1. The next Financial Hurricane
What can complex networks teach us
about financial fragility and systemic risk?
Bath Royal Literary and Scientific Society
Bath, 08/03/17
Antonio Cabrales
University College London

2. Structure of presentation
Introduction
Causes of financial crises
The financial system as a complex network
How to manage risk sharing versus contagion
Concluding comments

3. I. Introduction
Financial crises arise when some financial institutions or assets suddenly lose a large part of their value.
There are a number of different types of crises:
Banking crises (runs or related difficulties).
Speculative bubbles and crashes (stock markets, real estate)
Currency crises; isolated crises and contagion.
Systemic crises: a large number of institutions or assets behave in a non-sustainable way.

4. Consequences of systemic crises
Peak to trough measures (Reinhard and Rogoff 2009)
Real house prices:
average fall 35.5 percent, average duration 6 years.
Real equity prices:
average fall 55.9 percent, average duration 3.4 years.
Real per capita GDP:
average fall 9.3 percent, average duration 1.9.
Unemployment:
average rise 7 percent, average duration 4.8 years.
Increase in public debt (3 years after a banking crisis) :
average rise 86 percent.

5. II. Why do financial crises occur. II
II. Why do financial crises occur? II.1 Financial markets are naturally fragile
Intermediation of funds creates asset-liability mismatches
Banks provide liquidity to depositors as the timing of use of funds is uncertain
Banks earn by lending to illiquid long-term investments
Asset-liability mismatch can lead to bank runs
These can be a self-fulling prophecy: depositors/lenders withdraw in response to (expected) withdrawals by others.
Investors choices are strategic complements: incentives to coordinate decisions => strategic complementarity

6. Imperfect knowledge and limitations in human reasoning:
II.2 Amplifying factors
Imperfect knowledge and limitations in human reasoning:
Overestimated assets values.
Learning and herding lead to more volatility.
High leverage contributes to financial crises.
If funds are borrowed to invest more, then potential gains increased but also potential losses.
Moreover, bankcruptcy risk arises => contagion and increased systemic risk.
Regulatory failures
Movement of liabilities off balance sheets via securization.
CDS and other OTC markets non-transparent.
Basel II led to procyclicality.
Outright fraud (Madoff case).

7. II.3 Trade-offs: risk sharing versus contagion
Financial systems have important trade-offs:
Deeper intermediation and markets improve allocation of resources and of risks.
Imperfect information creates liquidity problems when confidence is lost.
Asymmetric information creates incentives that contribute to adverse outcomes (moral hazard, adverse selection).
Possibility of bad outcomes from asset-liability mismatches that become sour.

8. III. The financial sector as a complex network
A complex network is a set of relationships (links) between entities (nodes).
It can be represented graphically, or
analytically, aij = the strength of exposure of entity i to j.
In our case, entities are firms, relationships are financial exposures to one another.

9. III. The financial sector as a complex network
A bank fails if liabilities exceed her capital plus net revenues, typically because net revenues become very small or negative as investments do not pay or have losses.
Alexei Karas, Gleb Lanine, Koen Schoors

10. the combined roles of network topology and shock characteristics
Analytical results
Numerical extension
Empirical context

11. the combined roles of network topology and shock characteristics$
$+2% XX $
$+4%

12. the combined roles of network topology and shock characteristics
Diversification 7 6 8 $
$+2% XX XX $
$+1%

13. the combined roles of network topology and shock characteristics
Contagion 7 6 8 $
$+2% XX XX $
$-98%

14. the combined roles of network topology and shock characteristics
Link density
Modularity
Size and distribution heterogeneity
Degree heterogeneity

15. the combined roles of network topology and shock characteristics
Size distribution
Target
Small tailed, α >1
Biggest firm
Most connected
Random
Fat tailed, α <1
log10(shock size)

16. the combined roles of network topology and shock characteristics
Analytical results
Cabrales A, Gottardi P & Vega-Redondo F (2017)
Risk-sharing and contagion in networks.

17. the combined roles of network topology and shock characteristics
Analytical results (Cabrales et al. 2017)
Modularity
<
> +
Intermediate Group Size

18. the combined roles of network topology and shock characteristics
Analytical results (Cabrales et al. 2017)
Density
> = +
<

19. the combined roles of network topology and shock characteristics
Analytical results (Cabrales et al. 2017)
Distribution-Heterogeneity
With
With
Never Intermediate – separate investment-commercial banks

20. the combined roles of network topology and shock characteristics
Take home message from analytics
The standard approach of stress test is flawed.
Looks at one big shock, whole distribution matters.
Some general results on exactly what matters.
Historical shocks to know what Works.
Regulation can sometimes distort the optimal choice.
E.g. protection to commercial Banks makes
Investment want to join.
Banks become too “embroiled” because “too connected
to fail.

21. the combined roles of network topology and shock characteristics
Link density
Modularity
Capital requirements
Size heterogeneity
Degree heterogeneity

22. the combined roles of network topology and shock characteristics
Numerical extension
Loepfe, L., Cabrales, A., & Sánchez, A. (2013). Towards a proper assignment of systemic risk: the combined roles of network topology and shock characteristics. PloS one, 8(10), e77526.

23. the combined roles of network topology and shock characteristics
Modularity
Random
Mean defaults
Degree
Size

24. the combined roles of network topology and shock characteristics
Connectivity
Random
Mean defaults
Degree
Size

25. the combined roles of network topology and shock characteristics
Modularity
Connectivity

26. the combined roles of network topology and shock characteristics
Degree heterogeneity
Random
Mean defaults
Degree
Size

27. the combined roles of network topology and shock characteristics
Size heterogeneity
Random
Mean defaults
Degree
Size

28. the combined roles of network topology and shock characteristics
Capital requirements
Random
Mean defaults
Degree
Size

29. the combined roles of network topology and shock characteristics
Capital requierments

30. the combined roles of network topology and shock characteristics
Capital requirerments
Monte Carlo
Simulations

31. the combined roles of network topology and shock characteristics

32. the combined roles of network topology and shock characteristics
Standarized Regression Coefficients

33. the combined roles of network topology and shock characteristics

34. the combined roles of network topology and shock characteristics
Pure small tailed
4% fat tailed

35. the combined roles of network topology and shock characteristics

36. the combined roles of network topology and shock characteristics
Empirical context
We thank Stefania Vitali and Stefano Battiston
for sharing the Orbis data on corporate ownership.

37. the combined roles of network topology and shock characteristics
global banking network
Hale 2012
Connectivity

38. the combined roles of network topology and shock characteristics
corporate ownership network (core)
random
Highest
betweenness

39. the combined roles of network topology and shock characteristics
Conclusions
trade-off between diversification and contagion, need to know whole distribution.
transition from safe to risky regimes can be very sharp
small fraction of large shocks enough to change optimal configuration.
policies should target not only the firms, but also the links between them and whole structure/system.
the tools we surveyed can be used systematically.

40. the combined roles of network topology and shock characteristics
Thank you for your attention