DEWEY
MIT LIBRARIES
U(B>3i-
3 9080 02238 5212
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
FIRMS WANT TO BORROW MORE?
TESTING CREDIT CONSTRAINTS USING A DIRECTED
DO
LENDING
PROGRAM
Abhijit Banerjee
Esther Duflo
Working Paper 02-25
May 2002
REVISED: May 2008
Room
E52-251
50 Memorial Drive
Cambridge,
MA 021 42
This paper can be downloaded without charge from the
Social Science
Research Network Paper Collection
http://papers.ssrn.com/paper.taf ?abstract
id
:
-
3
16587
at
j
j
Do Firms Want
Borrow More?
to
Testing Credit Constraints Using a Directed Lending Program*
Abhijit V. Banerjeetand Esther Dufio*
Abstract
This paper uses variation in access to a targeted lending program to estimate whether
firms are credit constrained.
firms
may be
The
willing to absorb
basic idea
all
that while both constrained and unconstrained
is
the directed credit that they can get (because
cheaper than other sources of credit), constrained firms will use
while unconstrained firms will primarily use
these observations to firms in India that
it
it
to
it
expand production,
We
as a substitute for other borrowing.
became
may be
eligible for directed credit as
a
apply
result, of a
policy change in 1998, and lost eligibility as a result of the reversal of this reform in 2000.
Using firms that were already getting this kind of credit before 1998, and retained
in
2000 to control
for
time trends, we show that there
being used as a substitute
for
more production-there was a
these firms.
We
conclude that
large acceleration in the rate of
many
of the firms
for his
work
is
Services
for
collecting the data,
their
growth of
sales
and
profits for
must have been severely credit constrained,
to capital
was very high
for these firms.
Keywords:
G2
Banking, Credit constraints, India JEL: 016,
Sankarnaranayan
no evidence that directed credit
other forms of credit. Instead the credit was used to finance
and that the marginal rate of return
*We thank Tata Consulting
is
eligibility
help
in
understanding
Dean Yang and Niki Klonaris
the
Indian
banking industry,
for excellent research assistance,
and Robert Barro, Sugato Battacharya, Gary Becker, Shawn Cole, Ehanan Helpman, Sendhil Mullainathan,
Kevin Murphy, Raghuram Rajan and Christopher Udry
to the administration
for very useful
and the employees of the bank we studied
MIT CEPR and BREAD.
NBER, CEPR and BREAD.
'Department of Economics, MIT,
1
We
are particularly grateful
for their giving us access to
this paper.
'Department of Economics,
comments.
the data we used
in
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium
Member
Libraries
http://www.archive.org/details/dofirmswanttobor00bane2
Introduction
1
That there
are limits to credit access
is
widely accepted today as an important part of an
economist's description of the world. Credit constraints figure prominently in economic analyzes
of short-term fluctuations
and long-term growth.
1
Yet there
tight evidence of the
is still little
existence of credit constraints on larger firms in developing countries. While there
credit constraints in rural settings in developing countries,
the share of agriculture in the capital stock
it
is
is
even smaller.
evidence of
except in the very poorest countries,
not large enough for the credit constraints to have
in India, the share of agriculture in
quantitatively large effects:
in the capital stock
is
2
is
Moreover,
in
output
is
24% and
its
share
Banerjee and Duflo (2005), we argue that
not enough to show that the very smallest firms are credit constrained, since while they
are numerous, the share of capital in these firms
is
much power
too small to have
explaining the cross-country productivity differences.
On
the other hand,
if it
is
in
terms of
the
medium-
sized firms that are constrained, the productivity loss due to the missallocation of capital caused
by
credit constraints
may be
potentially very large: indeed,
to explain the entire productivity
Prima
facie, the idea that
it
such firms
showing that borrowing
though deposit rates are
less
may be
large
enough
gap between India and the US.
may be
severely constrained
is
certainly consistent with
Banerjee and Duflo (2005) survey evidence from
the available evidence.
countries,
we argue that
interest rates are often of the order of
than half as much, and defaults are
rare.
many
60%
less
developed
or above, even
This suggests that the
marginal product of capital in the firms paying these rates might far exceed the opportunity
cost of capital.
However
know whether the
irrational).
1
this
evidence
is
only partly convincing: the problem
firms that pay these rates are
somehow
atypical (smaller,
is
that
we do not
more desperate,
3
See Bernanke and Gertler (1989) and Kiyotaki and Moore (1997) on theories of business cycles based on credit
constraints and Banerjee and
Newman
(1993) and Galor and Zeira (1993) on theories of growth and development
based on limited credit access.
The estimation
of the effects of credit constraints on farmers
is
significantly
more straightforward
since
variations in the weather provide a powerful source of exogeneous short-term variation in cash flow. Rosenzweig
and Wolpin (1993) use
this strategy to
study the
effect of credit constraints
on investment
in bullocks in rural
India.
3
Although
it
is
worth saying that the studies from which these
interest rate
numbers come concern normal
There
also
is
some
direct evidence
on rate of returns on
McKenzie and Woodruff
capital.
(2004) estimate parametric and non-parametric relationships between firm earnings
capital.
less
Their estimates suggest huge returns to capital for these small firms: for firms with
than $200 invested, the rate of return reaches 15% per month, well above the informal
interest rates available in
pawn shops
month). These regressions
may
or through micro-credit
suffer
from of an
programs (on the order of
"ability bias"
3%
per
caused by a correlation between
investment level and rates of return to capital (Olley and Pakes (1996)).
De
and firm
To address
this issue,
Mel, Mckenzie and Woodruff (2006) randomly allocate small ($200 or so) capital grants to
microenterprises in Sri Lanka, and find that the returns to capital are also very high for those
firms: the average returns to capital
These firms
are,
is
as high as
4%
per month.
however, very small and unconnected to any bank. Unfortunately,
to imagine carrying out the
same experiment
for larger firms that are already
we take advantage
connected to the
banking
sector.
effect of
an influx of credit on investment and productivity of medium-sized firms
make
firms.
Thus,
in this
paper,
difficult
it
of a natural experiment to estimate the
in India.
4
We
use of a policy change that affected the flow of directed credit to an identifiable subset of
Such
policies
What makes
and policy changes are common
this case particularly interesting
registered firme (this
means that they
is
in
many
developing and developed countries.
that the firms affected by the policy are officially
none
are not part of the informal economy, although
of
these firms are listed on the stock market), fairly large by Indian standards, though not the
largest corporate entities: the average capital stock of firms in the 95th percentile in the
median
industry in India was Rs. 36 million (the exchange rate was about 45 rupees to a dollar), which
puts them at a size just above the category of firms that were affected by the policy change
(which required a capital stock between Rs. 6.5 and Rs. 30 million).
The advantage
of our approach
that
it
gives us a specific exogenous shock to the supply
of credit to specific firms. Its disadvantage
is
that directed credit need not be priced at
market
if
price.
is
This has two important, implications: First, firms
they are not credit constrained simply because
it
is
will
want directed
its
true
credit even
a cheaper substitute for market credit.
production loans to stable firms, not emergency loans.
4
The approach
spirit to that in
of Japanese
of looking directly at an identifiable shock to credit supply for a specific subgroup
Peek and Rosengren (2000), who look at the impact of a reduction
banks during the Japanese banking
crisis
on
is
similar in
in credit at U.S. subsidiaries
real estate activity in the U.S.
Therefore the fact that a firm
is
borrow more
willing to
at the
same
price
to them, cannot be seen as evidence that they are credit constrained
it is
when more
troublingly, a shock to the supply of directed credit might lead not just to
but also to more investment even
a firm
if
more borrowing
we develop a simple methodology based on
from price theory that allows us to deal with the inference problem suggested
paragraph.
The methodology
based on two observations:
is
Second, and
not credit constrained.
is
In the theoretical section of this paper
offered
rationed with respect
to this particular source of cheap credit, but not necessarily credit constrained.
more
is
first,
if
a firm
is
strained, then an increase in the supply of subsidized directed credit to the firm
ideas
in the previous
not credit con-
must lead
it
to
substitute directed credit for credit from the market. Second, while investment, and therefore
total production,
may
go up even
the firm
if
is
not credit constrained,
it
will
only go up
if
the
firm has already fully substituted market credit with directed credit.
We
test these implications using firm-level
firms in India.
We make
data we collected from a sample of medium
use of a change in the so-called "priority sector" regulation, under
which firms smaller than a certain limit are given priority access to bank lending. 5 The
experiment we exploit
is
a 1998 reform which increased the
eligible to receive priority sector lending
strategy
is
maximum
the rate of change in various firm
outcomes before and
included in the priority sector as a result of the
were already
size
first
below which a firm
is
(from Rs. 6.5 to Rs. 30 Million). Our basic empirical
a difference-in-difference-in-difference approach: that
for firms that
size
new
we focus on the changes
is,
reform for firms that were
after the
limit, using the
in the priority sector as a control.
We
in
corresponding changes
find that
bank lending and
firm revenues went up for the newly targeted firms in the year of the reform, relative to firms
that were already included.
bank
credit for borrowing
to firms that
last
had
We
no evidence that
find
this
was accompanied by substitution of
from the market and no evidence that revenue growth was confined
fully substituted
bank
credit for
market borrowing.
two observations are inconsistent with firms being unconstrained
Our second experiment
As argued
in their
earlier,
the
market borrowing.
uses the fact that a large fraction of these firms (specifically those with
investments higher than Rs. 10 million) that were included in the priority sector in 1998, got
5
Banks are penalized
sector.
for failing to lend
a certain fraction of the portfolio to firms classified
in
the priority
excluded again
We
in 2000.
find that
bank lending and firm revenues went down
for these firms,
both compared to the firms that had always been a part of the priority sector and to firms that
were included
and remained part of the priority sector
in 1998,
in 2000.
have two separate experiments that each allow us to estimate the
effects
Moreover because we
on credit and revenue
growth, we can implement an overidentification test which essentially ask whether the effect of
on revenue
credit
less likely that
medium and
the same in the two cases.
is
results easily pass this test,
making
much
it
the effects are generated by differential time trends in the productivity of small,
what we
large firms: for that to be the explanation for
would have to have reversed
We
The
also use this
in exactly the right
way
find,
the productivity trends
at exactly the right time.
We
data to estimate parameters of the production function.
estimate an
elasticity of sales
with respect to bank credit of 0.75. This ought to be a lower bound on the
elasticity of sales
with respect to working capital (since bank loans are only a part of working
capital,
we would expect working
when bank
up
loans go
capital not to go
up
in the
as a result of the policy shock).
We
substitution of market borrowing for bank borrowing, given
bank loans
must be
working
in total
well
above
capital, the elasticity of sales
same proportion
will
argue that
of their current level of investment,
if
bank loans
there
any
is
what we know about the share
of
with respect to total working capital
This would imply that firms have increasing returns
1.
as
in
the neighborhood
and therefore must be credit constrained (otherwise they
would keep borrowing).
Finally,
we
try to estimate the effect of the program-induced additional investment on profits.
While the interpretation of
large
this result relies
it
suggests a very
gap between the marginal product and the interest rate paid on the marginal dollar (the
point estimate
which
is
is
that Rs.
much too
1
more
in loans increased profits net of interest
for the elasticities
make
it
rewarded
we
for lending
In the second part of the
therefore study the allocation problem faced by a loan officer
more but
are also punished for defaults.
We
show that
who
this creates
incentive for targeting credit (a) towards firms that are the least likely to default which
also be
some
0.73,
particularly important to understand
the extra credit that the program generated was allocated.
theoretical section
payments by Rs.
large to be explained as just the effect of getting a subsidized loan).
These very high estimates
how
on some additional assumptions,
of the
more
is
an
may
profitable firms, but also (b) towards firms that are on the brink of
and therefore need to be bailed
default
on
profitability)
credit
hand, when the loan
want to give
all
growth
may
officer gets
out.
As
a result, an
OLS
regression of revenue (or
be biased downwards or upwards by selection.
On
the other
an unexpected inflow of additional loanable funds, he would
of that to the firms that are least likely to default, since the set of firms that
need to be bailed out remains unchanged and he was already taking care of those firms before
the inflow of extra credit. Assuming that the firms that are the least likely to default are also
among
the most productive firms,
we should
therefore expect to see the instrumental variables
estimate of the effect of credit growth on revenue growth to be
estimate,
The
when
the source of the variation
rest of the
paper
is
a program like the one
organized as follows:
is
much
we
stronger than the
are studying.
OLS
6
the next section describes the institutional
environment and our data sources, provides some descriptive evidence, and informally argues
that firms
may
be expected to be credit constrained
in this
environment. The following section
develops the theory that justifies our empirical strategy, and provides some useful insights for
interpreting
what we
find.
The next
We
section reports the results.
section develops the equations
we
estimate.
The penultimate
conclude with some admittedly speculative discussion of what
our results imply for credit policy in India.
Institutions,
2
Data and Some Descriptive Evidence
The Banking Sector
2.1
in India
Despite the emergence of a number of dynamic private sector banks and entry by a large number
of foreign banks, the biggest
banks
in India are all in the public sector,
i.e.,
they are corporatized
banks with the government as the controlling share-holder. In 2000 the 27 public sector banks
collected over
The
77%
particular
of deposits
and comprised over 90%
bank we study
is
of
all
branches.
a public sector bank, generally considered to be a good
bank. 7
While banks
ily
provide longer-term loans, financing fixed capital
is
primar-
the responsibility of specialized long-term lending institutions such as the Industrial Finance
6
is
in India occasionally
Note that we are
still
consistently estimating a causal effect of the extra credit with this experiment, but this
the effect on the extra credit on good firms, which are the ones affected by the reform.
'It is
consistently rated
among
the top
five
public sector banks by Business Today, a major business magazine.
Corporation of India. Banks typically provide short-term working capital to firms. These loans
are given as a credit line with a pre-specifled limit and an interest rate that
age points above prime.
The borrower draws from
the limit
is
set a
few percent-
when needed, and reimburses on a
quarterly basis. This paper therefore estimates the impact of short term capital loans, not that
of long
amount
term investment
moreover,
credit;
focuses on the working capital limit (which
the
is
of working capital financing available to the firm at any point).
The spread between the
firm's credit rating
interest rate
and other
many borrowers want
and the prime rate
characteristics, but cannot
charge interest only on the part that
is
is
fixed in advance based
be more than 4%. Credit
used and, given that the interest rate
on the
lines in
India
pre-specified,
is
as large a credit line as they can get.
Priority Sector Regulation
2.2
All
it
banks (public and private) are required to lend
sector",
40%
of their net credit to the "priority
which includes agriculture, agricultural processing, transport industry, and small scale
industry (SSI).
to specific
In
at least
If
banks do not
satisfy the priority sector target, they are required to lend
government agencies
money
at very low rates of interest.
