Causes and Effects of Financing

Constraints at the Firm Level P. Winker

ABSTRACT. This paper focuses on the empirical assessment profit maximization might lead to interest rates
of determinants and effects of financing constraints at the firm lower than the market equilibrium rate which, in
level. Using a standard model of credit rationing based on
asymmetric information firm age and size are found to be
turn, results in restricted access to bank loans, at
factors which should influence the probability of financing least for some firms.
constraints. Improving business conditions strengthen the If the incidence of credit constraints is not
degree of informational asymmetry. A unique panel of firm purely random, but rather concentrated on some
data for Germany, including direct information on financing groups of firms,2 and if firms have to react to such
constraints, is used for the econometric analysis. Firms’ size
and improving business conditions are found to have a
constraints due to an overall financing deficit,
significant effect. Furthermore, a significant impact on invest- changes in bank loan supply can have real effects.
ment and R&D expenditures cannot be rejected. The focus of this paper, therefore, is to identify
characteristics of firms more likely to face
restricted access to bank loans, and to test whether
1. Introduction this restriction imposes reactions on their invest-
ment and R&D expenditures.
If bank loans and other forms of external and
Among the characteristics which might influ-
internal financing are not perfect substitutes at the
ence the rationing probability are the age and size
firm level, restricted access to bank loans may
of firms, as well as their ongoing business rela-
result in a financing deficit. The approach of
tions to banks. There is large theoretical literature
asymmetric information, formalized for the loan
on this topic and it is beyond the scope of this
market by Stiglitz and Weiss (1981), offers a
paper to give an extensive review.3 Instead, it is
rationale for the existence of such a limited access
argued that the impact of firm age can be derived
to external finance. It is assumed that banks can
directly from the textbook Stiglitz and Weiss
classify firms according to their creditworthiness
model. Furthermore, a slight extension of the
only at a coarse level. The classification is based
model allows us to analyze the impact of business
on observable features, such as past success, age
expectations at the firm and sectoral level.
or duration of business relations. The actual prob-
The main contribution of this paper lies in the
ability of a single firm repaying its loan is
empirical assessment of the causes of financing
unknown to the banks; they merely know the dis-
constraints and their real effects. A unique firm
tribution of the repayment probabilities within a
panel data set for the German manufacturing
given group of firms. Under this setting, the loan
sector supplied by the ifo institute for economic
rate does not only influence the quantity of loan
research, Munich, is used for this purpose. In
demand, but also its quality due to adverse selec-
contrast to most of the data sets used in prior
tion or adverse incentive effects.1 Consequently,
research,4 the data here contain direct information
on the existence of financing deficits at the firm
Final version accepted on July 30, 1998 level.5 This feature of the data represents a distinct
advantage for two reasons. First, it avoids the use
Department of Economics of proxies for the existence of financing con-
University of Mannheim
D-68131 Mannheim straints, which was subject to the recent contro-
Germany versy about the Fazzari, Hubbard and Petersen
E-mail: [email protected] (1988) results in Kaplan and Zingales (1997) and

Small Business Economics 12: 169–181, 1999.

 1999 Kluwer Academic Publishers. Printed in the Netherlands.
170 P. Winker

Fazzari, Hubbard and Petersen (1996). Second, it Consequently, an increase of the interest rate will
enables us to analyze cause and effect of financing lead to a decrease in the overall repayment prob-
constraints separately. ability for the firms still requesting a loan. Adverse
The data for 1919 firms or operational units and selection takes place, which might be attributed to
the time period 1982–1991 also include informa- unrevealable information. Eventually, a further
tion on selected firm characteristics, such as increase in the interest rate will lead to a decrease
employment, sales, sector, investment and expen- in the bank’s expected return.
ditures for innovative activities. Furthermore, If firms can choose between projects with
estimates of the current business situation and differing risks and banks only know the distribu-
business expectations are covered. However, some tion of the risks over the projects, a similar effect
shortcomings of the data set discussed in section may arise: An increase of the interest rate will
3 require the use of proxies for some variables of give an incentive to choose riskier projects. The
interest, e.g. firm age. expected return to the bank might eventually
The paper is organized in the following way: decrease due to this effect. These adverse incen-
Section 2 provides a sketch of the Stiglitz and tives are also called “moral hazard”. They might
Weiss model and its extension, which allow us to also be described as effects of hidden action.
derive some hypotheses on the determinants of
financing constraints. The ifo firm panel used for
2.2. The Stiglitz-Weiss model
the empirical assessment is described in section 3.
The econometric models for these microdata are In order to derive the impact of some firm
presented in section 4; and the estimation results characteristics on the risk of facing financing
are discussed in section 5. constraints, a sketch of the basic model by Stiglitz
and Weiss (1981) for the adverse selection effect
is presented (cf. Blanchard and Fischer (1989), pp.
2. Theory of credit rationing
480ff, for some details).
Each firm, out of a continuum of firms,
2.1. Adverse selection and adverse incentives
disposes of exactly one project which requires
The more recent literature on credit rationing an initial endowment of K, where firms’ equity
stresses the argument of incomplete or asymmetric W < K. Consequently, external finance of B [
information (cf. Jaffee and Stiglitz (1990) and the K – W is necessary. It is offered by identical,
literature cited therein). Potential debtors know the competitive, risk neutral and expected profit
probability distribution of the returns of their own maximizing banks in form of standard loan con-
projects whereas the banks know only the aggre- tracts at interest rate r.6
gate distribution for all firms in a specific group. The project of firm i succeeds with probability
In standard loan contracts, the bank’s expected pi and gives a return of Ris. The return to a failed
return depends on the interest rate charged and project is equal to Rf for all firms. While the firms
on the repayment probability, since, in case of differ with regard to the success probability, the
insolvency, repayment is limited. If, for a group expected gross return is identical and equal to
of firms with identical expected gross returns, the R = pi Ris + (1 – pi)Rf for all projects. Consequently,
mean of the repayment probability is given, the a high return to a successful project Ris corresponds
bank’s expected return will decrease with a to a low success probability pi. If the project of a
growing variance of the repayment probabilities, firm consists of an investment plan including
while a firm’s expected net return, i.e. the gross several smaller projects, risk diversification will
return minus the repayment to the bank, will lead to higher overall success probability pi.
increase with the variance of the gross returns. The expected net return of firm i is given by
Now, consider a given group of firms with
E(pi) = pi[Ris – (1 + r)B] (1)
interest elastic loan demand. If a bank increases
= R – Rf – pi[(1 + r)B – Rf]. (2)
its interest rate, the expected net return for firms
with relatively riskless projects will become zero Thus, for given expected gross return R the
or negative. They will no longer seek a loan. expected net return of firm i decreases with the
 Causes and Effects of Financing Constraints at the Firm Level 171

