DEVELOPING SCENARIOS FOR THE INSURANCE INDUSTRY

Cambridge Centre for Risk Studies

Scenario Best Practices

DEVELOPING

SCENARIOS FOR

THE INSURANCE

INDUSTRY

Acknowledgements

Cambridge Centre for Risk Studies gratefully acknowledges

Lighthill Risk Network for supporting the research eorts

summarised in this report. The Centre is grateful for the

expertise provided by our research team, collaborators, and

subject matter specialists. Any misinterpretation in use of the

advice provided is entirely the responsibility of the Cambridge

Centre for Risk Studies.

Report Citation

Cambridge Centre for Risk Studies, in collaboration with

Lighthill Risk Network, 2020. Developing Scenarios for the

Insurance Industry. Cambridge Centre for Risk Studies at the

University of Cambridge Judge Business School.

Strong, K., Carpenter, O., Ralph, D. 2020. Developing

Scenarios for the Insurance Industry. Cambridge Centre for

Risk Studies at the University of Cambridge Judge Business

School and Lighthill Risk Network.

Cambridge Centre for Risk Studies

University of Cambridge Judge Business School

Trumpington Street

Cambridge

CB2 1AG

United Kingdom

[email protected]

Website and Research Platform

www.jbs.cam.ac.uk/risk

Research Team

Dr Andrew Coburn, Chief Scientist

Professor Daniel Ralph, Academic Director

Simon Rue, Director of Research & Innovation

Dr Michelle Tuveson, Chairman & Executive Director

James Bourdeau, Geopolitical Risk Research

Oliver Carpenter, Lead Environmental Risk Research

Jennifer Copic, Lead Governance Risk Research

Dr Jennifer Daron, Lead Technology Risk Research

Ken Deng, Lead Financial Risk Research

Timothy Douglas, Risk Modelling

Tamara Evan, Lead Geopolitical Risk Research

Taryn Hubbard, Risk Research

Oliver Pearson, Risk Markets Research

Dr Andy Skelton, Lead Risk Modelling

Kayla Strong, Lead Scenario Analytics

Dr Timothy Summers, Senior Data Scientist

Jayne Tooke, Communications and Events

William Turner, Data Science

Lighthill Risk Network

LM01.01.02

The Leather Market

11-13 Weston Street

London, SE1 3ER

United Kingdom

[email protected]

Website

https://lighthillrisknetwork.org/

Members

Matthew Eagle, Managing Director,

Head of GC Analytics (Guy Carpenter)

Premal Gohil,

Head of Innovation Partnerships (Liberty Mutual)

Paul Kaye, Head of Actuarial Practice (Aon)

Shree Khare,

Group Head of Catastrophe Research (Hiscox)

Trevor Maynard,

Head of Innovation (Lloyd’s of London)

Andrew Mitchell,

Head of Catastrophe Modelling (MSAmlin)

Delimoma Oramas-Dorta,

Catastrophe Risk Analyst (Guy Carpenter)

Cameron Rye, Research Manager (MSAmlin)

David Singh, Head of Exposure

and Portfolio Management (MSAmlin)

Keith Smith,

Research and Development Manager (Lloyd’s of London)

Dickie Whitaker,

Chief Executive Ocer (Lighthill Risk Network)

The views contained in this report are entirely those of the research team of the Cambridge Centre for Risk Studies, and do not imply

any endorsement of these views by the organisations supporting the research, or our consultants and collaborators. The results of the

Cambridge Centre for Risk Studies research presented in this report are for information purposes only. This report is not intended

to provide a sucient basis on which to make an investment decision. The Centre is not liable for any loss or damage arising from its

use. Any commercial use will require a license agreement with the Cambridge Centre for Risk Studies.

Cambridge Centre for Risk Studies

Scenario Best Practices

DEVELOPING

SCENARIOS FOR

THE INSURANCE

INDUSTRY

Contents

Introduction 5

1.1 The Insurance Industry 5

1.2 Why Use Scenarios? 5

1.3 Report Rationale and Intended Audience 6

1.4 A Framework for Scenario Development 7

Understanding Scenarios 8

2.1 What is a Scenario? 8

2.2 Types of Scenarios 9

Using Scenarios in the Insurance Industry 15

3.1 General Scenario Uses 15

3.2 Insurance Use Cases of Scenarios 17

3.3 Allocating Risk Capital 20

Shock Risk Scenario Development: Insurance Stress Testing 22

4.1 A Checklist for Scenarios 22

4.2 Scenario Development: a Stress Test Perspective 23

4.3 Considerations and Constraints 30

Scenario Development Case Study: Business Blackout 33

Future Perspectives on Scenarios 37

Glossary 38

References 40

Cambridge Centre for Risk Studies

Scenario Best Practices: Developing Scenarios for the Insurance Industry

1. Introduction

1.1 The Insurance Industry

The insurance industry comprises companies that

provide risk management by underwriting the risks of

individual entities and pooling them to spread the risk.

An insurance contract guarantees that, on the occurrence

of a specied uncertain future event, the insurer provides

a payment to the insured, and thereby assumes the risk.

In return, the insured (or policyholder) pays a premium,

or a regular fee, to the insurer for providing the coverage.

Insurance is critical in providing nancial security to

people and organisations performing their daily activities

and operations; or undertaking new and risky ventures.

There are several classes of insurance available to

accommodate a diverse range of customers with an array

of risks, which are often categorised by their asset type or

the entity being insured. Examples include property and

casualty, accident and health, or auto insurance.

In addition to oering traditional coverages for known

risks, supported by comprehensive industry experience

and understanding of their occurrence, insurers must

also continually adapt to emerging and uncertain risks.

Today, such risks are manifesting at an unprecedented

rate as the world is challenged by growing complexity

and interconnectedness between systems. To full the

consequent demand for insurance coverage against

emerging risks, insurers must rst evaluate such

uncertainty and ensure they can withstand potential

losses and operate sustainably. The insurance industry

shares key nancial and operational risk exposures with

the wider nancial sector, but is also uniquely exposed

to insurance risk, so internal assessments of risk must

account for the complexity of interactions between

both an insurer’s assets (premiums invested to cover

liabilities) and liabilities. One means of planning for

the future is using scenario analysis (or ‘stress testing’),

a longstanding practice within the industry, but which

continues to grow in importance as the market becomes

more aware of its benets and regulatory requirements

call for robust internal risk management.

1.2 Why Use Scenarios?

Scenarios are stories about how the future might

develop, aimed to stimulate exploration, understanding,

and discussion. Based on a coherent set of assumptions

about key deterministic relationships and driving forces,

scenarios describe plausible futures that are intended

to be scrutinised and debated. In the context of risk,

scenarios provide a tool to cope with uncertainty,

especially in the case of risks that are not well understood

or cannot be quantied or even identied. They provide

a systematic method for exploring how a complex and

diverse array of risks may impact an organisation, sector,

or economy; or in other words, how resilient these

systems are to potential disruptions. Scenarios question

whether organisations or communities can adapt to,

and even capitalise on, future changes, and stress their

existing capabilities to respond. This understanding

can be applied to support and rationalise decision

making about the future, and facilitate reporting,

management, and mitigation of risks. Scenarios are

valued for supporting creative thinking about plausible

futures, rather than attempting to accurately predict

individual outcomes.

In the insurance industry, these tools continue to

evolve in response to advancing consideration and

regulation of enterprise risk management, both within

the industry and for the insureds that the industry

indemnies. Scenarios are increasingly being used by

underwriters, analysts, risk managers, actuaries, and

other stakeholders in the (re)insurance community to

better understand the characteristics and consequences

of unknown, uncertain, or unexpected future events.

A critical distinction is made between scenarios that

examine emerging trends, which are of concern for

long-term strategic planning; and those that consider

catastrophic shocks, which trigger severe loss across a

potential range of insurance classes, and so represent

acute operational risks. This report is primarily written

to address the latter, and on the design and uses of shock

risk scenarios, which are increasingly in demand to

overcome the challenges posed in today’s and the future

risk landscape.

1. For a complete listing of the types of insurance, please refer to Multi line Data Denition

initiative (Cambridge Centre for Risk Studies, in collaboration with Risk Management

Solutions, Inc. 2018

Cambridge Centre for Risk Studies

1.3 Report Rationale and Intended Audience

This report intends to outline best practices for scenario

analysis within the insurance community, and to provide

a practical framework to assist practitioners engaging

with shock scenario development.

Within the report, we explore key features of, and

commonalities and dierences between, insurance

specic scenarios, and suggest how and for whom they

can be used eectively. Due to the varied applications

of scenarios and their associated development

methodologies, we focus the methodology discussion

on shock risk scenarios, and draw commonalities

from various shock scenario development techniques.

As a variety of scenario use cases have unique

requirements and maturity in the practice, development

methodologies do vary. However, we propose that

there is a general process which can be adapted and

modied for these various uses. By providing key

criteria and considerations for scenarios in the form of a

scenario development framework, we hope to equip the

reader with the necessary tools and context to develop

coherent, comprehensive, and intelligible scenarios,

which therefore eectively full their intended purpose.

This report provides insurance-specic

recommendations for scenario development, and

has been published in parallel with another report

which similarly outlines best practices for scenario

development in the disaster risk reduction community.

The insurance industry has a rich expertise in risk

assessment tools to price risk, employing scenarios

that tend to be expert-driven, scientically supported,

and product oriented. In contrast, within the diverse

community of disaster risk managers, scenarios

often serve to explore and incorporate the culture

and experience of various stakeholders in a more

participatory approach, for which emphasis may

be placed as much on what is learnt in the scenario

process as the end result. Developing the two reports in

parallel has provided valuable insight into the merits of

contrasting scenario approaches and has informed the

best practices advocated in both.

Report Aims

This report aims to address scenario best practices

through the following considerations:

The role scenarios play within the

insurance community;

The common and contrasting characteristics and

typologies of scenarios;

How scenarios are currently employed and might be

further or better used in methodologies to address

emerging risks;

Current scenario limitations and critiques specic to

the insurance community;

Stress test scenario guidance for catastrophe

risk analysis;

How scenario development and applications might

adapt in the future.

2. Report titled “Scenario Best Practices: Developing Scenarios for Disaster Risk Reduction”

(Cambridge Centre for Risk Studies, in collaboration with Lighthill Risk Network 2020)

Scenario Best Practices: Developing Scenarios for the Insurance Industry

1.4 A Framework for Scenario Development

Within this report, we propose a framework for scenario

development that denes eight core steps, outlined in

Figure 1 and discussed in further detail in Section 4.

This framework is intended as a point of reference to

assist and ensure ecacy in the scenario process, rather

than as a prescriptive method that must be followed

absolutely. Further, while it outlines a linear step-by-step

structure for clarity, we encourage the scenario process

to be an iterative one, in which stakeholder engagement

provides opportunities for review and revision to ensure

it succeeds in fullling its aims.

Step 1 Scope the Risk

Identify the risk to be addressed, or, if the risk is uncertain

or unknown, dene the issues or vulnerabilities that the

scenario exercise aims to expose. In the latter case, the risk

may be identied later in the process. This contextualises

the objectives and resultant decisions of the analysis.

Step 2 Conduct Background Research

Research the topic dened in Step 1 by consulting

relevant sources of (scientic) knowledge and all

associated stakeholders within and beyond the insurance

industry. If possible, consider each dimension of risk:

hazard, exposure, and vulnerability, to recognise how and

where impacts occur.

Step 3 Frame the Scenario(s)

Consider and dene the key aims, benets, and

characteristics of a scenario and its process.

Figure 2 outlines some of the key questions the

developer should ask when framing their scenario.

Sections 2 and 3 provide the context to inform

these considerations.

Step 4 Develop Candidate Scenarios

Compose a series of candidate scenarios that capture

a range of plausible futures. Summarise scenarios with

brief outlines and key variables, and explore contrasting

characteristics. Select scenarios to progress that will

challenge and achieve the desired objectives.

Step 5 Develop a Narrative

Expand the selected scenarios with descriptions

that are interesting, challenging, and plausible for

all stakeholders. Account for all dimensions of a

future event, including context, triggers, timelines,

geography, responses, and implications.

Step 6 Assess Impacts and Materiality

Assess the impacts within the insurance industry and

in wider macro systems. Consider complexities and

interconnectivities that may cause cascading impacts

beyond the expected. Dene what constitutes a material

impact in order to focus the analysis on materially aected

assets and areas of business.

Step 7 Communicate and Act

Communicate the key ndings to stakeholders via

meaningful qualitative and quantitative outputs. The

content and complexity should be tailored to the

audience. Include a clear indication of the extent to

which the results can be relied on to inform decisions and

actions to address the risk.

Step 8 Evaluate and Update

Evaluate whether the objectives of the exercise have been

achieved and iterate the process with stakeholder input to

ensure or enhance ecacy. Be aware that the possibility and

character of a scenario will change as controlling factors

evolve, as will its impact as the industry advances, and so it

should be updated to maintain relevance and utility.

Figure 1: Scenario development framework for the insurance industry

Cambridge Centre for Risk Studies

2. Understanding Scenarios

2.1 What is a Scenario?

Scenarios are descriptions of potentially plausible events

that may occur in the future, leading to a particular

set of outcomes. They are based on assumptions about

key driving forces, interconnections, and relationships,

and have the ability to capture the uncertainties and

complexities of a system in a coherent manner. Scenarios

are not intended to comprehensively describe the

future, but rather to highlight focal elements of dierent

plausible futures and to highlight the key factors that

will drive future developments. Sometimes the terms

scenario, projection, and prediction (as well as others

such as forecast and outlook) are used interchangeably,

but while all are tools to investigate the future, each is

nuanced in its meaning. A prediction can be dened as

a subjective (probabilistic) statement that something

will happen in the future, while a forecast is the most

likely expected development.

In contrast, a projection is

a (probabilistic) statement that something will happen

under certain conditions, allowing for signicant changes

in the boundary conditions that might inuence a

prediction. A scenario-based projection is a hypothetical

construct of what could possibly happen conditional

upon fundamental assumptions.

These assumptions

allow some of the uncertainties that complicate more

exact statements on the future to be set aside for the

benet of a scenario exercise. The dimensions of what

constitutes a plausible event changes as external forces

shift. As a result, the scenario process is inherently an

evolving one, and scenarios which have been developed

and are relied upon should be maintained and updated

regularly to reect current conditions.

Sometimes scenario development and scenario analysis

(also called scenario ‘thinking’ or ‘planning’) are

dierentiated. Development means speculating about

the uncertainty surrounding the future and envisaging

dierent plausible future outcomes, or, in other words,

to create ’memories of the future’.

Scenario development

is the necessary foundation for scenario analysis, and

the two are closely linked. Scenario analysis can be

understood as the integration of scenarios into decision

making. Here, we explore both scenario development

and analysis together as the scenario process and use the

terms collectively.

The Probable, Possible and Plausible

When considering the future, we often add

‘probability’, ‘possibility’, or ‘plausibility’

qualications to emphasise relevance or importance.

These notions are implicitly dened, but often

not clearly dierentiated and so are confused. Is a

plausible future also probable? Can one future be

more plausible than another? Should any conceivably

possible future be considered? Care should be taken

in using these terms in the description of scenarios.

Key elements of the three qualiers are summarised

here to establish a distinction between them, but this

concern cannot be resolved by reducing each to a

denition. Scenario users should note that these terms

do not have any universal value, and so should ensure

the distinction between them is made suciently clear

to be useful.

