EXPOSURE
The RMS Magazine with a unique perspective and clear mission “... to provide insight and analysis to help insurance and risk professionals innovate, adapt, and deliver.”
Insurance: The next 10 years
Mohsen Rahnama, Cihan Biyikoglu and Moe Khosravy of RMS look to 2029, consider the…
Times Square, New York City
Tag: RISK MODELING
Location, Location, Location: A New Era in Data Resolution
February 11, 2021The insurance industry has reached a transformational point in its ability to accurately understand the details of exposure at risk. It is the point at which three fundamental components of exposure management are coming together to enable (re)insurers to systematically quantify risk at the location level: the availability of high-resolution location data, access to the technology to capture that data and advances in modeling capabilities to use that data. Data resolution at the individual building level has increased considerably in recent years, including the use of detailed satellite imagery, while advances in data sourcing technology have provided companies with easier access to this more granular information. In parallel, the evolution of new innovations, such as RMS® High Definition Models™ and the transition to cloud-based technologies, has facilitated a massive leap forward in the ability of companies to absorb, analyze and apply this new data within their actuarial and underwriting ecosystems. Quantifying Risk Uncertainty “Risk has an inherent level of uncertainty,” explains Mohsen Rahnama, chief modeling officer at RMS. “The key is how you quantify that uncertainty. No matter what hazard you are modeling, whether it is earthquake, flood, wildfire or hurricane, there are assumptions being made. These catastrophic perils are low-probability, high-consequence events as evidenced, for example, by the 2017 and 2018 California wildfires or Hurricane Katrina in 2005 and Hurricane Harvey in 2017. For earthquake, examples include Tohoku in 2011, the New Zealand earthquakes in 2010 and 2011, and Northridge in 1994. For this reason, risk estimation based on an actuarial approach cannot be carried out for these severe perils; physical models based upon scientific research and event characteristic data for estimating risk are needed.” A critical element in reducing uncertainty is a clear understanding of the sources of uncertainty from the hazard, vulnerability and exposure at risk. “Physical models, such as those using a high-definition approach, systematically address and quantify the uncertainties associated with the hazard and vulnerability components of the model,” adds Rahnama. “There are significant epistemic (also known as systematic) uncertainties in the loss results, which users should consider in their decision-making process. This epistemic uncertainty is associated with a lack of knowledge. It can be subjective and is reducible with additional information.” What are the sources of this uncertainty? For earthquake, there is uncertainty about the ground motion attenuation functions, soil and geotechnical data, the size of the events, or unknown faults. Rahnama explains: “Addressing the modeling uncertainty is one side of the equation. Computational power enables millions of events and more than 50,000 years of simulation to be used, to accurately capture the hazard and reduce the epistemic uncertainty. Our findings show that in the case of earthquakes the main source of uncertainty for portfolio analysis is ground motion; however, vulnerability is the main driver of uncertainty for a single location.” The quality of the exposure data as the input to any mathematical models is essential to assess the risk accurately and reduce the loss uncertainty. However, exposure could represent the main source of loss uncertainty, especially when exposure data is provided in aggregate form. Assumptions can be made to disaggregate exposure using other sources of information, which helps to some degree reduce the associated uncertainty. Rahnama concludes, “Therefore, it is essential in order to minimize the uncertainty related to exposure to try to get location-level information about the exposure, in particular for the region with the potential of liquification for earthquake or for high-gradient hazard such as flood and wildfire.” A critical element in reducing that uncertainty, removing those assumptions and enhancing risk understanding is combining location-level data and hazard information. That combination provides the data basis for quantifying risk in a systematic way. Understanding the direct correlation between risk or hazard and exposure requires location-level data. The potential damage caused to a location by flood, earthquake or wind will be significantly influenced by factors such as first-floor elevation of a building, distance to fault lines or underlying soil conditions through to the quality of local building codes and structural resilience. And much of that granular data is now available and relatively easy to access. “The amount of location data that is available today is truly phenomenal,” believes Michael Young, vice president of product management at RMS, “and so much can be accessed through capabilities as widely available as Google Earth. Straightforward access to this highly detailed satellite imagery means that you can conduct desktop analysis of individual properties and get a pretty good understanding of many of the building and location characteristics that can influence exposure potential to perils such as wildfire.” Satellite imagery is already a core component of RMS model capabilities, and by applying machine learning and artificial intelligence (AI) technologies to such images, damage quantification and differentiation at the building level is becoming a much more efficient and faster undertaking — as demonstrated in the aftermath of Hurricanes Laura and Delta. “Within two days of Hurricane