Docstoc

determinants of bank lending performance in germany

Document Sample
determinants of bank lending performance in germany Powered By Docstoc
					Schmalenbach Business Review x Vol. 52 x October 2000 x pp. 344 – 362


Ralf Ewert/Gerald Schenk/Andrea Szczesny*

DETERMINANTS                 OF    BANK LENDING PERFORMANCE                                  IN
GERMANY**
Evidence from Credit File Data



ABSTRACT

We empirically identify factors that can explain the financial performance of bank lending
activities. We also analyze the individual bank’s evaluation of a loan’s risk. We use our
results to test theoretical hypotheses on the impact of certain parameters on credit terms
and distress probabilities. We find ratings act as an important factor in the bank’s lending
policy. Ratings reflecting higher risks lead to higher interest rate premia. The findings on
collateralization are less clear and do not fully support any of hypotheses that are
advanced to describe the role of collateral and covenants in credit contracts.



1 INTRODUCTION

Commercial banks often voice the opinion that the “traditional” credit business has
come under considerable competitive pressure. As a result, credit margins tend to
decrease and the profitability of lending becomes problematic. In banks with
several lines of business (e.g., German universal banks offering virtually all types
of financial services), granting credit to a firm seems to be viewed more as a “door
opener” for other transactions (e.g., investment banking activities), which banks
believe may be more advantageous. Thus, arguments of cross-selling are often
considered major factors in support of the lending business.

The new developments in risk management and bank supervision has exerted
additional pressure on the traditional lending business. The Basle commisssion for
bank supervision is in process of developing new standards on capital adequacy
of financial institutions, which will result in reform this particular area. Among
experts there is already widespread agreement on the fact that internal credit
ratings will be the future criterion for the equity requirements. Since we include

 * Prof. Dr. Ralf Ewert, Dipl.-Wirtsch.-Inf. Andrea Szczesny, Lehrstuhl für Controlling & Auditing,
   Johann Wolfgang Goethe-Universität Frankfurt, Mertonstr. 17, 60054 Frankfurt am Main, Prof. Dr.
   Gerald Schenk, Berufsakademie Heidenheim, Wilhelmstraße 10, 89518 Heidenheim.
** This paper is part of a research project on Credit Management in Germany initiated by the the Uni-
   versity of Frankfurt’s Center for Financial Studies (CFS). We would like to thank all the banks that
   participated in this project and their representatives for their willingness to cooperate on this
   research. We also thank Jörg Beißel, Kai Forst, Jan-Pieter Krahnen, Ulrich Rendtel, Bernd Rudolph,
   Wolfgang Schwerdt, and Martin Weber for comments on earlier drafts of this paper and helpful
   suggestions. We gratefully acknowledge very helpful comments from one anonymous referee of
   the Schmalenbach Business Review. We are responsible for all remaining errors.

344                                                                                    sbr 52 (4/2000)
                                                                  Bank Lending Performance


the banks’ internal credit ratings in our data set, we are able to address the role of
ratings in the lending business.

In our paper, we study the empirical determinants of bank lending performance.
We are particularly interested in answering the following questions: Are there any
empirical similarities between characteristic features of credit relationships and
measures for the success of the lending business? Do the internal credit ratings
predict liquidity crunches or defaults and thus help to improve the profitability by
pricing the underlying risk? Answering these questions will yield insights on the
empirical validity of theories that try to explain reality and can also help practi-
tioners seeking for alternative ways in which to improve the profitability of their
credit transactions.

To address these issues, we apply two measures as proxies for the “success” of
lending contracts:

• The first measure is based on the idea that a loan contract’s profit results from
  an interest rate premium over a rate at which the funds could have alternatively
  been invested. Hence, it is a rough measure of the surplus the bank could
  expect if there are no problems during the life of the credit contract.

• The second measure captures potential problems by looking at the frequency at
  which disturbances (e.g., delay of principal and/or interest payments by the
  borrower, technical default by the borrower, or even insolvency) occurred. Such
  disturbances imply either a definitive loss of payments for the bank, additional
  costs due to renegotiations, active involvement in the borrower’s firm policy,
  use of collateral, or perhaps all of these factors. Thus, the higher the frequency
  of such disturbances, the lower the profitability of a credit contract.

Our paper is linked to the recent literature on relationship banking 1. It contributes
to that literature in several ways: Concerning the surplus question, it augments the
existing literature by studying a different sample of data from German banks. Fur-
thermore, we were allowed to use confidential data contained in each respective
bank’s credit evaluation files. This makes possible the use of various measures
(e.g., the bank’s internal rating of a borrower) that are somewhat different from
(and more comprehensive than) traditional financial accounting measures. In addi-
tion, our study tries to incorporate aspects such as cross-selling and intensity of
competition as independent variables for the surplus question. Third, up to now
and as far as we know, the disturbance question has not been pursued in the rela-
tionship banking literature.

We study these questions by using regression techniques for panel data and dura-
tions analyses. A special feature of our analysis is that it includes not only data
from financial statements, but also from the individual credit contracts. Finally,
when we combine the results of our two regressions, we are able to test several
hypotheses regarding the use of credit contract variables (i.e., collateral and
bonding).


 1 See e.g. Petersen/Rajan (1994); Berger/Udell (1995); Blackwell/Winters (1997).

sbr 52 (4/2000)                                                                        345
R. Ewert/G. Schenk/A. Szczesny


The paper is organized as follows: In Section 2 we discuss the main questions and
theoretical background. Section 3 provides a short overview of the sampling pro-
cedures. Section 4 examines the surplus question. Section 5 concentrates on the
disturbance question. Section 6 contains a short summary of the findings and con-
cludes with some suggestions for future research.


2 THEORETICAL    FRAMEWORK

The theoretical framework of our analysis starts with a model of a neoclassical
credit market in which the terms of credits clear the market. If collateral and other
restrictions (covenants) remain constant, the interest rate is the only price mecha-
nism. With an increasing demand for credit and a given customer supply, the
interest rate rises, and vice versa. If risk is added to this model, the future interest
payments and repayments are stochastic. In this case a surcharge on the alterna-
tive investment without risk is calculated (interest rate premium) relative to the
underlying default probability 2. According to this theory, the corresponding failure
risk should affect the pricing.

Hypothesis 1:    The higher the failure risk of the borrower, the higher the interest
                 rate premium 3.

