargan test is a test of the overidentifying restrictions Žsee Sargan, 1988 .. To correct for heteroskedasticity, a two-step estimation procedure was used. D s denote dummy variables for size and for the period after 1989. The DPD program developed by Arellano and Bond Ž1988 . was used in the estimation. a Size categories had to be lumped together due to the small number of observations.

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

CF, medium after liberalization CF i t D m D after K i ty 1

13

14

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

coefficient on current cash flow and its standard error were unaltered when including the actual values of cash flow at t q 1 in the regressions.26 As expected, the size of the coefficients on cash flow decreases with firm size. The coefficient on the change in output is positive and significant in the OLS case, but not significant according to the GMM estimates These results remain essentially unaltered when using the changes in sales as the activity variable or when including lagged values of the change in output. The results in the first two columns suggest that the smallest firms benefited particularly strongly from the increased credit availability after 1989.27 For the next size category, the results also show a decline in cash-flow sensitivities, but this reduction is not significant at the 5% confidence level. According to the OLS estimates, the relationship between cash flow and investment increases after 1989 for medium and large firms. This is not confirmed by the GMM estimation.28 Since most of the firms in the sample are privately owned, the results do not change noticeably when examining only firms without public participation. However, the figures for the public enterprises are different: except for the case of large plants, none of the estimates indicate the presence of a significant relationship between cash flow and investment. These firms appear to operate in a very different environment: note that the activity variable does not enter significantly in the OLS case, either. These results are in line with our ex-ante expectations. Surprisingly, the GMM results show a decline in cash-flow sensitivities for large firms. Comparing firms with foreign participation with those that are purely Mexican-owned, the expected differences do not show up consistently neither in the size of the cash-flow coefficients nor in the changes after 1989. On the other hand, as expected, very small nonexporters appear to have benefited more than the smallest exporters from increased access to credit after 1989.29 Overall, for very small firms the effects of cash flow on investment decrease strongly after 1989, suggesting that financial liberalization resulted in an easing of financing constraints for these companies, while this is not the case for larger

26

Cash flows at t q1 did not enter significantly. See Harris et al. Ž1994. for a similar finding. It is interesting to note that in a survey carried out in 1991 ŽCANAME, 1991. among manufacturing firms in the electrical sector, of the firms classified as Amicroenterprises,B 10.7% indicated that access to financing was their major growth impediment, whereas 16.6 of the AsmallB firms named this factor as the most important one. Only 2.9% of the AmediumB and none of the AlargeB firms felt that financing constraints were the single most important limiting factor. Unfortunately, the survey does not provide the definition of the size categories. 28 Note that the GMM estimation results for all firms need to be interpreted with caution, since the Sargan test of overidentifying restrictions does not support the instruments used. 29 Babatz and Conesa Ž1997., using data from 71 stock-listed firms, do not find a significant difference in the cash flow sensitivities between exporting and nonexporting firms when estimating a similar specification prior to 1992. 27

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

15

firms.30 Some of the larger companies may have benefited from preferential credit distributed through public development banks prior to financial liberalization. As mentioned before, large firms were more likely to have had political connections facilitating access to credit before the process of liberalization was initiated; to some extent they were also able to finance themselves through the stock market. In addition, the rise in real interest rates may have resulted in an increased cost of finance for large firms that had access to credit prior to 1989, but not for those smaller firms that were essentially cut off from capital markets. This means that for big companies, financial liberalization had two-sided effects on the cost and availability of external funds.31

5. Fixed minimum project size and the role of collateral The estimations above overlooked two problems that have rarely been treated explicitly in the empirical literature: the likely presence of indivisibilities in investment and the possibility that a firm is completely cut off from credit markets. If, for example, an investment project requires a minimum size to be carried out, credit rationing may prevent a firm from undertaking the investment. In particular, in developing countries, the phenomenon of credit rationing may be more important than the more subtle issue of changes in the external finance premium Žsee Dailami and Giugale, 1991; Rama, 1993.. In addition, the empirical analysis above concentrated on the role of internal funds, and neglected the importance of collateral, which plays a salient role in the theoretical on borrower–lender relationships literature and in the descriptions of recent lending booms preceding crises. Empirically, the large number of observations with zero investment Žapproximately 13%. needs to be addressed. Although there are various potential reasons behind this phenomenon, we provide an explanation based on liquidity constraints. The model is a simple two-period moral hazard model with a risk neutral firm and a risk neutral lender as presented in Hoshi et al. Ž1993. and Holmstrom ¨ Ž1993., and modified to allow for variable investment size.32 30

Harris et al. obtain a similar result for Indonesia. Jaramillo et al. do not find an significant impact of financial liberalization in Ecuador. 31 To investigate the hypothesis that financial constraints are likely to have been higher in the total population of firms than in our sample Žwhich is biased toward more successful and growing enterprises., we ran similar regressions for those firms whose employment growth over the whole period was below average. The coefficient on cash flow for the two smallest size categories was indeed larger than for the total sample Ž0.18 and 0.12, respectively., and the change after 1989 more pronounced Žy0.18 and y0.05.. ŽAll coefficients were significant at the 1% level.. The coefficients for the two larger firm categories were the same as for the whole sample. 32 An extended version of the model is used by Holmstrom ¨ and Tirole Ž1998..

