POLICY RESEARCH WORKING PAPER 2694
Financial Development Microeconomic evidence
from 40 countries shows that
and Financing Constraints financial development aids
growth by reducing financing
International Evidence from constraints that would
otherwise restrict efficient firm
the Structural Investment Model investment.
Inessa Love
The World Bank
Development Research Group
Finance
October 2001
POLICY RESEARCH WORKING PAPER 2694
Summary findings
The relationship between the financial and real sides of 40 countries. The results show a strong negative
the economy has long been a topic of intense interest and relationship between the extent of financial market
debate. Love provides microeconomic evidence that development and the sensitivity of investment to the
financial development aids growth by reducing financing availability of internal funds (a proxy for financing
constraints that would otherwise restrict efficient firm constraints).
investment. Considering size effects, business cycles, and the legal
The author estimates a structural model based on the environment as plausible alternative explanations, the
Euler equation for investment using firm-level data from author finds the results to be robust in all cases.
This paper-a product of Finance, Development Research Group-is part of a larger effort in the group to study the
determinants of access to finance. Copies of the paper are available free from the World Bank, 1818 H Street NW,
Washington, DC 20433. Please contact Kari Labrie, room MC3-456, telephone 202-473-1001, fax 202-522-1155, email
address klabrie@worldbank.org. Policy ResearchWorkingPapers are also posted on the Web athttp://econ.worldbank.org.
The author may be contacted at ilove@worldbank.org. October 2001. (49 pages)
he Policy Research Working Paper Series disseminates the R hndings of work in progress to encourage the exchange of ideas about
development issues. An objective of the series Is to get the findings out qluickly, even if the presentations are less than fully polished. The
papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the autbors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the
countries they represent.
Produced by the Policy Research Dissemination Center
Financial Development and Financing Constraints:
International Evidence from the Structural
Investment Model.
Inessa Love*
JEL codes: G20, G31, 012, 016
Keywords: investment, financing constraints, financial development
*The World Bank, e-mail: ilove()worldbank.org. The views presented here axe the author's own
and not necessarily those of the World Bank or its member countries. I am grateful to Geert
Bekaert, Laarni Bulan, Charles Calomiris, Raymond Fisman, Ann Harrison, Charles Himmelberg,
Robert Hodrick, Glenn Hubbard, Margaret McMillaa, and Xavier Sala-i-Maxtin and Toni Whited
for helpful comments and discussions. This research was supported by a fellowship from the Social
Science Research Council Program in Applied Economics with funds provided by the John D. and
Catherine T. MacArthur Foundation.
1 Introduction
The relationship between the financial and real sides of the economy has long been a
topic of intense interest and debate. The potential importance of the financial sector
in promoting economic growth was recognized as early as Schumpeter (1912), though
this perspective was disputed by numerous economists over the decades that followed
(most notably by Lucas (1988)). Proper empirical work in assessing the relationship
between financial development and the real economy began only much later, with
the work of King and Levine (1993a,b) who reported cross-country evidence which
suggested that financial development affects economic growth by fostering produc-
tivity improvements. Since then, there has developed a large and growing literature
that examines the relationship between financial market development and various
economic outcomes, primarily utilizing the cross-country data and methodology pio-
neered by LLSV'(1997, 1998, 2000). LLSV argue that the development of financial
markets depends on a country's legal origin, which is largely exogenous to the coun-
try's future economic growth.2 However, this body of work is based almost entirelyon
cross-country analyses which always raises serious concerns about unobserved hetero-
geneity across data points. Furthermore, these country-level studies cannot properly
examine the channels through which finance affects growth, as this requires the micro-
level analysis of firm behavior.
The micro-level examination of the link between real and financial decisions of
firms has seen considerable work since the pioneering contribution of Modigliani and
Miller (1958), who showed that in a world of perfect capital markets, finance is irrele-
vant for real decisions. This view has been amended and disputed by richer theoretical
models, and empirical studies that have found a strong relationship between firms'
financial health and investment (see Hubbard (1998) for a recent survey). These
'Rafael La Porta, Florencio Lopez-de-Silanes, Andrey Shleifer, and Robert Vishny
2See Levine and Zervos (1998), Levine (1999) and Beck, Levine and Loyaza.(2000) for recent
cross-country studies. Several recent studies use time-series analysis, for example, Neusser and
Kugler (1998) and Rousseau and Wachtel (1998) address the causality issues, and Bekaert, Harvey
and Lundblad (2000) look at the effect of financial liberalization on growth.
2
financing constraints are generally attributed to capital market imperfections, stem-
ming from such factors as asymmetric information and incentive problems, which
result in differences between the costs of internal and external financing.
The results contained in this paper lie at the intersection of these two broad liter-
atures. Utilizing firm-level data, while taking advantage of cross-country variation in
financial market development, I show that financing constraints, measured by the sen-
sitivity of investment to internal funds, decrease with financial development. These
findings are robust to a wide variety of specifications, and to the consideration of a
range of alternative explanations. I also report a number of ancillary results that
provide further evidence on the importance of financial market development for firm
investment. In particular, I find that small firms are disproportionately more disad-
vantaged in less financially developed countries than are large firms, i.e. they have
relatively larger sensitivity of investment to availability of internal funds. Together,
these results provide a micro-level foundation for one of the commonly cited explana-
tions for the observed cross-country relationship between financial development and
economic growth. Namely, I provide evidence that an improvement in the function-
ing of financial markets will reduce firms' financing constraints. This will allow for
easier access to external funds for firms with good investment opportunities and this
improvement in capital allocation will in turn enhance growth.
The methodology used in this paper is based on the established literature on in-
vestment with financing constraints, which began with the work of Fazzari, Hubbard
and Peterson (1988). The first papers in this field, based on the Q-theory of invest-
ment, were based upon models that contained a number of very strong assumptions,
such as constant returns to scale, perfect competition, and perfect capital markets.
The assumption of perfect capital markets is particularly problematic for my paper,
as I explicitly assume that capital markets of the countries in my sample are at dif-
ferent levels of development, and therefore cannot be considered "perfect." I adopt
the Euler equation methodology, utilized by more recent contributions to the financ-
ing constraints literature, which has less restrictive assumptions than the previous
3
generation of models, including a relaxation of the perfect capital markets assump-
tion.3 The advantage of using the Euler equation methodology is that it explicitly
controls for growth opportunities captured by the marginal product of capital. In
this framework, the sensitivity of investment to the level of internal funds is inter-
preted as evidence of financing constraints. This sensitivity is allowed to vary with
the country-specific level of financial development using the interaction of the finan-
cial development index and a firm-level measure of internal funds. This interaction is
shown to be significantly negative, which implies that financial development reduces
financing constraints.
This paper builds upon several recent studies that similarly address issues on the
role of the financial system in stimulating economic growth using micro-data. The
work that is closest in spirit and methodology to that of my paper is Demirguc-Kunt
and Maksimovic (1998), which is the only other firm-level study that examines the
link between financial development and growth. In their paper, the authors first cal-
culate the proportion of firms in a country that were growing faster than they could
have using only internally generated funds. They find that this proportion is posi-
tively related to financial development and to legal system indicators. Although this
finding clearly suggests that more developed financial markets improve the availabil-
ity of external finance in the aggregate, it does not have any bearing on the issue
of allocation of capital within a country, as this would require identifying firms that
"should" be growing, given their investment opportunities. I am able to address this
issue by using a structural model which explicitly controls for growth opportunities
at the firm level.
3Euler equations for Investment have been estimated by numerous authors, with most studies
concentrating on US firms. See Whited (1992), Hubbard and Kashyap (1992), Hubbard, Kashyap
and Whited (1995), and Calomiris and Hubbard (1995) among others. The limited work utilizing
international data includes Bond and Meghir (1994) for the UK; Jaramilo et al. (1996) for Ecuador;
Harris, Schiantarelli, and Siregar (1994) for Indonesia; and Gelos and Werner (1999) for Mexico. The
only paper that estimated the Euler equation for several countries is by Bond et al. (1996), which
includes observations from Belgium, FRance, Germany and the UK. A related paper by Kadapakkam
et al. (1998) studied investment in six developed countries. They used a reduced form approach
combining Q-theory and sales "accelerator" together with cash flow and cash stock measures. Also,
they did not compare financing constraints across countries.
4
Rajan and Zingales (1998) use industry level data to show that industries that
require more external finance grow faster in more developed capital markets. Thus,
they claim that financial development affects growth by reducing the differential cost
of external finance. While their work is very innovative and yields a number of
interesting findings, it is based on several strong assumptions. One particularly strong
assumption that is implicit in their analyses is that growth opportunities are the same
for a given industry in all countries. That is, if an industry is not growing at the same
rate as it is in other countries, it is a failure of the financial markets. In other words,
the authors do not attempt to control for the growth opportunities available for each
industry at every point of time in each country. Again, the structural approach
adopted here allows me to address this issue and to explicitly control for such growth
opportunities.
Finally, Wurgler (2000) finds that financial development improves capital alloca-
tion by increasing the industry-level sensitivity of investment growth to value added
growth. Wurgler points out two reasons why firms in less developed financial markets
might not undertake the most profitable projects (and thus worsen capital alloca-
tion). First, insiders might not be able to distinguish good investment opportunities
due to the lack of information. In support of this proposition, he finds that more
firm-specific information in returns increases the sensitivity of investment to value
added growth. Second, insiders might not have incentives to undertake the most
profitable investments if their profits are expropriated; he provides evidence for this
hypothesis by showing that state ownership is associated with lower sensitivity and
minority rights with higher sensitivity.
Although my study is in a similar spirit to these previous papers, it improves
upon them in a number of ways. As noted above, since I use structural model, in
the form of investment Euler equations, I am able to control for future growth oppor-
tunities by explicitly including the marginal productivity of investment (a measure
of growth opportunities). Also, the model identifies the information set available at
each decision-making point, which allows for the specification of a valid instrument set
5
and the use of an appropriate estimation technique. In addition, the model allows for
the interpretation of estimated coefficients as structural parameters, which provides
an additional check on the plausibility of my results. iurthermore, by using firm-
level data, rather than industry-level aggregates, I exploit firm heterogeneity in the
productivity of capital. Since some firms will be more productive than others within
the same industry, allocating capital to the industry as a whole is not as efficient as
allocating capital to the most productive firms within each industry.
The rest of the paper is as follows. Section 2 presents the structural investment
model based on a dynamic optimization problem and discusses financing constraints.
Section 3 discusses the empirical model and estimation methodology, and Section
4 describes the data. Section 5 provides the main results including the analysis of
structural parameters. Section 6 presents several tests of alternative explanations,
including the size effect, business cycles and legal system. Section 7 presents the aux-
iliary results using single-country regressions. Finally, Section 8 provides conclusions
and directions for future research.
2 The Model of Investment
2.1 The Optimization Problem
The dynamic model of the firm value optimization is reproduced in this section. This
model is similar to models used in previous studies (listed in footnote 3), and follows
closely the specification in Gilchrist and Himmelberg (1998). The model is simplified
here because it ignores the possibility of debt financing. However, this simplification
does not affect the resulting first order conditions for investment, which are the focus
of this paper.4 In this model shareholders (or managers) are maximizing the present
value of the firm, which is equal to the expected discounted value of dividends subject
'Formally including debt into the problem results in a separate Euler equation for debt, see
Gilchrist and Himmelberg (1998) for derivation. However, the investment Euler equation is not
directly related to the debt Euler equation and is not affected by adding debt into the model.
