haircuts and repo rates: evidence from money market mutual
TRANSCRIPT
Aggregate Patterns of Repo FundingTheory
Empirical Results
Haircuts and Repo Rates:Evidence from Money Market Mutual Fund Filings
Arvind Krishnamurthy1 Stefan Nagel2 Dmitry Orlov2
1Northwestern University
2Stanford University
November 2010
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Funding of Shadow banks
MMMFBroker/Dealer
HedgeFunds
SPV(agency/non‐
agency)
Repo
Repo
ABCP
ABS
Treasuries
CorporatesecuriAes
ABCPconduit
Mortgages
Loans
ABS
“deposits”($1NAV)
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Tri-Party Repurchase Agreements
MMMF
Tri‐partyClearingAgent
$95m
Collateralworth$100m
$95m
Collateralworth$100m
Broker/Dealer
Haircut: 5% in this example
Repo rate: Interest paid by borrower on loan amount ($95m)
Daily unwind: Irrespective of repo term, each morning cashreturned to lender and security to borrower. Thus, intra-daycounterparty risk shifted to tri-party agent.
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Objectives
Which role did repo market play in financial crisis?
How big is repo funding? Often used federal Reserve data onprimary dealer repos includes inter-dealer repos(double-counting issue)“Run on repo” an in important part of the breakdown of“securitized banking” (Gorton and Metrick 2009)? Evaluatesize of repo funding with private-label ABS/MBS as collateral
How are repos structured and risks priced?
Participation constraints, haircuts, repo ratesEvaluate role of counterparty risk, collateral risk, ...View through lens of theories of collateralized lending andsecurity design
We obtain data on repo agreements of MMF from quarterlySEC filings (N-CSR, N-CSRS, N-Q) 2006Q4-2010Q2
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Example: Reserve Fund – Primary Fund
February 29, 2008 Repurchase Agreements
Notional Counterparty Rate Init. Rep. Collateral Coll. mkt.val.
1,000,000,000 Bear Stearns 3.28%, 2/29/08, 3/3/08 ABS, CMO, TRR, TR3 1,048,922,871450,000,000 Bear Stearns 3.33% 2/29/08 3/3/08 ABS, CMO 472,500,201500,000,000 Citigroup 3.23% 2/29/08 3/3/08 MNI, TRR 556,131,379140,000,000 Merrill Lynch 3.43% 2/29/08 3/3/08 WLR 146,599,1931,000,000,000 Morgan Stanley 3.29% 2/29/08 3/3/08 WLR 1,020,794,540...
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Data collection
Concentrated market: Biggest 10 MMF families control about60% of MMF assets under management
Aim: Collect data for 20 biggest MMF families
Completed so far:
BlackrockFidelityJPMorganReserve FundsMorgan StanleyVanguardDreyfusGoldman SachsFederated Funds
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Outline
1 Aggregate Patterns in Repo Funding
“Run on Repo” quantitatively important?
2 Theory: repo market participation, collateral choice, maturity,haircuts, repo rates
3 Empirical results
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Coverage of MMF Filings Sample
Quarter MMF Repo MMF Repo MMF Primarycollected Total Assets Dealer Repo($bn.) (FoF, $bn.) (FoF, $bn.) (NY Fed, $bn.)
2006Q4 (133)1 395 2312 34422007Q1 202 387 2372 36192007Q2 205 426 2466 38892007Q3 258 528 2780 38862007Q4 283 606 3033 41062008Q1 307 592 3383 42782008Q2 273 518 3318 42222008Q3 261 592 3355 39892008Q4 276 542 3757 32082009Q1 367 562 3739 27432009Q2 339 488 3585 25822009Q3 325 495 3363 24992009Q4 338 480 3259 24692010Q1 296 440 2931 24772010Q2 (66)1
1Incomplete coverage in 2006Q4 and 2010Q2Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Share of Collateral by Type (by value)
.4.6
.81
Sh
are
2007q1 2008q1 2009q1 2010q1Quarter
U.S. Treasury Agency Priv. ABS
Corporate Other
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Comparison with ABCP Issuance
0.0
5.1
.15
Priv. A
BS
Share
50
100
150
200
250
Issuance (
)
2007q1 2008q1 2009q1 2010q1Quarter
ABCP Issuance Priv. ABS Share
Issuance of 80day+ ABCP net of amount funded through FedCPFF programTotal contraction of ABCP outstanding ≈ $700bn. comparedwith pre-crisis repo with priv. ABS/MBS collateral ≈ $60bn.
