Loan supply surveys

LINK DOWNLOAD MIỄN PHÍ TÀI LIỆU "Loan supply surveys ": http://123doc.vn/document/561372-loan-supply-surveys.htm

5
binding in their model, liquidity constraints are not. As a result, the operation
of bank lending channel was restricted. In Hurlin and Kierzenkowski (2002)
the focus is on interest rate channel and pass-through of changes in official
rates to rates on loans, which is found to be very swift.
No studies so far have been dedicated to study the bank willingness to lend
and its impact on credit supply in the Polish economy. Pruski and
˙
Zochowski
(2006) and Brzoza-Brzezina, Chmielewski, and Nied´zwiedzi´nska (2008) report
on the high level of substitution between foreign currency and zloty lending,
which to some extent outweighs the impact of monetary policy on loan sup-
ply. Moreover, over the last decade Polish banks have been operating in the
environments of excess funding liquidity, which resulted from systematically
higher level of deposits than credit in the system. This two features of the
Polish banking sector may indicate that bank willingness to lend may be an
important driving force in the Polish credit market.
Bank lending surveys provide a powerful set of data to test different hy-
potheses about bank lending channel. In particular, questions about the rea-
sons of changes in lending policies are related to different types of risk, which
separately can be tested for their influence on banks’ willingness to lend. How-
ever, in this paper, due to short time horizon and low frequencies of answers
other than ”no-change” to questions on the lending terms and conditions as
well as reasons for changing them, we concentrate on answering whether, in
general, altering bank lending standards or conditions affect banks’ loan sup-
ply. Since we control for demand effects and individual bank effects, we for-
mulate and test the following main hypothesis:
H0: Tightening/ easing of bank lending policies leads to decrease/ increase
in individual bank loan supply
It is a necessary but not sufficient condition for the existence of risk tak-
ing channel. Also banks’ risk perception would have to change following the
change in the monetary policy stance. Although we do not test this in the
paper, we give some insights into the determinants of the changes of bank
lending policies, which seem to support our view that changes in perception
of risk by banks is an important driver of changes in lending policies. Since
according to Bernanke and Lown (1991); Woo (1999), changes in bank lending
policy are related either to capital constraints (for which we control) or to
shifts in perception of risks, our results provide some support toward the sig-
nificance of risk taking channel. Moreover, Rajan (1994) and Berger and Udell
(2004) demonstrate that banks tend to curb lending in economic downturns
by changing lending standards. Their results point to the importance of bank
lending policies to the broad economy and the business cycle.
4 Data
We used three types of data to verify the existence of monetary policy trans-
mission channels. These are the survey data on bank lending policy, individual
bank financial data and macroeconomic variables (see table 1).
6
The data on bank lending policy come from the Senior Loan Officer Opinion
Survey (SLOS), which has been carried out by the National Bank of Poland
(NBP) on a quarterly basis since December 2003. The survey questionnaire,
available from the NBP website
2
, resembles the questionnaire for ECB bank
lending survey. The results are published by the NBP (NBP, 2009).
In SLOS, 24 banks are asked whether they changed standards or terms on
loans over the previous quarter
3
. Separate questions address the situation in
the housing loan, other consumer loan and corporate loan markets. Banks are
asked to provide information on changes in lending standards with regard to
loans to large enterprises and to small and medium enterprises separately. We
used the responses of individual banks as a measure of changes in their lending
policy.
The Senior Loan Officer Opinion Survey is a qualitative survey. Partici-
pants may choose from a set of five options:
– the bank significantly eased its lending policy,
– the bank slightly eased it lending policy,
– the lending policy was unchanged,
– the bank slightly tightened its lending policy, or
– the bank significantly tightened its lending policy.