16

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

At time zero, t s 0, firm i has an opportunity to invest in a project that requires a minimum investment of size Imin . However, above Imin , the investment project can be carried out at any size I. At t s 1, the gross payoff from the investment is either RI Žin case of success. where R is the gross rate of return or 0 Žin case of failure.. The firm can influence the probability of success through its choice between two technologies. If it uses the efficient technology H, the probability of success is p H . Alternatively, it can use an inefficient technology L, with a probability of success p L - p H , which would leave BI dollars for the firm to use for unproductive activitiesrperquisites in both outcomes. The lender cannot observe the choice of the firm, which creates a moral hazard problem. Assume that the expected return is negative if the inefficient technology L is chosen, and positive in case the efficient technology H is used: p H RI y I ) 0 ) p L RI y I q BI Ž 5. The amount of cash that could be obtained by selling the firm’s assets in the second period is A; this represents the maximum that the firm can be forced to pay under liquidation. In the simplest case, in which the firm has no cash, it will borrow the whole amount needed for the investment project. A contract C s Ž ys , yf . between the lender and the firm, where yi is the amount that the investor is paid back in case of success Žs. or failure Žf., is viable if a number of restrictions are satisfied. The first is that the payments are feasible: ys F RI q A yf F A Ž 6. In addition, the following incentive compatibility constraint must hold to induce the firm to choose the efficient technology: p H Ž RI q A y ys . q Ž 1 y p H . Ž A y yf . G p L Ž RI q A y ys . q Ž 1 y p L . Ž A y yf . q BI Ž 7. For the lender to break even in expectations, the following condition must hold Žfor convenience the opportunity cost is assumed to be zero.: p H ys q Ž 1 y p H . yf G I Ž 8. It can be shown that in equilibrium yf s A, so that in case of failure of the project, the lender receives all assets of the firm. Using this insight, one can solve for the level of assets Žcollateral. that is necessary to undertake an investment of a given size: B A G I q I yp H R q p H Ž 9. pH y pL Note also that if the Õalue of the total assets of a firm is less than Imin y Imin w p H R y p H BrŽ p H y p L .x , the inÕestment project cannot be undertaken. Put differently, the amount of collateral determines the size of investment. Although this model is extremely simple, it captures some important aspects of reality not considered in other models. In particular, the possibility of credit rationing motivates frequent episodes of zero investment in a natural way.

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

17

In order to proceed with an empirical implementation, a few problems have to be overcome. First, the INEGI data do not include many financial variables, so that an accurate measure of a firm’s collaterizable net worth cannot be constructed.33 However, the data do provide a disaggregation of the capital stock into land and buildings and equipment. Therefore, in what follows, the value of real estate Žland and buildings. of the firms is used as a proxy for collaterizable assets. This choice is sensible given that real estate is the most widely used form of collateral for longer-term credits in Mexico. This is mainly due to problems with registries for movable capital; it is not feasible for lenders to ensure a unique claim on such types of collateral. Glaessner and Oks Ž1998. note that collateral Ausually takes the form of real estate equal to as much as three times the value of the loan.B 34 According to the predictions of the model, we would expect that, ceteris paribus, the probability of investment increases with value of the firm’s real estate and that the size of the investment depends on the value of the real estate. This suggests estimating a Tobit-type model. We again include the change in output to capture changes in profitability. The coefficient on cash flow is not constrained to be equal to the real estate coefficient since it is not clear that all of the internal funds available at the time of investment can be seized in case of bankruptcy. To avoid spurious correlation, capital expenditures on real estate were subtracted from investment, and net revenues from sales of land or buildings were deducted from cash flow. Formally, the model can be written as: Ii)t K i ty1 Ii t K i ty1

sb

s

D yi t K i ty1 Ii)t

K i ty1 s 0 if

qf

if

CFi t K i ty11 Ii)t K i ty1

Ii)t K i ty1

qu

RE i t K i ty1

q lt q n i q ´ i t ,

) Imin i

F Imin i

Ž 10 .

where RE i t denotes the value of the real estate owned by the firm. This is a Tobit model with fixed effects. Fixed effects are essential in order to control for unobserved heterogeneity, in particular concerning the minimum size of the firm’s

33 For example, the data do not cover information on debt, which would be needed to construct the value of net collaterizable assets. While the database contains information on interest payments, it would obviously be against the logic of this paper to assume a common interest rate for all firms in order to deduce the debt stock. 34 See also Black et al. Ž1996. for the importance of real estate for entrepreneurial decisions.