Probability refers to the concept of chance and

likelihood, leading to an ordinal ranking of more or less

likely futures. Any future is possible, but the selection

of a probable or improbable scenario depends on

the application.

Possibility is a claim of reality; whether a future

is potentially realisable or not. It is a binary

distinction but may be challenged by absolute

(violation of established laws) or contingent (lack of

realism) reasons.

Plausibility addresses the structure of an argument

and places value on the credibility of a future, which can

hold true even though the future itself may be factually

fallacious. This is therefore a cognitive notion. Scenarios

are challenged by the dierence in interpretation of

plausibility between developers and stakeholders.

3. (MacCracken 2001) 4. (MacCracken 2001; Van Vuuren et al. 2012)

5. (Mietzner and Reger 2005; Schwartz 1997) 6. (Van der Helm 2006)

Scenario Best Practices: Developing Scenarios for the Insurance Industry

2.2 Types of Scenarios

Scenario design and development processes can be

commonly distinguished and classied, based on

the development process, their purpose, or certain

characteristics.

In practice, these typologies are rarely

binary or independent, and instead can be imagined

as a multi-dimensional matrix with unique outcomes.

This section proposes a series of distinctions which

are commonly used to dene scenarios. We encourage

practitioners to consider them as they construct

scenarios in the context of their scenario aims, within

the process of ‘Framing the Scenario(s)’ (Step 3 in the

Scenario Development Framework). These typologies

are illustrated in Figure 2, and discussed in detail within

this section.

Aid Communication Demonstrate Due Diligence Identify Bias Sensitivity Analysis Addressing Uncertainty

To contextualise complex

risks and promote

stakeholder understanding

To assess risk exposures

and understand their

nancial implications

To explore partialities

that stakeholders may hold

towards certain decisions

To investigate the power

and variance of controlling

variables on a risk

To expand understanding

of the range of

plausible outcomes

Trend Risk Scenario Shock Risk Scenario

Slow-onset, trend phenomena

that emerge gradually over time

Sudden-onset, shock events

that occur quickly or unexpectedly

Exploratory - To ask ‘what if?’ Normative - To ask ‘what for?’

To stimulate imaginative thinking about the future

and widen understanding of available options

To better understand the path to desirable futures

and evaluate the impact of decisions

Figure 2: Framing a scenario – scenario typologies and applications

What is the scenario for?

How can the scenario benet stakeholders?

On what timescale does the risk materialise?

Which is the more important scenario outcome?

Who owns and contributes to the scenario process?

Understanding Tail Risks Understanding Emerging Risks Strategic Planning Accumulation Management

To identify and understand extreme,

low probability risks

To imagine and comprehend

new or evolving risks

To dene a resilient strategy

for the future that alleviates risks

To explore possible extreme or maximal

correlated losses to insurance portfolios

Participatory - Bottom-up, co-production of knowledge Expert-Driven - Top-down, analytical

To incorporate stakeholder culture, knowledge,

and experience in the process and end product

To deliver rigorous scientic descriptions

of plausible futures to decision makers

Probabilistic Deterministic

To estimate the likelihood of occurrence

based on the variance of quantied causal parameters

To speculatively explore phenomena

that involve a high degree of uncertainty

Is the scenario required to dene the likelihood of an outcome?

7. E.g. (Mietzner and Reger 2005; Henrichs et al. 2010; Van Vuuren et al. 2012)

Cambridge Centre for Risk Studies

Trend Risk Versus Shock Risk Scenarios

Scenarios can be developed to consider either sudden-

onset, shock events that emerge quickly or unexpectedly,

or slow-onset, trend phenomena that emerge gradually

over time. The type of risk dictates how the scenario is

developed and how it may be used.

Trend risks require users to consider temporality and

identify both short-term signals and long-term impacts.

The latter are likely to be less noticeable than short-

term events and are dicult to align with organisational

decision making, which tends to focus on much shorter

future outlooks; as a result, it can be more dicult to

incentivise mitigation measures for long-term threats.

If eective action is taken, there is the benet of having

enough time to determine the best method to mitigate or

alleviate the risk. Trend risks are typically not insurable

yet are a topic of increasing interest within the insurance

sector, for example, as seen in the corporate uptake of an

Environmental, Social, and Governance (ESG) agenda,

and in response to growing awareness of the increasingly

pronounced impacts of climate change. Insurers are

increasingly needing to make decisions on how to

address trend risks, whether it be through an adaptive

process to keep losses within an insurable window, or

the discontinuation of certain insurance policies. To best

support these decisions, trend risk scenarios remain an

important component of scenario analysis, regardless of

whether the risk is currently insured.

In contrast to trend risks, shock risks are sudden events

which may not have been anticipated and trigger impacts

that materialise rapidly. The focus of these scenarios is

The Origin of Trend Scenarios

Trend risk scenarios have historically been used

within operational or strategic planning and have

roots in policy and operational settings. The Cold

War futurist, Herman Kahn, and others at RAND

Corporation, are regarded as among the founders of

the construction and use of scenarios in the 1950s

and 1960s, primarily in the security arena (e.g., how

a country would function after a nuclear conict),

and encouraged the philosophy of how to ‘think the

unthinkable’.

He described a scenario as:

“a set of hypothetical events set in the future

constructed to clarify a possible chain of causal

events as well as their decision points”.

Other notable research was being

undertaken at the time by the

Stanford Research Institute,

who oered long-range planning

facilities to support their military

and business consulting. They were

among the rst to formalise and

utilise scenario planning techniques

for decision making by businesses

and governments.

Kahn’s scenario methodology

used mathematical models and

forecasting techniques and set a

precedent for the future of scenario planning.

the 1970s, this methodology was adapted by Pierre

Wack’s newly formed Planning Scenario team at

the Royal Dutch Shell Group for use in a corporate

setting, primarily to anticipate developments in the

volatile oil and gas markets. ‘Shell’ scenarios were

notable for their trend risk analysis and strategic

planning purposes, and followed seven broad steps

Decide drivers for change/assumptions

Bring drivers together into a viable framework

Produce seven to nine initial mini scenarios

Narrow down two to three scenarios

Draft the scenarios

Identify the issues arising

The value of Shell’s futuristic

approach was rst realised in the

1973 Arab–Israeli (Yom Kippur)

War, when the rst oil embargo

caught most companies by surprise.

Shell had considered and strategized

for the implications of an oil price

shock, and thus overcame the worst

shocks, and emerged from the crisis

as the sector leader.

Image 1: Herman Kahn of RAND Corporation

(National Archives, 1963)

8. (Kahn 1961) 9. (Amer, Daim, and Jetter 2013)

10. (Durance and Godet 2010) 11. (Jeerson 2012)

Scenario Best Practices: Developing Scenarios for the Insurance Industry

to identify events which might “shock” the system and

have a valuable role in tactical planning. The impact of

such scenarios is likely to be acute, rather than accruing

slowly as in the case of trend risks, and they are more

commonly hedged by conventional insurance products.

The impacts associated with a shock event will demand

an immediate response from the insurance community.

The focus of this report is on shock risk scenarios, which

have a wider variety of typologies and methodologies

specic to the insurance industry. The dialogue within

the report is written from a stress test perspective yet is

often applicable in trend risk discourses.

Although stress test scenarios in the form of shocks have

been used by the insurance industry for decades, their

use via modelled catastrophe theory was precipitated

in the insurance industry following Hurricane Andrew

in 1992. At the time, Hurricane Andrew was the most

expensive insured event recorded, and triggered over $16

billion in losses.

The nancial impacts caused eleven

insurers to go out of business in the state of Florida and

caused heavy strain on the global insurance community.

In the following years, the insurance focused credit

agency, A.M. Best, required American insurers to report

their modelled losses. Lloyd’s of London similarly

introduced their realistic disaster scenarios (RDS)

in 1995 (see Section 5 for an example RDS, Business

Blackout, used as a model case study for scenario

development). The intention of these scenarios was to

provide a communal exercise for insurance organisations

to stress their portfolios and identify any potential

weaknesses which do not meet regulatory thresholds.

The application of stress tests quickly evolved beyond

regulatory purposes and has become a valuable tool in

risk exploration and response planning.

Exploratory Versus Normative

An important distinction concerns the purpose of

scenario development. Scenarios range on a continuum

between exploratory, with the purpose of educating and

expanding awareness of plausible futures, to normative,

with a primary aim to facilitate decision making. Often,

the goal is to concurrently balance exploratory and

decision-based functions.

Exploratory eorts ask, ‘what if?’, as a helpful way to

create a ‘future memory’.

This approach explores a wide

and contrasting range of potentially plausible futures

as a function of diverging assumptions (in other words

questioning ‘what would happen if this happens?’), with

the aim of widening the scope of options considered by

users and stimulating imagination and creative thinking

about the future. The focus here is on learning about the

process under analysis, raising awareness, developing

a descriptive assessment of plausible futures, and

taking a specied issue or environment as the subject of

analysis.

Exploratory scenarios often apply a forecasting

approach, dening scenarios on the basis of a set of

imposed rules dened from the base year onwards.

In contrast, normative scenarios primarily ask, ‘what

for?’. To utilise such a scenario for decision making,

a more narrowly dened set of criteria and objectives

must be explicitly dened.

A normative approach

typically uses scenarios that are formulated in technical,

quantitative terms (and thus have less emphasis on

narrative) so that the paths to desirable, or undesirable,

futures can be analysed. The intention is to evaluate

the impact of a set of variants concerning specic

interventions (behaviours and decisions) relative to a

baseline, based on some form of valuation.

Such eorts

tend to focus on delivering a product, in the form of a

specic alternative to address a problem, or an advising

tool for evaluating alternatives.

A normative approach

can be more easily combined with a backcasting

approach (as opposed to forecasting), dening scenario

pathways only after rst describing the end-points and

reasons back from these end-points, and exploring short-

term decisions to make these changes happen.

12. (Grant 2015) 13. (Grant 2015) 14. (Tourki, Keisler, and Linkov 2013)

15. (Riddell et al. 2018; Tourki, Keisler, and Linkov 2013) 16. (Van Vuuren et al. 2012)

17. (Birkmann et al. 2015) 18. (Henrichs et al. 2010; Tourki, Keisler, and Linkov 2013)

19. (Tourki, Keisler, and Linkov 2013) 20. (Robinson 1990; Dreborg 1996)

Cambridge Centre for Risk Studies

Participatory Versus Expert-Driven

Another divergence in scenario design concerns who

‘owns’ the process, with a key distinction made between

top-down, expert-driven (or ‘analytical’) approaches

and bottom-up, participatory approaches. Both have

advantages and disadvantages, and each may serve

dierent purposes, although they are not necessarily

discrete, with eective scenarios often including

elements of both. We emphasise the importance

of participation in the scenario process, since this

dimension is often undervalued or excluded in insurance

applications of scenarios.

Expert-driven approaches have the objective of providing

rigorous descriptions of plausible futures, including

details that are well supported by available science.

Such approaches are oriented towards decision makers,

and as a result tend to neglect other stakeholders. They

are analytical in approach and allow for exploration

of large-scale phenomena, for example global climate

change, which may pose logistical diculties to

participatory approaches.

In a participatory approach, scenario developers

work together with stakeholders, namely the people

potentially aected by scenario outcomes, rather than

delivering scenarios as a top-down means of education.

Participation targets and integrates stakeholder needs

and values, and while scientic and technical knowledge

remains important, such approaches makes use of

cultural perspectives, knowledge, and experience beyond

the involved experts to dissect complex issues. No group

knows everything, and each will learn from others

through the scenarios, providing everyone is represented

in the discussion. Eective communication of scenario

information is much easier than accurate communication

of technical information (such as probabilistic risk). Such

diverse engagement is eective in developing community

understanding and investment in decision making,

builds trust, and encourages broader acceptance of the

ultimate scenario outcomes.

Participatory approaches

enable scenario developers to understand, examine, and

discuss the links between phenomena at dierent scales

– for example how global or sub-national trends relate to

the vulnerability in specic regions or municipalities.

Scientic and technological developments have driven an

increasing role of technology and expertise in scenario

approaches but demands for improved participation and

accountability; and criticism of technical expertise, have

also grown. Within the insurance industry scenarios

are generally expert-driven, although there has been a

recent eort to better involve stakeholders. Examples

include community partnerships and local risk planning

initiatives, or the rise of social impact bonds.

Further distinctions can be made, aligning with

exploratory or normative, and participatory or expert-

driven approaches, as follows:

Intuitive versus formal processes

Process design refers to how scenarios are developed,

or their methodological aspects, ranging from

intuitive to formal approaches. Intuitive processes

focus on qualitative knowledge and are participatory,

incorporating many perspectives from a wide range of

backgrounds and knowledge bases.

In contrast, formal

processes regard scenario development as an analytical

and systematic exercise, and so depend on quantitative

inputs to build conceptual or computational models.

They are exclusive in the way they only incorporate views

from specic stakeholders or areas of expertise.

Process versus product orientation

The scenario development process can be at least as

important for the user as the product, for example if

a scenario is intended to support a specic decision.

A process-oriented scenario includes the user in

development, so that they may learn from the experience

and feedback to enhance scenario ecacy. However,

in many contexts, scenarios are instead communicated

in a linear process, with an end product deliverable

such as a report and or a quantication of loss. In this

case the product is typically more important than the

process, with the potential advantage of reaching a target

audience beyond those participating in the process. If

this is the case, quality, transparency, and legitimacy

need to be emphasised in order to ensure scenarios are

relevant to the user community and can be readily used

for planning and decision making.

Qualitative versus quantitative

The distinction between qualitative and quantitative

information is clear in scenario development, but the

contrasting methods may be, and perhaps should

be, combined. Qualitative information, specically

narratives, provide logic to scenario assumptions and

help to dene plausible future developments in situations

where formal modelling is not possible. They provide

an eective way to derive information at dierent scales

or for dierent topics (for example regional scenarios

nested within global narratives). Quantication, via

modelling, adds scientic rigour to scenarios, expanding

on numerical estimates of futures developments (often

based on simulation tools) where relevant and reliable

information is available. Quantitative outputs can

strengthen communication through clear denitions

and rules.

21. (Star et al. 2016) 22. (Star et al. 2016) 23. (Birkmann et al. 2015)

24. For more information on social impact bonds see (Social Finance Inc. 2012)

25. (Van Notten et al. 2003; Tourki, Keisler, and Linkov 2013)

26. (Tourki, Keisler, and Linkov 2013) 27. (Van Vuuren et al. 2012) 28. (Henrichs et al. 2010)

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Deterministic Versus Probabilistic

Scenarios are also characterised as either deterministic

or probabilistic. A deterministic scenario is created by

selecting a specic set of parameters and conditions,

while a probabilistic scenario considers a multiplicity of

outcomes, each with its own probability of occurrence,

depending on the probability distribution of the input

parameters and conditions.

A deterministic scenario

treats the probability of occurrence as nite, whereas

probabilistic modelling is intended to address the

uncertainty with a ‘complete’ probability distribution of

synthesised events.

Deterministic scenarios are recognisable by their focus

on the causal chain of circumstances that will give rise

to unusual or extreme outcomes. They are an eective

means of exploring phenomena speculatively or

hypothetically when they are not very well understood

or there is a high degree of uncertainty. They can also be

very valuable for exploring emerging risks, specically

where market or policy responses are uncertain.

