1 sovereign, bank, and insurance credit spreads: connectedness and system networks m. billio, m....

Post on 18-Dec-2015

220 Views

Category:

Documents

2 Downloads

Preview:

Click to see full reader

TRANSCRIPT

1

Sovereign, Bank, and Insurance Credit Spreads: Connectedness and System

Networks

M. Billio, M. Getmansky, D. Gray A.W. Lo, R.C. Merton, L. Pelizzon

The research leading to these results has received funding from the European Union, Seventh Framework Programme FP7/2007-2013

under grant agreement SYRTO-SSH-2012-320270.

Funded by the European Union7th Framework Programme (FP7)

SYRTO

2

Objectives

• The risks of the banking and insurance systems have become increasingly interconnected with sovereign risk

• Highlight interconnections: • Among countries and financial institutions • Consider both explicit and implicit connections

• Quantify the effects of:• Asset-liability mismatches within and across

countries and financial institutions

3

Methodology

• We propose to measure and analyze interactions between financial institutions, sovereigns using:

– Contingent claims analysis (CCA)

– Network approach

4

Background

• Existing methods of measuring financial stability have been heavily criticized by Cihak (2007) and Segoviano and Goodhart (2009):

• A good measure of systemic stability has to incorporate two fundamental components: – The probability of individual financial

institution or country defaults– The probability and speed of possible shocks

spreading throughout the industry and countries

5

Background

• Most policy efforts have not focused in a comprehensive way on: – Assessing network externalities – Interconnectedness between financial institutions,

financial markets, and sovereign countries – Effect of network and interconnectedness on

systemic risk

Background: Feedback Loops of Risk from Explicit and Implicit Guarantees

Source: IMF GFSR 2010, October Dale Gray

6

7

Background

• The size, interconnectedness, and complexity of individual financial institutions and their inter-relationships with sovereign risk create vulnerabilities to systemic risk

• We propose Expected Loss Ratios (based on CCA) and network measures to analyze financial system interactions and systemic risk

Core Concept of CCA: Merton Model

• Expected Loss Ratio = Cost of Guar/RF Debt

= PUT/B exp[-rT] = ELR

• Fair Value CDS Spread = -log (1 – ELR)/ T

8

9

Moody’s KMV CreditEdge for Banks and Insurance Companies

• MKMV uses equity and equity volatility and default barrier (from accounting information) to get “distance-to- distress” which it maps to a default probability (EDF) using a pool of 30 years of default information

• It then converts the EDF to a risk neutral default probability (using the market price of risk), then using the sector loss given default (LGD) it calculates the Expected Loss Ratio (EL) for banks and Insurances:

EL Ratio = RNDP*LGDSector

• It calculates the Fair Value CDS Spread

Fair Value CDS Spread = -1/T ln (1 – EL Ratio)

Why EL Values?

• EL Values are used because they do not have the distortions which affect observed CDS Spreads

• For banks and some other financial institutions:• The fair-value CDS spreads (implied credit spreads

derived from CCA models, i.e. derived from equity information) are frequently > than the observed market CDS

• This is due to the depressing effect of implicit and explicit government guarantees

Why EL Values?

• In other cases, e.g. in the Euro area periphery countries, bank and insurance company CDS appear to be affected by spillover from high sovereign spreads (observed CDS > FVCDS).

• For these reasons we use the EL associated with the FVCDS spreads for banks and insurance companies which do not contain the distortions of sovereign guarantees or sovereign credit risk spillovers

Sovereign Expected Loss Ratio

• CCA has been applied to sovereigns, both emerging market and developed sovereigns

• Sovereign CDS spreads can be modeled from sovereign CCA models where the spread is associated with the expected loss value and sovereign default barrier

• For this study the formula for estimating sovereign EL is simply derived from sovereign CDS

EL Ratio Sovereign = 1-exp(-(Sovereign CDS/10000)*T)

• EL ratios for both banks and sovereigns have a horizon of 5 years (5-year CDS most liquid)

Linear Granger Causality Tests

ELRk (t) = ak + bk ELRk(t-1) + bjk ELRj(t-1) + Ɛt

ELRj(t) = aj + bj ELRj(t-1) + bkj ELRk(t-1) + ζt

• If bjk is significantly > 0, then j influences k

• If bkj is significantly > 0, then k influences j

• If both are significantly > 0, then there is feedback, mutual influence, between j and k.

