Public Liquidity, Bank Runs and Financial Crises
Wenhao Li
August 27, 2018 Public Liquidity Supply around the 2008 Financial Crisis
Dramatic increase in public liquidity, including reserves and treasuries.
0.45 Bank Reserves Treasuries Held By Domestic Private Investors 0.4 0.35 0.3 0.25 0.2 0.15
Liquidity/GDP ratio Liquidity/GDP 0.1 0.05 0 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 How relevant is public liquidity supply to the economy?
Quantify the liquidity channel: by holding more public liquidity, banks reduce losses during bank runs and fire sales.
I Holmstrom & Tirole 1998: Public liquidity alleviates systemic liquidity shocks.
Results: high relevance to both asset prices and output.
Different from credit channel, collateral channel, etc. Powerful without nominal and monetary frictions. Contributions
The first quantification of the liquidity channel.
Connection to the macroeconomy.
I Advancing QE1 to 2008 ⇒ ↑ 0.9% GDP
I Liquidity is effective only against bank run shocks, not non-financial recessions.
I Liquidity wealth effect, crowding out investment in the long-term
Explanatory power on asset prices.
I Model explains 43% monthly variations in liquidity premium. Benchmark: ∼ 10% for macro models (Del Negro et al. 2017).
I Model explains half of variations in the credit spread. Related Literature
Liquidity premium of U.S. treasury securities.
I Importance: (1) Indicator of financial distress (Longstaff 2004); (2) Informative on “moneyness” of the treasuries (Nagel 2016, Li 2018).
I Supply-premium relationship (Krishnamurthy and Vissing-Jorgensen 2012).
Financial sector balance sheet and macro.
I Bernanke and Gertler 1995; Kiyotaki and Moore 1997; Adrian and Shin 2010; Gertler and Karadi 2011; He and Krishnamurthy 2013; Brunnermeier and Sannikov 2014. Di Tella 2018.
Bank runs.
I Diamond and Dybvig 1983; Drechsler, Savov, and Schnabl 2014; Gertler, Kiyotaki, and Prestipino 2017; Moreira and Savov 2017. Outline
1 Model
2 Mechanism
3 Solution Method and Calibration
4 Asset Pricing Implications
5 Macroeconomic Implications Model Structure
Banks Households Government
Government Government No equity Bonds Bonds Wealth issuance (Bankers) Productive Capital Lump sum Government Wealth Taxation Bonds Productive Capital (lending to Bank Debt Bank Debt firms) Model Structure
Banks Households Government
Government Government Bonds Bonds Wealth (Bankers) Productive Capital Fire-Sale Lump sum Government Wealth Taxation Bonds Productive Capital Bank Run (lending to Bank Debt Bank Debt firms) Model Setup
Exogenous variables or functions in color.
Individual banker: maximize Z ∞ −ρt b E[ e log(ct )] 0 s.t. banker budget constraint.
Individual household: maximize Z ∞ −ρt h E[ e log(ct )] 0 s.t. household budget constraint. Investment and Production Technology
Firms = Productive capital. Modigliani-Miller holds.
K Firms make investment µt (growth of capital) according to the Q-theory.
Evolution of individual capital
dkj,t K K = µt dt − δdt + σ dBt − κ˜j,t dNt kj,t | {z } |{z} | {z } | {z } growth depreciation short-term fluctuations crisis shock
with crisis intensity λ> 0, and i.i.d. idiosyncratic shockκ ˜j,t ∈ {0, 1},
P(˜κj,t = 1) = θ ∈ (0, 1) Banker and Household Returns on Productive Capital
Bankers: the unit output of capital is A¯, with return
¯ K ¯K d(pt kj,t ) (A − φ(µt ))kj,t dRj,t = + dt pt kj,t pt kj,t | {z } | {z } ”capital gain” ”dividend”
Households: K K d(pt kj,t ) (A − φ(µt ))kj,t dRj,t = + dt pt kj,t pt kj,t
¯ ¯K K A > A ⇒ dRj,t > dRj,t ⇒ Bankers have incentives to raise funding from households. Government Liquidity and the Liquidity Premium
Government Bonds.
I Total supply Qt Kt . Balanced budget with lump-sum taxation. g g I All short-term with return dRt = rt dt
Illiquid assets.
I Selling price is only π ∈ (0, 1) fraction of the normal price. illiq illiq I Short-term with return dRt = rt dt.