6
to the capital accumulation and external financing constraints. The firm value is given
by:
Vt(Kt, (t) max Dt + Et Ot+s-Dt+s (1)
subject to
Dt= Il(Kt, t) - C(It, Kt) - It (2)
Kt+1 = (I1- 6)Kt +.It (3)
Dt > ° (4)
Here Dt is the dividend paid to shareholders and is given by the "sources equal uses"
constraint (2); Ot+,-,, is a discount factor from the period t + s to period t. In the
capital accumulation constraint (3) Kt is the beginning of the period capital stock,
It is the investment expenditure and 6 is the depreciation rate. The restricted profit
function (i.e. it is already maximized with respect to variable costs) is denoted by
[I(Kt, (4), where (t is a productivity shock.5 The adjustment cost of investment is
given by the function C(It, Kt), and is assumed to result in a loss of a portion of
investment. The financial frictions are introduced via a non-negativity constraint
on dividends (4), and the multiplier on this constraint is denoted At below. This
multiplier equals to the shadow cost associated with raising new equity, which implies
that external (equity) financing is costly and this extra cost is due to information or
'The profit function depends on the beginning of the period capital, and hence the implicit
assumption is that investment becomes productive only in the next period (i.e., a one period time
to build lag). I ignore the price of invesment which is replaced by fixed and time effects in the
estimation. I also ignore tax considerations due to data constraints.
7
contracting costs.6 This shadow cost is used in defining financing constraints, which
are discussed below.
2.2 The Euler Equation
The Euler equation derived from the above maximization problem (derivations are
available from the author) is given by:
1± (00) =tEt[et {(t ( +(1- ) ( ( )+) (5)
Here, a9C is the marginal adjustment cost of investment, '9,1 is the marginal "profit"
of capital, further referred as MPK, (the contribution of an extra unit of capital to
the firm's profits), and et = (t+i is the relative shadow cost of external finance
in periods t and t + 1. I refer to the factor et as "financing constraints" and discuss
it in a separate section below. The intuition behind this Euler equation is that the
marginal cost of investing today on the left hand side (given by the adjustment cost
and the price of investment goods, normalized to one) is equal to the discounted
marginal cost of postponing investment until tomorrow, on the right hand side. The
latter is equal to the sum of the foregone marginal benefit of an extra unit in capital,
given by MPK, plus the adjustment cost and price of investment tomorrow (again
normalized to one).
To arrive at the empirical model, one must identify empirical measures for financ-
ing constraints and MPK, specify a functional form for adjustment costs, linearize the
Euler equation and eliminate the expectation operator. These issues are addressed in
the subsections below.
'Several influential papers addressed the sources of information- or contracting-related frictions
in detail. See, for example, Jensen and Meckling (1976), Myers and Majluf (1984), Hart (1995) and
others. Here, these frictions are exogenous to the firm and are represented by the shadow value of
external finance. Another possible way to introduce financial frictions is by exogenously limiting the
amount of debt that the firm can raise at any point in time. This will create a shadow value of debt,
which has the same effect in the Euler equation as the shadow value of equity.
8
2.3 Financing Constraints
At the heart of the financing constraints theory is the factor et =-(t+i which
is the relative shadow cost of external finance in periods t and t + 1. If the shadow
cost of external funds is higher in period t than it is expected to be in period t + 1
(i.e. At > At+,), then Ot < 1 and it acts as an additional discount factor which makes
current period funds more expensive to use than the next period funds and therefore
induces the firm to postpone or reduce its investment. In this case we say that the
firm is "financially constrained," and et is the (degree of) financing constraints.7 In
perfect capital markets At = At+, = 0 for all t and hence Ot = 1 and the firm is
never constrained. With capital-markets imperfections, At depends on a vector of
state variables, including the productivity shock 't. Therefore, At is time-varying and
could be identified with some observable firm characteristics.
In the previous work several observable characteristics of the firm's financial health
have been used as proxies for the financing constraints. The most commonly used
variable was the cash flow. The problem with cash flow is that it is closely related
to operating profits and therefore also to MPK and will measure investment oppor-
tunities rather than, or in addition to, measuring the availability of internal funds
(i.e. the net worth). Using the terminology in Gilchrist and Himmelberg (1998) it
could be argued that the change in cash flow would simultaneously reflect a change
in "fundamentals" (increase in marginal productivity) and "financials" (increase in
the net worth of the firm, which will relax financing constraints).
As a measure of the internal funds, I use the stock of liquid assets, specifically
stock of cash and marketable securities scaled by total assets (hereafter referred to as
Cash Stock). The cash stock has an intuitive interpretation as "cash on hand" that
7If, on the other side, et > 1, the firm expects to be more constrained tomorrow (time t+1)
than it is today and at time t its investment will be unconstrained. In this case the firm is more
likely to invest at time t, since the discount factor ,3 is increased by the amount Ot (i.e. the interest
rate is lowered). Another possibility is that et = 1, because At = At+, # 0. But this seems very
unlikely in a stochastic model since At depends on a realization of the productivity shock. Even if
it is possible for some firms in some years to have Ot = 1, in estimating country-wide constraints
such a situation is very unlikely.
9
firms can use for investment if the opportunities arrive. One theoretical justification
for the cash stock measure appears in the Myers and Majluf (1984) model, where the
amount of cash holdings, which the authors call "financial slack," has a direct effect
on investment in the presence of asymmetric information. This slack allows firms
to undertake positive NPV projects, which they would pass if they do not have any
internal funds. This implies that if external financing is costly, there will be a positive
relationship between investment and cash stock, this is the relationship explored in
this paper.
Unlike the cash flow measure, the cash stock would proxy for the future growth
opportunities only in the presence of financing constraints. That is, firms that expect
high investment in the future, would accumulate cash stock to use up when the
opportunities arrive. Since holding cash is costly to the firms (because it diverts
resources from the productive use and offers zero return), the firms will accumulate
cash stock only if they expect to be financially constrained in the future. The evidence
consistent with this hypothesis is presented in Opler et al. (1999), among others.8
They find that firms hold liquid assets to ensure that they can keep investing when
outside funds are expensive and the firms that have lower cost of external financing
(large, dividend-paying, and firms with credit ratings) hold smaller stocks of liquid
assets.
I assume that the firm makes its decision for period t investment at the beginning
of that year (or, equivalently, the end of previous year). Therefore the appropriate
timing of the cash stock is t - 1, because the investment decision depends on how
much cash a firm has before starting the investment. I parametrize the financing
8Kim et al. (1998), Calomiris, Himmelberg and Wachtel (1995) and Calomiris and Himmelberg
(1996) also find that firms with lower costs of external finance maintain lower levels of financial work-
ing capital. Despite the growing empirical evidence on the "precautionary savings" by financially
constrained firms, this hypothesis still remains controversial, see for example Kaplan and Zingales
(1997); their view is disputed in Fazzari, Hubbard and Petersen (2000).
10
constraints as a function of cash stock as
Eit = aOi + aCashit-1,
where aoi is a firm-specific level of financing constraints (which enters into the fixed
effects) and Cashit-, is the cash stock. The sensitivity of investment to financial
health, measured by the parameter a, is the main focus of this paper. Recall from the
discussion above that under perfect capital markets, Eit = 1, hence a = 0 (i.e. invest-
ment is not related to internal funds). The larger the capital market imperfections,
the larger will be the sensitivity of investment to the amount of internal funds.
The main argument of this paper is that if financial development decreases capital
market imperfections, it should also decrease the coefficient a. In other words, the co-
efficient a is allowed to depend on the country-level measure of financial development
(hereafter FD), given by:
Oit = aoi + (a, + a2FD,)Cashjtij. (6)
Thus, the focus is on the interaction of FD and Cash, i.e. coefficient a2, and it is
expected to be negative, which will imply that financial development reduces the
sensitivity of investment to internal funds (i.e. financing constraints).9
2.4 Measuring MPK
The measure of MPK, derived from the profit maximization problem (derivations
are available from the author), is given by
MPK = 0K (7)
9In section 6.4 I present auxiliary results which allow for country-specific coefficients a,.
11
where s is a sales to capital ratio, 0 = , ak is the capital share in the produc-
tion function and ,L is a markup."0 This is a sales-based measure. An alternative
measure, which has been used in previous work (also derived in Appendix 2) is an
operating-profits measure. Although both measures are based on strong assump-
tions," the sales-based measure is less correlated with cash flow than the operating
profits measure (which basically is equal to cash flow). As discussed above, cash flow
would simultaneously proxy for change in "fundamentals" and "financials," therefore
I prefer to use the sales-based measure. As discussed in the previous section, I assume
that the firm makes the period t investment decision at the end of the period t - 1.
Therefore the appropriate timing for the sales to capital ratio is the end of period t.
2.5 Adjustment Costs
The adjustment cost function is given by C(It, Kt) = t2 - 9Kt V1)2Kt. This
adjustment cost function is slightly more general than the one used in the traditional
models because it includes lagged investment to capital ratio with an additional pa-
rameter g. It is added to capture strong persistence in investment to capital ratios
present in the data. This extended functional form allows for the more common form
with g = 0, which could be tested empirically. The intuition for this added term is
that it may be easier for the firm to continue investment at some fraction g of the
previous period ratio, since, for example, it has hired workers or made some other
arrangements which would be costly to cancel. Parameter vi could be interpreted as
some firm-specific level of investment at which adjustment costs are minimized. The
'0In the definition used here, the parameter ak (the capital share) is likely to be industry-specific,
and a markup (the measure of the market power) will be either industry or firm-specific. However
in the empirical work, the coefficient on sales to capital is assumed to be constant across all firms.
This will cause a measurement error, which is likely to bias the coefficient on the sales to capital
toward zero. This problem is ameliorated with the fixed effects, which capture the firm-specific level
of sales to capital.
"The operating-profits measure assumes that there are no fixed costs (i.e., reported Cost of Goods
Sold reflects only variable costs) and no quasi-fixed factors of production (such as R&D capital or
intangible assets). The sales-based measure assumes a Cobb-Douglas production function; while this
is a questionable assumption, the sales-based measure allows for quasi-fixed factors of production
and fixed costs.
12
marginal adjustment cost of investment is given by:
= av)(8)
AI t (Kt 9Kt_li)
2.6 Linearization and Expectations
Although the stochastic discount factor introduced by financing constraints, Ot
enters the Euler equation in a multiplicative form, in empirical work it is often easier
to estimate and interpret financing constraints when they enter additively. Similarly,
it is convenient to separate the discount factor ft in a linear term to allow for country-
and time-specific discount factors. Following Gilchrist and Himmelberg (1998), I
linearize the product of ft, Ot and the marginal benefit of investment (expression in
curly brackets in (5), here denoted as {'.t) using a first-order Taylor approximation
around the means. Since Ot could be above or below one, its mean should be a value
around one. Denoting the unconditional mean of the expression in curly brackets
as -y, and the average discount factor as 13, the approximation is given by (ignoring
constant terms):12
Otot {'t = Thet + I- {}t + ,yt- (9)
Finally, I assume rational expectations, which allows me to replace expectations
with realized values plus an expectation error eit. The error term is orthogonal to any
information available at the time when the investment decision is made. I assume
that the investment decision for year t is made at the beginning of that year (which
is equivalent to end of year t - 1). Therefore, the information available at the time of
decision is dated t - 1 since year t information does not arrive until the end of year
t. Then, the orthogonality conditions for this model are given by E[etIxt.8I = 0 for
s ) 1. This is equivalent to the assumption that the regressors are predetermined,
12Note that I implicitly assume that the covariance between the financing constraints factor and the
marginal benefit of investment (the term in {.}) is constant (in the empirical model this covariance
is captured by country-time dummies and fixed effects).
13
rather then strictly exogenous, and therefore require special estimation techniques
discussed in section 3.1.
3 Empirical Model and Estimation
I obtain the empirical model by substituting (6), (7), (8), and (9) into (5), and
replacing the expectation with the realization plus an error term. It is given by:
I = I + 2 I + 13 K + i4Cashit-1 + 035Cashi,t-IFD, + fi + d,,t + eit,
Kit Ki t+1 Sit-l it
(10)
where the coefficients are related to the structural parameters as:
01 = u 2 = d = 4 = a, -. a2 and d = 1 + (1-6)g.
d 2d 3ad' 4d ad
(11)
Here, fi denotes fixed effects,13 and d,,t denotes country-time dummies, that capture
aggregate shocks, including productivity, prices, and other macro shocks that are
allowed to be different for each country.
