Capital market imperfections before and after financial liberalization: An Euler equation approach to panel data for Ecuadorian firms

ELSEVIER

Journal of Development Economics Vol. 51 (1996) 367-386

JOURNAL OF Development ECONOMICS

Capital market imperfections before and after financial liberalization: An Euler equation approach to panel data for Ecuadorian firms Fidel J a r a m i l l o a Fabio Schiantarelli b,, A n d r e w W e i s s c Multipliea, Quita, Ecuador ~ Department qf Economics, Boston College, Chestnut Hill, MA 02167, USA " Boston Unicersity, Boston, MA, USA

Abstract Using a large panel of Ecuadorian firms, this paper analyzes the role of capital market imperfections for investment decisions, and investigates whether the financial reforms introduced in the 80s have succeeded in relaxing financial constraints. The model allows both for an increasing cost of borrowing, as the degree of leverage increases, and for a ceiling on the latter. The empirical results support the existence of significant capital market imperfections for small and young firms, but not for large and old firms. Moreover, the estimated equations do not provide evidence that financial reform in Ecuador has helped to relax financial constraints for small firms. JEL classification." E22; E44; O16 Kevwords: Financial liberalization; Capital markets; Investment

* Corresponding author. Tel.: (617) 552-3687; fax: (617) 552-2308: e-mail: SCHIANTA@ BCVMS.BC.EDU. I This paper was prepared under funding for the World Bank Research Project "Investment Decisions. Capital Market Imperfections, and the Effects of Financial Liberalization: the Ecuadorian and Indonesian Cases," RPO 676-72. A preliminary version was presented at the conference ~The Impact of Financial Reform," held at the World Bank, Washington DC, on April 2-3, t992. We are grateful to P. Beaudry, G. Caprio, J. Harris, and K. Lang fbr useful discussions. We have also greatly benefited from the comments by P. Honohan and by an anonymous referee. The views expressed here are solely the responsibility of the authors and should not be interpreted as reflecting those of the World Bank or of its shareholders. 0304-3878/96/$15.00 Copyright © 1996 Elsevier Science B.V. All rights reserved. PI1 S0304-3 878(96)t)0420-8

368

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

I. Introduction While there has been a growing literature and research effort in assessing the effect of liberalization on financial markets, less is known about the effects of these reforms on firms. The studies that exist are often descriptive in nature a n d / o r have concentrated on the macroeconomic picture. The purpose of this paper is to provide some evidence on the effects of financial liberalization in Ecuador by focusing on individual firms as the unit of analysis. In particular, we ask whether financial reform has relaxed the constraints faced by firms in obtaining external funds for investment. During the 1980s Ecuador introduced financial reforms to facilitate capital accumulation and growth. These reforms consisted mainly of the removal of administrative controls on interest rates, and the elimination or scaling down of directed credit programs. Our empirical investigation is based on a rich panel data set containing balance sheets and profit and losses statements for 420 Ecuadorian manufacturing companies during the period 1983-1988. The methodology developed here can, of course, be applied to analyze other programs of financial reform. In order to evaluate the effects of financial reform on firms' capital accumulation decisions, we must improve our understanding of the relationship between firms' financial and real choices, in the presence of imperfections in the capital markets. Such imperfections are due to the existence of administrative controls and regulations, and, at a deeper level, to asymmetric information and contract enforcement problems. A growing body of literature has examined how informational asymmetries may restrict access to outside funding and the implications thereof for a firm's investment decisions, even in the absence of administrative credit rationing. 2 Previous recent empirical work in this field has investigated whether investment models which do not allow for information asymmetries or incentive problems adequately describe the data, and whether the degree of misspecification varies across firms with different characteristics. That empirical work has used data from developed countries, for which panel data on firms were available, and it provides evidence against the perfect capital market paradigm. 3 One of the commonly used approaches is based on the estimation of the Euler equation for capital derived from a model with convex adjustment costs. The idea here is that the standard Euler equation will be misspecified for those firms that are financially constrained either because they have reached the (exogenous) maximum amount of debt allowed, or because they have exhausted retentions. A

2 At a theoretical level see the contributions by Stiglitz and Weiss (1981), Myers and Majluf (1984), and Bernanke and Gertler (1989). 3 Results and methods are reviewed in Schiantarelli (1995).

F. Jaramillo et aL /Journal of Development Economics 51 (1996) 367-386

369

correctly specified Euler equation would contain multipliers associated with the constraints, and their omission, abstracting from there would lead to misspecification. 4 The Euler equation approach is particularly appealing for developing countries (where stock markets are not well developed) because it does not require information on stock market values. Our formulation of the Euler equation departs from the perfect capital market paradigm for two reasons. First, we assume that agency problems can be thought of as giving rise to an increasing cost of borrowing above the safe rate, as the degree of leverage increases. Second, we introduce a ceiling on the m a x i m u m degree of leverage that firms are allowed. Using our panel of Ecuadorian firms we test whether these two forms of imperfection affect small and large (or young and old) firms differently, and whether they become less severe after financial liberalization. 5 The structure of the paper is as follows. In Section 2, 3, we develop the theoretical model for investment and we derive the appropriate Euler equation for the capital stock. In Section 4, we present econometric evidence on the importance of capital market imperfections for different categories of firms (small versus large, young versus old). In Section 5, we briefly describe financial reform in Ecuador, and we test whether it has relaxed financial constraints. We find that small firms were constrained both before and after financial liberalization, while we failed to find evidence that large firms were constrained in either period. Section 6 concludes the paper.

