7/27/2019 The Real Effects of Bank Capital Requirements
1/26
1
The Real Effects of Bank Capital Requirements
Matthieu BRUN 1, Henri FRAISSE2, David THESMAR3
July 4, 2013
Abstract: We measure the impact of bank capital requirements on corporateborrowing and expansion. We use loan-level data and take advantage of the transitionfrom Basel 1 to Basel 2. While under Basel 1 the capital charge was the same for allfirms, under Basel 2, it depends in a predictable way on both the bank's model and thefirm's risk. We exploit this two-way variation to empirically estimate the semi-elasticityof bank lending to capital requirement. This rich identification allows us to control forfirm-level credit demand shocks and bank-level credit supply shocks. We find verylarge effects of capital requirements on bank lending: 1 percentage point increase incapital requirement leads to a reduction in lending by approximately 10%. At the firmlevel, borrowing is reduced, as well as total assets (mostly working capital); we providesome evidence of impact on employment and investment. Overall, however, becauseBasel 2 reduced the capital requirement for the average firm, our results suggest thatthe transition to Basel 2 supported firm activity during the crisis period.
The opinions expressed in this document do not necessarily reflect the views of the Autorit de
Contrle Prudentiel.
1 ISODEV. Email: [email protected] . This paper was written when Mathieu Brun wasworking at the Banque de France.2 Autorit de Contrle Prudentiel. Email: [email protected] HEC Paris and CEPR. Email: [email protected]
7/27/2019 The Real Effects of Bank Capital Requirements
2/26
7/27/2019 The Real Effects of Bank Capital Requirements
3/26
3
sample, conditional on their risk category. Looking at these data we observe that banks
assign different default probabilities to firms in the same risk category, which confirms
that internal models vary widely across banks.
Identification rests on two strengths of our research design. First, capital requirements
for lending to a given firm are typically different for two different banks. This is
because Basel 2 allows banks to use their own internal model to evaluate the
probability of default of a firm. Hence, for given firm risk, we can compare the lending
of banks according to the penalties imposed by their internal models, for the same
firm observables. Second, our sample has a sizable number of firms that borrow from
several banks. For these firms, we can control for unobservable firm-level shifts in
credit demand, as the banking literature has done in other contexts (for example,
Kwhaja-Mian, 2005; Iyer,Peydro, Schoar, 2012; Jimenez, Ongena, Peydro, Saurina,
2012). Consistently with this literature, we find that making this adjustment does not
affect our results very much, suggesting that heterogeneity in bank loan portfolios is
not driving the results.
What we find.
Our analysis has caveats that we do our best to address. First, the goal capital
requirements reduce systemic risk. Systemic risk is beyond the scope of this paper,
which focuses only on lending effects. But the welfare benefits of having a safer
banking system may outweigh the costs of depressed lending that we measure.
Second, we measure essentially short-term effects. Because of data availability
constraints, our sample stops in the last quarter of 2011, or three full years after the
transition to Basel II is implemented. Our year-by-year results do not seem to suggest
that the negative impact on lending fades out. But it may be disappear after a longer
period. Last, the transition to Basel 2 that we study occurred in the beginning of 2008,
exactly when the financial crisis started to hit the global financial system. We do our
best to control for this: Our methodology rests on the comparison between firms that
received different treatment from the new regulation. In some specifications, we even
compare lending to the same firm by banks on which the regulation had a different
7/27/2019 The Real Effects of Bank Capital Requirements
4/26
4
impact. Our focus on France also helps a little, as French banks were not extremely
hurt by the crisis. But it could still be argued that our estimates are polluted by the
financial turmoil.
The literature: on regulation: Calomiris paper; macro papers. on Funding shocks:
Kwhaja-Mian, 2005 Iyer,Peydro, Schoar, 2012; Jimenez, Ongena, Peydro, Saurina,
2012). Like these paper, we use firm-time fixed effects to control for the fact that
banks may have different loan portfolios.
2. The Empirical Strategy
a. Calculating Capital Requirements
Since the Basel I accords in 1988, bank regulators around the world have asked banks
to use risk weights to assess the equity needs. The logic is the following. When a bank
makes an investment of 100 in a project (for instance a mortgage, or a corporate
bond), the regulator requires that the bank holds r x100 euros of equity capital, where r
is the equity requirement. Under Basel 1, the equity requirement for corporate lending
was 8%. This meant that if a bank had, say, 10bn of equity, it could not make more
than 125bn of corporate loans, unless, of course, it was ready to issue new stocks.
