Joint presentation with Michelle Tuveson and Dr Andrew Coburn from Cambridge Risk Center at the Conference Board Global Risk Conference in New York, 8 May 2013.
Links to conference website: http://www.conference-board.org/conferences/conferencedetail.cfm?conferenceid=2456
1. Analytical Frameworks
System shock analysis and complex network effects
The 2013 Global Risk Management
Pre-Conference Seminar
Michelle Tuveson, Executive Director, Cambridge Centre for Risk Studies
Andrew Coburn, Director External Advisory Board, Centre for Risk Studies
Dr Kimmo Soramäki, Founder and CEO, Financial Network Analytics
2. Analytical Frameworks: System shock analysis and complex network effects
Session Outline
Michelle Tuveson
Executive Director, Centre for Risk Studies, University of Cambridge
– A Framework for Managing Emerging Risks in International Business Systems
– Problem statement: emerging risks as a corporate problem, the Cambridge
Framework as a structure for approaching the problem
Dr Andrew Coburn
Director of External Advisory Board, Centre for Risk Studies, University of Cambridge
– Developing Scenarios for Managing Emerging Risks
– Methodology: structural modeling of scenarios and their consequences;
examples of scenarios for extreme oil prices
Dr Kimmo Soramäki
Founder and CEO, Financial Network Analytics
– Understanding Shock Effects on Business Systems and Investment Portfolios
– Solutions: networks and interactivity, investment portfolios, illustration of
network modeling
3. A Framework for Managing Emerging Risks
in International Business Systems
The 2013 Global Risk Management Pre-Conference Seminar
Analytical Frameworks: System shock analysis and complex network effects
Michelle Tuveson
Executive Director
Centre for Risk Studies, University of Cambridge
4. Some Recent Events Disrupting International Business
4
Hurricane Sandy 2012
impacted a region that generates 40% of US economy.
Flights from many airports disrupted. Eastern sea port
closures disrupted international shipping for weeks
Arab Spring 2011-12
Impacts on many international businesses. Increased fuel
prices. 22% of businesses globally reported that the unrest
has a negative impact on their business
Credit Crunch 2008
US housing price crash in 2007 caused liquidity crisis
impacting all major economies and triggering lengthy
recession , impacting global businesses
Japan Tōhoku Tsunami 2011
Killed 26,000, destroyed factories and
infrastructure, triggered Fukushima nuclear meltdown.
Disrupted supply chains for electronics and other high-tech
components
Swine Flu Pandemic 2009
caused international panic with initial reports of a high
virulence virus, leading to travel and business disruption
for many weeks
Thailand Floods 2011
Manufacturing regions in Chao Phraya flood plains
inundated disrupting supply chains for international
businesses . Large contingent business interruption claims
5. And the list goes on…
Volcanic eruption of Eyjafjallajökull, Iceland, 2010, closed airports across Europe for two
weeks. Business sectors worst hit, included fresh produce
providers, pharmaceuticals, and electronics
In 2010 piracy activity around Horn of Africa reached an unprecedented level of 490 acts
of piracy, and an estimated $12bn in costs incurred, leading to re-routing, delays, and
cost escalation for shipping routes between Europe and Asia
Unprecedented multi-national General Strikes were coordinated across
Portugal, Spain, Italy and Greece in November 2012, leading to impacts on air
travel, telecoms, and many other business sectors
7/7 2005 terrorist attack on London caused the closure of the City’s financial
centre, airports and local travel systems, and impacted international business activity
North American Blizzard of 2010 affected most of US with record snow
levels, suspending travel services, international flights and shipping with waves of
snowfall through Feb and March
Deepwater Horizon oil spill in 2010 made large parts of the Gulf of Mexico
unnavigable, caused damage to local industries and disrupted international business
connected to the region
SARS outbreak in 2003 disrupted airline passenger traffic for five months, depressing
tourism, travel and other business
5
6. The Problem
Modern corporate businesses are finding that their processes are
more prone to disruption than they expected
– Each geo-political event causes surprise
This is a result of globalization – corporate systems now reach across
the world and are impacted by many more hazards and localized
changes than ever before
Global business systems have been optimized to minimize cost – this
reduces safety margins
There is a new operational focus on ‘resiliency’
To understand and measure resilience requires a new framework
– The Cambridge Risk Framework
Many corporates are espousing new approaches to managing
‘emerging risks’
– The Cambridge Risk Framework aims to provide tools for this management
6
7. Japan Tōhoku Catastrophe
Disruption to Business Systems
7
“Sony's production and sales were severely affected by the earthquake and tsunami in Japan in March
last year.
