Risk Architecture: Application of Complexity Science

Neil Cantle, Principal, MillimanNeil.Cantle@milliman.com

Risk Architecture, Washington, DC, April18, 2012 2

Table of Contents

Background

The nature of risk

The sciences of complexity

Risk Appetite

Emerging Risk

Risk Architecture, Washington, DC, April18, 2012 3

BACKGROUND

Unravelling the complexities of risk

Risk Architecture, Washington, DC, April18, 2012 4

Why is studying risk so difficult?

Joining inputs to outputs is hardEffects are highly non-linearEverything is highly interconnectedHeuristics seem to become outdated quickly

Typical tools not really ideal– Reductionist– Linear– Single characteristic– Statistical

Risk Architecture, Washington, DC, April18, 2012 5

Level of Understanding

Symptoms

Causes

Sense-making

Understanding

Risk Architecture, Washington, DC, April18, 2012 6

Systems

Systems theory helps us to make sense of complex problemsHelps to uncover “complex” patterns…not chaosMany alleged “black swans” are just complex risks we didn't understand early in their developmentGain insights into future developmentSupport for the experts…spot the next crisisSciences developed across many disciplinesSystems tools help us to:– Identify and understand emergent properties– Describe how the system works in terms of the key

interactions of its components

Risk Architecture, Washington, DC, April18, 2012 7

Introduction to Systems

A set of components interconnected for a purpose

Input

Output

Risk Architecture, Washington, DC, April18, 2012 8

Complex System – Feedback, subsystems, etc.

Input

Output

Input

Output

Introduction to Systems

Risk Architecture, Washington, DC, April18, 2012 9

Introduction to Systems

Complex Adaptive System – Structure changes

Input

Output

Input

Output

Risk Architecture, Washington, DC, April18, 2012 10

Complex Adaptive System Characteristics

Has a purposeEmergence – the whole has properties not held by sub components Self Organisation – structure and hierarchy but few leverage pointsInteracting feedback loops – causing highly non-linear behaviourCounter-intuitive and non-intended consequencesHas tipping point or critical complexity limit before collapseEvolves and history is importantCause and symptom separated in time and space

Risk Architecture, Washington, DC, April18, 2012 11

Complexity from simple rules

Source: http://en.wikipedia.org/wiki/Double_pendulum

Equations of motion (where L denotes the Lagrangian, KE-PE)

Centres of mass:

Risk Architecture, Washington, DC, April18, 2012 12

RISK APPETITE

Linking inputs and outputs

Risk Architecture, Washington, DC, April18, 2012 13

Risk Appetite

Uncertainty = absence of precise and complete knowledge leading to consensus of future stateRisk = state of uncertainty for a participant where some of the possibilities involve an undesirable outcome (e.g. loss)Risk Appetite = “our comfort and preference for accepting a series of interconnected uncertainties related to achieving our strategic goals”Risk Limits = operational restrictions intended to maintain performance within risk appetite

Risk Architecture, Washington, DC, April18, 2012 14

Risk Appetite

Three main components:

Also need to specify which sources of uncertainty are un/acceptable (risk preferences)Need to understand how various factors cause variation in the outcome

Planned range of outcomes

Minimum acceptable outcomeReturn

period

Risk Architecture, Washington, DC, April18, 2012 15

The Risk Appetite Problem

Need to constrain multiple inputs...

...producing multiple outputs to be kept within appetite

...which flow through multiple complex adaptive interactions...

It is essentially a large, complex multi-objective optimisation and control challenge

Risk Architecture, Washington, DC, April18, 2012 16

Risk Appetite

Looked for a solution which can...– Cope with non-linear dependencies– Adapt and learn– Be communicated effectively to a wide range of

stakeholders– Suit a wide range of firms

Risk Architecture, Washington, DC, April18, 2012 17

Risk Appetite Process

Agree business goals for which uncertainty mattersDescribe how much uncertainty you are comfortable withIdentify the possible sources of uncertaintyDescribe how that system worksEstablish limits which maintain performance within desired range of uncertaintyCycle of measuring risk capacity and resource utilisation against appetite

Risk Architecture, Washington, DC, April18, 2012 18

Joining Top And Bottom

Use a combination of cognitive and data-driven methodsLeverage expert knowledge (using cognitive mapping)Technique can be easily embedded within ORSA/planning processResulting model remains in the language/style of the contributorsExplicitly links to Internal Model for Solvency II

Risk Architecture, Washington, DC, April18, 2012 19

Understanding The SystemKey Nodes

Key Drivers

Gaps

Rapidly elicit highly detailed description of risk profile and implicit dynamics

Risk Architecture, Washington, DC, April18, 2012 20

Sources of risk

Model links risk characteristics and indicators

Implemented in AgenaRisk™

Risk Architecture, Washington, DC, April18, 2012 21

Setting Appetite

Use propagation properties of Bayesian Networks

Setting an outcome here...

