Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Bank Leverage and Monetary Policy’s
Risk-Taking Channel:
Evidence from the United States
by
Giovanni Dell’Ariccia (IMF and CEPR)
Luc Laeven (IMF and CEPR)
Gustavo Suarez (Federal Reserve Board)
CSEF – Unicredit Conference, Naples, March 21, 2014
The views in this presentation do not necessarily reflect those of the IMF, IMF Board, Federal Reserve System, or its Board of Governors
Paper summary
Ask whether bank extend riskier loans when monetary policy’s
stance is easier
Employs loan-level confidential Fed dataset
Allows for ex-ante measure of loan riskiness
First to use disaggregated commercial bank data for US
Grounded in basic theoretical model
Look at how bank capitalization affects bank risk taking incentives
when monetary policy changes
Motivation
Many see easy monetary policy conditions in the 2000s as
major factor behind the crisis (Borio and Zhu, Adrian and Shin,
2009, Taylor, 2009)
Renewed debate: are low interest rates setting the stage for
future crises? (e.g., Rajan (2010))
Interest rate policy affects the quality and not just the quantity
of credit
Risk-taking channel of monetary policy (Borio/BIS)
Before crisis … Macro looked OK
-3
-2
-1
0
1
2
2000 02 04 06 08:
Q4
Output Gap2Core CPI Inflation
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
2000 02 04 06 08:
Q4
Euro area United States Average of other economies1
1 Japan omitted.2 Estimate of output gap using rolling Hodrick-Prescott filter.
9
Figure 8. Credit Growth and Monetary Policy(Selected countries that had a boom in the run -up and a crisis in 2007-08)
Sources: IMF International Financial Statistics, World Economic Outlook; staff calculations.Notes: Credit is indexed with a base value of 100 five years prior to the crisis.
0
50
100
150
200
250
0
1
2
3
4
T-5 T-4 T-3 T-2 T-1 T
United Kingdom 2007
Core inf lation
Credit (right axis)0
50
100
150
200
250
0
1
2
3
4
T-5 T-4 T-3 T-2 T-1 T
Ireland 2008
Core inf lation
Credit (right axis)
0
50
100
150
200
250
0
1
2
3
4
T-5 T-4 T-3 T-2 T-1 T
Spain 2008
Core inf lation
Credit (right axis)
0
50
100
150
200
250
0
1
2
3
4
T-5 T-4 T-3 T-2 T-1 T
Greece 2008
Core inf lation
Credit (right axis)
Credit Growth and Core Inflation
Pre crisis: a theory and policy gap
Macro models often ignored credit
Models with financial accelerators explored primarily how changes in
monetary policy affected the riskiness of borrowers
IC constraints generally bind, focus on quantity rather than quality
Little focus on risk attitude of the banking system
Banking literature focused on excessive bank risk-taking who:
Operate under limited liability
Are subject to asymmetric information
But this literature largely ignored monetary policy
Similar gap in policy making
Monetary policy considered financial stability the real of regulators
But regulators were focused more on individual banks than the system
Many observers have argued that monetary policy had an important role in
the recent crisis by providing intermediaries with the wrong incentives
Borio et al. (2008)
Several papers relate low interest rate environment to crisis
Overly loose monetary policy (Taylor, 2009)
Abundant liquidity – search for yield (Rajan, 2005, Acharya/Naqvi, JFE 2012)
Lending standards (Dell’Ariccia/Marquez, JF 2006, Gorton/He, RES 2008)
Increase in leverage and lower screening (Adrian and Shin, 2008, AER 2009,