In comparison, production of a probabilistic scenario is

possible when the underlying process is well understood,

and the causal parameters can be characterised with

estimates of their occurrence rates and distribution.

Each step in the causal chain has a dened distribution

of outcome likelihoods, and the model stochastically

samples from this distribution in many simulations. The

probability of an outcome and its uncertainty structure

is very sensitive to the assumptions made for the input

parameters.

By incorporating random variations into

the model, stochastic outcomes show a range of scenario

outcomes, and the likelihood of these permutations.

Techniques for probabilistic modelling are well

understood and documented, and are used in analysis,

such as natural catastrophe modelling, where the

subject phenomena have been comprehensively studied

and for which it is possible to estimate the uncertainty

distributions of the underlying variables.

There is overlap in deterministic and probabilistic

scenarios. For example, probabilistic modelling

can be used to generate a deterministic scenario,

typically such as the worst, best, or most likely case

events. Caution should be taken when comparing

the two types, as probabilistic scenarios still contain

deterministic attributes. Specically, probabilistic

scenarios require all potential outcomes to be dened,

yet in practice the universe is not a closed system.

There are outcomes of future probabilistic scenarios

which cannot be recognised at present; thus, it is not

possible to achieve a perfect probability estimation. Both

deterministic and probabilistic scenarios are used by the

insurance community.

Probability theory is critical in the industry, where

the likelihood of future events is required to develop

a policy or price a premium. This is true even where

signicant uncertainty exists, and it is in this case where

deterministic scenarios can be particularly important

to address uncertainty, providing reference points on

a journey towards probability. It is also important to

note that probabilistic scenarios are only useful when

understood, and where decision makers are not familiar

with the theory, it may be unhelpful or even misleading.

For example, return periods, or recurrence intervals,

are standard calculations for describing the magnitude

of potential events – such as a 1-in-100-year ood – but

are prone to misconceptions and misuses that are well

acknowledged but still widespread.

In cases where

communicating scenarios to non-experts is required, it

may be that probability should not be depended on, and

a deterministic approach could have greater value.

29. (Renard, Alcolea, and Ginsbourger 2013) 30. (Lorenz 1963) 31. (Serinaldi 2015)

Cambridge Centre for Risk Studies

Scenario Best Practices: Developing Scenarios for the Insurance Industry

3. Using Scenarios in the Insurance Industry

This section documents the various roles that scenarios play within the insurance industry and

explores who is using, or could better use, scenarios. The application of scenarios can be broadly

categorised into universal uses of scenarios, and those specic to the insurance industry.

3.1 General Scenario Uses

The action of developing a scenario carries a series of benets, some of which are simply a consequence of

the scenario building process. We highlight a series of outcomes that were often cited by insurance industry

representatives whose comments informed the development of this report, and were notably consistent,

regardless of a scenario’s intended use or application. These are individually discussed below.

Supporting Decision Making

Scenarios can be highly eective tools in support of

decision making, oering a creative and structured

mechanism to test and validate decisions in a scenario

planning process. When managers make decisions about

the future, concerning either near-term operations or long-

term strategy, they cannot only expect positive outlooks,

and must also be proactive in preparing for negative

events. This proactive outlook is a key characteristic of

enterprise risk management, contrasting with the typically

reactive approach of traditional risk management.

Scenarios facilitate discussion on how risks can be planned

for and be managed or mitigated eectively with robust

decisions and strategies. Further, decision makers must

factor in the associated risks on the belief of a certain set of

assumptions, the validity of which scenarios test.

Aiding Communication

Scenarios routinely function as a communication tool in

the industry, facilitating the sharing of ideas, risks, or

responses. Communication may occur formally during

the distribution process, or more informally during

the development and research phase. Scenarios are

especially valuable when discussing abstract ideas, or

complex risks, as they provide examples and context

to the issue, ensuring a consistent interpretation and

understanding. Furthermore, by providing a well-crafted

scenario which is mutually accessible to multiple parties,

consumers of the scenario have an equal platform

for communication across dierent sectors or areas

of expertise.

Demonstrating Due Diligence

The insurance industry is required to practice due

diligence to keep themselves and their insureds safe.

One component of due diligence is recognising what

risks exist, and where vulnerabilities exist within the

organisation. Scenarios assist in achieving this due

diligence, as they provide a systemic and comparable

platform for examining these risks, and a sheltered

sandbox to identify and test potential vulnerabilities.

These exercises can aid in answering questions such

as what silent exposure may exist, or if there is clash

potential on existing policies. By addressing areas

of uncertainty and taking informed risks, insurers

can demonstrate that potential consequences have

been considered and reasonable precautions have

been taken.

Identifying Bias

Scenarios provide a platform to explore hypothetical

outcomes and identify potential (dis)inclinations or

partialities that organisations have towards certain

situations and decisions. Taking a broad set of

scenarios mitigates well known behavioural eects like

conrmation and availability biases. Such an approach

also allows for alternative responses to be compared

to a baseline. By setting these processes up in advance,

insurers can be mindful of these issues and take

proactive measures to ensure that the decision-making

process remains objective. This yields a systemic benet

that is realised over time.

Cambridge Centre for Risk Studies

Scenarios to Support Robust Decision Making

Deep Uncertainty

Deep uncertainty exists when decision makers do

not know, or cannot agree on, a system in question

and its boundaries, the probabilities of model inputs,

or the consequences of interest and their relative

importance.

Such uncertainties persist for many drivers of global

change, including environmental, economic, or

technological developments. Societal perspectives

and preferences also change over time, including

stakeholders’ interests and their evaluation of

plans. As the future unfolds, plans are adapted to

developments, so decisions are part of the storyline

and an essential component of uncertainty.

Innovative analytical approaches

of ‘decision

making under deep uncertainty’ are emerging to

cope with uncertainty, and to help decision makers

evaluate robust and adaptive management strategies.

They help to build a consensus between stakeholders

with dierent values, priorities, and solutions, who

can agree on a decision for very dierent reasons.

The process reveals future threats, as well as

opportunities, confronting each plan. This paradigm

relies on exploratory modelling, typically involving

scenario approaches that harness speculation and

imagination to consider ‘unknown unknowns’.

Robust Decision Making

Robust decision making (RDM)

is one example of

a dened approach for analysing deep uncertainty,

which uses iterative, model-based scenario analysis.

The RDM methodology helps decision makers to

identify and improve robust strategies by testing

them against a very large exploratory scenario set

(of hundreds of possible futures) to reveal their

strengths and limitations. Statistical analyses of

model iterations identify the key conditions under

which strategies fail to satisfy their objectives. RDM

also has a participatory component, with stakeholder

deliberation used to dene (un)desirable outcomes,

and to rule out implausible scenarios.

Robust strategies will satisfy decision makers’

objectives in many scenarios, rather than being

optimal in any single future.

In other words, they

are ‘good enough’, rather than optimal options,

aiming to minimise regret rather than maximise

expected utility. RDM also helps to compare

strategies along other dimensions such as cost,

feasibility, and social acceptability.

RDM has been widely applied to explore where

deep uncertainty exists, including in the domains of

climate change

, risk and resource management

and insurance specic applications, for example the

feasibility of terrorism insurance

. An important

consideration of RDM is that it requires large

amounts of quantitative information and a high

degree of expert knowledge.

32.(Kwakkel, Walker, and Haasnoot 2016; Hallegatte et al. 2012) 33.(Haasnoot et al. 2013) 34.See (Haasnoot et al. 2013; Kwakkel, Walker, and

Haasnoot 2016) for discussions of various approaches to support decision making under deep uncertainty 35.(Tuck 2016) 36.(Olabisi 2017)

37.See (RAND Corporation 2019) for detailed use cases 38.(Olabisi 2017) 39.(Rozenberg et al. 2017) 40.(Rozenberg et al. 2017; Lempert et al.

2013) 41.(Lempert, Popper, and Bankes 2003; Lempert and Kalra 2011) 42.e.g. (Sayers, Galloway, and Hall 2012; Beven and Alcock 2012; Popper

et al. 2009) 43.(Lempert et al. 2013) 44.(Roxburgh 2009)

Sensitivity Analysis

Insurers are required to make assumptions and

estimates daily, considering many dierent variables and

evaluating their potential interaction with one another,

towards making a cost-benet analysis. Scenarios can

assist in this process, as they allow users a sheltered

environment to explore uncertainties and investigate

alternative futures which may arise, incorporating a

greater appreciation of the direct and indirect impacts of

their decision choices today.

Addressing Uncertainty

Scenarios expand understanding of a range of plausible

outcomes, each supported by a dening sequence of

events. Humans inherently expect that change will

occur gradually and that the future will reect the

past. By generating deeper insight into the underlying

drivers of change, scenarios may demonstrate how

and why changes could develop quickly and otherwise

unexpectedly, and which drivers do or do not have the

ability to cause consequential change.

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Systems Thinking

The capability to capture interconnectivities between

complex systems, or ‘system of systems’, is critical

to scenario planning. Systems thinking is a holistic

approach to address complex interconnections and causal

relationships, rather than on snapshots and independent

aspects, of a problem. Given the abundance of resources

on this topic, only a brief overview is provided here, with

an emphasis on the importance of wholly understanding

systems when using scenarios. The approach exposes

that which is not immediately obvious, providing a lens

to detect underlying controlling forces and relationships

between individual components, to understand the entirety

of a system. The iceberg analogy

(Figure 3) is a useful

way to illustrate systems thinking and enable practitioners

to appreciate the deeper perspective. As humans, we

typically notice events in the world around us (the ‘tip’ of

the iceberg) in a reactive and counteractive mode, only

seeing a small part of the underlying dynamics. Only when

we look below the (water) surface for patterns of behaviour

can the event be better understood with scenarios that

explore how interconnectivities may control the future.

Delving further into understanding these dynamics at a

structural level enables exploration of the structural level

of various risks facing an organisation.

3.2 Insurance Use Cases of Scenarios

In addition to the generic benets that scenarios provide

during the development and decision-making process,

scenarios have specic and unique benets to the variety

of users within the insurance industry (Table 1). This

report proposes that the insurance applications can

be categorised into four use cases. These uses overlap

and share common traits and should be recognised

as interconnected components. These use categories

are summarised in Table 2, and are described on the

next page.

(Enterprise) Risk Managers Regulators

Accumulation Controllers Claims Managers

Underwriters Risk Capital Controllers

Pricing Actuaries Business Planners

Advisors Investment Manager

Transaction Intermediaries Product Designers

Table 1: Example scenario users within the insurance industry

Use Category Description Example Uses Trend Shock

Understanding Tail Risk Identify extreme, low probability risks Reinsurance purchasing,

counterfactualanalysis

✓ ✓

Understanding

EmergingRisk

Identify new and undened risks and

policies

Understanding policy wording, trialling a

concept, problem verication

✓ ✓

Strategic Planning Identify risks and opportunities for

anorganisation

Short-term and long-term planning,

particularly with regard to strategic

response to opportunities or threats

✓

Accumulation Management Identify maximal potential losses and

potential for risks to scale

Probable maximum loss (PML)

identication, clashscenarios

✓

Table 2: Uses of scenarios within the insurance industry

Figure 3: Systems thinking – the iceberg analogy

(adapted from Senge, 1990; Van der Merwe, 2008)

Events

Patterns of Behaviour

Structure of

the system

Casual relationships,

driving forces,

assumptions

Increasing ability to inuence and learn

45. (Kauman 1980) 46. (Van der Merwe 2008)

Cambridge Centre for Risk Studies

Understanding Tail Risk

Within the insurance industry, a tail risk refers to an

extreme and highly unlikely event which has signicant and

often immediate consequences. The term ’tail’ originates

from statistics – for example, if outcomes are distributed

according to a standard bell curve (normal distribution),

then outcomes that are more than three standard deviations

below or above the expected outcome (the mean) only occur

in 0.3% of cases – these are the tails of the distribution.

These extremes, with either positive or negative impacts,

are, by denition, highly improbable and dicult to predict.

Within the insurance industry, the focus is typically on

the severe negative impacts that result in losses, although

recognising the improbable positive outcomes also are

of interest. As the exercise is hypothetical, variables can

be adapted and controlled to explore dierent potential

outcomes and their consequences. These scenarios assist in

answering the hypothetical question “what if?” and provide

a reference point to build from. As events unfold and the

probability distribution of a risk evolves, the denition of a

tail risk may also change.

Understanding Emerging Risks

An emerging risk is a new risk, changing risk, or novel

combination of risks for which the broad impacts, costs and

optimal management strategies are not yet well understood.

Emerging risks are driving an increase in insurance

exposure. Emerging risks oer opportunities for new

products in the market but are also a potential threat to

capital adequacy. Understanding the breadth of the risk and

its potential consequences can allow insurers to embrace

emerging risks and limit unintended exposure. Scenarios

are a tool in exploring these risks, as they provide a

platform to develop hypothetical events and consequences.

As emerging risks are fundamentally uncertain, scenarios

allow exploration of hypothetical events through which

these uncertainties can be stressed. This provides a gauge

of the potential hazard severity or scale, and what limits or

precautions should be taken at the time. When a specic

outcome has not occurred in reality, scenarios can be used

to see what pay-outs may be triggered, and how dierent

policies may be interrupted should such a risk emerge.

Strategic Planning

Strategic planning is the process by which an organisation

denes its strategy for the future, with consideration of

trend risks which have the potential to threaten strategic

goals. This can range from determining resources and areas

of investment, to setting priorities and targets. Specically,

scenarios are often used as they allow users to identify and

adjust factors which may impact these goals. Scenarios can

be used to project what may occur in the short- and long-

terms if conditions remain consistent, but also estimate

how the future might vary with various internal and

external inuences. Scenarios fundamentally support the

needs of communication across dierent stakeholders and

areas of expertise and provide a communal platform for

discussion and problem solving. When considering strategic

planning, members of an organisation can contribute to

scenario building, creating a holistic planning process.

While, as previously discussed, trend risks are typically

not insured via the current range of insurance products,

insurers are increasingly being called on to address

trend risks. Therefore, strategic planning should not

only be considered as relevant to an insurer’s internal

risk management, but also as an opportunity for insurers

to address trend risks that threaten the strategies of

organisations that serve the industry with business.

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Cyber Terrorism: AnEmerging Risk

Cyber terrorism has remained one of the most

notable emerging risks within the insurance industry

and continues to build its reputation as a potential

cyber insurance game-changer. Cyber terrorism

is dened by the Federal Bureau of Investigations

as: ‘any remediated, politically motivated attack

against information, computer systems or computer

programs, and data which results in violence against

non-combatant targets by sub-national groups or

clandestine agents’.

Cyber security professionals, governments, and law

enforcement are becoming increasingly concerned

with the potential for traditional terrorism groups

to exploit the cyber landscape to cause physical

destruction and human injury.

Insurers have

been reluctant to oer coverage for cyber terrorism

in their insurance products due to signicant

uncertainties surrounding the probability and

severity of cyber terrorism events. In a recent review

of 26 cyber armative policies, only 12% of cyber

insurance products oered terrorism coverage

in 2016.

The ambiguity around the process of

attribution of cyber attacks to terrorist actors has

also compounded this uncertainty.

