SLIDE 1
No Longer Too Big to Fail
EXTREMELY PRELIMINARY RESULTS
Antje Berndt Darrell Duffie Yichao Zhu ANU Stanford ANU Systemic Risk and Financial Regulation: 10 Years After Lehman Les institutions financi` eres face l` a r´ egulation Palais Brongniart, Paris, September 17, 2018
SLIDE 2 Big-bank credit spreads got much higher after the crisis
2002 2004 2006 2008 2010 2012 2014 2016 2018 50 100 150 200 250 LBOR-OIS spread (basis points)
(a) One-year LIBOR-OIS spreads
US banks European banks 2004 2006 2008 2010 2012 2014 2016 2018 50 100 150 200 250 300 year CDS rate
(b) 5-year CDS rates. Figure: (a) Spread between one-year USD LIBOR and one-year OIS (Fed funds). (b) Averages of the
5-year CDS rates of five U.S. banks (JPM, Citi, BAC, MS, GS) and of five European banks (Deutsche Bank, BNP, SocGen, Barclays, RBS). Data source: Bloomberg.
SLIDE 3 Is this consistent with the improved capitalization of big banks?
GS MS C BAC JPM WFC Tangible equity to assets (percent) 2 4 6 8 10 12 14 2007 2015
Ratio of tangible equity to assets. Data source: Holding company 10K filings.
SLIDE 4 The solvency buffers of big U.S. banks have gotten much larger
2002 2004 2006 2008 2010 2012 2014 2016 2018 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 Solvency ratio
Tangible equity divided by an estimate of the standard deviation of the annual change in asset value. Asset-weighted averages. Data: 10Ks of JPM, BOA, CITI, WF, GS, MS, ML, LB, BS, including preceding mergers, pro forma.
SLIDE 5 Presumably, lenders to large banks have reduced their beliefs in bailouts
◮ The EU Bank Recovery and Resolution Directive and Title II of the U.S.
Dodd-Frank Act have shifted expected insolvency losses from taxpayers to wholesale creditors.
◮ Conditional on the insolvency of a big bank, we estimate significantly reduced
market-implied probabilities of bailout.
◮ We estimate corresponding increases in credit spreads at a given distance to
default, and associated reductions in equity subsidies and subsidy-induced leverage.
SLIDE 6 Estimated 5-year CDS rates of big banks at a fixed distance to default
12/31/2002 12/31/2004 12/29/2006 12/31/2008 12/31/2010 12/31/2012 12/31/2014 12/30/2016 50 100 150 200 250 300
Credit spread (basis points)
Preliminary estimate for U.S. G-SIB holding companies at a distance to default of 2.
SLIDE 7 Sovereign uplifts have disappeared from big-bank credit ratings
Other firms GSIBs 2002 2004 2006 2008 2010 2012 2014 2016 2018 Ba Baa A Aa Aaa year Refined rating
Data source: Moody’s Investor Service. Ratings are adjusted for Watchlist and Outlook
SLIDE 8
Balance sheet at insolvency
assets V ∗ bonds deposits
The bank defaults when its assets get down to some endogenous level V ∗.
SLIDE 9
The bailout model
assets V ∗ bonds deposits bonds deposits assets V ∗ bailout capital bailout
The modeled bailout, if it occurs, injects enough government capital to increase the market value of the bonds to par, giving all equity to the government.
SLIDE 10
Unpredictable bailout
assets V ∗ bonds deposits recovered assets αV ∗ distress costs bond loss bond recovery deposits bonds deposits assets V ∗ bailout capital 1 − π b a n k r u p t c y π bailout
SLIDE 11
Conditional on no bailout: bankruptcy or bail-in
assets V ∗ bonds deposits recovered assets αV ∗ distress costs bond loss bond recovery deposits bailed in bonds bonds deposits assets V ∗ 1 − q b a n k r u p t c y q bail-in Reference: Chen, Glasserman, Nouri, and Pelger (2015); Neuberg, Glasserman, Kay, and Rajan (2016).