success probability pi. Consequently, firms with 2.3. A simple cohort model
riskier projects will ceteris paribus accept higher
interest rates on loans. If they have the alterna- Despite these adverse selection problems, banks
tive to invest their initial endowment W in a secure may still discriminate groups of firms according
asset, with rate of return r, risk neutral entrepre- to some observable characteristics. Both the
neurs will prefer their investment project as long as theoretical and empirical literature stress the
importance of the age of firms as mentioned in the
E(pi) ≥ (1 + r)W. (3) introduction. The model just introduced offers a
Thus, an increase in the interest rate r will ceteris straightforward option to motivate the influence
paribus result in a higher risk of the marginal of this variable without having to rely on infor-
project, i.e. the project at which an entrepreneur mation accumulation due to ongoing business rela-
is indifferent between undertaking it or choosing tions.
the secure asset. Let p* denote the success Therefore, it is assumed that in each period a
probability of the marginal project, i.e. E(pi) ≥ continuum of new firms with the features
(1 + r)W ⇐⇒ pi ≤ p*. Then, it follows that described above enters the market. Old firms
dp*/dr < 0. continue operation unless they go bankrupt. While
Now, the central assumption on asymmetric the density of success probabilities for new firms
information comes in. It is assumed that for a is given by f (p), it evolves over time for a cohort
group of firms with identical observable charac- of enterprises starting up business in the same
teristics banks know solely the distribution period, as firms with a low repayment probability
function F(p) of the pi in this group, while the have a high risk of leaving the market due to
firms themselves know their success probability bankruptcy. For a cohort of firms being in the
pi. It is impossible for the banks to distinguish market for n periods the density becomes
between the firms in such a group with regard to f (p)pn
fn(p) = . (5)
their risk. ∫ f (p)pndp
1
0
However, since only firms with a success
probability of less than p* will ask for a loan, the Figure 1 illustrates the development of the
bank’s expected return is given by repayment distribution for a cohort of firms. For
the example, f (p) was chosen as a curtailed normal
# pf (p)dp
p*
1
E(pb) = (1 + r)B density on [0.5, 1] with mean 1 and standard
F(p*) 0
deviation 0.15. As can be seen, the fraction of low
# (1 – p)f (p)dp
p*
1 risk firms increases, whereas the fraction of high
+ Rf
F(p*) 0 risk firms (p close to 0.5) decreases rapidly. This

F(p*) #
p*
f 1 obvious result is supported by the empirical
= [(1 + r)B – R ] pf (p)dp
0 findings on firm failures in Evans (1987), Dunne,
+ R,f
(4) Roberts and Samuelson (1989), Audretsch (1991,
1997) and Wagner (1994).
where F(p*) is the fraction of firms demanding a Using the model introduced in section 2, the
loan at the given interest rate r and f (p) denotes expected return of a credit to a newly founded firm
the density function corresponding to F(p). is given by equation (4), and to a firm operating
Using dp*/dr < 0 and differentiating (4) with for n years as
regard to r it can be shown that for any given
#
p*
density function f there always exists an interest 1
E(pb) = (1 + r)B pn + 1f(p)dp
rate for which the positive effect of higher repay- Fn(p*)kn 0
ment commitments due to the increase in the
#
p*
1
interest rate is overcompensated by the decrease = Rf (1 – p)pnf(p)dp, (6)
Fn(p*)kn 0
in the repayment probability. Profit maximizing
banks will choose r with dE(pb)/dr = 0, which is where p* is implicitly defined as above, Fn is the
not necessarily the market clearing price for a distribution function corresponding to fn and kn [
given loan supply. ∫01 f (p)pndp for n in N.
172 P. Winker

Figure 1. Densities of repayment probabilities.

The simple proof of the following proposition the overall influence on the marginal success
is given in Appendix A: probability for a given group of firms is positive.
The details are given in Appendix B. Hence, the
Proposition 2.1. If p* > 0, i.e. if there is a
bank optimal loan rate is larger in this case; loan
demand for loans at the given interest rate, then
supply will rise and the overall rationing proba-
for all n > 0 En(pb) > E(pb).
bility decrease. The additional information at the
Consequently, banks facing an excess demand firm level has an incentive compatible influence.
for loans at the given interest rate will, ceteris This effect, however, should not be confounded
paribus, i.e. after controlling for other observable with the effect of improving business prospects for
characteristics, prefer “old” firms. the whole group, which corresponds to a higher
mean expected return. This case is more likely to
be observable by banks, and then, reduces the
2.4. Business prospects and loan demand rationing probability for the whole group.
In order to analyze the impact of business Still, rationing might occur for some group of
prospects, another stochastic element is intro- firms. Due to the differences in the marginal
duced. Now, the return to a successful project of success probabilities, for any given interest rate,
firm i is given by (1 + di)Ris with di ∈ {–n, 0, n}, the share of firms with di = n asking for a loan
n > 0 and E(di) = 0. While the mean expected is larger than in the whole group of firms.
return over all firms within the group remains Consequently, since banks have no information on
unchanged, the expected project return is no the di, the probability of measuring financing
longer assumed to be identical over firms.7 constraints for an individual firm will rise if
However, only firms possess information about di business prospects are positive (di = n).
prior to investment. The information set of banks
remains unchanged introducing an additional 3. The ifo firm panel data
source of asymmetry in information.
An analysis of the expected return of firm i In order to undertake an empirical investigation of
along the lines depicted in equations (1) to (3) the impact of asymmetric information on credit
shows that the marginal success probability p* markets several approaches are possible. In prin-
rises for di = n and becomes smaller for di = –n. ciple, a hypothesis derived from the micromodel
This effect is not linear and it can be shown that can be tested in an aggregate model of the loan
 Causes and Effects of Financing Constraints at the Firm Level 173