In this context, collateral has no effect on pricing. The interest rate refers only to
the amount of credit without collaterization. For this reason, a bad lender who
would like to have the same nominal interest rate like a good lender is compelled
to offer more collateral 4. This leads to a negative relation between the amount of
collateral and the interest rate premium. (See hypothesis 2a further on in this
section.)

We now add more complexity to the simple credit market described above, taking
into account the fact that financial transactions are intrinsically characterized by
asymmetric information. Borrowers generally have private (internal) information
about their projects that is more accurate than the information possessed by
lenders. As a consequence, a lender could still be uncertain about the default risk
of a loan contract and have difficulties in assessing and controlling the nature and
behavior of the borrower.

This framework leads to phenomenons like adverse selection and moral hazard.
The adverse selection problem occurs if lenders try to protect themselves against
default risk by setting their contractual terms in a manner appropriate for the
expected average quality of their loan applicants. In this case, high-risk borrowers
could be encouraged to self-select into the loan applicant pool, while at the same
time low-risk borrowers could be encouraged to self-select out of the pool. Moral
hazard arises when borrowers who have internal information take hidden actions
that increase their default probability. The probability for adverse selection or
moral hazard increases with rising interest rates.

2 See e.g. Merton (1974).
3 See e.g. Diamond (1984).
4 See, for example, Rudolph (1984) or Saunders (1997).

346                                                                       sbr 52 (4/2000)
                                                                   Bank Lending Performance


A vast theoretical literature treats these topics 5. Stiglitz/Weiss (1981) show that
credit rationing could be one of the strategies that a bank uses to avoid adverse
selection. Rationing means that the loan amount granted is less than the amount
requested.

Other possible reactions manifest themselves in the terms of the credit contracts.
Using arguments from the agency and signaling theory, better firms can signal
their true quality by offering more collateral or covenants to bondholders 6. Better
firms know that they will not suffer severely from offering more collateral and
covenants, because they have a relatively low probability for the occurrence of
conditions under which covenants are violated, the bank might use the pledged
assets, or both. Thus, it pays for better firms to offer more collateral, covenants, or
both in exchange for lower interest rates. According to this theory, and already
suggested in the neoclassical model, we can formulate the following hypotheses:

Hypothesis 2a: A negative relation should hold between the amount of collateral
               and/or the existence of covenants and the interest rate premium.

Hypothesis 2b: We should further observe a negative relation between the amount
               of collateral and/or the existence of covenants and the distress
               probability.

Practitioners often use a contrary argumentation amounts to a “adverse signaling
argument”. According to their view, banks only require collateral and/or
covenants for relatively risky firms 7. If the firm is instead classified as low risk, the
bank dispenses with collateral and/or covenants.

Hypothesis 3a: According to the adverse signaling argument, a positive relation
               should hold between the amount of collateral and/or the existence
               of covenants and the interest rate premium.

Hypothesis 3b: A larger amount of collateral and/or the existence of covenants
               should be linked with a higher probability of default.

If pure financial contracting theory 8 is used instead, the resulting impact is only
clear for the individual firm but not in a cross-sectional analysis. According to this
theory, lenders can form rational and unbiased expectations on a firm’s future
prospects. There are firm-specific agency-problems that can be mitigated by the
use of collateral and/or covenants. Each firm chooses a specially designed credit
contract that maximizes firm value by trading off additional monitoring and
bonding costs against reductions in interest rate premiums.




 5 See e.g. Jensen/Meckling (1976); Stiglitz/Weiss (1981); Bester (1985); Bester/Hellwig (1989).
 6 For different contexts of such explanations see, e.g., Leland/Pyle (1977); Bester (1985, 1987); Chan/
   Kanatas (1985); John/Kalay (1985); Besanko/Thakor (1987); Ewert (1988).
 7 Bester (1994) gives some theoretical justification for this view.
 8 See e.g. Jensen/Meckling (1976); Myers (1977); Smith/Warner (1979).

sbr 52 (4/2000)                                                                                    347
R. Ewert/G. Schenk/A. Szczesny


Hypothesis 4:    For a single firm, the use of collateral and/or covenants should
                 reduce credit costs, resulting in a negative relation between these
                 variables.

However, in a cross-section that relation need not hold, because usually the firms
with the most severe agency problems (which presumably are high-risk firms) find
it advantageous to offer credit contracts including collateral and/or covenants.
Thus, there could be a positive cross-sectional relation between the observed
interest rate and the use of collateral and/or covenants, depending on which of
the two effects (reduction of individual credit risks versus use of collateral and/or
covenants by observably riskier firms) is stronger.

We can conclue that only the signaling and adverse signaling hypotheses yield
clear implications for the impact of collateral and/or covenants.

Building a house bank relationship is another way to reduce information asymme-
tries and thus avoid moral hazard. The closeness of the relationship between firms
and banks makes it possible for banks to price the default risk of firms in a more
accurate way 9. This leads to the following hypotheses.

Hypothesis 5a: Close relationships between banks and lenders are valuable and
               should lead to a decrease in interest rate premiums.

Hypothesis 5b: In the subgroup of firms in a close bank/customer relationship, the
               default probability should be lower.

The length of the bank/borrower relationship should affect the pricing. Petersen/
Rajan (1995) show that banks with a close relationship to borrowers can offer rel-
atively low interest rates at the beginning of the relationship purposely avoiding
adverse selection and moral hazard. In later periods they charge higher interest
rates to compensate the earlier concessions.

Hypothesis 6:    Taking the borrower’s quality as given, the interest rate premiums
                 should increase in time.

As we note in the introduction, the lending business is often viewed as just a
door-opener for other transactions. This perception ignores the profits from
lending activities.

Hypothesis 7:    The existence of cross-selling arguments should be linked with a
                 lower interest rate premium.

Some smoothing effects blur the relation between the above factors and the inter-
est rate premium. A close relationship between firms and banks gives banks the
opportunity to smooth interest rate premiums over time. According to Petersen/
Rajan (1995), banks follow this strategy to avoid adverse selection and moral
hazard.


9 Theoretical arguments can be found, for example, in Diamond (1989, 1991).

348                                                                           sbr 52 (4/2000)
                                                                  Bank Lending Performance


Fried/Howitt (1980) and Berger/Udell (1995) mention further reasons for smooth-
ing. For example, banks could offer insurance services when general interest rate
levels are high, or to firms in financial shortage 10. Furthermore, the current interest
rate level could affect the spreads 11.