January 1998, there was a change
in the definition of the small scale
industry sector.
Before this date only firms with total investment in plant and machinery below Rs. 6.5 million
were included. The reform extended the definition to include firms with investment
and machinery up
new
to Rs.
30 million.
In January 2000, the reform was partially
in plants
undone by a
change: firms with investment in plants and machinery between Rs. 10 million and Rs. 30
million were excluded from the priority sector.
The
priority sector targets
most banks): every
was 42%
seems to have been binding
bank we study
year, the bank's share lent to the priority sector
in 2000-2001).
It is
plausible that the
and Duflo (2000), calculated
is
(as well as for
very close to
40%
bank had to go some distance down the
quality ladder to achieve this target. Moreover, there
lending. Banerjee
for the
is
(it
client
the issue of the administrative cost of
that, for four Indian public banks, the labor
and
administrative costs associated with lending to the priority sector were about 1.5 higher per
rupees than that of lending in the unreserved sector. This
that lending to smaller clients
is
more
costly.
6
is
consistent with the
common view
With
in
we thus expect an
the reform,
increase in lending to the larger firms newly included
the priority sector, possibly at the expense of the smaller firms.
firms with investment
and machinery above 10 million were excluded again from the priority
in plant
these firms no longer counted towards the priority sector target.
the smaller clients to
fulfill its
priority sector obligation.
We
those firms declined relative to the smaller firms.
firms
When
and smaller
firms,
and evaluate whether any
was matched by a corresponding change
in sales
One
sector, loans to
The bank had
to go
back to
therefore expects that loans to
focus on the comparison between larger
relative
change
in loans
between these groups
and revenue.
Data Collection
2.3
The bank we
and
study, like other public sector banks, routinely collects balance sheets
account data from
loss
loan folder.
there
is
is
for
it
and compiles the data
renewal/extension of
its
profit
in the firm's
credit line,
and
also stored in the folder, along with the firm's initial application even
no formal review of the
physically impossible to put
With
firms that borrow from
Every year the firm also must apply
the paper-work for this
when
all
and
file.
The
more documents
folder
in
is
typically stored in the branch until
it.
the help of employees from this bank and a former
data from the loan folders from the
it is
clients of the
bank
bank
officer,
we
first
in the spring of 2000.
We
extracted
collected
general information about the client (product description, investment in plant and machinery,
date of incorporation of units, length or the relationship with the bank, current limits for term
loans, working capital,
and
profit
and
loss
letter of credit).
We
also recorded a
summary
of the balance sheet
information collected by the bank, as well as information about the bank's
decision regarding the
As we discuss
and
in
amount
more
of credit to extend to the firm
and the
interest rate
detail below, part of our empirical strategy called for a
it
charges.
comparison
between accounts that have always been a part of the priority sector and accounts that became
part of the priority sector in 1998, and the sample was selected with this in mind.
selected
all
We
first
the branches that primarily handle business accounts in the six major regions of the
bank's operation (including
information on
all
New
Delhi and Mumbai). In each of these branches,
we
collected
the accounts of the clients of the bank of firms which, as of 1998, had
investment in plant and machinery below 30 million Rupees. This gave us a total of 249 firms,
7
including 93 firms with investment in plants and machinery between 6.5 and 30 million rupees.
We
aimed to
collect
data
for the years
1996-1999, but
when a
older information
is
not always kept in the branch, so that old data gets "lost". Moreover, in some years, data
is
not collected for some firms.
had started
firms that
We
In the winter 2002-2003,
we
is full,
have data on lending from 1996 for 120 accounts (of the 166
their relationship with the
(of 191 possible accounts), 1998
folder
data
for
collected a
banks by 1996), 1997 data
for
175 accounts
and 1999 data
for
213 accounts.
217 accounts
new wave
(of 238)
of data
,
on the same firms
We
the impact of the priority sector contraction on loans, sales and profit.
in order to
study
have 2000 data
for
175 accounts, 2001 data for 163 accounts, and 2002 data for 124 accounts.
There are two reasons why we have
that
some
firms
had not had
data
less
their 2002 review
in 2000,
2001 and 2002 than in 1999.
when we re-surveyed them
it
is
15% among
firms with investment in plant and machinery above 10 million,
firms with investment in plant and machinery between 6.5 and 10 million, and
with investment in plant and machinery below 6.5 million. Thus,
it
Second,
in late 2002.
43 accounts were closed between 2000 and 2002. The proportion of accounts closed
is
First,
balanced:
20% among
20% among
firms
does not appear that there
sample selection bias would emerge from the closing of those accounts. 8
Table
and
1
presents the
summary
statistics for all
and
credit rationing (in the full sample,
change
in
in
data be used in the analysis of credit constraint
the sample for which
we have information on the
lending between the previous period and that period, which
is
the sample of interest
for the analysis).
2.4
Descriptive Evidence on Lending Decisions
In this subsection,
this evidence to
we provide some
argue that this
is
description of lending decisions in the banking sector.
an environment where credit constraints arise quite naturally.
The
Tables 2 and 3 show descriptive statistics regarding the loans in the sample.
of table 2
shows that,
in a
attrition in the 1998-1999 period
to be in our
data
set,
is
because our data
an account had to
still
be
in
for that period
existence in 1999.
implies that our sample only represents the survivor as of 1999. However, given that attrition
in
first
row
majority of cases, the working capital limit that the bank makes
The reason why we do not observe
collected retrospectively in 2000:
We use
response to second reform in 2000, there
is
again
little
is
was
This
not differential
concern that this sample selection biases the results.
available to the firm does not change from year to year:
65%
even in nominal terms for
it is
of the loans.
row
essentially non-binding:
2
This
shows that
the limit was not updated
in 1999,
not because the limit
is
is
in the six years in the sample,
set so
63%
high that
to
80%
of
the accounts reached or exceeded the credit limit at least once in the year: this means that the
borrower had drawn more from the limit in the course of a quarter than was available in the
credit limit.
This lack of growth in the credit limit granted by the bank
the Indian
that the
economy
demand
registered nominal growth rates of over
for
bank
credit should have increased
is
particularly striking given that
12% per
to firms in bur
up
in
sample come from
this
On
average
one bank and
98%
in
would suggest
from year to year over the period, unless
the firms have increasing access to another source of finance. There
using any other formal source of credit.
year. This
is
no evidence that they were
of the working capital loans provided
any
case, the
same kind
of inertia shows
the data on total bank loans to the firm. Indeed, sales have increased from year to year for
most firms (row
did the
2), as
maximum
authorized lending (a function of projected sales). Yet
there was no corresponding change in lending from the bank. In fact the change in the credit
limit that
was actually sanctioned systematically
the firm's needs as determined by the bank
recommended
It is
itself.
what the bank determined to be
fell
short of
On
average, the granted limit was
89%
of the
limit.
possible that
some
of the shortfall
was covered by informal
according to the balance sheet, total current
every year on average.
is
excluding bank credit increased by 3.8%
However, some expenses (such as wages) are typically not covered by
trade credit and, moreover, trade
heart of this paper
liabilities
credit, including trade credit:
credit,
could be rationed as well.
The question
that
whether such substitution operates to the point where a firm
is
is
at the
not credit
constrained.
In table 3,
factors that
we examine
in
more
detail
whether
might have affected a firm's need
we observe seems
to explain
to get an increase in limit
why
if
this
tendency could be explained by other
for credit.
had
risen, or
if
(3)
shows that no variable
a firm's credit limit was changed: firms were not more likely
they had hit the limit in the previous year,
sales (according to the bank itself) or their current sales
to sales
Column
had gone up,
if
if
their projected
their ratio of profits
their current ratio (the ratio of current assets to current liabilities, a
standard indicator,
in India as well as in
the US, of
secure a working capital loan
magnitude of changes, only an increase
increased. Turning to the direction or the
sales or current sales predicts
how
an increase
in
had
is)
in projected
granted limit, and only an increase in projected
be due to reverse causality, however:
sales predict the level of increase. This last result could well
the bank officers appear to be more likely to predict an increase in sales
when he
is
preparing
to give a larger credit extension to the firm.
Columns
and
Changes
clients.
more
5
6 in table 3 repeat the analysis, breaking the
in limits are
more frequent
for
younger
sample into recent and older
clients,
but they do not seem to be
sensitive to past utilization, increases in projected sales, or profits. This suggest that the
lack of information to
new
signal
may
not reflect that
all
the relevant information about the
firm
was already incorporated
3
Theory: the demand and supply of bank credit
Motivated by
in the lending decisions.
this evidence, the goal of this section
demand and supply
of subsidized
bank
credit.
The
is
to develop
sub-section on
some
intuition
demand
sketches the choice
problem faced by a firm that has (limited) access to cheap bank credit but can
We
the market at a higher rate.
affects the
it is
The
among
how
market borrowing of the firm as well as
reaction in the case where
where
are interested in
constrained in
it
its
about the
also
borrow from
increased access to cheap bank credit
revenues and
profits.
We
contrast
its
has unlimited access to market credit at a fixed rate with the case
its
access to market credit.
sub-section on supply then tries to understand the allocation of subsidized bank credit
users
who
all
want
it.
In particular
we analyze the
incentives facing a loan officer
and
the choices he will make. This will provide a framework to interpret our findings.
3.1
The demand
side:
the key to identifying credit constraints
Consider a firm with the following
a fixed cost
machinery).
C
fairly
standard production technology: the firm must pay
before starting production (say the cost of setting up a factory and installing
The
firm then invests in labor and other variable inputs,
capital invested in variable inputs, yield
R =
k rupees of working
f(k) rupees of revenue after a suitable period.
10
f(k) has the usual shape
—
it is
As mentioned above, the
credit
and more expensive
increasing and concave.
interesting case
credit
from other sources.
with respect to a particular lender
wants to borrow at that rate
We
is
if
there
will say that a firm is credit rationed
no interest rate
is
such that the amount the firm
r
and equal to an amount that the lender
strictly positive
is
We
is
willing
Essentially this says that the supply curve of loans from that lender to
to lend at that rate.
the firm
where the firm has access to both low cost bank
is
not horizontal at some fixed interest rate.
will say the firm is credit constrained if there is
that the firm wants to borrow at that rate
no
interest rate r such that the
equal to an amount that
is
amount
the lenders taken
all
together are willing to lend at that rate. This says that the aggregate supply curve of capital to
the firm
is
not horizontal at
Note that
a firm could
some
be
fixed interest rate.
credit rationed with respect to every lender without being credit
constrained in our sense. This can be the case, for example,
lenders, each willing to lend to
We
can get
which we
all
"market" and the "bank"
amount to the
The
is
we analyze
is
<r m
rate: r^
interest rate.
this in itself does not
A
is
possible scenario
is
is direct
We
will
show
bank
firms accepted the
evidence of credit rationing with respect
1.
The
The
horizontal axis in the figure
(or as little) as
it
11
is
in
the figure
step function represents the supply of
we assume that the
the firm wants to borrow at a given rate
much
and
in the next section that
To the extent that
depicted in figure
capital. In the case represented in the figure,
could borrow as
m
imply that they would have borrowed more at the
represents the marginal product of capital, f'(k).
if it
r
statutorily required to lend
measures k while the vertical axis represents output. The downward sloping curve
firm's profit
by
interest
.
purpose of working capital investment.
However
The amount
supply of
involves the firms in question being offered additional
additional credit being offered to them, this
market
infinite
reason to believe that the bank would have to
there was no corresponding change in the interest rate.
to the bank.
an
Denote the market rate of
.
Given that the bank
r^.
priority sector, there
below the market
policy change
credit, for the
is
the relevant intuition from the simple case where there are only two lenders,
will call the
set a rate that
there
no more than $10 at an interest rate of 10%.
the interest rate that the bank charges by
a certain
when
firm has access to
fcbo
units of
assumed to be an amount that would maximize the
wants at that
rate.
capital at the
rate r m
.
As a
bank
rate
result,
but was
ri,
consider what happens
is
the firm
if
equal to r m
entire additional
amount
offered to
though the amount
is
now
remain the case as long as
is
now allowed
it.
Moreover
The
reduced.
kbi
<
borrowing by bank loans. The firm's
its total
much
as
it
wanted
to borrow a greater amount, kbi, at the
bank
is
will
higher than
the firm will borrow the
r;,,
continue to borrow at the market interest
however
is
unchanged
It will
bank
of
borrow as much
Result
1:
If
the firm
the market rate), but
is
should always lead to a
up
as long as
can get from the bank but no more than
it
is
is
equal to
Tb-
We
if
>
kbi
ko
fall in its
market borrowing
maximum
maximum
possible
k\.
There
falls.
Of
However the
is
will
have no
course,
2,
if
If
in:
amount from the banks
amount from the market,
kbi
This has no
the firm
is
its
like at
(fc(,o)
< rm
it
wants
at
Profits will
.
and output
market borrowing.
effect
for a total
effect
r^
If
rj,
on outlay, output or
where the assumption
is
— rm
,
the
profits.
that the firm
an d supplements
investment of
ko.
it
is
with borrowing
Available credit
(since the total outlay
the rate r m ) and therefore total outlay expands to
profits.
10
credit constrained, an expansion of the availability of
k v \ were so large that F'(k p i)
go
will
In the initial situation the firm
on market borrowing
a corresponding expansion of output and
2:
the point where the
firm's total outlay
therefore credit constrained.
than what the firm would
Result
10
is
possible
from the bank then goes up to
is still less
be
faces will
it
can borrow as much as
borrowing from the market as long as
contrast this with the scenario in figure
borrows the
it
In this case
.
rationed for bank loans, an expansion of the availability of bank credit
rationed in both markets and
the
(i.e.,
the priority sector credit fully substitutes for
if
fc^,
summarize these arguments
not credit constrained
expansion of the availability of bank credit
We
This
fen.
go up because of the additional subsidies but
credit will have output effects in this setting
marginal product of capital
up only
at
be to substitute market
the firm will stop borrowing from the market and the marginal cost of credit
also go
is
outlay and output will remain unchanged.
The expansion
Tb-
where
until the point
Now
effect of the policy will
profits will
market
fcrj.
total outlay
The
ko-
it
at the higher
outlay in this equilibrium
Its total
.
Since at kbi the marginal product of capital
rate.
will
as
borrowed additional resources at the market rate
it
the marginal product of capital
rate,
borrow
free to
< r m then
,
this case as well.
12
bank
credit
there would be substitution of market borrowing in
will lead to
an increase
in its total outlay,
output and
profits,
without any change
in
market
borrowing.
