SYSTEMIC RISK IN CLEARING HOUSES: EVIDENCE FROM THE EUROPEAN REPO MARKET

Transcription

1 SYSTEMIC RISK IN CLEARING HOUSES: EVIDENCE FROM THE EUROPEAN REPO MARKET SECURITIES MARKETS TRENDS, RISKS AND POLICIES MILAN, FEB BOISSEL, DERRIEN, ORS, THESMAR (HEC Paris)

2 Motivation 2 We ask: Are CCPs resilient in time of market stress? Central Counterparties (CCPs), i.e. clearing houses, eliminate counterparty risk by inserting themselves between the two sides of a trade CCPs are at the center of policy makers attention EMIR and title VII of Dodd Frank Act Coeuré (ECB board member, 2014): The success of CCPs [ ] depends on the ability of CCPs to handle the risk they are concentrating What we find is consistent with In times of moderate stress CCPs work well But they appear to be less effective in times of high market stress

3 What we look at 3 Eurozone anonymous General Collateral (GC) CCPcleared repo market Collateral from a basket of Eurozone sovereign bonds Cash lender only knows country of the bond Not the maturity, not the coupon Borrower gives collateral to CCP and gets cash from CCP Cash lender lends to CCP against sovereign bond Each transaction is over-collateralized by a factor (the haircut) determined by the CCP Price of borrowing = Repo rate Baseline: Very safe transactions as counterparty default insured by collateral + CCP default fund

4 Our data 4 Data from two European repo platforms: ICAP/MTS Members: Large Banks (mainly EZ but also US) Data on all GC CCP-cleared repo transactions 11 nationalities of sovereign bonds GIIPS: Greece, Ireland, Italy, Portugal, Spain Non-GIIPS: Austria, Belgium, Finland, France, Germany, Netherlands Daily data on volumes and repo rates Average = 50bn of daily volume / EZ secured IB market = 200bn / CCP Cleared = 100bn Haircuts: One haircut per country chosen by CCP (LCH Clearnet) Sovereign risk We use CDS spreads on 5-year sovereign bonds

5 A simple framework 5 Assume risk neutral arbitrageur Option #1: lends to the ECB Rate = r ecb Option #2: lends against government bond of country c Rate = r c repo Unless triple default: Counterparty fails CCP fails Government fails Implicit assumption that when both counterparty and CCP default, lender ends up with the bond

6 A simple framework 6 No-arbitrage condition rr eeeeee = 1 ppλλλλ rr rrrrrrrr ppλλλλ. LLLLLL h (1) π = P(sovereign bond default) p = P(counterparty default sovereign bond default) λ = P(CCP defaults sovereign & counterparty default) LGD(h): LGD conditional on haircut h Rewriting (1) rr rrrrrrrr rr eeeeee = ppλλ LLLLLL h LLLLLL 0. (ππllllll 0 ) Sovereign CDS spread

7 Empirical framework 7 R c,t repo - R t ecb = a + b.cds c,t + γ t + δ c,m + ε c,t γ t = day fixed effect (common factor exposure) δ c,m = country-month FE (slow moving country factor) ε c,t are clustered across days Model says: b = LGD(h) x p x λ p = P(counterparty default sov. bond default) λ = P(CCP default sov. bond & counterparty default)

8 Repo rate sensitivity to sov. CDS 8 Run this regression at several key periods of the EZ crisis Coefficient only significant at the peak of the crisis Sovereign crisis

9 GIIPS vs. non-giips 9 This is entirely driven by GIIPS countries Similar effect on volume, albeit less robust CDS spread increases by 100bp Repo rate increases by 20bp

10 What explains the rise in sensitivity? 10 Repo-to-CDS spread sensitivity = increased in 2011 Try to disentangle between LGD( h) LGD( h) p. λ. LGD(0) Change in : were haircut changes too LGD(0) infrequent? Too small in 2011? p : change in counterparty risk? Did it increase? λ : or is it due to a change in CCP default risk?

11 Haircuts increased for GIIPS in Haircuts increased by CCP, by type of GC Haircut (Pct) Spain, Dec Spain, Sept Italy, Nov Date Spain Italy France

12 Haircuts do not reduce stress in Diff-in-diff tests around the three large haircut changes Look at the repo-to-cds spread sensitivity After vs. before the haircut change In country where haircut goes up vs. other countries Stress goes down when haircut goes up in 2010 Not in 2011

13 Change in counterparty risk p 13 p=(p Counterparty Default Sovereign default) Run regression CDS t members = a + b. CDS t GIIPS sov. + c.f t + ε c,t Where F t is a daily risk factor Calculated like in Pan and Singleton (2008) First principle component of CDS changes of 5 European sovereigns Under reasonable assumptions interpret b as a lower bound for p

14 Change in counterparty risk p 14 CDS t members = a + b. CDS t GIIPS sov. + c.f t + ε c,t In fact, p goes down in 2011 Consistent with an overall reduction of exposure to own sovereign by banks, in particular from GIIPS countries

15 Interpretation of results 15 Counterparty risk does not explain the rise in stress This suggests that λ, P (CCP default counterparty & sovereign default), increased in 2011 More evidence Run regression separately for CCP and bilateral trades in 2011

16 Conclusion 16 CCPs work well in times of moderate market stress ( ) Haircut high enough / conditional P(CCP default) 0 CCPs offer little protection in times of high market stress (2011) P(CCP default) very high No additional protection compared to bilateral repo CCPs no panacea in times when they are the most needed?