Scenarios have played a key role in the developing

the market for cyber terrorism coverage by

quantifying the potential exposure (particularly silent

exposure) insurers could face when a cyber terrorist

event occurs. Research commissioned by Pool Re, the

UK’s mutual terrorism reinsurance pool with the UK

Government, highlighted realistic cyber terrorism

scenarios that could result in signicant physical

damage and thus trigger silent exposure in property

lines. In response to the identication and estimated

quantication of this cyber terrorism insurance

gap, Pool Re have expanded their terrorism product

coverage to include explicit cover for property

damage and the resulting business interruption

caused by cyber terrorism.

Accumulation Management

Accumulation management, also described as

aggregation risk management, considers what risks exist

within a portfolio of insurance policies for large numbers

of claims to arise from the same underlying cause or

event. Scenarios of the events that could cause large

numbers of simultaneous, correlated losses identify the

potential systemic attributes that policy holders might

share. This allows insurers to test the loss potential for

their specic portfolio to those extreme losses. They also

enable the exploration of hypotheses of how the risk

might scale or increase in magnitude if a variable were

to change. By estimating how large an event might be

and how likely it is to occur, an insurer can eectively

protect their exposure through limits and exclusions. A

use of the scenario might be to help insurers design their

terms and conditions, product design, and the limits and

deductibles they might oer to protect themselves from

taking large systemic losses.

Similarly, in identifying what the maximum loss for

insurers might be, exposure limits can be introduced to

limit what is insured, or what inclusions and exclusions

are considered. This type of scenario is referred to as the

Probable Maximum Loss (PML) and is often required

when deciding a risk appetite or tolerance of loss for a

line of insurance business and can be used to regulate

the amount of new business that is underwritten in

the future.

Scenarios play a valuable role when discussing

accumulation and maximum loss, as they provide a trial

ground to explore what types of events would cause

broad impacts, especially across multiple portfolios,

or classes of insurance. They are also helpful when

considering limits and exclusions, as insurers can trial

what amount of risk they feel comfortable accepting, in a

controlled environment.

47. (Alford 2017) 48. (Broadhurst et al. 2017)

49. (Risk Management Solutions Inc, in collaboration with Cambridge Centre for Risk Studies 2016)

50. (Evan et al. 2017) 51. (Insurance Journal 2018)

Cambridge Centre for Risk Studies

3.3 Allocating Risk Capital

Capital modelling is used when deciding how much

capital should be priced and reserved to cover potential

risks.

The role of capital allocation modelling in the

insurance industry is typically focused on ensuring

capital requirements for regulatory purposes are met,

without signicantly impacting daily underwriting

decisions.

This is a method known as ‘Regulatory Risk

Based Capital’. Under Solvency II pillar one, insurers are

obligated to meet the Minimum Capital Requirement and

Solvency Capital Requirement.

To meet these requirements, insurers can either use

a ‘one size ts all’ prescriptive model provided by the

regulator, which uses a standard formula, or else develop

an internal capital model (ICM).

The use of partial or

fully internal capital models goes beyond the scope of

regulatory compliance by providing greater insight into

a company’s risk prole and is essential for navigating

towards ecient capital allocation.

To validate a modelling process of capital allocation,

insurers often develop scenarios to assess the resilience

of these capital reserves to shock and trend risks.

Scenario modelling is used to validate insurers decisions

on the allocation of capital across lines of insurance by

estimating their exposure to a disaster scenario.

Using clash aggregation modelling techniques, insurers

can test their capital allocation reserves based on

a disaster scenario that impacts multiple lines of

insurance. Clash modelling allows insurers to better

understand their potential losses across multiple

lines of business and develops a narrative for capital

requirements necessary to cover these losses.

Solvency II

In 2009, the European Union Insurance and

Occupational Pensions Authority (EIOPA),

nancial regulator body of the European Union,

introduced the EU Solvency II Directive. The

directive aims to homogenise EU insurance

regulation and focuses on an enterprise risk

management approach towards required capital

standards.

Solvency II is primarily concerned with

the level of capital reserves insurance companies

should hold to reduce the risks of insolvency.

This

legalisation was driven by the events of 2008, when

the Financial Crisis and subsequent Great recession

highlighted the necessity for insurance companies to

manage their capital allocation to remain solvent. To

ensure the solvency of EU insurance companies, the

Solvency II program has three main areas, known

as ‘pillars’:

Pillar 1: Financial Requirements

Pillar 2: Governance and supervision

Pillar 3: Reporting and Disclosure

Pillar 1, known as the ‘quantitative pillar’, puts

demands on the required economic capital that

insurance companies must hold.

The pillar

stipulates two thresholds that insurance companies

must adhere too: Solvency Capital Requirement

(SCR) and Minimum Capital Requirement (MCR).

The SCR requires insurers to estimate the level of

capital needed to meet quantiable risks on their

existing portfolios, including one year’s expected

new business.

MCR is the lower bound of the SCR

and is the level of capital in which regulators would

consider the insurer to be in signicant danger of

insolvency. The MCR is calibrated in the Solvency II

regulations as 85% value-at-risk over one year from

valuation date.

Pillar 2 is known as the ‘qualitative pillar’ which

sets out clear requirements regarding how the

quantitative objectives of pillar 1 should be achieved.

The pillar requires companies to design eective

systems of governance which are proportionate to

the nature, complexity and scale of operations.

insurer must have written internal policies in relation

to risk management systems and internal controls of

the organisation. Under the Own Risk and Solvency

Assessment (ORSA) area Solvency II, insurers must

action a signicant self-assessment of the risk they

are exposed to in the short and medium run.

ORSA

aims to go beyond the modelling requirements in

pillar 1 and for the insurer to think about additional

risks the insurer may be exposed to.

Pillar 3 outlines the reporting requirements of

company’s risk with respect to which information

should be disclosed such as risk exposure and

concentration, and the frequency with which this

information should be reported.

52.(Cummins 2000) 53.(GIRO Working Party 2016) 54.(Vandenabeele 2014) 55.(Krvavych 2018) 56.(Boonen 2017)

57.Formerly known as the Committee of European Insurance and Occupational Pensions Supervisors 58.( Vandenabeele 2014)

59.(House of Commons Treasury Committee 2018) 60.(Boonen 2017) 61.(Lloyd’s of London n.d.) 62.(Vandenabeele 2014)

63.(Financial Services Commission 2009) 64.(House of Commons Treasury Committee 2018) 65.(Vandenabeele 2014)

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Cambridge Centre for Risk Studies

4. Shock Risk Scenario Development:

InsuranceStress Testing

There are many types of scenarios which have been developed by the insurance industry, for particular

applications and with diering development methods. During the literature review and consultation

process for this report, a wide variety of techniques and development processes were identied

in surveys of insurers and stakeholders. There is no standardised approach, consensus, or single

proto-typical method in use. Instead, there are common features across multiple methodologies.

In this section we focus on the underlying similarities and propose a series of checkpoints to ensure

developed scenarios can be applied widely and meet the needs of the broader insurance community.

4.1 A Checklist for Scenarios

The rst checkpoint in the scenario development

process is identifying who the scenario users are,

and how they hope the scenario will be used in the

future. Within the insurance community there are an

abundance of potential users with individual needs

and criteria for a scenario. The best way to gauge these

needs is by talking with the intended audience or users

and taking the time to understand what they hope a

scenario to achieve. Supporting the direct feedback from

future users, scenario developers should also reect on

the following:

Background Knowledge refers to the existing

familiarity that users have with the subject. This is

an indicator for the type of narrative and level of

description that should be provided. If the scenario

is proposing an event that is novel to the user, the

scenario and its associated documentation should

provide additional resources regarding the current

risk landscape and use additional description when

suggesting cascading impacts. Comparatively, if the

user is very familiar with the proposed risk event, the

scenario can address a dierent level of detail and

specication. The terminology and language used

to describe the risk should also be adapted based

on the familiarity of the end user with the subject

matter. As a standard practice, the scenario should

include enough research and reference material so

that it remains widely accessible, regardless of the key

audience’s familiarity.

User Resource is the amount of time and human capital

that the user can be expected to dedicate to read and

interpret the scenario. The resource allowance should dictate

the length and detail of a scenario. Should the scenario be

intended for a user interested in new and emerging risks

over a long-term period, more resources may be allocated to

reviewing an extensive report and background material. If,

for example, the user is focused on the clash risk of a specic

area, there may be a greater urgency to understand potential

outcomes and losses, and so fewer resources would be

available to review an extensive report.

Existing Guidelines refer to the rules and regulations

that may have been put in place by the user regarding the

scenario’s development and presentation. This is most

commonly seen in the case of regulators and authorising

bodies, who are required to review and compare large

numbers of scenarios in a short period of time. Prior to

the development of the scenario, one should check to

ensure that available guidelines required by the user are

reected. These guidelines might include the types of

scenarios required, the number of variants included, and

the loss estimation process.

Audience refers to how the user intends to share the

material, and whether it will remain as an internal

resource, or be published externally for public use. Public

scenarios will require signicantly more documentation

and reference material than those for internal use, to

mitigate potential ambiguity.

Scenario Best Practices: Developing Scenarios for the Insurance Industry

4.2 Scenario Development: a Stress Test Perspective

The next several checkpoints in scenario development

relate to its construction. The intention of the report is to

provide an accessible resource and guide for developing

scenarios, but the process is varied and often attuned

to the specic needs and interests of stakeholders.

Accordingly, the steps discussed below are recommended

checkpoints to consider as a practitioner works through

the scenario process. Users should adapt the methodology

to create scenarios which work best for them, given their

audience, resources, and desired use. Further, while it

outlines a linear step-by-step structure for clarity, we

encourage the scenario process to be an iterative one, in

which stakeholder engagement provides opportunities

for review and revision to ensure it succeeds in fullling

its aims.

A case study of a worked example of scenario development

using this framework is provided in Section 5, based on

the 2015 Lloyd’s Business Blackout Report

, originally

developed to address emerging cyber risk within the

insurance industry.

There are various types of scenarios which can be used

within the insurance industry (Section 2.2). This section

focuses specically on stress test scenarios, which are

among the most widely used scenario types and carry the

greatest potential for external agency development. Many

of the recommendations made for stress test scenarios are

broadly applicable to other forms of scenario development.

A stress test is an exercise conducted within the insurance

and nance community to explore an organisation’s limits

and dene the magnitude of event that would ‘stress’ the

institution. This is often measured in terms of nancial

capital, with stress tests imagining extremely expensive

events leading to an inux of insurance claims or nancial

losses. Stress tests often focus on rare and extreme events

(tail risks) which result in a dramatic impact.

Within the insurance industry, stress tests are a common

requirement for external regulation and internal planning

and are a recognised means of evaluating the robustness

of an institution, both currently and in the future. The

process of developing a stress test involves both creating

a hypothetical event and identifying the consequences

that event would have upon the organisation/insurance

sector. This report suggests that the scenario development

process is best considered an iterative progression rather

than a linear trajectory. This is due to ongoing adaptations

and adjustments which are made as a researcher learns

more during the scenario’s development. The process is

summarised in Figure 4. The process can essentially start

at any phase within the process, though most will nd it

logical to rst identify a risk or question. Others may start

by formulating intended scenario outcomes and work

backwards, which is a reverse stress test.

Exploring Knowns and Unknowns

In 2002, the US Secretary of Defense, Donald

Rumsfeld, stated:

“There are known knowns. There are things we

know that we know. There are known unknowns.

That is to say, there are things that we now know

we don’t know. But there are also unknown

unknowns. There are things we do not know we

don’t know.”

Although the saying was intended for a national

security perspective, the same intention can be

applied to a broader risk perspective and our

understanding of potential hazards. Rumsfeld

correctly suggested that there are certain

unknowns which we are aware of and can in

turn eectively prepare for. Alternatively, there

are unknown unknowns, or unidentied risks,

which are dicult to imagine or plan for in

advance.

Within the insurance community, it

is the unidentied risks which carry the greatest

uncertainty and can have some of the most

consequential nancial impacts. Although risk

management is designed to consider potential

emerging risks, there are some risks for which

management is not possible.

The role of stress test scenarios within the

insurance industry is to identify what type

of extreme events might occur, and what

impacts would result. By shifting away from

what is likely or known and focusing on what

is plausible (either now or in the future), a

broader interpretation of the risk landscape can

begin to emerge. It is unlikely that scenarios

will identify an unknown unknown risk, but

they can encourage organisations to widen

their knowledge of their known unknown

risk environment and reduce the surprise of a

shock event.

When researching the subject, it is also helpful

to identify similar historical incidents, which

can provide precedent for the proposed event.

Alternatively, counterfactual analysis considers

events which might have occurred but didn’t.

Counterfactual analysis can be especially helpful

when considering event outcomes which have

never happened before.

By anchoring aspects

of the scenario in existing events, the scenario

maintains it credibility and relevancy.

66. (Cambridge Centre for Risk Studies 2015) 67. (Rumsfeld 2002)

68. (Kim 2012) 69. The Economist 2017)

Cambridge Centre for Risk Studies

Step 1: Scope the Risk

This step aims to contextualise the objectives

and resultant decisions of the process. It

is logical to approach the scenario process with the

statement of a problem or research question to be

answered, most simply by dening a specic risk facing

an organisation. For example, within the insurance

industry, a common theme is to better understand

potential or emerging risks and identifying how they may

aect the insurer’s portfolio. A research question typically

starts broadly (e.g. ‘what are my silent cyber exposures?’)

and evolves to a more specic question during

development (e.g. ‘what are my silent cyber exposures to

an industrial explosion event?’). The problem or question

will be rened throughout the research and development

process. Stress test scenarios are particularly helpful

when identifying exposure to unlikely but extreme events.

Although these events are dicult to assign a probability,

extreme scenarios are still valuable in investigating

interconnectivities and indirect impacts.

Where the risk is uncertain or unknown, the research

question should instead aim to dene the issues or

vulnerabilities that the scenario exercise aims to expose.

In this case, the desired outcome of the scenario process

is likely to identify one or more risks to an organisation.

A reverse stress test is an alternative way of framing a

question or problem, instead focused on the hypothetical

point of failure that would cause an organisation to

become unviable. The scenario then forces users to

identify the vulnerabilities which exist and might

cause such an eect. Reverse stress tests are helpful in

highlighting the potential for failure and addressing any

false sense of security concerning the robustness of a

business and its resilience to a shock.

Step 2: Conduct Background Research

Background research should include

consultation of the relevant sources of

knowledge, including the academic literature for a

scientic understanding of the topic, as well as the

knowledge of insurance industry experts and associated

stakeholders within and beyond the sector. Each

dimension of risk should be considered: the hazards,

whether acute, shock events, or slow-onset, trend

phenomena; exposure of an organisation, sector,

market to a hazard; and vulnerabilities at each of

these levels that have the potential to drive loss. A

distinction should be made between idiosyncratic risks

that aect a single organisation, and systemic risks

that drive a major collapse in a specic sector or the

broader macroeconomy.

Various stakeholders in the insurance community are

often willing to help at this stage and can provide nuanced

advice specic to certain risks. Speaking with subject

area experts outside the insurance industry can provide a

balanced perspective, as they have dierent perceptions

of risks, vulnerabilities, or potential impacts. Establishing

both an internal and external perspective is valuable

for this reason. This process alone may be valuable in

widening the view of known unknowns, highlighting risks

that do not yet feature on an organisation’s radar.