2. The model The thrust of the empirical analysis is to test for constraints that might be affecting the investment decisions of optimizing firms. The particular constraints that concern us are whether the firms face an upward sloping cost of funds, and

4 See Bond and Megir (1990), Whited (1992), and Hubbardet al. (1995), among others. 5 Most studies for developingcountries include various measures of aggregate credit as a regressor in (quasi) reduced form investmentequations, estimated using aggregate data for various LDCs (see Fry, 1988, for a review and Morisset, 1993, for a recent contribution).Exceptions are Tybout (1983), Nabi (1989), and Harris et al. (1994). Tybout (1983) uses three digits manufacturingindustry data for Colombia, partitionedbetween small and large firms, and tests whether the significanceof cash flow varies according to firms' size. Nabi (1989) uses a cross-section of small Pakistani firms and investigates how firms' characteristics affect their access to the credit market and their investment behavior. Harris et al. (1994) provide econometric evidence, based on panel data for Indonesian establishments,on the differentialrole of financialvariables,like cash flow and the degree of leverage, across firms (small versus large, members of an industrial group versus non-members),and between the pre and post financial liberalizationperiods.

370

F. Jaramillo et al./ Journal of Development Economics 51 (1996) 367-386

whether they face limits on their access to new debt. We let the cost of debt depend on the ratio of debt to capital. Moreover, we assume that the debt to capital ratio cannot exceed an exogenously given ceiling. When this constraint is binding, it is as if the firm faces an infinite cost for new debt. In order to test for these constraints, we embed them in an optimizing investment model. The model we formulate includes corporate and personal tax considerations. The firm is assumed to face a downward sloping demand for its product and adjustment costs for capital, and its objective is to maximize the value of shares for existing shareholders, Wt. This implies the maximization of the (tax adjusted) present value of dividends, net of the dilution of property rights over the stream of dividends for present owners due to new equity issues. The maximization problem is subject to non-negativity constraints on dividend payments, new share issues and debt, and to a ceiling on leverage, and can be written as: max Wt =

E t E/3/[ j= 0

~lt+jDt+j - stn+j]

(1)

subject to the following constraints: D r = (1 - " r , ) [ p t ( F ( K , , N t ) -A(Bt

8, z o , Bt

1

1,p~_lKt_l)] + ( B , - B t - , ) - P ~ I t + S'/ + C t

Kt=(1-6)K,_,

M,- pS

- G(K,,It) ) - wtN , -itB,_

+I t

(3)

s;'>_o

(4)

>- o

J where: / 3 / = 1-'I(1

+

(2)

R,+i/(l

(5

- z,+i))-1,

Tt = (1 -

m r ) l ( 1 - zt) and R, denotes

i=0

the required rate of return, Y, the tax price of dividends in terms of retentions, z, the corporate tax rate, m t the tax rate on dividends, zt the tax rate on capital gains, Pt the output price, K t the capital stock, N t labor, /t investment, w t the wage rate, i, the riskless interest rate, B, the stock of debt, p k the price of investment goods, C, the tax savings associated with depreciation allowances on existing capital goods, and 6 the depreciation rate. F ( K , , N t) denotes the gross value added production function and G ( K t , I t) is the additively separable adjustment cost function, with Gz > 0 and G H > O. We have included in the maximand an agency/financial distress cost function A ( B t 1, PtK- I K t - ~) which captures the premium paid by firms above the safe rate. The premium on external borrowing reflects the asymmetry of information be-

F. Jaramillo et al. / Journal of Det,elopment Economics 51 (1996) 367 386

371

tween the borrower and the lender, the difficulty of enforcing contracts, and dead weight losses to the lenders from bankruptcy. In these circumstances there is a potential conflict between the interests of the two parties which may result in the lender imposing on the borrower a cost of funds over and above the safe rate, and possibly other restrictions limiting borrowers' discretionary actions. These problems are undoubtedly more severe in LDC than in countries such as the US, Japan, and the UK, for which capital market imperfections have been found by previous researchers. For instance, there is a scarcity in LDC of sophisticated monitoring institutions that play a role similar to that of credit rating and auditing agencies in the US. Moreover, the nature of the legal system may be such that it is more difficult to enforce contracts. Since the greater the amount of debt in a firm's financial structure, the more severe the incentive problem becomes, we assume that agency costs increase with the stock of debt (A B > 0) at an increasing rate (A~,~ > 0). We also assume that agency costs are a decreasing function of the amount of collateralizable assets, represented here by the capital stock (A x > 0). Besides this premium on debt, we have allowed for another type of capital market imperfection, namely we assume that there is an upper limit, M t, to the debt to capital ratio that lenders consider acceptable. 6 Mt can vary over time and across firms. 7 It is well known that in this model a firm will not issue new shares and pay dividends simultaneously. Consider the case in which the firm issues debt but no new shares, and dividend payments are strictly positive in period t and t + 1. Assume also that the demand elasticity, e, is constant. The Euler equation for capital and the first order condition for debt can be written as: e-I ,ff

_( (1 -/xt)p[

(1-txt+')p[+'

(1- T,)p, /3,'+,y1+,(1

+ E,

(1- z,+,)

e , r , ( 1 - ~,)

A t"- -Rt

A K %(1

-

"q)PtPtk K23

(6)

6 Another type of capital market imperfectionthat can be easily incorporated into the maximization problem is the existence of a premium on new equity. The premium can be generated by an adverse selection argument, whereby only firms which are overvalued have an incentive to issue new shares (Myers and Majluf, 1984). In this paper, however, we will abstract from the lemon premium on new share issues and we will focus on the agency cost of debt. 7 In order to keep the notation simple, we do not include in the tbrmulae an index denoting firms.