Lending standards are defined as minimum acceptance criteria which must
be met by a prospective borrower to be approved for a loan, regardless of
the loan’s price and other terms the bank is willing to offer. Terms on loans,
defined as features of the loan contract which may be negotiated after loan
is approved, are broken down into six categories: spreads on regular loans,
spreads on high-risk loans, loan maturity, collateral, fees and maximum loan
amount. In the case of housing loans, banks are also asked about changes in
the required loan-to-value ratio.
In the NBP survey, the definition of terms and standards has been provided
to all participating institutions for clear demarcation between the two areas of
lending policy. Some potential for misinterpretation of the definition remains
and may lead for instance to reporting changes in lending standards as change
in terms on loans. We consider such behaviour to be manifested in the open
question on changes in other terms on loans (i.e., not explicitly mentioned
in the questionnaire)
4
. This is indicated by individual responses to the open
questions, in which banks sometimes note that they have actually changed
lending standards
5
. We filter our dataset for instances when a bank reported no
change in lending standards and simultaneously indicated a change in lending
2
http://www.nbp.pl/en/systemfinansowy/ankieta
en.pdf
3
Effective from October 2008 survey, the sample has been expanded to cover 30 banks
whose market share exceeds 80%. We did not include this expansion in our estimations as
the time series for additional banks do cover only a fraction of the credit cycle and may
have thus distorted the results.
4
Questions 3.7, 9.8 and 11.7 in the questionnaire.
5
Such as parameters in the scoring system, minimum eligible score, or minimum eligible
income.
7
standards in a corresponding open question on other terms on loans, and treat
such instances as a change in lending standards.
The volume of lending may be related both to the level and the change in
lending policy. We consider this possibility in construction of variables which
measure the impact of lending policy on bank lending behaviour.
We measure the impact of lending standards and loan terms on the volume
of new loans separately to allow for diverse responses of loan supply to these
factors. For lending standards, we construct two dummy variables which indi-
cate that a bank tightened or eased lending standards to allow asymmetries in
the response of loan volume to changes in lending policy. We do not take into
consideration the perceived size of change in lending policy, only its direction.
While terms on loans could have been treated similarly, with two dummies
being used to represent the tightening and easing of each of 6-7 categories of
terms, such approach would not be feasible with our small dataset.
The variable reflecting the changes in each bank’s terms on loans is an index
of general restrictiveness of loan terms
6
. The Senior Loan Officer Opinion
Survey measures only changes in lending policy, and not how conservative
bank lending policy is. There is no data on the actual restrictiveness of loan
terms offered by individual banks. We set our loan terms’ variable to 0 as
of the first edition of the Senior Loan Officer Opinion Survey (i.e. the third
quarter of 2003). For each bank in the sample, the starting point of ”zero”
restrictiveness is likely different
7
. Then, for period t the loan terms’ variable
T erms
t
would be given by the following formula:
T erms
t
= T erms
t−1
+
k

i=1
Ind
i
, (1)
where k is the number of categories of terms on loans (i.e., either 6 or 7) and
Ind
i
is an indicator variable such that:
– Ind
i
= 1 if the bank eased its lending policy with respect to the i-th
category of terms on loans,
– Ind
i
= −1 if the bank tightened its lending policy with respect to the i-th
category of terms on loans,
– Ind
i
= 0 otherwise.
For illustration, if in the first edition of SLOS a bank increased the spread
on regular loans and decreased the maximum available loan amount, our index
of restrictiveness would be -2. Then, if in the next period the bank decided to
decrease the maximum loan maturity, lower its loan extension fee and demand
6
We constructed a similar index for restrictiveness of lending standards, and performed
estimations using it instead of dummies for tightening and easing of lending standards.
Results did not differ materially from what is reported in this paper, and can be obtained
from the authors upon request.
7
The differences between banks’ lending policies at the beginning of the sample period
translates into individual effects.
8
less collateral, the index would consequently increase from -2 to -1. More suc-
cinctly, the index changes from one period to the other by the net number of
categories of loan terms with respect to which the bank changes its lending
policy.