SLIDE 12
Bail-in and bankruptcy have similar impacts on equity and senior bonds
assets V ∗ bonds deposits recovered assets αV ∗ distress costs bond loss bond recovery deposits bailed in bonds bonds deposits assets V ∗ Identification of q may require separate price data for identified bail-in bonds or CDS. 1 − q b a n k r u p t c y q bail-in Reference: Neuberg, Glasserman, Kay, and Rajan (2016).
SLIDE 13 Simplified model of a bank
◮ The bank’s assets in place satisfy
dVt = (r − k)Vt dt + σVt dZt, for a“risk-neutral” standard brownian motion Z, where r is the risk-free rate and k is the proportional rate of net revenue.
◮ Risk-free deposits of size D bear interest at rate R. ◮ Bonds have constant total principle P and coupon rate c, with an exponentially
decaying maturity structure and average maturity 1/m. (Leland, 1994)
◮ Maturing bonds are replaced with new issues at competitive market prices.
SLIDE 14 The equilibrium default time and bailout subsidy
◮ Extending Leland (1994) and D´
ecamps and Villeneuve (2014), there is a unique time-homogeneous Markovian equilibrium default boundary V ∗, which we solve explicitly.
◮ The market value of the bailout subsidy is
π Vt V ∗ −γ (V0 − V ∗ − H0), where V0 is the asset level at which bonds are par valued and equity value is H0, and where γ = r − k − σ2/2 +
σ2 .
SLIDE 15 The panel regression step
◮ For a given firm i, time t, fixing the default boundary V ∗, the market CDS rate is
proportional to the estimated no-bailout probability 1 − pit.
SLIDE 16 The panel regression step
◮ For a given firm i, time t, fixing the default boundary V ∗, the market CDS rate is
proportional to the estimated no-bailout probability 1 − pit.
◮ The distance to default dit(pit) of firm i at date t is the number of standard
deviations of annual asset growth separating log V0 from log V ∗.
SLIDE 17 The panel regression step
◮ For a given firm i, time t, fixing the default boundary V ∗, the market CDS rate is
proportional to the estimated no-bailout probability 1 − pit.
◮ The distance to default dit(pit) of firm i at date t is the number of standard
deviations of annual asset growth separating log V0 from log V ∗.
◮ For given pit and from 1.6 million observed CDS rates from 2002-2017 for 855
public firms including a subset B of GSIBs, we estimate log CDSit 1 − pit = α + βdit(pit) + γ1i∈B +
δm1t∈m + φ1i∈B, t ∈ post crisis + ǫit.
SLIDE 18 The panel regression step
◮ For a given firm i, time t, fixing the default boundary V ∗, the market CDS rate is
proportional to the estimated no-bailout probability 1 − pit.
◮ The distance to default dit(pit) of firm i at date t is the number of standard
deviations of annual asset growth separating log V0 from log V ∗.
◮ For given pit and from 1.6 million observed CDS rates from 2002-2017 for 855
public firms including a subset B of GSIBs, we estimate log CDSit 1 − pit = α + βdit(pit) + γ1i∈B +
δm1t∈m + φ1i∈B, t ∈ post crisis + ǫit.
◮ We also include crisis fixed effects, DSIB fixed effects, sectoral fixed effects, and
SLIDE 19 Fitting post-crisis reductions in bailout probabilities
◮ We allow non-zero bailout probabilities for big banks only:
pit = πpre, pre crisis = πpost, post crisis.
SLIDE 20 Fitting post-crisis reductions in bailout probabilities
◮ We allow non-zero bailout probabilities for big banks only:
pit = πpre, pre crisis = πpost, post crisis.
◮ We assume no post-crisis change in average default-risk premia for big banks
relative to other firms.
SLIDE 21 Fitting post-crisis reductions in bailout probabilities
◮ We allow non-zero bailout probabilities for big banks only:
pit = πpre, pre crisis = πpost, post crisis.
◮ We assume no post-crisis change in average default-risk premia for big banks
relative to other firms.
◮ We therefore search for πpre and πpost that generate a zero estimate for φ, the
big-bank post-crisis fixed effect.