Step 4

Step 5

Develop

Candidate Scenarios

Develop

a Narrative

Figure 4: Outline of the scenario development framework for the insurance industry

Step 1

Step 8

Step 3

Step 6

Step 2

Step 7

Scope

the Risk

Evaluate

and Update

Frame the

Scenario(s)

Assess Impacts

and Materiality

Conduct

Background Research

Communicate

and Act

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Step 3: Frame the Scenario(s)

Here, framing the analysis refers to

consideration and denition of the key aims,

benets, and characteristics of the scenario process. It is

the process of determining what type of scenario is most

appropriate for the current use case.

As discussed, to be eective, scenarios should blend

approaches and typologies to t their purpose. Figure 5

intends to guide this framing with a series of questions

that aim to provoke consideration of key practical

decisions in the process:

Aid Communication Demonstrate Due Diligence Identify Bias Sensitivity Analysis Addressing Uncertainty

To contextualise complex

risks and promote

stakeholder understanding

To assess risk exposures

and understand their

nancial implications

To explore partialities

that stakeholders may hold

towards certain decisions

To investigate the power

and variance of controlling

variables on a risk

To expand understanding

of the range of

plausible outcomes

Trend Risk Scenario Shock Risk Scenario

Slow-onset, trend phenomena

that emerge gradually over time

Sudden-onset, shock events

that occur quickly or unexpectedly

Exploratory - To ask ‘what if?’ Normative - To ask ‘what for?’

To stimulate imaginative thinking about the future

and widen understanding of available options

To better understand the path to desirable futures

and evaluate the impact of decisions

Figure 5: Framing a scenario – scenario typologies and applications

What is the scenario for?

How can the scenario benet stakeholders?

On what timescale does the risk materialise?

Which is the more important scenario outcome?

Who owns and contributes to the scenario process?

Understanding Tail Risks Understanding Emerging Risks Strategic Planning Accumulation Management

To identify and understand extreme,

low probability risks

To imagine and comprehend

new or evolving risks

To dene a resilient strategy

for the future that alleviates risks

To explore possible extreme or maximal

correlated losses to insurance portfolios

Participatory - Bottom-up, co-production of knowledge Expert-Driven - Top-down, analytical

To incorporate stakeholder culture, knowledge,

and experience in the process and end product

To deliver rigorous scientic descriptions

of plausible futures to decision makers

Probabilistic Deterministic

To estimate the likelihood of occurrence

based on the variance of quantied causal parameters

To speculatively explore phenomena

that involve a high degree of uncertainty

Is the scenario required to dene the likelihood of an outcome?

Cambridge Centre for Risk Studies

Step 4: Develop Candidate Scenarios

Developing candidate scenarios typically

occurs in tandem with the research process and

involves imagining the distinct features and narratives of

potential scenarios. This should be a free-thinking exercise

and includes exploring threat types and characteristics,

key variables and controlling factors, as well as expected

consequences and impacts. Informed by the background

research, a list of candidate scenarios should be formalised.

It is helpful at this stage to make the candidate list as

extensive as possible. Developers should consider the key

question or problems during this process and attempt

to list scenarios that satisfy the required criteria, while

ensuring they remain plausible and true to the reality as

understood from the background research. For each of the

candidate scenarios, consider what risks are measured, and

what types of consequences they may lead to. Once a series

of potential scenarios have been listed, they will be ranked

and selected for further development.

There are several methods to rank and select scenarios,

which are largely based on user preference and a scenario’s

intended application. Two common approaches include the

use of an impact uncertainty matrix, and the uncertainty

breadth approach. In the impact-uncertainty matrix (Figure

6), outcomes are projected upon a matrix which gages both

impact and uncertainty. Scenarios which have high impact

and high uncertainty are thought to benet the greatest

from further research. This is followed by scenarios which

have either a high impact or high uncertainty. Those which

are certain and unimpactful are deemed least appropriate.

The uncertainty breadth approach is an alternative method

which selects scenarios that cover the greatest range of

outcomes. In this case, multiple scenario variants would

be selected that are the most dierent from each other and

had the least amount of overlap. The uncertainty breadth

approach is especially valuable for emerging risks which

typically have more unknowns.

Determining a severity

It can be helpful to estimate what the frequency and severity

of the loss may be to a rough rst order. This process

identies extreme yet plausible events and can provide

some context to the discussion. Insurers often discuss size

and severities in terms of ‘return periods’, or the estimated

time interval between the occurrence of similar events.

For example, a 100-year return period event implies a

severity level that only has a 1% chance of occurrence in a

given year.

When determining the event severity, it is also helpful to

consider the size and scale of the (1) trigger event, and the

(2) market response. Often what generates a stress event

is a cascading set of incidents which amplify the triggering

incident. In this perspective, the trigger event may be

common, yet the response may be very uncommon, which

results in an overall extreme and unusual event. In most

cases, it is advised to not make the scenario too extreme,

as ‘game-over’ severities are not actionable. Although

there will be scenarios which can be much more severe or

stressing, events which result in the failure of society or

destruction of an industry as a whole, are often not relevant

for traditional stress test users. This is because, in such

cases, the insurance industry would be unable to support

such colossal losses. Scenarios should be kept potentially

plausible and broadly recoverable.

Step 5: Develop a Narrative

A detailed narrative expands the scenario

description to provide the required detail on

which to build the analysis. The narrative should include

a variety of dimensions of a future event, including

context, triggers, timelines, geography, responses,

and implications. When determining these factors,

it is helpful to once again refer to the intent of the

exercise expressed when scoping the scenario (Step 1).

Geography, timelines, and event triggers are particularly

relevant within the insurance context, as they determine

whether insurers bear liability and which policies will

be triggered.

It can be helpful to consider scenarios as stories which

have a beginning, middle, and end, and the narrative

guides the reader through the events and their

implications. The narrative is imperative for providing

logic and reasoning to the proposed events and assists

in making the scenario more applicable and relatable.

An eective narrative can also create broader interest

and relevancy, increasing the scenario’s utility. Research

conducted in the previous steps should be applied to help

guide the narrative based on historical precedence. In

providing rich descriptive details and elaborating on the

cause-and-eect processes, a more holistic impression of

the hypothetical event can be drawn, which assists when

determining its direct and indirect impacts. One way to

expand the narrative in an insurance specic context is

by using coverage trigger pathways.

High Importance

High Uncertainty

Low Importance

High Uncertainty

High Importance

Low Uncertainty

Low Importance

Low Uncertainty

Figure 6: Impact-uncertainty risk matrix

Increasing Uncertainty

Increasing Impact

70. A return period should not be interpreted as a prediction of when the event will occur.

Instead, a return period is the probability of occurrence.

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Coverage trigger pathways

Coverage trigger pathways are contextual details that

can be added to a scenario and are intended to stress

a specic type of insurance by impacting the insured

assets. An asset can be either tangible or intangible,

and can include: properties (location), contents, people,

products, mobile assets, organisations, projects, events,

trade agreements, contracts, or intellectual property.

The approach requires the users to view all the classes of

business insured in the footprint and identify potential

events which would stress that line. This can be done

in tandem with the traditional scenario narrative and

provides a series of statements to base impacts upon

when completing the scenario impact assessment.

Figure 7 illustrates an example of a scenario narrative,

and some coverage trigger pathways which could emerge

as specic insurance losses.

Designing a Scenario Narrative

The scenario narrative provides the context for a scenario and is needed when determining the direct and

indirect impacts of an event. When creating a scenario narrative, several variables should be considered.

Event Trigger: The narrative should specify the

trigger of a hazard event, for example the occurrence

of an earthquake, an act of terror, release of a new

malware, or industrial accident. Further, it should

specify any amplifying factors which might exacerbate

the event, or whether cascading events may be

triggered by an initial shock.

Location: Consider where the event trigger occurs

(if applicable) and how far reaching the impact is

felt. Furthermore, for certain types of insurance,

such as liability, the jurisdiction can dictate what

types of coverages are available, and the legal liability

organisations might hold. Indirect impacts should

also be considered in addition to primary impacts.

For example, if a terrorism event occurs in South

Korea, what international life insurance policies

might be triggered for tourists? What international

manufacturing operations would be impacted by the

interruption? With increasing global networks, event

impacts are not limited to the surrounding geographic

footprint and can quickly expand to a global scale.

Timing: The timing of an event occurrence, or a

timeline of events, should be outlined within the

scenario narrative. Variables such as the length of an

occurrence or the speed and eectiveness of a response

is critical for assessing the criticality of decisions.

Impact: The narrative should highlight who and

what is impacted by the occurrence, both directly

and indirectly, and how the severity of impacts is

distributed. Impacts, such as the eects on human

lives and livelihoods, business disruption, physical

damage and destruction, are unlikely to occur

uniformly across space and time. The narrative

should also consider individuals who face delayed

impacts, especially for occurrences which have a

longer timeline.

Recovery: Just as important as understanding the

cause and eect of the occurrence, the narrative

should also consider the recovery process. Specically,

the narrative should identify who is involved, the types

of resources recovered, and how long the recovery

takes. This directly inuences the impacts that the

occurrence has, and the scale of the event.

A downstream energy insurance claim is

triggered for pipeline physical damage

and business interruption.

Pipelines are damaged as a result of

mudslides, triggered from the hurricane.

A professional liability insurance claim is

triggered by the contractor.

A hurricane occurs in the Gulf of Mexico,

impacting the energy sector.

Scenario Narrative Coverage Trigger Pathways Impacted Insurance Line

An error by a contractor during a platform

inspection leads to structural failure during the

hurricane. Injuries are caused by the failure.

An upstream energy insurance claim is

triggered for platform physical damage

and business interruption.

Oshore marine platforms and accessories

aredamaged during the hurricane.

Figure 7: Hurricane coverage trigger pathway for the energy sector

71. (Cambridge Centre for Risk Studies, in collaboration with Risk Management Solutions, Inc. 2018)

Cambridge Centre for Risk Studies

Scenario variants

Once the scenario narrative is completed, its variants

can be considered. Variants are dierent versions of

the scenario where variables are adjusted to account for

uncertainty. Examples of adjustments can include the

magnitude, duration, or social reaction to the event. By

adjusting a variable, losses will increase or decrease.

Scenario variants are frequently used to show a

potential range of outcomes and their severity. The

number of scenario variants is dependent on time and

resources, but typically ranges from three to seven.

When considering the number of scenarios, the centre

stage eect

dictates that users are more prone to select

the middle scenario.

As a result, there is benet in

presenting an even number of scenarios, to dissuade the

tendency to select the centre as default.

Step 6: Assess Impacts and Materiality

Once the scenario event has been developed,

the next step is understanding what the

potential impacts are to an organisation within the

insurance industry, and/or systemic impacts to

the industry and wider social, environmental, or

economic systems. Consideration of dependencies and

interconnectivities between shocked systems with a

‘multi-threat’ approach is critical to understand the

hidden and cascading impacts beyond the expected (see

Section 3.1 on ‘systems thinking’). For example, the 2011

Tohoku earthquake triggered a tsunami that caused the

meltdown of the Fukushima Daiichi nuclear reactor

and the 2017 WannaCry cyber attack impacted the UK’s

National Health Service when sta were locked out of

their outdated operating system

Within the insurance community, specic attention is

given to the impacts on assets.

Assets are insured by an

insurance organisation, and should specic conditions

or circumstances arise resulting in nancial loss, a

payment would be made to the insured. Insurers will

have a catalogue of assets which they insure, and their

primary interest is how the stress test will aect these

assets. Depending on the type of asset, dierent types

of insurance

and insurance coverages

will be made

available. These categories of insurance are referred

to a ‘classes’ and are often underwritten together.

Accordingly, many stress tests are written to ‘stress’ a

specic insurance class, or series of classes.

When determining the assets which will be impacted by

the stress test, the following steps are recommended:

Determine where the impacts will occur

In the case of a physical event, this is the geographic

footprint of the direct and indirect impacts. For non-

physical events, this is identifying what systems or

networks might be impacted.

Identify what assets

will be impacted by the event

Consider what is insured within the footprint of

the scenario. Tangible assets would be physically

present within the footprint, while intangible assets

may be dependent on someone or something within

the scenario. Consider business operations, liability

agreements, contracts, and insured persons, in

addition to traditional physical assets.

Identify the incident rate

This might be the proportion of assets impacted, or the

replication of the risk. It can be helpful to consider the

incident rate as a severity scale, as it is rare that assets

all face the same degree of impact. Instead, gradation

can be applied to mimic impact variation.

Identify existing insurance

coverages available to asset

For the assets which are impacted by the stress test,

identify what form of insurance they would be covered

by. One asset may have multiple policies covering it for

dierent types of risks.

Upon identifying the aected assets and insurance

policies, the next step involves estimating the nancial

loss. Due to the various limits and deductibles which

are considered when paying an insurance claim, the

estimated total loss can be expressed in several dierent

forms. Common examples include macroeconomic loss,

ground-up loss,

gross loss,

and investment portfolio

losses.

These are summarised in Figure 8.

The methodology followed in calculating these loss types

varies depending on organisation and the amount of

detail available regarding regional assets. In some cases,

the specic assets and/or their exposure value may not

be known to the scenario developer. In these cases,

it is best to estimate based on insurance penetration,

historical losses, and known commercial values. If the

total insured value of an asset is available, one might

72. The centre stage eect refers the tendency for people to select the middle item when presented with a list of options (Valenzuela and Raghubir 2009a) 73 (Valenzuela and

Raghubir 2009a) 74. (Zastrow 2019) 75. (Rue et al. 2018) 76. Assets are anything tangible or intangible which is owned or can contribute value 77. Types of insurance are

products specic to a class of insurance. For example, within a marine insurance class, wet insurance includes cargo insurance, hull insurance, and marine liability insurance

78. Examples of insurance coverages include physical damage, bodily injury, or business interruption. 79. Examples of insurance classes include life, health, auto, property, or

marine. 80. Macroeconomic loss refers to losses felt by the whole economy, and considers how regional, national and global economies may react to the event 81. Ground-up

loss is the entire insurance loss, and does not include deductibles, limits or reinsurance. 82. Gross loss is the ceding company’s loss irrespective of any reinsurance recoveries

due. This can be interpreted as the ground up loss which has factored deductibles, limits and other connes, but doesn’t include reinsurance 83. Investment portfolio losses

refer to the cumulative losses seen in a collection of assets owned by an organisation or individual.

Scenario Best Practices: Developing Scenarios for the Insurance Industry

refer to the severity of the damage, and propose what

proportion of the asset’s total value would be aected.

Expert judgement can be a valuable resource when

estimating loss. Underwriters can be especially attuned

to relative asset values, and it is helpful to have their

guidance and review in this process. Depending on

the type of loss being calculated, deductibles, limits,

or reinsurance restrictions may also be applied to

the estimate.

Step 7: Communicate and Act

Eective communication of realised scenarios

controls the ecacy of the process. Although

it may appear trivial, a catchy name that provokes

interest is an often-overlooked scenario element that

can aid its successful uptake and communication. Then,

meaningful, comprehendible, and interesting outputs are

essential, and are likely to include both qualitative and

quantitative components addressing the scenario impacts

and materiality. It is important for the narrative to

explain and contextualise the results, and it may often be

helpful where multiple scenarios are explored to focus on

the most probable and use this as a reference to explore

others. The scenario outcomes can inform and prompt

decisions and actions where this is the intended purpose,

for example concerning an organisation’s strategic risk

management practices or to manage accumulation risk.