372

F. Jaramilloet al. / Journal of Development Economics51 (1996)367-386 % - Et[ flt°+,yt+,(1 + (1

-

Tt+l)it+l) ]

- E t [ flt+l'Yt+l(1- Tt+I)ABt]

At p~K,

0

(7)

where ~bt+ 1 = (~/t+ 1 ( 1 - - "rt+ 1)P,+ 1)/(7,( 1 -- ~',)P,), and /z, is the present value of tax savings associated with a sucre of new investment. Eq. (6) is the appropriate Euler equation when the firm finances its investment at the margin through retentions and new debt. Note that Eq. (6) contains on the right-hand side the unobservable multiplier A~' associated with the ceiling in the debt to capital ratio. To gain some intuition about Eq. (6) it is helpful to ignore adjustment costs. In that case the left hand side of Eq. (6) equals the marginal revenue product of capital, measured in units of output, which is assumed to be a decreasing function of capital. 8 The fight hand side of Eq. (6) represents the overall costs of financing an additional unit of capital. The first term is the standard Jorgenson's user cost of capital. The second to last term in brackets represents the change in future agency costs from an increase in the capital stock. If increases in capital have a negative effect on agency costs, then firms will increase investment as a means of reducing future agency costs. The last term in Eq. (6) captures the effect of the debt ceiling on investment. The more capital a firm has, the more it can borrow given any ceiling on the debt to capital ratio. If the ceiling on the debt to capital ratio is binding, and if debt is cheaper than retentions for tax reasons, an increase in the availability of debt finance will reduce the cost of capital and increase investment. This effect is dampened, although not reversed, if borrowing costs are close to the cost of financing investment through retentions, i.e. if At is small. To interpret Eq. (7), start by assuming that the ceiling on leverage is not binding (A t = 0). In this case Eq. (7) simply states that the benefit of issuing one unit of new debt to finance dividend payments (Tt, the first term in Eq. (7)) must equal the present value of the cost of repaying that debt plus interest (the second and third terms in Eq. (7)). When the ceiling is binding, benefits exceed cost. If the firms uses new equity to finance investment at the margin (together with debt), it can be shown that the Euler equation for capital would be almost identical to Eq. (6) with Yt and "Yt+l set equal to one. However, it is also possible, for a film with good investment opportunities, that at time t or t + 1 retentions are all used up (dividends are zero) and yet it is unprofitable to issue new shares. At the margin, investment will be financed only through debt. If this is the case at time t

8 Where adjustment costs are present, the left hand side of Eq. (6) represents the real gross marginal revenue product of capital, minus adjustment costs at time t, plus the saving for having to incur less adjustment costs at time t + 1, in order to obtain the same level of capital from t + 2 onward.

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

373

or at t + 1, additional unobservable non-negativity multipliers for dividends will appear in Eq. (6) and Eq. (7), and by setting them to zero the researcher introduces a misspecification in the Euler equation (see, for instance, Bond and Megir, 1990; Whited, 1992; Hubbard et al., 1995). 9 Although the dividend floor constraint may be empirically important, in this paper we assume that dividend payments are positive, and we concentrate on the imperfection related to the cost and availability of debt, which are of great importance in the context of a developing country. This choice is also motivated by the fact that our data set on Ecuadorian firms does not contain direct information on dividend payments.

3. From theory to testing: parametrizing the Euler equation for capital Consider the case in which the firm pays dividends, issues debt, and the ceiling on the degree of leverage is not binding. In this situation, since A t ' = 0, no unobservable multiplier will appear in Eq, (6). If we assume that the agency cost function is homogeneous of degree one in debt and in the capital stock, Eq. (7) will determine the optimal debt to capital ratio for each level of the discount factor, /3t°+~. To estimate Eq. (6) it is necessary to parametrize the agency cost function. Assume that: c

A(Bt

l,ptk_lKt_l)

82

2 ptk_lKt_l

(s)

where c > 0. This specification implies that we can think of the interest rate on debt issued at the beginning of period t to be equal to the safe rate i t, plus a premium per unit of debt that is linear in the degree of leverage, (cBt 1 ) / ( 2 P ~ - i K t - 1). In order to obtain an equation that can be estimated, we assume that the gross value added production function, F ( K t , IVt) , and the adjustment cost function, G(Kt, It), are homogeneous of degree one in their arguments, and that the adjustment cost function can be written as:

c(K,,1,)

-

2 Kt

(9)

where b > 0. Using Eqs. (8) and (9) in Eq. (6), together with the linear homogene-

9 More precisely, denote with A~ the multiplier associated with the non-negativity constraint for dividends. Yt should then be replaced by Yt + A° and Yt+ 1 by %+ 1+ AID+I in the definition of q6 in Eq. (6), and in Eq. (7).

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

374

ity assumption, and replacing expected values by actual values, we obtain the following equation: ~0

I,+, i/it+ I ~t°+ l(l

=al

-- (~) - -

Kt+ 1

I, K,

¢ ~-

K2

( 1 - / x , ) pt*

- 8)

(1 - T,)p,

(1 - / . i t + l)pt*+, (1-

Tr+,)p,+,

- ~

+ V,+l

7r,

ptK t

(lO)

% ( 1 - r , ) ( p t K , ) p, where: c~l=

1

b(e-1)'

a2=

b(e-1) ;

-c~ a3-

2b(e-1)

¢r, represents nominal operating profits (value added minus wage bill), Y, output, and vt+ l a serially uncorrelated forecast error. Note that Eq. (10) is exactly identified so we can recover the structural parameters of the model. We are particularly interested in the slope of the agency cost function, c. Profit maximization implies e > 1 and hence a 1 > 0, a 2 > 0, and a 3 > 0. Note also that if there is perfect competition, i.e. e ~ o ~ , then a l = l/b, c ~ 2 = 0 , and o r 3 = - c / 2 b . From Eq. (10) we can see that the effect of investment on agency costs is proportional to the square of the debt to capital ratio. Hence the larger ( B / p f K , ) 2 is, the greater is the reduction in future agency costs from increasing the current capital stock. This reduction of the cost of financing investment is captured by the last term in Eq. (10) (recall that a 3 is negative). Eq. (10) is incorrectly specified if the firm hits a binding ceiling on its debt to capital ratio. When the ceiling for the debt to capital ratio is reached, an additional variable containing the unobservable multiplier A~,~ should appear in the Euler equation (see Eq. (6)). However, even if the ceiling is binding, we can eliminate A'm using the first order condition for debt. If we continue to assume that the firm

l0 Note that linear h o m o g e n e i t y implies that F ( K t , N t ) - G ( I t, K t) = (F~: - Gtc)K t + F u N t - Gtlt. U s i n g the first order condition for labor and Eq. (9), we can obtain F x - G K = ( e ) / ( e - 1)~r t - l / ( e - I ) ( Y , ) / ( K , ) + b ( l ~ ) / ( K ~ ) , w h e r e 7r, are operating profits. Substituting this in Eq. (6) we obtain Eq. (10). Note that if the net p r o d u c t i o n function is h o m o g e n e o u s o f degree 1 + q~, the f o r m of the equation remains the same. H o w e v e r , n o w a z = ( l ) / ( b ( e l ) ) - ( q ~ ) / ( b ) and ~ and e cannot be separately identified.