We
hold
have assumed a particularly simple form of the credit constraint. However, both results
instead of the strict rationing
if
The
curve for bank credit.
interpret
it
to
be
The
result also holds
what happens
telling us
supply of cheap credit
is
—what
is
that in figure
firm
More
we
lenders, as long
more expensive sources of
to the
is
is
credit
when
the
drawn
as a horizontal line in figure 2
2,
is
that the supply curve of market credit in this figure
the supply curve of market credit
1,
more than two
generally, the key distinction
unconstrained) while in figure
is
there are
market credit
important
eventually becomes vertical.
is
if
upward supply
firms face an
expanded.
fact that the supply curve of
also not important
we have assumed here the
is
between
figure
1
and figure 2
always horizontal (which
the supply curve slopes up (which
why
is
is
why
the
the firm
is
constrained).
The
results also go
credit (for
through
if
example because bank
might be an increase
in
the market supply curve of credit
credit serves as collateral for
market borrowing as the
market
is
itself
a function of bank
credit). In this case, there
result of the reform but this should
be counted
as a part of the effect of the reform.
One
case where these results
market but not as
with this source).
little
If
the
as
it
fail is
when the
wants (because
it
firm can borrow as
much
as
wants from the
it
wants to keep an ongoing credit relationship
minimum market borrowing
constraint takes the form of a
minimum
share of total borrowing that has to be from the market and this constraint binds, a firm will
respond to the availability of extra bank credit by also borrowing more from the market,
to maintain the required
fail.
However
minimum
as long as there are
substitution of
bank
share of market borrowing.
some firms that
all
In this case, our result
are not at this constraint, there will be
1 will
some
credit for market credit. Therefore the direct test of substitution, proposed
below, would apply even in this case, as long as the
not bind for
in order
minimum market borrowing
constraint does
the firms.
Another case where the
results
would
the firm was choosing whether to shut
fail is if
down
the firm were not making a marginal choice:
or not,
and there was a
fixed cost of operating the
business, the availability of additional subsidized credit might be decisive
13
If
and
in this case, the
effect of
credit
subsidized credit on sales would be positive even
market and had not
of unconstrained firms
fully substituted its
if
the firm were unconstrained in the
market borrowing. Similarly a certain number
would shut down when deprived of
their access to subsidized credit.
This can be addressed by looking at what happened to the firms that were in our sample in
when the subsidy they were
2000,
collection, there
getting were removed.
no systematic difference
is
in exit rates
2000-2002 period. Indeed, rather surprisingly, attrition
observe in the sub-section on data
between large and small firms in the
actually slightly lower for bigger firms
This gives us some confidence that the results we show below are not driven by
in this period.
exit resulting
is
We
from the withdrawal of the subsidy.
i
The supply
3.2
The
side:
understanding lending behavior in Indian Banks
analysis of the supply side will help us build
some
empirical results. In particular we want to understand
cated to firms before and after the reform.
the reform?
more
How
the
is
credit or are
new
Which
intuition about
how
credit?
to interpret the
subsidized bank loans will be allo-
types of firms tend to get more credit before
credit allocated to firms after the reform?
more firms getting
how
Are some firms getting
Are the better firms or the worst firms getting the
marginal credit? Portfolio allocation by credit
officers in
a bureaucratic settings
is
potentially a
complicated problem which we are studying in some parallel research (Banerjee, Cole and Duflo,
2008). Here jve focus on an extremely simplified illustrative example, which provides
to
what we might learn from a more general analysis
The model
is
some
hints
of this problem.
intended to capture a very simple intuition: The two performance measures
for loan officers that are
most
easily observed are the
volume
of his lending
and whether the
loans got repaid. In a large bank, and especially in the highly bureaucratic Indian public sector
banks, this
probably
is
all
that the bank can use to give the loan officer incentives.
words, the only features of firm performance that the loan officer cares about
to borrow
and
The problem
their likelihood of default.
is
that
fact that there has
officer
it
At some
is
also
their willingness
what the bank cares about:
does not observe the ex ante likelihood of default but only the ex post
been default. This introduces a wedge between the incentives of the loan
and the incentives that the bank would have
officer to bail
level this
is
In other
liked
him
to have had, which leads the loan
out failing firms, whereas the bank would have preferred
14
them
to
fail.
3.2.1
We
A
start
simple model of loan allocation
from the model
However
in the previous section.
hand we make a couple of additional simplifying assumptions.
by the bank, equal to
interest rate charged
we
Second, since
If
Where we complicate
H
and
should
this
the model
L, in fractions po
now be
set rt, the subsidized
This simply makes the expressions
less ugly.
we ignore market lending
in
main conclusions would continue
an d
1
-
to hold.
by introducing the idea that firms come
is
The production function f(k)
pq.
in
two types,
of the previous section
interpreted as an expected production function (given that firms are risk neutral,
change does not
type) succeeds
it
is
1,
and correspondingly,
gets f(k). Otherwise
lives for 2
gets
it
periods and there
type
Assume
0.
L
is
pi <
When
1.
as before that f(k)
assume that the
,
a firm (of either
is
strictly concave.
no discounting between periods.
is
We
the second period the firm shuts down.
for
H the
For a firm of type
affect the analysis in the previous sub-section).
probability of success
Each firm
at
every firm started with a fixed amount of market credit (instead of zero)
credit constrained, all our
still
we
First,
find that the firms are indeed credit constrained,
everything we do.
but were
zero.
on the issue
in order to focus
At the end of
firm's probability of success
independent across the periods. Firms do not deliberately default, but
if
is
they cannot
they get
pay (they start with zero and do not retain earnings).
Lending on behalf of the bank
is
also 2 periods
is
by decided upon by loan
and once again, there
is
officers.
Each loan
no discounting between periods. Loan
officer's
tenure
officers are
given
incentives to lend out money,
and
to avoid default. Specifically, each loan officer starts his job
with a population of
new
firms assigned to
new borrower.
to allocate
it
size 1 of
In the second period he
officer is
per unit of default. 11
n When
amount
+ g and
is
unit to each
free to chose
how
is
C.
This assumption
is
a default.
a part of the reason
is
a loan in an Indian public sector bank
(like
why
This punishment
is
F
there are bailouts-it says
bank we study) becomes non performing,
it
triggers the
by the Central Vigilance Commission (CVC), the government body entrusted with
monitoring the probity of public
sector bank,
1
1
more information than the bank). Each unit that
penalized for any loan where there
possibility of an investigation
supposed to lend
is
given a portfolio of size
(since at this point, he has
unlent costs, the banker an
The loan
is
him and
officials.
The
CVC
is
formally notified of every instance of a bad loan
and investigates a fraction of them. There were 1380 investigations
15
of
bank
officers in
in
2000
a public
for credit
that the punishment
linear in the size of the default.
is
Since bailouts are a
way
to substitute
a probability of bigger future default for the certainty of a smaller current default,
making the
We
justify this
penalty convex enough in the size of the default would discourage bailouts.
assumption with the usual convenience argument
settings, the size of the first period loan
to
do with the industry that the firm
for linear incentives
schemes.
In real world
presumably depends on a range of factors that have
is in,
the interest rate in the market, the firm's access to
other sources of credit etc. For each such firm type, the optimal incentives for the loan officer
would require the penalty
ultimately bounded,
it
for default to
be convex over a different range. Since the penalty
cannot be globally convex
—
must therefore
it
also
is
be concave over other
ranges. Linear incentive schemes avoid the need to get these specific details exactly right for a
number
large
In the
of firms types,
first
each borrower
borrower has
type L.
If
he
in large bureaucracies.
period neither the loan officer nor the borrower knows the borrower's type;
a
is
failed
is
which makes them attractive
random draw from the population.
it is
common knowledge between
successful then with probability
that the borrower
a type
is
which makes him a type
H
H
.
With
with probability p\
the loan officer gets no signal the type
In analyzing this
model we
U
will focus
1
officer is the
— 7r,
=
if
the
is
a
all
know
that they
P0
+p (i^ p
\
> Po.We
is
call
that he did not
fail,
the firms on which
firms.
on the case where firms
in
is
1
capital. In the first period
unit to each borrower.
allocation problem the loan officer faces in period
does not have and has the discretion to use
both periods are willing to
given below). Therefore the
one who has to decide how to allocate the available
the loan officer has no discretion-he has to give
3.2.2
period,
the borrower and the lender that he
take the loans that they get offered (the exact condition for this
loan
first
both the lender and the borrower get a signal
it
probability
At the end of the
i.e.
2,
We
are studying the
when he has information
that the bank
it.
Analysis of lending decisions
Given that there
is
a large population of borrowers
the loan officer will have a fraction pon of borrowers
related frauds,
55%
of which resulted in
resulting from being investigated (there
major sanctions.
is
F
we know
who
is
are
that at the end of the
known
to be type
H and
period
have been
naturally thought of as the expected punishment
clearly a cost of being investigated even
16
first
if
you are innocent).
successful, a fraction (1
-
po){l
- pi)
known L
°f
types
who have
have also been successful, of an unknown type (type U). The
to
do about the firms that have
of
F
or bail out the firm by giving
The loan
the
new
officer will only bailout
loan. If the bailout loan
unit back to the
we
bank
require that /(/
such that f(l*
—
1)
so that
-
1)
=
/*.
>
/.
He can
failed.
it
when
if,
is
I,
How
and take
2nd
the firm succeeds in the
the firm only gets to invest
The minimum
never any reason to lend more than
it
what
punishment
his
I
—
1,
period,
it
can pay back
because
it
has to give
/*
>
loan size that will allow a successful bailout
1
to a firm that
/*
is l*
Since these firms are of type L, they are more likely
1.
other types of firms are willing to borrow more (which
giving
is
does not default right away. Therefore for a successful bailout
Obviously
I*,
who
rest,
big does the bailout loan need to be?
to default in the second period than either of the other types of firms.
between giving the firm
and the
decision he has to take
either report a default
a fresh loan. 12
it
first
all failed,
which generates a
is
is
what we
will
being bailed out.
Hence, as long as the
assume below), there
The
choice
is
is
therefore
possibility of a larger default in the future
nothing, which leads to a smaller but certain default now. Bailing out dominates
and
if
F>(l-pL )l*F
which clearly holds,
if
(and only
if)
l-PL<p
Assume
the
first
(1)
that this condition holds so that the loan officer always bails out those
period. There
is
no scope
for bailing
out in the second period because there
who
is
fail in
no future
in the relationship.
Given
among
firms.
the rest of the firms.
This °
gives them each
In other words,
12
if
I
Assume
—
~ p °>
9 ~'
f{
will
have
this the loan officer will
it.
g
~ PL
{
-
POT
in the
(I
-
po)(l
- PlY*
>
units.
Assume
units of capital left to allocate
among
known type
H
L
(2)
PQTT
is
divided equally
among
the
known
Since this also minimizes risk of default, this
Indian context
the
that
i+9-(i-Po)a- P L)i* >
This process of "evergreening" of loans by loan
been widely noted
-
that he divides this equally
the remaining capital
be happy to take
+
(see, for
officers
who
is
H type firms,
what the loan
they
officer
prefer not to have a default in their hands, has
example, Topalova (2004)), as well as elsewhere in the world.
17
should do. 13
Notice as long as this condition 2 continues to hold, this result does not depend on the size
Hence
of g.
is
as a result of a policy shift, the
if
amount
of subsidized credit available for lending
larger, the essential pattern of lending does not change:
H firms,
known type
also
be true
Result
credit
is
change the
The
down
as long as
Under assumptions
3:
rest
H firms get a bigger increment in their loan.
but now the type
g went
to give
and the
I*
if
H
known type
type
(i.e.
U
it is still
and
1
the loans go to the type
the case that
+g -
I
(1
— po)(l —
L and
to
This would
>
pi)l*
0.
2 the loan officer's optimal allocation of second period
+9 ~^ ~^
amount
firms an
type
'
,
Variation in the size of
firms) nothing.
set of firms that get loans in the
~ PL
^
g,
L
amount
firms an
within limits, does not
second period.
logic of this result is straightforward.
The
Hence he
loan officer wants to avoid default.
the existing firms that are in trouble but otherwise would like to focus entirely on
will bailout
the firms that are proven to be safe. Given that subsidized credit
be happy to take what he
is
is
scarce, these firms will also
offering them.
U
actually get a cut in their loan seems counterfactual
at least in the world of Indian firms. In our
data many firms show no loan growth, but few see an
The
prediction that the firms of type
actual decline. This
to default,
and as a
may be
because
if
the firm anticipates a large cut in
that loan size never goes
first
down
period loan was repaid.
If
we make the
as long as the first period loan
is
loan
is
Under the assumption that loan
4:
repaid, as well as assumptions
period credit
is
to give type
an increment oil* —
1
and
1
H firms an loan
and the
rest
it
will prefer
(i.e.
type
size
and
auxiliary assumption
is
as:
as long as the
period
first
the loan officer's optimal allocation of second
2,
increment of
U
down
never goes
first
and assume that g
repaid,
always large enough to allow this to happen, Result 3 would be restated
Result
loan,
want to commit to not cut loans between the
result loan officers
second period as long as the
its
firms)
fl
~'
~
~
"^
»
p
no increment. Variation
L
the type
firms
in the size of g,
within limits, does not change the set of firms that get increments in the second period.
13
the
We
H
are cheating a bit here.
loan officer
is
actually indifferent between dividing the capital equally
types and a range of other allocations where some
division
outcome
lobby the loan
is
The
is
socially efficient (because
officer for
/
is
H
types get more than others.
strictly concave)
each extra dollar and those
and
who have more
concave).
18
also the one that
to gain lobby
among
However the equal
would obtain
more (once
if
the firms
again, because
/
Implications of results
3.2.3
Under the conditions
1
and
2, this
very simple model therefore has several interesting implica-
tions.
1.
The
relation between loan growth
first
period revenue,
negative, or zero.
may be
second
period profits) in the cross-section of firms, can be positive or
first
The
and ex ante measures of firm performance (such as
firms that have the highest loan growth from the
either the best performing
(depends on how
I*
compares with
+g ~
(
H type firms or worst
~£
~
first
performing
) The intermediate
^
U
period to the
L
type firms
type firms get
no increments.
Note that
this
is
quite consistent with the descriptive evidence reported in section
we showed no systematic
relationship between measures of firm performance
where
2,
and probability of
a loan increment or amount of the increment.
2.