market (Winker (1996)) concentrating on rigidi- only in the context of innovative activities, the
ties of the price mechanism on this market.8 observations, for those firms and time periods
However, to test directly for the impact of firm when no innovation activities were required for
specific characteristics on their rationing proba- the current period, had to be excluded from the
bility, as well as the impact of financing deficits sample. Consequently, the variables measuring
on their investment decisions and innovation financing constraints comprise the decision to
expenditures, microdata have distinct advantages undertake projects requiring some funding, and the
if they cover the relevant variables. information whether this funding was available or
The data source used here for the empirical not. A possibly resulting selection bias was tested
investigation of determinants and effects of in a bivariate probit framework, modeling the
financing constraints is a unique panel of micro decision to undertake innovation activities on
data from West German manufacturing firms. The available data from the other business surveys. As
data combine observations of the monthly business the qualitative results did not change (cf. Winker,
survey, the annual investment survey and the 1996, p. 101), and as innovative activities and
annual survey on innovation activities of the ifo investment decisions are closely linked, it seems
institute, Munich (cf. Schneeweis and Smolny admissible to use this approximation.
(1996) for a detailed description of the dataset). Table I gives the percentage of firms charac-
Annual data for the time period 1982–1991 are terising themselves as financing constrained, either
available for 1919 firms and plants in manufac- by missing external or internal funds. The last
turing, which participated at least once in all three column gives the percentage of firms facing either
surveys, and still existed in 1984. constraint.
Missing observations reduce the number of The differentiation between missing external or
available observations to 6855 out of the internal funds seems questionable for small firms
maximum possible number of 19190. There are in particular; quite often, firms are denied a loan
three main sources for missing observations: firms because of missing equity. Whether a firm will
leaving the panel, new firms entering the panel, describe this state as rationing, by missing internal
and unwillingness of the firms to answer to all funds, rather than by missing external funds, is not
questions (or even to fill in the questionnaire) for clear. Hence, the empirical analysis is based on
some waves. The possibly most serious problem both rationing variables as well as on the variable
of attrition was tested for by including some attri- indicating either missing external or internal funds
tion dummies in the estimated equations without (MF ). Finally, the marked change in the level of
finding major effects. However, the sparse struc- all three financing constraints indicators in 1991
ture of the data set limits the usefulness of some has to be attributed to a change in the question-
panel estimators, such as fixed effects. naire, now allowing firms to differentiate between
The most prominent feature of the ifo firm modest and strong impediments. As both cate-
panel data for the purpose at hand is that it
contains direct information on financing deficits. TABLE I
In the survey on innovation activities firms are Impediments to innovations 1982–1991
asked whether they faced impediments to innova- Year Missing Missing Missing
tion activities. Besides other principal impedi- external funds internal funds funds
ments, such as too low an expected return, not
enough qualified employees or unwillingness of 1982 07.89% 36.09% 36.47%
employees and management, firms can select 1983 05.56% 29.12% 29.12%
1984 05.75% 29.61% 29.78%
“missing external funds” (MEF ) and “missing 1985 05.58% 28.89% 29.26%
internal funds” (MIF ). Thus, firms themselves 1986 04.78% 25.67% 26.57%
indicate whether they face financing constraints. 1987 04.64% 22.41% 22.87%
Therefore, avoiding the reliance on proxies, such 1988 03.42% 20.09% 20.54%
as cash flow, which might be less suitable (cf. 1989 03.56% 17.15% 17.31%
1990 03.06% 17.35% 17.69%
Faroque and Ton-That (1995)). 1991 13.00% 29.20% 30.00%
As the question on financial constraints appears
174 P. Winker