Hypothesis 8:      Loan rate spreads are relatively small in times of high general inter-
                   est rate levels, and vice versa.

In periods of high interest rates, capital costs are high, so banks have trouble
enforcing terms. In addition, the bargaining power implicit in the size of firms
affects the terms.

Hypothesis 9:      With increasing firm size, the interest rate premium should
                   decrease.

Low default rates and losses combined with high interest rate premiums do affect
the performance of the lending business. Therefore, we also analyze the relation
between the observed probability of problems and several determining factors,
especially the bank’s rating system. The results of both regressions (i.e., the
surplus and disturbance questions) must be considered in testing the respective
theories, since the analysis for the disturbance question is a direct test of the rela-
tions between several determining factors and the observed probability of prob-
lems. Thus, these results can either corroborate or contradict the results of the
premium-regression.


3 DATA

In our study we use the common data set of a research project on credit manage-
ment in Germany that was initiated by the Center for Financial Studies (CFS) in
Frankfurt. This section presents an overview of basic information on the data col-
lection that is necessary for the specific research questions of this paper.

The CFS research project on credit policy was performed in cooperation with six
leading German universal banks 12. This cooperation enabled researchers to sys-
tematically analyze the credit files of bank borrowers for the first time. The
research project was restricted to medium-sized firms with an annual sales
volumes between 50 and 500 millions DM 13.

The sample comprises a randomly chosen cross-section of 260 borrowers over the
seven years between 1992 and 1998, and includes an oversampling of potentially

10 See e.g. Fried/Howitt (1980) or Berger/Udell (1995). Fried/Howitt (1980) show that this smoothing is
   efficent for the bank, because the aquisition of new customers is more costly than customer main-
   tenance.
11 See for a description of the money illusion phenomenon Machauer (1999) and for empirical evi-
   dence Berger/Udell (1992) and Machauer/Weber (1998).
12 Bayerische Vereinsbank, Deutsche Bank, Commerzbank, DG Bank, Dresdner Bank, and West LB.
13 Because of possible distortions we exclude East German firms. For such distortions see Harhoff/
   Körting (1998).

sbr 52 (4/2000)                                                                                   349
R. Ewert/G. Schenk/A. Szczesny


distressed firms. One of the criteria for inclusion in this subset is at least one nega-
tive rating (rating 5 or rating 6; see section 4.1) by the borrower during the obser-
vation period.

Table 1: Descriptive statistics concerning the bank internal ratings

Rating                              Observ.    Total lines of    Availment     Collateral     Spread
                                    in percent credit in DM      in percent    in percent     in percent
                                               million (Mean)    (Mean)        (Mean)         (Mean)
1   (very good)                        3.75         15.117          49.77         39.93          3.01
2   (good/above average)              12.57         17.007          51.06         41.45          3.30
3   (average)                         26.26         16.523          63.77         31.30          3.41
4   (below average)                   29.41         14.444          74.36         39.76          3.72
5   (problematic borrower)            20.33         16.157          79.60         43.08          3.97
6   (loan in danger/loss of loan)      7.68         16.969          93.12         44.48          4.25


The complete credit files of each borrower served as the basis for the sample data
collection. This information is supplemented by additional information on the bor-
rower provided by different electronic data processing systems of the respective
bank. Apart from the bank’s internal rating, the data include information about the
terms of credits under current account, investment credits, discount credits, credit
by way of bank guarantee, and other credits, including all kinds of collaterization.
These data are further complemented by some firm characteristics such as the
legal form, branch of business, and important data taken from the firms’ annual
reports or balance sheets. In those cases in which a credit decision or investiga-
tion was documented for a borrower, all variables of interest 14 were collected.

To avoid a survivorship bias, the sample’s population had to include all borrowers
who matched the sampling criteria at some time during the observation period.
That is why some relationships started in the years after 1992 15.

Tables 1 and 2 provide some descriptive statistics.


4 THE SURPLUS         QUESTION

4.1 VARIABLES

This section describes the effects of different determinants on the pricing of
credits in current account. Like Berger/Udell (1995) and Harhoff/Körting (1998),
we restrict our analysis to this type of credit, since we expect more convincing
results than for other forms of credits such as investment credits 16. We choose


14 See the data collection scheme in Elsas et al. (1997).
15 Further the observations of some firms do not cover the whole period due to missing values.
16 Berger/Udell (1995) argue that terms of credits in current account are more influenced by borrower
   quality than are other types of credits such as investment credits. They point out that other credits
   in relation to credits in current account “… tend to be ‘transaction-driven’ rather than relation-
   ship-driven”.

350                                                                                       sbr 52 (4/2000)
                                                                Bank Lending Performance


Table 2: Descriptive statistics

Industrial sector (observations in percent):
Manufacturing                                         29
Machinery                                             22
Construction                                           9
Trade                                                 16
Other                                                 24
Credit Contracts (observations in percent):
Cross-selling (yes / no)                         17 / 83
Covenants (yes / no)                             66 / 34
Housebank relationships (yes / no)               38 / 62
                                      Mean     Std. Dev.
Total lines of credit
in DM million                         15.970      15.194
Availment in percent                      70          37
Collateralization in percent              39          49
Spread in percent                       3.61        1.46
Number of bank relationships             5.8         4.4
Sales in DM million                   162.59      186.32


the spread between the interest rate of the loan and the respective (3-month)
Frankfurt interbank offered rate (FIBOR) as the dependent variable of our panel-
regression 17. We regress this interest rate premium (IRP) on firm and credit vari-
ables and on additional control variables for bank-specific effects. We examine the
interest rate level by using a dummy variable HIGH, coded one in phases with
general high interest rate level, and zero otherwise, with a FIBOR of six percent as
cut off 18. We choose this dummy construction because during the observation
period, the interest rate level decreases along with the average quality of the bor-
rowers. Otherwise, due to this correlation it would have been difficult to separate
the two effects.

The rating (R12, R3, R4, R5, R6) reflects the bank’s individual evaluation of the
loan’s risk and is essentially a compact and comprehensive measure of various
quantitative and qualitative factors (e.g., the quality of the management, the
market position of the firm, and its future prospects). The different internal rating
systems of the five participating banks do not allow a homogeneous assessment of
the quality of the borrowers in the entire data record. Therefore, we had to trans-
form the individual rating systems into a uniform scheme 19.