18

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

investment project Imin i . ŽNote that the fixed effects n i and Imin i cannot be identified separately.. However, the estimation of a Tobit model with fixed effects is not trivial. Honore´ Ž1992. has developed an estimator that relies on the symmetry of the distribution of the latent variable,35 which we used here. Similarly to the approach followed in the previous section, a test was carried out on whether the postliberalization coefficients on cash flow and real estate differed from the ones for the period 1984–1994. The results are striking: the value of real estate ŽRE. has a significant effect on the investment decisions of all firms and the coefficient on RE is higher than the one on cash flow ŽTable 3..36 These qualitative results remain unaltered when modifying the model as to restrict attention only to investments above the maintenancerdepreciation threshold of 7%, i.e. replacing Ii t s 1 if ŽŽ Ii)t .r Ž K i ty1 .. ) d above.37 The pattern of cash-flow coefficients across firms, in turn, is similar to the one found earlier, and, in some cases, more in line with the a priori predictions. Cash flow matters for most types of firms: exceptions are public enterprises, small and very small foreign-owned firms, and very small exporting firms.38 As seen earlier, microenterprises show a drop in cash flow sensitivities after 1989. Exceptions again are those firms with foreign participation and those that are classified as exporters. Interestingly, consistent with the earlier results, for medium and large purely Mexican-owned firms, the correlation between cash flow and investment appears to increase after 1989. Is spurious correlation the reason for the high t-statistics on the real estate variable? One might suspect that non-real estate investment spending Žthe dependent variable. moves in tandem with investment in real estate. Current investment in real estate, in turn, could potentially contribute significantly to real estate shocks. If one does not control sufficiently for the common factor driving both types of expenditures Žexpected profitability., one might misinterpret the observed correlation. However, three facts speak against this: first, investment in real estate is only weakly correlated with other capital expenditures. Second, regressions including only those cases in which investment in real estate was zero gave qualitatively very similar results. Third, the main results were unaltered when

35

A brief description of the main idea behind the estimator is given in Appendix B. Cash could be regarded as collateral, so that one might expect the coefficients on real estate and on cash flow to be of similar size. However, as noted above, it is likely to be more difficult to repossess cash. 37 The importance of real estate was confirmed in the results of fixed-effects logit regressions presented in Gelos and Werner Ž1999.. These estimations were based on the original version of the theoretical model which does not allow for variable project size. 38 The coefficients on purely Mexican medium-sized firms and on large public firms are only significant at the 10% level. 36

Table 3 Tobit estimation with fixed effects Variables

Only private

Only with public participationa

Purely Mexican

Only wr foreign participation

Only non-export

Only export

0.43 Ž6.35 .

0.40 Ž5.95 .

0.91 Ž4.16 .

0.42 Ž4.40 .

0.43 Ž5.66 .

0.40 Ž5.93 .

0.37 Ž2.75 .

0.04 Ž2.97 .

0.04 Ž2.96 .

0.30 Ž0.98 .

0.03 Ž1.66 .

0.06 Ž3.36 .

0.04 Ž2.78 .

0.03 Ž1.31 .

0.18 Ž4.43 .

0.21 Ž4.05 .

–

0.19 Ž4.18 .

0.17 Ž1.28 .

0.19 Ž4.51 .

y1.98 Žy1.56 .

0.15 Ž3.46 .

0.14 Ž3.40 .

33.01 Ž0.00 .

0.20 Ž3.87 .

0.03 Ž0.73 .

0.15 Ž3.22 .

0.12 Ž2.00 .

0.04 Ž3.22 .

0.04 Ž3.19 .

y0.2 Žy0.48 .

0.05 Ž2.48 .

0.04 Ž2.30 .

0.03 Ž2.24 .

0.07 Ž3.23 .

0.05 Ž3.03 .

0.05 Ž2.52 .

0.04 Ž0.41 .

0.06 Ž2.21 .

0.05 Ž2.33 .

0.05 Ž2.58 .

0.06 Ž1.85 .

Cash flow, very small firms CF i t after liberalization D s D after K i ty 1

y0.14 Žy2.52 .

y0.17 Žy2.70 .