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Maturity percentiles (vw.)
05
01
00
15
02
00
25
0M
atu
rity
(b
usin
ess d
ays)
2006q3 2007q3 2008q3 2009q3 2010q3Quarter
99th 98th 95th 90th
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Maturity percentiles (ew.)
05
01
00
15
02
00
25
0M
atu
rity
(b
usin
ess d
ays)
2006q3 2007q3 2008q3 2009q3 2010q3Quarter
90th 80th 70th 60th
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Haircuts by Collateral Type (vw.)
24
68
Pe
rce
nt
2007q1 2008q1 2009q1 2010q1Quarter
U.S. Treasury Agency Priv. ABS
Corporate Other
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Average Repo Rate (vw.) and Fed Funds Rate/OIS
02
46
Pe
rce
nt
2006q3 2007q3 2008q3 2009q3 2010q3Quarter
Fed Funds Rate/OIS Average Repo Rate (vw.)
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Excess Repo Rate by Collateral Type (vw.)
−1
−.5
0.5
1P
erc
en
t
2007q1 2008q1 2009q1 2010q1Quarter
U.S. Treasury Agency Priv. ABS
Corporate Other
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
LiteratureModelResults
Haircuts and Repo Rates: Theory
Modigliani-Miller: Haircut and repo rate indeterminate
Haircut = leverageRepo rate = cost of debt
Theories of equilibrium haircuts with frictions
Geanakoplos (2009): Differences in beliefs between borrowerand lender about payoffs from collateral. Equilibrium haircutcreates default-free debtDuffie and DeMarzo (1999); Dang, Gorton, Holmstrom(2010): Asymmetric information about collateral payoffsbetween borrower and lender. Equilibrium haircut createsinformation-insensitive security (if sufficient concern aboutadverse selection).
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
LiteratureModelResults
Theory of Haircuts and Repo Rates
Common predictions of belief divergence and asy. informationstories
Haircuts should vary with risk of collateral, but repo ratesshould (mostly) notCounterparty risk should have little effect on haircuts and reporates
Theories miss some aspects that seem important in practice
Repo is not no-recourse: Repo lenders have recourse toborrowers balance sheet in event of defaultDifferences in beliefs and asy. information can not explainexclusion of high-risk counterparties and low-quality collateralfrom repo market
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
LiteratureModelResults
Model
Two dates, t0 and t1.
Single risky asset with t0 price P0 = 1 (partial equilibrium)
Borrower (trading desk in a bank) considers purchase of oneunit of risky asset with funding
1− h from repo lender (MMF), collateralized by risky asset,i.e., with haircut hh from bank (“equity”)
Four states of nature: At time t−1 just before date t1, thebank defaults with probability πd . Then, at t1, independent ofwhether the bank defaulted or not, the risky asset can betraded at price of R > 1 in the good state and L < 1 in thebad state.
Lenders are competitive. Lender, bank, and borrower arerisk-neutral.
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
LiteratureModelResults
Key assumptions
Belief divergence: Borrower and bank perceive the probabilityof the bad state to be pb, while lenders have a morepessimistic belief pl > pb, as in Geanakoplos (2009)
Equity financing friction: Bank discounts expected paymentfrom “trading desk” by by 0 ≤ α ≤ 1
Liquidation cost: Lender pays cost δ per $1 of face value ifshe has to take possession of the collateral in the event thebank defaults
Recourse: In the event of default on repo loan, lender hasrecourse to bank balance sheet, so unless bank defaults, thebank bears the losses on the risky asset in the bad state
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
LiteratureModelResults
Valuation and Objective
Repo: Lender has to offer face value (repurchase value) F (h)that satisfies
(1− h) =(1− πd)F (h) + πd [(1− pl)F (h) + pl min(L,F (h))]
− πdδF (h)
Equity: Borrower pays bank payment of E (h) in good state,which must satisfy
h =α[(1− pb)E (h) + (1− πd)pb(L− F (h))+
πdpb max(L− F (h), 0)]
The borrower’s objective is given by
maxh
(1− πd)(1− pb)(R − E (h)− F (h))
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
LiteratureModelResults
Results
Borrowers objective is piecewise linear: Borrower (weakly)prefers either zero leverage, risk-free repo with high haircut, orrisky zero-haircut repo
Non-participation: Zero leverage is preferred over risk-freerepo if
1− πdδ
α< 1
Haircut: Risk-free repo preferred over risky zero-haircut repo if
1− πdpl
1− πdpl − δπd
(1− πdpb
1− πdpl+
(α− 1)(1− pb)
1− πdpl
)> 1
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
LiteratureModelResults
Illustration of Non-Participation: Borrower Objective
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.031
0.032
0.033
0.034
0.035
0.036
0.037
Haircut
Base
Higher default prob.
