Вы находитесь на странице: 1из 15

EBS Business School

EBS Universität für Wirtschaft und Recht

Literature Review

Quantifying Risk Contagion with Social Network Analysis

Name: Julian Königseder


Address: Am Hang 4, 92272 Freudenberg
Submitted to: Prof. Ulrich Hommel PhD.
Submission Date: 31.05.2017


Quantifying Risk Contagion with Social Network Analysis i

Quantifying Risk Contagion with Social Network Analysis 1

1 Introduction
On September the 15th 2008, Lehman Brothers went bust. An institution in the
investment banking business for almost 160 years had to leave 28000 people unemployed,
and the world economy lost assets worth nearly 700 Billion Dollars. (Lehmann Brothers,
2008) (Benders, 2008).
This lets the world economy in a huge struggle. Banks had been rescued. Deficit spending
was practiced in high doses, for example, the “Konjunkturpakete” in Germany. Also, the
Euro, a highlight in the after-war economy began to fall rapidly.
After all, we learned from the subprime crises is there a way we can tell what happens if
banks go bankrupt. What happens when even bigger banks like the Deutsche Bank or
Blackrock are losing money? Is there a tool that can quantify risk contagion over the
world when banks, currency, security markets begin to fall?
Some models can help to make such estimates and identify potential consequences in
such situations. The key is social network analysis.
Moreover, because they are so important to look at them the following will describe what
they are, what the situation in research is and how these systems have a practical use. As
Social network analysis are becoming a booming field in the fintech industry and
marketing it is, also interesting to look at this area of study.
This small review will start explaining the basics of social network analysis. In chapter 3
there will be an overview about the literature available. First I will mention the theoretical
background then I will evaluate the field of simulation studies. This chapter will also try
to give an insight of how these studies can be proven. In the 4th chapter this review tries
to give an overview how social network analysis can be implemented to economic policy.
Chapter 5 concludes.

2 Introduction to Social Network Analysis (SNA) and Financial Networks Analysis


(FNA)
Social network analysis is not a new field of study and doing business. The first work
about graphically interpreting social networks was published in 1934 by Jacob L. Moreno.
In his book “Who shall survive” he lays the fundament on which social metrics is based
on. However, this area of science took really of since there are powerful computers to
analyze bigger networks, such as friend lists on Facebook, the relationships in the Bible
or the international finance sector.
Quantifying Risk Contagion with Social Network Analysis 2

There are many ways to use a social network. For example, to analyze “friend relations”
on social media (like Facebook), they can be used in as Communication networks or in
the financial industry to visualize markets or the debt network between banks. This work
will focus on the financial aspect of socio-metrics.
As stated before social networks are there to visualize relations between all parties in a
Network.
Normally they look the same. There are one or more cores, and these are connected to
other dots via lines and nodes. The lines represent the connection between the two parties.
A second use for social networks is to arrange specific parameters for each core and dot,
and test what happens. It is also possible to map a topology.
This, so called, stress testing is also an approach to build a business model and deliver,
as a service, stress test data for costumers. (FNA)
It is possible, for example, build a network of the relationships between the international
community and their trades. This network will have three big cores (China, USA,
Germany) which are connected to every other dot. The thickness of the line states how
much is traded. (Darren, et al.) The graphic for this is in Figure 1.

Figure 1 Network of the Worldtrade From (, Darren; Zoran; Levnajic; Vuk, Janjic; Rasa, Karapandza; Aleksandar,
Stojmirovic; Nataša, Pržulj, 2014)
Quantifying Risk Contagion with Social Network Analysis 3

By analyzing such networks, it is possible to understand the connections between the


different players. The most important factors of analyzing such network are the links, the
distribution and the segmentation.
This in mind FNA is very useful to determent what happens in a particular market
situations (FNA)

3 Study Review

3.1 Theoretical Studies


Over the last decades, several scientists studied the field of financial network analysis
(FNA) and tried to fit FNA into a theoretical context. The intention behind this work was
that it would be more realistic to study the banking system as a network than one bank in
isolation. Also, it was not sufficient to examine the financial sector in a simultaneous
manner, as it is done in game theory.
This thinking had the upside that it was possible to study risk spreading and the
consequences of illiquidity of one bank. Furthermore, it was possible to identify and
explain the danger of a whole network.