–

y0.16 Žy2.66 .

y0.04 Žy0.19 .

y0.14 Žy2.47 .

0.78 Ž1.01 .

Cash flow, small firms CF i t after liberalization D s D after K i ty 1

y0.03 Žy0.75 .

y0.03 Žy0.65 .

y32.6 Ž0.00 .

y0.07 Žy1.19 .

0.06 Ž1.56 .

y0.03 Žy0.64 .

y0.11 Žy1.54 .

Cash flow, medium-sized firms CF i t after liberalization D m D after K i ty 1

0.02 Ž1.59 .

0.02 Ž1.63 .

0.22 Ž0.55 .

0.05 Ž2.35 .

y0.01 Žy0.53 .

0.03 Ž1.96 .

y0.01 Žy0.32 .

Cash flow, large firms CF i t after liberalization D l D after K i ty 1

0.04 Ž1.95 .

0.04 Ž1.93 .

0.35 Ž0.36 .

0.07 Ž2.31 .

0.00 Ž0.21 .

0.04 Ž1.76 .

0.03 Ž0.80 .

0.01 Ž3.45 .

0.01 Ž3.36 .

0.00 Žy0.02 .

0.01 Ž3.04 .

0.01 Ž1.92 .

0.01 Ž2.75 .

0.01 Ž2.42 .

240.8 10,460

224.9 9841

124.8 619

194.2 7390

106.2 3070

194.7 7860

89.7 2600

RE i t K i ty 1 RE i t K i ty 1

D after after liberalization

Cash flow, very small firms Cash flow, small firms

CFi t K i ty 1

CFi t K i ty 1

Dm

Cash flow, medium-sized firms Cash flow, large firms

D yit K i ty 1 2

x test No. of observations

CFi t K i ty 1

Ds

CFi t K i ty 1

Dl

Dm

19

Dependent variable: Ž IrK . i t Žexcl. purchases of real estate .. T-statistics are given in parentheses. Year dummies included in all regressions Žcoefficients omitted .. The Newton optimization algorithm, as implemented in the OPTMUM routine of GAUSS, was used with a polynomial loss function. The program PANTOB described by Campbell and Honore´ Ž1991 ., kindly made available by Bo Honore, ´ was employed in the estimation. a For public enterprises, the number of observations was too small to use four size categories. Three classes were used instead, with those firms with up to 100 employees classified as AsmallB.

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

All

20

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

including the investment expenditures on real estate as an additional explanatory variable or when using lagged values of the real estate variable.39 Note that it is more difficult than in the case of cash flow to argue that the reason why real estate matters for investment is that its value is correlated with profit opportunities of the firm. Changes in profit opportunities could be idiosyncratic or aggregate. Idiosyncratic changes in the value of real estate can only come from purchases or sales, which were discussed above and were found not to be driving the results. By contrast, movements in land prices are likely to be correlated with general changes in business conditions. However, these types of aggregate effects are controlled for by the inclusion of time dummies. Lastly, cross-sectional variations in the firms’ stocks of real estate are unlikely to be systematically associated with differences in future profit opportunities. For most classes of firms, the importance of collateral appears to increase after 1989. This is not as surprising as it may seem. Although an increased availability of credit should have contributed to a reduction of liquidity constraints, there is no reason to believe that the informational and enforcement problems that motivate the use of collateral diminished after the liberalization of the financial sector. However, many firms that previously were completely cut off from any credit were now in principle eligible for bank loans. For them, possessing collateral became more important.40 These firms were new borrowers, whose risks were difficult to assess. On the banks’ side, credit expansion was not accompanied by a comparable increase in their technical capabilities.41 The type of lending conducted throughout most of the 1980s, namely the intermediation of resources to the public sector, had not fostered the development of credit-analysis techniques.42 Moreover, access of foreign banks to the Mexican market, which possibly could have promoted the implementation of more advanced credit risk monitoring practices, was tightly restricted. Due to lack of experience, technology and human resources, credit was extended mainly against collateral. There is little detailed information available on this issue, but Table 4 shows the percentage of collaterized loans over 20 million pesos taken over by the agency formed to recapitalize the banks after the crisis ŽFOBAPROA.. One would expect a lower-than-average reliance on collateral for the case of these larger loans, since they were presumably extended to larger commercial borrowers with a longer track record. Nevertheless, the vast majority of banks extended these credits against collateral. Of the 1022 loans for which this

39 One would expect the effect of cash flow to diminish with increased collateral value. An interaction term of cash flow and real estate in fact had the anticipated negative coefficient, but was not always significant. 40 In a firm survey conducted by the World Bank Ž1994., insufficient collateral was mentioned, together with high interest rates, as the main deterrent from investment. 41 See Gruben and McComb Ž1997.. 42 See Dıaz ´ de Leon ´ and Schwartz Ž1997. and Mancera Ž1997..