17 More precisely: We compute the FIBOR-rate for a loan by taking the monthly FIBOR average for
   the month that the credit was granted to the respective firm.
18 During the observation period HIGH indicates phases of in average nine percent interest rate in
   contrast to four percent below the cut off.
19 The rating systems of the five banks and the transformation mechanism are described in detail in
   Elsas et al. (1997). Some descriptive statistics are shown in table 1.

sbr 52 (4/2000)                                                                               351
R. Ewert/G. Schenk/A. Szczesny


We constructed a system with six categories: 1 equals very good, 2 equals
good/above average, 3 equals average, 4 equals below average, 5 equals proble-
matic borrower, and 6 equals loan in danger/loss of loan. The variable R12 reflects
categories 1 and 2, and variables R3 to R6 represent the categories 3 to 6.

Although other papers integrate collateral requirements by using dummy
variables 20 (e.g., one if collateral is pledged, zero if the loan is unsecured),
because we had access to the banks’ credit files, we were able to incorporate col-
lateral in a more detailed form. We use the collateralized percentage of the total
lines of credit (COLLAT). As a value for collateral we take the internal evaluation
of the liquidation value of collateralized assets on which banks base their deci-
sions. For covenants we use a dummy (COV) to account for the existence of such
provisions (e.g., direct and/or indirect dividend constraints 21). For collaterization
in our specifications, we assume a sequential negotiation of terms. In the first step,
banks fix collateral and/or covenants and then negotiate interest rates 22.

To consider cross-selling arguments, we form a dummy variable (CS) which we
code as one in case of cross-selling arguments in the credit file, and zero other-
wise. Since cross-selling arguments probably take effect at the beginning of a rela-
tionship, we also include a dummy that reflects observations in the first five years
of the relationship (CSYOUNG).

To account for possible house bank aspects (HB), we incorporate a dummy coded
one if the bank itself marked the relationship as a house bank relationship, and
zero otherwise 23. HBBAD examines bad borrowers in a house bank relationship.

To have this HB variable is an advantage in comparison with earlier studies in
which the duration (DURATION) of the bank/customer relationship is used as a
measure for the closeness of that relationship. We also use the latter variable in
our study to account for time effects in the relationship.

We further distinguish between the duration of the relationship in case of a house
bank relationship (HBDURAT) and without this relationship (NBDURAT). In addi-
tion, the number of banks which a firm borrows from, NUMBANK, is a proxy for
the closeness of the relationship between the bank and the borrower. This vari-
able can also be viewed as a measure of the quality of the borrower and as an
important indicator of firm size. Sometimes we use the number of bank relation-
ships as a proxy that investigates if bank competition is reflected in loan terms 24.
Firm size must also be considered by the standardized amount of total assets
(LN_TA) 25.


20 This is the procedure used by Berger/Udell (1995) and Blackwell/Winters (1997).
21 For a conceptual analysis of the efficacy of such constraints see John/Kalay (1982), Kalay (1982),
   Ewert (1986, 1987), Berkovitch/Kim (1990) and Leuz (1996).
22 This assumption is also made in the analysis of Harhoff/Körting (1998).
23 For a more detailed procedure concerning the house bank definition see Elsas/Krahnen (1998).
24 For a discussion of the number of banking relationships see Harhoff/Körting (1998) and
   Machauer/Weber (2000).
25 LN_TA = ln(total assets).

352                                                                                  sbr 52 (4/2000)
                                                                 Bank Lending Performance


Table 3: Hypotheses and Variables concerning the interest rate premium

Hypotheses                                                            Variable (Sign)
hypothesis 1:        risk premium                                     R3, R4, R5, R6 (+)
                     (relative to R12)
hypothesis 2a:       neoclassical credit market as well as
                     agency and signaling theory                      COLLAT (–)
                                                                      COV (–)
hypothesis 3a:       adverse signaling theory                         COLLAT (+)
                                                                      COV (+)
hypothesis 4:        pure financial contracting theory                COLLAT (–)
                                                                      COV (–)
hypothesis 5a:       housebank relationship, especially
                     in case of borrowers with relative
                     low quality                                      HB (–)
                                                                      HBBAD (–)
                                                                      NUMBANK (–)
hypothesis 6:        compensation over time,
                     especially in case of a housebank relationship   DURATION (+)
                                                                      HBDURAT (+)
                                                                      NBDURAT (+)
hypothesis 7:        cross selling arguments, especially
                     in case of young customers                       CS (–)
                                                                      CSYOUNG (–)
hypothesis 8:        money illusion                                   HIGH (–)
hypothesis 9:        bargaining power of large firms                  LN_TA (–)
                                                                      NUMBANK (–)


To identify whether banks apply different procedures to the pricing of loans
and/or to control for different sampling procedures, we use dummy variables for
the six banks (B1 … B6) that participated in our research project.

Table 3 gives an overview of the variables described above together with hypothe-
ses itemized in section 2.


4.2 RESULTS

The advantage of using panel data is that we can estimate a random effects
model. A random effects model is an appropriate specification if a number of indi-
viduals (firms in this case) are randomly drawn from large populations (here, bank
customers). This method offers the advantage of eliminating borrower- and time-
specific effects by adding separate random error terms for borrowers and time 26.
The results are shown in table 4.



26 See Baltagi (1995).

sbr 52 (4/2000)                                                                            353
R. Ewert/G. Schenk/A. Szczesny


Table 4: Regression of the interest rate premium (IRP)

