Slides

Mapping the UK interbank system
Given at Zurich, 14/09/12
Sam Langfield, (1) Zijun Liu (2) and Tomohiro Ota (3)
(1) UK Financial Services Authority and European Systemic Risk Board Secretariat. Email: samlangfield@gmail.com
(2) UK Financial Services Authority. Email: zijun.liu@fsa.gov.uk
(3) Bank of England. Email: tomohiro.ota@bankofengland.co.uk

What the UK interbank system looks like
Why it looks like this

‘focus on the wood, not the trees’

9,700,000
elements in the cross-section

163 variables:
prime lending (unsecured, secured)
equity
fixed income
CDS (bought, sold)
SFT, incl repo
derivatives (by underlying)
et al
by maturity (O/N,

Exposure Network
Panel A: Exposures Network:
Breakdown by Instrument
Funding Network
Panel B: Funding Network:
Breakdown by Instrument
derivati
ves
44%
unsecur
ed
24%
secured
4%
marketa
ble
16%
repo
66%
unsecure
d
29%
secured
5%
sft
11%
cds
-1%
Total exposures: £264bn
Total funding: £192bn

Exposure Network by Bank Type
UK
Commercial
Banks
Overseas
Banks
Panel B: Funding Network:
Breakdown by Instrument
Large UK
Banks
Derivative (51%)
Non-reporting
Derivative (37%)
Mkt securities (19%)
Building
Societies
Investment
Banks
Derivative (46%)
SFT (35%)
Note: Some small links are not shown in the graph.

Funding Network by Bank Type
UK
Commercial
Banks
Unsec Loans (86%)
Overseas
Banks
Panel B: Funding Network:
Breakdown by Instrument
Repo (64%)
Large UK
Banks
Unsec Loans (57%)
Non-reporting
Building
Societies
Investment
Banks
Repo (98%)
Note: Some small links are not shown in the graph.

Degree distribution
Figure 9
Degree: Exposures vs Funding Network
0 5
10 15 20 25
0 20 40 60 80 100
In-degree in Funding Network
UK banks Building societies Investment banks
Overseas banks Non-reporting banks
Markers' area are proportional to within-quintile mean of banks' total global liabilities
Sample: all 490 banks

Weighted degree distribution
Figure 10
Market Share: Exposures vs Funding Network
0
.05
.1
.15
0 .05 .1 .15
Funding Network: Market Share
UK banks Building societies Investment banks
Overseas banks Non-reporting banks
Markers' area are proportional to within-quintile mean of banks' total global liabilities
Sample: all 490 banks

Structure (1): Interbank Exposures
Generated using Financial Network Analytics

Structure (2):
Hub-spoke / Core-periphery
also see:
Borgatti et al (2002)
Craig and von Peter (2010)

No. of banks in core
Fitness of core-periphery model
No. of banks in core
Fitness of core-periphery model
Structure (3): Core-periphery fitness
14
12
10
8
Panel A:
Exposures networks
No. of banks in core (LHS)
Fitness of core-periphery model (RHS)
0.8
0.7
0.6
0.5
0.4
14
12
10
8
Panel B:
Funding networks
0.8
0.7
0.6
0.5
0.4
6
0.3
6
0.3
4
0.2
4
0.2
2
0.1
2
0.1
0
0
0
0

Broad Conclusion:
Interbank system is incomplete
and highly concentrated
Similar findings:
• Belgium (Degryse and Nguyen, 2010)Market-makers
absorbing demand-supply gaps
• Germany (Sacks, 2010; Craig and von Peter, 2010)
• Italy (Fricke and Lux, 2012)
• Mistrulli (2011)
• Degryse and Nguyen (2010)

Interpretation:
Why do we have core-periphery?
• A specialist collecting info on behalf of
others (Diamond-Dybvig monitoring)
• Market-makers and search costs (Duffie et
al, 2005)
• Star-shape is the efficient way to be
connected (Jackson and Wolinsky, 1996)

Source: BoE MFI stats
One final thought

Appendix

Interpretation (3):
Resilience: conflicting ideas
• Complete vs incomplete network: which is risky?
• Is core-periphery network risky?
• Resilient because core bank becomes a fire-stop
• Peripherals become more vulnerable against core
• Depends on size, depends on the number of links
• Identifying weakest links (Cont et al, 2011):
• 173 Contagious Links which can “kill” exposed bank
• 65 (out of 176) banks are exposed to contagious link

Structure (4):
Core composition
7
6
5
4
3
2
1
0
No. markets bank is in core
8
Red: UK banks
Blue: Investment banks
Yellow: Overseas banks
Individual banks

Testing Maximum Entropy
• Q: How reliable is the algorithm to populate
incomplete matrices?
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0% 5% 10% 15% 20% 25% 40% No Info

Maximum Entropy Algorithm:
How reliable is the method?
• Now we have an answer: the dataset is nearly
complete
Generate several incomplete matrices by
eliminating exposures smaller than x% of capitals
1. Populate the matrices by ME and RAS algorithms
2. Calculating Pearson correlations with the original
3. F test to check the Null hypothesis that ME method
does not improve the estimation
• populating the incomplete matrices by random numbers
for 1000 times to generate a distribution to compare
• Applying RAS for the randomly assigned matrices

k 1
bk 2
bk 3
bk 1
bk 2
bk 3
Why does ME work well?
• Hypothesis: ME is prone to expect coreperiphery
structure (even when it isn’t in fact)?
1
1
2 3
Actual matrix
2 3
ME’s estimate from total exposures
bank 1 0 0.6 0 0.6
bank 2 0 0 0.3 0.3
bank 3 0.1 0 0 0.1
0.1 0.6 0.3 1
bank 1 0.06 0.36 0.18 0.6
bank 2 0.03 0.18 0.09 0.3
bank 3 0.01 0.06 0.03 0.1
0.1 0.6 0.3 1