Dell’Ariccia et al. JET 2013)
Debate on whether ultra-low rates and the macro bailout are seeding the
ground for new crisis:
Rajan ( NY Times 2010)
Acharya /Yorulmazer (JFI 2007), Diamond/Rajan (JPE 2012), Farhi /Tirole
(AER 2012)
Post crisis
Existing empirical work
Several papers used non-U.S. data:
Ioannidou, Ongena, and Peydró (2009): Bolivia
Altunbas, Gambacorta, and Marques-Ibañez (2010), Maddaloni and
Peydró (RFS 2011): Lending standards euro area (and US)
Jimenez et al. (ECM forthcoming): Spain
Very few looked at U.S. data:
Lown and Morgan (JMCB 2006): lending standards (not significant)
Paligorova and Santos (2012), Delis et al. (2012): Differential spreads on
syndicated loans
Buch/Eickmeier/Prieto (2011): aggregate version of STBL
Preview of results
1. We use confidential loan-level data from the Fed’s Survey of
Terms of Business Lending to measure ex-ante bank risk-taking
for US banks
2. We find a negative relation between the level of short-term
interest rates and bank risk-taking
3. We find that the strength of this relationship depends on
banks’ capital structure
4. Results are statistically significant and robust. But economic
magnitudes are relatively small
Theoretical background
(At least) Two opposite forces link the policy rate with bank risk taking
Risk shifting: Higher deposit rates reduce profits in case of success (classic
effect in models with limited liability)
Portfolio rebalancing: Higher yields on safe assets reduce portfolio risk
(standard in asset allocation models)
Net effect theoretically ambiguous (although additional channels such as
leverage can help determine it; Dell’Ariccia, Laeven, Marquez JET 2014)
The first effect is the greater when limited liability more binding: the lower
the bank capitalization
So we should observe a differential effect of MP changes across banks
When rates are cut, less capitalized banks should increase risk taking more
Data: Survey of Terms of Business Lending
Loan-level data from the Fed’s Survey of Terms of
Business Lending (STBL)
Also Call Report data for individual banks
Macro variables at state level
STBL: All individual new loans extended on first
business week of middle month of quarter since 1977
400 banks / 60 percent of US banking system assets
Since 1997, the STBL has asked banks to report the
internal risk rating of each new loan ( σ )
Data: Survey of Terms of Business Lending
The internal risk rating for the loan is an increasing,
discrete index of loan riskiness:
1 = Minimal risk
2 = Low risk
3 = Moderate risk
4 = Acceptable risk
5 = Special mention or classified asset
Preliminary evidence
2.4
2.6
2.8
33.2
3.4
Ris
k o
f lo
an
s (
detr
en
ded
)
-4 -2 0 2 4Federal Funds Rate (detrended) (in %)
𝜎𝑘𝑗𝑖𝑡 = 𝛼𝑖 + 𝜆𝑗 + 𝛽𝑟𝑡 + 𝛾𝐾𝑖𝑡 + 𝛿𝐾𝑖𝑡𝑟𝑡 + 𝜃𝑋𝑘𝑡+𝜇𝑌𝑖𝑡+𝜌𝑍𝑗𝑡 + 휀𝑘𝑗𝑖𝑡
Measure of loan riskiness
Bank capital-asset ratio
Federal fund rate (start of quarter)
Loan-specific variables
Bank-specific variables (size)
Regional macro variables
•Tier 1 •Total CAR •Stock-mkt-cap-to assets
•Size •Maturity •Dummy secured
•Personal income •Change in CPI •Unemployment rate •Change in house prices
•Size (total assets)
Empirical model
Baseline regression 1
Dependent variable: Loan risk rating
(1) (2) (3) (4) (5)
Target federal funds rate
-0.016** (0.007)
-0.021*** (0.006)
-0.031*** (0.009)
-0.031*** (0.008)
-0.031*** (0.008)
Bank and region fixed effects?