We supplement our analysis with the balance-sheet and P&L data on indi-
vidual banks. The bank-level financial data come from the prudential reporting
system of the National Bank of Poland. All institutions with a Polish banking
licence, as well as branches of foreign banks in Poland, are required by the Act
on the National Bank of Poland to report a wide scope of financial information
with monthly or quarterly frequency. The data undergo a quality control, but
are not audited by independent parties. However, banks must supply amended
data should their regular auditor or the NBP find any inconsistencies or mis-
takes.
We use three bank-level variables to represent characteristics of individual
banks. The application of the variables is consistent with what is proposed
in the literature on lending channel (Berger and Udell, 2004; Hernando and
Martinez-Pag´es, 2001; Kishan and Opiela, 2000; Altunbas et al, 2009). The
Basel capital adequacy ratio is a measure of how well a bank is capitalized.
While the rules for calculation of this ratio have changed over time, its binding
minimum level of 8% remained unchanged, and the higher the ratio, the less
likely it is that bank experiences capital shortage. As a measure of liquidity we
use interbank gap, which we define as the ratio of the bank’s net position vis-
a-vis other banks (i.e., its gross claims on other banks minus gross liabilities
to other banks) to bank’s total assets
8
. Positive interbank gap indicates a
favourable net liquidity position, as the bank has excess funds to lend out
in the interbank market. However, if interbank gap is negative, it may be
either due to weak liquidity position or to strategic choice (to rely on) of
foreign funding, and in the Polish context this would mean chiefly intra-group
funding. Finally, the logarithm of the total number of accounts held at the
bank is set as our proxy for bank size.
As proxies for interest rates, we take average three-month Polish zloty
money market index (3-month WIBOR) and Swiss franc 3-month LIBOR.
The LIBOR rate is to represent foreign interest rate. It is economically relevant
because of significant share of bank credit in Poland was extended in foreign
currencies, especially in the Swiss currency. We also use the official GDP and
CPI data for Poland.
The flow of credit may be subject to seasonal fluctuations. A simple ex-
ample would be the much higher flow of consumer credit in November and
December than in other months due to Christmas shopping season. To control
for such fluctuations, we use seasonal dummies representing the first, second
and third quarter of a ear. We also correct for one merger which occurred
between participating banks in 2007 using additional dummies.
8
We also tested another measure of liquidity, the loan-to-deposit ratio, and got similar
results.
9
Banks which participate in the Senior Loan Officer Opinion Survey cover
approximately three-fourths of the respective loan markets in Poland (ca. 80%
of housing loan and corporate loan markets, and 65% of consumer loan mar-
ket). 21 of them are commercial banks, 2 are branches of foreign credit insti-
tutions, and one is a cooperative bank. Not all 24 banks are active in each
segment of the credit market.
Our sample consists of 13 institutions which extend housing loans, 18 in-
stitutions which are active in the corporate loan market and 16 institutions
issuing consumer loans. All major participants in the corporate and housing
loan markets are represented in this sample. In the consumer loan market,
some specialized banks, especially those which emerged as major players after
2004, do not participate in the survey and are not represented in the sam-
ple. Most institutions in the sample are owned by an ultimate foreign parent,
a situation characteristic for the Polish banking system. The composition of
our sample leads, by definition, to exclusion of new entrants which appeared
in some market segments, and therefore the results may not be valid for the
banking system as a whole.
We decided to drop the cooperative bank from the Senior Loan Officer
Opinion Survey sample of 24 banks due to the very limited geographical scope
of its operations. Since this bank accounts for a very small fraction of the
loan market, its exclusion does not cause any material loss of information.
We also removed two branches of foreign credit institutions, as they do not
have their own equity and are not obliged to meet the capital requirement
in the host country. Therefore, branches are not restricted in their lending
policy by leverage constraints faced by commercial banks. Moreover, the degree
of business independence of branches, as compared to commercial banks, is
markedly lower. This increases the likelihood that their lending policy may be
determined by the parent institution. Bearing these peculiarities of branches
in mind, we decided not to include them in the analysis. We also remove banks
which did not change their lending policy throughout the period of the study,
since they do not help in explaining variation of loan supply.
