The real effects of relationship lending
Ryan Banerjee (BIS)
Leonardo Gambacorta (BIS & CEPR)
Enrico Sette (Bank of Italy)
Banque de France
9 June 2017
The views expressed are the authors’ only and not necessarily those of the BIS,
the Bank of Italy or the Eurosystem
Motivation (1)
 Understand factors favoring the resilience of
economies during and after a crisis
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
2
Motivation (2)
 Focus on extent to which banks rely on Relationship
Lending (RL)
 RL=lending technology based on acquisition of soft
information about the borrower, through repeated /
close interaction. Long term implicit contract.
 RL helps access to finance for small / opaque firms
(Degryse et al. 2009 for a review).
 RL contributed to soften the transmission of shocks
(Bolton et al. 2016; Sette and Gobbi, 2015, Beck et al.
2015; Gambacorta and Mistrulli, 2015)
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
3
Motivation (3)
 Is RL a good banking technology when also banks are
under stress? What are the real effects of higher reliance
on RL?
 Effect of RL seems to depend on banks’ balance sheet
strength (Bolton et al. 2014; Sette and Gobbi, 2015;
Gambacorta and Mistrulli, 2015)
 Not clear that RL really helps if the crisis is protracted
/ systemic
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 28.2.2017
4
Motivation (4)
 Debate on change in bank business models and
diversification of financing mix of firms
 Extensive use of credit scoring by banks: more
importance to hard (quantitative) vs soft (qualitative)
information
 Diminish importance of bank funding for firms, but
possible trade-off if RL provides insurance during
crises
5
Motivation (5)
 In this paper: do firms that rely more on relationship
lending experience higher investment and
employment growth during the crisis?
 Test for impact on credit quantity and its cost
 Distinguish different types of loans
 Exploit detailed micro data for identification and
measurement of RL
 Test for heterogeneous effects
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
6
Related Literature
 Relationship lending in good and bad times
(Sette and Gobbi JEEA 2015, Gobbi and Sette RF 2015, Bolton et al. RFS 2016,
Beck et al. 2016; Gambacorta and Mistrulli, 2015)
 Real effects of credit shocks
(during crises: Chodorow-Reich QJE 2014, Cingano, Manaresi, Sette, RFS 2016,
Bentolila et al. 2015, Acharya et al. 2015, Amiti, Weinstein 2017 JPE forth.)
 Novelties of this paper:
 impact on investment and employment
 transmission mechanism: different types of credit
(working capital loans versus term loans)
 focus on different crisis periods
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
7
Empirical Strategy (1)
1. Show that RL leads to better access to credit,
distinguish between the 2 phases of the crisis

2008-2010: global financial crisis but Italian banks
heterogeneously affected (Panetta et al, 2010)

2011-2014: sovereign debt crisis
2. Test whether firms’ RL intensity has an effect on
investment and employment
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
8
Empirical Strategy (2)
Need to tackle possible endogeneity of RL:
Firms relying more on RL may be different along
several dimensions
 Point 1 (RL ensures better access to credit): rely on data at the
bank-firm relationship-level. Control for firm observable and
unobservable characteristics (time-varying).
 Point 2 (Real effects of RL):
 Control for observables and firm-fixed effects.