A
substantial part of loan growth under normal circumstances goes to firms that get bailed
out because they have failed (and are thus known to be bad). These firms are more likely
to
again than the average firm. Therefore, the
fail
growth
will
be biased downwards, since
it
OLS
estimate of loan growth on profit
confounds this (negative) selection
effect
and the
causal effect of loans. In contrast, the immediate impact of an unexpected policy change
that increases g
is
an increase
in credit flows to firms
firms in our model). Therefore,
on
profit using the policy
that are expected to do well (type
an instrumental variable estimate of the impact of loans
change as an instrument
for
change in lending
causal impact of extra lending on successful firms. This
us the
''local
average treatment effect" (LATE),
on the type of firms
The IV
for
H
i.e.
is
will give us
the
because the IV estimate gives
the effect of additional unit on credit
which credit actually changes
will therefore typically
represent a causal effect,
it is
be larger than the
a causal
effect
OLS
for
two reasons: While
19
does
within a selected group (in other words, the
"compilers" in this experiment will tend to have higher treatment effect than a
firm chosen from the population).
it
random
3.
The
set of firms that
have credit growth
is
magnitude of the credit inflow changes. This
and the loan
credit
them
if
more
effect of the
if
officer
available)
unchanged by the policy change-only the
is
always wants to give
because every firm wants more subsidized
to the safest firms (and to give
it
and therefore has no reason
to try to spread
it
more
to
around. All the
reform should therefore be on the intensive margin.
Empirical Strategy
4
Reduced Form Estimates
4.1
The
empirical work follows directly from the previous section and seeks to establish the facts
that will allow us to determine whether firms are credit rationed and/or credit constrained.
Our empirical
1998 and
its
strategy takes advantage of the extension of the priority sector definition in
subsequent contraction
in 2000.
The reform
the composition of clients of the banks: in the sample,
big firms have entered their relationship with the
banks was no more
affected
by sample
on big firms
likely to take
in
of the small firms,
and 28% of the
1998 or 1999. This suggests that the
and that our
after the reform
all
the outcomes
we focus on the proportional change
log(limit granted in year t-1).
limit faced
25%
results will not
be
selection.
Since the granted limit as well as
correlated,
bank
did not seem to have large effects on
by the firm
14
we
will consider, are
in this limit,
i.e.,
very strongly auto-
log(limit granted in year t)
As motivation, table 4 shows the average change
in the three periods of interest (loans
January 1998, between January 1998 and January 2000,
after
in the credit
granted before the change in
January 2000) separately
largest firms (investment in plant
and machinery between Rs. 10 million and Rs. 30
medium-sized firms (investment
in plant
and the smaller firms (investment
and machinery between Rs.
in plant
—
and machinery below Rs.
6.5
for the
million), the
and Rs. 10
million),
6.5 million).
For limits granted in 1997 the average increment in the limit over the previous years's limit
was 7%
larger for the small firms
compared
1
in
compared
to
medium
to the biggest firms. For limit granted in 1998
Since the source of variation in this paper
is
firms
and
and 1999,
it
2%
larger for small firms
was
2%
closely related to the size of the firm,
log to avoid spurious scale effects.
20
larger for
we express
all
medium
the variables
and 7% larger
firms,
2000, limit increases were smaller for
firms,
compared, once again to the smallest
for the biggest firms,
B
in table 4
14%
of
shows that the average increase
in the probability that the
After
but the biggest declined happened for the larger
all firms,
whose enhancement declined from an average
Panel
firms.
1998 and 1999 to
in
in the limit
0%
in 2000.
15
was not due to an increase
working capital limit got changed: big firms were no more likely to
experience a change in 1998 or 1999 than in 1997. This
the model in the previous section, which
tells
us that
consistent with implication 3
is
when loan
officers
from
need to respond to
pressure from the bank to expand lending to the newly eligible big firms, they prefer giving
larger increases to those
be
safe, rather
In Panel C,
which would have received an increase
than increasing the number of firms whose
we show the average
limit's
in
any case and are known to
are increased.
increase in the limit, conditional on the limit having changed.
The average percentage enhancement was
firms in 1997, smaller for the small firms than for the large firms in 1998
the same for the
medium
firms),
medium and
larger for the small firms than the
and larger
after 2000.
and 1999 (and about
The average enhancement
a change in limit declined dramatically for the largest firm after 2000
(it
large
conditional on
went from an average
of 0.44 to an average of slightly less than 0).
Our strategy
bank
of
will
be to use these two changes
credit to the
medium and
in policy as
a source of shock to the availability
larger firms, using firms outside this category to control for
possible trends.
We
start
by running the regression equivalent of the simple difference-in-differences above.
First use the data
log k blt
from 1997 to 2000 and estimate and equation of the form: 16
-
log fcwt-i
= a lkb BIG +
l
where we adopt the following convention
limit to firm
t
—
l
17
),
i
BIG
between Rs.
in year
is
a
t
Note that there
is
POST + llkb BIG, * POST + e lkm
t
for the notation:
(and therefore granted
dummy
6.5 millions
lkb
(i.e.,
kba
is
(3)
,
a measure of the bank credit
decided upon) some time during the year
indicating whether the firm has investment in plant and machinery
and Rs. 30 millions, and
no reason to expect a decline
in
POST
is
a
dummy
equal to one in the years
the loan limit for the larger firms
in
2000,
we
just expect
a bigger drop in the increase in limit between 1998-1999 and 2000.
16
All the standard errors are clustered at the sector level.
17
70%
of the credit reviews
happen during the
last 6
months
21
of the year, including
15%
in
December
alone.
1999 and 2000 (The reform was passed in 1998.
therefore affected the credit decisions for
It
the revision conducted during the year 1998 and 1999, affecting the credit available in 1999 and
2000).
entire
We
focus on working capital loans from this bank. 18
sample and
the loan.
We
We will
sample of accounts
in the
expect a positive y\ kb
for
We
estimate this equation in the
which there was no revision in the amount of
.
also run a regression of the
same form using a
dummy
for
whether the firms got any
increment as the dependent variable. The model predicts in this case that the coefficient of the
variable
BIG*' POST should be
an increment greater than
To study the impact
zero. Finally,
equation
(3) will
be estimated in the sample with
zero.
on bank
of the contraction of the priority sector
loans,
we use the
1999-2002 data and estimate the following equation:
-
log k bi t
where
BIG2
is
log k bi t-i
a
dummy
= a 2kb BIG2 +
t
firms that got
p,
.
Once
POST2 + e 2kbiU
(4)
t
millions,
and
POST2
is
a
dummy
equal to one in the years
positive increment
we
will also
and
for the
estimate a similar equation for an indicator for
in the limit.
we pool the data and estimate the equation:
log k
18
*
again, this equation will be estimated in the whole sample
whether the firms had any change
Finally,
POST2 + l2kb BIG2,
indicating whether the firm has investment in plant and machinery
between Rs. 10 millions and Rs. 30
2001 and 2002. 19
(3 2kb
m
-
log kbu-1
=
cyzkbBIG2
l
+ a 4kb MEDi + p3kb POST + f34kb POST2 +
j3kb BIG2 t
*
POST + likb MED
%kbBI.G2i
*
POST2 + jekbMEDi * POST2 +
t
l
*
POST +
t
t
t
Using total working capital loans from the banking sector instead leads to almost identical
e 3kbiu
(5)
results, since
most
firms borrow only from this bank.
19
Once
again,
we adopt the convention that we look
t—\. The reform was passed
in
at credit available in year
t,
and therefore granted
in
year
2000 and therefore affected credit decisions taken during the year 2000 and credit
available in the year 2001.
22
where
MED
is
dummy
a
indicating that the firm's investment in plant and machinery
and Rs. 10
Rs. 6.5 million
is
between
million.
After having demonstrated that the reform did cause relatively larger increases in bank loans
for the affected firm,
First,
(5).
we use
is
of other regressions that exactly parallel equations (3) to
the sample 1997-2000 to estimate:
yit
where yu
we run a number
-
= a ly BIG + PiyPOSTt + ^ y BIG
ytt-i
l
an outcome variable (such as
r
*
POST +
t
credit, sales, or cost) for firm
e lyU
(6)
,
in year
i
t.
Second, we
estimate:
logy,,
in the
-
log.i/it-i
sample 1999-2002
log yit
-
,
= a 2y BIG2i +
and
log y lt - x
finally
we
j3 2y
= a3y BlG2i + a 4y MED +
identified as
t
e 2yit
,
(7)
POST + p4y POST2 +
POST + l4y MEDi * POST +
*
t
POST2 + jeyMEDi * POST2 +
*
t
t
t
t
predicts that only impact of the reform
good
failed will also get
will
now
will not receive
an increase
an increment.
estimate equation y\ to
j/3
is
e 3yit
(8)
on the intensive margin: firms pro-
in loan to
but on which the credit
fail
Under the assumption
in the
separately in two sub-samples:
Sample
model,
it
is
not be affected
has no information
thus appropriate to
the sample with an increment in
BIG * POST)
(Heckman
(1979),
(1986), Angrist (1995)).
prediction that selection of firms getting positive increment
consistent with
officer
will
had previously
selection will not bias the results, because
uncorrelated with the regressors of interest (the variable
The
firms which
be bailed out, but that probability
and the sample without increment.
Heckman and Robb
Some
get a larger increase in their loan.
by the reform. The firms which did not
is
POST2 +
*
1997-2003 sample.
Our model
it is
3y
z
+ 75yJB/G2
limit,
y2y BIG2i
estimate:
l3y BIG2 t
in the
POST2 +
what we observe
in table 3
and
23
4.
is
uncorrelated to the reform
In particular, there isn't any evidence that
the probability of a change in the limit
affected
is
by the policy change.
be the case that the number of firms that get a change
but the type of the firms that get chosen
results in the selected sample.
after the reforms, failing firms
when we
Empirically,
the variables
However,
is
affected
this is
It
could of course
in the limit is unaffected
still
by the reform,
by the reform. This could then bias the
not what the model predicts. Both before and
and firms that have been
identified as efficient should
be selected.
regress pre-determined characteristics of firms with positive increment on
POST, BIG and BIG
*
POST
before and after the reforms,
we
see
no impact
(results omitted).
If
the assumptions in the model are right,
POST
limit,
and
BIG2
POST2
*
which provides a
sales, costs
and
all
BIG
*
Restricting the sample to firms
in limit will also increase the precision of the
profits for firms
coefficients of
the equations in the sample without change in
test of the identification assumptions.
with a positive increment
on
to be zero in
we should then expect the
estimates of the reform
which were actually affected by the reform.
Below, we describe the variables we use and their justification.
• Credit rationing
Following result
First,
1
,
we show that the
for a credit limit: the
we provide two
additional pieces of evidence to establish credit rationing:
firm used the extra funds they got, using a standard measure of utilization
logarithm of the ratio of total borrowing under the line of credit during
the year (in banker's parlance, the turnover on the account) to the credit limit.
check that the interest rate did not change (which
•
If
a firm were credit constrained, our theory
if it
credit for their
available (the data
on
i.e.
sales,
success:
tells
us that sales revenue would definitely go
were not, sales should only go up for firms that have already
of working capital,
year),
what would be expected)
Credit constraints
up, while
bank
is
Second, we
market borrowing. Given that we are looking
we expect the
fully substituted
at increases in the availability
increase in sales to take place the year the capital
shows that the working capital limits
is
becomes
turned over several times during the
the year after the limits was decided upon. To interpret the effect of credit expansion
we
posit a simple parametric relation between credit
Rn = Aukf
t
.
Note that
this
is
and
sales
revenue in the case of
a specific parametrization of the production function
24
introduced in the previous sub-section. 20 Taking
log
R =
log
it
logs:
A +
log k it
it
(9)
.
DifFerencing this equation gives:
-
log Ri t
for firms that
log J^t- 1
=
log
A
it
- log At- 1 +0[logk u -
succeeded in both periods. Assume that when firms
of sales such that log
Focusing on the
Rn = vn
first
(failing likely
does not involve zero
logfcit^i],
(10)
they get a small amount
fail,
but rather zero
sales,
experiment (credit expansion), the growth of bank credit
that were successful in period
1,
and
for
which the bank received a signal
profit).
21
for the firms
given an equation
is
of the form of equation (3).
In the absence of complete substitution between
same shape
a relationship of the
log
log
Ru - log flit-i =
log
in
credit
and market
credit, this implies
for capital stock:
= aiSkBIGi + l3 lSk POST + -n Sk RIGi
ku ~ log ha-!
which when substituted
bank
t
*
POST +
t
e
lku
(11)
,
equation (10) yields
Au - log At-i + O^skBIG, + 3 1Sk POST + j! Sk BIGi * POST + e lklt
t
t
).
(12)
If
we
restrict the
such firms
in
period
sample
t:
to firms with a positive increment in limits, there are
the firms that failed in period t-1 (which needs to be bailed out), and
the firms that succeeded in period
The 'latter
t-1
and about which the bank received a positive
firms are successful in both period, so their sales
the case of the failed firms the loan increment they get
to be bailed
20
This
is
out)-.
In period
t-1, they
get the failure
best thought of as a reduced form, derived from a
Cobb-Douglas function of the amount
working capital and
all
two kinds of
of
n inputs
x\,X2...x n
.
is
fixed
is
signal.
governed by equation
(it is
minimum
the
outcome (log(Rlt -i)
=
itjt-i-)-
12.
In
they need
In period
more primitive technology which makes output a
As long
inputs are purchased in competitive markets,
as the inputs
it
have to purchased using the
can be shown that the resulting indirect
production function has the form given above.
21
It
would simplify the exposition
proportional to the
avoid making
amount
lent.
to say that the failing
But we show below that
it.
25
this
outcome
is
is
simply a low
An
not a necessary assumption
so the sales
in this set
remain
up, so
we
t,
they get the success outcome with probability p (and this equal to
minimum
1
in
—
Thus, for firms that failed in period
p).
and the
capital a failed firm needs to be bailed out,
revenues
it
ki t
,
where ku
is
the
outcome with probability
neither the increase in loans nor the increase
more money
correlated with the reform, though they might get
is
in period t-1,
t-1,
failure
A
in
period
t
than
and big firms may be getting more money because they need more to survive. In
other words, for these firms, the increase in loans and revenues in both periods take respectively
the form:
log
ku - log
h^
=
+
ai Fk BIGi
lFk
POST +
t
u> lklt
(13)
,
and:
log
Combining
failed firms
we estimate equation
where
7isfc6<f>,
B* -
4>
is
= a 1FR BIG +
z
and successful firms
(3
lFR POSTt
in the loan
+ mkit,
equation (which
(14)
is
(6)
above
in the
Our
BIG * POST
by the fraction of successful
two can be interpreted
identification hypothesis
\ogA it and that a similar conditions
payoff
is
—
sample sample (with sales as the dependent variable,
in this equation) implies that 71/2
Thus, the coefficient in both equations are the causal impact of the reform
ratio of the
"jikb
the share of successful firms in the sample of firms with positive increment,
71.R the coefficient of
firms, multiplied
what we do when
3 in the sample of firms with positive increment) thus implies that
and estimating equation
and denoting
log Ikt-\
is
is
as
firms.