3.1.2 Insolvency Contagion


The primary literature in this field of study was published by Allen & Gale (2000). In
their article “Financial Contagion” they exposed the how banks in four different regions
reacted when the faced negatively correlated liquidity shocks.
In this particular situation, it could be optimal for banks to hold deposits in over regions.
However, as a financial crisis strikes in one region, the deposit of the other banks loses
value and can trigger crises in the other areas. Allen & Gale call this an incomplete
network.
For example, there is a system with four different regions (e.g. Germany, USA, Angola,
Galapagos Islands). Now a finical shock strikes in Germany and the American bank has
deposited in Germany; the Angolan Bank deposits in the USA and the Galapagos Bank
deposits in Angola. The deposits of the American bank depreciate and cause crises in the
U.S then it could happen that, even the bank on the Galapagos Islands, which is not
connected to the USA, can struggle.
Quantifying Risk Contagion with Social Network Analysis 4

This could be prevented if the network would be a complete network. The full network is
far more resilient and shock absorbing as the incomplete one. In this kind of network, all
banks have deposits in every other bank (see figure 2 for a visualization).



Figure 2 Complete and Incomplete Network From (Allen & Gale, 2000)

This can prevent financial crises in the other regions.


Although it is tough to generalize the results of the paper, partly because the real banking
system is far more complicated, as it is graphically shown in figure 3. Furthermore, the
results could not be reproduced. (Boss, Krenn, Puhr, & Summer, 2006).

3.1.1 Liquidity Gridlock


Around the time when Allen & Gale studied, FNA also Freixas & Rochet published a
paper about FNA. In their paper “Systemic Risk, Interbank Relations and Liquidity
Provision by the Central Bank” they used different banks in the various regions, likewise
Allen & Gale, and modeled them by the framework introduced in 1983 by Diamond &
Dybivig. The depositors can place investments in their home region. At a systemic level,
the banks can give credit lines to people who travel, so there is traffic between the banks.
This leads to coordination problems and multiple equilibria.
The first one where liquidity provision on system level emerge and a second one where
bank runs are happening.
They conclude that a consequence of a network can be a decile in market disciple because
banks feel safer.
The biggest problem with both studies is that a more general analysis is tough. Even
attempts were made, but theoretical literature is still very spare.

3.2 Simulation Studies


Like stated before it is tough to proof these theoretical studies. One way would be to
observe those domino effects in real life. However, there is the problem that in the most
Quantifying Risk Contagion with Social Network Analysis 5

cases the government steps into action and rescues banks, large enterprises or even whole
countries. This leads to a lack of uninfluenced observation.
To get around this problem, Furfine introduced so-called simulation studies. In the paper
“Interbank Exposures: Quantifying the Risk of Contagion” he tried to analyze payment
systems. Published papers quickly turned their head to banking systems. Several papers
were written about simulation model for banking systems. To biggest impact had Elsinger,
Lehar, & Summer in their work around 2006. This is also the study on which the next
parts are based.

3.2.1 The Simulation Model


This section is introducing a basic model of a banking system. It is build on Eisenberg &
Noe, 2001, Elsinger, 2009, Elsinger, Lehar, & Summer, 2013, Elsinger, Lehar, &
Summer, 2006, and Elsinger, Lehar, & Summer, 2006. Although it would be to much
content for such a small review, this section is trying to give an insight how these models
work.
First, assume that there is a banking system with banks, where banks denote in ! and ! ∈
# These banks have assets and liabilities. On the asset side, there are non inter-bank
assets $%&'( , where $%&'( is independent of the network, and interbank assets $%'( . On the
other side of the balance sheet, there are liabilities which are also divided by noninterbank
liabilities )%&'( and there are interbank liabilities )%'( .. Assume that there is only one
seniority.
Now where every variable is denoted it is possible to build a model.
The structure is represented by a # + # matrix ,. In this matrix -%. , where / ∈ # and -%. ≥
0, represents the liabilities between banks. Also, the diagonal elements equal zero.
Therefor,
7 7