5 Estimation
We estimate parameters of a reduced form model which attempts to identify
the impact of supply-side factors on loan growth in the Polish credit markets,
while controlling for loan demand effects. Given the oligopolistic features of the
bank loan market in Poland (Kozak and Pawlowska, 2008), in which borrowers
have very little bargaining power
9
, we treat loan demand as exogenous and we
assume that it can be described by a function of macroeconomic variables such
as interest rates, GDP and inflation. The choice of variables representing the
supply-side effects is based on literature presented in Section 2 and is described
in details in Section 4.
9
Other credit markets in Poland, such as corporate bond market, are at a very early
stage of development, and cannot be treated as substitute for bank loans.
10
Table 1 The description of the variables
Name Description of variable
DiffHLToA Quarterly change in housing loans normalised by assets
DiffHL Quarterly precentage change in housing loans
DiffCorpLToA Quarterly change in corporate loans normalised by assets
DiffCorpL Quarterly percentage change in corporate loans
DiffConsLToA Quarterly change in consumer loans normalised by assets
DiffConsL Quarterly percentage change in consumer loans
TermsLevelH Index of restrictiveness of terms on housing loans
StdTighteningH Dummy for tightening of lending standards on housing loans
StdEasingH Dummy for easing of lending standards on housing loans
TermsLevelCorp Index of restrictiveness of terms on corporate loans
StdTighteningCorp Dummy for tightening of lending standards on corporate loans
StdEasingCorp Dummy for easing of lending standards on corporate loans
TermsLevelCons Index of restrictiveness of terms on consumer loans
StdTighteningCons Dummy for tightening of lending standards on consumer loans
StdEasingCons Dummy for easing of lending standards on consumer loans
CAR Basel capital adequacy ratio of the bank
GapIBank Total interbank loans of the bank minus its total interbank borrowings,
as fraction of bank’s assets
LogAcc Logarithm of the number of accounts at the bank
GDPGrowth Real GDP growth rate (yoy)
Wibor3M Mean 3-month Warsaw Interbank Offered Rate over the quarter
LiborCHF3M Mean 3-month Swiss franc LIBOR over the quarter
CPI CPI inflation
Two types of models for each category of loans were estimated. In the main
models, we use the quarterly credit growth by loan type as dependent variables.
However, our sample of banks is very diverse, and some banks have revised
their business models markedly since the Senior Loan Officer Opinion Survey
was launched in 2003. As a result, some banks may have experienced relatively
high loan growth rates only due to the low basis. We employ a supplementary
model in which the loan growth rates are normalised by bank’s assets at the
beginning of the quarter to serve as a robustness check for the results from
the main model.
Both models are considered in a static specification
10
. To account for the
observed heteroscedasticity and serial autocorrelation in the data we applied
Prais-Winsten transformation (Baltagi, 2001).
11
Our main model is repre-
sented by Equation 2.
10
However, we check the estimates in the dynamic setting applying several competing
estimators (GMM estimator proposed by Arellano and Bond (1991), its System GMM ex-
tension (Blundell and Bond, 1998) with robust standard errors which utilise the correction
of Windmeijer (2005)). We use the levels of loan-to-assets ratio instead of the loan growth
normalised by assets as the dependent variable, and its lagged values as independent vari-
able. Nevertheless, the estimations did not prove to be statistically significant since the data
panel is almost quadratic. This validates the use of a static specification
11
A similar correction can be obtained using Feasible GLS. It has, however, been subject
to critique from Beck and Katz (1995) that standard errors yielded by Feasible GLS are
understated. Having attempted to apply Feasible GLS correction for the suspected patterns
of serial correlation in our data, we obtained standard errors which were by at least an
11
DiffLoanTypeToLoan
it
= β
1
StdEasingLoanType
i,t−n
+β
2
StdTighteningLoanType
i,t−n
+ β
3
TermsLevelLoanType
i,t−m
+
K+3

k=4
β
k
BankSpecVar
(k−3)
i,t−1
+
K+4+N

k=K+4
β
k
MacroVar
(k−K −3)
t
+ u
it
, (2)
u
it
:= µ
i
+ρu
it−1
+
it
is an autoregressive error component with random indi-
vidual effects µ
i
and common autocorrelation structure ρ.