Liquidity premium is illiq g `t = rt − rt Bank Runs: Asset Destruction
휅 = 1 휃 푗,푡 Capital destroyed
Banks: Shock 푑푁푡 hits
휿풋,풕: private info 휅푗,푡 = 0 1 − 휃 Capital not destroyed Bank Runs: Funding Withdrawal
휅 = 1 휃 푗,푡 Capital destroyed
Banks: Shock 푑푁푡 hits
휅푗,푡 = 0 1 − 휃 Capital not destroyed
Sticky deposits: No run. 1 − 훽
Not knowing which Active deposits: banks are solid, they 훽 have incentives to run. Bank Runs: Consequences
Bankruptcy and 휅 = 1 휃 푗,푡 Liquidation, Capital destroyed “run deposits” first
Banks: Shock 푑푁푡 hits
휅푗,푡 = 0 Fire Sales 1 − 휃 Capital not destroyed
Sticky deposits: No run. Funding 1 − 훽 withdrawal
Not knowing which Active deposits: banks are solid, they 훽 휿풋,풕 revealed have incentives to run. Endogenous Credit Limit and Dominant Strategies
Proposition 1 (Endogenous Credit Limit) Banks cannot raise debt during bank runs.
Intuition: during crisis shock dNt at time t, the expected loss of lending to a bank is O(1), but the benefit is O(dt).
Proposition 2 (Bank Run Game) If banks take leverage to hold productive capital, running on banks is a
weakly dominant strategy, for active deposits during crisis shocks dNt . Bank Runs and Fire Sales
Price of capital Crisis Shock 푝푡− Fundamental decline 푝 휅푡− = 푝푡− − 푝푡
푝푡 Market pressure 0 훼 푝푡
0 (1 − 훼 )푝푡 Time Banker Budget Dynamics
Individual banker j’s wealth dynamics
b dwj,t K R g g illiq illiq f d d b = xt µt + xt rt dt + xt rt dt + xt rt dt − xt dRt wj,t− | {z } | {z } | {z } | {z } | {z } capital return govt bond return illiquid asset return interbank lending return interest payment
b 0 ct K p α − b dt − (1κ˜j =1(1 − ε) + 1κ˜j =0(xt−κt− + 0 ∆xt− )) dNt . wt 1 − α | {z } | {z } consumption loss during a crisis
Banker’s effective funding withdrawal
d g illiq + ∆xt− = ( βxt− − (xt− + πxt− ) ) | {z } | {z } total funding withdrawal total liquidity insurance
g β < 1: higher xt− ⇒ smaller ∆xt−. Production
Then total output is Yt = ψt A¯ + (1 − ψt )A Kt , with A¯ > A | {z } per unit productivity of capital
where ψt is the fraction of bank holding of productive capital,
K wt xt ψt = K K wt xt + (1 − wt )yt
Liquidity supply affects
1) banker wealth wt drops in crises (↓ fire sales); K 2) bank lending xt (↓ risks). The Aggregate State Variable
b b h Aggregate state: banker wealth share wt = Wt /(Wt + Wt ), and aggregate capital Kt .
Aggregate growth of banker wealth share
dwt w b h b h = µt dt + (1 − wt )(σt − σt ) dBt − (1 − wt−)h(κt− − κt−) dNt wt− | {z } | {z } exposure to productivity shock exposure to bank run shock
Aggregate growth of productive capital follows
dKt K K = µt dt − δdt + σ dBt − θdNt Kt | {z } |{z} | {z } | {z } growth depreciation short-term fluctuations capital destruction loss Equilibrium Definition
Markov equilibrium with state variables wt and Kt , such that
I Consumption and portfolio choices are optimal.
I Market clearings: productive capital, government bonds, and the illiquid assets (zero supply)
I Aggregate wealth
∞ Z h b h −rs (s−t) Wt + Wt = pt Kt + Q(wt )Kt − Et [ e τs ds] |{z} t real wealth | {z } liquidity wealth
I Government budget balance
g Et [d(Q(wt )Kt ) + τt dt − Q(wt )Kt rt dt] = 0
I Resource constraint: Production = consumption + investment costs.