F. Jaramilloet al./ Journalof DevelopmentEconomics51 (1996)367-386

375

pays dividends, and it is using a positive amount of debt, solving for A'," from Eq. (7) and substituting in Eq. (6) we can obtain: I, + ~ o ~,t+,/3,+,(l8) -x,+,

( (1 --/xt)pt k

'~

_

11

x,

b(Z-

l)

(1 : r,-~p,

1,2

-}-

x,-9

q,,+,/3L,(1

-

~)

(1 - h t t + ~) p,k+ i (l -

rt+l)p,+l

~ r , p ,k

p~K, Pt + b ( ~ - 1) ce 2b(¢-

+

1)

ce b(e-

{ y,+,(1-rt+,)fi,'+,B/p~ y , ( l - r,)(p,~K,)2p, y,+ j(1 -

1)

Yt(l _

o "~ k ) r,+i)tS,+lB,'p,

,r,)(pkKt)2p,

{ [~,,- t3;;,~,+,(l + (l - ~,+,)/,+,)] b(~--~ 1)

T,(1 - r,)

8,p, ~ P~KtPt

-}- 12t +

l

(ll) Comparing Eq. (10) with Eq. (1 1), we see that two additional regressors (the last two in Eq. (11)) now appear in the Euler equation. They are a linear and a quadratic term in the debt to capital ratio. The debt to capital ratio affects the cost of capital in three ways. First, the observed debt to capital ratio is the ceiling on the debt to capital ratio, since we are explicitly assuming that the constraint is binding. When capital increases, this relaxes the constraint both now and in the future. Assuming that debt is the preferred form of finance when firms can borrow at the safe rate, an increase in current capital decreases (future) financing costs. These direct effects are captured by the last term in braces. However, increases in the debt to capital ratio will itself decrease the cost advantage of debt finance, because the premium that firms have to pay above the safe rate increases. This dampening effect due to an increase in agency cost is captured by the second to last term containing B,/(p,~Kt) 2, that enters the equation with a positive coefficient. Given the parametrization chosen, this effect dominates the decrease in the premium for debt that follows the increase in capital stock. Consequently, the coefficient on B,/( p,x K,)- changes sign when comparing a regime in which the

t l If y, - ill+ 1%+ i( 1 + (1 - r,+ I))i,+ I > 0, debt is the preferred source of finance, when firms can borrow at the safe rate (see the last term on the right hand side of Eq. ( I I )).

376

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

debt to capital constraint is not binding with a regime in which it is. The equation we will estimate can be written as: t+l 0t+,/3,+,(1 - 6 ) - Kt+ 1

It It 2 + K t K~

(1--l~t)pkt =O~1 (1-

~t)p,

Kt

~ t + l ~t°+ 1( I -- 6 )

(1 - / ~ t + l) ptk+ 1

7r,

(1 - Tt+l)p, +l

ptK,

Yt(1-"ct)(PtKt) P,

+or4( ['Yt- fl:+lyt+'(l 1+- 7"t)

/ pktKtP )

+

(12)

where: e otl=

1

b(e-1)'

oz2=

b(e-1)

cs ;

o~3= 2 b ( , ~ - l ) '

E'

a4

- 1)

The importance and the type of capital market imperfections are reflected in the significance and signs of the coefficients in front of the leverage terms in Eq. (12). If a 3 > 0 and a 4 < 0, then this can be taken as evidence that the firm faces an increasing premium for external finance, and that it has also hit the ceiling constraint for leverage. Note also that the model implies that - a 4 = a l = e/[b(e- 1)]. If we impose this restriction, the model becomes just identified and we can recover the structural parameters. If a 3 < 0 and a 4 = 0, then we return to the model of Eq. (10), in which the interest rate increases with the debt to capital ratio, but the ceiling is not binding. Finally, if a 3 = 0 and a 4 = 0, then we are back to a model in which both types of financial constraints are absent. The specification that allows for a premium above the safe rate that is linear in the debt to capital ratio, and for a ceiling on the latter may be thought of as a conveniently simple approximation to a model without an actual ceiling, but with a premium on debt that increases at a rapidly increasing rate. These two models are fundamentally observationally equivalent. Both of them, however, differ from the one based on the linear premium and, most importantly, from the model obtained under the assumption of perfect capital markets.

4. Capital market imperfections and firms' heterogeneity: econometric results In this section, we will discuss the empirical evidence obtained from estimating various versions of the Euler equation. In the specification and estimation of the