13

Data

• Sample: Jan 01-Mar12• Monthly frequency• Entities:

– 17 Sovereigns (10 EMU, 4 EU, CH, US, JA)– 59 Banks (31EMU, 11EU, 2CH, 12US, 4JA)– 42 Insurance Companies (12EMU, 6EU, 16US,

2CH, 5CA)• CCA - Moody’s KMV CreditEdge:

– Expected Loss (EL)

Slide 15

Mar 12

Blue InsuranceBlack SovereignRed Bank

Blue InsuranceBlack SovereignRed Bank

Slide 16

Mar 12

Blue InsuranceBlack SovereignRed Bank

Blue InsuranceBlack SovereignRed Bank

Network Measures

• Degrees

• Connectivity

• Centrality

• Indegree (IN): number of incoming connections • Outdegree (FROM): number of outgoing connections • Totdegree: Indegree + Outdegree

• Number of node connected: Number of nodes reachable following the directed path

• Average Shortest Path: The average number of steps required to reach the connected nodes

• Eigenvector Centrality (EC): The more the node is connected to central nodes (nodes with high EC) the more is central (higher EC)

Slide 18

Network Measures: FROM and TO Sovereign

17 X 102= 1734 potential connections FROM (idem for TO)

Slide 19

From GIIPS minus TO GIIPS

June 07

Blue InsuranceBlack SovereignRed Bank

March 08

Blue InsuranceBlack SovereignRed Bank

August 08

GreeceBlue InsuranceBlack SovereignRed Bank

SpainBlue InsuranceBlack SovereignRed Bank

December 11

March 12US

Blue InsuranceBlack SovereignRed Bank

IT

March 12

Blue InsuranceBlack SovereignRed Bank

Early Warning SignalsJa

n01_

Dec0

3M

ar01

_Feb

04M

ay01

_Apr

04Ju

l01_

Jun0

4Se

p01_

Aug0

4No

v01_

Oct

04Ja

n02_

Dec0

4M

ar02

_Feb

05M

ay02

_Apr

05Ju

l02_

Jun0

5Se

p02_

Aug0

5No

v02_

Oct

05Ja

n03_

Dec0

5M

ar03

_Feb

06M

ay03

_Apr

06Ju

l03_

Jun0

6Se

p03_

Aug0

6No

v03_

Oct

06Ja

n04_

Dec0

6M

ar04

_Feb

07M

ay04

_Apr

07Ju

l04_

Jun0

7Se

p04_

Aug0

7No

v04_

Oct

07Ja

n05_

Dec0

7M

ar05

_Feb

08M

ay05

_Apr

08Ju

l05_

Jun0

8Se

p05_

Aug0

8No

v05_

Oct

08Ja

n06_

Dec0

8M

ar06

_Feb

09M

ay06

_Apr

09Ju

l06_

Jun0

9Se

p06_

Aug0

9No

v06_

Oct

09Ja

n07_

Dec0

9M

ar07

_Feb

10M

ay07

_Apr

10Ju

l07_

Jun1

0Se

p07_

Aug1

0No

v07_

Oct

10Ja

n08_

Dec1

0M

ar08

_Feb

11M

ay08

_Apr

11Ju

l08_

Jun1

1Se

p08_

Aug1

1No

v08_

Oct

11Ja

n09_

Dec1

1M

ar09

_Feb

12

0

1000000

2000000

3000000

4000000

5000000

6000000

7000000

8000000

9000000

10000000

0

2000

4000

6000

8000

10000

12000

14000

EL # of lines

forecast

forecast

26

t=March 2008 t+1=March 2009; t = Jul 2011; t+1= Feb 2012Cumulated Exp. Loss ≡ Expected Loss of institution i + Expected losses of institutions caused by i

Early Warning Signals

Cumulative lossesMarch 09 February 12

Coeff t-stat R-square Coeff t-stat R-square# of in line# of out lines 0.40 2.92 0.23 2.2# of lines 0.87 3.5Closeness Centrality -0.63 -2.51 -0.15 -7.0

Eigenvector Centrality -0.15 -4.4 0.17 0.42

27

28

CDS data

29

Comparison CDS-KMV

30

Comparison CDS-KMV

31

CDS: Dec 11Spain

Blue InsuranceBlack SovereignRed Bank

Slide 32

Spain

Dec 11 : EL-KMV

Blue InsuranceBlack SovereignRed Bank

Slide 33

Blue InsuranceBlack SovereignRed Bank

CDS:Mar 12

IT

Mar 12:EL-KMV

US

Blue InsuranceBlack SovereignRed Bank

IT

35

Conclusion

• The system of banks, insurance companies, and countries in our sample is highly dynamically connected

• Insurance companies are becoming highly connected…

• We show how one country is spreading risk to another sovereign

• Network measures allow for early warnings and assessment of the system complexity

36

Thank You!

top related