¯ b h K (ψt A + (1 − ψt )A) · Kt = Ct + Ct + φ(µt )Kt Outline
1 Model
2 Mechanism
3 Solution Method and Calibration
4 Asset Pricing Implications
5 Macroeconomic Implications Liquidity Premium ∼ Price of Liquidity Service
Liquidity Premium ℓ = 푟푖푙푙푖푞 − 푟푔
Liquidity Supply
Equilibrium Liquidity Demand
0 Bank Holding of Public Liquidity
Figure: Determination of the liquidity premium at certain state (w, K) Liquidity Premium: Impact of Supply
Liquidity Premium ℓ = 푟푖푙푙푖푞 − 푟푔
Larger public liquidity supply
0 Bank Holding of Public Liquidity
Figure: Determination of the liquidity premium at certain state (w, K) Liquidity Premium: Impact of Demand
Liquidity Premium ℓ = 푟푖푙푙푖푞 − 푟푔
Higher Leverage, more bank debt
0 Bank Holding of Public Liquidity
Figure: Determination of the liquidity premium at certain state (w, K) Equilibrium Determination of Crisis Severity
Price Decline Jump 휅푝
Individual Aggregate consequence choice of fire-sale
Equilibrium Productivity
0 Bank Holding of Productive Capital
Figure: Determination of price decline κp in a bank run at state (w, K) Liquidity Supply and Severity of the Crisis
Price Decline Jump 휅푝
Larger Liquidity Supply
Productivity
0 Bank Holding of Productive Capital
Figure: Influence of liquidity supply at state (w, K) Liquidity Externality
Price Decline Jump 휅푝
Aggregate consequence of fire-sale
Considering externality
Productivity
0 Bank Holding of Productive Capital
Figure: Liquidity Externality Outline
1 Model
2 Mechanism
3 Solution Method and Calibration
4 Asset Pricing Implications
5 Macroeconomic Implications Solution Method
Main challenge: Jump depending on the unknown function p(w). I have designed an efficient functional iteration algorithm. Generality: more state variables; other macro finance/asset pricing/dynamic corporate finance models with endogenous jumps.
Price 푝(푤)
푤푡−
푝 휅푡−
푤푡 푤 휅푡−
Banker Wealth Share 푤 Calibration: Connections between Model and Data
Banks Households
Wealth Govt Bonds (Bankers) Govt Bonds Productive Capital Less Liquid Govt Assets MODEL Sticky Deposits Wealth Sticky Bank Debt Productive Capital
Active Runnable Active Deposits Bank Debt Calibration: Connections between Model and Data
Bank Holding Companies, Households, MMF, Pensions, Depository Institutions, Broker Dealers etc. Mutual Funds, etc.
Treasuries Equity Treasuries, Bank Reserves Corporate Bonds etc. Illiquid Govt Liabilities Insured DATA Deposits, $14t Wealth Long-term in 2007 Insured Bank debts Deposits, Loans Long-term (syndicated debts loan, commercial loan, etc.) ABCP, Repo, ABCP, Repo, MBS, etc. and other $4.6 t and other wholesales in 2007 wholesales funding funding Calibration
Parameters Choice Quantity to be matched Target Model
Production and Investment: A¯ Banker productivity 0.15 Investment to capital ratio 11% 11%
A Household productivity 0.125 Difference of return (A¯ − A)/pt 0.024 0.025 to prime loan rate - MMF rate
Bank Run and Fire Sales: λ Bank-run arrival rate 2.5% Frequency of financial crises ?0.025 0.025 β Fraction of bank run 25% Bank runnable funds/assets ?25% 25% ? α0 Fire sale market pressure. 21% Price pressure during fire sales. 21% 21%
θ Crisis shock size 10−5 Negligible value < 0.01 < 0.01 ε Bankruptcy leftover 10−3 Negligible value < 0.01 < 0.01
For Asset Prices: π illiquid asset resellability 11% Match average liquidity premium 22 bps 22 bps
Other macro parameters: Depreciation rate δ, discount rate ρ, growth volatility σK , investment adjustment cost χ are set to the standard values in the macro literature. Outline
1 Model
2 Mechanism
3 Solution Method and Calibration
4 Asset Pricing Implications
5 Macroeconomic Implications Data and Measurements
Public liquidity supply (1920-2016)
I Treasuries held by domestic private investors + Bank reserves
Financial sector leverage (1970-2016).
I Leverage data from He, Kelly, Manela (2017)
Liquidity premium (1920-2016)
I 1991-2016: Principal component of maturity matched Refcor Bonds - Treasury spreads (1-5 years), and Repo 3M collateralized - Treasury 3M
I 1920-1990: Banker Acceptance 3M - Treasury 3M. (Nagel 2016)
t t Method: At each month t, set Q (·) = Qdata, solve model and find w with t lvg(w) = lvgdata. Liquidity Premium – Time Series Predictions and Data
43% explained = 10% + 29% + 4% |{z} |{z} |{z} liquidity supply financial sector leverage interaction
1.6 Data Model Prediction 1.4 Spike in liquidity demand:
)
% 1.2 crisis and fragile banks Spike in liquidity
( supply
m
u
i 1
m
e r 0.8
P
y
t
i
d 0.6
i
u
q
i
L 0.4
0.2
0 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 Impact of Public Liquidity Supply on Liquidity Premium
Strong interaction effects: liquidity supply is more valuable in distress periods.