                                       Specification 1                         Specification 2
HIGH                                   –1.0112 ***       (0.0893)              –1.0093 ***       (0.0881)
R3                                      0.2699 *         (0.1510)               0.2516 *         (0.1499)
R4                                      0.4436 ***       (0.1642)               0.4437 ***       (0.1630)
R5                                      0.5279 ***       (0.1711)               0.8037 ***       (0.1989)
R6                                      0.8589 ***       (0.2432)               1.1192 ***       (0.2590)
LN_TA                                  –0.1673 **        (0.0766)              –0.1821 **        (0.0768)
COLLAT                                  0.0029 **        (0.0012)               0.0028 **        (0.0012)
COV                                     0.0268           (0.1228)              –0.0297           (0.1229)
CS                                     –0.0625           (0.1418)               0.0781           (0.1494)
CSYOUNG                                                                        –1.0211 ***       (0.3663)
HB                                     –0.2214           (0.1428)              –0.3084           (0.2191)
HBBAD                                                                          –0.6545 ***       (0.2344)
DURATION                                0.0070 *         (0.0041)
HBDURAT                                                                         0.0122 **        (0.0058)
NBDURAT                                                                         0.0018           (0.0052)
NUMBANK                                 0.0066           (0.0163)               0.0141           (0.0164)
B2                                      0.3638 *         (0.2151)               0.4934 **        (0.2200)
B3                                     –0.1482           (0.2662)              –0.0287           (0.2690)
B4                                      0.1125           (0.2531)               0.2186           (0.2557)
B5                                      0.0925           (0.2646)               0.1835           (0.2665)
B6                                      0.4555           (0.2890)               0.6193 **        (0.2940)
CONST                                   5.1619 ***       (0.8836)               5.2663 ***       (0.8854)
Obs.                                        682                                        682
R–Sq. within                             0.2427                                     0.2689
R–Sq. between                            0.2415                                     0.2401
R–Sq. overall                            0.2093                                     0.2222

Significant at the * 1 percent level / ** 5 percent level / *** 10 percent level.
Standard deviations in brackets.

The first specification shows a simple model, and the second specification pre-
sents the detailed hypotheses. Both specifications show clear results concerning
the ratings, thus supporting hypothesis 1.

The coefficients are strongly significant for all rating dummies and increase in the
rating number. The higher the risk, the higher the interest rate premium. A lower-
ing of one rating class goes hand in hand with an increase in interest rate premi-
ums of about 0.2 or 0.3 percent points.

When we test hypothesis 8 (if the loan rate spreads decrease with a rising interest
rate level), we find the expected negative sign. Phases of high general interest rate
levels differ from phases of low general interest rate levels by about 4 percent
points, which is generally accompanied by a 1 percent point difference in the
spreads.

The coefficient for the collateral variable COLLAT is positive and significant in
each of these regressions. The coefficient indicates only a slight effect on the


354                                                                                          sbr 52 (4/2000)
                                                                  Bank Lending Performance


spread: An increase from zero to 100 percent collateralization leads to an increase
in the interest rate premiums of only 0.3 percent points. The coefficient for COV is
not significant.

These results are not consistent with the combined agency and signaling argu-
ments outlined in hypothesis 2a. According to these arguments, we should observe
higher interest rate premiums for firms with lower collateral and fewer covenants,
contradicting the results in table 3. On the other hand, if the adverse signaling
hypothesis holds, then “good” firms are not required either to pledge much collat-
eral or to install covenants. Therefore, they should receive better credit terms. This
view is confirmed by our regression (hypothesis 3a).

The empirical results from using the pure contracting theory are consistent if we
hypothesize that the riskier firms find it more profitable to use collateral and to
install covenants, and that this effect cross-sectionally dominates the individual
interest-reducing effect of using such mechanisms (hypothesis 4). However, as
mentioned above, the premium-regression alone does not allow for a final judge-
ment on the three competing hypotheses.

Consistent with hypothesis 7, cross-selling in the first years of a relationship carries
a negative sign, indicating that chances of cross subsidies between lines of busi-
ness leads the bank to reduce the interest rates. Later on, it makes no difference.

Across all observations, we can identify a slight effect of the duration. This effect
is shown by specification 1. When there is a house bank relationship, the effect is
greater. Duration together with the house bank characteristic (HBDURAT) displays
a positive sign and an increase of the interest rate premium of 0.01 percent points
per year. This must be interpreted together with the result concerning the house
bank status in general. A firm with a close relationship to the bank (HB) can
expect better terms than other firms, but this advantage gets smaller over time.
This finding supports the findings of other empirical studies (hypotheses 5a and
6) 27. For bad borrowers the advantage of house bank relationships is considerable
(HBBAD). This confirms the supposition that banks are likely to help firms in
financial distress, expecting future earnings because of their house bank status.
Without a house bank status, competition among banks prevents such conces-
sions. Since we exclude observations of firms with obvious financial problems, all
results concerning house bank effects lose their significance.

The variable NUMBANK is not significant. Since Machauer/Weber (2000) show
that NUMBANK is essentially an indicator for firm size, the result does not come
as a surprise, because we include total assets as a direct measure of firm size in
our analysis. As the size of the firm increases, its bargaining power as a borrower
grows and lowers the interest rate premiums. Beyond the risk-reducing effect of
firm size covered by the ratings, an increase in total assets leads to a decrease in
the spreads (hypothesis 9) 28.



27 See Petersen/Rajan (1994); Berger/Udell (1995); Blackwell/Winters (1997).
28 This is consistent with the findings of Blackwell/Winters (1997).

sbr 52 (4/2000)                                                                        355
R. Ewert/G. Schenk/A. Szczesny


Spreads vary greatly between different banks. There is also a high variance within
each banks’ loan portfolio. Several reasons could account for this. On the one
hand, our standardization of the ratings may be not as accurate as we thought. On
the other hand, differences in the sampling process could have caused the relation
between spreads and the respective banks. Alternatively, real differences in the
arrangements of terms could be the reason.

By assessing the goodness of fit of the regressions, we must consider that we have
increased the variance of terms by raising the number of potentially distressed
firms. This oversampling stands out from other empirical analysis 29. Nevertheless,
our explanatory power with a R2 (overall) of about 0.20 lies in the middle range of
other results. If financial problems are obvious, banks react and change the terms,
for instance by deferment of payments or demanding additional collateral. It is
obvious that this procedure increases the variance of the terms of the contracts. If
we exclude these problematic oberservations, the R2 (overall) rises for example to
a value of 0.37.


5 DETERMINANTS       OF DISTRESS-PROBABILITIES

5.1 PROCEDURE      AND VARIABLES

To gain some insights into the determinants of the frequency of potential “distur-
bances” we perform a panel logit analysis with random effects. Our aim is to
investigate whether there is any systematic relation between characteristic features
of the loans and the occurence of problems.

