04 Lossmodel

Climate Change Uncertainty and Risk: from
Probabilistic Forecasts to Economics of Climate

Adaptation
Reto Knutti, IAC ETH
David N. Bresch, Swiss Re/ETH
Assistants: Maria Rugenstein, Martin Stolpe, Anina Gilgen
Reto Knutti / ETH Zrich | David Bresch / Swiss Re
Schedule
29.02.16 (1) Logistics, Introduction to probability, uncertainty and risk management,
introduction of toy model (RK, DB)
07.03.16 (2) Predictability of weather and climate, seasonal prediction, seamless
prediction (RK)
Exercise 1 (toy model)
14.03.16 (3) Detection/attribution, forced changes, natural variability, signal/noise,
ensembles (RK)
Exercise 2 (toy model)
21.03.16 (4) Probabilistic risk assessment model: from concept to concrete
application - and some insurance basics (DB)
28.03.16 Ostermontag (no course)
04.04.16 (5) Model evaluation, multi model ensembles and structural error (RK)
11.04.16 (6) Climate change and impacts, scenarios, use of scenarios, scenario
uncertainty vs response/impact uncertainty (RK, DB)
Exercise 3 (toy model), preparation of presentation
18.04.16 (7) Model calibration, Bayesian methods for probabilistic projections (RK)
25.04.16 (8) Presentations of toy model work, discussion (DB, RK)
Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re
Schedule
02.05.16 (9) Basics of economic evaluation and economic decision making in the
presence of climate risk (DB)
Exercise 4 (introduction to climada)
09.05.16 (10) The cost of adaptation - application of economic decision making to
climate adaptation in developing and developed region (DB)
Exercise 5 (impacts)
16.05.16 Pfingstmontag (no course)
23.05.16 (11) Shaping climate-resilient development valuation of a basket of
adaptation options (DB)
Exercises 6 (adaptation measures, preparation of presentation)
30.05.16 (12) Presentations of climada exercise, discussion (DB, RK)
Recap: Risk1 Management

Risk identification: Shared mental model, the prerequisite for awareness
perception is based on a shared mental model

wider sharing builds awareness
Risk analysis: Quantification, the basis for decision-making
Risk model: the quantitative expression of a shared mental model

allows to assess risk mitigation options
Risk mitigation: Prioritization based on metrics, options are to
avoid
prevent
transfer : Insurance puts a rice tag on risks incentive for prevention
or retain the risk
Risk = Probability
Severity
The Risk Management Cycle

Identify
Ana
lyze
Mitigation options:
Avoid
Reduce
Prevent
Transfer
Retain
5
Recap: Note on decision strategies

(a)
(b)
Source: UKCIP
In the face of high levels of uncertainty, which may not be readily resolved
through research, decision makers are best advised to not adopt a decision
strategy in which (a) nothing is done until research resolves all key
uncertainties, but rather (b) to adopt an iterative and adaptive strategy.
Natural Nat Cat damages on the rise

and: Massive gap between economic and insured damage
Natural catastrophe damages 1980-2013, in USD billion
450
Uninsured losses
400
Insured losses
350
10-year moving average insured losses
300
10-year moving average total economic losses
250
200
150
100
50
0
1970
1975
1980
1985
1990
1995
2000
Note: Amounts
Loss amounts
indexed
indexed
to 2012
to 2009
Source: Swiss Re
Source:
sigmaSwiss
catastrophe
Re, sigma
database
No 2/2010
2005
2010
Note on trend drivers
Ocean Drive, FL, 1926
The upward trend in natural catastrophe

damage is driven by:
Higher insurance penetration
Growing property values
Coastal value concentration
Higher vulnerabilities
Ocean Drive, FL, 2000
Climate change
Trend decomposition going forward ?
Economics of Climate Adaptation
Fire versus Nat Cat damage need for simulation
Fire vs. Nat Cat need for accumulation control
every
year
1 of 5000
houses
every
5000 years
Fire Risk
all houses
Nat Cat Risk