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

21

Table 4 Proportion of collateralized loans over 20 million pesos Name of bank

Percentage

Atlantico Banamex Bancen Bancomer Banorte Banpais BBV Bital Capital Cremi Interestatal Obrero Oriente Promex Pronorte Santander Mex. Serfin Union

100 80 62 76 71 n.a. 73 100 27 n.a. 90 n.a. n.a. 64 75 96 30 0

Source: FOBAPROA.

information is available, 60% were backed by collaterizable assets. As mentioned earlier, in most cases, the collateral consisted of real estate. Real estate prices had collapsed in the early 1980s, but experienced an enormous upswing since 1987. One Žpartial. description of the lending boom preceding the 1994r1995 crisis would be the following: rising real estate prices made it easier for firms to access credit, which allowed the completion of projects and improved the firms’ financial situation. This in turn lowered the cost of finance and led to further investment activity and higher demand for land. It is easy too see how such a Afinancial acceleratorB process can be self-reinforcing until it is interrupted by an economywide shock.43 Similar mechanisms may also help to explain the severity of the recent Mexican crisis. Even before the actual crisis, the share of nonperforming loans was rising, a situation that was worsened by the 1993 drop in real estate prices and a fall in stock market prices in 1994. This made lenders more reluctant to lend and increased incentives to engage in risky activities. ŽWith reduced collateral values, there is less to lose.. When finally the exchange-rate crisis hit the balance sheets of many firms severely, this effect in turn contributed to a decline in lending, giving

43 See, for example, Bernanke et al. Ž1996., Kiyotaki and Moore Ž1997. and Edison et al. Ž1998.. Schneider and Tornell Ž2000. construct a model that combines bailout guarantees with a financial accelerator mechanism in order to explain the dynamics of asset prices during lending booms.

22

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

rise again to similar financial accelerator effects as described earlier.44 A decline in borrower’s net worth in general increases the premium in the cost of external over internal funds, reducing investment even for firms with high-return projects, and potentially leading to an economy-wide decline in asset prices. These lower investment levels, in turn, decrease the availability of funds in the next period, which again depresses capital expenditures, and so forth. This reasoning is in line with the view expressed, among others, by Mishkin Ž1996., who, in discussing the Mexican 1994r1995 crisis, attributes an important role to balance-sheet effects and informational asymmetries. The presented interpretation underscores the riskiness of overcoming agency problems in borrower–lender relationships through the use of collateral whose value itself is prone to move with aggregate shocks.

6. Conclusion While financial constraints influenced investment behavior in the Mexican manufacturing sector throughout the examined period,45 financial liberalization resulted an easing of financing constraints for some, in particular small firms. For others, in particular, large purely Mexican firms, the reliance on internal funds increased, consistent with the interpretation that these firms lost privileged access to credit with financial deregulation. Collateral in the form of real estate played an important role in determining investment, even more so after 1989. One interpretation is the following: financial liberalization did not translate so much into a reduction in the premium of the cost of external funds over internal funds, but rather into an increase in the number of firms that were potentially eligible for credit. However, the poor state of the banks’ evaluating and monitoring capacities, together with prevailing legal and enforcement problems, led banks to rely heavily on collateral in their lending decisions: having real estate became more important for firms. Since this collateral-based lending probably increased the vulnerability of the financial sector, these facts highlight the need for a better understanding of the incentives guiding lending behavior in order to adopt effective banking regulation and supervision. One implication from the results is that one should not expect financial liberalization to result in an elimination of financial constraints. Enforcement difficulties and problems of asymmetric information in lender–borrower relationships, which constitute a main reason for financing constraints, are likely to remain important. Broader reforms are required to tackle these problems, for example in the area of bankruptcy laws and creditor protection. Similarly, it takes time to build screening and evaluation capacities on the banks’ side. 44 45

Apart from the drop in domestic demand, firms were hit by unhedged foreign currency liabilities. This conclusion is in line with that of ˙Is¸can Ž1998..

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

23

This may be relevant for an understanding of the severity of the Mexican crisis 1994r1995. In addition to increasing the financial system’s vulnerability to aggregate shocks prior to the crisis, the effects of the devaluation were exacerbated. Banks were suddenly stuck with large quantities of real estate, since many firms were hit hard by the increase in the peso value of their debts. Given the prevailing agency problems, the unwillingness of banks to continue lending possibly resulted in a Afinancial acceleratorB mechanism, which led the economy further into recession. The findings regarding the importance of real estate are consistent with the role often attributed to asset, in particular real estate prices in recent financial crises.