The Basel 2 accords made capital requirements much more heterogeneous across
banks and firms. These new agreements were published in 2004. The idea was to
reoptimize regulation so as to take into account the rising complexity of banking
activities (for instance, securitization), and also to avoid regulatory arbitrage by
adapting capital requirements to the riskiness of investment. Indeed, a flaw the Basel 1
was that, since capital requirements were the same for risky or safe corporations,
banks would have strong incentives to only lend to risky banks, so as to maximize
expected profits under regulatory constraints.
7/27/2019 The Real Effects of Bank Capital Requirements
5/26
5
Our study exploits one key feature of Basel 2: The capital requirement for a loan
depends in a predictable way on both the firm's risk and bank's model. When moving
to Basel 2, banks may be authorized to follow the "internal risk based" (IRB) approach,
whereby they can use their own estimate of the probability of default of the firm to
calculate the capital requirement. Remaining banks are required to take the
"standardized approach", where the capital requirement is a known function of the
firm's official rating.
Let us start with IRB banks. For a loan to firm f by bank b , the capital requirement is
then given by the following formula:
r bf = FIRB (PDbf , LGDbf , M bf ) (1)
where PD bf is firm f 's regulatory probability of default, as estimated by bank b , or by
the regulator (more on this below). LGD bf is the loss given default M bf is the maturity of
the loan. As banks do in most circumstances, we take LGD bf =.45 and M bf =2.5.F IRB (.) is
a known function.
The regulatory probability of default PD bf is not to be confused with the one directlyobtained from an internal model based on a scoring approach. The scoring model
allows putting each firm into a risk class and each firm of this class bears the same PD bf
which is then used for the computation of the capital requirement. In order to limit the
procyclicality induced by the Basel II framework, the French regulatory framework
imposes that the PD were computed through the cycle meaning the PD of a risk class
does not vary over time. However, the rating of a firm might vary over the cycle.
Therefore, firms might be associated to different class of risk overt time and mightdisplay a time varying PD.
We do observe the regulatory default probability PD bf for a large sample of bank-firm
linkages in 2008. We have the rating issued by the Bank of France, available for all
firms and all years in our sample. Using these observations, we calculate the average
effective regulatory PD per rating level. Therefore, we are able to map the Bank de
7/27/2019 The Real Effects of Bank Capital Requirements
6/26
6
France rating system into each internal rating system and to associate to each firm a
regulatory default probability each year. More formally, we have the relationship:
PDbf = 'b. f (2)
where ft is a set of dummies for each value of the firm's official credit rating (which is
discrete). b captures the fact that the model to evaluate the PD may differ from bank
to bank.
Banks that cannot use their own model of PD have to follow the standardized
approach. Under this approach, the capital requirement is direct and known function
of the official bank of France credit rating of the firm. For bank b and firm f , this leads
to:
r bf = Fstandardized ( f ) (3)
where F standardized (.) is a known function. Under the standard approach, all firms that
have the same official rating have the same capital requirement.
Overall, the capital requirement for each bank-firm linkage can be computed anddepends both on the firm's rating and the bank's model. This allows identifying the
impact of the capital charge separately from the firm's credit rating. Note that if all
banks followed the standardized approach, this would not be possible. Fortunately for
us, a large fraction of our sample includes banks that were authorized to adopt the IRB
approach. For these banks, the capital charge is not perfectly correlated with the firm
rating: It also depends on the bank's internal model, summarized by the vector b,
which we can identify using a specific dataset on PDs effectively used by IRB banks.
The identification also exploits the variation of capital requirement across banks
portfolio under the Basel II design. The capital requirement is also a function of the
bank's exposure (overall lending) to the firm and the turnover of the firm. Exposure
matters in that it allows loans to firms below 50m of turnover and below 1m of
exposure to be reclassified as "retail", and therefore benefit from lower capital
requirements. In sum, the Basel II leads to four regimes of capital requirement: the
7/27/2019 The Real Effects of Bank Capital Requirements
7/26
7
retail treatment versus the corporate treatment and the advanced approach
versus the standard approach. An internal model is validated by the regulator at the
level of the portfolio (retail versus corporate) of the judicial entity. Therefore within a
bank holding company, a firm exposure can be treated under the standard approach
by one entity and under the IRB approach by another.
We exploit these contrasting regimes in order to assess the impact of higher capital
requirement.
b. The Model
Our objective is to test whether equity requirements affect bank lending policy. We
focus on the change in lending between before and after the implementation of the
Basel 2 regulation, and ask whether changes in equity requirements are correlated
with changes in lending. We discuss here our exact estimation strategy, and discuss
the sources of identification in the next Section.