With respect to the coefficients in equation (10), the main hypothesis of this paper
is formally stated as:
Ho: 34)0and35< 0. (12)
That is, financing constraints are nonzero (for at least some countries) and they
decrease with financial development. (Note that /4 and 15 depend on previously
13Fixed effects arise in this structural model for several reasons. First, there are firm-specific
parameters for adjustment cost vi and for financing constraints aoi. Second, the omitted terms
that contain prices of investment goods and the conditional covariance of financing constraints and
marginal benefit of investment (discussed in 2.6) are replaced by the combination of time and fixed
effects. Third, the fixed effects capture a sample selection bias if the firms included in the sample
have different investment policy than the rest.
14
defined structural parameters a, and a2.) The focus is on the interaction of Cashit
(the firm-level variable measure of internal funds) and FD, (the country-level index
of financial development).
Hereafter, I refer to equation (10) as the "baseline model." I use the same frame-
work to test whether other measures affect financing constraints by replacing FD with
the index of interest (for example legal system indicators). For robustness tests, I add
additional interactions to the baseline model to see if the financial development effect
is still present when I control for other potential sources of financing constraints (such
as firm size or business cycles, described in section 6).
3.1 Estimation Methodology
The first issue in estimating this model concerns the presence of fixed effects. There
are several reasons for fixed effects to arise in this model (see footnote 13). The fixed
effects are correlated with regressors because the model contains lags and leads of the
dependent variable, therefore they need to be removed before the estimation, One
common procedure for removing fixed effects is mean-differencing. However, since the
regressors are not strictly exogenous (see discussion in section 2.6), mean-differencing
would create biased estimates. I use forward mean-differencing, which removes only
the forward mean, i.e. the mean of all the future observations available for each firm-
year. The forward mean-differencing preserves orthogonality between transformed
errors and untransformed original variables, which are used as instruments. Arellano
and Bover (1995) show that when moments are formed by summing over firms and
periods, as opposed to treating each year as a unique set of moments, and when there is
no serial correlation in the error term, the forward mean-differencing is more efficient
than the more commonly used first-differencing. In addition, the first-differencing
induces serial correlation in the errors and requires appropriate error-correction, while
the forward mean-differencing preserves the error structure.14
"4The forward mean-differencing, also referred to as the Helmert procedure, was used to estimate
investment models by Bond and Meghir (1994) and Gilchrist and Himmelberg (1998).
15
The country-time dummies, d,,t, are removed by country-time differencing of all
variables, i.e. regressors and instruments (the regressors are time-differenced after
the forward mean-differencing, but the order of transformations is not important).
As discussed in section 2.6, the expectation error eit is orthogonal to the informa-
tion available at the time when the investment decision is made, which I assume to
be t - 1. As noted above, after the forward mean-differencing, the transformed errors
are still orthogonal to the untransformed original variables dated t - s, where s > 1.
Therefore, I use the GMM procedure, implemented as IV (instrumental variables),
with t - 1 and t - 2 lags of instruments. The instruments are all the variables in the
regression, plus cash flow, cost of goods sold, industry dunmiies and the interactions
of cash, sales and investment with FD (see Table 2 for variable definitions).
In all regressions I use heteroskedasticity robust estimates of the standard errors,
which do not require an assumption of the independence of errors within the firm
(implemented with Stata's cluster option). To eliminate influential observations, I
exclude 1% on each side of the distribution for each of the variables in the regression
prior to transformations.
4 Data
All firm level data come from the Worldscope database, which contains data on
large publicly traded firms in which there is an investor interest. Using only large
publicly traded firms allows one to compare "apples to apples" across countries and
separate the effects of different financial and legal environments, which is the center of
attention here. An additional benefit of using these data is the attempt by Worldscope
to standardize accounting information to improve cross-country comparability.'5 The
drawback of the sample is that it does not have data on large non-public firms. Even
'5For example, if one company reports sales with included excise tax and another company ex-
cludes it, Worldscope corrects this difference and presents both with excluded tax. This is important
for my study because I use sales as a measure of MPK and want to have as much cross-country com-
parability as possible.
16
though I cannot extend the findings from the large public firms to all firms without
the appropriate data, there is a lot to be learned on the cross country differences in
this sample.
The firm data are available for 40 countries and cover over 7000 firms for the years
1988-1998 (however the years before 1991 and the year 1998 have fewer observations).
Details on the sample selection are given in Appendix 1. The coverage within countries
varies widely from as little as 1% of all listed domestic firms included (for India) to
as many as 82% (for Sweden), as calculated by LLSV (1997) who use the same
sample. Table 1 gives the list of countries in the sample with the number of firms
and observations per country. The number of firms in each country varies widely
across the countries, and the less developed countries are underrepresented. This
creates a problem with pooled cross-country estimation, though it is mitigated using
the empirical techniques discussed in the next section. The main firm-level variables
are investment and sales, scaled by the beginning of the period capital,"6 and stock
of liquid assets (cash stock). Other variables are defined in Table 2.
The main country-level indicator is an index of financial development, FD. It is
equal to the sum of the (standardized) indices of the stock market development,
STKMKT, and financial intermediaries development, FININT, which come from
Demirguc-Kunt and Levine (1996) (they refer to these indices as Indexl and Findexl
respectively). The STKMKT is the sum of three standardized measures: market cap-
italization over GDP (i.e. the size of the stock market), total value traded over GDP,
and total value traded over market capitalization (two measures of liquidity of the
market). The FININT is the sum of two standardized measures: the ratio of liquid
liabilities (M3) to GDP (i.e. the overall size of the credit market) and the credit going
16The model requires one to use the beginning of the period capital stock as a scaling factor for
calculating adjustment costs and MPK. One alternative is to use lagged capital stock (i.e. period t-1
used as the beginning of the period t capital stock). However, this would not be appropriate if there
are mergers, acquisitions, divestitures or other capital-changing events, which are hard to identify.
I use the approximate value given by the ending period capital, minus investment and depreciation
in that year, which is more robust to the capital-changing events, as discussed in Love (1999).
17
to the private sector over GDP (the amount of credit that is relevant to the firm's
financing).
Thus the FD index combines five important characteristics of financial markets
into a standardized measure, similar to the ones used in other studies. For a robust-
ness check, I also use the real growth of GDP (a country-year variable) as an indicator
of the business cycle conditions in each country, which are also thought to affect the
financing constraints (discussed in section 6). Table 2 lists the rest of the country-level
variables (and their sources), which are discussed in the relevant sections below.
Table 3 reports means and medians of the key variables, by country. Table 4
reports cross-country correlations of the country averages for these variables. Several
patterns stand out in these correlations. First, sales to capital, SK, is correlated with
investment to capital, IK, which could imply that countries with higher productivity
invest more (of course no claim is made about the causality of this relationship).
Second, FD appears to be positively correlated with the sales to capital ratio, SK.
This seemingly counter-intuitive result is likely due to differences in industry and
sample compositions across countries and should not be interpreted causally. Third,
FD is positively correlated with IK, although this correlation is not significant, so it is
not clear if financial development increases investment on average. Finally, the cash
stock is correlated with investment and sales. The interpretation of this relationship
is done using the regression analysis below. Most of the country-level institutional
characteristics are highly correlated with each other (Panel B, Table 4) and therefore
should not be included in the regressions simultaneously.
5 Main Results
As is clear from Table 1, the number of firms included in the sample varies widely
across the countries. The US and UK have more than 1000 firms per country, while
the rest of the countries have only 136 firms on average (Japan is the third largest
with over 600 firms). Such a prevalence of US and UK companies will overweight
18
these countries in the cross-country regressions and prevent smaller countries from
influencing the coefficients, especially when the variable of interest is the interaction
with the country-level financial development index. To correct for this I use two
approaches: the first is the rank-based approach, and the second is the weighted
regressions approach, discussed below.
The rank-based approach is based on the reasoning that to have a meaningful test
for the financial development effect on financing constraints, one needs to compare
apples to apples, i.e., large firms in one country to large firms in another country.'7
Therefore, the regressions include only the largest firms within each country. The
inclusion criteria are based on firm ranking, where rank 1 is given to the largest firm
in its country. Since there is no a priori criteria to select any specific number of firms,
I experiment with different cutoff points and report results for 50, 100, 150 and 200
largest firms.
The weighted regression approach assigns a country-specific weight, which is equal
to the inverse of the number of observations in each country. Countries with a lot of
observations get a smaller weight and countries with fewer observations get a larger
weight, so that the number of observations is equalized across all countries. This
method uses all the available observations, which results in efficiency gains.
The main results are based on the model given in (10) and are reported in Table
5. Models 1-4 use the rank-based approach with different cutoff points, and model 5
uses the weighted regression approach. All coefficients have their predicted signs and
are significant at conventional levels, with the most significant coefficients resulting
in the 150 largest firms regression (model 3) and weighted regression (model 5) where
all the coefficients are significant at the 1% level. The main variables of interest are
the cash stock and the interaction of cash stock with the financial development index,
FD. The main hypothesis, stated in (12), is that the cash coefficient is positive and
17It is plausible to argue that some small companies in the US have less access to external finance
than the few largest companies in, for example, Malaysia or Thailand, which in the sample period
enjoyed more attention from domestic and outside investors than many small US companies.
19
the interaction is negative. This result is obtained in all the regressions in Table 5.
This confirms the main claim of this paper, that financial development decreases the
sensitivity of investment to availability of internal funds, measured by the cash stock
(this sensitivity is interpreted as a proxy for the financing constraints).
5.1 Structural Parameters
The interpretation of the coefficients magnitudes is best done in terms of the
parameters of the underlying structural model, which are given by expressions in
(11). To identify the structural parameters I use the minimum-distance estimator,
described in Himmelberg (2000) (details on this estimator are available from the
author). There are 5 equations and 8 parameters, therefore not all the parameters
can be identified. I choose to identify (Q, g, a, a1, a2) and assume the values for the
remaining parameters. I assume a depreciation rate 6 = 0.12, which is the sample
average of the depreciation expense to capital ratio. The coefficient 0, which translates
sales to capital ratio into MPK in (7), is assumed to be equal to 0.23 (this corresponds
to the values for the capital share ak = 0.3 and markup IL =1.318). Finally, I assume
the value for the linearization parameter -y = 1.2, which is equal to the average
marginal benefit of investment, discussed in section 2.6.19 The resulting structural
parameters are reported at the bottom of Table 5. Although there is some variation
among the estimated parameters, the average values seem plausible. The average
discount rate /3 is equal to 0.8 (and several models imply a discount rate of 0.9),
which seems quite reasonable. Note that country and time specific discount rates are
captured by the country-time dummies so the estimated 3 represents an "average"
discount rate, which is hard to identify in panel data.
The parameters of the adjustment cost function have average values a =6.5 (this
l8The estimate of markup is taken from Hubbard, Kashyap and Whited (1995), and it corresponds
to a demand elasticity of -4.
19I assume that MPK is approximately equal to 0.2 (taken from Gilchrist and Hlimmelberg (1998))
and the marginal adjustment cost term is 0.2. Since the value of y depends on the values of other
parameters, I experimented with an iterative procedure when a was determined at every stage of
minimization using the parameter values at that stage. This produced qualitatively similar results.
20
excludes insignificant value from model 1) and g=0.23. The parameter g is quite
stable and always significant (which confirms the extended functional form for the
adjustment costs), while a varies quite a bit and is less significant in general. The
marginal adjustment costs function is then given by c = 6.5( I - 0.23 I - vi).
Since the average I in the data is approximately 0.18, the difference of the first two
terms is equal to 0.14. The magnitude of the marginal adjustment cost depends on
the parameter vi, which is not possible to estimate separately since it is included in
the fixed effects. For example, if vi is in the range 0-0.1, the marginal adjustment cost
will be in the range 0.25-0.9, which is in line with previous evidence (for example,
Hubbard and Kashyap, 1992).
