12th YOUNG ECONOMISTS’ SEMINAR
23rd DUBROVNIK ECONOMIC CONFERENCE
Dubrovnik, 4-6 June 2017
The Impact of Bank Shocks on
Firm-Level Outcomes and Bank Risk-Taking
Hans Degryse (KU Leuven, IWH Halle and CEPR)
Olivier De Jonghe (Tilburg University and NBB)
Sanja Jakovljević (CNB and KU Leuven)
Klaas Mulier (Ghent University)
Glenn Schepens (ECB)
Motivation
Finance is an important driver of economic growth.
E.g., King and Levine (1993), Guiso, Sapienza and Zingales (2004), Beck, Demirgüç-Kunt,
Laeven and Levine (2008), Rajan and Zingales (1998).
Banks are important providers of external finance to firms in general and
SMEs in particular.
Bank-loan supply shocks may hamper the availability of credit and impact
the real economy.
Main identification challenge: separating the firm-borrowing and the banklending channels.
E.g., literature in the area of monetary policy; banking; corporate finance.
1/22
Motivation (2)
Research questions and main findings
1. Can the identification of bank supply shocks be adjusted to include the many
single-relationship firms? Does it matter?
It can, and it does: the ranking of banks according to the magnitude of the shock
varies, and the real impact of the recent financial crisis is better captured.
2. Are bank-loan supply shocks only relevant in crisis periods?
The dispersion of bank supply shock estimates is also high in non-crisis periods.
3. Do bank-loan supply shocks impact firm performance and bank risk-taking?
Negative bank supply shocks result in lower growth and investment of firms, and
in risk-mitigating behavior by banks (and vice versa).
2/22
Identification of firm-borrowing and bank-lending
channels
Two methodological choices may limit the generality of conclusions on the
impacts of bank shocks:
1.
Identification based on exogenous shocks to loan supply
o drops in asset prices and real estate exposures of banks (Gan, 2007)
o nuclear tests and the collapse of the dollar deposit market (Khwaja and Mian, 2008)
𝐿𝑓𝑏𝑡 − 𝐿𝑓𝑏𝑡−1
≡ ∆𝐿𝑓𝑏𝑡 = 𝛼𝑓 + 𝛿 ∙ 𝐵𝑏 + 𝜀𝑏𝑓
𝐿𝑓𝑏𝑡−1
Does not say much on the behaviour of credit supply over prolonged periods
of time.
3/22
Identification of firm-borrowing and bank-lending
channels (2)
Two methodological choices may limit the generality of conclusions on the
impacts of bank shocks:
2.
Direct identification of loan supply shocks, on a sample of firms
borrowing from multiple banks
𝐿𝑓𝑏𝑡 − 𝐿𝑓𝑏𝑡−1
≡ ∆𝐿𝑓𝑏𝑡 = 𝛼𝑓𝑡 + 𝛽𝑏𝑡 + 𝜀𝑓𝑏𝑡
𝐿𝑓𝑏𝑡−1
Unusable in samples with plentiful single-bank firms.
(Ongena and Smith, 2000; Degryse, Kim and Ongena, 2009)
Single-bank and multiple-bank firms might not be similar.
4/22
Our methodology
We address the methodological challenges by
1.
Developing an indicator of bank-loan supply shocks which captures their
cross-sectional variation, over an extended time period
2.
Including firms borrowing from just one bank
3.
Considering the weighting structure suggested by Amiti and Weinstein
(2016)
Adding-up constraints: a bank cannot lend more without at least one firm
borrowing more and vice versa  Amiti and Weinstein (2016) suggest to
use weighted least squares
5/22
Our methodology (2)
We provide an alternative demand control which encompasses the vast
majority of firms.
Firm-time fixed effects are replaced by industry-location-size-time fixed effects
(2-digit NACE codes; 2-digit postal codes; deciles by total assets)
∆𝐿𝑓𝑏𝑡 = 𝛼𝑓𝑡 + 𝛽𝑏𝑡 + 𝜀𝑓𝑏𝑡
 ∆𝐿𝑓𝑏𝑡 = 𝛼𝑰𝑳𝑺𝑡 + 𝛽𝑏𝑡 + 𝜀𝑓𝑏𝑡
𝛼1𝑡 = 𝛼2𝑡 = ⋯ = 𝛼𝐹𝑡 , (1. . 𝐹) ∈ 𝐼𝐿𝑆
6/22
Our data
We use the following datasets of the NBB:
o Monthly bank-firm information on authorized credit to firms incorporated in
Belgium (reporting threshold: 25,000 EUR)
o Annual accounts of Belgian firms
o Monthly bank balance sheet information
Time coverage: 2002m1 – 2012m3
Estimation sample: around 17 mil. bank-firm time observations
7/22
Our data (2)
How relevant are firms borrowing from just one bank?
Figure 1. The number of firm borrowing relationships and their share in total loan volume
8/22
Our data (3)
Are single-bank and multiple-bank firms similar?