Higher liq. cost
Base case: R = 1.1, L = 0.8, pb = 0.2, pl = 0.5, α = 0.996,πd = 0.02, and δ = 0.1.
Alternative cases with πd = 0.04 or δ = 0.2.
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Counterparty Participation
Model
Model: Non-participation (zero leverage) is preferred overrisk-free repo if
1− πdδ
α< 1
Holding δ and α fixed, this implies participation constraintbased on πd
Empirically
Counterparty risk measure x (5yr Sr. CDS rate) as empiricalcounterpart to πd
We only observe participants, so not possible to estimateparticipation condition
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Distribution of Counterparty CDS Rates
0.0
5.1
.15
.2F
ractio
n
0 200 400 600 800 1000Counterparty 5−yr. Senior CDS rate (bps)
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Collateral Choice
Model
Non-participation (zero leverage) is preferred over risk-freerepo if
1− πdδ
α< 1
Holding α fixed, participation with high-δ collateral only if πd
low
δ likely higher with higher collateral risk
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Collateral Choice
Empirically
Collateral category q ∈ {0, 1, 2} ordered by collateral risk(Treasuries, Agencies, Others)
Choice of q modeled as function of latent variable q∗,
q∗ = a′0z + a1x + a2g + η, η|z , x , g ∼ N (0, 1)
where q = 0 if q∗ ≤ θ1, q = 1 if θ1 < q∗ ≤ θ2, q = 3 ifq∗ < θ3.
Macro variables z may help capture time-variation in δ
Government MMF dummy g , can be viewed as proxy forfunds that face extremely high δ for riskier collateral
Estimation of P(q = 0|x , z), ..., P(q = 3|x , z) with orderedprobit
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Collateral by Counterparty CDS Rate
0.2
.4.6
.81
Fra
ctio
n (
by v
alu
e)
< 50bps 50bps − 100bps 100bps − 250bps > 250bps
U.S. Treasury Agency Other
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Collateral Choice: Ordered Probit
Marginal effects on collateral-category probabilities
3mLIBOR-OIS spread and CDS rates in percentVIX in percent divided by
√250
(1) (2) (3)U.S. Treasury Agency Other
3mLIBOR-OIS -0.053 0.009 0.045(0.065) (0.010) (0.055)
VIX 0.137 -0.022 -0.115(0.035) (0.005) (0.032)
Govt. MMF dummy 0.570 -0.128 -0.442(0.023) (0.019) (0.022)
Counterparty CDS Rate -0.009 0.001 0.008(0.013) (0.002) (0.010)
Observations 7968 7968 7968
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Maturity Choice
Higher counterparty risk also likely implies increasedprobability of a future substantial revision of counterparty risk:With shorter maturity lender retains option to terminate,which reduces expected liquidiation costs
Shortening of maturity may be first response before reachingnon-participation status
Regressionlog(m) = b′0z + b1x + ν
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Maturity by Counterparty CDS Rate
05
10
15
Ave
rag
e I
nitia
l M
atu
rity
(d
ays,
va
lue
−w
eig
hte
d)
< 50bps 50bps − 100bps 100bps − 250bps > 250bps
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Maturity Choice: OLS
Dependent variable: Log of initial maturity
3mLIBOR-OIS 0.236(0.183)
VIX -0.161(0.138)
Custodian CDS Rate -0.147(0.223)
Counterparty CDS Rate -0.304(0.052)
Observations 7967Adjusted R2 0.024
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Maturity Choice: Probit
Marginal effects on probability that repo is overnight
3mLIBOR-OIS 0.012(0.044)
VIX -0.003(0.028)
Custodian CDS Rate -0.057(0.047)
Counterparty CDS Rate 0.092(0.019)
Observations 7968
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Haircuts
Model
Haircut set so that repo is riskless
Generalizing to more realistic setting where haircut makesrepo almost, but not entirely riskless: Collateral risk should beprimary influence, as higher haircut does alter πd and δ
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Haircuts
Empirically
Regression
h = c ′0z + c1x + c2m + c3w + ξ
with collateral risk measures w
Collateral volatility: Standard deviation of collateral index pricechanges in prior monthCollateral worst return: Worst daily price change of collateralindex in prior five years
