May 22, 2019
As risk modelers, we understand that with continuous advancements in science and engineering, excellence is a moving target. On a regular basis, we review new scientific research and data, choose among various assumptions, learn from recent catastrophes, and take advantage of improved technology to make measured changes to our risk models. Over time, this flexibility provides the means to derive progressively more realistic representations of various types of risk. Still, we readily acknowledge that models cannot represent reality perfectly. There is ever-present uncertainty inherent throughout the modeling process, various unmodeled sources of loss, and regions or perils not covered by models.
On the model user side, companies have been communicating a growing imperative to truly understand the risk analysis tools they use and the output they produce. They are more engaged with the modeling process than ever before, and many want the ability to adjust individual model components, to modify the output, or to implement a multi-modeling strategy within the same platform.
In the (re)insurance industry, “open modeling” has been getting a lot of attention in recent years, and this term can encompass several concepts. I wanted to take the time to explain AIR’s philosophy toward transparency, flexibility, and open platforms; provide a high-level overview of some of the steps we have already taken; and preview our future plans.
The first concept I want to address is that of transparency. In the past few years, we’ve seen a transformative movement to “open the black box,” so to speak, of proprietary catastrophe models. Spurred in part by major catastrophes that brought an increased awareness of model limitations and uncertainties, and partly by Solvency II and other regulatory directives on the need to fully understand and internally validate third-party or open source modeling tools, companies are seeking more information on the inner workings of models. This includes more visibility into hazard and vulnerability data, information on the quantity and quality of claims data used to build the model, and a thorough understanding of key assumptions and limitations, among many other issues.
AIR has been a strong and steadfast supporter of open and transparent data standards, starting with our introduction of the UNICEDE® data format more than 25 years ago. We believed then, as we do now, that inefficiency in the data transfer process is a major concern in the industry. Open standards facilitate the collection of high-quality exposure data, enable accurate and transparent data exchange throughout the insurance value chain, encourage innovation and deeper analytics, and promote interoperability between platforms. In addition to UNICEDE, our open data formats include CEDE™ and the Verisk Cyber Exposure Data Standard. AIR was a key supporter of the development and adoption of the ACORD data standard. We are actively supporting the Simplitium Open Exposure Database (OED) by freely providing our CEDE format as the database structure for models running on Oasis LMF, an open source, nonprofit loss modeling framework. And later this year, we will be publishing the CEDE data schema publicly, which will make it the first open source database format from a modeling firm. We hope this will help streamline risk management practices and encourage model development from academic institutions and smaller organizations, especially to fill gaps in current model coverage.
In addition, over the past several years we have taken a closer look at how we approach model documentation to provide richer insights and more transparency, not only increasing the amount of detail we offer but also using client feedback to improve how we present the information. We also place a concerted emphasis on how we communicate and prepare our clients for model change, as we understand the tremendous impact it can have throughout an enterprise. Within our Touchstone® platform, the Hazard Analytics Module offers location-level event intensities (LLEI) for historical and simulated events, allowing companies to do a deeper model validation at a more granular level. For the damage component, AIR provides extensive details on the relative vulnerabilities of different construction/occupancy class combinations.
Next up is flexibility, which is a broad concept to be sure, but in this case we’ll specify it to mean the ability to modify an existing catastrophe model. There are many reasons companies may want to do this. Based on claims experience or internal research, a company may believe that a model is overstating or understating risk for certain types of events, specific regions, lines of business, construction classes, etc. Or a company may want to capture a source of loss (for example, a secondary peril or a coverage) that is not explicitly modeled. Our perspective is that we want to empower our clients to truly own their risk. This means we are committed to delivering the flexibility that helps each company use their unique insight and experience in making decisions that make sense for their business. In Touchstone, users can modify losses by geography, event parameters, and even by individual events or for individual locations. Ground-up losses can also be adjusted for primary exposure characteristics such as occupancy, construction, year built, and number of stories, which essentially gives users the ability to modify the vulnerability component.
Beyond loss modification, we are hearing increasing demand from our clients for the ability to modify individual modeling components. We will explore additional ways to allow for the adjustment of event frequency and severity and vulnerability functions in Touchstone, and even to bring in custom vulnerability functions. An important potential application of hazard modification is to analyze the impact of climate change, which is something we’re actively exploring, in addition to the possibility of providing forward-looking climate-conditioned catalogs.
On Open Platforms
Finally, I want to address open platforms, which has been a trending topic in the industry for several years now. While AIR models continually expand to different countries and perils, there are still gaps in model coverage that may leave companies underprepared for high losses or hinder them from growing their business. We are seeing an increased demand to support third-party models to cover non-modeled risk or to enable multi-modeling, which is not only a scientifically rigorous way to address the uncertainty from using any single view of risk but is also a catalyst for innovation and healthy competition in the market.
Our perspective on the matter is simple. We recognize that no single model, AIR's or any other, represents "the answer." Rather, each model is a best formulation based on the culmination of current knowledge. However, there may be scientifically plausible alternatives to various elements of a given model, and we encourage companies to continually scrutinize and challenge the models they use.
Touchstone was architected to be an open platform with open data standards, and we allow users to incorporate third-party models and data to provide risk assessment capabilities where probabilistic models do not yet exist, or to offer alternative perspectives to existing AIR models. Last year, we took this concept one step further with the introduction of Model BuilderTM to simplify and streamline the model development process. Insurers, reinsurers, and brokers can leverage the expertise of their own scientists and engineers to build and distribute custom models, while academic, NGO, and other commercial entities can create their own licensable products. The custom models can be deployed in Touchstone along with AIR models to take advantage of AIR’s sophisticated financial module, for a seamless multi-modeling approach. Our vision is for Touchstone to connect model developers and consumers and provide a centralized repository to share custom models.
Where we pause in the open platform discussion is when we come upon the desire for the ability to “mix and match” model components, such as the hazard module from one vendor with the damage module of another. Model development requires a careful and intensive calibration process to ensure that the components work together to result in robust and defensible loss estimates. For example, if one set of damage functions that rightly reflect not just engineering expertise but also loss adjustment practices is replaced by an alternative set that have not been so calibrated, the resulting losses will not make sense. Similarly, if the model's damage functions have been calibrated to one hurricane wind field formulation, model output will not be useful if another wind field formulation is substituted, unless the damage functions, too, are recalibrated.
Transparency, flexibility, and openness are new standards propelling the industry toward innovation and more robust risk management. While I believe this to be one of the most exciting and meaningful trends in (re)insurance, one that we are delighted to be a part of, I wanted to point out that a core responsibility (and challenge) is to maintain a scientifically credible, internally consistent view of risk.
A key element of ensuring that adjusting a model or implementing a custom view of risk is beneficial—and not counterproductive—is a commitment to model education. The user must have a thorough understanding of a model’s data sources, its physical and statistical underpinnings, how its components work together—as well as an intuitive sense for the model as a whole. Many people are surprised to learn that AIR spends up to half the time used for developing a model on its validation. Much more of the responsibility of validating a modified or custom model will be placed on the model user.
This brings up a related topic regarding the transfer of risk. Catastrophe modeling has created a common currency to enable risk transfer, as all parties have a shared understanding of its value. Modified or custom models may degrade the confidence in the currency, so users need to clearly document and justify their view of risk to internal and external stakeholders, including regulators. I don’t point this out to discourage innovation or competition, but it is an important caveat, and one that requires a thoughtful and measured approach from both model providers and model users.
I look forward to seeing how the industry continues to evolve, and can assure you that AIR is committed to evolving with you.