Using bank-firm loan-level data, we estimate the following equation:
Lbf = f + b + .rbf + Xf + bf (3)
where Lbf is the growth of lending by bank b to firm f . rbf is the change in regulatory
capital requirement between before and after the reform. Since it is constant, equal to
8% under Basel 1, we take without loss of generality the risk weight in the "post"
(Basel 2) period as our measure of change. f is a firm fixed effect, designed to capture
the fact that some firm (for instance, risky firms) may simultaneously experience an
increase in risk weight, and a reduction of lending, without any causal link between the
two. As we discuss below, the model with fixed effects is identified only for firms
borrowing from several banks. b is a bank-specific fixed effect, which controls for
bank-wide lending decisions; It is identified because the same bank lends to several
firms. X f are sets of firm specific controls designed to capture observable differences in
7/27/2019 The Real Effects of Bank Capital Requirements
8/26
8
lending policy that are unrelated to regulation (firm profitability, size, credit rating). bf
are error terms, which we cluster at the bank b level.
The null hypothesis is that =0. In this case, bank lending is not constrained byregulation, for two possible reasons. First, the bank is well capitalized. In this case, the
requirement does not matter, as the bank has enough equity to make the loan, as long
as it is NPV>0. Since the NPV of the loan is firm specific (it does not depend on the
bank), it should be absorbed by the firm fixed effect. Second, the bank has little equity,
but can issue more without friction. Under these conditions, the MM theorem is valid:
controlling for loan NPV (the firm fixed effect), the capital requirement does not
matter, since the cost of capital is unaffected by the fraction of equity used to finance
the bank.
One last issue with equation (3) is that we need to estimate r bf at the bank-firm level
under the Basel 2 regime. To do this, we use equation (1), which connects r bf with the
probability of default PD bf , and equation (2), which connects PD bf with the Bank of
France credit ratings that we observe in our data. At this stage we need the
coefficients b, which establish the correspondence between the ratings and bank-specific default probabilities. To calibrate b, we use a dataset that reports the PDs
effectively used by banks. Since this dataset is not exhaustive but covers a small subset
of firms, we calculate the average PD per category of credit rating, and use these
numbers to impute the PD for all bank-firm match.
c. Identification
Our estimation technique allows for a sharp identification of the effects of regulation
on bank lending.
First, equation (3) is a first difference equation. Bank and firm fixed propensities to
borrow/lend are already absorbed in the implicit level equation. In its simplest form,
(3) is identified of off the correlation between loan growth and capital requirement
7/27/2019 The Real Effects of Bank Capital Requirements
9/26
9
increase. At this stage, however, the problem remains that risky firms (those with high
capital requirement in the post Basel 2 period) may be borrowing less after Basel 2 was
implemented. The problem is especially acute since the post Basel 2 period in our
sample coincides with the beginning of the financial crisis, which may have had a
stronger effect on the demand and supply of credit to risky firms.
Second, the coefficient on the capital charge, , is identified, even when one controls
for credit rating of the firm. For now, assume no firm fixed effect f in equation (3).
This is because, for a firm with a given credit rating f , different banks face different
capital requirements, depending on the model they use. There are two useful sources
of variation here. First, the difference between the standard and the IRB approaches:
As long as b differs between the two approaches, a bank under IRB faces a different
capital requirement than a bank under the standard approach, even if the firm's credit
rating is the same. Second, within the IRB approach, bank b and b' also face different
capital charges for given rating as long as the models they are using are different
enough: b has to be different from b' . This source of identification is the core of our
identification strategy.
Formally, our identification technique can be understood by combining equations (1)-
(3). To clarify exposition, let us assume that the Basel formula F(.) is a linear function of
the default probability only (in our empirical approach we take the other inputs as
given and constant, like most banks do in practice): we posit F(x)=ax. Let us then take
two firms: f , which borrows from bank b , and f' , which borrows from bank b' . Assume
both firms have the same credit rating . Then, equation (3) for both firms writes:
Lbf = a.b + . + bf (4)
Lb'f' = a.b' + . + b'f' (5)
In these two equations, we abstract from bank fixed effects (because they do not
affect the reasoning here) and from firm fixed effects (to which we return in the next
paragraph). Substracting the (5) from (4), we obtain that:
Lbf - Lb'f' = a.(bb' ) + (bf - b'f' )
7/27/2019 The Real Effects of Bank Capital Requirements
10/26
10
which shows that the coefficient of interest is identified as soon as banks b and b'
have different models: b b' . The key to our identification of separately from the
direct effect of ratings is that banks have different models. We establish this by (1)
controlling for ratings and (2) showing that b b' holds in the data. This discussion
remains the same if b follows the standardized approach: Then, it is key for
identification that b' follows IRB.
