After a decade-long drought of major hurricane landfalls in the U.S., 2017 has certainly captured the attention of insurers. Together, Hurricanes Harvey, Irma, and Maria present a rare opportunity for insurance carriers to gain underwriting insight, evaluate any potential disconnects between what underwriters think they’re writing and what claims adjusters pay out, and understand how each company’s unique underwriting guidelines align with industry losses. This year’s storms have ensured that insurance carriers paid dearly for this information; the question is whether they will take full advantage of it.
Because they made landfall in three very different regions, Harvey, Irma, and Maria will offer valuable insights useful for modeling by enabling answers to questions such as:
- Does your exposure and policy data input into the model align with that from loss reports?
- What impact do building codes (and enforcement thereof) have on expected damage?
- Are engineering predictions for structure resilience accurate?
- Does the range of damage to different types of construction align with loss data received?
- Does the customer selection process during underwriting demonstrate a different expected loss profile than what was received?
- Is there claims leakage outside the expected amounts?
- How well does policy wording hold up against claims adjustment?
- Are policy holders on track for being made whole through business interruption/loss of use coverage utilization?
It is too soon to perform this analysis, but the ‘golden’ opportunity is approaching. Typically, for a proper post-event claims analysis, more than 85% of claims should be closed; to reach that point after a major event can take several months. Given the complexity of some of the events this year, the February to April 2018 timeframe should be when most carriers begin their analysis.
Evaluating Data Quality
One of the first tasks (which can be done before claims are fully closed) is evaluating the quality of the data input into models. Coding of policy and exposure information into databases is often done by a junior member of the team, and errors can occur. Miscodes of construction, basement type, total insured value (TIV), deductibles, etc., can skew the representation of the exposure and impact model results (garbage in, garbage out).
Data audits have decreased with the expense reductions that have occurred over the past decade, and these naturally-occurring data issues may not have been proactively addressed. A claims report offers the opportunity to compare loss assessment information with what was entered into systems.
Understanding Your Losses
Once a sufficient portion of claims are closed, an analysis can begin. By comparing underwriting guidelines, model output, and regional differences, you can begin to identify where losses may have been unexpectedly higher or lower than what was estimated by the model; anything with more than a standard deviation difference should be investigated. Were there coding issues? Was there a significant underwriting guideline in place that was not captured in the data? Is there anything unusual regarding the exposure that models may not be able to quantify?
After the 2004/5 storms, I did an analysis on a carrier’s claims and found a few large discrepancies. One was that we were seeing losses lower than two standard deviations from those the vendor was estimating. Upon further investigation we discovered policy wording in our form that significantly decreased business interruption losses below the industry average. This allowed the client to continue work without standard downtimes as there was a substantial difference in how losses would be adjusted, which led to increased activity in that line of business. Armed with the understanding of why the losses were lower, appropriate adjustment to estimated costs could be applied.
For another storm, losses were found to be in excess of more than seven standard deviations from what was expected. Further investigation revealed a manufacturing roof defect arising from insufficient building code adherence that led to roof failure from wind speeds far less than what should have been experienced. Underwriting guidelines were put into place to address the manufacturing defect and business continued. Being able to break down losses and engage in detailed detective work is somewhat of an art, but can yield dividends in understanding of risk.
Evaluating Claims Leakage
It is of critical importance, as well, to listen to claims departments. An understanding of what worked (and didn’t work) on policy forms could give rise to further understanding of claims leakage. For large events, there may be claims for which the complexity of damage can make it difficult to accurately ascertain how much of a loss was covered under a policy versus what may not have been covered (such as from the flood peril).
Sometimes uncovered losses may be paid because protesting them could cost more than the dollar amount of the claim, especially when claims adjuster resources are limited post-event. AIR has adjustable leakage assumption values that modify estimated losses, and a proper analysis can be done to calibrate loss.
Similarly, allocated loss adjustment expense (ALAE) is a non-modeled loss for which you need to account. Once claims are sufficiently closed (you may want to wait until re-opens get a chance to work their way through), an approximation of non-modeled ALAE can aid accurate loss estimations for large events—an important consideration when placing reinsurance cover.
Seize the Opportunity
Insurance exists to help customers recover after an event. Being able to accurately plan for and estimate potential losses post-event facilitates proper financial loss prediction. After a decade without major landfalling storms, the events of 2017 allow for a better understanding of your portfolio. Don’t let this opportunity pass you by. If resources are too tight to devote to a proper analysis, AIR provides a claims analysis service; your account representative can aid both in answering questions from your analytics team and/or setting up a claims analysis project.