Next, I analyze the financing constraints factor, e, which is defined as a function
of the structural parameters in (6). The sensitivity of e to change in cash stock
depends on the country's level of FD. Thus, a firm in a country with high FD (i.e.
one standard deviation above the mean of FD) has close to zero sensitivity of e to
the change in the cash stock.20 This implies that firms in countries with high financial
development, such as US, UK and Japan, are not financially constrained. This is not
surprising given that the sample mainly consists of large publicly traded firms. For a
firm in a country with an average FD, such as Spain, Prance and Israel, a one standard
deviation change in cash stock results in 6.5% decrease in the financing constraints
factor e, which could be translated as a change in cost of capital from 11% to 19%.
On the other side, for firms in countries with low FD (i.e. one standard deviation
below the mean), such as Mexico, Brazil, or Chile, a one standard deviation decrease
in cash stock will decrease e by 13%, which implies an increase in the cost of capital
from 11% to 28%. Although these calculations are rough approximations, and appear
a bit on the high side (i.e. imply quite large changes in cost of capital), they suggest
that financial development has a very large and economically significant effect on the
20For simplicity, in all the calculations I assume that the financial development index has a mean
of zero and standard deviation of one. The actual mean and standard deviation are -0.03, and 1.14
respectively. I also use values al = 1 and a2 = -1, while the average values from Table 5, ignoring
insignificant estimates, are 0.96 and -0.99 respectively.
21
financing constraints.
6 Tests of Alternative Explanations
6.1 Size Effect
One potential problem with the main results (that financial development decreases
financing constraints) is the omitted effect of company's size on its financing con-
straints. Firm size has been commonly used to identify firms that are more likely
to be financially constrained (see Schiantarelli (1995) for a survey). The small firms
are more likely to suffer from financing constraints (i.e. have larger coefficients on
the financial variables) because information asymmetries are larger. If more finan-
cially developed countries have larger firms, as argued by Kumar, Rajan and Zingales
(1999), then the estimated FD effect could be attributed to the differences in the
firm size rather than financial development. To test this, I add the interaction of size
(measured by the log of total assets in US dollars) with cash stock to the baseline
model:
Kit =: 'K +)2K + 03 K + 34Cashi,t-j + 35Cashi,t-FDr (13)
+i36Cash,,t-,Sizei,tij + fi + deXt + eit
The test now is that the main hypothesis (04 > 0 and 35 < 0) still holds and also
06 < 0, i.e. financing constraints are smaller for larger firms.2' The instruments now
include size and interaction of size with investment, cash and sales.
The results with weighted regressions are presented in Table 9. I first test whether
size has any effect when included by itself (i.e. in the model (13) only Cashi,ti Sizei,t-j
2"Note that size enters Euler equation only as interaction with the cash stock (proxy for the
financing constraints), and not in levels. If there is any level effect, it is captured by the fixed
effects. One problem this size test is the nature of the sample, which mainly consists of the large
firms. Nevertheless, there is some variation in size among the firms in most countries, which is
exploited for this test. At worst, the sample selection creates bias against finding any size effect.
22
interaction is included). Model 1 in Table 6 shows that there is a significant size ef-
fect, that is larger firms have smaller cash coefficients. For example, for the firm with
the mean size, the cash coefficient is equal to 0.12, and for the firm with size equal
to one standard deviation below the mean the cash coefficient equals 0.3, almost a
triple increase in the investment sensitivity. Next, I include the interaction of cash
with financial development in addition to the size interaction (model 2, Table 6) and
find that both interactions are significant at 1%. This confirms that the financial de-
velopment effect is not caused by the differences in the size of the firms and that both
size and financial development have an independent effect on financing constraints.
This methodology allows me to address another interesting empirical question: Is
the size effect equal in all countries, or is it related to financial development? The
intuition is that the largest firms in less financially developed countries could still
enjoy an abundance of external finance (obtained through access to external capital
markets, or political connections), while smaller firms will be comparatively more
disadvantaged. I test this with an augmented model:
- I - + -2 S 3 3+ 4Cashit-, + f5Cashi,t-±FD, (14)
Kit Ki't+1 Ki't-1 Kit
+/36Cashi,t-iSizei,t-1 + 37Cashi,t-1Sizei,t-1FD, + fi + d,,t + eit
The test now is on the triple interaction coefficient, and I expect /7 > 0, that is
financial underdevelopment has more effect on the small firms (i.e. less negative
effect on the large firms), and all previous hypothesis are expected to hold: /34 > 0,
/35 < 0 and /6 < 0.
The results are reported in model 3, Table 6. All 4 coefficients /4 - j7 (cash, cash
interaction with FD, cash interaction with size, and triple interaction of cash, size
and FD) have their predicted signs and are significant at the 1% level. This suggests
that financial development has a differential effect on firms of different sizes. That is
the small firms are affected significantly more than the large firms are.
For a robustness check of this result, I define a dummy variable, which is equal
23
to one if the size of the firm (measured as the log of total assets) is smaller than
the median size in its country (note that this definition reverses the signs on the
size interactions). This definition is more robust to differences in firm size across
countries, but is less robust to differences in the sample size across countries (because
for the countries with the small samples, some of the large firms are classified as
small). The results, in model 4, although slightly weaker than in model 3, confirm
the above conclusion that small firms are significantly more affected by FD.
To quantify the relative difference in financing constraints of large and small firms
I use estimates from the model 3. Thus, in a country with the average FD, the average
size firm has cash coefficient of 0.15, while the small firm (i.e. the size of one standard
deviation below the average) has the coefficient of 0.38. However, in a country with
low FD (i.e. one standard deviation below the average), the average size firm has
cash coefficient of 0.19, while the small firm has the coefficient of 0.58. Thus the size
effect (the difference in coefficients of firms with different sizes) is about 35% larger
for a country with low financial development, relative to a country with an average
financial development.
Another way I address this question is by splitting the sample on high and low
financial development subsamples (based on the median FD) and estimating cash and
size coefficients (i4 and 06) from the model in (14) separately for each subsample.
These results are presented in Table 7. For the high FD sample, neither cash nor cash
interaction with size are significant, while for the low FD sample both are large and
very significant. This shows that even in the sample of large publicly traded firms,
the financing constraints are a significant issue for the countries with a low level of
financial development, which not only have higher financing constraints on average,
but have disproportionately larger constraints for the smaller firms.
24
6.2 Business Cycles
In this section I consider how the differences in the countries business cycles could
affect my results on financing constraints. Recall from the discussion above that the
main idea behind the financing constraints theory is that the information asymmetry
between borrowers and lenders creates the agency costs, which manifest themselves in
the wedge between internal and external financing costs. These costs decrease with an
increase in the borrower's net worth because, for example, an increase in the personal
stake decreases the incentives to misallocate the funds. Since the net worth is likely
to be procyclical, the agency costs will decline in booms and rise in recessions. In
other words, external financing is easier to obtain during good times (when profits
are high and balance sheets are healthy).22
One potential concern with my result on the financial development effect is that
the sensitivity of investment to internal funds could reflect different stages in the
countries' business cycles, rather than the average level of financial development.
That is, over the short period covered by my sample, it could happen that countries
with low level of financial development happen to be in recessions on average.
To test this possibility, I include the interaction of the real GDP growth rate,
grGDPd, a measure of the economic conditions in the country, with the firm-level
measure of the internal funds, the cash stock. Since the effect of the economic growth
is expected to manifest itself in the same time period as investment, i.e. growth at time
t is expected to affect investment at time t, the interaction timing is Cashjt-1grGDP,t,
where as before the cash stock represents available liquid assets at the beginning of
the period t (i.e. end of period t - 1). If the economic boom periods (i.e. periods
with high GDP growth) are associated with the lower level of financing constraints,
this interaction is expected to be negative.
The results are presented in Table 8. The interaction of GDP growth with cash
22This intuition has been formalized and tested on US data by Bernanke and Gertler (1989),
Gertler and Hubbard (1988), Kashyap, Lamont and Stein (1994) and Gertler and Gilchrist (1994),
among others.
25
stock is indeed negative and significant at the 5% level in three out of 4 models (it is
only marginally significant, at 17%, in model 3). This implies that favorable economic
conditions do improve financing constraints, in line with previous evidence on US
data. The effect of financial development on financing constraints is not significantly
affected by the addition of the GDP growth interaction: the FD interaction remains
significant at 1% in model 2 and at 2% in model 4. This robustness check confirms
that the overall level of the country's financial development is a significant predictor
of the firm's financing constraints, even after controlling for the business cycle effects.
6.3 Legal System Indicators
The distinguishing feature of the modern corporation is the large set of laws that
create the environment in which it operates (Zingales, 2000). The firm's financing
activity is largely based on the financial contracts or securities and the defining feature
of these securities are the rights that they bring to their owners (Hart, 1995). The
differences in these rights and their enforcement across the countries have been a focus
of recent developments in the literature, pioneered by LLSV (1997, 1998). They argue
that "the [legal] protection investors receive determines their readiness to finance
firms," and show that the legal environment has large effects on the size and breadth
of capital markets across countries. LLSV also point out that the legal systems of
most countries could be categorized into several broad legal families, which come from
the English, French, German, and Scandinavian origin. The countries inherited these
legal traditions from their colonizers, and the consequent development of the legal
system largely depends on this "origin." The apparent exogeneity of this legal origin
to subsequent economic development has been used to reinforce the arguments about
the causality from the financial development to economic growth (see Levine, 1999
and Beck et al., 1999).
It is easy to argue that better legal protection of investors should allow for more
efficient contracts and their enforcement, which in turn should reduce the cost of
external finance. The goal here is to test whether the legal variables are associated
26
with decreasing sensitivity of investment to availability of internal funds. This will
imply that better legal protection has the "real" consequences, i.e. allows for more
efficient capital allocation by diminishing financing constraints. The previous research
(LLSV and others) identified several legal system indicators such as the efficiency of
the legal system, the rule of law, the risk of expropriation, corruption, and legal origin
dummies (see Table 2 for variable definitions and original sources of this data). These
indicators measure different aspects of the legal environment. For example, efficiency
and the rule of law measure the quality of the law enforcement, i.e. how well the
laws on the books are enforced by the courts. Corruption measures the distortions
introduced by the courts and the government into the functioning of the financial
and real sectors of the economy. The accounting standards measure the quality of
information available to investors and should therefore reduce the external financing
costs associated with information availability.
I use the baseline model in (10) and replace FD with each of the legal indicators.
As shown in the Table 9, each indicator has a negative effect on the cash coefficient
when included by itself (all models with odd numbers). The results for legal origin
are also consistent with the previous evidence: FRench origin increases financing con-
straints (model 11) and English origin decreases the constraints (model 13). However,
when any of these indicators are included together with FD, they become insignificant,
while FD continues to be highly significant (all models with even numbers). Given
these results, it appears that the index of financial development is a better summary
measure of the differences in the cost of external finance than the individual legal in-
dicators are. In other words, the legal system differences are already reflected in the
level of financial development, and so the legal system affects the capital allocation
only indirectly, through better functioning capital markets.
6.3.1 Legal Origin as Instrument for Financial Development
The causality of the financial development and growth correlation has been debated
since the first empirical study of this relationship by King and Levine (1993). Sceptics
27
of the finance-growth link have pointed out that the financial systems simply responds
to the demands of the growing economies and therefore is endogenous to growth
(Lucas, 1988). Others argued that the financial development could be a leading
indicator of growth as financial markets anticipate the increased economic activity
and develop in anticipation of this activity (Rajan and Zingales, 1998). The potential
endogeneity of the financial development is a valid concern in the country-level or
even in the industry-level study (as financial markets could respond to the anticipated
growth of some individual industries). However, this endogeneity becomes less likely
in the firm-level study and financial development could safely be considered exogenous
to the growth of any given firm. Nevertheless, the test with legal origin as instrument
for the financial development could provide a useful robustness check on the results
and it is the goal in this section.