Third, we use firms borrowing from multiple banks to control for unobservable firm-
level credit demand shocks - and not just observed rating. This is represented in
equation (3) by the fixed effect f . For each firm who borrows from different banks,
identification relies on the comparison between bank lending depending on the bank's
capital charge. We can do this because the capital charge depends on both the firm
and the bank (see discussion above). This approach is well known in the banking
literature (see for instance: Khwaja and Mian, 2008, Iyer et al, 2011, Jimenez et al,
2012).
3. Data
a. Bank-Level Data
Our sample is made of the six largest banking groups operating in France. They
represent around 80% of the total asset of the French banking system. The sample
related to the estimation of equation (3) has 256 banks owned by one of these six
groups.
b. Loan-level data
7/27/2019 The Real Effects of Bank Capital Requirements
11/26
11
We start from a large dataset of bank-firm linkages available at the Bank of France
(Centrale des Risques ). A linkage between bank b and firm f exists and is reported in
the data as soon as bank b has an exposure of more than 25,000 euro to firm f . The
data is thus exhaustive above a certain exposure threshold. The dataset is quarterly,
and provides us with the total exposure, as well as the identifiers of the bank and of
the firm. Exposure is lumpy on the way up, but continuous on the way down. Exposure
increases abruptly when the bank extends a new loan to the firm, and then goes down
progressively as the firm repays the loan. We define "lending" as the quarterly change
in exposure. We then set lending to zero if exposure decreases - in the data such
occurrences typically corresponds to principal repayment.
We then create a quarterly panel firm-bank pairs. We include all firm-bank pairs that
appear at least once in the large linkage dataset between 2006Q1 and 2012Q4. We
then assume that the pair exists throughout the period that we study (2006Q1 -
2012Q4). If, at a given date, the pair is absent from the linkage data, we posit that
exposure is equal to zero.
We then merge with firm-level accounting and rating information, also available fromthe Bank of France ( Centrale des Bilans ). Such information is updated annually.
Accounting information follows the tax forms that firms have to fill in and provides us
with extremely detailed data on the balance sheet and the income statement. Credit
ratings are awarded by a special unit at the Bank of France, which is in charge of
maintaining the public credit registry. The credit registry covers a vast number of firms.
Because we want to focus on the effect of the Basel 2 implementation, we collapse the
dataset into two sub-periods: 2006Q1-2007Q4 (before the reform), and 2008Q1-
2012Q4 (after the reform). For each bank-firm pair, we take the sum of lending in each
sub-period. We then take the log of the sup-period lending plus one. Thus, if banks
does not lend during one period, the log is zero.
We restrict the sample of firms to the firms that provide balance sheet and income
statement over the entire period. We take averages of all firm observables. We end up
7/27/2019 The Real Effects of Bank Capital Requirements
12/26
12
with 421,901 bank-firm pairs. We provide summary statistics for the resulting dataset
in Table 1A.
We finally merge the data with bank-level information on the transition to Basel 2. Thisdata provides us with the transition date (2008Q2 for all banks), and the approach
approved by the French supervisor: IRB vs standardized. We focus on banks that are
continuously present throughout the sample. On the sample on which regression (3) is
estimated, 256 banks are present. All of them do not have exposures within each of
the four Basel II regimes. As for the retail portfolio, 142 out 239 transit to the IRB
approach. As for the corporate portfolio, 165 out 237 adopt the standardized
approach. In terms of bank-firm relationship, out of 429,901 pairs that we follow
between before and after the Basel 2 implementation, 291,162 transit to IRB, while
the remaining 130,739 follow to the standardized approach.
Firms that borrow from several banks make up a large fraction of our sample. 91% of
the bank-firm pairs in our sample (387,961 out of 429,901) correspond to the firms
borrowing from at least two different banks. Over the period, 73% of the firms have
had an exposure to several judicial entities. These observations are especially useful asthey allow us to include a firm specific fixed effect in equation (3), thereby controlling
for differences, across banks, in the unobservable riskiness of firms. In Table 1B, panel
A, we report summary statistics for two subpopulations of our loan-level sample: loans
in which the firm borrows from several banks, and loans in which the firm borrows
from one bank only. We see that the statistics are similar in the two subsamples, even
though the differences in means are statistically significant. We will control for this
observables in our regressions. But more convincingly, we will show that regressions
results, before controlling for firm fixed effects, are identical in the two subsamples.