Table 1. Characteristics of single-bank and multiple-bank firms
Mean
Firm-bank-month
observations
Firms
T-stat (p-value) of
difference in means
Age (in years)
Single-bank firms
Multiple-bank firms
12.70
24.01
9,705,534
1,367,672
183,885
5,752
166.36 (0.000)
Total assets (in mil. EUR)
Single-bank firms
Multiple-bank firms
1.74
29.44
9,705,534
1,367,672
183,885
5,752
10.98 (0.000)
Number of employees, FTE
Single-bank firms
Multiple-bank firms
4.17
57.67
9,705,534
1,367,672
183,885
5,752
18.30 (0.000)
Fixed assets/total assets
Single-bank firms
Multiple-bank firms
0.52
0.36
9,705,534
1,367,672
183,885
5,752
-132.69 (0.000)
Loan size (in mil. EUR)
Single-bank firms
Multiple-bank firms
0.30
1.33
9,705,534
1,367,672
183,885
5,752
32.77 (0.000)
Multiple-bank firms are on average older, larger, have lower investment ratios and
borrow larger credit amounts.
9/22
Our data (4)
What do we gain by considering the ILS grouping?
Figure 2. The number of ILS borrowing relationships and their share in total loan volume
10/22
Our data (5)
What do we gain by considering the ILS grouping?
Figure 3. The average credit growth rate
11/22
Estimation and verification of bank-loan supply
shocks: multiple-bank firm sample
Within the multiple-bank firm sample, we estimate the following set of
equations:
𝑖
∆𝐿𝑓𝑏𝑡 = 𝛼𝑖𝑡 + 𝛽𝑏𝑡
+ 𝜀𝑓𝑏𝑡 ,
𝑖 =∙, 𝐿, 𝐼𝐿, 𝐼𝐿𝐴, 𝐼𝐿𝐶, 𝐼𝐿𝑅, 𝐼𝐿𝑆, 𝐹
Demand controls: location,
industry,
age,
current assets,
risk (interest coverage ratio; debt; Altman Z),
size.
12/22
Estimation and verification of bank-loan supply
shocks: multiple-bank firm sample (2)
We then compare the bank shocks from various demand specifications with
the “standard” bank shocks:
𝑖
𝐹
𝛽𝑏𝑡
= 𝛿 ∙ 𝛽𝑏𝑡
+ 𝜇𝑡 + 𝜀𝑏𝑡 ,
𝑖 =∙, 𝐿, 𝐼𝐿, 𝐼𝐿𝐴, 𝐼𝐿𝐶, 𝐼𝐿𝑅, 𝐼𝐿𝑆, 𝐹
Table 2. Multiple-bank firm sample: comparison of credit demand controls
𝛽𝑏𝑡
𝐹
𝛽𝑏𝑡
𝛽𝑏𝑡
𝛽𝑏𝑡
𝛽𝑏𝑡
𝛽𝑏𝑡
𝛽𝑏𝑡
𝛽𝑏𝑡
𝛽𝑏𝑡
𝛽𝑏𝑡
𝛽𝑏𝑡
1.001*** 0.998*** 1.009*** 1.048*** 1.023*** 0.993*** 1.039*** 1.046*** 1.022*** 1.066***
p-value coef.=1
0.983
0.951
0.671
0.068
0.210
0.680
0.083
0.085
0.124
0.135
Adjusted R²
0.725
0.729
0.769
0.765
0.758
0.742
0.784
0.735
0.791
0.837
Spearman's r.c.c.
0.747
0.750
0.789
0.788
0.798
0.783
0.799
0.793
0.813
0.851
ILS provides the best fit;
The model performs equally well during crisis periods.
13/22
Estimation and verification of bank-loan supply
shocks: multiple-bank ILS sample
As we extend the sample towards multiple-bank ILS groups, our bank
shock estimates are departing from the “standard” bank shocks.
Table 3. Multiple-bank firm setup vs. multiple-bank ILS setup
0.752***
0.576***
2.66e-10
1.34e-09
Adjusted R²
0.787
0.632
Spearman's r.c.c.
0.778
0.669
p-value coef.=1
The ranking of banks based on the size of the shocks depends on the sample used.
We can meaningfully relate our bank shock estimates to:
o Tightening of lending standards (Bank Lending Survey)
o Growth in interbank liabilities
14/22
Variation of bank-loan supply shocks
Fluctuation of bank shocks might be high also in non-crisis periods!
Figure 4. Dispersion of estimated bank shocks (p75-p25)
15/22
How (in)efficient are Khwaja and Mian (2008)
estimates in our sample?
Not so much: suggestive of limited lending asymmetry.
Figure 5. The actual and predicted bank credit growth rates using KM(2008)
16/22
Application of bank-loan supply shocks
We analyse how bank-loan shocks relate to
1.