-%. = $.6'( and -%. = )%6'(


%89 .89

where $.6'( and )%6'( are denoting the nominal values of the claims from other banks and
liabilities in contrast to $%'( and )%'( .
All studies working on this topic use this as a base. Even there are two types of studies
available, one in which an insolvency event is modeled an one where the insolvency is
Quantifying Risk Contagion with Social Network Analysis 6

the base, and it is studied what happens to the system. (Upper, 2011). See Alessandri,
Gai, Kapadia, Mora, & Puhr, 2009 and Upper, 2011 for a deeper dive into the world of
simulation studies.
One of the biggest threats is the data source to feed the models and run the simulations.
Although we live in a world of big data, banks do not like it to publish all their information.
In some countries, e.g. Germany, there are loan registers. If this kind of data is available
is possible to run Monte Carlo shock scenarios with $%&'( .
This is the best way to observe the loss resulting from a financial crisis. Assume of the
bank loans as Bernoulli-distributed random variables. By this assumption, it is possible
to conduct a Monte Carlo simulation, where loans either get paid or result in a total loss.
Sure this is not like in reality, where credits are often partly paid, but is a simplified
method to predict loss scenarios. This method used by both Elsinger, Lehar, & Summer,
Risk Assessment for Banking Systems., 2006 and Gauthier, Lehar, & Souissi, 2012.
As stated before, not all countries have these loan registers. Although it is possible to
conduct testing and simulation. The key element it stock market data. It is possible to
estimate market values, volatilities, drift parameters and the correlation between stock
prices.
Using this information Lehar (2005) developed a system that could estimate future asset
values. Furthermore, Elsinger, Lehar, & Summer (2006) used his work to simulate the
systemic risk of the British banking network.
All these models are very precise, although they need a lot of computing power, in
normal cases, they underestimated the effect of contagion.
These models where used more over time and got more and more complex.

3.2.2 Verification of Empirical Findings


As mentioned before, it is very hard to observe these impacts, because most of the time a
bank, a “system relevant” company or a state goes bankrupt or has to close its doors the
government steps in and rescues the bank, company or country, at least in Europe.
However, the discussion about this is not the core of this work.
Some papers tried to confirm the simulation models. Especially Upper (2011)
summarized the methods and found out that insolvency contagion is caused by interbank
exposure.
A remarkable effort to this field of study was also made by the Austrian National Bank,
where Helmut Elsinger did a lot of research during his work for the OeNB. He concluded
Quantifying Risk Contagion with Social Network Analysis 7

that it is nearly impossible to model scenarios where there is a significant amount of


contagion. (Elsinger, Lehar, & Summer, Risk Assessment for Banking Systems., 2006)
(Elsinger, Lehar, & Summer, 2006).
As it turns out that it is hard to validate this information, it was implemented to the
systemic analysis of the OeNB (see Boss, Krenn, Puhr, & Summer, 2006).

3.2.3 The next Step


The main issue of all these model, developed before the US subprime crisis, was to see
what happens when an actual crisis happens and which institutes are necessary to rescue.
Then the subprime crisis striked and all models were proven wrong.
Studies predicted that it would need 30% losses along the whole system so that domino
effect contagion becomes a thing to deal with.
Eventually, nobody expected that the subprime crisis and that following the Euro-crisis
would tear down the modern banking business, or even whole countries (Greece, Portugal,
Italy, Spain, Ireland). The estimated loss in the year 2008 was 945.000.000.000 Dollar -
which is, to be fair, much money – but got even more in the year 2009 where the IMF
estimated the losses at 4.000.000.000.000 Dollar, this is also far less what it had cost the
world (Wallstreet Online, 2013).
However, in financial terms, this is not quite so much. For example, the balance sheet of
the Deutsche Bank had around 6,000 Billion Dollar in assets.
So why did these models all failed to predict this loss?
Well as it turns out, it is not “fully understood even at theoretical level” (Glasermann &
Young, 2016).
The problem is that the theoretical background was studied before the crises, even though
many studies were published after the financial crises, it needs more studies to understand
this topic better. (Glasermann & Young, 2016)
The next step would be to do more research about this topic and to implement basic
network theory into economic policy.
A step further would be to implement social network analysis into banks and companies
to estimate the risk of assets.

4 Economic Policy
Although it seems to be very practical to use network analysis to build a modern economic
policy, this method is barely used.
Quantifying Risk Contagion with Social Network Analysis 8

The biggest problem seems to be that more work should be done on this topic to develop
more precise models. However, with more precision more computing power is needed.
This leads to the need of more resources at supercomputers, which are very rare.
However, there are some aspects of which FNA could be used to make a better economic
policy.
The most important thing is to identify critical institutions, which even when little shocks
happen are likely to be unable to pay their debts or pay out their customers.
If it would be possible to map the international financial network it may be possible to
find out which bank or company must be rescued first. (Amini, Cont, & Minca, 2012)
As a possible way to conduct such a survey would be a so-called “threat index” introduced
by Demange (2012).
With this index, it is feasible to view the reactions of a financial network, and it is possible
to see which bank is the first to collapse.
As it is not optimal to target the weakest institutions in the market, but that with the
highest threat index.
On first sight, this seems to be very unlogical. However, if we assume that the Deutsche
Bank is not the weakest firm on earth, mainly because the German government would
rescue it, but if assume that it has a very high threat index, so more regulations can target
that bank. Also, weak firms can breathe more and maybe get stronger. Another
opportunity of FNA is stress testing. Currently only used at small companies it would be
very interesting to see what happens if the authorities look at the banking business at a
whole and not only as single bank.