Panel B, Table 1B investigates the difference between banks that adopt the standard
and banks that adopt the IRB approach. The difference between these two categories
might contribute to identification: two firms in the same risk class, but borrowing from
two banks following different approaches, will face different equity requirements. Our
empirical strategy will test if this difference in equity requirements is systematically
7/27/2019 The Real Effects of Bank Capital Requirements
13/26
13
related to differential lending growth. In Panel B, we report the difference in loan-level
data for these two subsamples. We find that the summary statistics are similar, even
though, once again, averages are significantly different. It may thus be argued that the
two subsamples are different. We will deal with this concern by showing that
regression results are similar whether we include, or not, banks transiting to the
standard approach. Thus, the standard/IRB difference do not importantly contribute to
our identification.
c. Common Counterparties
The first dataset ("common counterparties") allows us to measure how bank models
for default probabilities differ across banks, for given firm characteristics. This dataset
is a survey run in 2007 by the French prudential supervisor, on a sample of firms that
borrow from at least two different French banks. For each of these common
counterparties, banks report the default probability that they are using to calculate the
capital charge of the loan. Thus, this dataset allows to directly compare the PD models
that banks are using for the same firms. The dataset contains a lot of additional
information, such as the LGD used by the bank (almost always equal to 45%), and firm-
level characteristics, such as the credit rating given by the Bank of France.
We use these data to calculate a bank-specific correspondence between credit rating
and default probabilities. We first collapse the 11 bank of France rating categories into
6 categories: This is because, to attribute equity requirements under the standardizedapproach, the regulator only uses these 6 categories. We then calculate, for each bank
and each of the 6 risk categories, the average regulatory PD of the firms in the sample
(not the effective probability of default, but the one that banks assign according to the
CC survey). This is how we obtain the b parameter in equation (2). We then use this
bank-specific parameter to impute, in our entire sample (described in Section above),
the default probabilities for each bank-firm pair. We then transform these default
probabilities into capital charges using the Basel formula (1). We provide summary
7/27/2019 The Real Effects of Bank Capital Requirements
14/26
14
statistics on the resulting capital charges in Table 1A. For the average firm, the
transition to Basel 2 led to a decrease in capital charge by approximately 1.9
percentage points (from 8% to 6.1 on average). This is consistent with the general idea
that Basel 2 aimed at fostering SME financing. It should be noted however that the
cross sectional s.d. of the change in equity requirements is about 2%, so that a
significant fraction of the firms in our sample experienced increases capital charges.
We exploit this variation in our empirical Section.
Key to our identification strategy is that banks have different models for default
probabilities, i.e. that the b are different across banks. We check this in Table 2,
where, for each class of credit rating separately, we regress the imputed PD on a full
set of bank dummies. We then test whether we can reject the null hypothesis that all
banks use the same model, i.e. that the b are all equal, conditional on the rating. The
first three columns of Table 2 show that, conditional on firm ratings, there is
substantial heterogeneity in probabilities of default. Columns 4-8 show that the null
hypothesis that all bank fixed effects are equal is strongly rejected by the data for all
ratings groups. Moreover, bank fixed effects explain a considerable amount of the
variance of PDs. The unexplained variation comes from the fact that we average
observations across quarters of a sub period. Since banks may change credit rating, we
take the integer part of the average rating, as an approximation for the average ration
across periods. This creates composition effects that add noise to our construct.
d. Firm-level data
Last, we collapse the dataset constructed in Section 3.a. into firm-level data. The
objective here is to evaluate the impact of the Basel 2 reform on firm-level decisions
and outcome such as capital structure and investment. In particular, an important
question is whether firms that can borrow more from one bank because of the
transition to Basel 2 end up borrowing more in total, or whether they reduce they
borrowing from other banks. We start from the loan-level data described in Section
7/27/2019 The Real Effects of Bank Capital Requirements
15/26
15
3.a. For all observations corresponding to the same firm, we then take the average of
all variables. Firm-level accounting variables are not affected by this procedure, as they
are by definition the same across all firm-bank linkages corresponding to the same
firm.
Summary statistics are reported in Table 3. They show that few composition effects
arise, when compared with Table 1A. The average change in capital charge is the same:
a decrease by about 2.2 percentage points after Basel 2 was implemented. Sales grow
by 10% on average. Lending which accordingly to our definition cumulates the new
loans on the two sub periods- increases by 147%. Note that this large increase is due to
our definition of change in lending which implies for a firm borrowing X in the post
period and nothing in the pre period a change of log(1+X) and which considers a bank-
firm relationship as soon as an exposure appears on the total period. On the sample of
firms that borrow in both periods, the lending increases by 67%. In the following
section, we will test the robustness of our results considering alternative definitions of
what a bank-firm pair is. All these numbers do not differ widely from those obtained in
Table 1A: This is because multi-bank borrowing firms are not very different from
single-bank firms, as already observed in Table 1B. We have added a few more firm-
level accounting variables in this Table. Exposure is the total exposure to all banks,
averaged over quarters and differentiated. Consistently with more borrowing (lending
increases), total bank exposure increases by 58%.