12
The independent
variables were at least one period lagged to avoid endogeneity.
Similarly, our supplementary model is given in the static form by Equation
3.
DiffLoanTypeToAssets
it
= β
1
StdEasingLoanType
i,t−n
+β
2
StdTighteningLoanType
i,t−n
+ β
3
TermsLevelLoanType
i,t−m
+
K+3

k=4
β
k
BankSpecVar
(k−3)
i,t−1
+
K+4+N

k=K+4
β
k
MacroVar
(k−K −3)
t
+ u
it
, (3)
For our static specifications, we followed a two-step procedure of estimation
in order to get the best possible estimator – consistent and efficient. We tried
random effects (RE), fixed effect (FE) or pooling method (POOL) and chose
the most appropriate one according to the following procedure:
1. (RE or FE vs POOL) First, we verify the hypothesis of no individual effects
in the model (2) by Breusch-Pagan Lagrange Multiplier test for random
individual effects and ANOV A F test based on comparison of within and
pooled models. We also perform the joint LM test for random individual
effects and first-order serial correlation proposed by Baltagi and Li (1991).
2. (RE vs FE) Secondly, in case individual effects may prove important in
explaining variability of banks loan supply, we check if RE procedure per-
forms well in a statistical sense resulting in the most efficient estimators.
To test whether random effect specification gives consistent estimators of
parameters, we employ the Hausman test, comparing GLS estimates of RE
model and the fixed effects (within) model.
6 Results
According to the proposed estimation approach we obtain two specifications
of the model for each segment of the loan market.
order of magnitude lower than in any other estimation. Hence, we decided to use only the
Prais-Winsten estimators.
12
We performed additional test of the stability of ρ across panels estimating the analogous
equation with panel specific correlation in the error term u
it
:= µ
i
+ ρ
i
u
it−1
+ 
it
.
12
6.1 Housing loans
We proposed the following model explaining variability of housing loans growth.
DiffHL
i,t
= β
1
TermsLevelH
i,t−2
+ β
2
StdChangeH
i,t−2
+β
3
CAR
i,t−1
+ β
4
GapIBank
i,t−1
+ β
5
LogAcc
i,t−1
+β
6
GDPGrowth
t−1
+ β
7
Wibor3M
t−1
+β
8
CPI
t−1
+ β
9
LiborCHF3M
t−1
+ u
it
, (4)
As to lending standards, we allow the loan growth to respond asymmetri-
cally to tightening and easing of lending policy.
Table 2 presents the parameter estimates yielded by various panel model
procedures applied to Model 4. Summary of statistical tests which we con-
ducted for all our models is given in Table 4. The data give strong evidence
that individual, bank specific effects determine dynamics of housing loans.
Breusch-Pagan and ANOV A F tests reject the hypothesis of no individual
effects. In the Hausman test, we did not reject that RE gives consistent esti-
mates of parameters. Application of the Prais-Winsten estimators is supported
by serial correlation present in the data, as well as significant between-group
heteroscedasticity.
In an analogous model where the growth of housing loans is normalised by
assets, we also favour the RE procedure, corrected for heteroscedasticity and
serial correlation. Such approach is confirmed by the results of Breusch-Pagan
and Hausman tests. Table 3 summarizes the results for this model.