In section 5.4 below,
=
6f\sk4>-
for the successful
we
discuss that the
an IV estimate of the impact of bank loan on revenues.
that for successful firms
log Ait-x
=
ai A BIGi
+ 0i A POSTt + tit
t
(15)
true for failed firms (neither the failed payoff nor the successfull
correlated with the interaction
BIG * POST).
This amounts to assuming that the rate of change of
A
(which
is
a shift parameter in
the production function) did not change differentially for big and small firms in the year of
the priority sector expansion.
Under
this
assumption
expansion of the priority sector on sales revenue.
26
7/? gives
the reduced form effect of the
If
we consider the
entire
sample instead
positive increment, the reasoning
coefficients of
BIG * POST
if
looking only at the firms that have received a
would be exactly the same, except that
in the sales
in that case,
the
equation and in the loan equations are both multiplied
by the share of successful firms on which the loan
officer
has received a signal in the entire
sample.
Similar calculations lead to an equation of the
priority sector contraction (2000-2002),
log
If
A ~
log
it
where the
Au -i =
same form,
similar to equation (8) for the
identification hypothesis
a 2A BIG2 +
l
firms are credit constrained, 7i/{ should be positive
(3 2A
and
POST2
•yok
t
is
that
(16)
.
should be negative, while
if
no
firms are credit constrained 71^ will only be positive for those firms that have fully substituted
market
We
credit,
and
72;? will
be negative only for those firms that had no market credit
therefore also estimate a version of equation (6) in the sample of firms
liabilities
exceed their bank credit.
72/? in this
A
sample should be
final piece of
If
the firms were not credit constrained, the value of 77? and
evidence comes from looking at
is
profit.
Profits are expected to increase
credit constrained or not (since the interest
down), but the extent of the increase in profit
is
nevertheless interesting, since
payments go
it
can give us
of the firm's marginal return to capital.*'"
Empirical Strategy: Testing the Identification assumptions
4.2
The
interpretation of the central result on sales growth crucially depends on the assumptions
made
equation (15) and (16). Likewise, the interpretation of the other results depends -on
in
the assumption that the error term
BIG * POST
reasons
why
sectors
in equation (6)
this
assumption
and
may
is
not correlated with the regressors, most importantly
BIG2
*
POST2
may be
In the.
However, there are
in equation (7).
many
may be
differently
different sectors,
and these
not hold. For example, big and small firms
by other measures of economic policy (they could belong to
affected
22
total current
zero.
regardless of whether the firm
some indication
whose
initially.
affected by different policies during this period).
working paper version of
(.he
paper,
we derive and diseuss interpretation
various assumptions.
27
of
t.he
results
on
profits
under
The
fact that
we have two experiments
effect of the priority sector regulation
The two reforms went
affecting different sets of firms help distinguishing the
from trends affecting
and did not
in different directions
different groups of firms differentially.
affect all the firms identically.
Credit
constraints would predict 71 /? in equation (6) to be positive and 72/? in equation (7) to be
negative. Moreover, the ratio
The same reasoning
•^La
-
and -^- should be equal.
of course applies to equations (5)
~ ^3L
experiments), as well, so that the ratios ~^-,
over-identification test:
the observed patterns
if all
,
should also
these equalities are satisfied,
come from the
fact that the
and
it
(which combine the two
(8)
all
be equal. This
is
a natural
would be extremely implausible that
time trends are different for small and large
firms.
Even
we would
these tests work,
if all
still
need to worry about the possibility that, being
labeled as a priority sector firm affects the sales and profitability of a firm over and above
on credit access.
effects
the right to manufacture certain products
concern by using profit before tax in
among
firms,
the small firms,
24%
One
do.
all
44% manufacture
control strategy
of excise taxation. Second,
reserved for the SSI sector.
is
specifications.
We will
address the
first
The second concern could be a problem:
a product that
would be to leave out
that are reserved for SSI. Unfortunately,
in 1998.
exempt from some types
First, SSI firms are
its
is
all
reserved for SSI.
Among
the big
firms that manufacture products
we only know what products
Excluding firms that manufactured SSI reserved products
in
the firm manufactured
1998 does not change the
>
results.
However
after 1998
A way
which
is
and
it
remains possible that some of the big firms moved into reserved product
this increased their sales
to resolve this issue
is
and
profits.
to focus on a different test of the identification assumption,
to estimate equations (6) to (8) for
all
the different outcomes variables separately in
two subsamples: one subsample made of the firm-year observations where there was no change
in the granted limit
from the previous year to the current year, and one subsample made of
firms where there was a change.
products on the SSI
list
change after the reform.
If
there
is
an effect of just becoming entitled to produce the
even the big firms that had no change
We
in
the granted limit should
therefore test whether the coefficient of
BIG*POST is statistically
indistinguishable from zero in the sample of firms that did not get a change.
above this
is
consistent under the assumptions of the mode.
28
show a
As we discussed
Results
5
Credit
5.1
•
Credit Expansion
Panel
ables.
23
We
(columns
granted
A
in table 5 presents the results of estimating
start with a variable indicating
(1)),
(3) for several credit vari-
whether there was any change in the granted limit
and two dummies indicating whether there was an increase or a decrease
Consistent with the model and the evidence
limit.
equation
we
in the
discussed above, there seem to be
absolutely no correlation between the probability of getting a change in limit and the interaction
BIG * POST.
Moreover, even the main effects of
seem
variables in this regression
There
is
also
no
whether the
to affect
effect of the interaction
BIG
and POST'&xe very small: none of the
was granted a change
file
in limit or not.
on the probability of getting an increase or a decrease
in the limit.
we look
In the columns (4), to (7)
by the bank. 24
at limit granted
As the
descriptive
evidence in table 4 suggested, relative to small firms, loans from this bank to big firms increased
significantly faster after 1998 than before: the coefficient of the interaction
in the
complete sample, and 0.27
sample
in the
of these coefficient are statistically significant,
credit for the
which there
and indicate a
dummy
POST
for
is
BIG
is
-0.22,
In
columns
(6)
and
(7)
,
BIG, although the
we
restrict the
The
almost the same (0.26) and
23
The standard
is
errors in
all
stage for the
is
average enhancement
enhancement than small firms (the
The gap completely
actually larger in absolute value than
difference
IV estimation
be the
coefficient
in the
is
small).
sample to observations where we have data on future
sales (which will
first
Both
large change in the availability of
with a standard error of 0.088).
closed after the reform (the coefficient of the interaction
the coefficient of the variable
in limit.
negative). Before the expansion of the priority sector,
large firms were granted smaller proportional
coefficient of the variable
any change
is
sample of firms that were reviewed. There was a decline
for small firms (the
medium and
for
POST*BIG is 0.095,
still
of the impact of
bank loans on
sales).
significant.
regressions are adjusted for heteroskedaticity
and clustering
at the firm
and sector
levels.
If,
this
instead,
simply
we use
the
sum
of the limits from the entire banking sector,
reflects the fact that
most firms borrow only from one bank.
29
we obtain
virtually identical estimates:
Credit contraction
•
we present the
In panel B,
result of estimating equation (4).
the contraction on the probability that the limit
is
Here again, we find no
changed (column
effect of
which reinforces the
(1)),
claim that the decision to change the limit has nothing to do with the priority sector regulation.
However, the probability that the limit
is
cut goes up significantly for the largest firms after the
reversal of the reform in 2000 (the coefficient
magnitude of the change
to the
(column
in limit
for big firms after
2000
is larger,
we
MED
firms
*
POST2
became
is
the coefficient
and
positive
less likely to
-0.12)
significant in
column
POST
in
(6)
and
(1):
The
the regression).
in
25
(7)).
may be because
It
main
coefficient
Relative to other firms,
experience a change in limit after 2000.
The
1998 and 1999.
to ^§kb (the corresponding
73^;,
is
and the sample with
(columns
sales
*
Turning
yearly decline in the limit
than the average yearly increase in limit
present the interaction coefficients
is
is
The average
-0.44).
presented in the tables, but were included
effects are not
of
(4),
sample where we have data on
results are very similar in the
In panel C,
the coefficient
(5),
BIG2
in limit, the coefficient of the interaction
negative both in the entire sample (in column
a change
0.119, with a standard error of 0.033).
is
medium
they have
experienced relatively large changes in the two years before.
The
in
on the magnitude
effect
column
(whole sample) and
(4)
change
in the
(5)
in the limit granted
by the banks are presented
where the
was changed). During the
(the sample
expansion of the priority sector, the limits of both
more than that of small
firms,
firms.
both of which became
The impact
medium and
of the reform
limit
large firms increased significantly
was similar
During the contraction, large
eligible.
for
firms,
medium and
who
experienced a significant reduction in their credit limit relative to small firms.
(who did not
lose eligibility) also suffered a decline but the coefficient
for large firms.
that of
25
The sample
The
earliest
column
BIG * POST
size
on loans but not on
26
(In
effect
is
(5) for
-0.48.
example, the coefficient of
Only the
later
is
significant).
much
is
MED
*
large
lost eligibility,
Medium
firms
smaller than that
POST2
is
-0.18, while
26
drops in this column since we are not using the data from the
last year
when we have data
medium
firms based on the
sales.
on medium firms may come from the
data we have on them (1997).
Some
of
fact that we, classified firms as
them have almost
certainly
by the bank as large firms, even though we are treating them as medium
30
grown since and are now being treated
firms.
5.2
Evidence of Credit Rationing
Table 6 presents evidence on credit rationing.
B
experiment, and panel
Columns
(1) to (3)
As
before, panel
A
focuses on the expansion
focuses on the contraction experiment.
present the results for the interest rate.
The
first
column shows
second column logarithms, and the third column replaces the difference
indicating whether the interest rate
fell
in
between the two years.
rt
—
levels,
by a
r t -\
the
dummy
There seems to be strong
evidence that the interest rate did not decline for big firms (relative to small firms) as they
entered the priority sector.
interaction
BIG
*
POST
is
In
all
three samples and for
insignificant in panel A,
relative increase of the interest rate, rather
point estimate
is
is
,all
three measures
we consider, the
and the point estimate would suggest a
than a decrease. In the complete sample,
0.073, with a standard error of 0.17.
27
in levels,
the
In logs the coefficient of the interaction
0.002, with a standard error of 0.011. In panel B, the coefficient of
BIG2 * POST2
is
likewise
insignificant in all the specifications.
This shows that the fact that big firms are borrowing more from the banks after the expansions and
lending.
less after
To complete the argument we
credit they get
When wo
and
the contraction,
when
there
is
is
also
not explained by a
fall in
the interest rate on bank
need to show that firms actually use the additional
an expansion. 28
To look
at this,
we compute
use this variable as the dependent variable, the coefficient of
insignificant
limit utilization.
BIG * POST
is
negative
both during the expansion and during the contraction.
This results are far from definitive, due to the limited number of observations for which the
data on turnover
is
available.
29
However, the evidence available suggests that firms did make
use of the extension in credit without a change
This suggests that firms are
in interest rate.
willing to absorb the additional credit at the rate at
which
it
is
turn to sales and profit data to assess whether firm's activity
offered
is
by the bank.
We now
constrained by their limited
access to credit.
27
28
line,
29
The average change
This
is
in interest rate in
sample period was 0.34, with a standard deviation of
0.86.
not automatic, since under the Indian system the bank gives the firms an extension of their credit
but firms only pay
for the
amount they
actually draw.
For example, we do not present the results for loan utilization
very to few observations on turnover in each
cell in this restricted
31
for firms
sample.
whose
limit changed,
because we have
Evidence of Credit Constraints
5.3
Table 7 present evidence on credit constraints.
•
Credit Expansion
In panel A,
column
(1),
In order to keep the table
we
by looking at the impact of the credit expansion on
start
manageable, we present only the coefficient of the interactions, which
are the coefficients of interest (the coefficients of the
Of note among unreported
in absolute value
The
and
sales.
coefficients
insignificant in
coefficient of the interaction
effects are available
the coefficient of the
is
all
main
specifications
BIG * POST
is
and
for all
variable,
which
is
small
dependent variables.
0.194 in the sample with a change in limit,
with a standard error of 0.106. In the sample where there
increase disproportionately for large firms:
'TOST"
upon request).
is
no change
in limits, sales did not
the coefficient of the interaction
0.007,
is
with a
standard error of 0.074. This supports our identification assumption that the difference in the
annual rate of growth of
The
An was
not differentially affected in the year 1999.
increase in sales suggests that firms were not only credit rationed, but also credit con-
strained, unless
we
We
reliable
do not have
are in the case where
data on market
ence between total current
liabilities
bank
credit,
credit completely substituted for
but we have a proxy
and the bank
limit.
column
In
or smaller).
The
sample (0.168): the increase
full
coefficient of
in sales is
BIG * POST
is
credit.
for trade credit, the differ(2)
we
restrict the
to firms that, according to this measure, have not stopped using trade credit
has not become
market
(i.e.,
similar as
not due to firms that had
first
this
what
it
sample
measure
is
in the
completely substi-
tuted away from trade credit. Moreover, note that very few firms drop from the sample where
we
focus on firms that have positive non-bank liability
borrowing), which in
and
(2)
itself
(i.e,
we drop
suggests that substitution cannot be easy.
firms without any market
The
results in
column
(1)
together with the previous results establishing credit rationing, suggest that firms are
credit constrained: sales increased for firms that
substituted entirely. Below,
have been
little
we use the magnitude
substitution of
Although finding an
effect
bank
on
credit for
profit
effect
on
profit
is
had non-bank
credit,
and very few firms
of the estimates to argue that there
market
would not be
constraints (since part of the effect on profit
magnitude of the
still
comes
credit.
sufficient to establish the presence of credit
directly
from the subsidy), establishing the
a useful complement to the results on sales.
32
seems to
Using the
logarithm of profit as the dependent variable presents the difficulty that this variable
defined whenever profit
We
negative.
is
can thus only estimate the
on
effect
is
not
profit for firms that
have a positive profit in both periods, which introduces sample selection and makes the profit
regressions difficult to interpret.
To avoid
this problem,
we look
(defined as sales-profits), which
always defined. The
is
effect
on
profit for
any particular firm or
can then be recovered from the estimate of the reform on sales and costs,
for the average firm
without sample selection
bias.
comparable magnitude: the
with change in
impact of the reform on the logarithm of cost
at the direct
limit,
The
increase in sales
coefficient
and only 0.005
on the
is
accompanied by an increase
BIG * POST
interaction
sample without change
in the
is
0.187 in the sample
in limit.
