Preparing effective rating agency presentations is important to most companies, but when it comes to articulating catastrophe risk and large loss potential, there can be a disconnect between what insurers and reinsurers think the rating agencies want and what the agencies actually want.
Executive SummaryUsing her frequent interactions with rating agencies as a guide, catastrophe risk expert Karen Clark dispels some myths about what agencies want to know about the use of catastrophe models by carriers they rate and about carrier cat loss potential. It’s not about the model, she says, but about the model assumptions selected and the credibility of the loss estimates.
For full disclosure, the author has never worked for a rating agency. This article is based on frequent interactions and detailed discussions with the major rating agencies over the past several years to determine what they really want to know and what’s most important for demonstrating effective catastrophe risk management. It highlights what these investigations have revealed and dispels some common misconceptions, such as:
- If you’ve been using one model for a while, you have to continue using that model. (False)
- You will be viewed in a better light if you blend models. (False)
- You have to use at least one of the three major vendor models. (False)
Switching Your Primary Model
The rating agencies don’t dictate which models to use or compel you to continue using a model you’ve used in the past. Of course, rating agencies do expect a detailed explanation of why you’ve decided to use another model, particularly if the numbers are significantly different. Rating agencies want to guard against “model shopping” to find the lowest numbers.
One reason companies can get locked into one model is not having strong scientific reasons for choosing that model in the first place. If the model selection is based simply on a general belief that one model vendor is the best, for example, or because other companies are using that model, it will be difficult to explain why that model is no longer being used.
Rating agencies understand that the model vendor does not equal the model. The loss estimates from different model versions from the same model vendor can differ widely—sometimes even more significantly than the loss estimates from two different vendors. For example, the loss estimates generated by the RMS V9 hurricane model may have been closer to the loss estimates from the contemporaneous version of the AIR hurricane model than they were to the loss estimates generated by RMS V11.
This decision must be based on scientific evaluations of the model assumptions and detailed tests of the model output. For example, how do the basic assumptions on event frequency and severity compare to historical data? Are the assumptions conservative or do they seem to underestimate the risk? What are your assumptions with respect to future events?
When you fully understand the historical data and the fundamental risk characteristics of each peril region, you’re in a strong position to support changes in your model choices. For the most part, the models are based on publicly available historical data and scientific literature.
It’s your responsibility to understand the data and the risk—independently of the model—so that you can determine whether or not a particular model version is an adequate representation of that risk.
Theoretically, an “ensemble” approach is better than a single model approach, but this is true in two specific situations. One is when a model is being used to make a forecast, for example, tomorrow’s weather or the projected track for an actual hurricane.
But the catastrophe models do not make forecasts—they provide estimated probabilities of loss based on historical data. They are not trying to predict where the next hurricane will make landfall but rather to estimate where hurricanes are more likely to make landfall.
It’s also true that if there were hundreds of credible catastrophe models, the output from those models might provide the full range of uncertainty around the model output, which would be very useful for insurers and reinsurers. But there are only a handful of models at best for specific peril regions, and most of the existing models employ a similar methodology and therefore do not provide a complete picture of uncertainty.
Because the rating agencies expect companies to have their own views of risk, the models are assumed to be a starting point. Model blending is a more expensive way to develop your own view than adjusting one model. You can examine multiple models without implementing model blending for pricing and portfolio management.
Through the process of evaluating different models you will have more insight into why one model may be better than another as your initial starting point and a more credible analysis for the rating agencies. The rating agencies know model blending or simply weighting the model output does not necessarily produce a better outcome.
Rating agencies realize there is no best or “right” model. The third-party vendor models have improved over the past two decades, but because they are limited by the available scientific data, the models will never be accurate despite the false precision of the model output.
Data limitations have another consequence as well. The model vendors continue to add assumptions that cannot be sufficiently quantified given the limited data. As a result, the model output has become more volatile and prone to human error.
The lack of transparency means you have to do a lot of due diligence to ensure your risk management decisions are not subject to the biases and mistakes in the third-party models.
For the rating agencies, an external model may be the starting point, but you own the risk. Even the model vendors are strongly encouraging you to take ownership of the risk and not rely on their models.
One way to reduce reliance on the vendor models is to implement different tools and alternative ways of looking at your loss potential. For example, do you monitor specific scenarios and Characteristic Events (CEs)? Do you know how much total exposure you have to any one event? Do you know where you could have an outsized loss relative to your peer companies?
The more you know about your loss potential and can convey to the rating agencies, the better. The rating agencies have not required companies to use third-party models, but they would like to see a credible process for understanding and quantifying your large loss potential.
Rating agencies ask you to provide a lot of numbers, but contrary to popular belief, they have relatively few prescriptive requirements. In general, they want to know three things:
- You understand the catastrophe risk in the peril regions where you have significant exposure.
- You have credible methodologies for quantifying your large loss potential.
- You have a robust process for managing the risk.
Based on discussions with the rating agencies, they do not require specific models or even that you use an external third-party model. An effective rating agency presentation will demonstrate your knowledge of the risk, show the results of different approaches for quantifying your loss potential, and illustrate how you control and manage the risk.