No Yes Yes Yes Yes
Region controls? No No Yes Yes Yes
Bank controls? No No No Yes Yes
Loan controls? No No No No Yes
R2 0.001 0.169 0.170 0.170 0.183
Obs 994,287 994,287 994,287 994,287 994,287
Baseline regression 2
Dependent variable: Loan risk rating
(1) (2) (3) (4)
Target federal funds rate -0.004 (0.009)
-0.003 (0.015)
0.014 (0.016)
Time effects
Tier 1 capital ratio 0.267 (0.429)
1.109*** (0.372)
Tier 1 capital ratio × target federal funds
-0.317*** (0.082)
-0.389*** (0.081)
Total capital ratio 0.661 (0.395)
Total capital ratio × target federal funds
-0.239*** (0.077)
Market cap 0.575*** (0.199)
Market capitalization × target federal funds
-0.077** (0.034)
R2 0.183 0.183 0.197 0.188
Obs 994,287 994,287 994,287 994,287
Baseline regression 2
Dependent variable: Loan risk rating
(1) (2) (3) (4)
Target federal funds rate -0.004 (0.009)
-0.003 (0.015)
0.014 (0.016)
Time effects
Tier 1 capital ratio 0.267 (0.429)
1.109*** (0.372)
Tier 1 capital ratio × target federal funds
-0.317*** (0.082)
-0.389*** (0.081)
Total capital ratio 0.661 (0.395)
Total capital ratio × target federal funds
-0.239*** (0.077)
Market cap 0.575*** (0.199)
Market capitalization × target federal funds
-0.077** (0.034)
R2 0.183 0.183 0.197 0.188
Obs 994,287 994,287 994,287 994,287
Results so far
Monetary tightening:
Decreases bank risk taking
Less so for lowly capitalized banks (consistent with risk shifting)
Results statistically significant and robust
Various capitalization measures
Inclusion of several controls
Fixed effects specification
Economic effect relatively small
Average risk rating 3.43 with 0.85 s.d.
At sample mean Tier 1 capital ratio a 1 s.d. increase in rates reduce risk by
about 1/12 of the rating s.d. (100bp hike would reduce rating by about 0.04)
Effect increases to 1/10 of s.d. at one s.d. below Tier 1 mean
Loans under commitment
Firms may draw on pre-approved credit lines
Actual new loan quality may differ from a bank’s chosen mix,
when there are monetary policy and other macro surprises
Exclude loans made under commitment
About 25 percent of observations
Results roughly the same, but improve a bit in size and
significance
Main identification concerns
Monetary policy may react to risk taking
Financial stability considerations may lead FOMC to cut rates when risk is
high to shore up banks’ balance sheets
Simultaneous causality (although less of a concern for new loans than for
stock)
Bank risk rating endogenous to monetary policy
Loan officers more optimistic during expansions
Underestimate risk
GDP correlated with policy rate
Higher rates may correspond to periods of euphoria (low risk ratings)
Endogeneity of monetary policy
Examination of minutes
Search for keywords
Pre-2007, little evidence that bank risk taking played significant role in MP
decisions
Consistent with official line (“it is regulators’ job”)
Focus on sub-samples for which concern less serious
States not correlated with national cycle (also answers second concern)
State with little income volatility
Small/local banks less exposed to national trends (exclude top quintile)
Banks in states without large banks
Frequency of keywords in FOMC minutes
Keyword
# of times the keyword
was used in FOMC
meetings from
1997Q2—2011Q4
# of times the keyword
was used in FOMC
meetings from
1997Q2—2006Q4
# of times the keyword
was used in FOMC
meetings from
2007Q1—2011Q4
Frequency of times the
keyword was used in
FOMC meetings from
1997Q2—2006Q4
Frequency of times the
keyword was used in
FOMC meetings from
2007Q1—2011Q4
Conservative Liberal Conservative Liberal Conservative Liberal Conservative Liberal Conservative Liberal
Bank risk 0 0 0 0 0 0 0.000 0.000 0.000 0.000
Banking risk 0 0 0 0 0 0 0.000 0.000 0.000 0.000
Banking sector 10 14 1 1 9 13 0.026 0.026 0.450 0.650
Banking system 15 19 3 3 12 16 0.077 0.077 0.600 0.800
Condition of the banking system 2 2 2 2 0 0 0.051 0.051 0.000 0.000
Financial conditions 112 351 74 187 39 167 1.897 4.795 1.950 8.350
Financial stability 14 17 0 0 14 17 0.000 0.000 0.700 0.850
Financial system 11 19 1 2 10 17 0.026 0.051 0.500 0.850
Health of the banking system 0 0 0 0 0 0 0.000 0.000 0.000 0.000
Risks to the financial system 1 1 0 0 1 1 0.000 0.000 0.050 0.050
Stability of the financial system 2 3 0 0 2 3 0.000 0.000 0.100 0.150
Systemic 2 4 0 0 2 4 0.000 0.000 0.100 0.200
Systemic risk 0 0 0 0 0 0 0.000 0.000 0.000 0.000
Troubles of the banking system 1 1 0 0 1 1 0.000 0.000 0.050 0.050
Notes: Frequency is determined as the number of times a word has been used within a time period divided by the number of quarters in that time period. Conservative = the number
of reports the word appears in (if a word appears several times in a report, that's not counted). Liberal = the total number of times the word appears in the reports.