Laura striking Louisiana at the end of August 2020,” says Rahnama, “we had been able to assess roof damage to over 180,000 properties by applying our machine-learning capabilities to satellite images of the affected areas. We have ‘trained’ our algorithms to understand damage degree variations and can then superimpose wind speed and event footprint specifics to group the damage degrees into different wind speed ranges. What that also meant was that when Hurricane Delta struck the same region weeks later, we were able to see where damage from these two events overlapped.” The Data Intensity of Wildfire Wildfire by its very nature is a data-intensive peril, and the risk has a steep gradient where houses in the same neighborhood can have drastically different risk profiles. The range of factors that can make the difference between total loss, partial loss and zero loss is considerable, and to fully grasp their influence on exposure potential requires location-level data. The demand for high-resolution data has increased exponentially in the aftermath of recent record-breaking wildfire events, such as the series of devastating seasons in California in 2017-18, and unparalleled bushfire losses in Australia in 2019-20. Such events have also highlighted myriad deficiencies in wildfire risk assessment including the failure to account for structural vulnerabilities, the inability to assess exposure to urban conflagrations, insufficient high-resolution data and the lack of a robust modeling solution to provide insight about fire potential given the many years of drought. Wildfires in 2017 devastated the town of Paradise, California In 2019, RMS released its U.S. Wildfire HD Model, built to capture the full impact of wildfire at high resolution, including the complex behaviors that characterize fire spread, ember accumulation and smoke dispersion. Able to simulate over 72 million wildfires across the contiguous U.S., the model creates ultrarealistic fire footprints that encompass surface fuels, topography, weather conditions, moisture and fire suppression measures. “To understand the loss potential of this incredibly nuanced and multifactorial exposure,” explains Michael Young, “you not only need to understand the probability of a fire starting but also the probability of an individual building surviving. “If you look at many wildfire footprints,” he continues, “you will see that sometimes up to 60 percent of buildings within that footprint survived, and the focus is then on what increases survivability — defensible space, building materials, vegetation management, etc. We were one of the first modelers to build mitigation factors into our model, such as those building and location attributes that can enhance building resilience.” Moving the Differentiation Needle In a recent study by RMS and the Center for Insurance Policy Research, the Insurance Institute for Business and Home Safety and the National Fire Protection Association, RMS applied its wildfire model to quantifying the benefits of two mitigation strategies — structural mitigation and vegetation management — assessing hypothetical loss reduction benefits in nine communities across California, Colorado and Oregon. Young says: “By knowing what the building characteristics and protection measures are within the first 5 feet and 30 feet at a given property, we were able to demonstrate that structural modifications can reduce wildfire risk up to 35 percent, while structural and vegetation modifications combined can reduce it by up to 75 percent. This level of resolution can move the needle on the availability of wildfire insurance as it enables development of robust rating algorithms to differentiate specific locations — and means that entire neighborhoods don’t have to be non-renewed.” “By knowing what the building characteristics and protection measures are within the first 5 feet and 30 feet at a given property, we were able to demonstrate that structural modifications can reduce wildfire risk up to 35 percent, while structural and vegetation modifications combined can reduce it by up to 75 percent” Michael Young, RMS While acknowledging that modeling mitigation measures at a 5-foot resolution requires an immense granularity of data, RMS has demonstrated that its wildfire model is responsive to data at that level. “The native resolution of our model is 50-meter cells, which is a considerable enhancement on the zip-code level underwriting grids employed by some insurers. That cell size in a typical suburban neighborhood encompasses approximately three-to-five buildings. By providing the model environment that can utilize information within the 5-to-30-foot range, we are enabling our clients to achieve the level of data fidelity to differentiate risks at that property level. That really is a potential market game changer.” Evolving Insurance Pricing It is not hyperbolic to suggest that being able to combine high-definition modeling with high-resolution data can be market changing. The evolution of risk-based pricing in New Zealand is a case in point. The series of catastrophic earthquakes in the Christchurch region of New Zealand in 2010 and 2011 provided a stark demonstration of how insufficient data meant that the insurance market was blindsided by the scale of liquefaction-related losses from those events. “The earthquakes showed that the market needed to get a lot smarter in how it approached earthquake risk,” says Michael Drayton, consultant at RMS, “and invest much more in understanding how individual building characteristics and location data influenced exposure performance, particularly in relation to liquefaction. “To get to grips with this component of the earthquake peril, you need location-level data,” he continues. “To understand what triggers liquefaction, you must analyze the soil profile, which is far from homogenous. Christchurch, for example, sits on an