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Haircuts: Collateral Risk and Counterparty Credit Risk
Collateral indices
Treasuries: Barclays US Treasury IndexAgency: Barclays US MBS indexPrivate-label MBS/ABS: Barclays US ABS home equityCorporate: Barclays US Corporate Investment GradeCommercial Paper: Fed St. Louis Commercial paper (priceindex constructed from yields)Certificates of Deposit: BBA 3m LIBOR (price indexconstructed from yields)Munis: Barclays Municipal BondEquity: S&P500 index
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Decomposing Variation in Haircuts
Nonparametric approach: Dummy variable regressions
Time dummies (year-month)Time dummies interacted with collateral, counterparty,maturity dummies
Incremental R2 DGF F p-value
Collateral×Time 0.38 219 37.51 0.00Counterparty×Time 0.02 801 1.67 0.00Maturity×Time 0.02 50 7.01 0.00Full 0.67 1114 15.05 0.00
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Haircuts: Regression
3mLIBOR-OIS -0.607(0.159)
VIX -0.197(0.099)
Initial maturity 0.004 0.003(0.001) (0.001)
Collateral volatility 1.546 1.657(0.255) (0.271)
Collateral worst return -0.282 -0.243(0.020) (0.022)
Counterparty CDS Rate 0.167 0.131(0.063) (0.056)
Year-month dummies N YObservations 7521 7521Adjusted R2 0.357 0.384
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Repo rates
Model
Haircut set so that repo is riskless, but in more general casewhere risk not entirely eliminated, risk-neutral valuationimplies repo rate
r =πd
1− πdpl − πdδ(λ(h) + δ)
where
λ(h) = pl
(1− L
1− h
)is the expected loss (to the lender) given default.
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Repo rates
Empirically
Empirical measurement of λ(h)
Gaussian expected loss
φ(−h/σ)
Φ(−h/σ)σ
where σ is measured by the standard deviation of collateralindex price changes in prior monthHaircut-adj. collateral worst-return: Worst daily price changeof collateral index in prior five years in excess of haircut
Regression
r = d ′0z + d1m + d2x + d3λ(h) + d4(x × λ(h)) + ξ
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Decomposing Variation in Excess Repo Rates
Incremental R2 DGF F p-value
Collateral×Time 0.32 219 29.49 0.00Counterparty×Time 0.06 806 1.59 0.00Maturity×Time 0.04 50 17.42 0.00Full 0.62 1119 13.14 0.00
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Excess Repo Rates: Regression
3mLIBOR-OIS 0.146 0.145(0.083) (0.083)
VIX -0.080 -0.075(0.064) (0.066)
Custodian CDS Rate 0.069 0.059(0.075) (0.075)
Initial maturity -0.000 0.000 -0.000 0.000(0.000) (0.000) (0.000) (0.000)
Counterparty CDS Rate -0.031 -0.041 -0.069 -0.086(0.039) (0.036) (0.073) (0.070)
Expected loss λ(h) 0.092 0.089 0.081 0.073(0.013) (0.013) (0.015) (0.014)
Haircut-adj. collateral worst return -0.012 -0.007 -0.025 -0.014(0.006) (0.006) (0.010) (0.009)
λ(h)× Counterparty CDS rate 0.010 0.013(0.013) (0.013)
Haircut-adj. coll. worst return × Counterparty CDS rate 0.013 0.009(0.010) (0.009)
Year-month dummies N Y N YObservations 7065 7065 7065 7065
Adjusted R2 0.131 0.222 0.133 0.224
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo
Aggregate Patterns of Repo FundingTheory
Empirical Results
Participation, Collateral Choice, MaturityHaircutsRepo Rates
Conclusion
Aggregate amounts of repo funding provided by MMF
Private-label ABS/MBS completely disappear as collateral in2008/2009, but aggregate amount small relative to contractionin ABCP: “Run on repo” may be symptomatic for relatedproblems, but by itself not a major factor in breakdown ofshadow bank financing
Terms of tri-party repo agreements
Counterparty risk in repo markets affects participation,collateral choice, and maturityConditional on participation and collateral choice, counterpartyrisk has little influence on haircuts and repo ratesCollateral risk is main driver of haircuts and repo rates, and“long memory” of realized tail events seems to matter inaddition to recent volatilityElevated levels of haircuts and repo rates still persist sincecrisis
Arvind Krishnamurthy, Stefan Nagel, Dmitry Orlov Repo