0.2 of house value
0.2 of house value

Experience or burning cost approach

Take historic data (often only few years), calculate expected value
Pro: easy to calculate and easy to communicate
Con: only looking backward, (very) incomplete sampling of phase space
Damage
Underwriter
experience
Frequency
Scenario
Definition: A scenario is a snapshot that describes a possible and plausible
future. Scenario analysis is a systematic approach to anticipate a broad
range of plausible future outcomes
Scenario analysis is used in general
as a risk management tool to assess the potential impact of an event or
development to anticipate and understand risks
as a tool to spot new business opportunities and to discover strategic
options
as foresight in contexts of accelerated change, greater complexity and
interdependency
for evaluation of highly uncertain events that could have a major impact
to steer mitigation strategies, implementation and monitoring by reviewing
and tracking different possible developments
Forecast
Scenario
Focuses on certainties,
Focuses on uncertainties,
disguises uncertainties
Conceals risks
Results in a single-point
projections
legitimizes recognition of
uncertainties
Clarifies risk
Results in adaptive
understanding
Sensitivity analysis
Quantitative > qualitative
Diversity of interpretations
Qualitative > quantitative
Forecast
The present The path
Scenarios
The future
Current
realities
(mental maps)
Alternative
future images
Probabilistic modeling
Construct all possible states, requires system understanding
Pro: full sampling of phase space
Con: huge effort
Note: still scenarios, NOT forecasts
Damage
Probabilistic
modeling
Underwriter
experience
Scenario
Considerations
Frequency
Natural Catastrophe Risk Assessment Model
Natural Catastrophe Risk Assessment Model

Hazard
How strong?
How frequent?
Vulnerability
Assets
Damage function
Exposure database
Whats the impact?
Where?
What?
Probabilistic multi-hazard risk assessment
climada
Open-source (GitHub, on MATLAB and Octave), makes use of a wide variety of open-access databases and
provides various interfaces (open architecture)
Widely used, basis for several peer-reviewed publications, past and on-going collaborations with ETH Zrich,
MeteoSwiss, Uni Bern, NCAR, TNC, consultants https://github.com/davidnbresch/climada
climada: High-resolution (1x1km) damage model
Bangladesh
climada
climada: High-resolution (1x1km) damage model
Bangladesh
climada
Tropical cyclones in the Indian ocean
historic
~ 25 years
probabilistic
~ 5000 years
Understanding the hazard how strong
Tropical cyclones in the North Atlantic
historic
~100 years
Tropical cyclones in the North Atlantic
historic
~100 years
probabilistic
~10000 years
Tropical cyclones in the West Pacific
historic
~ 50 years
probabilistic
~ 5000 years
Tropical cyclones in the Southern hemisphere
historic
~ 50 years
probabilistic
~ 5000 years
Note on validation
Both the historical dataset and
especially the probabilistic set
need to be carefully validated,
as an example, we show the
validation of intensity
distributions:
historical catalogue, 34
storms (upper panels)
probabilistic event set, 3391
storms (lower panels)
red line shows an extreme
value distribution (Gumbel)
Tropical cyclone intensity the wind field (1/3)
http://en.wikipedia.org/wiki/Hurricanes

We use the Holland wind field model
The 1-min sustained wind at
gradient wind level (boundary
layer height & no surface
effects) is modelled using the
Holland 2008 approach. It
models the first-order vortex of a
tropical cyclone.
The translational speed (also
called celerity) is added
geometrically.
Holland, G. J., 1980: An analytic model of the wind and pressure

profiles in hurricanes. Monthly Weather Review, 108, 1212-1218.
Vickery, P.J. and D. Wadhera, 2008: Statistical models of Holland

pressure profile parameter and radius to maximum winds of
hurricanes from flight-level pressure and H*wind data. J. Appl.
Meteor. Clim.