The twin disasters resulted in supply chain disruptions and a shortage in power supply in Japan, forcing
Sony to curtail production.
Its fortunes were hurt further by floods in Thailand later in the year, which saw its factories in the
country being affected.”
8. The Cost of Disruption
Examples of daily cost impact of a disruption in a company’s supply network
being $50-$100 million
– Rice and Caniato (2003)
Studies of ‘long-run’ equity values of companies following disruption to supply
chain show:
– Average abnormal stock returns of -40% for firms suffering disruptions
– Shareholders lose average of 10% of their stock value at announcement
– 14% increase in equity risk in the year following a disruption announcement
– Firms do not quickly recover from the negative effects of disruptions
– Source: Hendricks & Singhal, 2005 (sample of 827 disruption announcements made during 1989–2000)
2004 Survey of top executives at Global 1000 firms showed supply chain disruptions and
associated operational and financial risks to be single greatest concern
– (Green, 2004)
Current trends in best practice for managing the risk of international disruption:
– Cost management and efficiency improvements
– Supply base reduction
– Global sourcing
– Sourcing from supply clusters
– Source: Craighead et al., 2007, The Severity of Supply Chain Disruptions: Design Characteristics and Mitigation Capabilities
8
9. The Current Challenge of Managing ‘Emerging Risk’
Modern businesses face a large number of ‘Emerging Risks’
Many companies maintain an emerging risk committee or have
a formal monitoring system in place
– Much of this work is ad-hoc
‘Emerging Risks’ also include emerging recognition of long-
standing threats
Is there a systematic process to assess and evaluate the entire
range of threats?
How are these threats best managed?
Can we also assess the positive opportunities and upside
potential that might be presented by new threats?
What financial products or techniques could best answer the
corporate demand for de-risking global business?
9
10. Catastrophe Modeling Meets Complex Systems
The Centre for Risk Studies arises from shared interests by the
participants in exploring areas of intersection between
– Catastrophe modeling and extreme risk analytics
– Complex systems and networks failures
Advance the scientific understanding of how systems can be made
more resilient to the threat of catastrophic failures
10
Air Travel Network Global Economy
To answer questions such as:
‘What would be the impact of
a [War in Taiwan] on the [Air Travel Network] and how would this impact the [Global Economy]?
Regional Conflict
11. Business Activity as a System of Systems
11
Air Travel Network Cargo Shipping Networks
Communications Networks
12. Networks, Attacks, and Residual Modeling
A framework for assessing the consequences of an event on a system network
12
Network ‘Attack’ Residual
Describe the topology of the network
as nodes and links
Baseline efficiency of the network
quantified through standard metrics
of Value Function:
• Connectivity
• Reference path length
• Diameter
• Social Welfare
Degradation of the network through
localized impairment or removal of
nodes and links
Attack measured by ‘k-cut’ metrics
Post-attack network either static or
adaptive
• Network may be fragmented after an attack
Adaptive response of a network adjusts
traffic and relationships
May introduce congestion
Changes in Value Function are
measured as a result of the attack
13. Components of
Cambridge Risk Framework
13
Threat Observatory
Network Manager
Analytics Workbench
Strategy Forum
http://www.CambridgeRiskFramework.com
14. Cambridge Risk Framework
Threat Taxonomy
14
Famine
Water Supply Failure
Refugee
Crisis
Welfare System
Failure
Child
Poverty
HumanitarianCrisis
AidCat
Meteorite
Solar Storm
Satellite System
Failure
Ozone Layer
Collapse
Space
Threat
Externality
SpaceCat
Other
NextCat
Labour Dispute
Trade Sanctions
Tariff
War
NationalizationCartel
Pressure
TradeDispute
TradeCat
Conventional War
Asymmetric War
Nuclear
War
Civil
War
External
Force
GeopoliticalConflict
WarCat
Terrorism
Separatism
Civil
Disorder
AssassinationOrganized
Crime
PoliticalViolence
HateCat
Earthquake
Windstorm
TsunamiFloodVolcanic
Eruption
NaturalCatastrophe
NatCat
Drought
Freeze
HeatwaveElectric
Storm
Tornado &
Hail
ClimaticCatastrophe
WeatherCat
Sea Level Rise
Ocean System Change