gories are summarized, the change in level results, more likely to be observable by banks and, there-
in a large extent, from a different threshold when fore, lead to increased loan supply and less
financing constraints are realized. In the estima- rationing.
tion, this level effect is catched up by the year The impact of financing constraints is analyzed
dummy for 1991. Furthermore, the results are through changes in innovation and investment
robust against leaving out the observations for expenditures. In order to control for the firm size
1991. such expenditure data have to be normalized. The
In order to test the hypotheses derived in the preferable normalization by value added or capital
previous section, the explanatory variables of stock is not possible as these variables are not
central interest are the age and business prospects contained in the data set. Therefore, a normaliza-
of the firm. Unfortunately, the age of the firms is tion by sales had to be used. The share of expen-
not contained in the data set. Instead, firm size is ditures on innovation activities in sales is denoted
measured by the number of employees (FE ). by INQ, while IQ denotes the share of investment
Given the industrial sector, which is controlled for expenditures.
by including sectoral dummies in the estimated In addition to the dummies for financing
100
equations, FE is a reasonably good proxy for the deficits, the degree of capacity utilization (Q–1 ) is
age of the firm, in the sense that firms starting up used to control for cyclical effects at the firm
business are comparably small (cf. Evans, 1987; level. In particular, cash flow is assumed to depend
Dunne, Roberts and Samuelson, 1989; Audretsch, on capacity utilization in the previous period.
1991, 1997; and Wagner, 1994).9 However, it does Finally, all specifications include time dummies
not solely measure the direct impact of age to control for macroeconomic conditions and – if
described in the previous section, but may also not stated otherwise – sectoral dummies to control
comprise information on equity base, direct access for sector specific effects. Table II provides means
to capital market or its legal form (cf. Gertler and standard deviations of the variables used for
and Gilchrist, 1994, pp. 313f), influencing the estimation.
rationing probability in the same direction as the
simple age effect. Furthermore, effects of diversi-
4. Empirical specification
fication may be covered. For some specifications,
a diversification dummy (DIV ) is added, which Based on the arguments given in section 2, the
equals one for multi-product firms, and zero following model of the indicator variables for
otherwise. financial constraints (MF ) is obtained
While the age of the firm has to be proxied, a
MF = f (SI, BP, CB, X ), (7)
distinct advantage of the ifo firm panel is the fact
that it includes estimates of the current business where SI is a measure of firm size reflected by
situation as well as expectations about the future log FE or size dummies SK1, . . . , SK4 for firms
development. While banks may know whether the with less than 50, 50–199, 200–999 and more than
current business situation of the firm is good (B+), 999 employees, respectively; BP captures business
medium (B0) or bad (B–), it is probably less well prospects; CB reflects the current business situa-
informed about the expectations at the firm level.10 tion; and X includes other variables to control for
In particular, firms with negative business observable characteristics and macroeconomic
prospects (B&) have no incentives to report it to effects. The firms’ industrial sector and a diversi-
their banks.11 Consequently, positive business fication dummy are used in some specifications,
prospects (B%) must also be considered as private while year dummies are always included to mirror
information to the firm. Both variables are used to macroeconomic influences, such as interest rates,
model the impact of asymmetric information along general wage and price trends or general business
the lines depicted in subsection 2.4. Ceteris conditions.
paribus, the risk of facing financing constraints As the information about financing constraints
should be positively correlated with positive in the data set is qualitative, limited dependent
business expectations (B%) at the firm level, while variable models such as Probit estimation are used
positive expectations at the sectoral level (B%) are for the estimation of equation (7).
 Causes and Effects of Financing Constraints at the Firm Level 175

TABLE II
Descriptive statistics for the used variables

Variable Scale x σx

Miss. external funds MEF dummy 00.047

Miss. internal funds MIF dummy 00.214
Missing funds MF dummy 00.218
Employees in firm log FE log. 02.494 00.695
< 50 employees SK1 = 1 dummy 00.121
50–199 employees SK2 = 1 dummy 00.275
200–999 employees SK3 = 1 dummy 00.384
Diversification DIV dummy 00.684
Good business sit. B+ dummy 00.203
Bad business sit. B– dummy 00.165
Improving bus. sit. B% dummy 00.094
Worsening bus. sit. B& dummy 00.114
Investment ratio IQ share 00.051 00.066
Innovation ratio INQ share 00.030 00.052
100
Capacity utilization Q–1 percent 83.162 12.813

The impact of financing constraints on invest- contrast to most of the cited studies using OLS
ment behaviour and innovative activities is tested estimates.
following the approach by Fazzari, Hubbard and The investment ratio (IQ) is analyzed based on
Petersen (1988), Hoshi, Kashyap and Scharfstein a dynamic model of the firm allowing for effects
(1991), Blundell, Bond, Devereux and of rationing on goods and labour markets (cf.
Schiantarelli (1992), Elston and Albach (1995) Smolny, 1993). The resulting equation to be
among others. In these studies, standard invest- estimated is given by
ment functions or models for R&D expenditures
IQ = b0 + b1MF + (b2 + b3MF )Q–1
100
are augmented with variables proxying financing
+ b4X + e,
iv
(9)
constraints, such as firm size, cash flow or access
to the capital market. Although the estimation of where Xiv, again, captures other control variables.
(7) will show a significant influence of such Apart from a level effect (b1), as for the expendi-
observable individual characteristics, a large part tures for innovative activities, the model allows
of the variance remains unexplained. for the effect of financing constraints to depend
Therefore, it seems preferable to use a direct on the degree of capacity utilization as expressed
approach which is feasible due to the unique infor- by b3. For example, firms facing a low degree of
mation on financing constraints in the ifo firm capacity utilization may react more to financing
panel. Then, the risk of biased estimates of the constraints, both for a cash flow effect and for
impact of financing deficits on real activity due lower expected future demand.13 Again, the esti-
to bad proxies (cf. Faroque and Ton-That, 1995, mation is performed as Tobit model due to the left
p. 312) can be avoided. censoring of IQ.
The estimation equation for the ratio of inno- The existence of adjustment costs may justify
vation expenditures to sales (INQ) is specified as12 including lagged observations of the endogenous
variables in the estimation equations. However, for
INQ = a0 + a1MF + a2Xie + e, (8)
innovation expenditures in particular, the sparse
where Xie captures other explanatory variables structure of the panel considerably reduces the
such as firm size, business prospects, sectoral number of available observations in this case.
dummies, mean of innovation ratio of the sector Nevertheless, some results of this dynamic spec-
in a given year etc. As the variable INQ is left- ification are also reported.
censored by zero, a Tobit model is estimated in
176 P. Winker