alluvial plain, which means there are multiple complex layers of silt, gravel and sand that can vary significantly from one location to the next. In fact, across a large commercial or industrial complex, the soil structure can change significantly from one side of the building footprint to the other.” Extensive building damage in downtown Christchurch, New Zealand after 2011 earthquakeThe aftermath of the earthquake series saw a surge in soil data as teams of geotech engineers conducted painstaking analysis of layer composition. With multiple event sets to use, it was possible to assess which areas suffered soil liquefaction and from which specific ground-shaking intensity. “Updating our model with this detailed location information brought about a step-change in assessing liquefaction exposures. Previously, insurers could only assess average liquefaction exposure levels, which was of little use where you have highly concentrated risks in specific areas. Through our RMS® New Zealand Earthquake HD Model, which incorporates 100-meter grid resolution and the application of detailed ground data, it is now possible to assess liquefaction exposure potential at a much more localized level.” “Through our RMS® New Zealand Earthquake HD model, which incorporates 100-meter grid resolution and the application of detailed ground data, it is now possible to assess liquefaction exposure potential at a much more localized level” — Michael Drayton, RMS This development represents a notable market shift from community to risk-based pricing in New Zealand. With insurers able to differentiate risks at the location level, this has enabled companies such as Tower Insurance to more accurately adjust premium levels to reflect risk to the individual property or area. In its annual report in November 2019, Tower stated: “Tower led the way 18 months ago with risk-based pricing and removing cross-subsidization between low- and high-risk customers. Risk-based pricing has resulted in the growth of Tower’s portfolio in Auckland while also reducing exposure to high-risk areas by 16 percent. Tower’s fairer approach to pricing has also allowed the company to grow exposure by 4 percent in the larger, low-risk areas like Auckland, Hamilton, and Taranaki.” Creating the Right Ecosystem The RMS commitment to enable companies to put high-resolution data to both underwriting and portfolio management use goes beyond the development of HD Models™ and the integration of multiple layers of location-level data. Through the launch of RMS Risk Intelligence™, its modular, unified risk analytics platform, and the Risk Modeler™ application, which enables users to access, evaluate, compare and deploy all RMS models, the company has created an ecosystem built to support these next-generation data capabilities. Deployed within the Cloud, the ecosystem thrives on the computational power that this provides, enabling proprietary and tertiary data analytics to rapidly produce high-resolution risk insights. A network of applications — including the ExposureIQ™ and SiteIQ™ applications and Location Intelligence API — support enhanced access to data and provide a more modular framework to deliver that data in a much more customized way. “Because we are maintaining this ecosystem in the Cloud,” explains Michael Young, “when a model update is released, we can instantly stand that model side-by-side with the previous version. As more data becomes available each season, we can upload that new information much faster into our model environment, which means our clients can capitalize on and apply that new insight straightaway.” Michael Drayton adds: “We’re also offering access to our capabilities in a much more modular fashion, which means that individual teams can access the specific applications they need, while all operating in a data-consistent environment. And the fact that this can all be driven through APIs means that we are opening up many new lines of thought around how clients can use location data.” Exploring What Is Possible There is no doubt that the market is on the cusp of a new era of data resolution — capturing detailed hazard and exposure and using the power of analytics to quantify the risk and risk differentiation. Mohsen Rahnama believes the potential is huge. “I foresee a point in the future where virtually every building will essentially have its own social-security-like number,” he believes, “that enables you to access key data points for that particular property and the surrounding location. It will effectively be a risk score, including data on building characteristics, proximity to fault lines, level of elevation, previous loss history, etc. Armed with that information — and superimposing other data sources such as hazard data, geological data and vegetation data — a company will be able to systematically price risk and assess exposure levels for every asset up to the portfolio level.” “The only way we can truly assess this rapidly changing risk is by being able to systematically evaluate exposure based on high-resolution data and advanced modeling techniques that incorporate building resilience and mitigation measures” — Mohsen Rahnama, RMS Bringing the focus back to the here and now, he adds, the expanding impacts of climate change are making the need for this data transformation a market imperative. “If you look at how many properties around the globe are located just one meter above sea level, we are talking about trillions of dollars of exposure. The only way we can truly assess this rapidly changing risk is by being able to systematically evaluate exposure based on high-resolution data and advanced modeling techniques that incorporate building resilience and mitigation measures. How will our exposure landscape look in 2050? The only way we will know is by applying that data resolution underpinned by the latest model science to quantify this evolving risk.”