I use the baseline model (10) and include FD interaction with cash as a regressor.
However, now I do not include FD interactions in the instruments list and replace
them with legal origin dummies and their interactions with the firm level variables.
Thus, only the component of financial development that is explained by the legal
origin is allowed to influence the investment sensitivity. The results presented in
the Table 10 are remarkably similar to the main results in the Table 5, with slight
decrease in the significance of the cash coefficients, but the FD interaction continues
to be significant at 1% level in all the regressions (except "rank 50," which was only
significant at 10% before and now is significant at 5%) 23 The conclusion of this
section is that the main result is unchanged with the use of legal origin as instrument
for the financial development.
6.4 Single Country Regressions
This section describes an alternative way to address the relationship between
financial development and financing constraints. Recall that with the cross-country
23It is also interesting to note that the Hansen test of overidentifying restrictions is not rejected
at 1% level in the models that use legal origin as instrument for FD.
28
regressions the financing constraints are parametrized as a linear function of the index
of financial development, FD, given by equation (6). An alternative approach is to
allow each country to have different levels of financing constraints (measured by the
coefficient on cash stock), given by Oit = aoi + acCashit-1. The country-specific cash
coefficient, a,, is obtained in the first stage regressions, where the Euler equation is
estimated separately for each country using the model:
i I +02 I +03 K + a,Cashit-, + dt + fi + eit. (15)
As before, fi denotes fixed effects (see footnote 13), and dt denotes time dummies.
Then, I estimate the second-stage regression, in which the coefficients a are regressed
on the country-level index of financial development (FD) using the model:
'a = bo + bIFD, + e,. (16)
The main hypothesis now is that b, < 0, that is, the first stage estimates of the cash
coefficients, -a, are negatively related to the index of financial development, FD,. The
second stage regression in (16) is estimated by OLS.24
The single-country regressions are not as efficient as cross-country regressions be-
cause they require estimating 200 coefficients (5 per country for 40 countries) rather
than estimating only 6. However, this approach has a few advantages: first, it is
completely unrestricted in a sense that all the coefficients are allowed to vary across
countries. Second, it allows one to estimate the average level of financing constraints
for each country, while the cross-country regressions leave a "black box" feeling be-
cause only the slope, bl, is estimated.
Table 11 reports the results of estimating (15) separately for each country. The
24Since the dependent variable is estimated in stage 1, for proper inference the generated regressors
adjustment is required. However, due to the nature of this methodology, such an adjustment proves
to be quite complicated and it is ignored in this version of the paper. Therefore, the errors reported
in stage 2 are not asymptotically correct and are used as an approximation.
29
cash stock coefficients range from zero (25th percentile) to 0.35 (75th percentile),
with a mean of 0.17 and a median of 0.11. These statistics are in line with the
cash coefficients estimated in the cross-country regressions (which were varying from
0.10 to 0.17, Table 5). However, most of the cash coefficients are not significant at
conventional levels, often due to the small sample size in the individual countries.25
Despite the problem with low efficiency, the coefficients are consistent and present
interesting patterns, analyzed below.
Using the model in (16), I regress the country-level cash coefficients on the FD
index. This results in a coefficient of -0.18, significant at 1% level (the standard
error is 0.05 and R squared is 0.35). The scatterplot of cash coefficients and financial
development, with predicted values from the above regression (the straight line) are
given in Figure 1. It shows that cash sensitivities exhibit a clear negative relationship
with financial development. This confirms the result obtained in section 5, which
found negative coefficients on the interaction of cash stock and FD.
There is one visible outlier on the plot, Korea (KR), and for a robustness check I
ran the above regression without it, which results in a coefficient of -0.16, significant
at 1% (the standard error is 0.04 and R squared is 0.37). It is also interesting to
note that the effect of the financial development on the cash coefficient, given by the
interaction term in cross-country regressions (which varied from 0.08 to 0.15, Table
5), is similar to the effect found here in the single country regressions (the slope
of 0.16-0.18 in the second stage). This make the single-country regressions a useful
robustness test, despite the problem of low efficiency of individual coefficients.
7 Conclusions
This paper shows that financing constraints, measured by the sensitivity of in-
vestment to the availability of internal funds, are significantly negatively related to
25This problem is exacerbated by the fact that less developed countries, that are expected to have
larger coefficients, have a small number of observations, while more developed countries (which have
more observations) are expected to have coefficients close to zero.
30
financial development. This negative effect remains after controlling for firm size and
the country's business cycles, which also affect financing constraints. I also find that
small firms are disproportionately more disadvantaged in less financially developed
countries than are large firms. I also find that legal system indicators (the efficiency of
the legal system, the risk of expropriation, corruption, or legal origin) are negatively
related to investment sensitivity (the measure of financing constraints). However,
they lose significance when financial development is added to the regressions. This
implies that the legal system affects financing constraints indirectly, through better
developed financial markets. The impact of FD on financing constraints is unchanged
when legal origin is used as an instrument for financial development.
The paper makes contributions to two strands of literature. First, it contributes to
the investment literature by estimating a structural investment model and confirming
the presence of financing constraints for a broad range of countries. This paper
extends the only existing cross-country study, in Bond et al. (1997), which studies
investment in four developed countries. Second, and more important from a policy
perspective, this paper contributes to the economic development and growth literature
by showing that financial development diminishes financing constraints by reducing
information asymmetries and contracting imperfections. The decrease in financing
constraints allows firms to invest according to their growth opportunities and therefore
improves capital allocation.
31
Appendix 1. Sample Selection
All countries in the Worldscope database (May 1999 Global Researcher CD) with
at least 30 firms and at least 100 firm-year observations are included in the sample
(the exception is Venezuela (VE) which is included with 80 observations only); former
socialist economies are excluded. This results in a sample of 40 countries. The sample
does not include firms for which primary industry is either financial (one digit SIC
code of 6) or service (one digit SIC codes of 7 and above).
In addition I delete the following (see Table 2 for variable definitions):
- All firms with 3 or less years of coverage;
- All firm-years with missing CAPEX, PPENT, Sales, and cash;
- Observations with zero PPENT (200 obs);
- Observations with negative KBEG (277 obs), Cash/Ta or COGS (27 obs);
- Observations with IK > 2.5 (1% of all obs);
- Observations with SK > 20 (5% of all obs) ;26
- Observations with Cogs/K > 20 (80 obs.);
- Observations with Cash/Totass >0.6 (1% of all obs);
- 50% of all US firms with at least 4 years of data available was selected by random
sample.27
The resulting dataset has about 59,500 observations, the number of observations
by country is given in Table 1.
26This rules excludes firms for which capital is not a big factor in production. Half of these were
in the US and UK; Japan, France and Denmark totaled 25%.
27The original sample for the US had over 25,700 observations (firm-years) while for all other
countries at most there are 12,000 for the UK, 5,000 for Japan, less then 1,000 for most countreis
(see Table 1). Even after the sampling, the US has the most data available.
32
References
Allen, F. and D. Gale, "Comparing Financial Systems." Campbridge MA: MIT
Press 1999.
Arrelano, M. and 0. Bover, 1995, "Another Look at the Instrumental Variable
Estimation of Error Component Models," Journal of Econometrics 68, pp. 29-51.
Beck, T. and R. Levine, 2000. "New Firm Formation and Industry Growth: Does
Having a Market- or Bank-Based System Matter?" World Bank working paper series
2383.
Beck, T., R. Levine and N. Loyaza, 2000, "Financial Intermediation and Growth:
Causality and Causes," Journal of Monetary Economics, Vol. 46, pp. 31-77.
Bekaert G., C. Harvey, and C. Lundblad, 2000, "Emerging Equity Markets and
Economic Development," NBER working paper 7763.
Bernanke, B. and M. Gertler, 1989, "Agency Costs, Net Worth, and Business
Fluctuations," American Economic Review, 79, pp.14-31.
Bernanke, B., M. Gertler and S. Gilchrist, 1996, "The Financial Accelerator and
the Flight to Quality," Review of Economics and Statistics, 78(1), pp. 1-15
Bond, S. and C. Meghir, 1994, "Dynamic Investment Models and the Firm's
Financial Policy," Review of Economic Studies, 61(2) pp. 197-222.
Bond, S., J. Elston, J. Mairesse, and B. Mulkay, 1997, "Financial Factors and
Investment in Belgium, France, Germany and the UK: a Comparison Using Company
Panel Data," NBER working paper 5900.
Calomiris, C. and C. Himmelberg, 1996, "Investment Banking Costs as a Measure
of the Cost of Access to External Finance," Mimeograph, Columbia University.
Calomiris, C., C. Himmelberg, and P. Wachtel, 1995, "Commercial Paper, Cor-
porate Finance, and the Business Cycle: a Microeconomic Perspective," Carnegie-
Rochester Conference Series on Public Policy 42, pp. 203-50.
Calomiris, C. and G. Hubbard, 1995, "Internal Finance and Firm-Level Invest-
ment: Evidence from the Undistributed Profit Tax of 1936-1937," Journal of Business,
33
68 (5) pp. 443-82.
Demirguc-Kunt, A. and R. Levine, 1996, "Stock Market Development and Finan-
cial Intermediaries: Stylized Facts," World Bank Economic Review 10, pp. 291-321.
- - - - - -1999, "Bank-Based and Market-Based Financial Systems: Cross-
Country Comparisons", working paper 2143, World Bank.
Demirguc-Kunt, A. and V. Maksimovic, 1998, "Law, Finance and Firm growth,"
Journal of Finance, 53(6) pp. 2107-31.
______, 2000, "Funding Growth in Bank-Based and Market-Based Financial
Systems: Evidence from Firm-Level Data," World Bank working paper series 2432.
Fazzari, S., G. Hubbard and B. Peterson, 1988, "Financing Constraints and Cor-
porate Investment," Brookings Papers on Economic Activity, 78 (2), pp. 141-95.
______, 2000, "Investment-Cash Flow Sensitivities are Useful: a Comment
on Kaplan and Zingales," Quarterly Journal of Economics, 115(2), pp. 695-705.
Gelos, G. and A. Werner, 1999, "Financial Liberalization, Credit Constraints, and
Collateral: Investment in the Mexican Manufacturing Sector," IMF working Paper,
WP/99/25.
Gertler, M. and S. Gilchrist, 1994, "Monetary Policy, Business Cycles, and the
Behavior of Small Manufacturing Firms," Quarterly Journal of Economics, 109(2),
pp.309-340.
Gertler, M. and G. Hubbard, 1988, "Financial Factors in Business Fluctuations,"
in "Financial Market Volatility: Causes, Consequences, and Policy Recommenda-
tions," Federal Reserve Bank of Kansas City, 1988.
Gilchrist, S. and C. Himnmelberg, 1998, "Investment, Fundamentals and Finance,"
NBER Macroeconomics Annual 1998, MIT Press.
Gross, D., 1995, "The Investment and Financing Decisions of Liquidity Con-
strained Firms," MIT dissertation.
Hart, O., 1995, "Firms, Contracts, and Financial Structure." London, Oxford
University Press, 1995.
34
Harris, J. R., F. Schiantarelli and M. G. Siregar, 1994, "The Effect of Finan-
cial Liberalization on the Capital Structure and Investment Decisions of Indonesian
Manufacturing Establishments," The World Bank Economic Review, 8.
Himnelberg, C., 2000, "Estimating Structural Investment Models with Panel
Data: Identifying Net Worth and Balance Sheet Effects," Mimeograph, Columbia
University.
Himmelberg, C., R.G. Hubbard, and I. Love, 2000, "Investor Protection, Owner-
ship, and Investment," Mimeograph, Columbia University.
Hubbard, R.G., 1998, "Capital Market Imperfections and Investment," Journal
of Economic Literature, 36 (3).