We first flag those observations with obvious problems 30. An observation is
marked if one of the following events occured:

• Initiation of formal insolvency proceedings

• Utilization of collateral by the bank

• Valuation adjustments of the bank’s claims

• Initiation or planning of restructuring activities by the bank

• Termination of the bank’s engagement

These criteria comprise a broad spectrum of potential problems that might occur
during a credit engagement. Each criterion involves specific costs that lower the


29 For example Petersen/Rajan (1995) and Berger/Udell (1995), who have no oversampling of poten-
   tially “bad” borrowers.
30 The analysis in this section concentrates on the complete credit engagement (i.e., the firm that is
   granted credit by a bank) rather than looking at single credits. The reason is that potential distur-
   bances can hardly be traced to any single credit, but are regularly caused by the firm’s total debt
   and the investment program.

356                                                                                    sbr 52 (4/2000)
                                                        Bank Lending Performance


bank’s return from lending to the respective firm. We note that our classification
procedure does not rely on the firm rating.

When we apply these criteria to our sample, we find that problems occur in 27
percent of our observations, (coded “1”) and 73 percent of our observations are
without problems (coded “0”). We have multiple failure-per-subject data, since in
some cases problems do not end with the termination of the bank’s engagement.
In these cases, restructuring activities have been successful.

Table 5: Hypotheses and Variables concerning the default probability

Hypotheses                                                     Variable (Sign)
hypothesis 1:     risk premium                                 R3, R4, R5, R6 (+)
                  (relative to R12)
hypothesis 2b:    agency and signaling theory                  COLLAT (–)
                                                               COV (–)
hypothesis 3b:    adverse signaling theory                     COLLAT (+)
                                                               COV (+)
hypothesis 4:     pure financial contracting theory            COLLAT (–)
                                                               COV (–)
hypothesis 5b:    House bank relationship                      HB (–)
                                                               NUMBANK (–)


We base our set of independent variables on the set of variables for the premium-
regressions in section 3, but we make some modifications. We do not include an
indicator for firm size, since the risk-reducing effect of firm size is captured by the
rating, and bargaining power should not affect the probability of problems. We
exclude DURATION in the logit analysis (specification 1) due to endogeneity. The
duration models (specifications 2-4) specify the duration effect by assumption.

In addition to the COLLAT-variable we include two variables that describe the form
of collaterization: MORTGAGE indicates the securization by mortgages and GUA-
RANT is coded “1” if guarantee commitments are given by other persons or firms.


5.2 RESULTS

Table 6 shows the results of four specifications. In the first specification we esti-
mate a logit model with a lagged dependent variable in which the time lag is one
year, i.e., the period between the observed values of the ratings, collateral and so
on and the occurence of problems is one year. A simultaneous examination of the
dependent and independent variables is not advisable, since the ratings are set if
problems become visible. Thus, the causality is inverted.

Specifications 2-4 show the results of a duration analysis. We formalize duration
models by specifying a probability density function for the problem-free period in
the credit engagement. We also include explanatory variables such as collateral to
incorporate additional determinants of this probability.

sbr 52 (4/2000)                                                                     357
R. Ewert/G. Schenk/A. Szczesny


A coefficient with a positive sign in table 4 indicates a positive effect on the prob-
ability of default. For example, a firm with rating of 6 is likely to have problems
earlier than a firm with rating of 5.

Different assumptions about the effect over time lead to different models. Since
we do not know the exact effect of duration, we choose three specifications. The
exponential duration (specification 2) says that the probability of the customer
leaving the non-problem period is the same no matter how long the relationship is
problem-free. The Weibull density (specification 3) accounts for an increasing or
decreasing probability. The Cox proportional model (specification 4) needs no
special assumption 31. The panel logit analysis requires regularly distributed obser-
vations over time. Therefore, we include in the analysis only the last observation
per year and per firm. In contrast, the duration models consider all observations
per year and per firm.

The ratings are highly significant in all specifications. The higher the rating classifi-
cation, the higher the probability for the occurence of problems. Like the spreads
in the previous section, the differences of distress probabilities among the banks
are large. As we did in section 4.2, we offer different explanations. Our standardi-
zation of the ratings could be responsible for the the differences. Or differences in
the sampling process could have caused this result. But the scales and the signifi-
cance indicate that there are large differences in the existing rating systems, even
after standardization. In light of the latest discussion of new standards on capital
adequacy of financial institutions, this result emphasizes the oft-stated doubts
about using internal ratings as criteria for equity requirements. Many financial
experts believe use of these criteria threatens the principle “same business, same
risk, same rules” 32.

Furthermore, the collateral variable COLLAT is significantly related to the probabil-
ity of distress in specification 1. A higher percentage of collateralized credits is
associated with a lower probability of problems in the following year. In the dura-
tion models, the sign indicates the same effect even if the result is not significant.

Our results on securization by mortgage (MORTGAGE) indicates that banks
demand this form of securization if the engagement involves more risk. This is not
surprising, since mortgages are mostly easier to use and might improve the recov-
ery rate if there is a default.

The duration models show clear results concerning covenants (COV). If the con-
tract involves covenants, the probability of problems occuring is less than if there
are no covenants.

Thus, the results for the monitoring and bonding variables contradict the adverse
signaling hypothesis but are consistent with the combined agency-signaling
theory. However, when we return to the results of the premium regressions, if the

31 For an applied survival analysis see e.g. Hosmer/Lemeshow (1999). Kiefer (1988) gives a brief
   overview.
32 The results of a survey among financial experts in Germany to the subject of the new capital ade-
   quacy framework (ZEW, April 2000) show that 51 percent of the experts express this opinion.

358                                                                                 sbr 52 (4/2000)
                                                                         Bank Lending Performance