5 Conclusion
Financial Network Analysis is a fascinating tool to analyze banking systems. As stated
above, it is useful to analyze banks in relation to other banks and not as single.
However, it is very hard to prove the models. First, there is very less data what happens
when one institute fails. This is because we rescue system-relevant banks. The second
thing is that these models did not predict the impact of the financial crisis.
Thus, the models are not sufficient to build a modern economic policy, with all their
effects on banks.
Another thing to bear in mind is that FNA is a very young field of study. The most models
that we use to explain the market are very old. For instance, the market model, one of the
best to describe what happens in capitalistic market is over 120 years old. Maybe we work
Quantifying Risk Contagion with Social Network Analysis 9

in 120 years’ time more with this model. Nevertheless, FNA is a big step in understanding
how the banking system works. Nowadays it is needed to survey and map the massive
amount of data produced every day in a systemic way to see what is happening. One more
thing is the use of FNA by private firms. One of the leaders in this field is Financial
Network Analytics Ltd., a London based company. In 2015, they released a Platform to
monitor and visualize risk and shocks at stock and currency markets.
I like this idea very much because it is not only useful to study this topic and get better
estimates what happens in a crisis, it is also useful to use the results we got from older
studies. By this path also private firms, like small investment banks without the
computing power of a central bank, can use network analysis.
In conclusion, I follow the most recent reports and studies about this topic. The topic is
very young and it needs a lot of computing power to get proper results. Furthermore, the
financial crisis was a big problem for this field of study. The fact that the most models
could not predict the tremendous impact that it had was enormous blow for FNA.
Quantifying Risk Contagion with Social Network Analysis 10

Reference List
Alessandri, P., Gai, P., Kapadia, S., Mora, N., & Puhr, C. (2009). Towards a Framework
for Quantifying Systemic Stability. International Journal of Central Banking.
Allen, F., & Gale, D. (2000). Financial Contagion. The Journal of Political Economy, 1-
34.
Amini, H., Cont, R., & Minca, A. (2012). Stress Testing the Resilience of Financial
Networks. International Journal of Theoretical and Applied Finance.
Benders, W. (15. 09 2008). Lehman Brothers muss Konkurs beantragen . Handelblatt.
Boss, M., Krenn, G., Puhr, C., & Summer, M. (2006). Systemic Risk Monitor: A Model
for Systemic Risk Analysis and Stress Testing of Banking Systems. Financial
Stability Report.
Bundeszentrale für politsche Bildung. (kein Datum). bpb. (B. f. Bildung, Herausgeber)
Abgerufen am 12. 02 2017 von http://www.bpb.de/nachschlagen/lexika/lexikon-
der-wirtschaft/20275/offene-volkswirtschaft
Cifuentes, R., Ferrucci, G., & Shin, H. S. (2005). Liquidity risk and contagion. (B. o.
England, Hrsg.) BoE Working Papers.
Cohen-Cole, E., Patacchini, E., & Zenou, Y. (2010). Systemic Risk and Network
Formation in the Interbank Market. CAREFIN Research Paper .
Cont , R., & Wagalath, L. (2016). Fire Sales Forensics: Measuring Endogenous Risk.
Mathematical Finance.
Cont, R., & Wagalath, L. (2013). Running for the Exit: Distressed Selling and
Endogenous Correlation in Financial Markets. Mathematical Finance.
Darren, Zoran, Levnajic, Vuk, J., Rasa, K., Aleksandar, S., & Nataša, P. (kein Datum).
Rasas arbeit.
Demange, G. (2 2012). Archive ouverte en Sciences de l'Homme et de la Société.
Abgerufen am 25. 05 2017 von Contagion in financial networks: a threat index:
https://hal.archives-ouvertes.fr/file/index/docid/662513/filename/wp201202.pdf
Diamond , D., & Dybvig, P. (1983). Bank Runs, Deposit Insurance, and Liquidity. The
Journal of Political Economy.
Quantifying Risk Contagion with Social Network Analysis 11