Hubbard, R.G. and A. Kashyap, 1992, "Internal Net Worth and the Investment
Process: An Application to US Agriculture," Journal of Political Economy, v.1001 3).
Hubbard, R.C., A. Kashyap, and T. Whited, 1995, "Internal Finance and Firm
Investment," Journal of Money Credit and Banking, 27 (4) pp. 683-701.
Jaramillo, F., F. Schianterelli, and A. Weiss, 1996, "Capital Market Imperfections
Before and After Financial Liberalization: An Euler Equation Approach to Panel
Data for Ecuadorian Firms," Journal of Development Economics, vol. 51(2), pp.
367-86.
Jensen, M. and W. Meckling, 1976, "Theory of the Firm: Managerial Behaviour,
Agency Costs, and Ownership Structure," Journal of Financial Economics, 3(4), pp.
305-60.
Kadapakkam, P., P. C. Kumar, and L. Riddick, 1998, "The Impact of Cash Flow
and Firm Size on Investment: The International Evidence," Journal of Banking and
Finance, 22, pp.293-320.
Kaplan, S. and L. Zingales, 1997, "Do Investment-Cash Flow Sensitivities Provide
Useful Measures of Financing Constraints?" Quarterly Journal of Economics 112(1)
pp. 169-215.
Kashyap, A., 0. Lamont, and J. Stein, 1994, "Credit Conditions and the Cyclical
35
Behavior of Inventories," Quarterly Journal of Economics, vol. 109 (3), pp. 565-592.
Kim, C., D. Mauer, and A. Sherman, 1998, "The Determinants of Corporate
Liquidity: Theory and Evidence," Journal of Financial and Quantitative Analysis,
v.33(3).
King, R.G. and R. Levine, 1993a, "Finance and Growth: Schumpeter Might be
Right," Quarterly Journal of Economics, 108(3), pp. 717-37.
_- -______ 1993b, "Finance, Entrepreneurship, and Growth; Theory and Evi-
dence," Journal of Monetary Economics, 32, pp. 513-42
Kumar K.B., R. Rajan, and L. Zingales, 1999, "What Determines Firm Size,"
NBER working paper 7208.
La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny, 1997, "Legal De-
terminants of External Finance," Journal of Finance, pp. 1131-50.
-_-____-__, 1998, "Law and Finance," Journal of Political Economy 106-6, pp.
1113-55.
______, 2000, "Investor Protection and Corporate Governance," Journal of
Financial Economics, 58.
Levine, R., 1999, "Law, Finance and Economic Growth," Journal of Financial
Intermediation 8, pp. 8-35.
Levine, R. and S. Zervos, 1998, "Stock Markets, Banks and Economic Growth,"
American Economic Review, 88 (3), pp. 537-56.
Love, I., 1999, "The Capital Accumulation Identity-How Well does it Hold in the
Data: Evidence from Compustat and Worldscope Databases," Mimeograph, Columbia
University.
Lucas, R.E., 1988, "On the Mechanics of Economic Development," Journal of
Monetary Economics XXII, pp. 3-42.
Modigliani, F. and M. Miller, 1958, "The Cost of Capital, Corporation Finance,
and the Theory of Investment," American Economic Review, 48 (3), pp. 261-97.
36
Myers, S. and N. Majluf, 1984, "Corporate Financing and Investment Decisions
When Firms Have Information That Investors Do Not Have," Journal of Financial
Economics, 13(2), pp. 187-221.
Neusser, K. and M. Kugler, 1998, "Manufacturing Growth and Financial Devel-
opment: Evidence from OECD countries," Review of Economics and Statistics. 80
pp. 636-46.
Opler, T., L. Pickowitz, R. Stultz, and R. Williamson, 1999, "The Determinants
and Implications of Corporate Cash Holdings," Journal of Financial Economics, 52(1),
pp. 3-46.
Rajan, R. and L. Zingales, 1998, "Financial Dependence and Growth," American
Economic Review, 88(3), pp. 559-86.
Rousseau, P. and P. Wachtel, 1998, "Financial Intermediation and Economic Per-
formance: Historical Evidence from Five Industrialized Countries," Journal of Money
Credit and Banking 30(4), pp. 657-78.
Schiantarelli, F., 1995, "Financial Constraints and Investment: A Critical Review
of Methodological Issues and International Evidence" in Peek, J., and E. Rosengren
eds. "Is bank lending important for the transmission of monetary policy?" Conference
Series, no. 39. Federal Reserve Bank of Boston, pp. 177-214.
Schumpeter, J.A., orig. 1912, The Theory of Economic Development, trans. 1934
(Harvard U.P., Cambridge, MA).
Stulz, R., 2000, "Financial Structure, Corporate Finance and Economic Growth,"
International Review of Finance, (2000) pp. 11-38.
Whited, T., 1992, "Debt, Liquidity Constraints, and Corporate Investment: Evi-
dence from Panel Data," Journal of Finance, 47 (4), pp. 1425-60.
Wurgler, J., 2000, "Financial Markets and Allocation of Capital," Journal of Fi-
nancial Economics, October 2000.
Zingales, L., 2000, "In Search of New Foundations," Journal of Finance, 55(4),
pp.1623-53.
37
Table 1. Sample Coverage Across Countries
Country Number of Percent of total Number Average number
Country code observations observations of firms of years per firm
Argentina AR 198 0.003 28 7.1
Austria AT 454 0.008 55 8.3
Australia AU 1571 0.026 197 8.0
Belgium BE 561 0.009 71 7.9
Brazil BR 687 0.012 94 7.3
Canada CA 3382 0.057 391 8.6
Switzerland CH 1043 0.017 132 7.9
Chile CL 411 0.007 55 7.5
Colombia CO 150 0.003 20 7.5
Germany DE 3970 0.067 468 8.5
Demnark DK 1045 0.018 126 8.3
Spain ES 947 0.016 114 8.3
Finland Fl 747 0.013 84 8.9
France FR 3274 0.055 402 8.1
United Kingdom GB 9931 0.166 1129 8.8
Hong Kong HK 969 0.016 142 6.8
Indonesia ID 531 0.009 84 6.3
Ireland IE 427 0.007 47 9.1
Israel IL 152 0.003 29 5.2
India IN 1507 0.025 269 5.6
Italy IT 1149 0.019 132 8.7
Japan JP 4646 0.078 624 7.4
South Korea KR 1264 0.021 187 6.8
Mexico MX 502 0.008 69 7.3
Malaysia MY 1476 0.025 205 7.2
Netherlands NL 1280 0.021 147 8.7
Norway NO 680 0.011 84 8.1
New Zealand NZ 315 0.005 43 7.3
Peru PE 101 0.002 17 5.9
Philippines PH 271 0.005 43 6.3
Pakistan PK 418 0.007 72 5.8
Portugal PT 254 0.004 42 6.0
Sweden SE 1162 0.019 137 8.5
Singapore SG 841 0.014 122 6.9
Thailand TH 1045 0.018 177 5.9
Turkey TR 145 0.002 23 6.3
Taiwan TW 405 0.007 83 4.9
USA US 10422 0.175 1247 8.4
Venezuela VE 81 0.001 11 7.4
South Africa ZA 1151 0.019 135 8.5
Total 59565 7537
Average number of firms per country 188
Average number of firms per country, excluding US and GB 136
Median number of firms per country, excluding US and GB 114
Table 2. Variable Definitions:
Abbreviation Description
Firm Level variables (from Worldscope)
PPENT Property Plant and Equipment, net of depreciation
CAPEX Capital expenditure
DA Depreciation and Amortization expense
K Beginning period capital = PPENT-CAPEX+DA
IK, I/K Investment to Capital ratio = CAPEX / K
SK, S/K Sales to Capital ratio = Sales / K
Cash Cash plus equivalents scaled by Total Assets (or scaled by K for robustness checks)
CF Cash Flow (Net income + DA), scaled by K
COGS Cost of goods sold, scaled by K
Size Log of total assets in US dollars
Rank Ranking based on size of PPENT (first, ranked by year, then averaged over the years), largest firm
in each country has rank equal to one (described in section 5.1).
Weight Weight is a country-level variable equal to one over the number of valid observations per country
(described in section 5. 1).
Industry For manufacturing industries the dummies are on a two digit SIC level and for the rest of
dummies industries they are on a one digit level.
Country-Level variables
STKMKT Stock market development is Index 1 from Demirguc-Kunt and Levine (1996), equals to the sum
of (standardized indices of) market capitalization to GDP, total value traded to GDP, and turnover
(total value traded to market capitalization).
FININT Financial intermediary development is Findexl from Demurguc-Kunt and Levine (1996), equals to
the sum of (standardized indices of) ratio of liquid liabilities to GDP, and ratio of domestic credit
to private sector to GDP.
FD Financial Development = STKMKT+FININT.
Legal Origin Country's legal origin categorized into 4 groups: English, French, German or Scandinavian, from
LLSV (1998).
Efficiency, Efficiency of legal system and Rule of Law are two measure of the quality of law enforcement,
Rule of Law from LLSV (1998).
Expropriation Risk of expropriation is the risk of outright confiscation or forced confiscation by the government,
from LLSV (1998).
Corruption The measure of corruption, from LLSV (1998).
GNP PC Log of GNP per capita in US dollars in 1994, World Development Report 1996.
grGDP Annual real growth rate of GDP, IFS
Table 3. Descriptive Statistics for Key Variables
Summary statistics by country for main variables. Variables definitions are given in Table 2. Outliers (far away Max or
Min) are underlined.
Cash I/K S/K Financial Development
Country mean median mean median mean median FD FININT STKMKT
Argentina 0.08 0.04 0.19 0.13 1.6 1.2 -1.38 -0.79 -0.59
Austria 0.10 0.07 0.25 0.20 4.4 3.0 -0.27 -0.12 -0.15
Australia 0.08 0.05 0.26 0.18 3.3 2.3 0.42 0.23 0.19
Belgium 0.10 0.08 0.25 0.20 4.0 3.7 -0.82 -0.35 -0.47
Brazil 0.08 0.04 0.12 0.09 1.7 1.0 -1.04 -0.75 -0.29
Canada 0.07 0.02 0.23 0.18 3.1 1.5 0.03 -0.06 0.09
Switzerland 0.14 0.11 0.23 0.15 3.9 2.6 2.2 1.45 0.75
Chile 0.07 0.04 0.21 0.16 1.6 1.3 -0.75 -0.29 -0.46
Colombia 0.08 0.04 0.26 0.14 3.6 1.9 -1.6 -0.72 -0.88
Germany 0.08 0.05 0.31 0.25 5.5 4.8 1.68 0.3 1.38
Denmark 0.15 0.14 0.24 0.21 4.4 3.7 -0.49 -0.12 -0.37
Spain 0.06 0.04 0.14 0.09 2.8 1.8 -0.14 0.11 -0.25
Finland 0.09 0.08 0.38 0.21 5.0 3.2 -0.41 0.12 -0.53
France 0.12 0.09 0.27 0.21 6.5 6.0 0.1 0.31 -0.21
UK 0.09 0.06 0.22 0.17 4.6 4.0 1.68 0.45 1.23
Hong Kong 0.16 0.11 0.25 0.16 3.5 2.1 2.01
Indonesia 0.14 0.10 0.37 0.23 3.9 2.6 -1.17 -0.46 -0.71
Ireland 0.15 0.12 0.26 0.17 4.4 3.1 -0.45
Israel 0.11 0.10 0.30 0.24 3.7 3.0 0.01 -0.07 0.08
India 0.04 0.03 0.27 0.19 3.5 2.6 -0.7 -0.44 -0.26
Italy 0.12 0.09 0.26 0.17 4.4 3.0 -0.64 -0.13 -0.51
Japan 0.19 0.17 0.22 0.19 4.1 3.3 3.3 1.31 2.02
South Korea 0.08 0.06 0.31 0.23 3.9 3.0 0.84 -0.21 1.05
Mexico 0.08 0.06 0.11 0.10 1.6 1.3 -0.85 -0.71 -0.14
Malaysia 0.08 0.05 0.23 0.16 2.7 1.8 1.19 0.29 0.9
Netherlands 0.10 0.05 0.24 0.20 5.1 3.8 0.66 0.34 0.32
Norway 0.14 0.12 0.33 0.22 3.6 2.1 -0.15 0.03 -0.18
New Zealand 0.04 0.02 0.17 0.13 3.3 2.8 -0.53 -0.2 -0.33
Peru 0.09 0.04 0.22 0.14 1.5 1.3
Philippines 0.12 0.07 0.37 0.22 2.6 1.4 -1.15 -0.61 -0.54
Pakistan 0.11 0.04 0.26 0.19 4.8 2.5 -1.28 -0.46 -0.82
Portugal 0.06 0.03 0.22 0.13 3.3 2.0 -0.67 -0.06 -0.61
Sweden 0.12 0.08 0.31 0.19 4.7 3.8 -0.31 -0.21 -0.1
Singapore 0.19 0.15 0.28 0.22 3.6 2.5 1.6 0.56 1.04
Thailand 0.06 0.03 0.40 0.23 4.5 2.5 0.36 -0.02 0.38
Turkey 0.13 0.07 0.56 0.50 7.5 5.8 -1.2 -0.59 -0.61
Taiwan 0.13 0.09 0.20 0.14 2.5 1.8 0.64
US 0.09 0.04 0.24 0.19 4.8 3.8 1.35 0.14 1.21
Venezuela 0.09 0.06 0.21 0.13 1.5 1.1 -1.26 -0.52 -0.74
South Africa 0.09 0.06 0.22 0.19 4.7 3.8 0.25 -0.23 0.48
Mean 0.10 0.07 0.26 0.19 3.74 2.73 -0.03 -0.06 0.09
Median 0.09 0.06 0.25 0.19 3.78 2.60 -0.29 -0.12 -0.18
Std 0.04 0.04 0.08 0.06 1.33 1.20 1.14 0.51 0.79
Table 4. Correlations
Correlations of country-level means and medians of the firmn level variables and country's institutional characteristics.
