Photo: © Marcos Juarez, 2012

NatCatSERVICE® and Geo Risks Research

Jan Eichner – Geo Risks Research, NatCatSERVICE

NatCatSERVICE

One of the world‘s largest databases on natural catastrophes The Database Today  From 1980 until today all loss events; for USA and selected countries in Europe all loss events since 1970.  Retrospectively, all great disasters since 1950.  In addition, all major historical events starting from 79 AD – eruption of Mt. Vesuvius (3,000 historical data sets).  Currently ca. 35,000 data sets

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

NatCatSERVICE

Database Structure – Peril Families (updated structure following IRDR DATA Project: work in progress!)

Family

Main event

Sub Peril

Geophysical

Earthquake

Earthquake (Ground shaking)

Volcanic eruption

Fire following

Meteorological

Landslide

Tropical cyclone Winter storm (i.e. extra-trop. cyclone) Tempest/Severe storm Hail storm Lightning Tornado Local windstorm (i.e. orographic storm) Sandstorm/Dust storm Blizzard/Snowstorm

General flood

Heat wave

Flash flood

Cold wave / frost

Extreme temperature

Storm surge

Extreme winter conditions

Drought

Glacial lake outburst flood

Tsunami

Mass movement dry Hydrological Storm

Volcanic eruption Subsidence Rockfall

Climatological

Flood Mass movement wet

Wildfire

Drought Subsidence Avalanche Landslide

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Wildfire Unspecified

NatCatSERVICE

Loss Events Worldwide 2014 Geographical overview Floods United Kingdom, Dec 2013–Feb 2014

Winter damage USA, Canada, 5–8 Jan

Floods Bosnia and Herzegovina, Serbia, Croatia, Romania, 13–30 May

Typhoon Rammasun China, Philippines, Vietnam, 11–22 Jul

Severe storms USA, 18–23 May

Severe storms France, Belgium, Germany, 7–10 Jun

Drought USA, 2014

Winter damage Japan, 7–16 Feb Typhoon Kalmaegi China, Philippines, Vietnam, 12–20 Sep

Flash floods USA,11–13 Aug Hurricane Odile Mexico, 11–17 Sep Severe storms USA, 2–4 Apr

980 Loss events

Severe storms USA, 27 Apr–1 May

Cyclone Hudhud India, 11–13 Oct

Drought Brazil, 2014

Floods India, Pakistan, 3–15 Sep

Severe storms USA, 3–5 Jun

Earthquake China, 3 Aug

Source: Munich Re, NatCatSERVICE, 2015

Loss events Selection of catastrophes Overall losses ≥ US$ 1,500m

Geophysical events

Hydrological events

(Earthquake, tsunami, volcanic activity)

(Flood, mass movement)

Meteorological events

Climatological events

(Tropical storm, extratropical storm, convective storm, local storm)

(Extreme temperature, drought, wildfire)

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE – As at January 2015

NatCatSERVICE

Loss Events Worldwide 1980 - 2014 10 costliest events ordered by overall losses

Date

Event

Affected countries

Overall losses in US$ m

Insured losses in US$ m

original values

original values

Fatalities

Earthquake, tsunami

Japan

210.000

40.000

15.880

Hurricane Katrina, storm surge

United States

125.000

62.200

1.720

17.1.1995

Earthquake

Japan

100.000

3.000

6.430

12.5.2008

Earthquake

China

85.000

300

84.000

Hurricane Sandy, storm surge

Bahamas, Cuba, Dominican Republic, Haiti, Jamaica, Puerto Rico, United States, Canada

68.500

29.500

210

Earthquake

United States

44.000

15.300

61

Floods, landslides

Thailand

43.000

16.000

813

6-14.9.2008

Hurricane Ike

United States, Cuba, Haiti, Dominican Republic, Turks and Caicos Islands, Bahamas

38.000

18.500

170

27.2.2010

Earthquake, tsunami

Chile

30.000

8.000

520

23/24/27.10.2004

Earthquake

Japan

28.000

760

46

11.3.2011 25-30.8.2005

23-31.10.2012 17.1.1994 1.8-15.11.2011

Source: Munich Re NatCatSERVICE, 2015 © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

As at: January 2015

NatCatSERVICE

Loss Events Worldwide 1980 - 2014 10 costliest events ordered by insured losses

Date

Event

Affected countries

Overall losses in US$ m

Insured losses in US$ m

original values

original values

Fatalities

Hurricane Katrina, storm surge

United States

125.000

62.200

1.720

Earthquake, tsunami

Japan

210.000

40.000

15.880

Hurricane Sandy, storm surge

Bahamas, Cuba, Dominican Republic, Haiti, Jamaica, Puerto Rico, United States, Canada