...tells us what the states ought to be here

Risk Architecture, Washington, DC, April18, 2012 22

Propagating evidence

Setting desired appetite level translates into information about underlying limitsE.g. Counterparty credit...

Risk Architecture, Washington, DC, April18, 2012 23

Monitoring

Use propagation properties of Bayesian Networks

...gives us an estimate of risk level here

Entering observed values here...

Risk Architecture, Washington, DC, April18, 2012 24

Monitoring risk levels

Entering actual indicator values gives information about risk levels versus appetite

Risk Architecture, Washington, DC, April18, 2012 25

Risk Appetite

Proposed approach:– Embraces systems thinking approach– Is scalable from small/simple to large/complex– Can apply to any type of firm– Reacts naturally to emerging information– Provides a basis for setting AND monitoring limits– Can make use of expert knowledge until data available– Retains a form of use and interest to business people– Translates “risk” into business terms

Risk Architecture, Washington, DC, April18, 2012 26

Digression – Operational Risk

Hard to engage business in discussions about loss curves...

...Bayesian Networks provide a basis for retaining the business input

Risk Architecture, Washington, DC, April18, 2012 27

EMERGING RISK

Seeing the wood for the trees

Risk Architecture, Washington, DC, April18, 2012 28

Emerging Risk

Simple risks are easy(ish) to spot earlyComplex risks more difficult– Visible factors may not yet be obviously linked to risk

outcomes– Adaptation makes it hard to adjust monitoring to maintain

focus

Information

metrics

Cognitive techniques

Evolutionary techniques

Risk Architecture, Washington, DC, April18, 2012 29

Emerging Risk

Risk registers typically force the assignment of a label to each entryBut the entries are often not that simpleBy using a more granular labeling approach it is still possible to aggregate the informationTechnique from biology permits analysis of:– Which entries are “like” each other– Understanding of how risk scenario characteristics evolve– Clues about potential future scenarios

Risk Architecture, Washington, DC, April18, 2012 30

An overview of Evolutionary Risk Approach

Enterprise risk as an evolutionary processHow can we model the risk evolution process What insight can evolution of risks provide– A rigorous classification system with relationships– A guide to emerging, dynamic and systemic risks– A unique organizational risk lineage

Risk Architecture, Washington, DC, April18, 2012 31

Overview of Cladistics and Phylogeny

This methodology identifies small groups of highly related risks which share a common ancestorThe evolutionary history of each of these groups can then be accurately tracedThen their relation to other groups investigatedBy understanding the phylogeny of the risks we can:– Determine where evolution is most prolific – Detail path dependency and co-evolution of risk– Identify the most active characteristics to manage– Create focused scenarios for emerging risk modelling

Risk Architecture, Washington, DC, April18, 2012 32

Cladistic approach

ScenarioCharacteristics

1 2 3 4 5 6

A N N N N N NB Y Y N N N YC Y N Y Y Y YD Y N Y N Y N

Most parsimonious solution

Risk Architecture, Washington, DC, April18, 2012 33

Leverages Existing Information

The technique builds on (enhanced) risk register infoSimply extend range of characteristic classifiers

Risk Architecture, Washington, DC, April18, 2012 34

Evolutionary Risk Profile

We can identify risks which share similarities, common evolutionary paths and identify clues about future development

Risks can be studied for a part of the company, or the whole

Risk Architecture, Washington, DC, April18, 2012 35

Risk Profile

BU 1

BU 2

BU 3BU 4

BU 5

Risk Architecture, Washington, DC, April18, 2012 36

Interpreting Evolutionary Properties

Look at tree shape– areas of cascading bifurcation are likely areas for more

evolution and therefore emerging risks

Identify branches that have the most characters/adaptation– They are more likely to adapt again

Find characters that evolve most frequently– Is there a character or pattern that is responsible?

Are any risks/branches losing characters, ask why?– Risks should generally increase in complexity

Are there any characters gained in sequence/coevolution? – Understand this pattern as a possible clue to new risks

Risk Architecture, Washington, DC, April18, 2012 37

Summary

Systems approach helps to study the complexity before making simplificationsHelps to triangulate multiple insights (data, experts, etc.)Helps to incorporate adaptation and non-linearityThink of companies as a collection of people not machinesFocus on outcomes not just the “how”