Variables definitions are in Table 2. Panel A: below the diagonal are Pearson correlation coefficients, with two outlier
countries excluded: JP (Japan) and TR (Turkey). Above the diagonal are Spearnan correlations (robust to outliers) with all
observations included. Including outliers for Pearson correlations results in significant correlation for FD and Cash (due to JP
which is an outlier on both of these) and nonsignificant correlation for SK and FD (due to TR, which has very high SK and
low FD). Panel B: pearson correlations with all countries. (Excluding Japan makes correlaitons of GDP PC with FD and
FININT significant at 6% and 2 % respectively; also correlation between FININT and Log GDP becomes insignificant.) P-
values are in parenthesis; bold are significant at 5% or better, underlined are significant at 10%.
Panel A. Cross-Country Correlations of Firm Level Variables
Country Means Country Medians
FD Cash IK SK FD Cash IK SK
FD 0.15 0.02 0.34 0.21 0.21 0.44
(0.35) (0.87) (0.04) (0.22) (0.22) (0.006)
Cash 0.21 0.32 0.36 0.21 0.36 0.32
(0.23) (0.04) (0.02) (0.22) (0.02) (0.04)
IK 0.09 0.27 0.55 0.29 0.35 0.59
(0.59) (0.099) (0.0002) (0.096) (0.03) (0.0001)
SK 0.41 0.24 0.49 0.45 0.23 0.56
(0.015) (0.14) (0.002) (0.008) (0.15) (0.0003)
Panel B. Correlations of Country-Level Institutional Characteristics
FD FINTNT STKMKT Efficiency Corruption Expropr. Accounting GNPPC
FININT 0.90
(0)
STKMKT 0.95 0.73
(0) (0)
Efficiency 0.52 0.52 0.45
(0.001) (0.001) (0.005)
Corruption 0.51 0.55 0.40 0.83
(0.001) (0.001) (0.01) (0)
Expropriation 0.57 0.64 0.39 0.72 0.83
(0.001) (0) (0.02) (0) (0)
Accounting 0.34 0.28 0.34 0.31 0.41 0.36
(0.05) (0.09) (0.04) (0.06) (0.01) (0.03)
GNP PC 0.56 0.61 0.46 0.74 0.87 0.84 0.46
(0) (0.0) (0.004) (0) (0) (0) (0.005)
GDP US 0.54 0.41 0.49 0.20 0.26 0.46 0.15 0.42
(0) (0.01) (0.002) (0.22) (0.11) (0.003) (0.39) (0.01)
Table 5. Main Results on Financial Development and Financing Constraints
The dependent variable is IK,, the model is given in (10); variable definitions are in Table 2. The estimation is by GMM (IV),
country-time and fixed effects are removed prior to estimation (see Section 3.1). Instruments are first and second lags of IK, SK,
Cash, CFK, COGS, interactions of FD with IK, SK and Cash, and industry dummies. The firms are ranked based on the size of
PPENT (described in Section 5). In the weighted regression, weights are equal to a value of one divided by the number of
observations per country. Structural parameters as functions of estimated coefficients are given in (11). They are identified using
minimum distance estimator (see Section 5.1). The Hansen test is a test of overidentifying restrictions; reported are p-values
(this test is not available for weighted regressions). Heteroskedasticity adjusted standard errors in parentheses; ***,**,* and a
represent significance at 1%, 5%, 10% and 15% respectively.
Model: 1 2 3 4 5
50 largest 100 largest 150 largest 200 largest All, weighted
I'K,+, 0.688 0.671 0.543 0.571 0.273
(0.171) .. (0.137) (0.132) .. (0.125)*4* (0.135)
/Kt- 0.201 0.200 0.208 0.204 0.203
(0.022) (0.016) .. (0.014) (0.014) (0.018) ..
S/K, 0.011 0.018 0.020 0.020 j 0.042
(0.011) (0.007) (0.007) (0.006) (0.009)
Cash,-, 0.132 0.081 0.124 0.102 0.174
(0.064) (0.051)a (0.046) .. (0.048)* (0.062)..
Casht i*FDc -0.119 -0.136 -0.110 -0.082 I -0.149
(0.046)* (0.039) (0.035) (0.034) (0.046)
Constant 0.000 0.000 -0.001 -0.001 j -0.003
(0.001) (0.001) (0.000) (0.000) I (0.002)*
N obs 6488 10477 12474 13922 ! 21278
N firms 1436 2335 2791 3111 I 4794
0.000 0.010 0.103 0.086 j 0.208
RootMSE 0.127 0.128 0.121 0.123 ! 0.130
Hansen test 0.340 0.001 0.001 0.002 NA
Structural parameters:
beta 0.935 0.908 0.708 0.750 ' 0.331
(0.288) (0.227) .. (0.201) .. (0.193) (0.173)
g 0.240 0.238 0.238 0.235 i 0.215
(0.034) (0.024) (0.020) .. (0.019) (0.022)
alfa 16.825 9.719 7.069 7.600 1.718
(20.454) (5.195) (3.978) (4.091) * (1.135)
al 2.371 0.855 1.189 1.000 j 0.802
(2.428) (0.601)a (0.589) (0.572)* (0.341)
a2 -2.129 -1.448 -1.052 -0.798 j -0.685
(2.139) (0.666)* (0.525) (0.458) (0.266) ..
Table 6. Size Effect
The dependent variable is IKt; the models are described in section 6.1. The Size is equal to the (log of) total assets in US
dollars in models 1-3 and "Small" dummy in model 4 (durnmy is equal to one if total assets are less than the country's own
median level of total assets). The estimation is by GMM (IV), country-time and fixed effects are removed prior to
estimation (see Section 3.1). Instruments are first and second lags of IK, SK, Cash, CF, COGS, size and size interactions
with Cash, IK and SK, interactions of FD with IK, SK, Cash, and size, and industry dummies. All the regressions are
weighted regressions, weights are equal to a value of one divided by the number of observations per country.
Heteroskedasticity adjusted standard errors in parentheses; ***,**,* and a represent significance at 1%, 5%. 10% and 15%
respectively.
Model: 1 2 3 4
I/Kt+1 0.484 0.310 0.295 0.348
(0.121) (0.137) (0.137) (0.129)
1/K,-} 0.201 0.202 0.202 0.203
(0.018) (0.019) (0.019) .. (0.018)
S/K, 0.035 0.039 0.040 0.036
(0.009) (0.009) .. (0.009) (0.009)
Cash,, 0.616 0.721 0.870 0.081
(0.194) (0.204) (0.233) (0.075)
Casht.l*Size,, -0.095 -0.105 -0.134 0.200
(0.032) (0.034) ... (0.040) ... (10.110) *
Cash, ,*FDC -0.146 -0.700 -0.048
(0.047) (0.201) (0.060)
Cash,- *Size *FD, 0.099 -0.206
(0.032) (0.095)
Constant -0.001 -0.002 -0.002 -0.002
(0.001) (0.002) (0.002) (0.002)
N obs 21777 21278 21278 21348
N firms 4934 4794 4794 4794
R2 0.137 0.198 0.1998 0.1865
Root MSE 0.136 0.131 0.131 0.133
Table 7. Sample Splits
The dependent variable is IKl; variable definitions are in Table 2. High FD and Low FD are samples split on the
median FD (reported in Table 3). The estimation is by GMM (IV), country-time and fixed effects are removed
prior to estimation (see Section 3.1). Instruments are first and second lags of IK, SK, Cash, CFK, COGS and
industry dummies. All the regressions are weighted regressions, weights are equal to a value of one divided by the
number of observations per country. Heteroskedasticity adjusted standard errors in parentheses; ***,**,and *
represent significance at 1%, 5% and 10% respectively.
High FD Low FD
Model: 1 2 3 4
I/Kt,l 0.505 0.472 0.427 0.424
(0.190) (0.175) (0.126) .. (0.122)
I/Kf 0.209 0.211 0.198 0.198
(0.018) (0.018) ... (0.030) (0.029)
S/K, 0.025 0.025 0.048 0.048
(0.008)''' (0.008) .. (0.015)''' (0.015)'''
Casht-l 0.014 0.197 0.262 1.124
(0.051) (0.180) (0.104) (0.364)..
Casht.i*Size,l -0.033 -0.165
(0.030) (0.061)
Constant 0.000 0.000 -0.003 -0.003
(0.002) (0.001) (0.003) (0.003)
obs 18106 18106 3671 3671
N firns 3930 3930 1004 1004
R2 0.131 0.147 0.171 0.169
Root MSE 0.128 0.127 0.141 0.142
Table 8. Business Cycles and Financing Constraints
The dependent variable is IK, the model is given in (10) with added interaction of cash stock with country-year real GDP
growth rate; variable definitions are in Table 2. The estimation is by GMM (IV), country-time and fixed effects are
removed prior to estimation (see Section 3.1). Instrunents are first and second lags of IK, SK, Cash, CFK, COGS,
interactions of FD and GDP growth with IK, SK and Cash, and industry dummies. The firrns are ranked based on the size
of PPENT (described in Section 5). In the weighted regression, weights are equal to a value of one divided by the number
of observations per country. Heteroskedasticity adjusted standard errors in parentheses; ***,**,* and a represent
significance at 1%, 5%, 10% and 15% respectively.