Acknowledgements The authors are indebted to Michael Boozer, Eduardo Borensztein, William Brainard, Steven Fazzari, Rafael Gamboa, Ann Harrison, Michael Krause, Stefan Krieger, Giuseppe Moscarini, William Nordhaus, Gustav Ranis, Luca Rigotti, Christopher Sims, T.N. Srinivasan, Skander Van Den Heuvel, seminar participants at Banco de Mexico, Yale University, the IMF, and the World Bank, as well as an ´ anonymous referee for helpful comments. Part of the research for this paper was conducted while Gaston Gelos was a visiting researcher at Banco de Mexico. He is ´ grateful to Agustın and ´ Carstens and Moises ´ Schwartz, then at Banco de Mexico, ´ to Miguel Cervera and Abigail Duran ´ from INEGI. Alejandro Cano and Jose´ Carlos Rodriguez provided very helpful assistance.

Appendix A. Construction of the variables Capital Stock: The survey includes replacement cost values for five categories of fixed assets: machinery equipment, buildings, land, transport equipment and other. However, due to the variability of these series, we opted for not using these values. Instead, a perpetual inventory method based on reported investment figures was used, with the replacement cost numbers for 1984 as the initial stocks. The assumed depreciation rates are zero for land, 4% for buildings and 7% for all other assets. InÕestment: Investment is defined as purchases minus sales of used and new assets plus improvements on existing assets plus capital assets produced for own use. Machinery and transport equipment investment were deflated by the mid-year machinery price index, other investment by the mid-year wholesale price index, purchases of land by the mid-year Mexico City Land Price Index, construction expenditures by a construction mid-year price index. Cash flow: Mexican law requires every firm to pay out 10% of profits to its employees. We multiplied the profit-sharing figures by 10 and added reported

24

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

depreciation Žwhich in most cases reflects accounting, not economic values. to obtain cash flow at the firm level. Obviously, these figures are problematic, since profit-sharing is never negative. However, only 3.4% of the observations are equal to zero. In rare cases, firms were allowed to depart from the rule prescribing 10% profit-sharing; in general, such deviations were more likely if a firm faced a difficult situation and committed itself to reinvest its profits. Obviously, these cases tend to bias the coefficient on cash flow towards zero. The mid-year wholesale price index was used to deflate cash flows. Output: In the calculation of output values, a correction for maquila services Žsubcontracting work. had to be undertaken.46 In general, output of the firm rendering subcontracting services is counted as output from the company paying for the services. Therefore, following Grether Ž1994., income for maquila services was added, and maquila costs were subtracted from the reported value of manufactured products. This correction may not be accurate in all cases. Mid-year producer price indices at the four-digit disaggregation level were used for deflation. In order to eliminate outliers, establishments with zero or missing capital were eliminated entirely from the sample. In addition, plants that reported values for the change in output, investment and real estate, scaled by the lagged capital stock, in the top and bottom three percentiles, were discarded, as well as establishments with less than three employees.

Appendix B. Honore’s estimator for Tobit models with fixed effects The method is based on a generalization of Powell’s Ž1986. trimmed least squares estimators for Tobit models without fixed effects. The estimators are semiparametric; no parametric form for the disturbances has to be assumed. Heteroskedasticity across individuals is permitted. Consider the case of two time periods. The data is assumed to be generated as transformations of unobserved latent variables Y1) and Y2) given by Yt ) s a q X t b q ´ t

for

t s 1,2,

where X 1 and X 2 are K-dimensional vectors of explanatory variables, b is the parameter vector of interest, a is the fixed effect, and e 1 and e 2 are error terms. The econometrician observes Ž Yi t , X i t .: t s 1,2, i s 1, . . . , n4 where Yi t s max 0,Yi t) 4 , and Yi t) and X i t are distributed as given above. Honore´ shows that if e 1 and e 2 are i.i.d. conditional on Ž X 1 , X 2 , a ., then the distribution of Ž Y1) ,Y2) . conditional on Ž X 1 , X 2 . is symmetric around the 458-line through ŽŽ X 1 y X 2 . b ,0.. 46

AMaquila servicesB in the survey do not denote factories Žmaquiladoras. operating at the Mexican– U.S. border under special tax preferences.