The impact of terms on housing loans on their supply was found to be
significant. Net change in the restrictiveness index for loan terms by -1, which is
equivalent to tightening of one of six terms on housing loans, led to a slowdown
of quarterly housing loan growth by 0.34 to 1.01 percentage points after two
quarters. The marginal effect of tightening of terms on loans was comparable to
that of 0.07-0.08 p.p. rise in market interest rates. In contrast, credit standards
have little impact on loan growth, regardless of the direction in which they are
changed.
Housing loan supply does not depend on capital adequacy of banks. This
can be explained by the fact that most banks in the sample
13
were controlled
or even fully owned by foreign financial institutions. Many Polish banks had,
at least until the 2008-2009 financial market turmoil, an almost unrestricted
access to capital from their parent institution. Some rapidly expanding banks
could have operated close to the regulatory minimum capital requirement of
8%, because they had been reassured by the (implicit or explicit in the business
plans) commitment of the parent institution to provide additional capital if
needed.
The evidence on how bank’s liquidity position may affect the growth of its
housing loan book is mixed. Our main model suggests that the relationship
is positive, i.e. the more liquid the bank, the faster should loan book expand.
13
Eleven banks out of total of 13 banks in the sample.
13
Table 2 Model of growth of housing loans
Prais-Winsten
(PSAR1)
Prais-Winsten
(AR1)
RE FE
StdTighteningLag2 -0.0377 -0.0199 -0.0485 -0.0458
(0.127) (0.496) (0.169) (0.210)
StdEasingLag2 -0.0045 -0.0007 0.0020 0.0081
(0.834) (0.982) (0.947) (0.797)
TermsLevelLag2 0.0101 0.0021 0.0034 0.0048
(0.047) (0.636) (0.022) (0.098)
CARLag1 0.0083 -1.0909 0.6195 0.4872
(0.992) (0.328) (0.112) (0.255)
GapIBankLag1 0.3901 0.2974 0.1831 0.2201
(0.020) (0.090) (0.146) (0.257)
LogAccountsLag1 -0.0872 -0.0530 -0.0501 -0.0995
(0.000) (0.083) (0.001) (0.063)
WIBOR3MLag1 0.0403 -0.0075 -0.0002 -0.0033
(0.085) (0.676) (0.993) (0.869)
LIBOR3MLag1 -0.1209 -0.0424 -0.0445 -0.0455
(0.002) (0.083) (0.018) (0.018)
CPILag1 0.0093 0.0229 0.0132 0.0149
(0.413) (0.027) (0.343) (0.294)
GDPGrowthLag1 0.0272 0.0207 0.0172 0.0161
(0.015) (0.017) (0.060) (0.083)
DummyQ1 -0.0105 0.0152 0.0163 0.0170
(0.581) (0.427) (0.642) (0.628)
DummyQ2 0.0027 0.0207 0.0100 0.0104
(0.898) (0.294) (0.765) (0.758)
DummyQ3 0.0293 0.0478 0.0475 0.0470
(0.097) (0.009) (0.159) (0.168)
Constant 1.0782 0.9012 0.6552 1.3534
(0.001) (0.048) (0.013) (0.073)
Notes: parameters significant at 5% level are reported in bold. Parameters significant at 10%
level are reported in italics. Critical significance levels are reported below the parameters.
Source: own calculations
Our supplementary model hints that the relationship is negative: housing loan
books grow faster in less liquid banks. The lack of clear evidence may be
due to the fact that banks met very few restrictions when accessing foreign
funding, especially on an intra-group basis. The other reason for slower housing
loan portfolio growth in banks with the positive interbank gap could be their
business strategy, not only concentrated on the loan market but balanced by
more secure investment opportunities on the interbank market.