5. Results of estimations useful in this context, due to the number of con-

secutive observations per firm being small as well
5.1. Causes of financing constraints as the variance of the financing constraints
dummies for a single firm over time.
Some results of the estimation of equation (7) are
The dependent variable MF is a dummy with
shown in Table III. Except for the random effects
value 1 if the firm faces a situation of financing
specification (3), individual heterogeneity is con-
constraints for the current period. Therefore, a
trolled for only by the inclusion of sectoral and
positive coefficient for an explanatory variable
time specific means calculated for the exogeneous
indicates ceteris paribus an increased risk of facing
variables without the respective firm. The inclu-
financing constraints. The t-statistics for the esti-
sion of fixed effects does not seem to be very
mated coefficients are included in parentheses.
All three specifications in Table IV show a
TABLE III highly significant negative effect of firm size as a
Financing constraints and firm size proxy for the age of firms. The larger a firm
Dependent variable: MP
became, the lower is ceteris paribus the risk of
facing financial limitations. This confirms the
Pooled probit Random theoretical result and the approach used in the
effects literature to proxy financing constraints by firm
(1) (2) (3) size.
Constant 0.418 –1.485 –2.122
Current business situation shows the expected
(1.34) (–5.49) (–3.26) effect as good current businesses (B+) reduce the
rationing risk, whereas bad current businesses (B–)
log FE –0.551 SK1 0.670 0.704
(–15.71) (10.43) (5.20)
tend to increase it. The reference category is
defined by the firms considering their business
SK2 0.379 0.313
(7.55) (3.16) situation as average. Current business situation can
SK3 0.008 –0.05
be observed by banks and is used as a screening
(0.17) (–0.62) variable. Firms with different current business
situations will belong to different segments of the
B+ –0.286 –0.297 –0.368
(–5.22) (–5.97) (–4.62)
loan market. Consequently, the sectoral and time
specific mean of current business situation should
B– 0.185 0.252 0.270
(3.45) (5.22) (3.55) not have a relevant influence. This is confirmed by
B% 0.256 0.194 0.198
(not reported) specifications including these means.
(3.81) (3.20) (2.08) An inverse effect is found for the future
B% –0.557 –0.376 –0.606 business prospects representing asymmetric infor-
(–1.08) (–0.80) (–0.90) mation effects. If a firm expects an improvement
B& 0.001 0.045 0.123 of its business situation (B%) it has to face a
(0.13) (0.79) (1.40) higher risk of rationing compared to the reference
B& –0.730 –0.480 –0.645 category of firms expecting no change in their
(–1.82) (–1.37) (–1.29) business prospects. This effect is rather strong
r – – 1.452 compared to the contrary effect of improved
(29.556) business expectations for the whole sector. Only
if about half of the sector expects improving
Lik. ratio x2 651.999 566.213 2729.701
d.f. 041 043 0046 businesses, the net effect for a single firm with
B% = 1 will be zero. On the other hand, if a
FE = firm employment. worsening is expected for the whole sector, the
SK = employment size classes. financing limitations for a single firm become less
B = current and expected business situation. binding ceteris paribus, whereas a worsening of
XX = sectoral and time specific mean of variable XX without
firm. own business prospects has no significant effect.
All specifications include dummies for industrial sector and This finding confirms the theoretical analysis of
time. section 2.4 and supports the importance of asym-
 Causes and Effects of Financing Constraints at the Firm Level 177

TABLE IV
Effects of financing constraints on innovation expenditures

Dependent variable: INQ

Pooled tobit

(1) (2) (3) (4)

Constant 0.026 0.020 0.026 0.014

(6.73) (1.41) (1.83) (0.92)
MF –0.016 –0.037 –0.011 –0.007
(–6.72) (–5.91) (–4.49) (–2.33)
INQ 0.683 –0.260 –0.255 –0.006
(11.81) (–2.29) (–2.26) (–0.42)
INQ–1 – – – 0.410
(20.49)
Sectoral dummies without with with with
Firm size dummies without without with with
Changing business prospect dummies without without with with
r 0(0.064 0(0.063 0(0.063 0(0.055
(88.19) (88.11) (88.12) (68.68)
Log-likelihood 4842.5 4912.2 4952.3 3321.8

MF = dummy for financing constraints.

XX = sectoral and time specific mean of variable XX without firm.
t-statistics in parenthesis.

metric information about future developments as effects did not differ substantially from the ones
source of financing constraints. presented in Table IV.
In specification (3) the potential problem of
unobserved individual heterogeneity omnipresent
5.2. Effects of financing constraints
in panel data econometrics was tackled by the
estimation of a random effects model (cf. Butler Table V summarizes the results of the estimation
and Moffitt, 1982). The estimated coefficient ρ is of equation (8). Again, all specifications include
large. The Durbin-Wu-Hausman specification test time dummies to capture macroeconomic trends.
(cf. Davidson and MacKinnon, 1993, pp. 389ff ) In contrast to most of the above cited studies, a
indicates significance at the 1% level. The increase pooled tobit estimator is used to take account of
in some of the parameter estimates is contributed the left-censored data. In (1) a parsimonious
to the fact that the a priori standard deviation of specification of equation (8) is used, whereas (2)
the residuals is not normalized to 1 anymore. and (3) in addition include some of the explana-
Instead, the standard deviation is given by tory variables for the financing constraints, such
√r2 + 1 ≈ 1.77. However, the qualitative results as firm size dummies and dummies for changing
are not changed. business prospects, which might also have a direct
In order to assess the robustness of the results, impact on innovation expenditures. Finally, spec-
some specification tests with respect to the ification (4) includes the lagged endogenous
possible heteroscedasticity of the error terms, non variable, reducing the number of available obser-
random attrition and non random selection were vations from 4623 to 2743 due to the sparse
performed (cf. Winker, 1996), p. 99f.). Although structure of the data set.
the hypothesis of heteroscedasticity with respect For all specifications, a highly significant
to some of the sector dummies and a partially non negative influence of financing constraints on
random selection could not be rejected, the esti- innovation expenditures is found. This effect
mation results of models correcting for these remains significant even after including some
178 P. Winker

TABLE V
Effects of financing constraints on investment

Dependent variable: IQ
Pooled tobit

(1) (2) (3) (4)

Constant –0.010 –0.001 0.016 0.009

(–1.10) (–0.03) (0.96) (0.53)
MF –0.054 –0.055 –0.055 –0.044
(–3.75) (–3.82) (–3.81) (–3.08)
100
Q–1 /100 0.035 0.041 0.024 0.018
(3.94) (4.64) (2.55) (1.89)
100
Q–1 /100 × MF 0.061 0.062 0.065 0.053
(3.50) (3.56) (3.73) (3.04)
IQ 0.782 0.001 –0.017 –0.066
(14.89) (0.01) (–0.166) (–0.65)
IQ–1 – – – 0.321
(19.49)
Sectoral dummies without with with with
Firm size dummies without without with with
Changing business prospect dummies without without with with
r 00(0.071 0(00.070 0(00.070 0(00.068
(105.21) (105.20) (105.20) (102.68)
Log-likelihood 6723.1 6821.1 6844.5 6651.4

MF = dummy for financing constraints.