Dierig, C. (2. 02 2016). Die Welt. (W. GmbH, Herausgeber) Von


https://www.welt.de/wirtschaft/article152863219/Was-hat-Amerika-was-
Frankreich-fehlt.html abgerufen
Education Group GmbH. (kein Datum). Geschäftsprozesse Aufbauoptimierung. (E. G.
GmbH, Hrsg.) Abgerufen am 27. 11 2016 von Traditionelle Formen der
Aufbauorganisation: http://www.eduhi.at/dl/trad_aufb.pdf
Eisenberg, L., & Noe, T. (2001). Risk in Finacial Systems. Management Science.
Elsinger, H. (2009). Financial Networks, Cross Holdings, and Limited Liability . (A. N.
Bank, Hrsg.) OeNb Working Papers .
Elsinger, H., Lehar, A., & Summer, M. (2006). Risk Assessment for Banking Systems.
Management Science.
Elsinger, H., Lehar, A., & Summer, M. (2006). Using Market Information for Banking
System Risk Assessment. International Journal of Central Banking.
Elsinger, H., Lehar, A., & Summer, M. (2013). Network Models and Systemic Risk
Assessment. Cambridge, Great Britain: Cambridge UniversityPress.
Freixas, X., & Rochet, J. C. (2000). Systemic Risk, Interbank Relations and Liquidity
Provision by theCentral Bank. DNB Staff Reports, 39.
Furfine, C. (1999). INTERBANK EXPOSURES: QUANTIFYING THE RISK OF
CONTAGION. (B. f. Settlements, Hrsg.) BIS Working Papers.
Gai, P., & Kapadia, S. (2010). Contagion in Financial Networks. Proceedings of the Royal
Society.
Gauthier, C., Lehar, A., & Souissi, M. (2012). Macroprudential Capital requirements and
Systemic Risk . Journal of Financial Intermediation.
Georgi, C. P. (2016). Principles of Management Managing: Organisational Structure and
Culture. Slides of the Lecture. Östrich-Winkel.
Glasermann, P., & Young, P. (2016). Contagion in Financial Networks. Journal of
Economic Literature.
Gorten, G., & Metrick, A. (2012). Getting up to Speed on the Financial Crisis: A One-
Weekend-Reader’s Guide. Journal of Economic Literature.
IMF . (2008). Global Financial Stability Report: Financial Stress and Deleveraging:
Macro-Financial Implications and Policy.
Jones, G., & George, J. (2014). Contemporary management. New York, USA: McGraw-
Hill Education.
Quantifying Risk Contagion with Social Network Analysis 12

Lehar, A. (2005). Measuring Systemic Risk: A Risk Aanagement Approach. Journal of


Banking & Finance.
Lehmann Brothers . (2008). 2007 Anual Report. Anual Report, Lehmann Brothers, New
York.
Mohr, D. (9. 02 2017). DEStatis. (S. Bundesamt, Herausgeber) Abgerufen am 12. 02 2017
von
https://www.destatis.de/DE/PresseService/Presse/Pressemitteilungen/2017/02/P
D17_045_51.html
Nöcker , R. (17. 06 2006). FAZ.net. (F. A. Zeitung, Hrsg.) Abgerufen am 27. 11 2016
von Matrix-Organisationen Albtraum mit zwei Dimensionen:
http://www.faz.net/aktuell/beruf-chance/matrix-organisationen-albtraum-mit-
zwei-dimensionen-1331078.html
Reuters, Die Zeit. (19. 01 2017). Die Zeit Online . (D. Zeit, Herausgeber) Abgerufen am
12. 02 2017 von http://www.zeit.de/wirtschaft/2017-01/brexit-banken-
verlagerung-arbeitsplaetzen-eu
Shin, H. S. (2010). Risk and Liquitity. Oxford, GB: Oxford Univerity Press.
Upper, C. (2011). Simulation methods to assess the danger of contagion in interbank
markets. Journal of Financial Stability.
Wallstreet Online. (09. 09 2013). Kosten der Finanzkrise Deutschland einer der größten
Krisen-Verlierer. Abgerufen am 29. 05 2017 von Wallstreet Online:
http://www.wallstreet-online.de/nachricht/6322659-kosten-finanzkrise-
deutschland-groessten-krisen-verlierer