Consider how and to whom the scenario is delivered;

this may include internal and or external stakeholders.

Internal partners may include dierent underwriting

bodies, risk managers, or strategic planning parties,

while external partners might include regulating

authorities, governing bodies, or investing partners.

When communicating the scenario, one should consider

the audience’s background on the risk topic and the

scenario development process and ensure that they have

enough background information to interpret the ndings.

Importantly, it must be made clear what the scenario is

intending to accomplish, and how the ndings should

be interpreted. The scenario is not a prediction of future

events, but instead a thought exercise into plausible

futures. Further, stakeholders should feed back to ensure

a scenario and its results are plausible and relevant to

their needs. It is common for changes and experiments

to occur at this stage, especially if dierent variations or

severities of the scenario are being trialled. Most notably,

some may be interested in identifying what variables

will ‘tip’ the balance to the point that insurers would be

unable to nancially recover from the event.

Step 8: Evaluate and Update

To complete the scenario process, it

is encouraged to evaluate whether the

objectives of the exercise were achieved. This is

likely to involve a consultation with participants and

associated stakeholders, to review whether a scenario

is plausible and if it can be useful in developing the

users’ understanding of a risk and informing decisions.

Listen to contrary opinions, as a method to overcome

groupthink and build on insights from a range of

sources. Consider if the stress test scenario answers the

initial problem, and if the severity of the event meets

the intended targets. This process has been expressed

in a linear step-by-step process, but it is critical to use

this evaluation process to identify and address any

weaknesses in the scenario and applied analytics in an

iterative process.

Sometimes, the most interesting and insightful

scenarios are those that initially appear to be the least

probable. Scenarios have varying lifespans, intended

to be discarded after the exercise is nished or to be

kept and reused over a period of years.

Therefore, it is

important to acknowledge that scenarios are dynamic.

The possibility and character of a scenario will change

over time as controlling factors evolve, as will its impact

on the insurance industry – including practices, product

oerings, and exposures – and wider society – including

social, economic, and environmental characteristics –

and so it should be updated periodically to maintain

relevance and continued utility.

84. (Roxburgh 2009)

Co-share Limit

Deductible or Self

Insured Retention (SIR)

Total Insured Value

Gross Loss

Ground Up Loss

Figure 8: Dimensions of insurance loss

Cambridge Centre for Risk Studies

4.3 Considerations and Constraints

When discussing scenarios, there is potential to overemphasise their mechanics and impact without

providing a balanced critique. This section of the report is dedicated to highlighting the limitations of

scenarios and raising awareness of appropriate precautions that should be considered.

Constraints of Probability

Probability is the likelihood of an event occurring and

has guided extensive decision making in the insurance

industry. The application of probabilities to scenarios

has also been a point of contention, however, as there

are concerns that it may introduce false condence and

awareness to areas of high uncertainty. To appreciate

these concerns, it is helpful to explore probability

as a concept. Probability is usually understood via a

frequency calculation: the number of ways of achieving

a specic outcome, divided by the total number of

possible outcomes. Underpinning probability theory, the

Kolmogorov axioms state that you must know the event

space, or all potential outcomes, in order to dene the

likelihood of one specic outcome. This perspective implies

that probability requires the subject exist within a closed

system, where all potential future events can be identied

at present and can be accounted for when considering

likelihood.

As this is not an option in practice, insurers

often accept the ergodic axiom, which suggests that the

future is going to reect the past and present and use that

to guide their prediction of the future.

This introduces

a series of problems when considering emerging risks,

or events with limited historical precedent, and fails

to prepare users for events which have never been

experienced before. In these environments, probabilities

which are dependent on historical events may be

misleading, or may imply an over-condence that all

potential outcomes have been explored and addressed.

Common critiques of the use of probability when

considering future risks are not aimed at eliminating

the use of probabilities, but instead at addressing the

uncertainty and assumptions that they bring.

A related and more theoretical concern is that these

probabilistic scenarios remain deterministic in their

assumptions. This is because the probabilities of future

events are only identifying specic outcomes, which

are determined by the modeller.

For example, when

considering the possibility of a selected risk occurring

in a specic location, one may reduce all outcomes to a

short list of, say, four outcomes of dierent severities

and likelihoods. While this may be helpful in simplifying

the analysis of that risk, it may also lead to an incorrect

sense of certainty that the future may only materialise

as one of these four outcomes, and introduces a new

risk of not accurately representing the uncertainty of the

future.

Furthermore, one must recognise that probability

applications do have a place but cannot always be applied.

For some risks, the probability is essentially irrelevant, as

the outcome is too crucial. Examples include extreme tail

risks, or highly unique events which cannot be aggregated

into a composite group.

In such situations, probabilities

are less important than recognising who is responsible

as the risk holder of last resort, and understanding their

preparation and response to that risk given its potential

impact and available resources.

Cognitive Bias

When designing and using a scenario, one must

consider the eects of cognitive bias on decision making.

Cognitive bias is an umbrella term which describes

the input of subjectivity onto decision making and our

interactions with the environment. These biases can

guide what risks we choose to focus upon, and how we

interpret given information.

When discussing cognitive bias, it is helpful to provide

some perspective as to why it is important, and why it

occurs. The world provides the brain with a complex

environment with enormous amounts of information

to process. This complexity is compounded when

decisions need to be made. To reduce the eort and

delay until a decision is made, cognitive biases develop

based on interactions with the environment. These

cognitive simplications are commonly referred to

‘heuristics’ and can be thought of as mental shortcuts or

our ‘intuition’.

While an individual’s cognitive biases and developed

heuristics likely developed for adaptive reasons to

simplify decision making, these intuitive judgements

based on intuitive probability and frequency judgements

have the potential to lead to judgement errors. Although

many of these errors are unavoidable, we can be mindful

of their impact and take precautionary steps if needed.

As a rst example of a bias, we pick out overcondence

which describes the tendency to be overly optimistic

in ignorance of contrary evidence, often in the context

of setting a goal. Nobel laureate Daniel Kahneman

identies overcondence as “the most signicant of the

cognitive biases”.

85. (Derbyshire 2017a) 86. (Derbyshire 2017a; Davidson 1991) 87. (Roxburgh 2009)

88. (Derbyshire 2017a) 89. (Derbyshire 2017a) 90. (Derbyshire 2017b) 91. (Kahneman 2011)

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Below we discuss two common biases inuencing scenario development.

Availability Bias

Decisions on probability and judgements are based on the

ease with which relevant instances come to mind

Frequent occurrences of an event mean individuals have

plenty of relevant experience to draw from when judging

the probability of an event, so basing judgments on

availability is sensible and people’s frequency judgments

are often very accurate. However, availability can be

biased if our experience of past events does not reect the

true frequency or if an event is easier to recall for a reason

other than its frequency.

These biases can be introduced by the environment or the

individual. Rare events are often given disproportionate

publicity and are correspondingly more mentally

available than their true frequency would merit.

Similarly, events that individuals have experienced

personally with an emotional response are much more

readily available.

The availability bias is helpful to consider when

evaluating risks, as we generally underrepresent common

risks which do not receive as much attention while we

over represent recent heavily reported risks, or risks

which we have experienced ourselves. When applying

the availability bias to scenario development, one should

attempt to understand the viewpoint users of the scenario

have, to create scenarios that make the risk accessible

and personal.

Anchoring Bias

Final estimates or judgements are often reached by

adjusting away from an initial “anchor” value, but these

adjustments are often insucient.

Anchoring bias occurs when we focus decision making on

an initial piece of information. An initial value is used as

a reference when evaluating and comparing additional

information. The outcome may be skewed or inuenced

by the initial, possibly arbitrary, anchor. An example of

this may be seen when estimating catastrophic losses.

We often base our estimated losses on a historical

precedent and are unprepared when a substantial

variation occurs, such as was seen in Hurricane Andrew

or the World Trade Centre terrorism attack. Stress

test scenarios are routinely used to introduce higher

anchoring values, which insurers can plan against.

Accordingly, stress test scenarios should be gauged to

stress the system and introduce new precedents for

insurers to plan against.

Communication Failures and Misinterpretations of Outcomes

Scenario modelling outcomes are most eective when

there is a governance structure in place that promotes

clear lines of communication to external and internal

stakeholders. Eective communication enhances risk

management by providing information that may be used

to enhance processes, identify emerging issues, and aid

in overall decision making.

If scenario analysis is not

communicated suciently, the outputs of the scenario

could be ignored or the misinterpreted which then

negatively impacts business performance.

One of the most common misinterpretations which was

reported via this report’s consultation process was the

interpretation of results as predictions rather than as

possible futures. Scenarios in an insurance context are

not developed to assist insurers in preparing for a specic

incident, as it is highly improbable that the specic

outcome will occur. Instead, they must be interpreted as

thought exercises. By ensuring that the communication

eorts reect this dierentiation, scenarios can be

eectively used for their intended purpose.

When to Avoid Scenarios

In certain cases, uncertainty surrounding a risk

may be so great that a scenario to explore it

cannot provide any reliable insights. Scenarios

may be harmful in generating a groupthink of

their own, whereby they inform the belief that

only a dened set of outcomes are possible.

Practitioners should be wary of thinking that they

have all reasonable scenarios when there are quite

dierent possibilities that remain unknown.

92. (Tversky and Kahneman 1974) 93. (Tversky and Kahneman 1974)

94. (North American CRO Council 2013) 95. (Roxburgh 2009) 96. (Roxburgh 2009)

Cambridge Centre for Risk Studies

Scenario Best Practices: Developing Scenarios for the Insurance Industry

5. Scenario Development

Case Study: Business Blackout

This section presents an example of a s

cenario’s development process, using

the method proposed in Section 4 of this report. The example scenario is

based upon the 2015 Lloyd’s Business Blackout scenario

, developed to

address emerging cyber risk within the insurance industry. The case study

illustrates a practical example of each of the eight s

cenario development steps.

Step 1: Scope the Risk

The purpose of the Business Blackout

scenario was to develop a scenario that would

explore the potential for losses to occur in many dierent

lines of insurance as a result of a cyber attack against

the US power grid. Of concern was the potential for

insured parties to experience losses that are indirectly

covered within insurance policy terms and conditions,

or are ambiguous, not explicitly excluded in policy

wordings, known as ‘silent’ cyber coverage. The scenario

sought to identify how much silent cyber coverage might

be triggered under certain conditions and to explore

whether there might be a need to take more proactive

actions to identify and clarify this coverage. In this

case, the risk – a cyber attack – was clearly dened and

explicitly motivated the exercise, with the objective to

explore the uncertainty around the impact of a severe

event. A specialist working group of researchers and

insurance specialists drew up a long list of potential

candidate scenarios and prioritised a scenario where

there was a cyber attack that would cause a widespread

power outage in the electrical grid in the United States.

To assist with the scenario planning, the development

team developed a problem statement. What are the

potential insurance impacts of an extreme power loss

caused by a cyber attack on the US electrical power grid?

Step 2: Conduct Background Research

Several dierent topics were researched

to better understand the risk landscape.

Dominant questions for discussion included:

What insurance policies might pay out in the event

of a lengthy power outage, and how would it be

dierent if the proximate cause were to be a malicious

cyber attack?

How much of a threat does cyber pose to critical

national infrastructure, and what evidence is there for

cyber hackers targeting the energy sector?

□ Is it possible for a cyber attack to cause a power

outage, and how severe could it potentially be?

□ What sort of vulnerabilities could be exploited,

and what are the implications for the geographical

extent and duration of outages that would occur?

How often and extensive have outages been to the US

electrical grid from non-cyber causes?

□ How did it occur and what was the impact?

□ Have there been “near misses”

from cyber attacks

which have occurred before?

Where are the system vulnerabilities within an

electrical grid?

□ How does the electricity grid work, what is its

structure and generating capacity?

□ What cyber security measures are in place to

protect the generators, transmission systems, and

distribution networks for electricity grids?

□ What are future developments for the electricity

grid (i.e. Smart Grids)

What are the planned responses to an electrical

grid outage?

□ Are components of the response susceptible

to errors?

97. (Cambridge Centre for Risk Studies 2015)

98. Near misses refers to an event which had the potential for harm but did not materialise.

Cambridge Centre for Risk Studies

What type of cyber insurance (both silent and

armative) would be impacted by a cyber

physical scenario?

□ How big would this be?

When approaching these questions, the team consulted

various experts who were familiar with the topic.

These experts were found through research networks,

recommendations, and introductions, and included

members from the insurance industry, the government,

the energy industry, and the cyber security community.

Each expert brought unique insight to the scenario, and

were aware of dierent vulnerabilities, histories, and

responses. In addition to the subject matter experts, a

wide number of resources were also utilised, including

government records, media, and reference databases.

Research and background consultation was not restricted

to this period of the exercise, but continued throughout

the scenario’s development.

Step 3: Frame the Scenario

In the initial stages of development, the

development team participated in a thought

exercise to decide what scenario methodology to use. The

team discussed the framing questions and agreed in the

scenario’s direction. These framing questions entailed:

What is the scenario for?

The scenario is being developed to understand an

emerging risk, cyber, which remains a novel and

uncertain threat for which there is only a short history

of historical precedents.

How can the scenario benet stakeholders?

The scenario is intending to address uncertainty and

expand the insurance industry’s understanding of

cyber risk and insurance exposures.

On what timescale does the risk materialise?

The scenario is intended to be a sudden-onset, shock

risk scenario.

What is the scenario outcome?

The scenario intends to ask ‘what if’, and is primarily

an exploratory scenario, imagining a future event for

which there is no close precedent.

Who owns and contributes to the

scenario process?

The scenario is intended to be an expert-driven,

top-down exercise, and delivered via a report,

hosted in the public domain, for repeated use by

insurance practitioners.

Is the scenario required to dene the

likelihood of an outcome?

The scenario involves a high degree of uncertainty and

is speculative in nature and is therefore deterministic.

It includes three scenario variants of varying severity

and probability, expressed as annual probabilities (for

example, 1 in 50 years).

Step 4: Develop Candidate Scenarios

During the research process, the development

team compiled potential candidate scenarios

for processes that might trigger an electrical grid failure.

These candidate scenarios were based on discussions, near

misses, historical events, and key system vulnerabilities,

and initially these were very loosely dened. Several

emerged from the list of possible scenarios as viable

options. These received further investigation to consider

their plausibility and were then reviewed by panels of

cyber security experts to assess their credibility. This led

to the development of three candidate scenarios.

Hackers manipulate electricity generators’ control

networks, causing the generators to malfunction. The

generators become damaged, and lead to a power outage.

Hackers interfere with the energy market and falsify

a series of purchases. This leads to an oversupply

combined with a decrease in demand in the market,

triggering bankruptcy and market failure.

Specic brands of phasor measurement units are

manipulated with known software vulnerabilities,

causing randomised damage across generators. The

response to these attacks, including investigating their

cause and replacing the generators, lasts for years.

The three scenarios were reviewed at a workshop

attended by external stakeholders and subject matter

experts and received feedback and recommendations

on each scenarios’ feasibility, severity, and interest.

Internally, the development team considered which

scenario best answered the question proposed and was

best positioned to address the uncertainty of a physical

cyber attack. Scenario 1 was considered the best suited

scenario and was further dened during the workshop.