Landfall correction (not part of Exercise)
A topography module corrects
the Holland model for large
scale topography, but still
assumes a gradient level wind,
i.e. at boundary layer height.
The approach of Vickery et al.
2009 corrects the wind for the
boundary layer and surface
properties.
Vickery, P.J. et al., 2009: A Hurricane Boundary Layer and Wind

Field Model for Use in Engineering Applications. J. Appl. Meteor.
Clim.
and an update of the Holland approach:

Holland, G. J., 2008: A Revised Hurricane PressureWind Model,
Monthly Weather Review, 136, 3432-3445.
Damage calculation (1/2)

The damage is calculated for each single asset at each location for each
scenario or event, so basically damage = value * damage function (see
below), looped over assets and events
damage = asset value * MDD * PAA
damage function
damage is the damage from ground up, from the first dollar, so to speak
asset value is the total value of the asset
MDD is the Mean Damage Degree (the damage for a given intensity at
an affected asset) - how strongly an asset is damaged. Range 0..1 (from
none to total destruction)
PAA is the Percentage of Assets Affected (the percentage of assets
affected for a given hazard intensity) - how many assets are affected.
Range 0..1 (from none affected to all affected)
Damage calculation (2/2)

So far, the hazard intensity did not show up in the calculation, did we miss
something? Well, the damage is a function of the hazard intensity, hence:
MDD = f(hazard intensity)
PAA = f(hazard intensity)
where hazard intensity is the hazard's intensity at each asset for each
event. Since the damage also depends on the asset type, we have in fact:
MDD = f(hazard intensity, asset type)
PAA = f(hazard intensity, asset type)
Mean Damage Ratio
Notes on damage function
Hazard Intensity
Uncertainty of the
hazard intensity
+
Uncertainty of the
damage
results in
Convoluted Distribution
Insurance - Basics
Insurability
Insurance is the mutual cover of a fortuitous, assessable need
of a large number of similarly exposed business
Alfred Manes, 1877-1963
mutuality: numerous exposed parties must join together to form a risk
community, to share and diversify the risk large number
fortuitous or randomness: time of occurrence must be unpredictable,
occurrence itself must be independent of the will of the insured
assessability: damage probability and severity must be quantifiable
similarly exposed business large number
plus: economic viability: private insurers must be able to obtain a riskadequate premium
Effect of insurance (1/4)

Time series of annual result
1.00
0.50
0.00
-0.50
average result 0.15, stddev 0.60
2030
2025
2020
2015
2010
-1.00

Time series of annual result (after prevention)
price for prevention
1.00
0.50
0.00
-0.50
average result 0.19 (+25%), stddev 0.56 (-8%), prevention price 0.01
effect of prevention: stabilize result, reduce volatility
2030
2025
2020
2015
2010
-1.00

Time series of annual result (after prevention and insurance)
price for prevention

and insurance
1.00
0.50
0.00
-0.50
average result 0.17 (+12%), stddev 0.43 (-29%), prev+ins price: 0.13
effect of insurance: reduce (extreme) volatility
2030
2025
2020
2015
2010
-1.00

raw
result
0.15
stdev
0.60
price*
+ prevention
0.19 (+25%) 0.56 (-8%) 0.01
cost-effective adaptation (net gain of 0.04 at cost of 0.01 )
+ insurance
0.17 (+12%) 0.43 (-29%) 0.01+0.12
substantial reduction of volatility, result increase even after deduction
of prevention cost and insurance premium affordable!
insurance alone
0.12 (-17%) 0.45 (-25%) 0.153
prevention (strongly) incentivizes insurance
*price is already taken into account in result
The main function of risk transfer
Insurance conditions
proportional:
non-proportional:
damageafter=damagebefore*share
damageafter=min(max(damagebefore-deductible,0),cover)
non-proportional
cover
from ground-up damagefgu (total area) to damage