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

377

various models, the error term is modeled as the sum of a firm-specific effect, a time effect common to all firms, and an idiosyncratic shock. To eliminate the firm-specific effect all the equations are estimated in first differences. Years dummies are also included in all specifications. To allow for the endogeneity of the regressors, we use the generalized method of moments. As a test of specification we also provide the Sargan-Hansen test of over-identifying restrictions. 1~ The model summarized in Eq. (12) (of which Eq. (10) is a special case) was estimated on a balanced panel of 420 firms in the manufacturing sector, covering the period 1983-1988. The panel is based on data collected by the Superintendent3 cia de Compafiias, and consist of balance sheets and profit and loss statements. In order to estimate Eq. (12), we must choose a proxy for the discount factor fit'+J. 14 We experimented with proxies based upon the interest rate on deposits, adjusted and unadjusted for a risk premium, and upon the lending rate. Results are not sensitive to the different measures of /3,+ 1. For the tax parameters we have used the statutory values. As we mentioned before, the significance of a 3 a n d / o r 4 represents deviations from the investment model under perfect capital markets. If a 3 < 0 and a 4 = 0, then the model would be consistent with imperfections related to agency/financial distress costs that make the borrowing rate depend upon the degree of leverage. If instead, a 3 > 0 and a 4 < 0, then in addition we expect to have binding credit constraints. However, even if capital market imperfections are pervasive, we need not expect that they are equally severe for all firms. It is reasonable to assume that small and young firms are more likely to be financially constrained. Lenders are likely to have more information about large and more established borrowers. Those borrowers also have more collateralizable wealth, which reduces the cost of default for the lender, and are likely to keep better records which helps the bank in monitoring the loan. Another reason why large firms may have less informational problems is that they are more likely to belong to industrial groups with bank associations. This association will decrease the problems of informational asymmetries in capital markets. In 42% of large firms in our sample either the owner of the company was a member of the board of directors of a bank or vice versa. None

12 See Arellano and Bond (1991) for the application of the GMM approach to panel data. We have used the DPD program for estimation (Arellano and Bond (1988)). 13 See Appendix A for further details. 14 Note that when the ceiling on the degree of leverage is not binding, we could obtain from the first order conditions for debt, Eq. (7), a relationship between the firm's discount factor, on the one hand, and the safe rate of interest and the debt to capital ratio, on the other. Assuming that the safe rate of interest and the tax parameters for the next period are known at time t, we could solve for fit'+ ~ and use the result in the Euler Eq. (10). We explored this option in the estimation of Eq. (10), but without much empirical success due to the non-linearities introduced in the problem. Moreover, when the ceiling constraint is binding, we cannot replace both flt°+t and Am in Eq. (12) in terms of observable variables (Am will appear in the definition of fl~'+l).

378

F. Jaramillo et al./Journal of Development Economics 51 (1996) 367-386

o f the small firms had either of these bank affiliations. 15 Size considerations may also affect the access to the directed credit programs at subsidized rates available in Ecuador, although it is difficult to say a priori which firms will benefit more. On the one hand, certain programs like the ones promoting exports, are likely to be more advantageous for large firms. The latter probably also enjoyed better political connections, which may be instrumental in obtaining access to directed credit. On the other hand, in Ecuador there were lines of credit that provided cheap access to long-term finance for small firms. Finally, the resources devoted to the provision of directed credit have actually decreased with the introduction of financial reform. W e will discuss at length the effect of financial liberalization in the next section. In order to test whether the likelihood and severity of financial constraints varies across firms, we have allowed the parameters that capture capital market imperfections to differ between small and large firms (or young and old ones). Summarizing, we find that the investment behavior of small firms is well characterized by a model in which interest costs are an increasing function of the debt to capital ratio and the constraint on their leverage is binding. W e fail to find evidence of either form of imperfections for large firms. In Table 1 we present estimates of the unrestricted version of Eq. (12) (the restriction - c~4 = c~ 1 has not been imposed) over the entire period. In column (A) we allow the coefficients of the financial variables, c~3 and c~4, to differ between small and large companies. We define large companies as those that have a value of capital stock (in machinery, plant, and equipment) greater than US$600000 at 1975 prices. They represent on average 22% of the total sample over the period. Note that we allow firms to transit between different categories by introducing an endogenous size dummy variable for the two categories and by interacting them with the last two financial terms in Eq. (12). The G M M estimation method we have adopted, that uses appropriate lagged variables as instruments, allows us to deal with the endogeneity of the size dummies and the other regressors. The coefficients of c~ 1 and a 2 that are functions of the adjustment cost parameter, b, and of the elasticity of demand, e, have the expected sign and are quite significant. ~6 The W a l d test of joint significance of all the coefficients on the leverage terms ( W 2 = 10.26 with four degrees of freedom), suggests a rejection of the model of perfect capital markets at the 5% significance level. A closer look at the results shows an interesting difference between large firms, on the one hand, and smaller firms, on the other. Note that the first number following the a s refers to the coefficient number, as defined in Eq. (12), and the second number refers to size (1 for large firms, 0 for small firms). Variables capturing firms' financial

,s This information was obtained as part of proprietary research on investment in business groups. 16We have allowed a l and c~2 (that are functions only of b and e) to differ across firms' categories. However, we cannot reject they are equal at the 1% significance level, using a Wald test.

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367 386

379

Table 1 Estimates of the Euler equation for capital, Eq. (12) (A) With size interactions dummies

(B) With age interactions dummies

(C) Partly restricted version of model in (A)

(D) Fully restricted version of model in (A)

Coeffi'cients o' 1 c~2 a3 o~4 o~31 o130 a41

0.205 (2.81) 0.038 (3.94)

0.218 (2.40) 0.039 (3.81)

-0.038 -(0.79) 0.022 (2.25) -0.835

0.012 (1.20) 0.022 (1.77) I).183

- (0.38)

a40

-0.884 - (1.25)

0.211 (2.84) 0.037 (3.55)

O. 165 (2.53) 0.035 (4.05)

0.028 (2.62)

0.013 (2.30)

(0.41 )

- 1.680 - (1.46)

- 1.046 - (1.50)

Parameters e

0.158 (7.46) 4.753 (8.99) 7.659 (6.70)

e b

Tests W1 W2 Sargan/Hansen

24.97 6 10.255 4 24.549 22

20.84 6 7.876 4 24.549 23

25.99 4 10.66 2 26.66 20

21.15 3 6.29 I 22.29 22

Instruments: 1/K, I / K 2, Y / K , L / K , ( L / K ) 2, DI(B/K), D I ( B / K ) 2, DO(B/K), DO(B/K) 2, time dummies and D1, all dated t - 2 in levels. L represents liquid assets; DI is the dummy lot large (old) firms, DO for small (young) firms. Asymptotic 't' statistics are shown in parenthesis. W l: Wald test /'or joint significance of all the regressors. W2: Wald test for joint significance of financial variables. The Sargan/Hansen test is distributed as a X 2. Degrees of fi-eedom are presented below the respective tests.