4. Results
a. The Effect of Basel 2 on Bank Lending
Table 4 reports the regression results of various specifications of equation (3). We start
with Panel A. In column 1, there are no firm-level controls, and no fixed effects. In this
very raw specification, we find a negative impact on lending of higher capital
requirement but this impact is not significant at conventional levels. Column 2 includes
7/27/2019 The Real Effects of Bank Capital Requirements
16/26
16
firm-level controls: the pre and post-period average bank of France rating discretized
into 6 categories, as well as the pre-Basel 2 level of log sales and ROA. These variables
are designed to control for firm size and financial health before the reform. Credit
ratings are also included because they serve to calculate the equity requirements, and
we want to make sure that the effect of requirements is not identified on firm credit
risk, but on its interaction with the model of the bank. Including these controls actually
almost doubles the size of our estimate which now turns out to be strongly significant.
In column 3, we further include a bank fixed effects, to alleviate the concern that
conservative banks (whose equity requirement is high under their own internal model)
may be linked to firms that do not borrow very much for other reasons. For instance,
mutual banks may be lending to SMEs that do not seek to grow very much, especially
in times of crises, and may also end up with higher capital requirements, either
because of conservative risk-management practices, or because they were only
allowed to adopt the standard approach (which has higher capital requirements). As
can be seen from Column 3, controlling for bank fixed effects does change our
estimate in a significant manner. We find that a two percentage point decrease in
capital charge (say, from 8 to 6%) leads to an increase by 8.74*0.02=17.5 percentage
point increase in bank lending. This corresponds to about 6% of the sample s.d. of bank
lending growth. Our model thus has a priori reasonable explanatory power for this
kind of microeconomic study. In aggregate, the model estimates that the Basel 2
reform has boosted corporate lending by 0.022*8.74=19 percentage points,
where2.2% is the average reduction in equity requirement coming from Basel 2. This is
14% of the increase in lending observed between the two sub-periods.
Columns 4-5 seek to control for credit demand shocks by including a firm-level fixed
effect in equation (3). The firm fixed effect also allows to control for unobserved
heterogeneity in firm-level credit risk that may affect a bank's decision to lend. Note
that, for this heterogeneity to matter, it would have to explain lending beyond what
the bank of France rating can do. Firm-level risk unobserved heterogeneity is identified
of off firms that borrow from several different banks (they constitute almost 73% of
7/27/2019 The Real Effects of Bank Capital Requirements
17/26
7/27/2019 The Real Effects of Bank Capital Requirements
18/26
18
b. Additional Tests
We then consider alternative measures of lending and definition of the bank-firmrelationship. In Table 2, recall that once an exposure of the bank b to the firm f was
observed over the whole period, the pair bank-firm was considered. In the panel A of
Table 5, we restrict ourselves to bank-firm pairs when the firm has taken out at least
one new loan from the bank within the entire period. As before, we take the same 5
econometric specifications. The impact of the capital requirement is then magnified
leading to a significant impact of higher capital requirement on lending growth in all
specifications. To measure loan growth at the intensive margin, we take lending
growth of firm-bank linkages for which we observe at least one loan in both periods.
We report the result of this regression in Table 5, Panel B. We find that the intensive
margin effect that we estimate is smaller than the overall effect estimated in Table 4.
The coefficient hovers around 3. This means that 2 percentage point decrease in the
equity requirement (-0.019 is the sample mean) would lead to a 6 points decrease in
lending at the intensive margin, much smaller than the 19 points obtained globally (in
Table 4, Panel A).
We then investigate the transition dynamics, and show that the effect of the reform
increases over the years until 2011 (we do not have firm controls for the year 2012).
We have to compare the lending granted in 2011, 2010 and 2009 taken separately to
the one made in the pre reform period. In order to compare the parameters associated
to the change in capital requirement to the same parameter estimated with the whole
post period we consider another definition of lending. Indeed, if we stick to our former
definition, the parameter will be less comparable since the likelihood to get a new loan
is of course greater when considering more years in the post period and that the
amount of new loans accrue over time. Alternatively, we now define lending as the log
of the average exposure over one year of the post period when considering the post
7/27/2019 The Real Effects of Bank Capital Requirements
19/26
19
period year separately. When we collapse all the post periods, we define lending as the
log of the average exposure over the entire post period.