68.500

29.500

210

6-14.9.2008

Hurricane Ike

United States, Cuba, Haiti, Dominican Republic, Turks and Caicos Islands, Bahamas

38.000

18.500

170

23-27.8.1992

Hurricane Andrew

United States, Bahamas

26.500

17.000

62

Earthquake

New Zealand

24.000

16.500

185

Floods, landslides

Thailand

43.000

16.000

813

Earthquake

United States

44.000

15.300

61

Hurricane Ivan, storm surge

United States, Barbados, Cayman Islands, Cuba, Dominican Republic, Grenada, Haiti

23.000

13.800

120

Hurricane Wilma

Bahamas, Cuba, Haiti, Jamaica, Mexico, United States

22.000

12.500

44

25-30.8.2005 11.3.2011 23-31.10.2012

22.2.2011 1.8-15.11.2011 17.1.1994 7-21.9.2004 19-24.10.2005

Source: Munich Re NatCatSERVICE, 2015 © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

As at: January 2015

NatCatSERVICE

Loss Events Worldwide 1980 - 2014 10 deadliest events

Date

Event

Affected area

Overall losses in US$ m

Insured losses in US$ m

original values

original values

Fatalities

12.1.2010

Earthquake

Haiti

8.000

200

222.570

26.12.2004

Earthquake, tsunami

Sri Lanka, Indonesia, Thailand, India, Bangladesh, Myanmar, Maldives, Malaysia

10.000

1.000

220.000

2-5.5.2008

Cyclone Nargis, storm surge

Myanmar

4.000

29-30.4.1991

Tropical cyclone, storm surge

Bangladesh

3.000

100

139.000

8.10.2005

Earthquake

Pakistan, India (Kashmir region), Afghanistan

5.200

5

88.000

12.5.2008

Earthquake

China

85.000

300

84.000

Jul - Aug 2003

Heat wave, drought

All of Europe

13.800

1.120

70.000

Jul - Sep 2010

Heat wave

Russia

20.6.1990

Earthquake

Iran

7.100

100

40.000

26.12.2003

Earthquake

Iran

500

19

26.200

140.000

400

56.000

Source: Munich Re NatCatSERVICE, 2015 © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

As at: January 2015

NatCatSERVICE

Loss Events Worldwide 1980 – 2014 Number of events Number

1 000

800

600

400

200

1980

1982

1984

Geophysical events (Earthquake, tsunami, volcanic activity)

1986

1988

1990

1992

1994

1996

Meteorological events (Tropical storm, extratropical storm, convective storm, local storm)

1998

2000

2002

2004

Hydrological events (Flood, mass movement)

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE – As at January 2015

2006

2008

2010

2012

Climatological events (Extreme temperature, drought, forest fire)

2014

NatCatSERVICE

Loss Events Worldwide 1980 – 2014 Overall and insured losses US$ bn

400

300

200

100

1980

1982

1984

1986

1988

Overall losses (in 2014 values)*

1990

1992

1994

1996

1998

2000

2002

2004

2006

Insured losses (in 2014 values)*

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE – As at January 2015

2008

2010

2012

2014

*Losses adjusted to inflation based on country CPI considering ROE of LCU and US$

NatCatSERVICE

Q:

How do we get estimates for direct economic losses?

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

NatCatSERVICE

Economic Loss Estimation

Five levels of information quality: 1. Info on insured loss in industrial countries, compiled by institutions such as PCS, Perils AG or various Insurance Associations 2. Partial info on insured loss in developing markets / countries 3. Info on total economic loss, often from governments (no info on insured loss) 4. Partial info on economic loss (e.g. impact on agriculture, infrastructure etc.) 5. Only description of event (e.g. number of houses damaged / destroyed by flood, storm, earthquake etc.)

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

NatCatSERVICE

Economic Loss Estimation Based on insurance market loss information (info level 1)

Economic loss estimation based on insured loss data is of best quality! …and easiest way to scale up

3. Modulation of economic loss based on event-specific information and/or NatCatSERVICE experience 2. Up-scaling of insured loss based on insurance penetration info

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

1. Insured loss info

NatCatSERVICE

Economic Loss Estimation Example for info level 1 (loss event in TX and OH) Convective storm event in Texas and Ohio in 2013 (Source: PCS) Major damage in TX from : Major damage in OH from:

flash floods and wind wind

Split by state and lines of business: TX: 16.5m (res) + 5.5m (com) + 9.7m (auto) = 31.7m US$ OH: 14m (res) + 5 (com) + 0.7m (auto) = 19.7m US$ Insured loss:

51.4m (PCS) + 30.2m (NFIP) = 81.6m US$ Assumptions on insurance penetration rates in affected areas: For Texas:

Insurance penetration residential: Auto and commercial penetration: NFIP penetration:

81% 95% 20%

For Ohio:

Insurance penetration residential: Auto and commercial penetration:

99% 95%

Assumption for limits / deductibles: © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

5%

NatCatSERVICE

Economic Loss Estimation Example for info level 1 (loss event in TX and OH)

Calculating the economic loss based on insured loss: TX:

OH:

Residential: Auto and commercial: Flood losses:

16.5m / (0.81 * 0.95) = 21m US$ (5.5m + 9.7m) / (0.95 * 0.95) = 17m US$ 30.2m / 0.20 = 151m US$

Plus assumption of ~25% infrastructure damage:

= 47m US$ 224m US$

Residential: Auto and commercial:

= =

14m / (0.99 * 0.95) (5m + 0.7m) / (0.95 * 0.95)

Plus assumption of ~10% infrastructure damage:

Combined direct economic loss estimate: 247m US$

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE – As at January 2015

15m US$ 6m US$

= 2m US$  23m US$  rounded: 250m US$

NatCatSERVICE

Q:

What is driving the losses?

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

NatCatSERVICE

Definition of Risk in the Insurance Industry

Risk ~ Hazard x Vulnerability x Exposure

All three factors can and will change over time!

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

16

NatCatSERVICE

Examples of Drivers of NatCat Losses El Niño

1914

La Niña

Jet stream during La Niña: Shift of tornado activity

2012

Exposure:

Vulnerability:

- Inflation

- Building codes

- Population increase/shift

- Improved materials

- Increase of wealth

- Expensive materials

- Increase of building stock - Flood zones © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Hazard: - Natural variability (rather short time scales)

- Climate change (long time scales)

NatCatSERVICE

Bias in Loss Data over Time

Two main biasing factors: a) Improved reporting of events (internet etc.) b) Increasing wealth, population and destructible assets (socio-economic growth) # of events

# of events

Reporting effect

Increasing exposure

Loss event magnitudes today today past

past

today

US$

Trend in # of events © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

US$

Trend in loss magnitudes

NatCatSERVICE

Bias in Loss Data over Time

How to overcome these biases when doing time series analysis of loss data?

1. Normalization of loss data  eliminates socio-economic trend bias

2. Introducing a lower threshold to normalized losses  eliminates reporting bias

Procedure is necessary when studying other factors of influence on loss data!

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

NatCatSERVICE

Q:

What to do with historic loss data?

Example: Thunderstorm losses in the USA

Sander, J., J. Eichner, E. Faust, and M. Steuer in: Weather, Climate, and Society, March 2013, DOI: 10.1175/WCAS-D-12-00023.1 © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Relation between climate variability and losses? Thunderstorm losses in USA (1970 - 2009)

Source: Munich Re NatCatSERVICE database

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Relation between climate variability and losses? Normalization of U.S. thunderstorm loss data

• To find potential climate signal in loss data time series one has to remove the signal of increasing destructible wealth (= socio-economic growth) Normalization of past direct economic losses to current levels of wealth:

𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛

𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡

= 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑦𝑦𝑦𝑦 𝑜𝑜𝑜𝑜 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒

𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤ℎ𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 ∗ 𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤ℎ𝑦𝑦𝑦𝑦 𝑜𝑜𝑜𝑜 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒

• Two proxies for wealth used: - building stock (BS) (number of home units) x (nominal median value of homes) - GDP (population) x (nominal GDP per capita) © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Relation between climate variability and losses? Normalization of U.S. thunderstorm loss data

Original direct thunderstorm losses, east of 109° W (east of the Rockies), March – Sept.

© Munich Re, 2012

Normalization using building stock as a proxy for destroyable wealth

Normalised thunderstorm losses (state-based)

© Munich Re, 2012

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Relation between climate variability and losses? Removing reporting bias in U.S. thunderstorm loss data

To ensure homogeneity of normalized loss events over time: Find threshold selecting sizeable normalized loss events that would have been detected at any time. Here: per-event threshold of US$ 250m (US$ 150m insured) in normalized loss associated with multi-state loss during all of the analysis period.

Normalized loss events exceeding US$ 250m account for 80% of the total loss aggregate in the analysis period.

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Relation between climate variability and losses? Removing reporting bias in U.S. thunderstorm loss data

Original direct thunderstorm losses, east of 109° W (east of the Rockies), March – Sept.