Atmospheric System
Change
Pollution
Event
Wildfire
EnvironmentalCatastrophe
EcoCat
Nuclear Meltdown
Industrial Accident
Infrastructure
Failure
Technological
Accident
Cyber
Catastrophe
TechnologicalCatastrophe
TechCat
Human Epidemic
Animal Epidemic
Plant
Epidemic
ZoonosisWaterborne
Epidemic
DiseaseOutbreak
HealthCat
Asset Bubble
Financial Irregularity
Bank
Run
Sovereign
Default
Market
Crash
FinancialShock
FinCat
15. Profile of each Macro-Threat Class
We are preparing a monograph on each of
the key threat categories:
State-of-knowledge summary of the science
Identify the leading authorities and publications
on the subject
Catalogue of historical events
Map the geography of threat
Define an index of severity (‘magnitude scale’)
Assess a first-order magnitude-recurrence
frequency (worldwide)
Provide illustrative ‘Stress Test’ scenarios of
large magnitude events
– For e.g. 1-in-100 (or 1-in-1,000) annual probability
System impact (vulnerability) knowledge
Assessment of uncertainties
15
16. Adopting Cambridge Threat Taxonomy
as an Industry Standard
In September 2013, Munich Re will be co-hosting a workshop
to review the CRS Threat Taxonomy v2.0 for use in emerging
risk management processes
Attendees include major corporations, model developers and
insurance companies
Objective is to produce a version 3.0 for use by Munich Re and
others for use as an industry standard
Others are welcome to participate
– Invitation to attend the workshop
– Or review the proposed standard during consultation stage
– Participants should be interested in adopting the standard for their
own use in risk management
16
17. Conclusions
Many international corporates now recognize the
importance of managing emerging risks in their global
business
Managing emerging risks needs a framework for
– Understanding the interlinkages in global business systems
– Assessing all the different types of threats that might impact
those business systems
The framework can be used to develop shock test
scenarios for use in risk management
17
18. Developing Scenarios
for Managing Emerging Risks
The 2013 Global Risk Management Pre-Conference Seminar
Analytical Frameworks: System shock analysis and complex network effects
Dr Andrew Coburn
Director of External Advisory Board
Centre for Risk Studies, University of Cambridge
19. Using Scenarios for Risk Management
Many companies use ‘what-if’ scenarios for
understanding and managing risk
Management science is well developed
– Use of scenarios in business strategy since 1960s
Scenario planning proved to create business value
– Companies like Shell place great value in their scenario
unit, and attribute it with anticipation of the 1970s oil
crisis, and rapid response to 2008 financial crisis
Scenarios
– Create management flexibility
– Improve resilience to a crisis
– Challenge management assumptions about status quo
19
20. Seven Key Lessons for Developing Scenarios
1. Make it plausible, not probable
2. Ensure that the scenarios are disruptive and challenging
3. Offer two scenarios for a situation, not one or three
4. Make the suite of scenarios equally likely
5. Quantify the consequences of the scenario
6. Ensure scenarios are ‘coherent’
7. Make the scenarios relevant to the management team
20
21. Example Scenarios Currently in Development
21
Cyber Catastrophe Risk
Major compromise of commercial and national infrastructure IT systems by
malicious worm attack
Geopolitical Conflict Risk
Regional conflict in South China Sea embroiling Western military powers and SE
Asian nations
Human Pandemic Risk
Virulent influenza pandemic causes 6 months of workforce absenteeism and
social and economic disruption
Civil Disorder Risk
Austerity-driven riots and strikes across multiple cities in several Eurozone
countries
22. Oil Supply Shock Analysis
22
Hypothetical Scenario of
a Geopolitical Crisis in Middle East
23. Disclaimer
This is a hypothetical scenario developed as a
stress test for risk management purposes
It does not constitute a prediction
The Centre for Risk Studies develops hypothetical
scenarios for use in improving business resilience
to shocks
These are contingency scenarios used for ‘what-if’
studies and do not constitute forecasts of what is
likely to happen
5/9/2013
24. System Shock Project
How might…
24
A geo-political event …impact the global price of
crude oil…
…and how would that affect
a typical investment portfolio..?