Table 6: Regressions of default probability

                                    Specification 1    Specification 2     Specification 3   Specification 4
                                       (Logit)         (Exponential)          (Weibull)          (Cox)
R3                                     2.5766 ***          0.9263             0.9442 *          0.9679 *
                                      (0.8450)            (0.5697)           (0.5693)          (0.5727)
R4                                     4.3778 ***          2.4397 ***         2.4464 ***        2.4270 ***
                                      (0.8962)            (0.5176)           (0.5174)          (0.5193)
R5                                     5.7353 ***          3.1969 ***         3.1909 ***        3.0994 ***
                                      (0.9416)            (0.5142)           (0.5140)          (0.5183)
R6                                     9.0224 ***          3.5558 ***         3.5503 ***        3.4517 ***
                                      (1.3630)            (0.5165)           (0.5162)          (0.5194)
COLLAT                                –0.0090 *          –0.0015             –0.0015          –0.0011
                                      (0.0050)           (0.0015)            (0.0015)         (0.0016)
MORTGAGE                               1.5283 ***         0.3531 **           0.3474 **        0.3518 **
                                      (0.5566)           (0.1518)            (0.1523)         (0.1659)
GUARANT                               –0.3045            –0.1450             –0.1284          –0.1643
                                      (0.4565)           (0.1273)            (0.1283)         (0.1376)
COV                                    0.1602            –0.3112 **          –0.3044 **       –0.3060 **
                                      (0.4233)           (0.1246)            (0.1245)         (0.1347)
HB                                    –0.3404            –0.1122             –0.1083          –0.1131
                                      (0.5239)           (0.1345)            (0.1348)         (0.1461)
NUMBANK                                0.0147             0.0160              0.0122           0.0102
                                      (0.0545)           (0.0149)            (0.0152)         (0.0165)
B2                                     6.6469 ***         1.5168 ***          1.5627 ***       1.5626 ***
                                      (1.2298)           (0.2605)            (0.2619)         (0.2819)
B3                                     3.4527 ***         0.8143 ***          0.9356 ***       0.9089 ***
                                      (1.1693)           (0.2922)            (0.3029)         (0.3219)
B4                                     1.0790            –0.1900             –0.1445          –0.3915
                                      (1.2180)           (0.3498)            (0.3512)         (0.3717)
B5                                     2.4537 **          0.4932 *            0.5685 *         0.5105 *
                                      (1.1266)           (0.2912)            (0.2956)         (0.3066)
B6                                     7.1679 ***         1.9333 ***          1.9339 ***       1.8843 ***
                                      (1.3623)           (0.2803)            (0.2807)         (0.3102)
CONST                               –10.4386 ***         –4.5772 ***         –5.0369 ***
                                     (1.6354)            (0.5778)            (0.6508)
Obs.                                       662               1160                1160             1160
Pseudo R–Sq. 33                         0.2327             0.2798              0.2774           0.2212

Significant at the * 1 percent level / ** 5 percent level / *** 10 percent level.
Standard deviations in brackets.

combined agency- and signaling hypothesis holds we should observe a negative
relation between the IRP and the use of collateral and/or covenants. Since this
observation is not supported by the premium regressions, the empirical results of
both regressions do not completely corroborate the combined agency- and signal-
ing theory.

33 It is problematic to calculate pseudo R-squareds because of a high percentage of censored data.
   For this reason the R-squareds seem low. For the calculation and discussion of the Pseudo R-
   squareds see Hosmer/Lemeshow (1999), p. 229.

sbr 52 (4/2000)                                                                                          359
R. Ewert/G. Schenk/A. Szczesny


One possible interpretation of the sign of COLLAT is that the monitoring and
bonding devices are actually useful in reducing debt-related agency problems.
Therefore, the incidence of disturbances should decrease the more collateral and/
or covenants are used. This finding is confirmed by our empirical results. At first
glance, this argument is consistent with the pure contracting theory, but only
because the predictions of that theory are somewhat vague regarding the sign of
the coefficients in a cross-section. We could obtain corroboration for our argu-
ments if the premium regressions in section 4 revealed a negative relation
between COLLAT and/or COV and the IRP, but this is not the case. When we take
all empirical results together, we obtain interpretations that are both consistent
and inconsistent with either of the three hypotheses. For the time being, any clear-
cut statement is impossible.


6 SUMMARY

This paper uses a unique data set from credit files of six leading German banks to
provide some empirical insights into determinants of bank lending performance in
Germany. We use panel techniques and methods for duration analyses to analyze
structural relations for interest rate premiums and for (broadly defined) distress
probabilities.

For the premium regressions we find that the banks’ rating is significant and posi-
tively related to the interest rate premium (as expected). Cross-selling arguments
are relevant mainly for the first years of a relationship and lead to a reduction of
the interest rate spread. In addition, our results for the interest rate premiums are
consistent with hypotheses derived from adverse signaling theories, i.e., higher
collateralization is associated with a higher premium.

For distress probabilities, the results for the rating variables are structurally consis-
tent with those of the premium regression, i.e., the higher the rating, the higher
the empirically observed occurrence of problems. However, the results for collat-
eralization do not corroborate those for the interest rate premiums. More collateral
and the existence of covenants are significantly associated with lower distress
probabilities. This is consistent with the combined agency and signaling theory,
but contradicts the adverse signaling argument that has proven to be useful in
explaining the premium results.

Taken together, the results of both regressions seem to imply that credit contracts
are priced lower where the risks are greater. This constitutes an empirical puzzle
that should be further analyzed by future research.

We also find that in both regressions, bank dummies are significant. This suggests
that there are important idiosyncratic elements in the rating and evaluation
systems of the respective banks. Thus, in light of the recent discussion on equity
requirements using internal ratings, we might need to harmonize some of the
internal rating process if we are to expect the principle “same business, same risk,
same rules” to hold.



360                                                                        sbr 52 (4/2000)
                                                                  Bank Lending Performance