structure

are

not statistically significant

for large firms

either individually

or

jointly. F o r s m a l l f i r m s , t h e s i t u a t i o n is d i f f e r e n t . H e r e w e f i n d a s i g n i f i c a n t p o s i t i v e c o e f f i c i e n t o n t h e q u a d r a t i c t e r m ( o ~ 3 0 ) w h i c h is c o n s i s t e n t w i t h a m o d e l in w h i c h

380

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

firms face both credit constraints and rising costs of borrowing. The coefficient on the linear term ( a 4 0 ) is negative as expected in this case, although not very precisely determined. This suggest that smaller firms are more likely to be affected by informational problems and not only have to pay an increasing premium for debt, but are also rationed in the credit markets. The W a l d test on the joint significance of a 3 0 and a 4 0 suggests that the perfect capital market hypothesis can be rejected at the 2.5% significance level for small firms. 17 A similar picture results if we divide firms by age. In column (B) of Table 1 we present results allowing for different estimates of a 3 and a 4 for young firms and old firms. Young firms are defined as those born after 1970, and represent 47% of the sample. 18 The coefficients on the financial variables are more significant for young firms (c~30 and a 4 0 ) than for old firms (ce31 and a41). A general word of caution is necessary, however, in interpreting these results. Our model is based on the twin assumptions of the existence of a premium above the safe rate that is a linear function of the debt to capital ratio, and on the existence of a ceiling on the latter. 19 It is possible that our results suggest not an actual ceiling, but the fact that the premium increases at a rapidly increasing rate with the debt to capital ratio. The model with a linear premium and a ceiling fits well, under this interpretation, because it is a good approximation to such a highly non-linear relationship. Moreover, insofar as the ceiling on the degree of leverage can differ for every firm-year observation, the model with the ceiling is much less restrictive than compared to those based only on the existence of a linear or quadratic premium and no ceiling. Thus, it is not surprising that a flexible ceiling fits the data better. Whatever the interpretation, it is clear, however, that for small and young firms we can reject the model based on the absence of capital market imperfections. Since the coefficients in front of the financial variables for large firms (c~ 31 and c~41) are not significant, we set them both to zero, obtaining the model in column (C). Both the magnitude and statistical significance of the coefficients representing financial constraints for smaller firms (c~30 and c~40) increase, further indicating that smaller firms suffer from both sources of capital market imperfections. The W a l d test of joint significance of c~30 and c~40 (see W2) suggests that the perfect capital market model can be rejected at the 1% significance level for smaller firms. From Eq. (12) the coefficient in front of the profit term and the one in front of the (linear) leverage term should be equal in absolute

17 We have also re-estimated the model in column (A), using different proxies for/3t"~ t. in particular we have added a risk premium of 5% to the safe interest rate, either for small firms or for both small and large firms. The results are virtually identical. 18 In this case, in o~ij, j refers to age ( j = 1 for old firms, and j = 0 for young firms). 19We have also allowed for the premium to be a quadratic function of the degree of leverage. In this case a cubic term in the debt to capital ratio should be added to the Euler equation for capital. However, when this specification was estimated on our data, the cubic term was always insignificant.

F. Jaramillo et al. ~Journal of Development Economics 51 (1996) 367-386

381

Table 2 Euler equation for capital and the effect of financial reform

(A)

(B)

0.190 (2.42) 0.041 (3.78) 0.051 (2.41) 0.036 (2.49) - 1.884 -(1.49) - 1.862 -(1.70)

0.164 (2.09) 0.032 (3.52) 0.013 (1.51) 0.014 (2.54)

25.02 6 12.87 4 15.98 17

18.741 4 6.547 2 21.995 19

Coefficients oil o~2 a 30a a 30 b a40a a40b

Tests W1 W2

Sargan/Hansen

See footnotes to Table 1. In the definition of instruments we allow lbr the pre- and post-liberalization split for the debt to capital ratio.

value and opposite in sign ( - a 4 0 = a 1). This restriction is not rejected by the data on the basis of a Wald test. When we impose this restriction, we obtain the model in column (D), We can then recover the structural parameters b, e, and c. All the coefficients have the expected sign and are statistically significant. The Sargan/Hansen test does not provide evidence against the specification and the choice of instruments. The structural parameters of the model are presented in Table 2, and all of them have reasonable magnitudes. The size of the elasticity of demand, e, indicates that firms in the manufacturing sector have some monopolistic power. More precisely, an e of 4.75 implies a mark-up over marginal cost of 26.6%. We are not aware of any other study that estimates the elasticity of demand for manufacturers in Ecuador. However, the magnitude of this parameter is similar to the one obtained by estimating the Euler equation for capital for developed countries (see for instance Galeotti and Schiantarelli, 1991, for results using US data). Q type of investment models derived under the assumption of quadratic adjustment costs have often yielded very high values of the adjustment cost parameters that imply huge costs of changing the capital stock. The value of b obtained here is reasonable and it implies that adjustment costs are approximately

382

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

6.2% of total sales in manufacturing during the 1984-1988 period, still substantial, but not unrealistically so. 2o Finally, the slope coefficient of the marginal agency cost for small firms, c, implies that the average premium over the safe rate (calculated at the average debt to capital ratio) is 10.2%. 2~ To get an idea of the importance of the premium, note that in 1986, for example, the average nominal borrowing rate was 30.75% and the real one approximately 9%.