$
25. Oil Price Shock Scenarios
25
Forcing Oil Price to an Unprecedented Low
Shale oil bonanza from large reserves in China turns China into a
net producer, causing rapid oil price collapse on global markets
Forcing Oil Price to an Unprecedented High
‘Arab Spring’ regime change in Saudi Arabia deregulates OPEC-
Swing oil production and triggers extreme oil price escalation
26. Project Team
26
Andrew Coburn
Michelle Tuveson
Danny Ralph
Simon Ruffle
Gary Bowman
Louise Pryor
Kimmo Soramäki
Samantha Cook
Christian Brownlees
With assistance from:
Peace and Collaborative Development Network
Ivan Ureta
Associate Prof in International Relations
Investment Fund
Will Beverley
Head of Macro Research
30. Oil Prices Driven by Global Growth
Prices of commodities tend to be:
• Log-normal-ish, but
• fat-tailed
• mean reverting
• with sudden jumps
Prices of commodities tend to be:
• well-correlated to global economy
• cyclical
• seasonal
31. Spot Price
($/B)
Initial Spot
Price ($/B)
Price Adjustment
Must be between 0 and 2
Price
Adjustment
Delay
Delta: PA Now -
PA Delay
Futures Oil
Price ($/B)
Initial Futures
Oil Price ($/B)
Difference
Futures/Spot
Futures/Spot
Price
Adjustment
Future delay
($/B)
Futures/Futures
Delay Price
Adjustment
Market
Sentiment market adj
Inital Market
Sentiment
market adj
output
<Prod - Cons 1
month delay (B/M)>
Ideal Production -
Consumption (B/M)
Ideal D/S - Actual
D/S (B/m)
Demand/Supply
Price Adjustment
Commercial
Inventory Adj
<Commercial
Inventory Flows
(B/M)>
Exogenous
event
Spot Price 1
Month Delay
($/B)
<Strategic Inventory
Flows (B/M)>
Strategic
Inventory Adj
<Prod - Cons 1
month delay (B/M)>
ST geopolitics
<Exogenous
event>
<OPEC Supply constraints:
Politics/embargos/wars
(B/M)>
Conversion Delay 1Exo Eve
Geopoltics
ST geo
Modeling of Crude Oil Spot Price
32. Scenario Initiation
Two months of initial unrest leads to
increasing levels of violence and anti-
government protest in Saudi Arabia
Initial dissatisfaction is driven by social
conditions but is rapidly taken up by
neo-Arab nationalism and minority Shia
Islamic fundamentalism
Suspicion of support to rebels being
provided by Shia groups in Middle
East, including Iran and Hezbollah
32
33. Seizure of Refineries and Oil Production
Mass-movement leads to loss of control
of major oil production facilities as
protestors occupy refineries
– Ras Taruna (0.5 m barrels/day)
– Yanbu (1m barrels/day)
– Multiple others
Many thousands of armed protestors
occupying sites, taking hundreds of
western workers as hostages
Military stand-off as Saudi and US forces
are unable to retake facilities without
jeopardizing civilian hostages
Sudden loss of production of over 1m
barrels a day (10% of Saudi output)
Political chaos as leadership falters
33
34. Initial State
Overthrow
Scenario Escalation Event Tree
34
Anti-western regime
established
US Military
Intervention
Iran Hezbollah
Response Regional
Escalation
None -
Forced Standoff