Such reliance of some Polish banks on intra-group funding and capital in-
jections from the parent can disturb the transmission of monetary policy and
provoke contagion from the home markets of parent banks to the Polish loan
market. Under such circumstances, and if capital and liquidity constraints
for Polish banks were binding, credit conditions with regard to housing loans
would depend on ability and willingness of foreign parent institutions to pro-
vide capital and funds to their Polish subsidiaries. This, in turn, is linked to
the capital position of a parent institution. As a result, Polish banks may be
forced to curb lending in case the financial condition of the parent company
deteriorates, even without any intrinsic reasons. Conversely, if capital and liq-
uidity constraints are not binding for Polish banks, they may not respond
to monetary impulses in an expected manner. A fall in deposits, induced by
14
Table 3 Model of growth of housing loans normalised by assets
Prais-Winsten
(PSAR1)
Prais-Winsten
(AR1)
RE FE
StdTighteningLag2 -0.0009 -0.0010 -0.0045 -0.0049
(0.380) (0.366) (0.031) (0.020)
StdEasingLag2 0.0007 0.0009 0.0040 0.0042
(0.421) (0.288) (0.025) (0.022)
TermsLevelLag2 0.0002 0.0002 0.0004 0.0005
(0.297) (0.372) (0.009) (0.004)
CARLag1 -0.0357 -0.0353 -0.0533 -0.0557
(0.166) (0.210) (0.026) (0.024)
GapIBankLag1 -0.0027 -0.0043 -0.0269 -0.0200
(0.692) (0.505) (0.004) (0.074)
LogAccountsLag1 0.0005 -0.0013 0.0005 0.0039
(0.700) (0.234) (0.699) (0.198)
WIBOR3MLag1 -0.0009 -0.0015 -0.0001 -0.0001
(0.308) (0.142) (0.919) (0.921)
LIBOR3MLag1 0.0030 0.0037 0.0026 0.0024
(0.014) (0.011) (0.020) (0.033)
CPILag1 -0.0002 -0.0002 -0.0028 -0.0026
(0.746) (0.656) (0.001) (0.001)
GDPGrowthLag1 0.0008 0.0009 0.0009 0.0010
(0.089) (0.067) (0.076) (0.071)
DummyQ1 -0.0025 -0.0023 -0.0033 -0.0033
(0.006) (0.018) (0.099) (0.103)
DummyQ2 0.0017 0.0015 -0.0002 -0.0001
(0.094) (0.154) (0.912) (0.993)
DummyQ3 0.0021 0.0019 0.0004 0.0005
(0.021) (0.052) (0.837) (0.813)
Constant 0.0009 0.0296 0.0083 -0.0377
(0.960) (0.086) (0.704) (0.383)
Notes: parameters significant at 5% level are reported in bold. Parameters significant at 10%
level are reported in italics. Critical significance levels are reported below the parameters.
Source: own calculations
Table 4 Summary of statistical tests
Main models Supplementary models
Test Housing
loans
Corporate
loans
Consumer
loans
Housing
loans
Corporate
loans
Consumer
loans
Breusch-Pagan test for
random effects
8.58 10.76 19.99 446.56 94.52 2.75
0.0034 0.0010 0.0000 0.0000 0.0000 0.0973
Hausman test of fixed vs.
random effects
4.86 6.83 24.29 0.72 35.78 11.50
0.9932 0.6659 0.0833 0.9999 0.0019 0.7773
F-test for poolability of the
data (H0: no individual ef-
fects)
2.66 4.76 4.17 17.98 7.84 2.62
0.0022 0.0000 0.0000 0.0000 0.0000 0.0013
Panel-specific serial corre-
lation
22.27 3.16 78.69 206.82 2.11 25.45
0.0000 0.0753 0.0000 0.0000 0.1465 0.0000
Joint LM test of random ef-
fects and serial correlation
24.77 26.58 82.24 517.13 101.94 25.47
0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Modified Wald test for
groupwise heteroscedastic-
ity
11756.8 1257.4 846.4 998.2 1115.2 673.3
0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
Note: p-values reported in italics.
Source: own calculations