5. The effect of financial liberalization The importance of capital market imperfections for different types of firms has been highlighted and empirically tested in the last section. It is likely that the results we have obtained reflect both asymmetric information and contract enforcement problems and the importance of administrative controls on interest rates and on the allocation of credit. These controls were widely used through the first half of the 1980s. In this section we discuss whether financial liberalization has relaxed financial constraints for Ecuadorian firms. This was the presumption of the early literature on financial liberalization, which suggested that freeing the interest rates from controls that kept them artificially low would increase the supply of loanable funds, and alleviate problems of credit constraints. 22 Whatever the macro effect may be, it is not clear that liberalization will necessarily relax financial constraints for all classes of firms. Even after the elimination of administrative constraints, problems of imperfect information and imperfect controls remain and it is possible that certain firms may still face a rising premium for external finance. Moreover, one must pay attention to the impact for different classes of firms of the reduced importance of subsidized credit programs that accompanies financial reform. For instance, those small firms that were getting very little credit at low interest rates from subsidized credit programs before liberalization and were not borrowing on the free market, were likely to have been severely constrained prior to liberalization. If they then borrowed on the free market after liberalization the constraints were likely to have been relaxed. Small firms that were getting large amounts of subsidized credit are likely to have been unconstrained in the preliberalization regime. Whether a small firm was constrained or unconstrained before liberalization depends largely on whether the amount and price of subsidized credit available to that firm exceeded its demand

20 Note that adjustment costs equal (b/2)(lt2/Kt). If we divide by Y,, we get (b/2)(lt 2/ K t 2)(K t / Y,). This formula is used to calculate the size of adjustment costs reported in the text, using sample averages for the investment rate and the capital-output ratio. 21 The premium is equal to (cBt)/(2ptkKt). In the calculation we have used the average leverage ratio for small firms over the entire period. 22 See McKinnon (1973) and Shaw (1973).

F. Jaramillo et al./ Journal of Del,elopment Economics 51 (1996) 367-386

383

for credit at that price, as well as upon the availability and cost of credit to that firm in the tree market. Consequently, financial liberalization could decrease the severity of credit constraints for some firms while increasing their severity for others. It could even be the case that all small firms find their financial constraints increased (or decreased) after liberalization. In the context of our model, one way to assess the impact of liberalization is to analyze whether there is a structural change in the coefficients of the variable that capture the rising cost of funds schedule, and the tightness of the ceiling on the degree of leverage. However, financial reform in Ecuador has been a multi-stage process. This implies that there is some degree of arbitrariness in the choice of the breaking point of the sample to conduct tests of structural stability. A major change, however, occurred in August 1986 when all interest rates were completely treed and the real rate reached positive levels (of approximately 10% in 1986 and 1987) for the first time in many years. Serious attempts were instituted to begin removing subsidized credit programs. These included export credits under FOPEX (Fund for Export Promotion) and subsidized lines of credit for small firms like FOPINAR (Fund for Small Industry and Handicraft). Between 1985 and 1988 outlays by FOPINAR fell by roughly 44% in real terms and by roughly 50% as a fraction of GDP. We have experimented with classifying 1986 as a pre- or post-reform year. The empirical results are not sensitive to this choice and we present below the estimates for the latter case. In testing for structural change in the context of Eq. (12), if liberalization relaxes the ceiling constraint, then the linear term in the leverage ratio should not enter the equation (o~4a = 0). 23 Moreover, the coefficient on the quadratic term on the leverage ratio (c~3a) should switch sign from positive to negative if firms still have to pay a premium that increases in the debt to capital ratio. If liberalization helps in mitigating informational problems that cause an increase in the premium for external finance, then we would expect ~Y3a to be smaller in absolute value than c~3b, its value before liberalization. If both oL3a and cx4a are not significant, this would mean that the perfect market paradigm cannot be rejected after liberalization. In the previous section we have argued that capital market imperfections affect mainly small firms. A relevant question is then whether financial liberalization has improved their access to external finance. Results are presented in column (A) of Table 2 for the model that corresponds to column (C) in Table 1, but allows the coefficients on the financial variables to differ pre- and post-liberalization. The evidence suggests that small firms suffered from a ceiling constraint and an increasing cost of borrowing both before and after liberalization. Moreover, we

2! a relers to the value of the coefficients after liberalization, and h bet~)re liberalization.

384

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

cannot reject the hypothesis that the coefficients capturing capital market imperfections are identical in the two sub-periods at the 1% significance level. 24 Finally, if we impose the restriction that - a 4 0 b = - a 4 0 a = a l , as the theory suggests, we can again recover the structural parameters. The results for the restricted version of Eq. (12) are presented in column (B). The estimates in column (B) suggest that the slope of the marginal agency costs function for small firms, c, decreases slightly after liberalization (from 0.175 to 0.155). However, the difference is not statistically significant. Moreover, if we compare the average debt to capital ratio for small firms in the pre-liberalization years relative to the post-liberalization years, we observe that it decreases approximately by 11% (from 1.41 to 1.18), suggesting that if small firms face binding constraints in their access to debt both before and after liberalization (as the econometric results indicate), then financial reform did not increase the maximum amount of borrowing allowed for a given value of collateral. 2s The degree of leverage also decreased by 17% for large firms (from 1.16 to .86). Both responses were likely to be due to the increase in the cost of borrowing and the decrease in its availability, following a decrease in the aggregate supply of credit due to the reduction in the role of the government as a major supplier of funds (to be precise, our results imply that large firms were not at the borrowing ceiling in either period). Results presented here suggest that financial reform did not help in improving the access to credit markets or in reducing significantly the premium for external finance for small firms. This may be because of the existence in Ecuador in the pre-liberalization period of programs like FOPINAR, structured to enhance the access to finance for small firms. The elimination of these programs with financial liberalization is likely to have tightened credit constraints for some of the firms that benefited from them. Moreover, the evolution of the macroeconomic situation in Ecuador may help in explaining our results. In fact, the increase in the real interest rate in Ecuador was interrupted by a severe inflationary episode in 1988, following a major earthquake and fiscal mismanagement. Finally, the second part of the 80s saw a reduction in the funds provided by the Central Bank to the commercial banks (to support directed credit programs) that was not fully compensated by the increase in financial savings during the years of positive interest rate. As a result, the real supply of credit to the private sector decreased in real terms between 1986 and 1988, and it is likely that this decrease has influenced the effect of financial reform on different categories of firms.