after conditions, conditions get applied in the
following sequence:
proportional
1) coinsurance
2) deductible
3) cover (we show the overspill)
4) share (see also 1-share)
hence we calculate as follows:
damage after conditions=min[
max(damagefgu*{1-coinsurance}deductible,0),cover] *share
Note on insurance conditions the reality
Portfolio of assets
Policy 1
Site 1
...
Policy 2
Site 2
Coverage 1
...
Site m
Coverage 2
...
Policy n
e.g. ZIP code 12345
Coverage p
e.g. Building
Damage accumulation
In reality, risk transfer happens on several (repeated) hierarchical levels,

such as:
Forms of insurance
Risk transfer can be agreed upon based on different triggers:
indemnity1, also called incurred or occurred damage
parametric, also called index
modelled (well, a form of parametric)
and with different partners, such as:
policyholders from macro (e.g. large corporates in Texas) to micro
(e.g. smallholder farmers in Ethiopia Example)
insurers (reinsurers insure them)
other reinsurers, called retrocession
capital market, called insurance-linked security (ILS) or
often also Cat Bond ( Example)
public sector (PPP Example)
1specific
or market-share
Microinsurance case study Ethiopia
http://www.swissre.com/rethinking/crm/The_R4_Rural_Resilience_Initiative.html
Cat
case
study Amrica
North
CatBond
Bond
North
America
http://www.swissre.com/rethinking/crm/experts_on_multicat_mexico.html
Efficient monopolies
Source: Efficient Monopolies. The Limits of Competition in the European

Property Insurance Market, Thomas von Ungern-Sternberg,
Oxford University press, 2004.
ISBN 0-19-926881-9
Efficient monopolies
40% more
expensive for
the same cover
Efficient Monopolies. The Limits of Competition in the European Property Insurance Market, Thomas von Ungern-Sternberg, Oxford University press, 2004.

04 Lossmodel

Uploaded by

Copyright:

Available Formats

04 Lossmodel

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

04 Lossmodel

Uploaded by

Copyright:

Available Formats

Climate Change Uncertainty and Risk: from

Probabilistic Forecasts to Economics of Climate

Reto Knutti / ETH Zrich | David Bresch / Swiss Re

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Recap: Risk1 Management

perception is based on a shared mental model

Risk analysis: Quantification, the basis for decision-making

Risk model: the quantitative expression of a shared mental model

Risk mitigation: Prioritization based on metrics, options are to

transfer : Insurance puts a rice tag on risks incentive for prevention

or retain the risk

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

The Risk Management Cycle

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Recap: Note on decision strategies

Natural Nat Cat damages on the rise

10-year moving average insured losses

10-year moving average total economic losses

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Note on trend drivers

Ocean Drive, FL, 1926

The upward trend in natural catastrophe

Ocean Drive, FL, 2000

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Fire versus Nat Cat damage need for simulation

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Fire vs. Nat Cat need for accumulation control

Nat Cat Risk

0.2 of house value

Experience or burning cost approach

The present The path

Natural Catastrophe Risk Assessment Model

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Natural Catastrophe Risk Assessment Model

Whats the impact?

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Probabilistic multi-hazard risk assessment

climada: High-resolution (1x1km) damage model

climada: High-resolution (1x1km) damage model

Tropical cyclones in the Indian ocean

Understanding the hazard how strong

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Tropical cyclones in the North Atlantic

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Tropical cyclones in the North Atlantic

Tropical cyclones in the West Pacific

Tropical cyclones in the Southern hemisphere

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Tropical cyclone intensity the wind field (1/3)

Tropical cyclone intensity the wind field (2/3)

Holland, G. J., 1980: An analytic model of the wind and pressure

Vickery, P.J. and D. Wadhera, 2008: Statistical models of Holland

Tropical cyclone intensity the wind field (3/3)

Vickery, P.J. et al., 2009: A Hurricane Boundary Layer and Wind

and an update of the Holland approach:

Reto Knutti / IAC ETH Zurich | David Bresch / Swiss Re

Damage calculation (1/2)

Damage calculation (2/2)