Swift restitution of
pro-Western regime Insurgency
Iranian state-backed
military invasion
Annexation of
regional caliphate
Lengthy military
campaign
China backing for
military action
Israeli counter-strikes
and broader ections
Western coalition
forces deployed
Russia annexes areas
of Islamic influence
Other coincidental or triggered consequences can increase the severity of a scenario
A
C
D
E
B
35. Conflict Escalation Across ‘the Oil Corridor’
Potential for scenario to
escalate into broader
regional conflict
‘Oil Corridor’ contains a third
of the world’s oil
Worst case sees prolonged
conflict across entire region
36. Arab Spring Timelines
Libya
First protests (15 Feb 2011)
UN Recognition (16 Sep 2011)
End of violence (23 Oct 2011)
251 days
36
Egypt
First protests (25 Jan 2011)
Mubarak resigns (11 Feb 2011)
Protests end (30 June 2012)
18 days (523 days of unrest)
Tunisia
First protests (18 Dec 2010)
Regime Change (14 Jan 2011)
Protests end (9 Mar 2011)
27 days (82 days of unrest)
Yemen
First protests (27 Jan 2011)
Ceasefires and Transitions
End of protests (27 Feb 2012)
397 days
Syria
First protests (15 Mar 2011)
736 days (ongoing)
37. Oil Production
OPEC produces 40% of the
world’s 80 mbbl/d oil and
holds three quarters of the
world’s 1.6 tr bbl reserves
Oil consumption is well-
correlated to global economy
– with cyclical and seasonal
patterns
Oil Corridor accounts for a
third of all oil production
OPEC follows Oil Corridor lead
37
Saudi
Arabia, 1
0
Rest of
OPEC, 2
3
Non-
OPEC, 4
5
0
10
20
30
40
50
60
70
80
90
Millionsofbarrelsofoilperday
World Oil Production
millions of barrels a day
Total 80 mbbl/d
Total World
Saudi Arabia
Other OPEC
Middle Eastern Oil Corridor
38. OPEC Swing
Saudi Arabia controls the ‘OPEC-
Swing’
OPEC Swing is a pricing
regulatory mechanism
– releases more reserves as price
rises
It damps sudden price rises and
constrains market volatility
In this scenario, the OPEC Swing
mechanism is effectively disabled
It enables prices to follow market
sentiment rather than economic
fundamentals
38
39. Market Reaction: The Black Bubble
Market reactions are severe
Negative sentiment feedback and
pessimistic commentary results in
a ‘black bubble’
Oil prices peak at $500 a barrel for
3 days
Release of government strategic
reserves and political commentary
reduces oil pricing to below $300
Sustained period of high oil prices
39
40. Modeled Impact on Oil Price
$0
$100
$200
$300
$400
$500
$600
1 11 21 31 41 51 61 71 81 91 101
OilPriceperbarrel
Crisis (Days)
Oil Price during Saudi Arabia Crisis Scenario
Attack on
Ras Tanura
Attack on
Yanbu
‘OPEC Swing’
failure
Note – this is a ‘what-if’ illustration of
potential extreme price patterns not a
prediction or estimation of an actual
outcome
Duration of military action
41. Scenario Durations and Impacts
41
0%
5%
10%
15%
20%
25%
30%
35%
0 20 40 60 80 100 120 140
A
B
C
D
E
Duration: Months before restoration of normal oil production
Impact:
