Catastrophe Models: Learning from Superstorm Sandy

Transcription

1 Catastrophe Models: Learning from Superstorm Sandy January 2013 Lockton Companies Although Superstorm Sandy was only a Category 1 hurricane, it made landfall on October 29 as the largest Atlantic hurricane on record. By October 31, catastrophe modeling firms EQECAT and AIR estimated Sandy s insured losses would reach $10 to $15 billion, with RMS only indicating that Sandy s losses would exceed those experienced in 2011 from Hurricane Irene ($4.3 billion) i. Two weeks later, RMS estimated Sandy s insured losses at $20 to $25 billion, which is well-aligned with the Swiss Re estimate of $20 and $25 billion in insured losses. While the RMS and Swiss Re figures match up, was an event like Sandy considered in the models BEFORE it happened? To address this question, let s consider the following ingredients of Sandy: Track RMS includes simulated events that closely match Sandy s unusual veer toward the Northeast. Sandy s unusual track also increased the impact of storm surge. This article was written by Jeff Tennis Manager Catastrophe Analytics jtennis@lockton.com While the RMS and Swiss Re figures match up, was an event like Sandy considered in the models before it happened? L O C K T O N C O M P A N I E S

2 Storm surge RMS claims its model represented Sandy s storm surge reasonably well, but has provided few details supporting this claim. The yellow and red markers on the map identify south areas experiencing at least a 12-foot storm tide. In fact, a 16-foot surge was observed at Raritan Bay on Staten Island, New York. When analyzing this location in RMS, the following was noted. RMS simulated more than 2,300 storms in this area. The maximum storm surge produced was more than 22 feet. Sixty-two storms produced a 16-foot or greater storm surge: According to RMS, Sandy s low central pressure (940 millibars) and slow forward speed (11.2 mph) emphasized storm surge heights. Of the simulated events generating a 16-foot surge: On average, the simulated storms were a category 3 hurricane. Only 14 simulated events had a forward speed within +/- 25% of Sandy s observed forward speed. Inland flood RMS stated inland flooding is not a key driver of loss, although areas in Maryland received more than 12 inches of rain during the storm. Inland flooding is not captured with the RMS model. Indirect losses The area hardest hit by Sandy accounts for 23% of national GDP ii. Sandy devastated the electrical and transportation network throughout the Northeast, generating substantial business interruption claims. Although models quantify direct business interruption, measuring indirect business interruption is far more difficult. Dennis Sugrue of Standard & Poor s stated, The complexity of business interruption claims, where there are losses arising from disruption to the policyholders business, makes it difficult to model and estimate losses. Modelling agencies can model BI losses within their catastrophe modelling software, but the diversity of policies covering BI means a generic modelling approach to loss estimates is highly problematic. (Incorporating business interruption losses into model estimates) can be very difficult, because the specific policies can be so nuanced as to what covers a loss or when you can claim on those business interruptions. iii 2

3 January 2013 Lockton Companies Many insureds will have contingent business interruption claims, due to power and/or transportation disruption, as well as supply chain disruptions due to Sandy. Other sources of loss which vendor models ignore include losses due to prevention of access to a facility (ingress/egress, civil authority) and infrastructure losses iv v. Throughout the industry, there is agreement these sources of loss will be significant. Although catastrophe models fail to consider all possible sources of loss, it would be naïve to fully discredit them. Insurance professionals must recognize model deficiencies and supplement where needed. Some considerations include: When did the last major event in the area of interest occur? Longer time intervals between events will increase the uncertainty in loss estimates. For example, the last hurricane to cause widespread damage in New York was Hurricane Agnes 40 years ago in What are potential sources of indirect losses, such as widespread power failures, transportation disruption, dam/levee failure, etc.? Where are critical suppliers or distribution centers located? p Events like Sandy will continue to expose deficiencies in catastrophe models, particularly for areas with a limited historical record. To illustrate this point, consider the New Madrid seismic zone pictured. The last major New Madrid earthquake occurred 200 years ago. Given the limited historical record, it is highly unlikely any model can accurately predict the estimated cost of a significant New Madrid earthquake, but models are still a useful tool to evaluate how bad bad can get. Events like Sandy will continue to expose deficiencies in catastrophe models, particularly for areas with a limited historical record. 3

4 In fact, a FEMA study suggests a 7.7 earthquake in the New Madrid region would cause $300 billion of direct economic damage, while also suggesting indirect losses could reach $600 billion. Insurers would bear 60-80% of the economic damages from this event, due to the relatively high take-up rate of earthquake insurance in the New Madrid region vi. Indirect loss contributors from this event could include: Transportation infrastructure disruption, including more than 3,500 damaged bridges. Fifteen major bridges would be unusable. 425,000 breaks/leaks to interstate gas pipelines. Severely inhibited road, rail, air and river travel. Substantial damage to utility infrastructure, leaving millions without water or electricity. 42,000 personnel required for nearly 1,500 search and rescue teams vii. 4

5 January 2013 Lockton Companies In this example, the indirect losses are significantly greater than the direct losses. Albeit rare, these types of events heighten the need to utilize catastrophe models with a critical eye. Considering the direct and indirect losses provides a comprehensive view of catastrophe risk, as Sandy highlighted. What can be done to supplement modeled outputs? Understand the policy coverages. Does the policy provide coverage due to service and utility disruption, contingent business interruption, extra expenses, civil authority or ingress/egress? Although models do not explicitly account for these loss sources, adjustment factors can be applied to modeled output if these are concerns for the insured. Quantify the completeness and accuracy of data inputs. Lockton offers a data completeness report for any insured, importance-weighing the critical data elements for catastrophe models. Lockton s report highlights data elements which are present or missing from the analysis. Pinpoint the location of critical suppliers and infrastructure dependencies. Is your client dependant on a utility company located in a CAT-prone area? As the Thailand floods emphasized, are critical suppliers located in a CAT-prone area? If a critical supplier is disrupted for weeks, or even months, how would this impact your insured? 5

6 Stress test catastrophe models. What return period do the limits of insurance equate to? Are the covered locations in areas prone to flood, or located near a levee in a high-risk area? Does your client have a concentration of exposures very near each other, which could easily consume the limits of insurance? How sensitive are the modeled outputs to storm surge and/or demand surge? Catastrophe models now allow the user many options for modeling storm surge, such as the assumed NFIP take-up rates, or if there will be leakage of storm surge losses paid by wind-only policies. It is critical to know which options were modeled by the broker and carrier. Lockton performed an analysis demonstrating that storm-surge-related losses can differ by nearly 500% depending on the modeling options selected. Recognize models will always be wrong and there will always be nonmodeled factors. Each catastrophe has unique circumstances that no model will ever capture. For instance, governors of the states impacted by Sandy ordered property insurers to enforce all-other-peril (AOP) deductibles, rather than percentage hurricane deductibles post-sandy. If an insured has a 5% hurricane deductible for a $500,000 home, its hurricane deductible is $25,000, while the AOP deductible may have been only $500 or $1,000. This issue alone creates a significant gap between the modeled loss estimates and reality. i ii iii iv v vi vii 6

7 7 January 2013 Lockton Companies

8 Our Mission To be the worldwide value and service leader in insurance brokerage, employee benefits, and risk management Our Goal To be the best place to do business and to work Lockton, Inc. All rights reserved. Images 2013 Thinkstock. All rights reserved.