150 largest All, weighted
Model: 1 2 3 4
I/K,+}1 0.705 0.517 0.427 0.360
(0.145) (0.135) (0.116) (0.125)
U/K,., 0.198 0.202 0.193 0.194
(0.014)"' (0.014) (0.018) (C.018)
S/K, 0.018 0.021 0.026 0.025
(0.006) (0.006) (0.006) (0.007)
Cash,1 0.117 0.186 0.203 0.232
(0.043) "' (0.048) (0.062) (0.068)
Cash,-l*grGDP,, -1.720 -1.827 -1.578 l -2.424
(0.881) (0.870) (1.150) (1.228)
CashtI*FDc -0.108 -0.117
(0.037) (0.049) 2
Constant -0.001 -0.002 -0.002 -0.003
(0.001) (0.001)a (0.001) (0.002)
N obs 12923 12411 22061 21549
N firTns 2935 2794 4973 4832
R2 0.000 0.126 0.145 0.167
Root MSE 0.131 0.122 0.136 0.136
'is significant at 17%
2in model IV the FD interaction is significant at 2% and grGDP interaction is significant at 5%
Table 9. Legal System Indicators and Financing Constraints
The dependent variable is IK,, the "baseline" model is given in (10) with FD interactions replaced or supplemented
with each of the Indicator variable interactions (the rest of coefficients are not reported). Variable definitions are in
Table 2. The estimation is by GMM (IV), country-time and fixed effects are removed prior to estimation (see Section
3.1). Instruments are first and second lags of IK, SK, Cash, CFK, COGS, interactions of FD and appropriate Indicator
with IK, SK and Cash, and industry dummies. All the regressions are weighted regressions, weights are equal to a
value of one divided by the number of observations per country. Heteroskedasticity adjusted standard errors in
parentheses; " andb represent significance at 1%, 5%, 10%, 15% and 20% respectively.
Cash,-, Cash,-,*Indicator, Casht, *FD,
Model: Indicator:
I Efficiency 0.587 -0.052
0.234 0.027
2 Efficiency 0.338 b -0.021 -0.123
0.245 0.030 0.050
3 Rule of Law 0.633 -0.060
0.245 0.027
4 Rule of Law 0.455 -0.037 b -011 1
0.240 0.028 0.045
5 Corruption 0.670 -0.066
0.254 0.029
6 Corruption 0.479 -0.041" -0.100
0.260 0.031 0.045
7 Expropriation 1.208 -0.119
0.480 0.051
8 Expropriation 0.850 -0.078 a -0.097
0.495 0.054 0.045
9 Accounting 0.793 _0.010 a
0.437 0.006
10 Accounting 0.364 -0.003 -0.118
0.448 0.006 0.042
11 French 0.089 a 0.191
0.060 0.114
12 French 0.189 -0.044 -0.163
0.078 0.132 0.052
13 English 0.230 -0.191
0.080 0.100
14 English 0.177 -0.037 -0.141
0.072 0.096 0.046
Table 10. Legal Origin as Instrument for Financial Development
The dependent variable is IK,, the model is given in (10); variable definitions are in Table 2. The estimation is by GMM
(IV), country-time and fixed effects are removed prior to estimation (see Section 3.1). Instruments are first and second lags
of IK, SK, Cash, CFK, COGS, interactions of Legal Origin dummies with IK, SK and Cash, (note that Legal Origin
replaces FD in the intsrument set). The firms are ranked based on the size of PPENT (described in Section 5). In the
weighted regression, weights are equal to a value of one divided by the number of observations per country. The Hansen test
is a test of overidentifying restrictions, reported are p-values (this test is not available for weighted regressions).
Heteroskedasticity adjusted standard errors in parentheses; ***,**,* and a represent significance at 1%, 5o/, 10% and 15%
respectively.
Model: 1 2 3 4 5
50 largest 100 largest 150 largest 200 largest All, weighted
l/K,+, 0.627 0.692 0.665 0.670 0.473
(0.109) (0.095) (0.090) (0.084) (0.100)
1IK. 0.207 0.205 0.209 0.204 0.202
(0.020) (0.016) (0.014) (0.014) (0.019)
S/K, 0.016 0.017 0.015 0.016 0.029
(0.008) (0.006) (0.006) (0.005) (0.007)'--
Cash,-, 0.118 0.078 0.109 0.093 0.144
(0.058) (0.049) (0.048) (0.049) (0.059)
Casht,*FDc -0.108 -0.133 -0.118 -0.099 -0.148
(0.051) (0.043) ... (0.041) (0.039) ... (0.058)
Constant -0.001 -0.001 -0.001 -0.001 -0.001
(0.001) (0.001) (0.001) (0.001) (0.001)
N obs 6499 10502 12498 13961 21348
N firms 1433 2332 2788 3108 4794
2 0.042 0.004 0.026 0.020 0.133
RootMSE 0.124 0.129 0.127 0.129 0.137
Hansen test 0.197 0.019 0.029 0.036 N/A
Table 11. Single Country Regressions
The dependent variable is IKt; the model is given in (15), Section 7.1; variable definitions are in Table 2. The
estimation is by GMM (IV), country-time and fixed effects are removed prior to estimation (see Section 3.1).
Instruments are first and second lags of IK, SK, Cash, CFK, COGS, and industry dummies. Constants are
included, but not reported since they are very close to zero and never significant. Heteroskedasticity adjusted
standard errors in parentheses; ***, *, * and a represent significance at 1%, 5%, 10% and 15% respectively.
IfK, 1IKt.L S/K_ Casht-. Number
Code Coeff. St.error Coeff. St.error Coeff. St.error Coeff. St.error of obs.
AR 0.96 0.318 0.21 0.096 -0.097 0.057 0.77 0.600 69
AT 0.40 0.467 0.17 0.051 -0.004 0.018 0.34 0.230 199
AU 0.80 0.321 0.26 0.042 -0.001 0.009 0.17 0.131 654
BE 0.26 0.178 ' 0.14 0.097 0.087 0.032 0.71 0.336 215
BR 0.46 0.120 0.36 0.076 0.003 0.007 0.11 0.055 206
CA 0.62 0.130 0.28 0.044 0.033 0.011 0.01 0.102 1500
CH 0.41 0.304 ' 0.22 0.083 0.047 0.018 -0.10 0.113 438
CL 0.51 0.211 0.37 0.184 0.033 0.048 0.45 0.411 162
CO 0.33 0.089 0.00 0.116 0.030 0.025 0.70 0.612 35
DE 0.97 0.107 0.18 0.036 0.000 0.006 -0.06 0.099 1825
DK 0.23 0.209 0.18 0.039 0.028 0.016 0.40 0.195 462
ES 0.78 0.100 0.25 0.061 -0.002 0.016 0.07 0.185 386
Fl 0.76 0.243 0.23 0.049 0.021 0.015 ' -0.01 0.349 297
FR 0.63 0.125 0.16 0.037 0.012 0.008 0.05 0.121 1358
GB 0.58 0.157 0.15 0.024 0.022 0.006 0.16 0.062 4084
HK 0.56 0.316 0.18 0.081 0.053 0.027 0.25 0.162 245
ID 0.12 0.295 -0.06 0.078 0.040 0.042 0.44 0.282 149
IE 0.72 0.136 . 0.20 0.062 0.014 0.014 0.08 0.258 167
IL 0.37 0.143 0.09 0.077 0.054 0.047 -0.03 0.477 32
IN 0.46 0.145 0.08 0.067 0.035 0.016 0.27 0.653 315
IT 0.63 0.177 0.25 0.056 0.029 0.018 0.04 0.106 521
JP 0.83 0.143 0.28 0.040 0.005 0.009 -0.06 0.086 1263
KR 0.39 0.205 0.27 0.089 0.027 0.018 -0.75 0.558 136
MX 0.76 0.137 0.14 0.056 0.033 0.020 0.06 0.236 187
MY 0.86 0.284 0.19 0.072 0.028 0.020 ' -0.30 0.233 450
NL 0.50 0.119 0.29 0.066 0.017 0.009 -0.03 0.152 589
NO 0.65 0.114 0.24 0.094 -0.003 0.009 0.11 0.227 262
NZ 0.32 0.255 0.15 0.095 0.021 0.026 -0.10 0.226 117
PE 0.49 0.128 0.49 0.108 0.071 0.043 -0.05 0.414 25
PH 0.60 0.271 0.39 0.087 0.040 0.032 0.48 0.428 65
PK 0.34 0.268 0.23 0.086 0.020 0.025 0.87 0.928 103
PT -0.15 0.180 0.05 0.072 0.062 0.041 0.75 0.367 48
SE 0.81 0.152.. 0.23 0.058 0.024 0.013 0.31 0.162 473
SG 0.90 0.214 0.18 0.079 0.018 0.012 a -0.36 0.324 229
TH 0.18 0.184 0.14 0.085 0.013 0.014 0.22 0.229 236
TR 0.19 0.141 0.40 0.224 0.045 0.015 0.06 0.597 17
TW 0.60 0.255 0.05 0.181 0.100 0.045 0.39 0.535 69
Us 0.55 0.114 0.28 0.036 0.030 0.007 0.25 0.069 3942
VE 0.34 0.189 0.60 0.099 am 0.081 0.021 0.22 0.537 22
ZA 0.48 0.123 0.26 0.065 0.007 0.005 0.01 0.104 516
Mean 0.53 0.19 0.22 0.08 0.027 0.02 0.17 0.30 552
Quartiles:
25% 0.36 0.13 0.15 0.05 0.013 0.01 -0.01 0.13 114
50% 0.53 0.18 0.21 0.07 0.027 0.02 0.11 0.23 233
75% 0.73 0.25 0.27 0.09 0.040 0.03 0.35 0.42 484
Figure 1. Cash Coefficients and Financial Development
PK
AR BE PT
.5 % JD
c~~coB
4- VE-CLHU
0) THEu
0 AU GB
TR MX
0)~~~~~~~~~~~I
o 0-FiD j
o N CH
(U MY
o ~~~~~~~~~~~~~~SG
-.5
KR
-1 0 1 2
Financial Development
Regression line: -0.18, significant at 1% (the standard error is 0.05 and R squared is 0.35)
Without KR: -0.16, significant at 1% (the standard error is 0.04 and R squared is 0.37)
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS2681 On the Duration of Civil War Paul Collier September 2001 P. Collier
Anke Hoeffler 88208
Mans Soderbom
WPS2682 Deposit Insurance and Financial Robert Cull September 2001 K. Labrie
Development Lemma W. Senbet 31001
Marco Sorge
WPS2683 Financial Policies and the Prevention Frederic S. Mishkin October 2001 R. Vo
of Financial Crises in Emerging 33722
Market Economies
WPS2684 From Monetary Targeting to Inflation Frederic S. Mishkin October 2001 R. Vo
Targeting: Lessons from Industrialized 33722
Countries
WPS2685 Monetary Policy Strategies for Frederic S. Mishkin October 2001 R. Vo
Latin America Miguel A. Savastano 33722
WPS2686 Education, Earnings, and Inequality Andreas Blom October 2001 S. Benbouzid
in Brazil, 1982-98: Implications for Lauritz Holm-Nielsen 88469
Education Policy Dorte Verner
WPS2687 Geographic Patterns of Land Use Kenneth M. Chomitz October 2001 S. Hendrickson
and Land Intensity in the Brazilian Timothy S. Thomas 37118
Amazon
WPS2688 Aid, Shocks, and Growth Paul Collier October 2001 A. Kitson-Walters
Jan Dehn 33712
WPS2689 Global Trade and Food Safety: John S. Wilson October 2001 L. Tabada
Winners and Losers in a Fragmented Tsunehiro Otsuki 36896
System
WPS2690 Ringing in the 20th Century: Scot Wallsten October 2001 P. Sintim-Aboagye
The Effects of State Monopolies, 37644
Private Ownership,and Operating
Licenses on Telecommunications
in Europe, 1892-1914
WPS2691 Evolution of Earnings and Rates of Gladys L6pez-Acevedo October 2001 M. Geller
Returns to Education in Mexico 85155
WPS2692 Introduction to Property Theory: David Ellerman October 2001 B. Mekuria
The Fundamental Theorems 82756
Policy Research Working Paper Series
Contact
Title Author Date for paper
WPS2693 Helping People Help Themselves: David Ellerman October 2001 B. Mekuria
Toward a Theory of Autonomy- 82756
Compatible Help