% of
world’s oil
production
affected
Short
Revolution
Successful US
Intervention
US fights well-
resourced insurgency
Iranian invasion
Regional
Conflagration
Duration
Impact
42. Sectors Worst Affected
42
Code Sector Subcode Industry Groups Correlation with Oil Price Shock
10 Energy 1010 Energy High + 3
15 Materials 1510 Materials High - -3
2010 Capital Goods Medium - -2
2020 Commercial & Professional Services Low - -1
2030 Transportation High - -3
2510 Automobiles and Components Medium - -2
2520 Consumer Durables and Apparel Medium - -2
2530 Consumer Services Medium - -2
2540 Media Medium - -2
2550 Retailing Medium - -2
3010 Food & Staples Retailing High - -3
3020 Food, Beverage & Tobacco Medium - -2
3030 Household & Personal Products Medium - -2
3510 Health Care Equipment & Services Low - -1
3520 Pharmaceuticals, Biotechnology & Life Sciences Low - -1
4010 Banks Medium - -2
4020 Diversified Financials Medium - -2
4030 Insurance Medium - -2
4040 Real Estate Medium - -2
4510 Software & Services Low - -1
4520 Technology Hardware & Equipment Low - -1
4530 Semiconductors & Semiconductor Equipment Medium - -2
50 Telecommunication Services 5010 Telecommunication Services Low - -1
55 Utilities 5510 Utilities Medium + 2
35 Health Care
40 Financials
45 Information Technology
20 Industrials
25 Consumer Discretionary
30 Consumer Staples
Few sectors are not negatively impacted by a severe oil price
43. Understanding the Implications of a High Oil Price
Businesses can trace the implications of high oil prices on
all their business operation costs and opportunities
Sectoral impacts have marginal differences
Affects overall macro-economic environment
– Transportation of all goods to market cause spirals of cost
inflation
– Severe curtailment of demand through increased pricing
– Recessionary forces
– Alternative sources of energy become more attractive and
economically viable
A major impact is investment portfolio asset movements
43
44. What Other Scenarios Should a Business Consider?
As an alternative to contingency planning for a world of
extreme high energy prices, there are scenarios for
extreme low prices of energy
– The Shale Oil Bonanza
These may have opposite implications and contingency
requirement
There are also several scenarios for extreme impacts on
business systems and operational continuity that are
plausible
– Pandemics; cyber-catastrophes; severe weather; environmental
collapse;
Drives emphasis on flexibility of thinking, and resiliency
to cope with unexpected shocks
44
45. Conclusions
Scenarios are useful tools for business planning to
challenge assumptions about the status quo
Can be used as stress tests to a five-year plan and as
contingency plan requirements
Scenarios have proved their business value in helping
businesses have more agile reactions to unexpected
events
The Cambridge Centre for Risk Studies will be publishing
and releasing scenarios for use with models of networked
business systems to fully understand potential effects
45
46. Understanding Shock Effects on
Business Systems and Investment Portfolios
The 2013 Global Risk Management Pre-Conference Seminar
Analytical Frameworks: System shock analysis and complex network effects
Dr Kimmo Soramäki
Founder and CEO
Financial Network Analytics
47. Systemic Risk ≠ systematic risk
The risk that a complex system composed of many interacting
parts fails (due to a shock to some of its parts).
Domino effects, cascading failures, financial interlinkages, … ->
i.e. a process in the financial network
News articles mentioning “systemic risk”, Source: trends.google.com
47
Not:
48. Network Theory
Main premise of network theory:
Structure of links between nodes
matters
Large empirical networks are
generally very sparse
Network analysis is not an
alternative to other analysis
methods
Network aspect is an unexplored
dimension of ANY data
48
49. 49
For example:
Entities:
100 banks
Variables:
Balance sheet items
Time:
Quarterly data since 2011
Links:
Interbank exposures
Information on the links allows
us to develop better models for
banks' balance sheets in times of
stress
Networks brings us beyond the Data Cube
"The Tesseract"
50. Observing vs Inferring
Observing links
– Exposures, payment flow, trade, co-
ownership, joint board
membership, etc.
– Cause of link is known
Inferring links
– Observing the effects and inferring a
relationship e.g. via correlations
– Cause of link is unknown
– Time series on asset prices, trade
volumes, balance sheet items
50
51. Inferring Links from Asset Prices
Issues:
– Prices vs Returns (arithmetic vs log)
– Controlling for Common Factors (PCA)
– Correlation (Pearson, rank, ...) vs dependence (partial
correlations, tail, normal, regimes)
– Time period (short vs long)
– Significant and Multiple Comparisons -correction
-> Goal is to uncover 'links' or relationships that form a network
52. Benefit of Visualization
52
Mean of x 9
Variance of x 11
Mean of y ~7.50
Variance of y ~4.1
Correlation ~0.816
Linear regression:
y = 3.00 + 0.500x
Anscombes Quartet: Constructed in 1973 by Francis Anscombe to
demonstrate both the importance of graphing data before analyzing it and
the effect of outliers on statistical properties
53. Visualizing Correlations
Calculate pairwise correlations for 31
ETFs in various geographies and asset
classes
(465 correlations)
Color code correlations:
Problem:
We are making many estimates, some
of which are likely false positives
-1 +1
2007-2008
2012-2013
54. 54
Example - Distribution of correlation in 30 trials
with random numbers
20 pairs 50 pairs
100 pairs 200 pairs
55. Significant Correlations
Keep statistically significant correlations
with 95% confidence level
Carry out 'Multiple comparison' -
correction -> Expected error rate <5%
Problem:
Heatmaps can be misleading due to
human color perception
2012-2013
Last month
61. Minimum Spanning Tree
A Spanning Tree of a graph is a subgraph that:
1. is a tree and
2. connects all the nodes together
Minimum spanning tree (MST) is a spanning tree with shortest length. Length
of a tree is the sum of its links.