REFERENCES

Baltagi, Badi H. (1995): Econometric Analysis of Panel Data.
Berger, Allen N./Udell, Gregory F. (1992): Some Evidence on the Empirical Significance of Credit
   Rationing, in: Journal of Political Economy, Vol. 100, pp. 1047 – 1077.
Berger, Allen N./Udell, Gregory F. (1995): Relationship Lending and Lines of Credit in Small Firm
   Finance, in: Journal of Business, Vol. 68, pp. 351 – 381.
Berkovitch, Elazar/Kim, E. Han (1990): Financial Contracting and Leverage Induced Over- and Under-
   Investment Incentives, in: Journal of Finance, Vol. 45, pp. 765 – 794.
Bester, Helmut (1985): Screening vs. Rationing in Credit Markets with Imperfect Information, in: Ameri-
   can Economic Review, Vol. 75, pp. 850 – 859.
Bester, Helmut (1987): The Role of Collateral in Credit Markets with Imperfect Information, in: Euro-
   pean Economic Review, Vol. 31, pp. 887 – 899.
Bester, Helmut (1994): The Role of Collateral in a Model of Debt Renegotiation, in: Journal of Money,
   Credit and Banking, Vol. 26, pp. 72 – 86.
Bester, Helmut/Hellwig, Martin F. (1989): Moral Hazard and Equilibrium Credit Rationing: An Overview
   of the Issues, in: Bamberg, Günter/Spremann, Klaus (Eds.): Agency Theory, Information, and Incen-
   tives, pp. 135 – 166.
Besanko, David/Thakor, Anjan V. (1987): Collateral and Rationing: Sorting Equilibria in Monopolistic
   and Competitive Credit Markets, in: International Economic Review, Vol. 28, pp. 671 – 690.
Blackwell, David W./Winters, Drew B. (1997): Banking Relationships and the Effect of Monitoring on
   Loan Pricing, in: Journal of Financial Research, Vol. 20, pp. 275 – 289.
Chan, Yuk-Shee/Kanatas, George (1985): Asymmetric Valuations and the Role of Collateral in Loan
   Agreements, in: Journal of Money, Credit and Banking, Vol. 17, pp. 84 – 95.
Diamond, Douglas W. (1984): Financial intermediation and delegated monitoring. Review of Economic
   Studies, Vol. 51, pp. 393 – 414.
Diamond, Douglas W. (1989): Reputation Acquisition in Debt Markets, in: Journal of Political Economy,
   Vol. 97, pp. 828 – 861.
Diamond, Douglas W. (1991): Monitoring and Reputation: The Choice between Bank Loans and
   Directly Placed Debt, in: Journal of Political Economy, Vol. 99, pp. 688 – 721.
Elsas, Ralf/Henke, Sabine/Machauer, Achim/Rott, Roland/Schenk, Gerald (1997): Empirical analysis of
   credit relationships in small firms financing: Sampling design and descriptive statistics, Working
   Paper 98 – 06 (Center for Financial Studies Frankfurt).
Elsas, Ralf/Krahnen, Jan-Pieter (1998): Is Relationship Lending Special? Evidence from Credit-File Data
   in Germany, Working Paper 98-05 (Center for Financial Studies Frankfurt).
Ewert, Ralf (1986): Rechnungslegung, Gläubigerschutz und Agency-Probleme.
Ewert, Ralf (1987): The Financial Theory of Agency as a Tool for an Analysis of Problems in External
   Accounting, in: Bamberg, G./Spremann, K. (eds.): Agency Theory, Information, and Incentives, pp.
   281 – 309.
Ewert, Ralf (1988): Finanzierungsrestriktionen, Kreditverträge und Informationsasymmetrie, in: Heil-
   mann, W. et al., (eds.): Geld, Banken und Versicherungen, (Versicherungswirtschaft e.V.), pp. 829 –
   843.
Fried, Joel/Howitt, Peter (1980): Credit Rationing and Implicit Contract Theory, in: Journal of Money,
   Credit, and Banking, Vol. 12, pp. 471 – 487.
Harhoff, Dietmar/Körting, Timm (1998): Lending Relationships in Germany: Empirical Results from
   Survey Data, in: Journal of Banking and Finance, Vol. 22, pp. 1317 – 1353.
Hosmer, David W./Lemeshow, Stanley (1999): Applied Survival Analysis.
Jensen, Michael C./Meckling, William H. (1976): Theory of the Firm: Managerial Behavior, Agency Costs
   and Ownership Structure, in: Journal of Financial Economics, Vol. 3, pp. 305 – 360.
John, Kose/Kalay, Avner (1982): Costly Contracting and Optimal Payout Constraints, in: Journal of
   Finance, Vol. 37, pp. 457 – 470.
John, Kose/Kalay, Avner (1985): Informational Content of Optimal Debt Contracts, in: Altman, E./Sub-
   rahmanyam, M. (eds.): Recent Advances in Corporate Finance, pp. 133 – 161.
Kalay, Avner (1982): Stockholder-Bondholder Conflict and Dividend Constraints, in: Journal of Finan-
   cial Economics, Vol. 10, pp. 211 – 233.


sbr 52 (4/2000)                                                                                   361
R. Ewert/G. Schenk/A. Szczesny


Kiefer, Nicholas M. (1988): Economic Duration Data and Hazard Functions, in: Journal of Economic Lit-
   erature, Vol. 26, pp. 646 – 679.
Leland, Mayne E./Pyle David H. (1977): Informational Asymmetries, Financial Structure and Financial
   Intermediation, in: The Journal of Finance, Vol. 23/2, pp. 371 – 387.
Leuz, Christian (1996): Rechnungslegung und Kreditfinanzierung.
Machauer, Achim (1999): Bankverhalten in Kreditbeziehungen.
Machauer, Achim/Weber, Martin (1998): Bank Behaviour based on Internal Credit Ratings of Borrow-
   ers, in: Journal of Banking and Finance, Vol. 22, pp. 1355 – 1383.
Machauer, Achim/Weber, Martin (2000): Number of Bank Relationships: An Indicator of Competition,
   Borrower Quality, or just Size?, Discussion Paper No. 2000/06, Center for Financial Studies Frankfurt.
Merton, Robert C. (1977): An analytic derivation of the cost of deposit insurance and loan guarantees,
   in: Journal of Banking and Finance, Vol. 1, pp. 3 – 11.
Myers, Stewart C. (1977) Determinants of Corporate Borrowing, in: Journal of Financial Economics, Vol.
   5, pp. 147 – 175.
Petersen, Mitchell A./Rajan, Raghuran G. (1994): The Benefits of Lending Relationships: Evidence from
   Small Business Data, in: Journal of Finance, Vol. 49, pp. 3 – 37.
Petersen, Mitchell A./Rajan, Raghuran G. (1995): The Effect of Credit Market Competition on Lending
   Relationships, in: The Quaterly Journal of Economics, Vol. 110, pp. 407 – 443.
Rudolph, Bernd (1984): Kreditsicherheiten als Instrumente zur Umverteilung und Begrenzung von Kre-
   ditrisiken, in: zfbf, Vol. 36, pp. 16 – 43.
Saunders, Anthony (1997): Financial Institutions Management: A Modern Perspective, 2nd. ed.
Stiglitz, Joseph E./Weiss, Andrew (1981): Credit Rationing in Markets with Imperfect Information, in:
   American Economic Review, Vol. 71, pp. 393 – 410.
Smith, Clifford W./Warner, Jerold B. (1979): On Financial Contracting: An Analysis of Bond Covenants,
   in: Journal of Financial Economics, Vol. 7, pp. 117 – 161.




362                                                                                     sbr 52 (4/2000)

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:2
posted:11/5/2012
language:Unknown
pages:19