 Show balancing of covariates
 IV
 Several robustness
9
Data
 3 main datasets
1. Credit Register: includes all exposures of a banks towards an
individual if these are above 30k euros (75k euros until 2008)
Distinguish 3 types of loans: revolving credit lines (overdraft), term
loans (mortgages, leasing), loans backed by receivables
For a representative sample of banks includes information on
interest rates
2. Cerved (firm register): balance sheet information of
incorporated firms
3. Supervisory reports: bank balance sheet data, useful for
heterogeneous effects
10
Data
 Merge Credit Register, Firm Register (CERVED) using
unique firm tax identifier and Supervisory reports for
bank characteristics using unique bank identifier
 Select random sample of 10% of CR firms
 Data span 2004-2014
 Non-financial firms
 Multiple bank relationships
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
11
Descriptive stats I
Table 1: Descriptive Statistics: Bank-firm level
Mean
Median
Standard deviation
25th percentile
75th percentile
No. of observations
∆Log
(Total
credit)
-0.490
0
36.08
-11.84
7.891
314649
∆Log
(Revolving
credit lines)
0.281
0
43.44
0
0
278883
∆Log
(Term
loans)
-8.771
-9.245
67.97
-37.94
11.57
169803
Interest rate
on
revolving
credit lines
10.89
9.235
7.609
6.939
12.57
204030
Interest rate
on term
loans
4.459
4.377
1.805
3.010
5.789
136484
Relationship
duration (in
years)
5.740
6
3.625
3
8
314649
Log
(relationship Log credit
duration)
granted
1.704
12.95
1.946
12.77
0.717
1.091
1.386
12.13
2.197
13.59
314649
314649
Drawn
credit
/credit
granted
56.78
60.09
32.42
30.81
85.11
314649
Share of
revolving
credit lines
used
23.90
13.22
27.66
5.245
31.03
314649
12
Descriptive stats II
∆Log
(Total credit)
2.019
Investment
Rate (growth
rate of fixed
assets)
17.98
Intangible
investment/ Total
investment
19.31
∆Log
(Employment
costs)
4.039
Credit weighted
log relationship
duration in 2006
1.400
Credit weighted
log relationship
duration
1.461
Return on
assets
0.550
Leverage (Debt
/ total assets)
80.26
EBITDA/
Interest
expense
6.717
Log (total
assets)
8.042
0
-0.775
3.3
3.574
1.479
1.512
0.432
84.33
3.354
7.885
31.14
75.82
71.91
24.45
0.511
0.583
4.923
16.59
13.92
1.264
-12.49
-9.442
30.43
-5.082
1.062
1.060
-0.281
70.97
1.652
7.149
16.70
15.62
0
12.41
1.811
1.900
1.869
93.11
6.972
8.788
82692
82314
68064
79420
65398
82692
80551
82633
82294
82689
Mean
Median
Standard deviation
25th percentile
75th percentile
No. of observations
13
Results on credit at the bank-firm relationship level
Use Khwaja-Mian-type identification (2008, AER)
∆𝑌𝑖,𝑗,𝑡 = 𝑅𝐿𝑖,𝑗,𝑡 + 𝑅𝐿𝑖,𝑗,𝑡 *D(Crisis 1)+𝑅𝐿𝑖,𝑗,𝑡 *D(Crisis 2)+b X+𝛾𝑖,𝑡 + 𝜀𝑖,𝑗,𝑡
where Y is ∆ (log credit) or interest rate, X vector of controls,
𝛾 fixed effects
 RL measure of relationship lending is (log) length of the
relationship (standard in the literature)
 Potentially endogenous, so important to control for firm
time*varying unobservables
 compare credit by different banks to the same firm at
the same time
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
14
Effects of relationship banking on lending
Relationship durationt-1
(1)
∆Log
(Total
credit)
(2)
∆Log
(Total
credit)
0.493**
(0.200)
-0.245
(0.292)
1.111***
(0.348)
-0.208
(0.341)
-14.33***
(0.427)
0.0453***
(0.00500)
0.0534***
(0.00487)
Yes
Yes
314649
0.401
Relationship duration t-1*D(Post 2008)
Relationship duration t-1*D(Post 2011)
Log credit granted t-1
Drawn/grantedt-1
Share revolving credit lines t-1
Bank*Time fixed effects
Firm*Time fixed effects
Observations
R-squared
-14.33***
(0.427)
0.0452***
(0.00500)
0.0534***
(0.00485)
Yes
Yes
314649
0.401
(3)
(4)
∆Log
∆Log
(Revolving (Revolving
credit lines) credit lines)
1.189***
(0.195)
-13.03***
(0.605)
0.0991***
(0.00850)
-0.610***
(0.0274)
Yes
Yes
268953
0.382
0.702**
(0.306)
0.906**
(0.429)
-0.489
(0.337)
-13.03***
(0.606)
0.0992***
(0.00850)
-0.610***
(0.0274)
Yes
Yes
268953
0.382
(5)
∆Log
(Term
loans)
(6)
∆Log
(Term
loans)
0.151
(0.336)
-0.549
(0.823)
1.038
(1.087)
-0.215
(0.944)
-9.018***
(0.800)
0.00262
(0.0315)
0.416***
(0.0231)