A Model Operation
September 05, 2018EXPOSURE explores the rationale, challenges and benefits of adopting an outsourced model function Business process outsourcing has become a mainstay of the operational structure of many organizations. In recent years, reflecting new technologies and changing market dynamics, the outsourced function has evolved significantly to fit seamlessly within existing infrastructure. On the modeling front, the exponential increase in data coupled with the drive to reduce expense ratios while enhancing performance levels is making the outsourced model proposition an increasingly attractive one. The Business Rationale The rationale for outsourcing modeling activities spans multiple possible origin points, according to Neetika Kapoor Sehdev, senior manager at RMS. “Drivers for adopting an outsourced modeling strategy vary significantly depending on the company itself and their specific ambitions. It may be a new startup that has no internal modeling capabilities, with outsourcing providing access to every component of the model function from day one.” There is also the flexibility that such access provides, as Piyush Zutshi, director of RMS Analytical Services points out. “That creates a huge value-add in terms of our catastrophe response capabilities — knowing that we are able to report our latest position has made a big difference on this front” Judith Woo Starstone “In those initial years, companies often require the flexibility of an outsourced modeling capability, as there is a degree of uncertainty at that stage regarding potential growth rates and the possibility that they may change track and consider alternative lines of business or territories should other areas not prove as profitable as predicted.” Another big outsourcing driver is the potential to free up valuable internal expertise, as Sehdev explains. “Often, the daily churn of data processing consumes a huge amount of internal analytical resources,” she says, “and limits the opportunities for these highly skilled experts to devote sufficient time to analyzing the data output and supporting the decision-making process.” This all-too-common data stumbling block for many companies is one that not only affects their ability to capitalize fully on their data, but also to retain key analytical staff. “Companies hire highly skilled analysts to boost their data performance,” Zutshi says, “but most of their working day is taken up by data crunching. That makes it extremely challenging to retain that caliber of staff as they are massively overqualified for the role and also have limited potential for career growth.” Other reasons for outsourcing include new model testing. It provides organizations with a sandbox testing environment to assess the potential benefits and impact of a new model on their underwriting processes and portfolio management capabilities before committing to the license fee. The flexibility of outsourced model capabilities can also prove critical during renewal periods. These seasonal activity peaks can be factored into contracts to ensure that organizations are able to cope with the spike in data analysis required as they reanalyze portfolios, renew contracts, add new business and write off old business. “At RMS Analytical Services,” Zutshi explains, “we prepare for data surge points well in advance. We work with clients to understand the potential size of the analytical spike, and then we add a factor of 20 to 30 percent to that to ensure that we have the data processing power on hand should that surge prove greater than expected.” Things to Consider Integrating an outsourced function into existing modeling processes can prove a demanding undertaking, particularly in the early stages where companies will be required to commit time and resources to the knowledge transfer required to ensure a seamless integration. The structure of the existing infrastructure will, of course, be a major influencing factor in the ease of transition. “There are those companies that over the years have invested heavily in their in-house capabilities and developed their own systems that are very tightly bound within their processes,” Sehdev points out, “which can mean decoupling certain aspects is more challenging. For those operations that run much leaner infrastructures, it can often be more straightforward to decouple particular components of the processing.” RMS Analytical Services has, however, addressed this issue and now works increasingly within the systems of such clients, rather than operating as an external function. “We have the ability to work remotely, which means our teams operate fully within their existing framework. This removes the need to decouple any parts of the data chain, and we can fit seamlessly into their processes.” This also helps address any potential data transfer issues companies may have, particularly given increasingly stringent information management legislation and guidelines. There are a number of factors that will influence the extent to which a company will outsource its modeling function. Unsurprisingly, smaller organizations and startup operations are more likely to take the fully outsourced option, while larger companies tend to use it as a means of augmenting internal teams — particularly around data engineering. RMS Analytical Services operate various different engagement models. Managed services are based on annual contracts governed by volume for data engineering and risk analytics. On-demand services are available for one-off risk analytics projects, renewals support, bespoke analysis such as event response, and new IP adoption. “Modeler down the hall” is a third option that provides ad hoc work, while the firm also offers consulting services around areas such as process optimization, model assessment and transition support. Making the Transition Work Starstone Insurance, a global specialty insurer providing a diversified range of property, casualty and specialty