SEASONAL FORECASTING- A RE/INSURANCE PERSPECTIVE Sub-seasonal to Seasonal Prediction Workshop December 2 nd 2010 Matthew Foote Research Director Willis Analytics [email protected]
Jan 27, 2016
SEASONAL FORECASTING- A RE/INSURANCE PERSPECTIVE
Sub-seasonal to Seasonal Prediction Workshop
December 2nd 2010
Matthew FooteResearch DirectorWillis [email protected]
2
Managing Financial Extremes -the (re)insurer’s world view
0 1000Return Period (years)0
1
Loss
($b
illio
ns)
3
Risk and risk transfer – a quick summary
When faced with risk there are three commonly available options: AvoidanceAcceptanceTransfer
Risk Transfer - the basic principle of insuranceA common pool of premiums is collected from many insuredsLosses of the few are then paid from the common poolThe contribution to the pool must reflect the amount of risk that each insured brings to the poolRisk can be transferred from individuals to an insurance companyRisk accepted by the insurance company may be transferred again to a reinsurance company
“Risk”
“Hazard”
Avoidance Acceptance Transfer
4
Natural Catastrophe and Insurer (in)Security
Example insurer: Capital requirement gross of reinsurance under EU Solvency 2 QIS4
UnderwritingMarketDefaultOperational
77%20%0%3%
SCR 362,306,041
Underwriting 329,678,231Catastrophe 300,000,000
Market 69,468,900Default 0Operational 8,788,400
Risk Based Capital Regulation via Modeling Extremes
5
insured losses*
Insurance losses and the challenge of extreme weather
•Insurance losses from natural catastrophes are increasing (but vary variable) – why?•Insurance is at the interface between ‘weather’ and ‘climate’•It’s all about the decision maker - moving from the theoretical to the operational forecast
•How to effectively include the complexity of process-response?•Statistics are useful (and simple relationships between statistics – e.g. SST/ACE)– but purely empirical studies have shown limited value for decision making
2010 total economic catastrophe losses USD 222 billion Insured: USD 36 billion(Swiss Re) – but in line with 20 yr average – quake dominant,hurricane low
*source: Swiss Re, Sigma no 1/2008
number of events*
6
What do re/insurers need from seasonal forecasting?
Non-marine (re)insurers are interested in land! Activity rate forecasts are not enough – we need to understand landfall
Severity is important but not everything – is ACE the most useful parameter?
Temporal and spatial clustering is equally worrying – and can significantly affect price / capital
77
Regimes & Dynamic Allocation of Capital
NOAA Hurdat reanalysis: Storms in a box since 1851 – Gero Michel
Changing Regimes
8
It’s a question of timing…
Most insurance and reinsurance contracts are annual
(some are longer, some shorter but these are exceptions)
Inception dates ‘cluster’ – many on 1st Jan, others on 1st April, June, July
Insurers and reinsurers are therefore taking a snapshot of expected exposure up to 12 months in advance of time
Effective forecasting requires confidence in trends at least 4 months ahead – pricing, capacity, structuring
Increasingly, concerns regarding climatic factors – rates of event frequency, clustering potential
9
OK – hurricanes I get, but what about the rest?
Current research shows that we could include our understanding of the physical dynamical process to the operational forecasting of hurricanes at seasonal / decadal scales (Met Office, Exeter, Reading ECMWF, NCAS, NCAR etc. etc. etc.)
Some re/insurers are (probably, but they are unlikely to let on…)
But…
What about European winterstorm? Chinese winter storms? US drought? Central European Flood? Global / regional Correlations? Anti-correlations?
Lead times for re/insurance decisions are 12 months (should they be?)
And will (re)insurers embrace seasonal forecasting en masse?
10
Seasonal forecasts within operational climate services
What is needed is a coherent knowledge base and operational source of credible data across regions
Effective linkage of catastrophe modelling and seasonal forecasts
Critical to know “When is a forecast successful?” Let’s take 2010 hurricane season as an example…
19 named storms (second largest year after 2005 in the period of named storms, and in the top of all recorded storm counts)
Numerous forecasts suggested high activity quite early on (e.g. CSU Dec 2009 ‘be more active than the average for the 1950-2000 seasons ‘’11-16 named storms
But… landfalling low impact, limited clustering – intensity not always indicator of impact (Ike???) – so what do you measure as ‘skill’?
11
Making forecasting relevant
• Willis Hurricane Index – improving the representation of key hurricane processes and calibrated against the Willis Energy Loss Database
• Developed to assess GoM hurricane damage potential
• Coupled to NCAR WRF high resolution climate / weather models – but currently short term
1212
Integrating earth system modelling into global forecasting
Thanks to: Prof. Pier Luigi Vidale, Dr Jane Strachan, NCAS-Reading, Prof David Stephenson, Dr Renato Vitolo, Exeter University, Dr Aidan Slingsby, City University
We (re/insurers) know that the models need to better represent global scale climate systems (oceanic and atmospheric) within extreme weather models
E.g. Madden-Julian Oscillation influence on Asian rainfall, European drought
ENSO / East Atlantic wave formation
Predictability? Forecast Validation?
13
Towards Operational Climate Forecasting
14
Historic and Future UKMO Hurricane Activity Forecasts
2005
2009
2010
15
UK Met Office 2010 hurricane forecasts – June 2010
Current conditions (sea surface) and dynamical models were used to generate forecasts for the coming season
The forecast indicated an enhanced risk (29.3%) of exceeding 25 tropical storms during June–November 2010
16
Embrace the uncertainty!
We know that there will always be random elements in extreme weather (for insurance this is good!)
Initiation, genesis, shear, track, rainfall, clustering
Models are as models do…
What is critical is for decision makers to be given the knowledge around the forecasts (predictions?) that make those decisions more relevant and can be underwritten (intellectually) by those providing the knowledge
Ensemble models and ensemble knowledge are key to improved uptake in the industry
17
And if all else fails…
“Astrometeorology defines Mercury as the planet of wind; the mercurial mind as relating to human intellect and the rapid nerve transmission in the brain can be seen in the atmospheric condition as rapid air movement. “
http://www.rhegeds.freeserve.co.uk/gales.htm
18
And finally…