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

25

This symmetry is used to propose orthogonality conditions that must hold at the true parameter values. The proposed estimators for b are then defined by the minimization of objective functions that have as first order conditions the sample analogs of these orthogonality conditions. The estimators are proved to be consistent and asymptotically normal. Monte Carlo results show that the asymptotic results give a good approximation of the small sample distribution if N G 200. Note that the i.i.d. assumption concerning the e ’s is not as restrictive as it may appear, since the fixed effect can capture some dependence between the error terms. References Arellano, M., Bond, S.R., 1988. Dynamic Panel Data Estimation using DPD—A Guide for Users. Institute for Fiscal Studies, London. Arellano, M., Bond, S.R., 1991. Some tests of specification for panel data: Monte Carlo evidence an application to employment equations. Review of Economic Studies 58, 277–297. Atiyas, I., 1992. Financial Reform and Investment Behavior in Korea: Evidence from Panel Data. Mimeo, The World Bank. Babatz, G., Conesa, A., 1997. The Effect of Financial Liberalization on the Capital Structure and Investment Decisions of Firms: Evidence from Mexican Panel Data. Mimeo. Bernanke, B.S., Gertler, M., 1989. Agency costs, net worth, and business fluctuations. American Economic Review 79, 15–31. Bernanke, B.S., Gertler, M., Gilchrist, S., 1996. The financial accelerator and the flight to quality. Review of Economics and Statistics 78, 1–15. Black, J., de Meza, D., Jeffreys, D., 1996. House prices, the supply of collateral and the enterprise economy. Economic Journal 106, 60–75. Bond, S.R., Elston, J., Mairesse, J., Mulkay, B., 1997. Financial Factors and Investment in Belgium, France, Germany and the UK: A Comparison Using Company Panel Data, NBER Working Paper No. 5900. Caballero, R.J., Leahy, J.V., 1996. Fixed Costs: The Demise of Marginal q. NBER Working Paper No. 5508. Campbell, J.R., Honore, ´ B.E., 1991. Pantob Instructions. Mimeo. ., 1991. Monografıa CANAME ŽCamara Nacional de Manufacturas Electricas del Sector de ´ ´ ´ Economica ´ Manufacturas Electricas, Mexico City. ´ Caprio Jr., G., Atiyas, I., Hanson, J.A., 1996. Financial Reform—Theory and Evidence. Cambridge Univ. Press, Cambridge; New York; Melbourne. Chan-Lau, J.A., Chen, Z., 1998. Financial crisis and credit crunch as a result of inefficient financial intermediation—with reference to the Asian financial crisis. IMF Working Paper 98r127. Chirinko, R.S., 1997. Finance constraints, liquidity, and investment spending: theoretical restrictions and international evidence. Journal of the Japanese and International Economies 11, 185–207. Dailami, M., Giugale, M., 1991. Reflections on Credit Policy in Developing Countries—Its Effect on Private Investment. World Bank Policy, Research and External Affairs Working Paper WPS 654. The World Bank, Washington. Dıaz ´ de Leon, ´ A., Schwartz, M.J., 1997. Crisis management and institutional change aimed at the prevention of future crises. In: von Furstenberg, G.M. ŽEd.., The Banking and Financial Structure in the Nafta Countries and Chile. Kluwer Academic Publishers, Boston. Edison, H.J., Luangaram, P., Miller, M., 1998. Asset Bubbles, Domino Effects and ‘Lifeboats’: Elements of the East Asian Crisis. International Finance Discussion Papers, No. 606, Board of Governors of the Federal Reserve System.

