Catastrophe modeling for the life industry is developing rapidly
The September 11 terrorist attacks forever changed the insurance industry's perception of risk. Property insurers were forced to actively manage a new catastrophe peril, and workers' compensation, group life, personal accident, short and long-term disability insurers were forced to acknowledge the implications of catastrophe risk on their books of business for the first time. Catastrophe modelers immediately adapted to the changing risk environment, overcoming numerous technical challenges to develop new tools to assess and manage the peril of terrorism, and the life and casualty exposures at risk.
The first terrorism model, released by AIR in 2002, estimated not only property losses from potential terrorist attacks, but also injuries and fatalities. This enabled insurers to probabilistically estimate potential catastrophic workers' compensation losses for the first time. This was quickly followed by the first probabilistic analysis of a life portfolio.
Since then, several additional modeling techniques have been developed to further advance catastrophe analysis of life and casualty portfolios.
Advances in modeling
Prior to the September 11 terrorist attacks, catastrophe reinsurance for life and casualty insurers was readily available at reasonable rates.
In the aftermath of the attacks, the market changed abruptly. Catastrophe life and casualty reinsurance was suddenly expensive and hard to come by. Furthermore, terrorism isn't the only risk that casualty insurers face. Earthquakes, like terrorist attacks, occur without warning and carry a similar potential for injuries and loss of life.
The impact of the terrorist attacks on the life industry can be compared to the impact of Hurricane Andrew on the property industry. Since Andrew, collecting, maintaining and analysing high quality, detailed data about exposed risks has become a top priority for property insurers. Now, life and casualty insurers must take the same steps to understand and manage catastrophe risk. Catastrophe modelers are facilitating the effort with the release of innovative new modeling techniques and tools.
Injuries and fatalities
Earthquakes don't kill people - buildings do. Hence the methodology developed for estimating the number of injuries and fatalities that result from catastrophic events can, in large degree, be leveraged from the experience modelers have in estimating damage to structures. Damage to non-structural elements, such as the shattering of windows or the collapse of suspended ceilings, will result in injuries of a certain severity. Damage to structural elements - fallen beams, for example - can cause major injuries and even death. Complete building collapse will be responsible for large numbers of fatalities. The shifting of contents can also be a major contributor to injuries in the event of an earthquake.
For terrorist attacks employing conventional weapons, the damage footprint can be quite small but still result in a large number of severe injuries and fatalities. Engineering research and weapons effects models can be employed to estimate the number of injuries and fatalities for events of varying intensity. Once the number of injuries and fatalities by location is determined, insured losses are calculated, accounting for all policy conditions and reinsurance. Estimating the injuries and fatalities resulting from terrorist attacks using weapons of mass destruction requires a somewhat different approach. For example, AIR uses a government-created tool called Hazard Prediction and Assessment Capability (HPAC) to estimate contaminant levels and fatalities from a variety of chemical, biological, radiological and nuclear scenarios. HPAC was recently used by the US government in preparation for the war in Iraq to model the potential impact of the use of weapons of mass destruction in the conflict.
The location of individual risks is the most important information insurers need in order to analyse the risk to their portfolios. This is a challenging proposition for the life industry as, unlike property, people are often on the go, without a fixed location. To address this challenge, AIR has developed new techniques in data analysis that help group and individual life, personal accident and disability insurers augment and make intelligent assumptions about their exposure data.
Unlike workers' compensation policies, life policies cover individuals 24 hours a day. Hence, the primary challenge with modeling a group life portfolio is estimating where people are located throughout the day. Some insurers have little specific information about the location of their clients' employees. In many cases, information about where employees live will not be readily available and an insurer may only have the corporate headquarters as a single data point, even if the company has thousands of employees located in many locations across the country.
To help insurers determine where risks are located throughout the day, AIR has compiled a proprietary database containing detailed information on more than six million buildings and nine million businesses in the US. For clients with no location information other than the name of the business, AIR uses this database to determine where its offices are located and the number of employees in each office. Detailed information on size, construction and occupancy is also incorporated - information critical for modeling losses.
Catastrophes can happen at any time of day, therefore knowing the location of the insureds at night is also important. When home address information is not known, AIR can use US Census Bureau data to distribute employees among their residences. To increase accuracy, information on average commute distance is used to refine the distribution. Once the location of all employees is estimated, AIR's database of residential properties is used to estimate the type of structures (e.g., wood frame, masonry) these insureds live in to determine their vulnerability.
For insurers with specialty exposures such as voluntary accident and business travel, policyholders are on the move, which makes their exact location more uncertain at any given time. Reasonable assumptions need to be made using various data sources to estimate worst loss scenarios and probabilistic loss estimates. For example, in the case of voluntary accident, issues such as vacation time, multiple shift operations and travel must be taken into account. For example, AIR recently analysed business travel risk for a client using data on business trip characteristics from a number of government and private sources. Through these types of data augmentation exercises undertaken in coordination with the client, it is possible to create an exposure data set that provides a representation of their portfolio suitable for modeling.
Estimating potential losses for long and short-term disability from an extreme event poses additional challenges for a catastrophe modeling company.
In addition to injury severity, the duration of the injury is important for estimating the financial payout. A further challenge is incorporating the complex benefit structure associated with disability claims.
In the event of an injury, disability claims kick in after an 'elimination period' that typically represents the employee's number of sick days.
Short-term disability policies generally provide a weekly percentage of one's salary for 13 to 26 weeks beyond the elimination period. Once short-term disability benefits expire, long-term disability begins, paying a monthly percentage of one's salary for an extended period of time, typically two or five years, or until age 65. Actuarial models are employed to make estimates of average disability claims cost based on the average duration of injuries and monthly payment.
Life insurance executives are placing increasing importance on more detailed policy information and accumulation control. GIS-based mapping and modeling tools are designed to assist these insurers meet their objectives.
Life reinsurers are demanding details about the exposures they cover and are asking for estimates of potential loss. This also puts the onus on primary insurers to know where their exposures are located, the value of those exposures and the potential losses. For many years, rating agencies such as AM Best have required property and casualty insurers to provide details about their catastrophe exposures. This type of information may well be required from life insurers in the future.
In just the last two years, catastrophe modelers have made significant progress in catastrophe analysis for life and casualty portfolios, leveraging their more than 15 years of experience in modeling property portfolios.
From the first terrorism and workers' compensation models to recent breakthroughs in data analysis for group life, accident and disability insurers, catastrophe modelers have made it possible for life insurers to effectively manage their catastrophe risk in a new and rapidly changing environment.
- Uday Virkud, PE is a senior vice president and Frank Fischer is a senior account executive at AIR Worldwide Corp, a risk modeling company based in Boston, MA.