Firm-level outcomes
Growth (growth in total assets)
Investment (growth in fixed assets)
Employment (growth in number of FTE employees)
∆𝑌𝑓𝑡 = 𝛿 ∙
(𝐾𝑀, 𝑊)
𝑏
𝜃𝑓𝑏𝑡−1 𝛽𝑏𝑡−1
17/22
+ 𝜇𝑡 + 𝛾𝑓 + 𝜀𝑓𝑡
Firm-level outcomes
Table 4. Bank credit supply estimates and firm-level outcomes
Khwaja and Mian (2008) approach
Multiple-bank ILS sample
ILST shock
FT shock
Multiple-bank firm sample
ILST shock
FT shock
Amiti and Weinstein (2016) approach
Growth
Investment
Employment
Growth
Investment
Employment
0.0327**
(0.0128)
0.0367***
(0.0101)
0.0941**
(0.0395)
0.122***
(0.0311)
-0.0209
(0.0240)
-0.0138
(0.0187)
0.0973***
(0.0132)
0.0456***
(0.00757)
0.179***
(0.0408)
0.0820***
(0.0234)
0.0429*
(0.0245)
0.0357***
(0.0136)
0.0287
(0.0240)
-0.00316
(0.0215)
0.0830
(0.0662)
0.00983
(0.0592)
0.0139
(0.0348)
0.0147
(0.0312)
0.105***
(0.0223)
0.0434**
(0.0171)
0.137**
(0.0614)
0.0420
(0.0473)
0.103***
(0.0320)
0.0775***
(0.0243)
The weighting structure matters.
In the multiple-bank ILS sample, both methodologies work well in identifying shocks
to growth and investment, albeit of somewhat differing economic magnitudes.
18/22
Firm-level outcomes (2)
Table 5. Bank credit supply estimates, firm growth and investment
Khwaja and Mian (2008) approach
Growth
ILST shock
FT shock * Crisis
Investment
Growth
Investment
0.0327**
(0.0128)
0.0273**
(0.0129)
0.112***
(0.0382)
0.0941**
(0.0395)
0.0862**
(0.0399)
0.165
(0.118)
0.0973***
(0.0132)
0.0788***
(0.0135)
0.263***
(0.0414)
0.179***
(0.0408)
0.147***
(0.0418)
0.459***
(0.128)
0.0367***
(0.0101)
0.0383***
(0.0103)
-0.0233
(0.0295)
0.122***
(0.0311)
0.129***
(0.0318)
-0.0932
(0.0910)
0.0456***
(0.00757)
0.0468***
(0.00783)
-0.0143
(0.0235)
0.0820***
(0.0234)
0.0853***
(0.0242)
-0.0395
(0.0725)
ILST shock * Crisis
FT shock
Amiti and Weinstein (2016) approach
When the crisis effects are considered in the multiple-bank ILS sample,
our methodology captures them better than the “standard” one.
Same conclusion holds for smaller and more indebted firms.
19/22
Application of bank-loan supply shocks (2)
We analyse how bank-loan shocks relate to
2.
Bank risk-taking (at the extensive margin)
Portfolio-weighted average of firm-level Altman Z score
Loan share
𝑍𝑏𝑡
𝑒𝑥𝑡
− 𝑍𝑏𝑡−1
𝐿𝑒𝑥𝑡
𝑏𝑡 /
𝑖𝑛𝑡
(𝐾𝑀, 𝑊)
+ 𝜇𝑡 + 𝜔𝑏 + 𝜀𝑏𝑡
(𝐾𝑀, 𝑊)
+ 𝜇𝑡 + 𝜔𝑏 + 𝜀𝑏𝑡
= 𝛿 ∙ 𝛽𝑏𝑡−1
𝐿𝑖𝑛𝑡
𝑏𝑡−1 = 𝛿 ∙ 𝛽𝑏𝑡−1
20/22
Bank risk-taking
Table 6. Bank-loan supply estimates and bank risk-taking
Khwaja and Mian (2008) approach
Entries
Altman Z score
Share
Exits
Altman Z score
Share
Amiti and Weinstein (2016) approach
Full period
2002m1 2008m9
2008m10 2012m3
Full period
2002m1 2008m9
2008m10 2012m3
-0.00140**
(0.000614)
-0.00109*
(0.000659)
-0.00394**
(0.00155)
-0.00323***
(0.000775)
-0.00291***
(0.000797)
-0.00340*
(0.00186)
0.456***
(0.0571)
0.514***
(0.0662)
0.0903**
(0.0414)
0.281***
(0.0370)
0.320***
(0.0420)
0.0137
(0.0317)
-0.00287***
(0.000855)
-0.00298***
(0.000953)
-0.00278**
(0.00136)
-0.00560***
(0.00128)
-0.00622***
(0.00143)
-0.000419
(0.00161)
-0.308***
(0.0596)
-0.133***
(0.0391)
-1.656***
(0.535)
-0.334***
(0.0574)
-0.194***
(0.0490)
-1.327***
(0.401)
New firms in the portfolio more risky and are taken on faster;
riskier firms also dropped from the portfolio, but more slowly:
risk-taking behaviour by banks with more positive supply shocks.
21/22
Conclusion
We develop a methodology to identify bank supply shocks in the presence of
a multitude of single-borrowing firms.
Firms borrowing from lenders with a more negative bank-loan supply shock
have lower growth, investment and (to some extent) employment rates,
and vice versa.
Banks with more positive supply shocks take on more risk, and vice versa.
22/22
Thank you for your attention!