A Malware infects generator control rooms and allows

hackers to disable safety systems which would usually

protect the generators from desynchronisation events.

Generators are manipulated to overheat and catch re.

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Step 5: Develop Narrative

Upon deciding the basis for the event, the

surrounding narrative and details began

to develop. The development team conducted further

research to identify the location for such an event to

occur and cause maximum impact, and how long the

crisis might last before emergency provisions became

available. A list of potential impacts was also developed,

largely based upon smaller scale precedents. A narrative

was created for the scenario to provide the structure for

the loss modelling and to communicate the outcomes.

The following is a summary of the scenario’s narrative:

“A piece of malware (the ‘Erebos’ trojan) infects

electricity generation control rooms in parts of the

North-eastern United States. The malware goes

undetected until it is triggered on a particular day when

it releases its payload which tries to take control of

generators with specic vulnerabilities. In this scenario

it nds 50 generators that it can control, and forces

them to overload and burn out, in some cases causing

additional res and explosions. This temporarily

destabilises the North-eastern United States regional

grid and causes some sustained outages. While power

is restored to some areas within 24 hours, other parts

of the region remain without electricity for a number of

weeks. Economic impacts include direct damage to assets

and infrastructure, decline in sales revenue to electricity

supply companies, loss of sales revenue to business and

disruption to the supply chain”.

To compliment the narrative and assist in the modelling

process, a series of coverage trigger pathways were

also proposed. These statements aided in determining

what the asset was, and what line of business was being

triggered by the impact.

For example, when discussing damage to generators, the

coverage trigger pathway stated:

Commercial generators are damaged by the electrical

re, and are non-repairable due to the extent

of damage.

Step 6: Assess Impacts and Materiality

With the narrative in place, the next step was

calculating the event consequences. Within

this scenario, the development team decided to calculate

ground up loss. A portfolio loss guide was also written

for how insurers could use the scenario to estimate

the exposure of, and associated losses to, their own

portfolios. Losses were calculated individually for each

asset type.

An example of the loss calculation processes used for

perishables is described below.

Determine where the impacts will occur: The

power outages occur in 15 North Eastern U.S. states, in

regions dependent on compromised generators.

Identify what assets will be impacted by

the event: Perishables, which require refrigeration,

are routinely found in culinary, hospitality, grocery,

and medical properties. Marine cargo and chemical

facilities also routinely store perishable products.

Identify the incident rate: Of all potential

properties, assume a fraction do not carry any

perishables. Assume an additional fraction have an

emergency generator or alternative power source. Of

the remaining properties, we assume that they do not

all have the same length of outage, with some lasting

days, while others’ weeks. A severity gradient was

applied to account for this variation of damage.

Identify existing insurance coverages available

to asset: The insurance class for the perishable assets

depend on the sector from which they arise, either

marine insurance or commercial property insurance.

The coverage is assumed to be property damage in

both cases.

Calculate the loss based on asset value: The

loss values were calculated based upon the number

of damaged perishables, the ratio of damage, and

the value of the perishable. The sum of these values

($21.4bn) generated a rst order estimate of the loss.

During the loss calculation process, interviews were

conducted on an ongoing basis. This assisted in

developing parameters upon which to base estimates.

99. (Cambridge Centre for Risk Studies 2015)

Cambridge Centre for Risk Studies

Step 7: Communicate and Act

Once losses were calculated, the complete

scenario and results, were reviewed by the

initial consultation group. This aided in fact checking the

proposed events and outcomes and adding additional

recommendations that may initially have not been

included. The industry loss estimation methods were

reviewed with a number of insurers before the scenario

was nalised. Examples of questions posed during the

iterative process include:

Is the outage length plausible from a cyber attack?

Are there any missing logical steps in the narrative?

Are the losses in the range which you would expect?

Are parameters used in the modelling consistent with

your experience with losses in the area?

Through the publication of the report, the ndings were

communicated to community stakeholders. A public

launch was held to mark the release of the report and

disseminate it to a wider audience.

Step 8: Evaluate and Update

The scenario has been widely applied

within the insurance industry by a variety of

practitioners and was adapted for further use in cyber

models. The scenario remains accessible and relevant

years after development, although users of this scenario

have called for an update to maintain its validity in the

changing cyber and insurance markets.

The physical property insurance market has grown

and changed in terms of how it covers cyber risk. The

scenario will be updated as part as a periodic review

process, to ensure it remains useful.

Scenario Best Practices: Developing Scenarios for the Insurance Industry

6. Future Perspectives on Scenarios

Scenario development is a constantly developing eld of research. Scenario methodologies and

applications have advanced substantially in the past decades, with further changes and growth to come.

Two of the greatest changes anticipated within the insurance modelling community are the increased

availability of data, and the potential for customised modelling supported by machine learning.

Traditionally, insurance scenario modelling has been

dependent on the availability of structured internal data,

produced on a pre-determined frequency (monthly,

annually, quarterly etc.),

100

and desktop research

conducted by underwriters and sta. Alternatively,

external catastrophe modelling approaches aim to

understand potential future losses due to catastrophic,

tail risk events, with synthetic loss events in the future.

The ‘Fourth Industrial Revolution’ has spawned a new

generation of data collection tools that presents an

opportunity for the insurance industry’s use. Advanced

data mining techniques have generated a substantial

increase in the availability and frequency of large and

unstructured data sets that can be used for real-time

scenario analytics. Much of the data is third-party

mined from digital platforms such as social media and is

therefore used mainly as a rich source of data for insights

into consumer behaviour.

101

With an estimated 2.5 quintillion bytes of data being

created daily, the insurance industry is increasingly

using big data and machine learning techniques

to analyse large and unstructured data sets which

may aid the scenario development process.

102

Open-

sourced data management tools such as Apache and

Hadoop are increasingly being used by insurers to

disseminate structured and unstructured data sets in

short time horizons. This has aided in the development

of real-time analytics in the insurance industry.

103

Real-time behavioural data opens new opportunities

in understanding behavioural risks that have been

previously dicult to forecast.

Consumer specic data also heralds the possibility for

personalised scenario modelling, based on real-time

monitoring and visualisation of consumer behaviour.

Examples of customised data collection have been

increasingly seen in auto and health insurances, which

have been introducing personal monitoring devices to

provide more accurate premiums to consumers.

104

The insurance industry is lagging behind many other

sectors in investment into AI and machine learning,

with only 1.33% of insurance companies investing in AI

in 2017 compared with 32% of internet and software

companies.

105

However, a survey by Deloitte in 2017

estimated that 95% of insurance executives intend to

start or continue to invest in AI capabilities. Indeed,

investment in AI in the insurance industry increased by

68% between 2011 and 2014.

106

As investment in machine

learning and AI tools continues to grow, scenario

development will increasingly involve these tools to

create more complex and comprehensive scenarios.

100. (Clarke and Libarikian 2014) 101. (Clarke and Libarikian 2014; Allen 2018)

102. (Schroeck and Shockely 2012) 103. (Schroeck and Shockely 2012)

104. (Insurance Business UK 2018) 105. (Deloitte Digital 2017) 106. (Deloitte Digital 2017)

Cambridge Centre for Risk Studies

7. Glossary

Accumulation: The combined total risks triggered by a

single hazard event.

Backcasting: A scenario technique which identies

a future state and then works backwards to determine

possible paths to this future from the present state.

Baseline Scenario: An assumed future with no explicit

deviations, used to compare possible futures against.

Cascading Events: An event which triggers an event of

another type, to cause a more extreme event.

Counterfactual Scenarios: An alternative outcome of

a historical event, had a specic hypothetical intervention

occurred, or a real intervention had not.

Decision Making Under Deep Uncertainty: An

emerging paradigm comprising a variety of approaches

to cope with uncertainty with robust decisions.

Deep Uncertainty: Exists when various parties to

a decision do not know, or cannot agree on, how a

complex system works, the probability of possible future

states, and how important the various consequences of

interest are.

107

Deterministic Approach: An examination

of the impacts of a single event, dened by the

scenario developer.

Due Diligence: Care which is taken to avoid harm or

damage to other persons or property.

Emerging Risk: An emerging risk is a new risk,

changing risk, or novel combination of risks for which

the broad impacts, costs and optimal management

strategies are not yet well understood.

Forecast: The most probable prediction that something

will happen in the future.

Hazard: A natural or anthropogenic process or

phenomenon that may result in negative social,

economic, or environmental consequences. A hazard

event is the manifestation of a hazard in a particular time

and place.

108

Insurance: A contract where an entity agrees to pay an

agreed amount in the case of a specic event occurring.

Insurance is a form of risk transfer.

Narrative: A descriptive summary of the events

occurring within a scenario. Also known as a storyline.

Prediction: A subjective (probabilistic) statement that

something will happen in the future.

109

Probabilistic Approach: An examination

of all potential outcomes, and their estimated

likelihood (probability).

Projection: A probabilistic statement that something

will happen under certain conditions, allowing for

signicant changes in the boundary conditions that

might inuence a prediction.

110

Resource Allowance: The amount of resources that

can be spent on a specic task. Examples of resources

can include time, money or persons.

Risk: A hazard event with the potential to cause

economic and/or societal losses, determined

probabilistically as a function of hazard, exposure

and vulnerability.

Scenario: Plausible descriptions of how the future

might develop, based on a coherent and internally

consistent set of assumptions about key deterministic

relationships and driving forces.

111

Shock Risk: Extreme, sudden-onset hazard events that

emerge quickly or unexpectedly.

Stress Test: A scenario designed to assess the

idiosyncratic vulnerabilities of an organisation to a

hazard event.

Tail Risk: Rare events that occur more than three

standard deviations from the expected, most probable

outcome. The term can be used more generally to

describe highly unlikely extreme outcomes.

Trend Risk: Slow-onset hazard phenomena that

emerge gradually over time.

Uncertainty: A potential outcome with an unknown

probability, which is therefore uncontrollable.

Vulnerability: “The conditions determined by physical,

social, economic and environmental factors or processes

which increase the susceptibility of an individual,

a community, assets or systems to the impacts of

hazards”.

112

107. (Kwakkel 2017) 108. (UNISDR 2017) 109. (MacCracken 2001)

110. (MacCracken 2001) 111. (Van Vuuren et al. 2012) 112. (UNISDR 2017)

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Cambridge Centre for Risk Studies

8. References

Alford, Carlos L. 2017. ‘The Department of Defense

Eort to Counterting the Cyberterrorism Threat:

Is the Threat Real or Hyperbole?’ National Defense

University: Joint Forces Sta College. https://www.hsdl.

org/?abstract&did=805424.

Allen, Marshall. 2018. ‘Health Insurers Are Vacuuming Up

Details About You — And It Could Raise Your Rates’.

ProPublica. 17 July 2018. https://www.propublica.org/

article/health-insurers-are-vacuuming-up-details-about-

you-and-it-could-raise-your-rates.

Amer, Muhammad, Tugrul U. Daim, and Antonie Jetter. 2013.

‘A Review of Scenario Planning’. Futures 46 (February):

23–40. https://doi.org/10.1016/j.futures.2012.10.003.

Beven, Keith J., and Ruth E. Alcock. 2012. ‘Modelling

Everything Everywhere: A New Approach to Decision-

Making for Water Management under Uncertainty’.

Freshwater Biology 57 (s1): 124–32. https://doi.

org/10.1111/j.1365-2427.2011.02592.x.

Birkmann, Joern, Susan L. Cutter, Dale S. Rothman,

Torsten Welle, Matthias Garschagen, Bas van Ruijven,

Brian O’Neill, et al. 2015. ‘Scenarios for Vulnerability:

Opportunities and Constraints in the Context of Climate

Change and Disaster Risk’. Climatic Change 133 (1): 53–68.

https://doi.org/10.1007/s10584-013-0913-2.

Boonen, Tim J. 2017. ‘Solvency II Solvency Capital

Requirement for Life Insurance Companies Based on

Expected Shortfall’. European Actuarial Journal 7 (2):

405–34. https://doi.org/10.1007/s13385-017-0160-4.

Broadhurst, Roderic, Hannah Woodford-Smith, Donald

Maxim, Bianca Sabol, Stephanie Orlando, Ben Chapman-

Schmidt, and Mamoun Alazab. 2017. ‘Cyber Terrorism:

Research Review: Research Report of the Australian

National University Cybercrime Observatory for the

Korean Institute of Criminology’. SSRN Scholarly Paper ID

2984101. Rochester, NY: Social Science Research Network.

https://papers.ssrn.com/abstract=2984101.

Cambridge Centre for Risk Studies. 2015. ‘Lloyd’s Business

Blackout Scenario’. Emerging Risk Report. Cambridge

Centre for Risk Studies at the University of Cambridge

Judge Business School. https://www.jbs.cam.ac.uk/

faculty-research/centres/risk/publications/technology-

and-space/lloyds-business-blackout-scenario/.

Cambridge Centre for Risk Studies, in collaboration with

Lighthill Risk Network, 2019. Scenario Best Practices:

Developing Scenarios for Disaster Risk Reduction.

Cambridge Centre for Risk Studies at the University of

Cambridge Judge Business School.

Cambridge Centre for Risk Studies, in collaboration with Risk

Management Solutions, Inc. 2018. ‘Multi-Line Insurance

Exposure Management Data Denitions Document v1.0’.

Cambridge Centre for Risk Studies at the University of

Cambridge Judge Business School. https://www.jbs.cam.

ac.uk/leadmin/user_upload/research/centres/risk/

downloads/crs-geac-data-denitions-document.pdf.

Clarke, Richard, and Ari Libarikian. 2014. ‘Unleashing the

Value of Advanced Analytics in Insurance’. McKinsey

& Company. https://www.mckinsey.com/industries/

nancial-services/our-insights/unleashing-the-value-of-

advanced-analytics-in-insurance.

Cummins, J. David. 2000. ‘Allocation of Capital

in the Insurance Industry’. Risk Management

and Insurance Review 3 (1): 7–27. https://doi.

org/10.1111/j.1540-6296.2000.tb00013.x.

Davidson, Paul. 1991. ‘Is Probability Theory Relevant for

Uncertainty? A Post Keynesian Perspective’. Journal

of Economic Perspectives 5 (1): 129–43. https://doi.

org/10.1257/jep.5.1.129.

Deloitte Digital. 2017. ‘From Mystery to Mastery: Unlocking

the Business Value of Artical Intelligence in the Insurance

Industry’. Deloitte Digital. https://www2.deloitte.com/

content/dam/Deloitte/xe/Documents/nancial-services/

Articial-Intelligence-in-Insurance.pdf.

Derbyshire, James. 2017a. ‘The Siren Call of Probability:

Dangers Associated with Using Probability for

Consideration of the Future’. Futures 88 (April): 43–54.

https://doi.org/10.1016/j.futures.2017.03.011.

———. 2017b. ‘Potential Surprise Theory as a Theoretical

Foundation for Scenario Planning’. Technological

Forecasting and Social Change 124 (November): 77–87.

https://doi.org/10.1016/j.techfore.2016.05.008.

Dreborg, Karl H. 1996. ‘Essence of Backcasting’.

Futures 28 (9): 813–28. https://doi.org/10.1016/S0016-

3287(96)00044-4.