Yes
Yes
138698
0.397
-9.018***
(0.799)
0.00247
(0.0315)
0.416***
(0.0230)
Yes
Yes
138698
0.397
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
15
Effects of relationship banking on interest rates
(1)
(2)
Interest rate on Interest rate on
revolving credit revolving credit
lines
lines
Relationship durationt-1
0.913***
(0.0742)
Relationship duration t-1*D(Post 2008)
Relationship duration t-1*D(Post 2011)
Log credit granted t-1
Drawn/grantedt-1
Share revolving credit lines t-1
Bank*Time fixed effects
Firm*Time fixed effects
Observations
R-squared
-0.742***
(0.0442)
0.00397***
(0.00100)
-0.0359***
(0.00291)
Yes
Yes
219763
0.555
1.180***
(0.0917)
-0.322***
(0.0821)
-0.0968
(0.0861)
-0.742***
(0.0441)
0.00392***
(0.00101)
-0.0359***
(0.00292)
Yes
Yes
219763
0.555
(3)
Interest rate on
term loans
(4)
Interest rate on
term loans
0.0143
(0.0110)
0.0830***
(0.0135)
-0.0586***
(0.0192)
-0.0688***
(0.0222)
-0.185***
(0.0135)
-0.00286***
(0.000297)
0.00313***
(0.000347)
Yes
Yes
125791
0.719
-0.185***
(0.0134)
-0.00284***
(0.000299)
0.00314***
(0.000346)
Yes
Yes
125791
0.719
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
16
Extensions & Robustness
 Heterogeneous effects by firm and by bank
 Effect of RL is stronger if banks have higher capital
 Effect of RL has limited heterogeneity across firm
characteristics (leverage and profitability)
 Robustness:
 Include interactions of dummy crisis with all controls
 Subsample of relationships which are more intensely used
(drawn/granted > 50%)
 Control for the presence of past-due loans
17
Real effects
 Get to the firm level
 Construct credit-weighted average length of relationships
 We estimate the following model
𝑦𝑖,𝑡 = 𝛽𝑅𝑒𝑙𝐿𝑒𝑛𝑑𝑖,𝑡 + 𝛾𝐹𝑖𝑟𝑚 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡 + 𝛿𝑡 + 𝜃𝑖
 3 main dep vars: credit growth, investment rate, change in labor cost
 RelLend is credit-weighted average log duration of relationships (weights
are share of credit of each banks).
 Firm-level controls include: leverage, roa, ebitda/interest expenses, firm
size, z-score (prob default)
 Include firm fixed effects and time fixed effects
18
Real effects
 Problem: potentially endogenous
 Test for sorting
 Fix RL at 2006 (before the crisis) and add interactions with
crisis dummies + firm fe
 Use IV: instrument is the difference between average length
and the average length of relationships with banks involved in
M&As in 2006 (Hong and Kacperckyk, 2009)
 Intuition: Change in average length of relationship, conditional
on firm FE and firm time varying controls uncorrelated with
firm unobservables
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
19
A test for the presence of sorting in bank-firm relationship
Table 7: A test for the presence of sorting in bank-firm relationships
1st
Quartile
Leverage
(Total debt/Total assets)
Return on assets
EBITDA/Value added
Zscore
∆Log (Total credit)
Investment rate
∆Log (Labour costs)
85.14
(0.32)
0.53
(0.06)
40.57
(0.14)
5.71
(0.28)
15.58
(0.4)
39.9
(0.17)
14.8
(0.4)
2nd
Quartile
83.15
(0.20)
0.60
(0.07)
36.88
(0.07)
5.42
(0.1)
9.53
(0.24)
28.69
(-0.01)
8.34
(0.18)
3rd
Quartile
81.66
(0.11)
0.72
(0.10)
35.98
(0.07)
5.24
(-0.01)
7.04
(0.17)
24.64
(-0.08)
6.93
(0.13)
4th
Quartile
Standard
deviation
80.34
(0.03)
0.75
(0.10)
34.26
(0.02)
5.13
(-0.08)
5.27
(0.11)
23.73
(-0.09)
5.17
(0.05)
15.74
5.19
48.92
1.63
32.21
61.36
24.23
Standardized difference (Imbens-Wooldridge 2009) in parentheses
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
20
Real effects of relationship lending at the firm level
∆Log (Total credit)
VARIABLES
Average interest rate on
total credit
Investment Rate
(2)
(3)
(4)
(5)
Weighted relationship
duration*D(Post 2008)
4.435***
4.260***
-26.02***
-27.01***
3.893***
(0.661)
(0.661)
(2.515)
(2.490)
(1.267)
(1.282)
(1.690)
(1.674)
(0.514)
(0.507)
Weighted relationship
duration*D(Post 2011)
1.158*
1.011*
-10.40***
-10.75***
1.811
2.015*
12.50***
13.57***
0.457
0.531
(0.616)
(0.611)
(2.526)
(2.519)
(1.172)
(1.163)
(1.568)
(1.564)
(0.499)
(0.497)
Firm leverage
EBITDA/interest expenses