SLIDE 22 Fitting post-crisis reductions in bailout probabilities
◮ We allow non-zero bailout probabilities for big banks only:
pit = πpre, pre crisis = πpost, post crisis.
◮ We assume no post-crisis change in average default-risk premia for big banks
relative to other firms.
◮ We therefore search for πpre and πpost that generate a zero estimate for φ, the
big-bank post-crisis fixed effect.
◮ πpre and πpost cannot both be identified, so we estimate πpre for stipulated πpost.
◮ For example, setting πpost = 0.2, we estimate πpre = 0.65. ◮ For πpost = 0.0, we estimate πpre = 0.55.
SLIDE 23 Estimated 5-year CDS rates of a big bank at a fixed distance to default
12/31/2002 12/31/2004 12/29/2006 12/31/2008 12/31/2010 12/31/2012 12/31/2014 12/30/2016 50 100 150 200 250 300
Credit spread (basis points)
U.S. G-SIBs at a distance to default of 2, for πpost = 0.2 and fitted πpre = 0.65.
SLIDE 24 Total tangible assets of the largest U.S. banks
1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 1 2 3 4 5 6 7 8 9 10 11
Total assets (trillions of dollars)
Data source: Tangible assets, from 10Ks of JPM, BOA, CITI, WF, GS, MS, LB, BS. JPM and BOA include preceding mergers, pro forma.
SLIDE 25 Market-to-book equity ratios of big banks
Dealers: GS−MS−LEH−BSC−MER Banks: C−BAC−JPM*−WFC 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 1 2 3 4 year Market−to−book equity ratio
Asset-weighted averages. J.P. Morgan includes preceding mergers, pro forma.
SLIDE 26 Average ratio of GSIB estimated bailout subsidy to equity market value
Dec01 Dec03 Dec05 Dec07 Dec09 Dec11 Dec13 Dec15 Dec17 0.5 1 1.5 2 2.5 3 3.5
Average subsidy-to-equity ratio
For πpost = 0.2 and fitted πpre = 0.65, average of BoA, MS, C, JPM, GS, BNYM, WF.
SLIDE 27 Some prior work on post-crisis declines in TBTF subsidies
◮ Acharya, Anginer, Warburton (2016). ”We find that passage of Dodd-Frank Act
did not significantly alter investor expectations of future government support for large financial institutions.”
◮ Neuberg, Glasserman, Kay, and Rajan (2018). For Europe, an increase in
CDS-implied bail-in protection of senior debt in 2014, reversed in 2016.
◮ Atkeson, d’Avernas, Eisfeldt, and Weill (2018). For a stylized composite U.S.
bank and the Gordon dividend-discount model based on historical aggregate U.S. bank accounting returns, an estimated post-crisis 23% decline in the market-to-book ratio associated with bailout subsidies.
SLIDE 28 Appendix: Equilibrium default time
◮ A default time τ is an equilibrium for bank (V0, P, c, D, r, R, k, σ, α, π) if:
- 1. τ = inf{t : hτ(Jτ)t = 0}, where hτ(Jτ) is the equity price process implied by the
default time τ and the bond issuance price process Jτ.
- 2. Jτ is the bond issuance price process implied by the default time τ.
◮ An equilibrium default time τ is time-homogenous and Markovian if
τ = inf{t : Vt ≤ V ∗}, for some constant V ∗, the associated default boundary.
◮ This is effectively the solution concept of D´
ecamps and Villeneuve (2014).
SLIDE 29 Appendix: Theoretical default boundary with bailout or bankruptcy.
For the case D < αV ∗, V ∗ = γ
r −κ(cP+RD)
r
− D + π(V0 − H0)
r+m
− πP + (1 − π)D
- 1 + γ(1 − π)(1 − α) + γπ + ηα(1 − π)
, where η = r − k − σ2/2 +
- (r − k − σ2/2)1/2 + 2(m + r)σ2
σ2 . For the case D > αV ∗, V ∗ = γ
r −κ(cP+RD)
r
− D + π(V0 − H0)
r+m
− πP
.