XX = sectoral and time specific mean without firm.
t-statistics in parenthesis.

explanatory variables for the financing constraints tion of equation (9) for the investment to sales
dummy. Therefore, it is not surprising that the ratio, as no more preferable scale variables, such
effects on innovation expenditures found using the as capital stock are available in the data set. The
available direct information on financing con- cyclical component of sales is partially captured
straints is larger than those reported in the litera- by the capacity utilization variables. Again, all
ture using indirect approaches. In fact, the mean specifications include time dummies to capture
ratio of innovation expenditures to sales amounts macroeconomic trends including wages and prices.
to 3% in the whole sample and is reduced to less In (1) a parsimonious specification of equation (9)
than 2% for the constrained firms.15 is used, including variables for the degree of
The sectoral and time specific mean of the capacity utilization and the sectoral and time
dependent variable calculated without the value of specific mean of the endogenous variable. The
100
the respective firm (INQ) loses importance as variable Q–1 is used with one lag in order to
soon as sectoral dummies and other individual reduce a possible simultaneity bias. Specifications
characteristics of the firm are added. Hence, it (2) and (3), in addition, include some of the
might be concluded that INQ mainly reflects explanatory variables for the financing constraints,
heterogeneity at the sectoral level. while (4) again, presents a simple dynamic spec-
The results of the dynamic specification (4) are ification.15
not directly comparable to the other results due Again, a highly significant negative impact of
to the smaller sample. However, the order of financing constraints is found on investment
magnitude of a permanent financing constraints expenditures. Hence, it may be concluded that
effect is similar. the possibly subjective assessment of financial
Table V summarizes the results of the estima- constraints at the firm level is, in fact, closely
 Causes and Effects of Financing Constraints at the Firm Level 179

connected to their real decisions; in other words, The framework of the Stiglitz and Weiss model
financial constraints do matter! was used to demonstrate the impact of age, size
In contrast to the estimation results for the and business relations of firms on the probability
innovation expenditures, a significant effect of the of being rationed on the credit market.
degree of capacity utilization is found for the Furthermore, the microdata of the ifo institute
investment to sales ratio. The influence becomes allowed for constructing a variable closely corre-
smaller if firm size dummies and dummies for lated with the informational asymmetry between
changing business prospects are included ((3) and the firm and its possible creditors.
(4)). The effect of capacity utilization differs The empirical results confirm the impact of
between rationed and non rationed firms. Both these variables. Ceteris paribus, older firms face
the financing constraint dummy itself, and the a lower risk of being rationed on the credit market,
interaction term with the utilization variables are whereas asymmetric information due to improving
significant in all specifications. business expectations will increase this risk.
While the dummy MF has a negative impact on The answers of the firms on the question about
100
investment, the interaction with Q–1 shows a impediments to innovation allow the conclusion
positive sign, indicating a higher sensitivity of that credit rationing is a relevant phenomenon for
investment with regard to the degree of capacity the German credit market. Furthermore, the impact
utilization for rationed firms. This might be of financial limitations on the activities of firms
explained by the fact that the rationing dummy leads to the deduction that rationing on credit
mirrors the existence, but not the extent of markets might be important for a deeper under-
financing constraints. Thus, firms with a ceteris standing of the monetary transmission mechanism.
paribus higher degree of capacity utilization in the It will be the subject of further research to
last period might face a smaller financing deficit assess the quantitative importance of credit
due to an increased cash flow (this effect is used rationing on an aggregate level (cf. Winker, 1996).
by Fazzari, Hubbard and Petersen (1988) to
estimate the impact of financing constraints on
Acknowledgements
investment).
The overall effect of financing constraints is Research was supported by the Deutsche
not uniform. While firms with a low degree of Forschungsgemeinschaft, Long-Term Research
capacity utilization reduce their investment expen- Project 178. I am indebted to the ifo institute
ditures heavily when facing financing constraints, for economic research, Munich, for supplying
a slight increase can be found for firms with near anonymous micro data, to B. Fitzenberger for his
to full utilization of capacities. code for the random effects probit estimation, and
Again, the significant influence of the financing to T. Büttner, B. Fitzenberger, E. Schröder, and
constraints dummy and the interaction terms with W. Smolny as well as participants of seminars and
the degree of capacity utilization persist, if, instead conferences in Konstanz, Maastricht, Mannheim,
of the general financing constraints dummy, a Osnabrück, Rostock and Toulouse for valuable
dummy for missing internal or missing external comments on earlier drafts of this paper. The
funds is used. The same holds true for the dynamic constructive comments of two anonymous referees
specification. of this journal helped to improve the paper con-
siderably.
6. Conclusion
Appendix
A microeconometric method was used in this
paper to assess the relevance and impact of credit A. Proof of Proposition 1
rationing at the firm level. The main advantage Let f̃ and f̃n be given by
of this approach is that rationing of individual 1 1
f̃ (p) = f (p) and f̃n(p) = pnf (p)
firms can be explained by individual characteris- F(p*) Fn(p*)kn
tics of the firms including asymmetric informa- on [0, p*], where kn [ ∫01 f(p)pndp for n ∈ N. f̃ and f̃n are
tion. densities, i.e.
180 P. Winker

# #
p* p*
Now, equation (13) becomes
f̃(p)dp = f̃n(p)dp = 1. (10)
0 0
∆p*[(1 + r)B – Rf ] = δ{R – [1 – p* + ∆p*)]Rf }
Furthermore, f̃ (0) = f̃n(0) = 0 by definition. Setting Cn = = δ[R – (1 – p*)Rf ] + ∆p*δRf
F(p*)/(Fn(p*)kn) and e = min{n√(1/Cn), p*} one obtains ⇔ ∆p*[(1 + r)B – (1 + δ)Rf ] = δ[R – (1 – p*)Rf ].