60 Public Liquidity/GDP decreased uniformly by 10%
50
40
30
20
10 Liquidity Premium Difference(bps) Premium Liquidity
0 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 Outline
1 Model
2 Mechanism
3 Solution Method and Calibration
4 Asset Pricing Implications
5 Macroeconomic Implications Matching 2008 Financial Crisis: Shocks
Real GDP (Detrended at 1.4%)
푑퐵 푡 -6.3 -0.7 -0.7 -1.0 -0.4 -1.1 -0.7 0.3 2 (%)
푑푁푡 0 1 0 0 0 0 0 0
0
%
n
i
e
g
n
2
a
−
h
C Change Change in %
Contribution
4 − of 푑푁푡 = 3.9%
6
−
2006 2008 2010 2012 2014 Matching 2008 Financial Crisis: Model v.s. Data
Real GDP (% Diff from 2007) Financial Equity (% Diff from 2007)
1 Data Data Model Model 0 0 −1 −20 −2 −3 Change in % Change in % −40 −4 −5 −60 −6 2008 2010 2012 2014 2008 2010 2012 2014
Credit Spread (Diff from 2007) Liquidity Premium (Diff from 2007)
Data Data
Model 0.8 Model 1.0 0.6 0.5 0.4 Change in % Change in % 0.2 0.0 0.0
2008 2010 2012 2014 2008 2010 2012 2014 What if all shocks come from dBt?
Real GDP (Detrended at 1.4%) Financial Equity (Detrended at 1.4%)
1 Data Data Model Model 0 0 −1 −20 −2 −3 Change in % Change in % −40 −4 −5 −60 −6 2008 2010 2012 2014 2008 2010 2012 2014
Credit Spread (Diff w.r.t. 2007) Liquidity Premium (Diff w.r.t. 2007)
Data Data
Model 0.8 Model 1.0 0.6 0.5 0.4 Change in % Change in % 0.2 0.0 0.0
2008 2010 2012 2014 2008 2010 2012 2014 Liquidity Supply and Two Counterfactual Experiments
Total Liquidity/GDP 0.3 Treasuries Held By Domestic Private Investors 0.30 Counterfactual Experiment 1 Bank Reserves Counterfactual Experiment 2 0.25
0.2 0.20 0.15 Liquidity/GDP 0.1 Public Liquidity/GDP Public 0.10 0.05
0.0
2008 2010 2012 2014 0.00 2008 2010 2012 2014 Counterfactual 1: 10% Increase of Liquidity/GDP at 2008
GDP Financial Equity 6
1.5 Counterfactual − Baseline Counterfactual − Baseline 5 4 1.0 3 2 0.5 1 Counterfactual Difference (%) Difference Counterfactual (%) Difference Counterfactual 0 0.0
2008 2010 2012 2014 2008 2010 2012 2014
Credit Spread Liquidity Premium
0.02 Counterfactual − Baseline Counterfactual − Baseline 0.00 −0.02 −0.04 −0.06 Counterfactual Difference (%) Difference Counterfactual (%) Difference Counterfactual −0.08 −0.10 2008 2010 2012 2014 2008 2010 2012 2014 Counterfactual 2: Increase of Liquidity/GDP in 2012,2013
GDP Financial Equity
0.3 Counterfactual − Baseline 1.0 Counterfactual − Baseline 0.8 0.2 0.6 0.4 0.1 0.2 0.0 Counterfactual Difference (%) Difference Counterfactual (%) Difference Counterfactual −0.2 −0.1 2008 2010 2012 2014 2008 2010 2012 2014
Credit Spread Liquidity Premium
Counterfactual − Baseline Counterfactual − Baseline 0.00 0.00 −0.05 −0.05 −0.10 −0.10 Counterfactual Difference (%) Difference Counterfactual (%) Difference Counterfactual −0.15 −0.15 2008 2010 2012 2014 2008 2010 2012 2014 The Liquidity Wealth Effect
Total wealth = productive capital + value of bonds - taxation ∞ Z h −rt t = productive capital + E[ e (··· ) · `t · Qt Kt dt] 0 | {z } liquidity wealth
0.01 liquidity wealth/GDP (%) 10 growth rate of K (% diff) 0 average discounted output (diff)
8 -0.01
-0.02 6
-0.03 4 -0.04
2 -0.05
0 -0.06 0.5 1 1.5 2 2.5 3 3.5 4 public liquidity/GDP Policy Implications from the Liquidity Insurance Channel
Results Policy Implications
A. Injecting liquidity earlier → 50 times benefits. Prompt
B. Liquidity → Vulnerability to real shocks. Only for financial shocks.
C. Liquidity wealth effect → ↓ investment Short-lived Summary: How relevant is public liquidity?
Answer: highly relevant, through the liquidity channel.
Approach: build and quantify a new microfounded macro model.
Results: connection to both asset prices and the macroeconomy.
I Explain 43% variations of monthly liquidity premium, and 47% credit spread.
I Advancing QE1 to 2008 ⇒ ↑ 0.9% GDP, persistent for years.
New implications of liquidity policy: prompt, short-lived, and only against financial crises.