© Munich Re, 2012

Normalization using building stock as a proxy for destroyable wealth

Normalised thunderstorm losses (state-based)

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

© Munich Re, 2012

© Munich Re, 2012

Selecting sizeable multi-state loss events (> $250m) to ensure homogeneity in detection.

Normalised thunderstorm losses from events > US$ 250m (state-based)

Relation between climate variability and losses? NCEP/NCAR reanalysis data  NCEP/NCAR reanalysis data: symbiosis of climate model and measurements Chosen grid points for analysis:

Worldwide grid with1.875° x 1.915° spatial and 6h temporal resolution, selecting 1970 – 2009, March – September

 Parameters for severe convective storm events:  CAPE - Convective Available Potential Energy  DLS - Deep-Layer Wind Shear

(temperature and humidity)

(up- and down-draft winds, supports convection)

 From this we calculate TSP (Thunderstorm Severity Potential) TSP := �𝟐𝟐 × 𝑪𝑪𝑪𝑪𝑪𝑪𝑪𝑪𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎 𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍𝒍 𝟏𝟏𝟏𝟏𝟏𝟏 𝒉𝒉𝒉𝒉𝒉𝒉 × DLS6km AGL-GL

(J. Sander, 2011)

[J kg-1]

 Severe thunderstorm forcing environments defined by very high values of TSP ≥ 3,000 J kg-1, corresponding to the 99.99th percentile of distribution. © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Relation between climate variability and losses? Correlation btw. TSP environment and loss data

Correlations may appear in… …FREQUENCY of events or …INTENSITY of events or …BOTH Hence, we have to look at both at the same time: - NUMBERS of threshold exceedences per time step (counts) - SUMS of INTENSITIES exceeding the thresholds per time step (aggregated values)

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Standardized seasonal aggregated value

Standardized seasonal count

Relation between climate variability and losses? Correlation btw. TSP environment and loss data

count of TSP per grid point > 3,000 J kg-1 count of norm. loss events ≥ $250m (BS) count of norm. loss events ≥ $250m (GDP)

Seasonal count: TSP, norm. economic losses aggregate of TSP per grid point > 3,000 J kg-1 aggregate of norm. loss events ≥ $250m (BS) aggregate of norm. loss events ≥ $250m (GDP)

Seasonal aggregate: TSP, norm. economic losses BS, GDP: different normalization approaches using either building stock (BS) or GDP (GDP) as a proxy for wealth

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Relation between climate variability and losses? Correlation btw. TSP environment and loss data 7-year running means

How will CAPE and shear develop in the future? Over past 40 years CAPE in North America shows a clear trend that correlates very well with observed warming in the area over the same time span.

Source: NOAA NCEP/NCAR reanalysis data, wmax > 42 m/s © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

29

Rise of March-Sept. aggregate of wmax in line with rise of specific humidity in northern hemisphere (WORK IN PROGRESS) Six-hourly wmax aggregated per March – September season (1970-2009) from analysis domain (NCEP/NCAR reanalysis). Threshold of SQRT(CAPE) = 42 m s-1 (corresponding to CAPEml ~ 1,764 J kg-1) was applied. Produced from the new global high-resolution, quality-controlled land surface database HadISDH, available at: http://www.metoffice.gov. uk/hadobs/hadisdh/online material.html For infromation on HadISDH see Willett, K.M. et al., 2013, Clim. Past. 9, 657-677.

Findings…

• Increase in variability and mean level of severe thunderstorm-related normalized large losses (USA east of Rockies, 1970 – 2009, March – Sept.) • Changes in losses reflecting increasing variability and mean level in thunderstorm forcing, i.e. changing climatic conditions. This finding contradicts the opinion that changing socio-economic conditions are the only driver of change in thunderstorm-related losses. • Changes coincide with rise in low-level specific humidity and in seasonally aggregated potential convective energy. These effects seem consistent with the modeled effect from anthropogenic climate change that other studies have demonstrated. Further research that is underway • Can we identify variability signals in other perils & loss data / in other regions? • Can we identify variability signals similar to the observation also in climate change projections? © 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

NatCatSERVICE

In Summary Tasks / interests of Geo Risks Research & NatCatSERVICE

o Collect and analyze worldwide data and info on all types of NatCat losses o Find correlations between loss patterns and patterns on the hazard side (i.e. meteorological, hydrological, geophysical, …) o Learn about economic consequences of temporal changes in these patterns o Learn about impact of both, climate change and climate variability o Eventually use this info to assist the steering of insurances‘ NatCat business o …but also as an indicator for changing risks in society and economy in general!

© 2015 Münchener Rückversicherungs-Gesellschaft, Geo Risks Research, NatCatSERVICE

Thank You for Your Attention!

33