For comparison, we also present the results on directly estimating the
column
in the
•
(4).
The
effect
on
sample with change
profit
in limit
The
very large.
is
is
0.54,
in cost of
profit,
equation
in
BIG * POST
coefficient of the interaction
with a standard error of 0.28.
Credit Contraction
Panel
B
presents the estimate of the effect of the credit contraction on the sales and costs
of firms with investment in plant
firms as a control)
and Rs. 30
and machinery between Rs.
million.
10 million (using
In the sample where there
coefficient of the interaction
BIG2*POST2
of 0.207). Here again, there
is little
was a change
BIG2
*
POST2
the other
in limit, the
negative and large (-0.403, with a standard error
is
evidence of substitution.
The
result
the analysis to the sample of firms that have some market borrowing.
interaction
all
in the cost equation
is
is
similar
The
if
we
restrict
coefficient of the
negative and similar to the effect on sales
(-0.374).
In the
sample where there was no change
either on sales or
•
Pull
on
in limit, in contrast there is
tests
Table 8 present the results of estimating equation
MED
*
and we estimate separately the
POST, BIG2 * POST2
and
MED
*
(8)
for sales
and
costs.
coefficients of the interactions
POST2
the firm's investment in plant and machinery
We
effect
costs.
sample and overidentification
entire period,
no significant
is
(where
MED
is
a
dummy
BIG
*
use the
POST,
indicating that
between Rs. 6.5 million and Rs. 10
also present in the table the ratios of the interaction coefficient in the
33
We
million).
outcome equation
and
to the corresponding coefficient in the loan equation (from table
In the sales
of the
and cost equations, the
MED POST
*
and
BIG2
separately, they lose significance).
coefficients
POST
*
The
BIG2 * POST2
coefficient of the interaction
20%
of the
POST2
coefficient
and
(7)).
when introduced
interactions are positive (though
significant and, while negative, the coefficient of the interaction
*
panel B, column
have the expected pattern: both the coefficients
and
BIG2
5,
The
insignificant.
MED
*
is
negative
POST2
is
only
coefficients are similar in the full
sample and the sample "without substitution.
Formally, the overidentification test does not reject the hypothesis that the implied effect of
credit on the sales
if
we look
and cost variables
at the sales equation in
is
the
column
same
(1),
for all the
sources of variation. For example,
the ratio between the coefficients in the sales
equation and the corresponding coefficients in the loan equation are similar (they range between
0.73 and 0.83), and the test does not reject the hypothesis that they are equal.
makes
it
This result
very implausible that the estimated coefficient reflect differential trends arising from
other, unobserved, factors.
Taken together, these
credit constraints.
The
results present a consistent picture
sales of the firms affected
resulted in an expansion in credit, and decreased
of firms that
which suggests that firms face
by the reform increased when the reform
when the reform
led to a contraction.
was affected by the expansion, but not the contraction, behaved
These
firms in the expansion, but like an unaffected firms in the contraction.
together suggest that
it is
unlikely that the effects are driven
like
by time trends
A
subset
the affected
results taken
affecting different
firms differentially. Furthermore, these results are concentrated in the firms that experienced a
change
A
in loans,
which makes
last piece of
it
unlikely that the effect
important evidence
in the probability of default
:
is
officially qualified as
driven by differential trends.
whether a credit expansion
the increase in profits (and sales)
strategies pursued by the large firms.
Performing Assets (NPAs).
is
Since
it
is
may
associated with an increase
otherwise reflect more risky
we use data on Non
In order to answer this question,
takes at least a year for a loan that has gone bad to be
an NPA, we treat the years 1998 and 1999 as the "pre" period, the year
2000 and 2001 as the period following the expansion, and 2002 as the period following the
contraction. In 1998
to small firms,
and 1999, 1%
became NPA. 5.5%
of the loans to
of the
medium and
medium and
34
large firms,
large firms,
and
and
5%
4%
of the loans
of the small firms
that were not
NPAs
NPAs
1999 became
in
for the loans to big firms, the difference
is
2000 or 2001. While the growth
in
3%
very small. Conversely,
medium
seem
firms that were not
NPAs by
to have led an unusually large
2001 became
number
NPAs
in 2002.
NPA
faster
is
of the loans to the largest
and 2%
firms (with investment in plant and machinery above 10 million)
in
and
of those to small
Additional credit does not
of firms to default.
Instrumental Variables Estimates: the impact of bank credit on sales and
5.4
profit
The
discussion in section 4 suggests that equation (3) to (5) and (6) to (8) respectively form the
first
stage and the reduced form of an instrumental strategy of estimating the impact of
loan on sales (or any other outcome variable
coefficient of the interaction
sector expansion on the
volume of loans
to
in the
equation
in
good
sample of firms with an increment, the
(3) is
the causal impact of the priority
firms, multiplied
by the fraction of good firms
The
coefficient of the interaction
BIG * POST
sales, multiplied
by the fraction of good firms
in the sample.
Assuming that the only impact
impact of credit (which we
will verify later), this indicates
in this
on
BIG * POST
y):
bank
sample.
of the reform on sales
that, controlling for
bank loans on the
due to
is
BIG
its
POST, BIG * POST
and
sales of the firms that are
the reduced from and and
first
known
stage equation
is
is
is
the causal effect of the reform
thus a valid instrument for the impact of
to be good (the ratio of the coefficients in
equal to the impact of credit on sales for good
firms).
In this last sub-section,
of
bank loans on
sales, costs
Column
squares estimate,
the instrument
coefficient
is
we
(the instrument
BIG2
*
columns
(1)
and
in the
POST2
is just,
already saw. Finally, column
and
profit.
For comparison, we also present the weighted least
IV estimate
a
(3)
in
of the effect of
sample with a change
with a standard error of 0.37.
the previous one, which
BIG * POST
and
(1) presents the
BIG* POST
0.75,
present (in table 9) instrumental variable estimates of the effect
Column
in
bank loans on
loan in the 1997-2000 period.
(2) uses
using
The
the "contraction" experiment
the 1999-2002 period). This estimate (0.73)
way
sales,
is
very close to
to restate the result of the overidentification test that
uses the entire period and three instruments
BIG2 * POST2). The
coefficient
(2) (0.76).
35
is,
we
(MED * POST,
once again, very close to what
it
was
in
If
firms do not increase market credit in proportion to the increase in
bank
estimates of 9 (the elasticity of sales with respect to bank credit) provide a lower
(the elasticity of sales with respect to overall credit) for these firms:
To
credit, these
bound on
9
see why, rewrite equation
(10), to obtain:
log
=
Rit
log
An +
log k bu
- 01og Jg.
(17)
Differencing over time:
log
flit
We
- logfl^! = logA it do not observe log
^-
log
The term
0[\og -Mf
- log
fli t
?
b
.
log Ait-i
~l
,
]
log
-
+
-
e[\ogk bit
log.fc
W t-i] - <?[log|| -
^ffy and therefore estimate an equation
log
which
=
A^-j
is
~
6[log k bit
log
+v
is
g^ffr]-
(18)
of the form:
(19)
it
omitted when estimating equation
The one exception
be positively affected by the reform.
kbit--]]
lo
(19),
should typically
the case where the firm
is
credit
constrained and access to market capital increases so fast as a function of access to bank capital
that total capital stock goes up faster than bank capital-which seems rather implausible. This
suggests that 6 will be a lower
The
bound
for 9.
with respect to bank loans,
elasticity of sales
8,
gives
some additional information about
the plausibility of the firms having substituted the bank loans for the market credit. To see this,
note that
_
Afl_fc
A
A7c
If
there
is
substitution of
=
Afl
Ak +
bank
b
kb
A/c m
credit for
+ km _
-
A
1
1
market
krn
+ Akm /Ak b
credit, A/c
m
is
kb
negative while
Akb
i
s
positive
In this (possibly hypothetical) scenario
9>e[l + ^}.
(20)
h
9
is
equal to about 0.75. Using the Prowess data
short run
bank debt over
total liability in Indian
implying a ratio of ip of about
1.
set,
Topalova (2004) estimates that the ratio of
manufacturing
is
about 0.5 from 1996 onwards,
Therefore in this scenario our estimate would suggest an 8
36
of above 1.5 in the neighborhood of the current capital stock, implying that the firm
>
credit constrained (a 8
1
suggests local increasing returns and, hence the firm must
borrow more) and therefore unwilling
Column
to cut
The
coefficient
from the result
column
0.35). In
The estimate
OLS
is
(5),
a
Panel
B
whole sample (the
we go back
is
to
all
that
it
though
less precise,
coefficient
on
the firms, and
sales
smaller than the
IV
is
we include
to
credit.
gives an exclusive right to produce
higher (0.93) but very imprecise.
little
estimate, which
is
somewhat smaller and
is
in the
want
the sample to firms that do not produce SSI products, since, as
(4) restricts
mention before, one advantage of SSI status
goods.
back on any form of
must be
it is
we
some
not statistically different
with a standard error of
0.50,
no change
firms with
Finally the last
in limit.
column present the
estimate, consistent with our model predicts.
The
present the estimate of the effort of bank loans on costs.
here axe, again, very close to each other, and just a
little
estimates
we obtain
smaller than the effect of the loans on
sales.
We
can use these estimate to get a sense of the average increase
rupee in loan. The average loan (averaging across years and firms)
days of sales)
.
in profit
is
Rs.
caused by every
8,680,000 (about 45
Therefore, using the coefficients in column (3), an increase of Rs. 100,000 in the
loan corresponds to an increase in Rs. (610,000 in sales, and Rs. 537,000 increase in costs. This
implies an Rs. 73,000 increase in profit for the average firm, after repaying interest.
In panel C,
we
present, for the sake of comparison, the direct
log(profit), despite the fact that these regression suffer
omission of the firms with negative profits.
The
IV estimate
of loans on
from the sample selection induced by the
estimates vary between 1.79 and 2.00. Taking
1.79 as the estimate of the effect of the log increase in loan on log increase in profit, an increase
100,000 in lending causes a
of Rs.
2%
increase in profit.
At the mean
profit
(which
is
Rs.
3,670,000), this would correspond to an increase in profit of Rs. 72,000 after repaying interest,
which
is
Can
very similar to what we found using cost and sales as the dependent variables.
a net return of
72%
or
73% be
explained by the subsidy implicit in the program?
After correcting for default risk and administrative costs and using a cost of capital of 12%,
we estimate the
which
is
cost of lending to the priority sector for Indian sector public banks to be 22%,
higher than the
16%
the firms actually pay. But a subsidy of
on a net return of 72%. Indeed the excess return would
37
still
6% makes
be sizeable
if
very
we were
little
dent
to decide
that the public sector banks are pricing their credit wrong and that they should charge
month (42% per
3%
per
year) which seems to be the going rate on trade credit for these relatively large
firms.
The
private return on an extra rupee of loans to firms in this sample
+ 16%). The
social return
is
about 83% (the social cost of capital
Both these returns are correctly read
cost).
therefore the right
number
how
credit should
is
what the
90% (72%
what happens
The
success.
if
the
private return
impact of a shock to the
to use for calculating the short-run
bank's balance sheet, while the social return
close to
higher than the private
as answers to the question:
bank lends an extra rupee to the firms that have had a history of
is
6%
is
is
social planner should use in deciding
be allocated. The magnitude of these numbers
tells
us that
known highly
profitable opportunities remain uno.xploite.d in the Indian economy.
However
since this
over-state the impact
the long run not
all
is
the return to captial in case of success, both of these numbers probably
we would expect
"good" firms
the bank will have to lend to
will
new
it is
the bank's lending was permanently raised since in
remain that way (markets change, managers
firms of
Moreover these returns cannot be read
model
if
unknown
is
and
quality in order to identify future winners.
as the average return
bank loan
clear that the average
retire)
much more
on a dollar of bank lending. In our
likely to
go to a "bad" firm than the
marginal loan that results from the expansion of a directed lending program. Consistent with
this
we
find that in the
data the
OLS
increases are both smaller than the
positive, but only 0.28.
insignificant.
The OLS
Although due
and the IV estimates, the
Our estimate
IV.
estimates of the effect of loan increases on sales or profit
IV
estimates.
For
sales,
the
OLS
estimate
estimates of the effect of loans on profit
to large standard errors,
we cannot
is
significantly
is
even smaller and
reject the equality of the
difference clearly goes in the right direction that
OLS
is
OLS
smaller than
therefore should be seen as the causal estimate of the marginal value of an
additional dollar lent, as long as there are enough credit constrained good firms in the bank's
portfolio.
It.
is,
however, not obvious that
factor (capital, in the
to
form
we should
of machines, say).
pay wages (because paying labor
is
think of these returns as the return to any specific
The most common use
of this
money
is
probably
the one thing one cannot use trade credit for), but
possible that getting access to this extra
money
38
will also
it is
impact the borrower's ability to get
more trade
credit
and hence expand the
of the extra rupee will then be
The observed
firm's use of other inputs as well.
some combination
of the effect of extra labor
and the
effect
effect of
the extra units of the other factors.
Conclusion: Policy Issues
6
The evidence presented
in this
paper suggests that
severely credit constrained during 1998-2002.
tale
about what happens when banks, as
As shown
in the
way
it
in section 2,
it is
allocates credit,
this
relatively large firms in India
in India, are largely publicly
and one could imagine
is
were
might be tempting to see this as a cautionary
It
true that the particular public sector
optimality in the allocation. Indeed this
However
many
owned.
bank we study
is
quite rigid
this leading to substantial deviations
what the model
from
in section 3 predicts.
During the period of our study, and especially
cannot be the whole story:
during the period covered by the later experiment (2000-2002), private banks were quite active
in the
— almost a quarter of the total credit to firms
Indian banking sector
from private banks, including a number
in
the economy
came
of multinational banks. If the entire underlending
was
a product of the irrationality of the public bank, any of these private banks could have stepped
in
- the firms
in
our sample are but a drop
in
the ocean compared to the total lending of any
one of the private or multinational banks operating
urban
areas, certainly
credit
and perhaps
had the option
did.
The
in India.
Our
firms, all based in relatively
of approaching a non-public sector
interesting question
is
why
bank
for additional
nevertheless, they did not invest
much
more, especially given the enormous profitability of additional investment.
One
yet have
possible answer
is
enough resources
that the local private banks were
to lend to these firms
policy of public ownership, albeit indirectly.