26

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

Fazzari, S.M., Hubbard, R.G., Petersen, B.C., 1988. Financing constraints and corporate investment. Brookings Paper on Economic Activity 1, 141–195. Fazzari, S.M., Hubbard, R.G., Petersen, B.C., 2000. Investment-cash flow sensitivities are useful: a comment on kaplan and zingales. Quarterly Journal of Economics 115, 695–706. Gale, D., Hellwig, M., 1985. Incentive-compatible debt contracts: the one-period problem. Review of Economic Studies 52, 647–663. Gelos, R.G., 1998. Fixed Investment in the Mexican Manufacturing Sector: Adjustment Costs, Credit Constraints, and the Effects of Financial Liberalization, PhD Thesis, Yale University. Gelos, R.G., Werner, A., 1999. Financial Liberalization, Credit Constraints, and Collateral: Investment in the Mexican Manufacturing Sector, IMF Working Paper 99r25. Gilchrist, S., Himmelberg, C.P., 1998. Investment, Fundamentals and Finance, NBER Working 6652. Glaessner, T.C., Oks, D., 1998. NAFTA, Capital Mobility, and Mexico’s Financial System. Policy Research Working Paper 1984. The World Bank, Washington. Grether, J.M., 1994. Trade Liberalization, Market Structure, and Performance in Mexican Manufacturing 1985–90. Mimeo. Gruben, W.C., McComb, R.P., 1997. Liberalization, Privatization, and Crash: Mexico’s Banking System in the 1990s. Economic Review, Federal Reserve Bank of Dallas. Harris, J.R., Schiantarelli, F., Siregar, M.G., 1994. The effect of financial liberalization on the capital structure and investment decisions of Indonesian manufacturing establishments. The World Bank Economic Review 8, 17–48. Holmstrom, ¨ B., 1993. Financing of investment in Eastern Europe: a theoretical perspective. Mimeo, Yale University. Holmstrom, ¨ B., Tirole, J., 1998. Financial intermediation, loanable funds, and the real sector. Quarterly Journal of Economics 108, 663–692. Honore, ´ B.E., 1992. Trimmed LAD and least square estimation of truncated and censored regression models with fixed effects. Econometrica 60, 533–565. Hoshi, T., Kashyap, A., Scharfstein, D., 1993. The choice between public and private debt: an examination of post-regulation corporate financing in Japan. NBER Working Paper 4421. Hubbard, R.G., 1998. Capital-market imperfections and investment. Journal of Economic Literature 36, 193–225. ˙Is¸can, T.B., 1998. Financing constraints and investment decline in Mexico. Manchester School 68, 24–43. Jaramillo, F., Schiantarelli, F., Weiss, A., 1997. Capital market imperfections, financial constraints and investment: econometric evidence from panel data for ecuador. Journal of Development Economics 51, 367–386. Kaplan, S., Zingales, L., 2000. Investment-cash flows sensitivities are not valid measures of financing constraints. Quarterly Journal of Economics 115, 707–712. Khamis, M., 1996. Credit and exchange-rate-based stabilization. IMF Working Paper 96r51. Kiyotaki, N., Moore, J., 1997. Credit cycles. Journal of Political Economy 105, 211–248. Krugman, 1998. Bubble, boom, crash: theoretical notes on Asia’s crisis. Mimeo, MIT. Mancera, M., 1997. Problems of bank soundness: mexico’s recent experience. In: Enoch, C., Green, J.H. ŽEds.., Banking Soundness and Monetary Policy. International Monetary Fund, Washington, DC. McKinnon, R.I., 1973. Money and Finance in Economic Development. Brookings Institution, Washington, DC. McKinnon, R.I., Pill, H., 1997. Credible economic liberalizations and overborrowing. American Economic Review, Papers and Proceedings of the Hundred and Ninth Meeting of the American Economic Association, 189–193. Mishkin, F.S., 1996. Understanding financial crises: a developing country perspective. NBER Working Paper 5600. Modigliani, F., Miller, M., 1958. The cost of capital corporate finance, and the theory of investment. American Economic Review 48, 261–297.

R.G. Gelos, A.M. Wernerr Journal of DeÕelopment Economics 67 (2002) 1–27

27

Nabi, Y., 1989. Investment in segmented capital markets. Quarterly Journal of Economics 104, 453–462. Powell, J., 1986. Symmetrically trimmed least squares estimation for tobit models. Econometrica 54, 1435–1460. Rama, M., 1993. Empirical investment equations for developing countries. In: Serven, L., Solimano, A. ŽEds.., Striving for Growth after Adjustment—The Role of Capital Formation. The World Bank Regional and Sectoral Studies, Washington, DC, pp. 107–146. Sargan, J.D., 1988. Testing for misspecification after estimating using instrumental variables. In: Maasoumi, E. ŽEd.., 1988. Contributions to Econometrics: John Denis Sargan, vol. 1. Cambridge Univ. Press, Cambridge; New York; Melbourne, pp. 213–235. Schaller, H., 1993. Asymmetric information, liquidity constraints, and canadian investment. Canadian Journal of Economics 26, 552–574. Schiantarelli, F., 1995. Financial constraints and investment: a critical review of methodological issues and international evidence. In: Peek, J., Rosengren, E.S. ŽEds.., Is Bank Lending Important for the Transmission of Monetary Policy. Federal Reserve Bank of Boston, Boston, pp. 177–214. Schneider, M., Tornell, A., 2000. Balance Sheet Effects, Bailout Guarantees and Financial Crises. NBER Working Paper 8060. Shaw, E., 1973. Financial Deepening in Economic Development. Oxford Univ. Press, New York. Stiglitz, J., Weiss, A., 1981. Credit rationing in markets with imperfect information. American Economic Review 71, 393–410. Townsend, R., 1979. Optimal contracts and competitive markets with costly state verification. Journal of Economic Theory 21, 265–293. Tybout, J.R., 1983. Credit rationing and the investment behavior in a developing country. Review of Economics and Statistics 24, 598–607. World Bank, 1994. Mexico Country Economic Memorandum—Fostering Private Sector Developments in the 1990s. Report No. 11823-ME.