f̃n(p) < f̃ (p) for all p ∈ [0, e[. (11)
>0 >0
From equation (10) it follows the existence of p0 ∈ ]0, p*[
with f̃n(p) = f̃ (p); hence Cn p0n f (p0) = f (p0) and finally Consequently, sgn(Dp*) = sgn(d), for d = n > 0 the marginal
success probability will increase, whereas for firms with
f̃n(p) > f̃ (p) for all p ∈ ]p0, p*[. d = –n < 0 it will decrease. Furthermore, comparing Dp* for
In particular, p0 is unique, and again with equation (10) it d = n and d = –n one finds
follows ν[R – (1 – p*)Rf]
∆pν* =
(1 + r)B – (1 + ν)Rf
# #
p0 p*
– ( f̃n(p) – f̃n(p))dp = ( f̃n(p) – f̃(p))dp.
ν[R – (1 – p*)Rf]
= |∆p*–ν|.
0 p0
>
Due to the monotony of the integral operator one obtains (1 + r)B – (1 + ν)Rf
This result indicates that the quality of the risk pool of loan
# #
p0 p*
– p( f̃n(p) – f̃n(p))dp < p( f̃n(p) – f̃(p))dp, demanding firms will improve if the additional stochastic
0 p0
element is introduced. Nevertheless, rationing might still occur
and with (1 + r)B > Rf the proposition En(pb) > E(pn) results and is concentrated on good business prospect firms as
for n ≥ 1 and p* > 0. compared to their share in the whole sample.