—
It is,
this puts the
A
more
in their infancy
blame on the pre-liberalization
this period. It also
plausible version of this
seems
over time and most non-public sector banks do not yet have
it.
may be much more
39
less plausible
argument points to
the fact that lending to the small-scale sector requires specific expertise that
existing public sector banks, once privatized,
and did not
however, belied by the fact that these banks
were investing heavily in government bonds throughout
in the case of the multinational banks.
still
is
only acquired
This would suggest that the
effective
than the present crop
of private banks, precisely because they have the requisite experience. 30
There are however good reasons not to be quite so optimistic. Stein (2001) has argued that
the inability to lend effectively to small borrowers
have a natural tendency to be
large, in order to
is
in the very
nature of being a bank: banks
spread out idiosyncratic
On
risk.
being larger necessarily increases the distance between the owners and the
who
the other hand,
many
loan officers
deal with small borrowers. Since loan officers need to take decisions about relatively large
amounts
of
money
that do not belong to them, and defaults are costly for the bank, 31
important that the loan
officers
have the right incentives.
distance between the owner and the loan officer grows.
by
restricting the
on
easily
domain
very
it is
This obviously gets harder as the
Banks deal with
this
of the loan officer's authority: in particular, by
problem
making
in part
rules,
based
measured characteristics of the borrower, about how much they can borrow and by
penalizing the loan officer for defaults. As in our model, this discourages the loan officer from
lending, unless the firm
is
a very sure bet. This obviously limits the discretion the loan officer
enjoys and makes his lending less effective, but
it
covers the bank. 3 "
An
obvious social cost
is
33
that small firms have a hard time borrowing.
This
is
not to say that some characteristics of the India economy such as the cost of enforcing
a loan contract are not important in understanding
30
why no one wants
to lend to these firms.
This also suggests that while the public sector banks are probably over-staffed, the extent of over-staffing
be over-estimated
if
we
directly
may
compare private and public banks, because private and public sector currently
play very different roles. Banerjee et
al.
(2004) contains an overall assessment of the performance of the Indian
public sector.
31
Defaults are also quite
common,
at least in India.
they are supposed to be (and actually are, at least
in
Working
capital loans in India are not nearly as safe as
the US). This
is
because the borrower can easily
sell off
the inventories that are supposed to be securing the loan before he defaults, and hide the proceeds. While this
potentially actionable, inefficiency of the legal system discourages going after borrowers.
commercial banks have a
lot of
It is
is
in the
is
Maximum
is
is
that most
form of working capital loans.
therefore not surprising that the existing rules in India leave
In particular, projections of future profits (an area
decision.
result
non-performing assets (estimated to be as much as 10% of total assets) despite
the fact that most of their lending
32
The
permissible bank finance
is
little
room
where judgement tends
to
for
independent decision-making.
be important) have no place
in
the
calculated as a percentage of projected sales. In turn, the guideline
that projected sales should not exceed current sales plus 15%.
33
Berger
et
al.
(2001)
show that
in the
US, the increasing concentration
significantly reduced, access to credit for small firms.
40
in
banking after deregulation, has
But there are many other countries with
same kinds
It is
similar dysfunctionalities
where we would expect the
of results to apply.
therefore important not to lose track of policy changes that would
make
it
easier to
lend to small firms in developing countries by focussing entirely on the privatization issue.
In particular,
(some states
it
may
help to set up special courts for the speedy disposition of default cases
in India are
experimenting with this model, and Visaria (2006) finds that this debt
recovery tribunals do reduce default and interest rates charged on loans).
to
improve the system of recording
the same asset
may be
titles to,
and
It is also
important
liens on, property, to avoid the possibility
that
used to secure multiple loans. Severe punishments for those involved in
asset-stripping and other types of fraud will also
41
make
lenders
more forthcoming.
References
Angrist, Joshua (1995) "Conditioning on the Probability of Selection to Control Selection Bias",
NBER
Technical Working Paper, No. 181.
Banerjee, Abhijit and Esther Dufio (2000) "Efficiency of Lending Operations and the Impact
of Priority Sector Regulations",
MIMEO, MIT.
Banerjee. Abhijit and Esther Dufio (2005) "Growth Theory through the Lens of Development",
in
Economics Handbook of Economic Growth,
Banerjee, Abhijit,
Shawn Cole and Esther Dufio
Forum, Volume
Banerjee, Abhijit,
1,
Vol.
1,
Part A, 473-552.
(2004) "Banking Reform in India" India Policy
Brookings Institution.
Shawn Cole and Esther Dufio
(2008)
"Are the monitors over-monitored:
Evidence from Corruption, Vigilance, and Lending in Indian Banks"
Banerjee, Abhijit and
Andrew Newman
MIMEO, MIT.
(1993) "Occupational Choice and the Process of De-
velopment", Journal of Political Economy 101(2): 274-298.
Berger, Allen,
Nathan
Miller, Mitchell Peterson,
Raghuram Rajan and Jeremy
"Does Function Follow Organizational Form:
Large and Small Banks",
MIMEO,
Stein (2001)
Evidence from the Lending Practices of
Harvard University.
Bernanke, Benjamin and Mark Gertler (1989) "Agency Costs, Net Worth, and Business Fluctuations",
Galor,
American Economic Review, 79(1):14-31.
Oded and Joseph
Zeira (1993) "Income Distribution and Macroeconomics", Review of
Economics Studies 60:35-52.
Heckman, James (1979) "Sample Selection Bias
as a Specification Error", Econometrica 42:670-
693.
Heckman, James and Richard Robb (1986) "Alternative Methods
Selection Bias in Evaluating the Impact of Treatment on
from
Self- Selected
Samples, ed. H. Wained,
42
New
for Solving the
Outcomes"
,
in
Problem
of
Drawing Inferences
York: Springer Verlag.
Hardman Moore
Kiyotaki, Nobuhiro and John
Economy
(1997)
"Credit Cycles" Journal of Political
105(2): 211-48.
McKenzie, David and Christopher Woodruff (2004)
u
Do Entry
Costs Provide an Empirical Ba-
Poverty Traps? Evidence from Mexican Microenterprises" Working Paper, Stanford
sis for
University.
Olley, Steven
and Ariel Pakes (1996) "The Dynamics of Productivity
Equipment Industry", Econometrica 64
(6):
in the
Telecommunications
1263-1297.
Peek, Joe and Eric Rosengren (2000) "Collateral Damage: Effects of the Japanese
on Real Activity
90
(1):
Rosenzweig,
in the
United States," with Joe Peek.
Bank
Crisis
The American Economic Review
30-45.
Mark and Kenneth Wolpin
(1993) "Credit Market Constraints,
Consumption
Smoothing, and the Accumulation of Durable Production Assets in Low: income Countries:
Stein,
Investment
Bullocks in India", Journal of Political Economy, 101 (21): 223-244.
in
Jeremy (2001) "Information Production and Capital Allocation: Decentralized Versus
Hierarchical Firms"
,
forthcoming
in
Journal of Finance.
Topalova, Petia (2004) "Overview of the Indian Corporate Sector:
1989-2002"
IMF
working
paper WP04-64.
Visaria, Sujata (2006) "Legal
Debt Recovery Tribunals
Reform and Loan Repayment: The Microeconomic Impact of
in India"
IED Working Paper
43
157,
Boston University
Figure
1
F\k)
rm
i
"60
kb\
k
^0 *62
Fi gure 2
F\k)
rm
r
kbo
kbx
k
k
k
x
Table
1:
Descriptive statistics
Change! t)-(l-l)
levels
entire
sample
change
in loans
entire
sample
change
not missing
(1)
in
loans
not missing
(2)
(3)
(4)
LOANS AND INTEREST RATES
PANEL
A:
working
capital
87.66
96.29
10.29
7.46
loan (this bank)
(237.04)
(258.2)
(59.92)
(55.32)
1226
928
966
928
log( working capital
loan) (this bank)
working
loans
(all
3.44
0.07
0.07
(1.5)
(.24)
(.24)
1208
928
928
928
87
97
10
7
(246)
(273)
(69)
(67)
1102
807
842
807
capital
banks)
log(working capital loans)
(all
3.39
(1.47)
banks)
other bank loans
positive
other bank loans
(level)
interest rate
log(interest rate)
3.36
3.41
0.06
0.06
(1.48)
(1.51)
(.26)
(.26)
1085
807
807
807
0.0120
0.004
0.0000
-0.007
(.11)
(.06)
(.14)
(.1)
1748
807
1748
807
1.65
2.23
0.00
-0.62
(25.86)
(36.54)
(22.54)
(30.9)
1748
807
1748
807
15.75
15.58
-0.32
-0.32
(1.63)
(1.59)
(.94)
(.94)
1142
896
876
856
2.75
2.74
-0.02
-0.02
(.18)
(.19)
(.16)
(.17)
1142
896
878
858
Notes:
1
-Columns
1
and 2 present the mean level of each variable, with the standard error
and the number of observations on the third
2-Columns 3 and 4 present the mean change in each
and the number of observations on the third line.
3.
in
parentheses
line.
All Values are expressed in current Rs.10,000.
variable, with the standard error in parentheses
Table
1
(continued) Descriptive statistics
Change t-t-1
levels
entire
sample
change
in
loans
entire
sample
i
;hang e
not missing
(I)'
PANEL
B:
(2)
in
loans
not missing
(4)
(3)
CREDIT UTILIZATION AND FIRM PERFORMANCE
account reaches the
0.72
0.69
-0.01
-0.01
limit
(.45)
(.46)
(.44)
(.44)
522
380
247
233
log(tumover/limit)
2.15
2.15
0.09
0.11
(.95)
(.96)
(.72)
(.71)
384
308
170
167
709.33
820.70
108.64
86.66
(2487.24)
(2714.88)
(653.62)
(598.64)
1259
746
1041
739
Sales
log(sales)
log(sales/Ioan ratio)
5.49
5.64
0.17
0.09
(1.44)
(1.46)
(.53)
(.45)
1248
740
1029
732
2.19
2.18
-0.01
0.02
(.89)
(.87)
(.53)
(.49)
1004
740
751
732
36.51
42.49
6.08
4.04
(214.11)
(237.16)
(61.32)
(58.3)
1259
747
1043
741
net profit
log( costs)
*
5.45
5.61
5.45
5.61
(1.45)
(1.45)
(1.45)
(1.45)
1245
739
1245
739
Notes:
1
-Columns
1
and 2 present the mean
and the number
level
of each variable, with the standard error
2-Columns 3 and 4 present the mean change in each
and the number of observations on the third line.
3. All
in
parentheses
of observations on the third line.
Values are expressed
in
current Rs. 10,000.
variable, with the standard error in parentheses
—
C3
o
o
CS
"3
o d
co
m
r^ *o
d d
o
o
CS
vo
r~
ci
o
o
o
«
r-~
<^-
^
CS 00
d d d d
r-;
o
o
r^
cn
t^ t^ t^ 't
<6
<zi d>
d>
rr)
\o
c-«
c>
d
in r^
\o rci ci
m
>o
d
r-~
— O° o
Co g 5"
O °
O M
— S
o
w
g;
X;
- o N
«
in t^ in cc
c> ci
c->
r~-
<z>
Co 3 c^
S
2
o m
o >n
o 2
o
ci
m o
O
vq
t--
;—• oo --o
n ^ T
OS
w m
O o o m
o
r-;
.
c6 ci ci
ci
op
'5
•
CJ>
r^ hi m
^D O 00
d d d d d d d
1- r\0 \o >o
a
«
o
*d-
oo
>c;
O
,j,
^ rr
O o
aON
^
3
sO
,
00
o m
vo vo
CS
I-;
o
>c
in
00
00
CO
^-^
o
d o
d d d d d d
CM
VO
—
-r
d O
00
ca
c
'
T3
CD
>
o
£^
B.
J2
si
H
r3
"O
E
u
5
j5
T3
<D
u
.p
m
au
-o
1)
«
3
-^
E
1!
C3
o
O
"I)
~
§
_o
£
u
.—
^a
O
CS
d
^
c
^
CS
T3
-o
JJ
— —
T3-C
c/i
:r
r3
u
I
ft.
O a
*J
u
s
lM
1L
60
00
U
&
C
vi
c
c P
-a
ZbM
p
O
d
E
V V
<D
—
s
"J
>-,
c
^!
o
u
G
U o
oo
3 s
CJ
g
r3
03
s y > X
0J
o
o
CS
u
-J
ea
CS
^O
o
ca
rt
<1>
«3
«
5
uJ
2 -
U
E
'J
ri
.
—
—
—
j
—
«
-
CD
J3
*^
—
O O O o
m ^
—
o ^
o o s
r--
•sj-
00 CN
>o r^
t--
•^
O
o
*
o
9>
O
^
u
KJ
tu
O c
cn cO oo
\d o 2!
d d d o
vq
o o q
o fj
o
c_
j=
5
r-~
CN
u s
E O
u
O
S-,
o
u
O
c
o
— in
^ °.
O O O s
o
cd
5
^
*
o m- -~r
in in o
2
—
'-'
in
"i
— o
\d r^
e>
q
~
— rn °^
2T
vO O
O O O in
o
"*
*0
dodo
do©
J;
.
— rg N
o o o o
1
.
.
II
c—
~
CN
<D
"
C
c
|
r-
*_i
3
c
"J
<u
S
m
— O rn
— ^
o o o o
m in in ^ri S
O
O S
O ™
o
On *0
^j
d d s
O O O o
=1
O
—
en
ID
o
>
>>
VO in
o
d d d
U
ex
e
vC
o-
i.
£
s
3
s
X
<
o in
2 -*
c^ ro o
Q d d d
—»
UJ
^
CN
J d
o
CN
so
o
c
3^
vC
--"•
t~-
,
^*
<N r^
vo
.c
in
O o
d d d
o -r
d d d o
o
d d d q
^
3~
d d o O
^-
CD
=
<s>
-J
S3
'Z-
c
'J
CD
a>
13
en
"J
^r
—
u<
C
L-i
O
—
VO VO
^-^
2C
d
3
u c
o en
u
u
PL,
c/3
i-t
CD
>-
o
Z 3
c
-a
CD
"S
_C
en
o
^5
~~
11)
w*
_o ta
CD
^:
t-
<
;
CD
Z^
ed
o.
^r:
>
CJ
ed
CD
c_
*—
^-v
in oc
in in —
IN
m
D o'
m
(N rn O CN
c
O rn fS o
d d d o a d d d
d d d o a d d d o
<
UJ
UJ
<
5
<
<
u
u
2
z
3
z
u
u
DJ
UJ
5
s
a
rn in 00
UJ •* CN •—
'-^
r^i
1
.