Durance, Philippe, and Michel Godet. 2010. ‘Scenario Building:

Uses and Abuses’. Technological Forecasting and Social

Change, Strategic Foresight, 77 (9): 1488–92. https://doi.

org/10.1016/j.techfore.2010.06.007.

Evan, Tamara, Eireann Leverett, Simon Rue, Andrew

Coburn, James Bourdeau, Rohan Gunaratna, and Daniel

Ralph. 2017. ‘Cyber Terrorism: Assessment of the Threat to

Insurance’. Cambridge Risk Framework Series. University

of Cambridge: Cambridge Centre for Risk Studies. https://

www.jbs.cam.ac.uk/leadmin/user_upload/research/

centres/risk/downloads/pool-re-cyber-terrorism.pdf.

Financial Services Commission. 2009. ‘Basel II: Pillar 2 - The

ICAAP & The SREP’. Guidance Note. Gibraltar: Financial

Services Commission. http://www.fsc.gi/uploads/legacy/

download/adobe/banking/noteiccaapsrep.pdf.

GIRO Working Party. 2016. ‘Integrating Pricing & Capital

Modelling’. https://www.actuaries.org.uk/system/les/

documents/pdf/Crozet.pdf.

Grant, Matthew. 2015. ‘What Can the Insurance Market Teach

Banks About Stress Tests? | The RMS Blog’.

The RMS Blog (blog). 2015. https://www.rms.com/

blog/2015/08/03/what-can-the-insurance-market-teach-

banks-about-stress-tests/.

Scenario Best Practices: Developing Scenarios for the Insurance Industry

Haasnoot, Marjolijn, Jan H. Kwakkel, Warren E. Walker, and

Judith ter Maat. 2013. ‘Dynamic Adaptive Policy Pathways:

A Method for Crafting Robust Decisions for a Deeply

Uncertain World’. Global Environmental Change 23 (2):

485–98. https://doi.org/10.1016/j.gloenvcha.2012.12.006.

Hallegatte, Stéphane, Ankur Shah, Robert Lempert, Casey

Brown, and Stuart Gill. 2012. ‘Investment Decision

Making under Deep Uncertainty’. World Bank Group,

Policy Research Working Papers, 6193. https://doi.

org/10.1596/1813-9450-6193.

Henrichs, Thomas, Monika Zurek, Bas Eickhout, Kasper

Kok, Ciara Raudsepp-Hearne, Teresa Ribeiro, Detlef van

Vuuren, and Axel Volkery. 2010. ‘Scenario Development

and Analysis for Forward-Looking Ecosystem

Assessments’. Ecosystems and Human Well-Being: A

Manual for Assessment Practitioners, 151–219.

House of Commons Treasury Committee. 2018. ‘The Solvency

II Directive and Its Impact on the UK Insurance Industry:

Bank of England Response to the Committee’s Third

Report of Session 2017-2019’. Second Special Report of

Session 2017-19. House of Commons.

https://publications.parliament.uk/pa/cm201719/

cmselect/cmtreasy/863/863.pdf.

Insurance Business UK. 2018. ‘How Telematics Is Transforming

the Car Insurance Landscape’. Insurance Business UK.

17 December 2018. https://www.insurancebusinessmag.

com/uk/opinion/how-telematics-is-transforming-the-car-

insurance-landscape-119099.aspx.

Insurance Journal. 2018. ‘Pool Re Extends Retrocession

Program to $2.9B with Cyber Terrorism Now

Covered’. Insurance Journal. 1 March 2018.

https://www.insurancejournal.com/news/

international/2018/03/01/482001.htm.

Jeerson, Michael. 2012. ‘Shell Scenarios: What Really

Happened in the 1970s and What May Be Learned for

Current World Prospects’. Technological Forecasting and

Social Change 79 (1): 186–97. https://doi.org/10.1016/j.

techfore.2011.08.007.

Kahn, Herman. 1961. ‘On Thermonuclear War’. Princeton:

Princeton University Press.

Kahneman, Daniel. 2011. Thinking, Fast and Slow. Macmillan.

Kauman, Draper L. 1980. Systems One: An Introduction to

Systems Thinking. Pegasus Communications.

Kim, Seong Dae. 2012. ‘Characterizing Unknown Unknowns’.

In . North America, Vancouver, British Columbia, Canada.

Newtown Square, PA: Project Management Institute.

https://www.pmi.org/learning/library/characterizing-

unknown-unknowns-6077.

Krvavych, Yuriy. 2018. ‘Making Use of Internal Capital Models’.

presented at the ASTIN Colloquium, The Hague, May 21.

https://web.actuaries.ie/sites/default/les/erm-resources/

making_use_of_internal_capital_models.pdf.

Kwakkel, Jan H. 2017. ‘Managing Deep Uncertainty:

Exploratory Modeling, Adaptive Plans and Joint Sense

Making’. Integration and Implementation Insights (blog).

14 August 2017. https://i2insights.org/2017/08/15/

managing-deep-uncertainty/.

Kwakkel, Jan H., Warren E. Walker, and Marjolijn Haasnoot.

2016. ‘Coping with the Wickedness of Public Policy

Problems: Approaches for Decision Making under Deep

Uncertainty’. Journal of Water Resources Planning and

Management 142 (3): 01816001. https://doi.org/10.1061/

(ASCE)WR.1943-5452.0000626.

Lempert, Robert J., and Nidhi Kalra. 2011. ‘Managing Climate

Risks in Developing Countries with Robust Decision

Making’. In .

Lempert, Robert J., Steven W. Popper, and Steven C. Bankes.

2003. Shaping the Next One Hundred Years: New Methods

for Quantitative, Long-Term Policy Analysis. https://www.

rand.org/pubs/monograph_reports/MR1626.html.

Lempert, Robert J., Steven W. Popper, David G. Groves, Nidhi

Kalra, Jordan R. Fischbach, Steven C. Bankes, Benjamin

P. Bryant, et al. 2013. ‘Making Good Decisions Without

Predictions: Robust Decision Making for Planning Under

Deep Uncertainty’. Product Page. 2013. https://www.rand.

org/pubs/research_briefs/RB9701.html.

Lloyd’s of London. n.d. ‘What Is Solvency II?’ Lloyd’s of

London. Accessed 12 December 2018. https://www.lloyds.

com/market-resources/regulatory/solvency-ii/about/

what-is-solvency-ii.

Lorenz, Edward N. 1963. ‘Deterministic Nonperiodic Flow’.

Journal of the Atmospheric Sciences 20 (2): 130–41.

https://doi.org/10.1175/1520-0469(1963)020<0130:DNF

>2.0.CO;2.

MacCracken, Mike. 2001. ‘Prediction versus Projection -

Forecast versus Possibility’. WeatherZine, 26 February

2001, 26 edition. https://sciencepolicy.colorado.edu/zine/

archives/1-29/26/guest.html.

Mietzner, Dana, and Guido Reger. 2005. ‘Advantages and

Disadvantages of Scenario Approaches for Strategic

Foresight’. International Journal of Technology

Intelligence and Planning 1 (2): 220. https://doi.

org/10.1504/IJTIP.2005.006516.

North American CRO Council. 2013. ‘Scenario Analysis

Principles and Practices in the Insurance Industry’. North

American CRO Council. http://crocouncil.org/images/

CRO_Council_-_Stress_and_Scenario_Testing_Paper_

FINAL.pdf.

Olabisi, Laura Schmitt. 2017. ‘Dealing with Deep Uncertainty:

Scenarios’. Integration and Implementation Insights

(blog). 5 January 2017. https://i2insights.org/2017/01/05/

deep-uncertainty-and-scenarios/.

Popper, Steven W., James Grin, Claude Berrebi, Thomas

Light, and Endy M. Daehner. 2009. ‘Natural Gas and

Israel’s Energy Future’. Product Page. 2009. https://www.

rand.org/pubs/technical_reports/TR747.html.

Renard, Philippe, Andres Alcolea, and David Ginsbourger.

2013. ‘Stochastic versus Deterministic Approaches’. In

Environmental Modelling, 133–49. John Wiley & Sons,

Ltd. https://doi.org/10.1002/9781118351475.ch8.

Riddell, Graeme A., Hedwig van Delden, Graeme C. Dandy,

Aaron C. Zecchin, and Holger R. Maier. 2018. ‘Enhancing

the Policy Relevance of Exploratory Scenarios: Generic

Approach and Application to Disaster Risk Reduction’.

Futures 99 (May): 1–15. https://doi.org/10.1016/j.

futures.2018.03.006.

Cambridge Centre for Risk Studies

Risk Management Solutions Inc, in collaboration with

Cambridge Centre for Risk Studies. 2016. ‘Managing

Cyber Insurance Accumulation Risk’. Risk Management

Solutions Inc, in collaboration with Cambridge Centre

for Risk Studies at the University of Cambridge Judge

Business School. https://www.jbs.cam.ac.uk/leadmin/

user_upload/research/centres/risk/downloads/crs-rms-

managing-cyber-insurance-accumulation-risk.pdf.

Robinson, John B. 1990. ‘Futures under Glass: A Recipe for

People Who Hate to Predict’. Futures 22 (8): 820–42.

https://doi.org/10.1016/0016-3287(90)90018-D.

Roxburgh, Charles. 2009. ‘The Use and Abuse of Scenarios’.

McKinsey & Company. November 2009. https://www.

mckinsey.com/business-functions/strategy-and-corporate-

nance/our-insights/the-use-and-abuse-of-scenarios.

Rozenberg, Julie, Cecilia M. Briceno-Garmendia, Xijie

Lu, Laura Bonzanigo, and Harry Edmund Moroz.

2017. ‘Improving the Resilience of Peru’s Road

Network to Climate Events’. WPS8013. The World

Bank. http://documents.worldbank.org/curated/

en/691821490628878185/Improving-the-resilience-of-

Perus-road-network-to-climate-events.

Rue, Simon, Jennifer Daron, Jennifer Copic, Éireann

Leverett, and Tamara Evan. 2018. ‘Submission to

UK Parliamentary Joint Committee on the National

Security Strategy: Cyber Security: Critical National

Infrastructure Inquiry’. UK Parliamentary Joint

Committee on the National Security Strategy. http://

data.parliament.uk/writtenevidence/committeeevidence.

svc/evidencedocument/national-security-strategy-

committee/cyber-security-critical-national-infrastructure/

written/81366.html.

Rumsfeld, Donald. 2002. ‘Press Conference US SoD’. NATO

HQ, Brussels, June 6. https://www.nato.int/docu/

speech/2002/s020606g.htm.

Sayers, Paul B., Gerald E. Galloway, and Jim W. Hall.

2012. ‘Robust Decision-Making under Uncertainty ?

Towards Adaptive and Resilient Flood Risk Management

Infrastructure’. In Flood Risk, 281–302. ICE Publishing.

https://doi.org/10.1680/fr.41561.281.

Schroeck, Michael, and Rebecca Shockely. 2012. ‘Analytics:

Real-World Use of Big Data in Insurance’. IBM. IBM

Institute for Business Value (blog). 2012. https://www.

ibm.com/services/us/gbs/thoughtleadership/big-data-

insurance/.

Schwartz, Peter. 1997. Art of the Long View: Planning for the

Future in an Uncertain World. New Ed edition. Chichester:

John Wiley & Sons.

Serinaldi, Francesco. 2015. ‘Dismissing Return Periods!’

Stochastic Environmental Research and Risk Assessment

29 (4): 1179–89. https://doi.org/10.1007/s00477-014-

0916-1.

Social Finance Inc. 2012. ‘Social Impact Bonds: An Overview’.

Social Finance. Boston, MA: The Rockefeller Foundation.

https://www.socialnance.org.uk/sites/default/les/

publications/small.socialnancewpsinglenal.pdf.

Star, Jonathan, Erika L. Rowland, Mary E. Black, Carolyn A. F.

Enquist, Gregg Garn, Catherine Hawkins Homan, Holly

Hartmann, Katharine L. Jacobs, Richard H. Moss, and

Anne M. Waple. 2016. ‘Supporting Adaptation Decisions

through Scenario Planning: Enabling the Eective

Use of Multiple Methods’. Climate Risk Management

13 (January): 88–94. https://doi.org/10.1016/j.

crm.2016.08.001.

The Economist. 2017. ‘A Lloyd’s Report Urges Insurers to Ask

“What If?”’ The Economist, 19 October 2017. https://www.

economist.com/nance-and-economics/2017/10/19/a-

lloyds-report-urges-insurers-to-ask-what-if.

Tourki, Yousra, Jerey Keisler, and Igor Linkov. 2013.

‘Scenario Analysis: A Review of Methods and Applications

for Engineering and Environmental Systems’.

Environment Systems & Decisions 33 (1): 3–20. https://

doi.org/10.1007/s10669-013-9437-6.

Tuck, Laura. 2016. ‘Embracing Uncertainty for Better

Decision-Making’. Text. Ppps. 3 October 2016. http://

blogs.worldbank.org/ppps/embracing-uncertainty-better-

decision-making.

Tversky, Amos, and Daniel Kahneman. 1974. ‘Judgment under

Uncertainty: Heuristics and Biases’. Science 185 (4157):

1124–31. https://doi.org/10.1126/science.185.4157.1124.

UNISDR. 2017. ‘Terminology on Disaster Risk Reduction’.

UN Oce for Disaster Risk Reduction. 2 February 2017.

https://www.unisdr.org/we/inform/terminology.

Valenzuela, Ana, and Priya Raghubir. 2009a. ‘Position-Based

Beliefs: The Center-Stage Eect’. Journal of Consumer

Psychology 19 (2): 185–96. https://doi.org/10.1016/j.

jcps.2009.02.011.

Van der Helm, Ruud. 2006. ‘Towards a Clarication of

Probability, Possibility and Plausibility: How Semantics

Could Help Futures Practice to Improve’. Foresight 8 (3):

17–27. https://doi.org/10.1108/14636680610668045.

Van der Merwe, Louis. 2008. ‘Scenario-Based Strategy

in Practice: A Framework’. Advances in Developing

Human Resources 10 (2): 216–39. https://doi.

org/10.1177/1523422307313321.

Van Notten, Philip W. F, Jan Rotmans, Marjolein B. A

van Asselt, and Dale S Rothman. 2003. ‘An Updated

Scenario Typology’. Futures 35 (5): 423–43. https://doi.

org/10.1016/S0016-3287(02)00090-3.

Van Vuuren, Detlef P., Marcel TJ Kok, Bastien Girod, Paul

L. Lucas, and Bert de Vries. 2012. ‘Scenarios in Global

Environmental Assessments: Key Characteristics and

Lessons for Future Use’. Global Environmental Change 22

(4): 884–95.

Vandenabeele, Tim. 2014. ‘Solvency II in a Nutshell’. Milliman.

http://nl.milliman.com/uploadedFiles/insight/2014/

solvency-ii-nutshell.pdf.

Zastrow, Mark. 2019. ‘Fukushima Bosses Cleared over Nuclear

Disaster’. Nature. https://doi.org/10.1038/d41586-019-

02822-y.

Cambridge Centre for Risk Studies

Cambridge Judge Business School

University of Cambridge

Trumpington Street

Cambridge

CB2 1AG

T: +44 (0) 1223 768386

F: +44 (0) 1223 339701

[email protected].uk

www.jbs.cam.ac.uk/risk

Join our LinkedIn group at

Cambridge Centre for Risk Studies