For instance, in Table 5, Panel A, we use as dependent variable the change betweenthe log of average post reform exposure and the log of average exposure in the pre
period. In Table 5, Panel B, we use as dependent variable the change between the log
of average exposure in 2009 and the log of average exposure in the pre period. Panel C
focuses on 2010 and Panel D on 2011. In each panel, we take the whole sample and
report results for all the 5 specifications used in Table 4. Looking at all specifications,
we conclude that the effect became more pronounced over the years. This is
consistent with the financial crisis, a time where it was notoriously difficult for banks to
raise external equity.
c. The Effect of Basel 2 on Corporate Outcomes
We now test for the impact of the regulation change on firm-level outcomes. To do
this, we run the following regression at the firm level:
Yf = .rf + Xf + f (6)
where Yf is the change in firm policy (debt, investment etc.), and X f is a set of firm
level controls (the same as in Tables 4-6). rf is the average of all changes in equity
requirement, across all banks to which the firm f is linked. The f are assumed to be
heteroskedastic. We report the results in Table 7.
Note that, since we focus on firm-level outcome (not loan level ones), this
methodology does not permit to control for bank nor firm fixed effects. This is a
limitation of our study. What gives us confidence in our estimates, however, is that the
estimates, with or without bank and firm FE, tend to be stable (see Table 4, Panel A).
This suggests that the potential biases that may arise from bank or firm-level
7/27/2019 The Real Effects of Bank Capital Requirements
20/26
20
heterogeneity (bank-level credit supply shocks, firm-level credit demand shocks etc)
are not too large.
We first show that the reform affected overall firm borrowing. We have shown inTables 4-6 that the change in capital requirement affects bank lending, but it may be
the case that firms can substitute shrinking lending by one bank with borrowing from
another bank. It may alternatively be the case that the ability to borrow from one
particular bank (because of decreasing capital charge) would cause the firm to borrow
less from other banks. In both cases, even though the reform is shown to have an
impact at the loan level, it may have no impact at the firm level. We check this in
columns 1-3, Table 7, using our two different measures of firm borrowing.
In column 1, we look at average firm borrowing across all linkages. This measure is the
closest to the regressions already run in Tables 4-5, except that we aggregate at the
firm level. For each firm and each date, we calculate the sum of loans made across all
bank-firm linkages. We then sum this measure across dates of each sub period. Finally,
we take the log and differentiate. This measure is exactly identical to the loan growth
measure used in Table 4-5 for firms who are linked to only one bank (about half of oursample). As shown in column 1, we obtain an estimate of the impact of the reform that
is slighty lower than the baseline estimates (-3.17 in table 7 to be compared with -5.8
in table 4, suggesting that there is little substitution within our sample given the
precision of the estimates).
In column 2, we use as dependent variable average total bank exposure. In the initial
bank-firm linkage data, we aggregate total exposure across all banks at the firm-
quarter level. We then take average total exposure across quarters of each subperiod,
take the log and differentiate. We find in column 2 that the effect estimated in column
1 is slightly lower than the one displayed in table 6, column 2 at the bank-loan level.
The fact that the estimate has similar size for stocks and flows suggests that the
maturity of the loans tends to be short, so that a reduction in lending would quickly
lead to a reduction in the stock of debt.
7/27/2019 The Real Effects of Bank Capital Requirements
21/26
21
In columns 4-7, we show that a decrease in average equity requirement leads to the
expansion of a firm's assets, working capital and employment. In column 4, we find
that a 2 ppt increase in capital requirement leads to an increase in total assets, as well
as current assets, by about 1 ppt. The effect is a bit smaller, although with a similar
order of magnitude, for employment: For the average firm in our sample, the
transition to Basel 2 helped create 0.31 x 0.022 = 0.68 additional ppt of employment.
This is to be compared with a 3.2% average employment growth in our sample.
5. Conclusion
Although present at the heart of the policy debate on the banking regulation, the
impact of higher capital requirement on lending and real outcome has been rarely
examined in the academic literature at the micro level.
This paper evaluates such impact. The implementation of the Basel II regulatory
framework in 2008 in France led banks to substantially modify the regulatory capital
associated to each credit line of their corporate portfolio. Exploiting the French
national credit register and the internal bank rating models, we are able to match eachloan at the firm-bank level with a risk weighted asset charge applied by the bank. We
identify the impact of the capital requirement by contrasting the lending of several
banks charging different regulatory capital to the same firm. These charges differ
across banks because banks are either under different regulatory regimes (e.g.
advanced, foundation or standard approach of Basel II) or simply because they differ in
their internal models. Therefore our approach controls for demand effect.
