If we are to start this project of lifting the veil on insurance speak somewhere, we might as well start with the letter A, and the acronym AAL. Let's start with the basics: AAL stands for Average Annual Loss, and is used in modeling catastrophic (usually Property, but also applicable to other lines) events over what can seem like hair-raising timespans: tens and hundreds of thousands of years are commonly simulated to arrive at the final average.
Why do we care about the average of 500,000 years of simulated windstorms, when we are underwriting a 12 month policy for a loss free account? The answer lies in the nature of insurance itself: the Law of Large Numbers. This concept describes that the more data you include in any query, the more closely it resembles any possible whole group.
For example, think about a group of homeowners seeking property coverage for their Italian summer residences. They present you with proof of two loss free years, 2023 and 2022. The Italian villas look stunning, well taken care of, and there is clear pride of ownership. Why are they seeking special coverage? Well, they are high networth individuals and don't care for some of the standard exclusions, far fetched things hidden in exclusion paragraphs. They want true all risk coverage, and with fantastic weather scores across the board you gladly grant it. The next year every single one burns down.
Turns out these villas were located on the side of the world’s second most active volcano, Mount Etna. Including even one more year of loss data would have shown an eruption in 2021. If data since 2000 had been used, the underwriter would have seen 12 years of eruptions, so in a binary decision an average year would be just as likely to have an event as not. A far fetched exclusion for volcanic activity for this all too homogeneous group of policyholders suddenly becomes absolutely necessary.
AALs differ from this example by looking forward not backward, as they are based on simulations of what is to come, not just loss experience at any given location. Historic facts still are a major component, but come with significant drawbacks. Most perils have limited records, some dating back to the 19th century, others only to the 1980s. Especially older records are not continuous or based on constant definitions. A forward looking simulation satisfies the Law of Large Numbers much better than using any such limited backward looking data set.
The other benefit of a predictive simulation is the ability to alter conditions. Without going too far into the topic certain changing climate conditions have to be built into any assumption, if the outcome is to be taken seriously. Droughts, frequency and severity of storms, rising sea levels all contribute to changing probability of natural catastrophic events.
What an AAL cannot deliver is insight into how a specific account is prepared to handle these stressors, be they changing or a consistent extrapolation of historic circumstances. Take our volcanic villa dwellers. An AAL analysis would have quickly shown both frequency and severity potential from volcanic activity. But we can assume that they, having built their houses on the slope of a known volcano, are aware of this threat, and have put measures into place to reduce the risk of a partial or total loss.
This is where a PML calculation becomes a necessary tool in any underwriter’s toolbelt. PML stands for Probable Maximum Loss. Note that it is not Possible, but Probable. There should always be the assumption of the possibility of a total loss, not only from a natural catastrophe, but from fire and any other covered peril. An AAL analysis will show the likelihood of a total loss in addition to the actual expected average loss. A PML analysis in comparison takes risk mitigation factors into account.
New roofs for windstorms, pipes for water losses, even the choice of which trees are planted along the perimeter of a property in a wildfire prone area all should be considered when thinking through the loss potential of a property. Oftentimes these elements then are factored into a set of eligibility criteria (what can we not live without?), rating factors (what makes for a superior risk?) and terms and conditions (what deductible or self insured retention - SIR - is applied?). This latter set of considerations ensures risk transfer between policy holder and insurer only occurs for the desired risk amount above a limit that might contain frequency or attritional losses better paid for by the policy holder themselves. A certain frequency of smaller expected losses would make the pricing inefficient, both for the policy holder and the insurer. At the same time exposures with low frequency but above a comfort point for the insurer lend themselves well for a separate risk transfer, for example to a facultative reinsurer.
The PML analysis and its resulting account considerations in turn feed into the AAL: higher deductibles or SIRs reduce the exposure at any given location, as do limits below the total exposure. While an AAL should take into account the total exposure, it also needs to show the net TIV, aka the exposure without any primary layers retained one way or another by the policy holder and any top layers either retained or transferred again. It will then show the difference in AAL between the “whole” and the “actual” exposure, and can inform the next round: what additional steps can be taken to protect the property and improve the risk quality.
I hope this explanation resonated with you! If you have additional comments or questions, please send them my way! And if you have a villa in Sicily you need a few risk thoughts on, I am happy to travel and take a closer look! I was fortunate enough to spend a winter vacation in Sicily as a kid, and it was honestly worth sweeping up Volcanic soot on the veranda every morning…
As always, this publication expresses my personal thoughts. If you find any incorrect information, please let me know and I will update accordingly.
Do your models take into account what has transpired since covid with the rising and lowering of material costs? What about the labor rates of the roofs, plumbing systems and other aspects of the properties?
Are these figured into the the statistical analysis?
One question I have come to ask on TIV (Total Insured Value), is why an insurance company doesn't actually factor in the total cost to build back the property as a whole. A couple of exclusions in the calculations I see are the cost for demolition and dirt work to rebuild. When I have asked what the TIV is, the answer is that it is the limit of insurance on the dec pages. The following question is what does the limit of insurance mean which is dependent on the coverage RCV (Replacement Cost Value) ACV (Actual Cash Value), in RCV terms, this typically means the cost to rebuild the structure as is and typically as if it's a fresh piece of ground untouched. Some policy's deal with this in the form of a 5-10% limit of coverage A as additional insurance for debris removal and some don't.
Ok, I read the article and now I am more interested in how you extrapolate the data for big insurance even more.
In googling "Is the US running out of trees" the first two articles that came up contradict each other. The first from of course a hard wood company says "no", https://nwh.com/hardwood-blog/history-of-us-forest-resource-management/#:~:text=Absolutely%20not.,over%20the%20past%20100%20years.
The 2nd from a study out of Harvard says "yes", actually in 300 years, https://www.pollutionsolutions-online.com/news/green-energy/42/breaking-news/how-long-until-there-are-no-more-trees/36067#:~:text=Alarming%20new%20research%20conducted%20by,in%20just%20over%20300%20years.
In my neighborhood, when it was constructed in the 50 - 60's, they only planted one or two different species of trees. Maybe it was 4, I don't remember but they look beautiful and they are dying. What I learned in the last two years is that there are actually male and female trees. I also learned that through human learning/intervention we have destroyed ecosystems by planting mainly either male tree dominated area's or female dominated area's. The problem being that they survive off of each other.
I say all this because I read the RMS article and I understand the idea behind pulling the data from the government, but do you in your job ever dig deeper or is there anyone in insurance that is actually digging deeper? I think relying on governmental data has shown to have flaws and potentially be skewed in certain ways. Is there a bigger picture thinktank or model that takes into account deeper data dives? For example, the college effect, the US has made a massive push for everyone to go to college, thus driving down the thought of working in a manual labor job. Every contractor I talk to his hurting for skilled workers let alone workers that will actually show up for day 2 of the job. So now the US is filled with college educated folks, who either can't or don't want to pick up a hammer. Is this data in the algorithm?