Source: FOMC Minutes
State/US income correlation
Dependent variable: Loan risk rating
(1) States with
high correlation
with US GDP
(2) States with
low correlation
with US GDP
(3) States with
high correlation
with US GDP
(4) States with
low correlation
with US GDP
Target federal funds rate
-0.037** (0.014)
-0.021*** (0.004)
-0.018 (0.016)
0.009 (0.008)
Tier 1 capital ratio 0.707 (0.918)
-1.153* (0.637)
0.995 (0.930)
-0.736 (0.590)
Tier 1 capital ratio × target federal funds
-0.226* (0.131)
-0.349*** (0.095)
R2 0.212 0.147 0.212 0.147
Obs 561,642 432,645 561,642 432,645
Endogeneity of rating system
Harder to address. Do not have a fully satisfactory answer. Yet…
Results on states with low income correlation with US GDP
growth (and hence MP) comforting.
Control explicitly for GDP growth and recessions
Focus on deviations from “regional” conditions:
FF target minus state CPI
Deviations from “regional” Taylor rule
Again, results remain roughly the same
Aren’t crises different?
Monetary policy more likely to react to risk taking during distress
Also, banks may behave radically differently during crises
Split sample 1997-2007 and 2008-2010
Split sample in years with many/few bank failures
Results “die” in crisis years
Crisis and distress
Dependent variable: Loan risk rating
(1) Crisis years
(2) Non-crisis
years
(3) Years with many bank
failures
(4) Years with few bank failures
Target federal funds rate -0.031* (0.016)
0.008 (0.011)
0.026 (0.024)
0.017* (0.008)
Tier 1 capital ratio 0.757 (0.703)
0.056 (0.605)
-0.712 (0.507)
0.148 (0.727)
Tier 1 capital ratio × target federal funds
0.263* (0.133)
-0.549*** (0.109)
0.063 (0.187)
-0.394*** (0.096)
R2 0.200 0.194 0.192 0.206
Obs 254,761 739,526 348,329 645,958
Additional Results
This are not in this version of the paper (apologies to Hans)
Multinomial logit estimations and other tests to take into account
non-linearities (such as the effect of the lowest rated loans)
Link our risk rating variable to future NPLs
Better controls for macro conditions (throughout the paper. We
already had some in this version). Taylor residuals etc.
Results remain qualitatively the same
Conclusions I
Evidence that risk-taking by banks is negatively correlated with the level of short-term interest rates
This negative relationship is stronger for banks with higher capital ratios
Results are statistically robust, but economically small
Conclusions II
If taken at face value, little reason to alter design of monetary policy
Yet, there are other ways for banks to take risk
Liability side: Leverage, Maturity mismatches
Off-balance-sheet activities
End
Thank you