24 We obtain very similar results when we partition firms according to age instead of size. 25 The higher value of the debt to capital ratio for small firms in the pre-liberalization period can partly be explained by the existence of subsidized credit programs they could benefit from. After financial reform, their leverage decreases, but remains higher than for large firms.

F. Jaramillo et al./Journal of Development Economics 51 (1996) 367-386

385

6. Conclusions The estimation of Euler equations for the capital stock suggests that capital market imperfections have different effects for different firms, both pre- and post-liberalization. In particular, the model that allows both for a premium above the safe interest rate that increases linearly with the debt to capital ratio, and for a ceiling on the latter, appears to fit the situation faced by small (young) firms. As we have discussed above, this specification may fit well because it is a good approximation to a model in which the premium or debt increases at a rapidly increasing rate with the degree of leverage, even if there is no ceiling on the latter. There is no evidence that large (old) firms encounter significant capital market imperfections. These results should not be interpreted as supporting government programs of directed credit. Although informational asymmetries hinder the allocation of credit in a free market, there are other well known problems associated with directed credit. To choose between these two systems we need to directly compare them. but this goes beyond the objective of this paper. Our econometric results are not supportive of a substantial regime change after liberalization. Financial reform in Ecuador does not seem to have had much effect on the financial constraints faced by small firms when making investment decisions. The continuing presence of informational imperfections, and the decrease in resources devoted to promoting cheap access to credit for small firms after liberalization are part of the explanation, as well as the lack of expansion of the real supply of credit to the private sector. The main limitation of our investigation is that we do not have data for enough years before and after the implementation of financial liberalization to allow us to fully assess its effects, especially in the longer term. Further work is needed using longer panels, and looking at the experience of different countries.

Appendix A. Data The econometric estimation was based on a balanced panel of 420 firms in the manufacturing sector for which there was continuous information for sales, capital stock, value added, profits, and other key variables for the study. Firms selected had to satisfy standard consistency checks and additional criteria, including non-negative capital stock or value added. The panel is based on information collected by the "Superintendencia de Companias" (SC) of Ecuador. SC is an official agency that controls corporate activities. The data includes yearly balance sheet and profit and losses information for the period 1984-1988. The balance sheets also include information of the revaluation of assets allowed by the government to account for inflation and exchange rate depreciation. The definition of investment includes plant and machinery, buildings, and others (excluding land), and has been obtained by taking the difference between the gross capital

386

F. Jaramillo et al. / Journal of Development Economics 51 (1996) 367-386

stock at historical cost. The capital stock measure used in the equations is instead the revalued measures of the net capital stock. Our measure of debt includes both long-term and short-term debt, mainly obtained from banks and other financial institutions. Trade debt is not included. As a proxy for the output price we have used the two-digit wholesale price index. As a proxy for the price of investment goods, we have used the aggregate investment deflator. Both series are regularly published by the Banco Central del Ecuador. The statutory tax rates and depreciation allowances were used in calculating % and the present value of tax savings associated with a unit of new investment, /z t. The personal tax rate, m t, has been set equal to 20% and the effective rate on capital gains, z t , equal to zero.

References Arellano, M. and S. Bond, 1988, Dynamic panel data estimation using DPD: a guide for users, Institute for Fiscal Studies Working Paper No. 88/15. Arellano, M. and S. Bond, 1991, Some tests of specification for panel data: Monte Carlo evidence and an application to employment situations, Review of Economic Studies 58, 277-297. Bemanke, B. and M. Gertler, 1989, Agency costs, net worth and business fluctuations, American Economic Review 79(1), 31-41. Bond, S. and C. Megir, 1990, Dynamic investment models and the firm's financial policy, The Review of Economic Studies 61, 197-272. Fry, M.W., 1988, Money, interest, and banking in economic development (The Johns Hopkins University Press, Baltimore). Galeotti, M. and F. Schiantarelli, 1991, Variable markups in a model with adjustments costs: econometric evidence for U.S. industry, C.V. Start Center for Applied Economics W.P. No. 91-44. Hams, J., F. Schiantarelli and M. Siregar, 1994, The effect of financial liberalization on the capital structure and investment decisions of Indonesian manufacturing establishments, The World Bank Economic Review 8, 17-47. Hubbard, G.R., A. Kashyap and T.M. Whited, 1995, Internal finance and firm investment, Journal of Money, Credit, and Banking, forthcoming. McKinnon, R.I., 1973, Money and capital in economic development (The Brookings Institution, Washington, DC). Morisset, J., 1993, Does financial liberalization really improve private investment in developing countries?, Journal of Development Economics 40, 133-150. Myers, S.C. and N.S. Majluf, 1984, Corporate financing and investment decisions when firms have information that investors do not have, Journal of Financial Economics 13, 1417-1426. Nabi, I., 1989, Investment in segmented capital markets, Quarterly Journal of Economics 54, 453-462. Schiantarelli, F,, 1995, Financial constraints and investment: a critical review of methodological issues and international evidence, in "Is Bank Lending Important for the Transmission of Monetary Policy?", edited by J. Peek and E.S. Rosengren, Federal Reserve Bank of Boston. Shaw, E.S., 1973, Financial deepening in economic development (Oxford University Press, New York). Stiglitz, J.E. and A. Weiss, 1981, Credit rationing in markets with imperfect information, American Economic Review 71,393-410. Tybout, J.R., 1983, Credit rationing and investment behavior in a developing country, Review of Economics and Statistics 65, 598-607. Whited, T.M., 1992, Debt, liquidity constraints, and corporate investment: evidence from panel data, Journal of Finance (September), 1425-1460.