CTv
r-j
C^'
-1-
*
C/3
c/o
r3
_u
en
J3
3
"-"'
UJ
c-;
_
<
ON
CN
d d
C/3
d d
<
H
cd
o
c
a>
<
o
>-•
z
i5
L^
u
o
o
-i
0.
a)
C/J
c
o
>-
Z
•
i
(3
Q
IU
o
B
^
en
(D
D.
d>
-C
CD
CD
CD
^
cej
UJ
Z -
C
O
Ji
?5
>
o
> z 5
3
en
1
•u
C/3
UJ
CD
CD
3
"ob
Z 3
E
3
o
ll
CM
o
c
>-
5
—C
en
^
o
C3
en
>
I
c
CD
c
"J
—
-3
J=
z
u
m
en
c^
T3
on
u o
CD
c
u
ex
o
<
>
<
a
c —
o
c
jr
d d
> d d
™
CJ
i
m
m
CD
Oen
O
en
•*
m *
i^-
P
su
o
~:
—
>
m
<
C/3
—
c/3
Ce>
t>-
UJ
H
<
3
g -a
c
'w
o -C
en
u
u
r3
0X1
3
—
c
a
U
D
ID
^5
o
<
a
—
<
a^
C/3
cn
o
,
1/3
u
<
2
O
P
<
CN
in \D
»n
""''
£
en
"^2
—
en
*-j
~
03
C
CD
Cm
IT
§
u
o
-c
S
=
__
e
t/3
-
OD
a.
3
o
'-1
E
c
-
en
'J
n.
o
"J
CD
*3i
if
u
3
O
,^-
c
c
O (N rn !•
~ -. == S
O O O ^v
c
o
>
C3
5 U s
c
o U
a
JS
u
S
a
7^
o
O
en
C-
cn
en
c3
;>
.0
"O
T3
VO
^
o
a
g
3
O
u
CN
en
m
> ^
c
CD
:*
Table
4:
Average change
in limit
Years
1996-1997
1998-1999
2000-2002
Firm's category
A. Average change in limit
0.110
0.075
0.070
(.021)
(.013)
(.014)
small
medium
0.040
0.093
0.011
(.032)
(.030)
(.025)
0.093
0.147
0.000
(.064)
(.040)
(.031)
biggest
B.
Proportion of cases where lim lit was not
0.701
0.724
(.043)
(.031)
(.027)
medium
0.667
0.608
0.798
(.088)
(.055)
(.040)
0.625
0.692
0.769
(.183)
(.075)
(.053)
biggest
C. Average change
ch, anged
0.701
small
in limit,
conditional on change
0.366
0.252
0.253
(.045)
(.035)
(.045)
small
medium
0.119
0.237
0.053
(.093)
(.068)
(.124)
0.248
0.479
-0.002
(.137)
(.062)
(.138)
biggest
Notes:
1-The
at
first
row of each panel presentsthe average of log( working
date t)-log( working capital limit granted
2-Standard errors
in
capital limit granted
date t-1).
parentheses below the average.
3-Number of observations
in the third
row of each
4-"Small firms" are firms with investment
"Medium
at
firms" are firms with investment
in
in
panel.
plant and machinery
below Rs
plant and machinery above
and below Rs 10 million. "Biggest firms" are firms with investment
above Rs 10 million.
in
Rs
6.5 million
.
6.5 million.
plant and machinery
— o rr o
o r™ o
C- ^
° w° w°
rs
1
P
«_.
"jT
C
lU
»
0>
A
2
C_
-3
•
"i
O
cc
u-im^av or^ ^.ON«
'-c
t
;
on
s
S3
>
«
—
— z~- o —
o -r*
t ^ ^ ^ r
° S° S° w
c- (N
2"
-
•
-
CO
JS
U
o
|
o
^
o 3
.0^.0
«
E_
(0
oD
p
V
^
3
s
0)
Oft
o
ao
55
p
r3
(4
*U
P
5
3.
r3
o
>
0£
'
f*.
fN
^
yi
^r
^
— —
^^
£T
^
*~"
*a
C^
O
Vi
S
o
-S
t<
«-
—
i
§
Jtf
s
P
a.
t/3
^
_
00
^1 00 L,
—° W°
r»
u
c
OJ
EL
^ s ^ s ^ ^
s s
©
© S
o ?° s_>
os
[^
-—
fvi
*3"
-~ >—
Qn
O ^ (N N vO
| g S g 2 g S
°°
"*
r--
*
r*.
«*
—
r-
§
£
—
O
(4
-J
'
1
in vi
so \Q
-a "§-•
-a
«
IS
T S
^ —T
CT «- rC; —
o
o H W1
o - - 2
^
°^
VO
r-
-
-
o
"— (N
"o
cr^r
s s
i
s
8S
1
la
>>
C
;
s
:
.32
;
s *
i
;
3
_n
c3
fc
3D JD
a
^3
P
3 a
fl
«
§
S a
2
5
bo
p
o
o
o
rs^ss
— O
_ >
rt
Q.
r^
i^i
2.
„
o ~ °
r~-
^
c=
31
_
§
—
5.2
c
-
—
C
v ^
B
w > «
4J
4J
^
'
P
P
2 S S «
!J
°° ^T ^
^
- C
>°
«
°
c s
° s
« ° s
— ^
o 2
3o o° 3
«
^TT
—
o^ogo^
O ^ O
O ^ o
r~-
^.
on
J2
rz:
>
3 W
p
*5
S
,_>
—
O
(fl
5_,
S
S
-1
z a
PUP
P
^ —
P
_
"
U.
H
("N
<^i
s
P
g
iz
w
u -
n
>> o. >,
p
r aS
—
^3
e3
cr cr
""
i 5
5 -td
p
_
?
3
_^
tfcH
'vi
09
%.
•
_5 3 S9
3 u
'
—
15 J
-p
"
M
...
go-.
o
§
>-
1
p
t^
-a
'
_g
e
'
P g
5 5
1
P I
5
,o
<
u
4)
5
-
— z: —
© o°
u
c
<
•
^
1
g O
cs
5
*=>
"^
—
O
<=>
00
o
(N
cj
p
•a
^.
u
H
'-,
a
3
a
"2 O 2 O g
© £» —
^ o -^ ° O ° _>
^j-
u
eo
_^
w
—
— UJ
©
& _
'
<=>
II
so
(N
<
4)
5
S
5d
*S
Of]
_c
1
jr*
r-~
_
O
or
^j
rt
u
o
-
2
o
(N
-*"
fN
(N
£T __
00
•
ftT
— m c* —
^ £
o
© m
so
f^ IT
~; fN
<^
_>
*^"
.v,
.„j,
c~ ^
2 r« ©
© " ©
so
w.
r--
cr
00
00
^
^
cr
u-!
^
00 ^J
Zo 3
'->
^
2
'
r„ ^r
—
v-,
1
1
G
u
^;
>
u
E
p
|
OO
60
©
_0
_».
°
J.
" — ©-r
_ so
oc
—
_;
O
r^
qv
rv]
o.
°°
p
o - o £ o S
-
^j,
fN
o =:
2 E
o
_x
in
m
n
2 — 2
q
o S
o
~s
^
'
t/3
o
5
a
.3
"5S
u o
S
2 o
(N
C
*7
—
« o
-j
-
ei
"*
b
•
e
e
1)
N „
^ 00 _, rn o °° (N
q s °
o q o _. ° o ° o
sb sd
sf ^ o£ (2
Os
-
-
>,
e c
.S
U
o
=
E
<u
OS
ca
3
3
^
00
» O
a
_
W
o 5 o 5 « t«
s q § q s ^
00
00
38
4>
§ 8
«
4)
u
TZ
00
i>
(A
4)
u
oo
_c
DO
s
"^
4J
3 -
C§ £
rsi
2T
"^
rn
r©
©
=>
)£>
m
Tt
© o ° — °
2-
*""""
4T
H
ki
rt
w
V)
-3
u
a
V-
V
<U
0)
'j
M
53
4>
.a
j=
o
S
—w
n. -r
—
c S
-s
5
u
5
E
53
tn
o
o
c
>
2 °
$
-a
3
"
u
r-)
s.
U
jO
«
^
Mh-a
oj
P
S
±1
„
_3
,s
— „
o
-
O °
1
3a^
u
S
rS
u o
4)
«)
S
a o
e
a
o
^ ^ o
©
—
£
S
o
°
if
5 c
2 «
00 ~
g
« _ ^
S
in
a>
J
S
tl
— ~u
^
ii
<3
<S
cl)
y
=
t
?
g
"do .1
£
V.
u B a c E
5
^ t^ -1 9 n E
^ £ ^
2 — (N m E—
S
<
8.
CO
J5
^J-
aII3
M «
H 10
Table
7:
Credit constraints: Effect of the reform on sales, sales to loan ratios, and profits
(OLS
regressions)
Dependent variables
log(sales/Ioans),
Log( costs),
Log( profit),
log(sales/loans),_.
-log(cost),.,
-log(profit),..
OLS
OLS
OLS
(3)
(4)
(5)
Log(sales),-log(sales),.|
Complete Sample
Sample without
substitution
OLS
OLS
(1)
A. t= 1997-2000
1.
Sample with Changes
in limit
post*big
2.
Sample without Change
0.168
-0.065
0.187
0.538
(.106)
(.118)
(.104)
(.097)
(.281)
152
136
152
151
141
in limit
post*big
3.
0.194
0.007
0.022
0.007
0.005
0.280
(.074)
(.081)
(.074)
(.064)
(.473)
301
285
301
301
250
Whole sample
post*big
0.071
0.071
-0.016
0.068
0.316
(.068)
(.069)
(.075)
(.055)
(.368)
453
421
453
452
391
-0.403
-0.387
0.143
-0.374
-0.923
(.207)
(.196)
(.206)
(.279)
(.639)
168
150
169
168
151
-0.092
-0.045
-0.092
-0.048
0.170
(.108)
(.128)
(.108)
(.086)
(.56)
401
380
401
399
321
-0.143
-0.113
-0.016
-0.101
-0.253
(.111)
(.134)
(.162)
(.094)
(.496)
569
530
570
567
472
B.t=l 999-2002
1
Sample with Changes
in limit
post2*biggest
2.
Sample without Change
post2*biggest
3.
in limit
'
Whole sample
post2*biggest
Notes:
1.
Each panel
is
a separate regression.
Each column presents
a regression
of column heading on the variables
listed in
each panel
2. The dummy "post" is equal to 1 in years 1999 and 2000, zero otherwise.
The dummy "post2" is equal to 1 in years 2001-2002 zero otherwise.
3. The dummy "big" is equal to 1 for firms with investment in plant and machinery larger than Rs 6.5 million, zero otherwise.
The dummy "biggest" is equal to 1 for firms with investment in plant and machinery larger than Rs 10 million.
4-Standard errors (corrected for clustering
5-In addition
at
the sector level) are in parentheses
from coefficient displayed, the regressions
in
panels
5-In addition from coefficient displayed, the regressions in panels
A1-A3
BI-B3
below the
include a
include a
coefficient.
dummy for post and a dummy for big.
dummy for post2 and a dummy for biggest.
^f
—
rn
")
O
(J
r.
O
o
oO "Bo
o
-
o
00
©
°
8
«
S3
—8
-
1
i
O° O°
*_>
°
CT
-_>
in
vd
"S
=
S3
•-C
3
»->
00 -s
a
t2
c
M
£
vi
ed
ff-M
—
c
i
°
!
U
00
c§
I
c
c
'
3
©'
:
,_>
©
'"'
©
o
©
&
o
£
-S
a
~
'J
c3
wGO
O
_3
5
_c
>
a.
E
00
o
(N
1)
-J
~
=
N
S
S
<2
o-
^
u .a
S3
E
CO
_o
CJ
00
1 —i
|
S3
S3
b
-C
1
s
c_
B
©
3^ O
°
3
(N
©' OO
°
©
3
cd
t/j
*o
(N
o ON
o
o
E
53
"H
<u
on
Ti 7!
S 2
Cj
|SR
o
? r
—
2
*j o
(N
u
op
cu
s
15
C
tg
s
""*
—
o
— o
a
u
1 e
EX
B
<_>
5
s.
-
£
a o
J2
*-*
r "a
- ~
O
_ - a
2
tr
rs
eS
nf
O
—
cN
a]
u
a
O
!~
o
u
°
T?
(N
o
.2 -S
<^
5
o
,2
5
jr
OJ
3
o
a)
=i
aj
ra
>
*-•
-3
>
—
.3 ~°
—
IN t—
'
£ ^
D- 22
vi .S
'
—
+-(
a
_£i
S
E
3
vi
IN
C
vj
O
d.
*
en
3
Xr,
3
60
vj
<u
c
*
C
c
60
3
— "
5
'j
E
a
oo
"3,
E
a
U
J
O
o
3
a
03
-
-5 c2
rN
cs
t-
•a
U
u
u
^J
3
c
,
-3
3
"
o £
^ K —"
do
u
&l
3
c
oo
oo
C8
o->
1
T?
E
&
"
V)
3
(N
°-
©
NO
§5
2
VI
>
60
op
LS
t-
C
—
a
3
O
C3
a.
—
e
<u
-a
c
60
00
-a
60 tS
u
^1
c
c
>
U
o
o
_
3
t-~
fS
"1
—
FC
(N
_;
O
-* Zi
(N
w
<=?
<N
=
-T <N
&
£
E
(J
£
2
ON
o
"O
3
>> -3
-a
•-
fi
r~
-~-
*n
oo
v©
2
in
a3
"
"oo
C3
-
<N
C3
d
c
60
u
oT 'c
c
^
=
"S
S
oo
IS
2
'S.
01
si
_©
,J
e
J^
.Si
oo g>
K
-^
C
60
>
.=
—
B3
°
U
c
vi
•
oo
T
60
e
.2
u
u
o
a
>
?
>
00
2
J4
IB
Oil
o
60
g
si
o
U
U
"
rt
u
^-
c
>=
_2
i
ir
g
D.
-a
>
60
3
—
3
3
o
u
o
O
c
a
E
E
E
3
-a
00
!>?
cs
g
U
3
£O0 ^
E
3
S
E
*
_3
_3
_3
e
U
—c —c
O
VJ
VI
60
^iw
00
^*J
w
u
w
m
M-
QJ
CO
P
3
o
o
c
—
a
a
.5
vj
5
2
O O O
CJ
3
=
a
u
oo ti
oo K
-^
y"
-^ u
1>
VJ
00
v-^J
u
V3
vi
U 3
VI
^2
>
5
~
^5 3oo_C
pj
a.
5
,o
*p. _Q.
T^ ts
_3
a.
O
VI
o
c
u
i
,
DO
oo
u
C-
E
.ti
~
v5
o O
-j
ri
O u
60
D.
-
3
u
i_
—
73
.
=
U o
"-J
'g
=
^ « «
VI
5
.3:
*^
3
t-i
c
a g
u —
a
—
'^
•J
-, .J
o
a.
J2-
<N
a_
o
b
=
I
C
f^)
>n