insurance to customers worldwide, has been operating an outsourced modeling function for two and a half years. “My predecessor was responsible for introducing the outsourced component of our modeling operations,” explains Judith Woo, head of exposure management at Starstone. “It was very much a cost-driven decision as outsourcing can provide a very cost-effective model.” The company operates a hybrid model, with the outsourced team working on most of the pre- and post-bind data processing, while its internal modeling team focuses on the complex specialty risks that fall within its underwriting remit. “The volume of business has increased over the years as has the quality of data we receive,” she explains. “The amount of information we receive from our brokers has grown significantly. A lot of the data processing involved can be automated and that allows us to transfer much of this work to RMS Analytical Services.” On a day-to-day basis, the process is straightforward, with the Starstone team uploading the data to be processed via the RMS data portal. The facility also acts as a messaging function with the two teams communicating directly. “In fact,” Woo points out, “there are email conversations that take place directly between our underwriters and the RMS Analytical Service team that do not always require our modeling division’s input.” However, reaching this level of integration and trust has required a strong commitment from Starstone to making the relationship work. “You are starting to work with a third-party operation that does not understand your business or its data processes. You must invest time and energy to go through the various systems and processes in detail,” she adds, “and that can take months depending on the complexity of the business. “You are essentially building an extension of your team, and you have to commit to making that integration work. You can’t simply bring them in, give them a particular problem and expect them to solve it without there being the necessary knowledge transfer and sharing of information.” Her internal modeling team of six has access to an outsourced team of 26, she explains, which greatly enhances the firm’s data-handling capabilities. “With such a team, you can import fresh data into the modeling process on a much more frequent basis, for example. That creates a huge value-add in terms of our catastrophe response capabilities — knowing that we are able to report our latest position has made a big difference on this front.” Creating a Partnership As with any working partnership, the initial phases are critical as they set the tone for the ongoing relationship. “We have well-defined due diligence and transition methodologies,” Zutshi states. “During the initial phase, we work to understand and evaluate their processes. We then create a detailed transition methodology, in which we define specific data templates, establish monthly volume loads, lean periods and surge points, and put in place communication and reporting protocols.” At the end, both parties have a full documented data dictionary with business rules governing how data will be managed, coupled with the option to choose from a repository of 1,000+ validation rules for data engineering. This is reviewed on a regular basis to ensure all processes remain aligned with the practices and direction of the organization. “Often, the daily churn of data processing consumes a huge amount of internal analytical resources and limits the opportunities to devote sufficient time to analyzing the data output” — Neetika Kapoor Sehdev, RMS Service level agreements (SLAs) also form also form a central tenet of the relationship plus stringent data compliance procedures. “Robust data security and storage is critical,” says Woo. “We have comprehensive NDAs [non-disclosure agreements] in place that are GDPR compliant to ensure that the integrity of our data is maintained throughout. We also have stringent SLAs in place to guarantee data processing turnaround times. Although, you need to agree on a reasonable time period reflecting the data complexity and also when it is delivered.” According to Sehdev, most SLAs that the analytical team operates require a 24-hour data turnaround rising to 48-72 hours for more complex data requirements, but clients are able to set priorities as needed. “However, there is no point delivering on turnaround times,” she adds, “if the quality of the data supplied is not fit for purpose. That’s why we apply a number of data quality assurance processes, which means that our first-time accuracy level is over 98 percent.” The Value-Add Most clients of RMS Analytical Services have outsourced modeling functions to the division for over seven years, with a number having worked with the team since it launched in 2004. The decision to incorporate their services is not taken lightly given the nature of the information involved and the level of confidence required in their capabilities. “The majority of our large clients bring us on board initially in a data-engineering capacity,” explains Sehdev. “It’s the building of trust and confidence in our ability, however, that helps them move to the next tranche of services.” The team has worked to strengthen and mature these relationships, which has enabled them to increase both the size and scope of the engagements they undertake. “With a number of clients, our role has expanded to encompass account modeling, portfolio roll-up and related consulting services,” says Zutshi. “Central to this maturing process is that we are interacting with them daily and have a dedicated team that acts as the primary touch point. We’re also working directly with the underwriters, which helps boost comfort and confidence levels. “For an outsourced model function to become an integral part of the client’s team,” he concludes, “it must be a close, coordinated effort between the parties. That’s what helps us evolve from a standard vendor relationship to a trusted partner.”