Log (firm total assets)
Z-Score
(7)
(8)
∆Log (Labour costs)
(1)
Return on assets
(6)
Intangible investment/
Total investment
(9)
4.440*** -8.222*** -8.989*** 4.314***
(10)
4.289***
0.301***
-0.547***
0.245***
-1.371***
0.455***
(0.0435)
(0.153)
(0.0817)
(0.0998)
(0.0351)
-0.135***
0.459***
-0.0910**
-1.227***
(0.0193)
(0.0869)
(0.0388)
(0.0588)
0.0515***
(0.0160)
0.149***
-0.375***
0.280***
-0.102***
0.0395***
(0.0191)
(0.0532)
(0.0358)
(0.0310)
(0.0104)
-10.73***
7.131***
-28.98***
21.28***
-4.436***
(0.628)
(2.635)
(1.357)
(1.500)
(0.506)
-2.695***
4.345**
-1.872**
-1.186
-0.229
(0.451)
(1.723)
(0.854)
(1.168)
(0.360)
Time fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Firm fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
62995
62797
62995
62797
62837
62644
35013
61110
61110
60987
R-squared
0.194
0.194
0.653
0.653
0.245
0.244
0.401
0.274
0.275
0.274
21
Instrumental variable estimation
(1)
(2)
(3)
(4)
(5)
∆Log
(Total credit)
Average interest
rate on total
credit
Investment rate
Intangible
investment/
Total
investment
∆Log (Labour
costs)
8.914***
-8.999
8.920***
-5.143
6.430***
(1.551)
(7.686)
(2.874)
(4.886)
(1.166)
-1.780
-14.97**
2.646
13.78***
0.613
(1.364)
(6.677)
(2.532)
(4.109)
(1.069)
Time fixed effects
Yes
Yes
Yes
Yes
Yes
Firm fixed effects
Yes
Yes
Yes
Yes
Yes
Observations
62797
62797
62644
34912
60987
R-squared
0.193
0.652
0.244
0.398
0.274
Kleibergen-Paap weak
identification F-statistic
743.39
743.39
738.58
534.15
703.22
Weighted relationship
duration*D(Post 2008)
Weighted relationship
duration*D(Post 2011)
Intuition of instrument: M&A destroy some relationships and this changes
the average duration. Change exogenous conditional on firm fixed
effects.
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
22
IV: supporting the validity of the instrument
Table 9: A test for the presence of sorting in firm relationships with banks that merged in 2006
Leverage
(Total debt/Total assets)
Return on assets
EBITDA/Value added
Zscore
1st
Quartile
79.32
2nd
Quartile
81.92
3rd
Quartile
84.08
4th
Quartile
84.97
(-0.03)
0.82
(0.11)
35.33
(0.04)
5.02
(-0.15)
(0.12)
0.78
(0.11)
33.96
(0.02)
5.23
(-0.02)
(0.26)
0.64
(0.08)
36.6
(0.07)
5.51
(0.16)
(0.30)
0.36
(0.03)
41.81
(0.16)
5.75
(0.30)
Standard
deviation
15.74
5.19
48.92
1.63
23
Other results
 Robustness checks:
 year by year regressions
 weighted regs by value added (aggregate effects)
 Firm heterogeneity: some evidence that highly-leveraged
and highly-profitable firms receive more insulation effects
on credit, investment and employment. Results on highly
leveraged firms are weaker.
 Bank heterogeneity: well-capitalised RL banks protect more
their clients
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
24
Alternative interpretation? Loan evergreening?
 Could this be evidence of «loan evergreening»? Banks
captured by firms keep lending to avoid that these
default.
 Boundary between RL and evergreening may be fuzzy.
 Yet:
 1. results on bank heterogeneity: RL stronger if banks
more capitalized. Stronger incentives to loan
evergreening for low capital banks
 2. limited evidence on firm heterogeneity, but stronger
incentives to evergreen loans to weaker firms
25
Conclusion - Main points to take away
1. Relationship lending ensured firms a steadier flow of
credit during both crises
2. Firms more reliant on RL invest and increase
employment (relatively) more than other firms during
both crises
3. Insulation effects of RL remained pretty stable in the
2nd phase of the financial crisis (sovereign debt crisis)
4. Bank capital influences the effect of relationship
lending in affecting credit supply, investment and
employment
5. Some evidence of heterogeneous effects across firms
in firm-level regressions.
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
26
Thanks for your attention!
Enrico Sette
Bank of Italy
Email: [email protected]
my research page:
https://sites.google.com/site/settenrico/home/research
“The real effects of relationship lending” –
Banerjee, Gambacorta and Sette – 09.06.17
27