62. Re-positioning the Assets
We lay out the assets by their
hierarchical structure using Minimum
Spanning Tree of the asset network.
Shorter links indicate higher
correlations. Longer links indicate
lower correlations.
Negative correlations are shown as
red links and positive correlations as
black.
Absence of links marks that asset is
not significantly correlated with
anything
Interactive chart at:
http://www.fna.fi/demos/conference-board/charts/correlation-network.html
63. Data Reduction for Clarity
Node color indicates identified
community.
Missing links (clusters) denote
no significant correlation.
Interactive chart at:
http://www.fna.fi/demos/conference-board/charts/correlation-tree.html
64. Extensions
Principal Component Analysis and Correlation
regimes
GARCH -based forecasts
Alternative link definitions:
Granger causality, partial correlation, tail
dependence
Outlier detection and alert systems
Stress testing
65. Partial Correlation
Partial correlation measures the degree of association between two random variables, controlling
for other variables
We build regression models for daily returns of e.g. Oil and Gold based on all other assets of
interest and look at the correlation of their model residuals (i.e. what is left unexplained by the
other factors) -> Partial correlation
Model 1: Regress Gold on all other assets except Oil
Model 2: Regress Oil on all other assets except Gold
Gold residuals = vector of differences between observed Gold values and values predicted by
Model 1
Oil residuals = vector of differences between observed Oil values and values predicted by Model 2
Partial correlation between Oil and Gold is the correlation between Oil residuals and Gold residuals
65
66. Partial Correlation Network
Network of statistically significant
partial correlations of monthly returns
for a wide set ETFs during 2007-2013
Link width is value of partical
correlation (range up to 0.85)
We can use the partial correlations to
undestand linkages within a standard
portfolio stress test model
We organize the network on the basis
of distance from the shocked node:
67. The Network for an Oil Shock
Interactive chart at:
http://www.fna.fi/demos/conference-board/charts/oil-shock-01.html
68. Shocking Multiple Nodes
We use multivariate percentiles (based on the multivariate normal
distribution) to simultaneously shock Financials, German Stocks and Gold
First we estimate the mean and covariance matrix of these three asset
returns from theobserved data.
Then, for the first percentile, we find the shocks x, y, and z such that the
joint probability P(XLF < x AND EWG < y AND GLD < z) = 0.01 and the
marginal probabilities are equal, i.e., P(XLF < x) = P(EWG < y) = P(GLD < z)
A similar calculation finds the 99th percentile.
69. The Network for Multiple Shocks
Interactive chart at:
http://www.fna.fi/demos/conference-board/charts/triple-shock-01.html
70. Is it Correct?
We develop a model where we use the network structure to estimate many
small models (some of which are based on estimates)
We see how well cascading predictions works by predicting values for a out
of sample data set whose values are known.
We compare results to a normal linear model
Result: Predictions based on partial correlation network are as good for
single asset shock, and just slightly worse for multiple asset shock
-> The partial correlations do open up the model and provide more insights into asset
dynamics and interdependencies
Caveats: shocks outside 'normal' bounds may not exhibit same behavior. Shocks to
correlations, volatilities are not covered.
71. Summary
Correlation networks can provide visual insights into market
dynamics
Partial correlation networks can provide visual insights for
portfolios stress testing
72. Blog, Library and Demos at www.fna.fi
Dr. Kimmo Soramäki
kimmo@soramaki.net
Twitter: soramaki