Depending on our definition of lending at the extensive margin, on new loans or
exposure- we find an increasing impact ranging from 3 to 8 percent of an increase of
capital requirement by one percent. This impact translates to real corporate outcomes
such as employment and investment.
7/27/2019 The Real Effects of Bank Capital Requirements
22/26
22
References
Admati, Anat, DeMarzo, Peter, Hellwig, Martin and Pfleiderer, Paul, 2010, "Fallacies,
Irrelevant Facts and Myths in the Discussion of Capital Regulation: Why Bank Equity is
Not Expensive", Stanford GSB WP 2065
Aiyar, Shekhar, Calomiris, Charles and Tomasz Wiedalek, 2012, "Does Macro-Pru Leak?
Evidence from a UK Policy Experiment", NBER WP 17822
Hanson, Sam, Kashyap, Anil and Stein, Jeremy, 2011, "A Macroprudential Approach to
Financial Regulation", Journal of Economic Perspectives , Vol 25(1), pp 328
Iyer, Raj, Lopes, Samuel, Peydro, Jose-Luis and Antoinette Schoar, 2011, "The Interbank
Liquidity Crunch and the Firm Credit Crunch: Evidence from the 2007-2009 Crisis",
mimeo UPF and MIT
Jimnez Gabriel, Steven Ongena, Jos Luis Peydr and Jess Saurina, 2012, "Credit
supply and monetary policy: Identifying the bank balance-sheet channel with loan
applications, American Economic Review, 102 (5), 2301-2326.
Kashyap, Anil and Stein, Jeremy, 2004, "Cyclical Implications of the Basel II capital
standardss", Economic Perspectives , Federal Reserve Bank of Chicago, vol. 28
Kwhaja, Asim, and Mian, Atif, 2008, "Tracing the Impact of Bank Liquidity Shocks",
American Economic Review , vol 98(4)
7/27/2019 The Real Effects of Bank Capital Requirements
23/26
23
Figures and Tables
7/27/2019 The Real Effects of Bank Capital Requirements
24/26
24
7/27/2019 The Real Effects of Bank Capital Requirements
25/26
25
Table 1A: the different Basel II regimes
Note: The unit of observation is a bank-firm linkage. Basel II considers four different regimes ofcapital requirement: the retail exposure (less than one million of exposure and 50 millions ofturnover), the corporate exposure (more than one million of exposure or more than 50millions of turnover). The retail portfolio and the corporate portfolios of a given entity canbe treated either under the standard approach or the advanced approach. This table displaysthe number of banks, the number of exposures, the total amount of exposures in each of thefour regulatory regimes. We start from the universe of bank-firm linkages with bank exposureabove 25,000. These data are quarterly. For a pair to be followed throughout 2006:1 - 2012:4,we require it to be in the bank-firm linkage dataset at least once in the period (2006:1 -2012:4). We then match these data with annual firm-level accounting information. In our finaldataset, we have 421,901 bank-firm linkages. The average exposure is computed over the2008:1-2012:4 period. The regulatory regime of a given entity for a given regulatory portfoliohas been provided by the French supervisory authority.
7/27/2019 The Real Effects of Bank Capital Requirements
26/26
Table 1B: Summary Statistics of the Loan-Level Dataset
Note: The unit of observation is a bank-firm linkage. Change in log lending is defined as thedifference between the log of the sum of new loans granted to the firm f by the bank b in thepost period and the log of the sum of new loans granted to the firm f by the bank b in the preperiod pre period. For the other variables, "Change in X" the difference between the averageof X in the post reform period (2008:1 - 2011:1) and the average of X in the pre period (2006:1- 2007:4). "Pre reform X" is the average of X in the pre period (2006:1 - 2007:4). To constructthese averages, we start from the universe of bank-firm linkages with bank exposure above25,000. These data are quarterly. For a pair to be followed throughout 2006:1 - 2012:4, werequire it to be in the bank-firm linkage dataset at least once in the period (2006:1 - 2012:4).We then match these data with annual firm-level accounting information. Finally, we use 2007
bank-risk class averages from the "common counterparties" and the Banque de France ratingof the firm to impute regulatory PDs and capital charges for each bank-firm pair and eachquarter. We then take the average of all variables over the two sub-periods (pre and postreform), and then differentiate. In our final dataset, we have 421,901 bank-firm linkages.