B. Derivation of results for section 2.4

Notes
In section 2.4 an additional stochastic element was introduced
1
capturing firm business expectations. Given di ∈ {–n , 0, n} Jaffee and Stiglitz (1990) offer a survey of the theoretical
the expected net return of firm i becomes according to literature on credit rationing based on asymmetric information.
2
equation (1) For example, a survey of the Deutsche Ausgleichsbank
(1995) indicates that of the East German firms newly founded
E(pi) = pi[(1 + di)Rsi – (1 + r)B] in 1990, almost one half (49.2%) reported financing deficits
= pi[Rsi – (1 + r)B] + di pi Rsi . in 1994.
3
Using R = pi Rsi + (1 + pi)Rf equation (1) now reads Cf. Gertler (1988, 1992) and Jaffee and Stiglitz (1990).
Vitols (1998) and Deeg (1998) describe the specific features
E(pi) = R – Rf – pi[(1 + r)B – Rf ) + di pi Rsi . (12) of the German banking system.
4
Obviously, E(pi) depends positively on di. For di = n the Cf. e.g. Oliner and Rudebusch (1992) and the literature
expected return rises. Consequently, the cut-off probability cited in Kaplan and Zingales (1997) and Audretsch and Elston
implicitly defined by (1997).
5
The ZEW innovation panel described in Egeln, Licht and
E(pi) ≥ (1 + r)W Steil (1997) includes a similar variable, but so far covers only
will also rise. Ceteris paribus, the marginal success probability a short time span.
6
will be larger for firms with di = n. Just the opposite holds Diamond (1984) and Williamson (1986) show that standard
for di = –n. If the distribution of di is independent of the loan contracts are the optimal way to raise funds if monitoring
distribution of pi, the share of firms with di = n demanding a is costly.
7
loan will be larger than in the whole group of firms with As for the standard Stiglitz and Weiss model it is assumed
otherwise identical characteristica. that the return to a successful project is always large enough
The marginal success probability is given if (12) holds with to pay back debt plus interest while the return to a failed
equality. Let p* denote the marginal success probability for project does not cover the liability towards the bank.
8
firms with di = 0. Then, the change Dp*due to di ≠ 0 is Examples of applications to the German credit market are
calculated according to Kugler (1987) and Winker (1998).
9
Using a dataset containing information both on firm size
R – Rf – (p* + Dpi) [(1 + r)B – Rf ] + d(p* + Dp*)Rsi and age, Egeln, Licht and Steil (1997) find a highly negative
= (1 + r)W impact of both variables on the risk of facing financing con-
⇔ Dp*[(1 + r)B – Rf ] = d(p* + Dp*)Rsδ (13) straints.
10
where Ris denotes the normal (i.e. for d = 0) return to a The questionnaire asks for a judgment on current pro-
successful project with success probability equal to p* + Dp*. duction and for business prospects during the next half-year
Using again R = (p* + Dp*)Rsδ + ([1 – (p* + Dp*)Rf, one above purely seasonal effects.
11
obtains Fortunately, the anonymous data of the ifo panel are not
flawed by these incentives, as the expectations expressed in
R – [1 – (p* + ∆p*)]Rf the questionnaire show a high positive correlation over time
Rsδ = .
p* + ∆p* with their ex post realizations.
 Causes and Effects of Financing Constraints at the Firm Level 181
12
For a related specification and some theoretical back- Fazzari, S. M., R. G. Hubbard and B. C. Petersen, 1995,
ground see, e.g., Himmelberg and Petersen (1994), pp. 44ff. ‘Financing constraints and corporate investment: Response
13
As the planing horizon for innovation expenditures is to Kaplan and Zingales’, NBER working paper No. 5462.
longer than for investment, the capacity utilization variables Gertler, M. L., 1988, ‘Financial Structure and Aggregate
are omitted from specification (9). In some preliminary tests Economic Activity: An Overview’, Journal of Money,
100
Q–1 exhibited now significant impact on IQ. Credit, and Banking 20, 559–588.
14
The results do not differ essentially if the general Gertler, M. L., 1992, ‘Financial Capacity and Output
financing constraints dummy is replaced by dummies for con- Fluctuations in an Economy with Multi–period Financial
straints by internal and external finance, respectively. Relationships’, Review of Economic Studies 59, 455–472.
15
The sample size is reduced moderately from 5651 to 5379, Gertler, M. L. and S. G. Gilchrist, 1994, ‘Monetary Policy,
as the observations in the investment survey are more dense Business Cycles, and the Behavior of Small Manufacturing
in time. Firms’, Quarterly Journal of Economics 109, 309–340.
Himmelberg, C. P. and B. C. Petersen, 1994, ‘R&D and
References Internal Finance: A Panel Study of Small Firms in High-
tech Industries’, The Review of Economics and Statistics
Akerlof, G. A., 1973, ‘The Market for “Lemons”: Quality 76, 38–51.
Uncertainty and the Market Mechanism’, Quarterly Hoshi, T., A. Kashyap and D. Scharfstein, 1991, ‘Corporate
Journal of Economics 84, 488–500. Structure, Liquidity, and Investment: Evidence from
Audretsch, D. B., 1991, ‘New-firm Survival and the Japanese Industrial Groups’, Quarterly Journal of
Technological Regime’, The Review of Economics and Economics 106, 33–60.
Statistics 73, 441–450. Jaffee, D. M. and J. Stiglitz, 1990, ‘Credit Rationing’, in B.
Audretsch, D. B. and A. Roy Thurih, 1997, New-firm survival: M. Friedmann and F. H. Hahn (eds.), Handbook of Monet-
Industry versus Firm Effects. Discussion Paper No. 97, 63, ary Economics, Vol. 1, Amsterdam: Elsevier, pp. 837–888.
Tinbergen Institute, Amsterdam. Kaplan, S. N. and L. Zingales, 1997, ‘Do Investment-cash
Audretsch, D. B. and J. A. Elston, 1997, ‘Financing the Flow Sensitivities Provide Useful Measures of Financing
German Mittelstand’, Small Business Economics 9, 97–110. Constraints?’, Quarterly Journal of Economics 112,
Blanchard, O. J. and S. Fischer, 1989, Lectures on 169–215.
Macroeconomics, Cambridge, MA: MIT Press. Kugler, P., 1987, ‘Credit Rationing and the Adjustment of
Blundell, R., S. Bond, M. Devereux and F. Schiantarelli, 1992, the Loan Rate: An Empirical Investigation’, Journal of
‘Investment and Tobin's Q’, Journal of Econometrics 51, Macroeconomics 9, 505–525.
233–257. Oliner, S. D. and G. D. Rudebusch, 1992, ‘Sources of the
Butler, J. S. and R. Moffitt, 1982, ‘A Computationally Financing Hierarchy for Business Investment’, The Review
Efficient Quadrature Procedure for the One-factor of Economics and Statistics 74, 643–654.
Multinomial Probit Model’, Econometrica 50, 761–764. Oppenländer, K. H. and G. Poser, 1989, Handbuch der Ifo-
Deeg, R., 1998, ‘What Makes German Banks Different?’, Umfragen, Berlin: Duncker & Humblot.
Small Business Economics 10, 93–101. Schneeweis, T. and W. Smolny, 1996, Das ifo
Deutsche Ausgleichsbank, 1995, Ost-Betriebe: Zwischen Unternehmenspanel 1980–92. Discussion Paper No. 32,
Wachstumsdynamik und Liquiditätsengpässen (Wissen- CILE, Konstanz.
schaftliche Reihe, Deutsche Ausgleichsbank, Bonn). Smolny, W., 1993, Dynamic Factor Demand in a Rationing
Diamond, D. W., 1984, ‘Financial Intermediation and Context. Theory and Estimation of a Macroeconomic
Delegated Monitoring’, Review of Economic Studies 51, Model for the Federal Republic of Germany, Heidelberg:
393–414. Physica.
Dunne, T., M. J. Roberts and L. Samuelson, 1989, ‘The Stiglitz, J. E., 1992, ‘Capital Markets and Economic
Growth and Failure of U.S. Manufacturing Plants’, Fluctuations in Capitalist Economies’, European Economic
Quarterly Journal of Economics 104, 671–698. Review 36, 269–306.
Egeln, J., G. Licht and F. Steil, 1997, ‘Firm Foundations and Stiglitz, J. E., and A. Weiss, 1981, ‘Credit Rationing in
the Role of Financial Constraints’, Small Business Markets with Imperfect Information’, American Economic
Economics 9, 137–150. Review 71, 393–410.
Elston, J. A. and H. Albach, 1995, ‘Bank Affiliations and Firm Vitols, S., 1998, ‘Are German Banks Different?’, Small
Capital Investment in Germany’, ifo Studien 41, 3–16. Business Economics 10, 79–91.
Evans, D. S., 1987, ‘The Relationship Between firm Growth, Wagner, J., 1994, ‘The Post-entry Performance of New Small
Size, and Age: Estimates for 100 Manufacturing Indus- Firms in German Manufacturing Industries’, Journal of
tries’, The Journal of Industrial Economics 35, 567–581. Industrial Economics 42, 141–154.
Faroque, A. and T. Ton-That, 1995, ‘Financing Constraints Williamson, S. D., 1986, ‘Costly Monitoring, Financial
and Firm Heterogeneity in Investment Behaviour: An Intermediation, and Equilibrium Credit Rationing’, Journal
Application of Non-Nested Tests’, Applied Economics 27, of Monetary Economics 18, 159–179.
317–326. Winker, P., 1996, Rationierung auf dem Markt für Unter-
Fazzari, S. M., R. G. Hubbard and B. C. Petersen, 1988, nehmenskredite in der BRD, Tübingen: Mohr & Siebeck.
‘Financing Constraints and Corporate Investment’, Winker, P., 1998, ‘Sluggish Adjustment of Interest Rates and
Brookings Papers on Economic Activity 88, 141–195. Credit Rationing’, Applied Economics, forthcoming.