When better models meet better data: Realizing a longstanding market vision

Working with the insurance and reinsurance market, we have seen a consensus emerge around what the market wants from catastrophe models: greater realism, higher spatial and temporal fidelity, and clarity on how losses vary across financial perspectives and contract terms.

Our customers asked for models that better reflect real-world damage patterns, account for mitigation measures, and robustly cover emerging perils that are becoming increasingly material, such as wildfire, flood, and severe convective storms.

More recently, the market has raised a second, equally important challenge: advanced models need high-quality exposure details as inputs to produce the most accurate modeled loss output. Collecting consistent, objective, and up-to-date exposure data at scale has long been one of the industry’s hardest problems.

Today, that challenge is being addressed, and we are excited to announce the launch of Moody’s new data offering within Enhanced Risk Data (ERD), called Exposure Enrichment.

What is Enhanced Risk Data and Exposure Enrichment?

Enhanced Risk Data on Moody’s Intelligent Risk Platform™ (IRP) contains hazard data layers across many perils, along with Exposure Enrichment for the United States. Exposure Enrichment provides richer building-level attributes for residential properties in the U.S. and comes fully integrated into the IRP.

Exposure Enrichment brings together AI models from both Moody’s RMS risk modeling and Moody’s CAPE property intelligence teams to build a richer list of primary and secondary modifiers for the U.S.

What’s unique about Exposure Enrichment is the data quality and also how frequently the data is updated—as and when aerial imagery is refreshed—utilizing proprietary mappings to inject exposure details directly at the point of model execution. This ensures that high-quality primary and secondary modifiers missing from the portfolio are populated before model execution, generating modeled losses with greater accuracy than a portfolio using missing, unknown inputs.

Why does improved, up-to-date input data matter more than ever?

Some building attributes tend to remain stable over time, such as occupancy, floor area, or number of stories; others can change over time and materially influence risk, such as roof age, roof condition, roof covering, the presence of solar panels, or proximity to vegetation and overhanging trees. These characteristics are difficult to capture reliably with traditional reporting processes and can quickly become outdated.

Exposure Enrichment covers both stable and changeable attributes by providing objective, up-to-date building-level characteristics, aligned directly with model inputs. This enables model users to benchmark and improve the data reported by policyholders, cedants, or brokers, helping to identify gaps, inconsistencies, or overly optimistic representations of risk.

Breakthrough in the use of high-confidence property attributes

Using observed claims experience consistently shows that certain property characteristics can have a disproportionate impact on loss. For example, wind or hail claim frequency for properties with poor or severely degraded roof conditions can be multiple times higher than for properties with roofs in good or excellent condition.

When these differences are captured objectively and fed into models with the necessary resolution to differentiate effectively, the resulting loss estimates then become more explanatory, more defensible, and more closely aligned with real-world outcomes. This is where the combination of high-definition (HD) modeling and Enhanced Risk Data’s Exposure Enrichment capability delivers its greatest value.

To illustrate, let's use an example applied to a single residential property in Florida (see Figure 1 below), which shows a sequence of snapshots, from a (year 2020) to f (year 2025).

Using these snapshots, it is possible to observe how the property’s roof deteriorated over time, was replaced in 2023, and, by 2025, the roof was surrounded by higher tree density.

The extraction of exposure attributes is achieved using state-of-the-art AI engines at scale, for nearly all U.S. residential properties, with commercial properties to follow.

Figure 1. Typical temporal evolution of exposure characteristics for a single-family dwelling.

Now, consider a second key insight: assume this property has a constant insured value of $1 million. The modeled premium for this home increased from about $1,500 in 2020 to $1,900 in 2022 (a 20% increase), then dropped to just below $1,000 in 2025 (more than 40% decrease).

This example illustrates how premiums can be adjusted over time for the same property—or how risk differentiation can lead to different premiums for two nearby homes that may look similar on paper but have very different expected average annual losses, with one being almost twice as risky as the other.

Does precise risk characterization still matter when moving from a single location to an account or a portfolio? The answer is yes, and for two main reasons:

For portfolios coded with overly optimistic (or pessimistic, though this is less frequent) exposure characteristics, adopting an objective and up-to-date view of property attributes can reveal a general low (or high) bias. Hence, yes, in this case, it really matters.

When portfolios are coded with reasonable primary characteristics such as occupancy, construction class, year built, and number of stories, leaving the so-called secondary modifiers unknown, we should typically expect that, on average, portfolio metrics do not greatly change.

This is also due to the validity of the Moody’s RMS building inventory characteristics used in the case of the ‘unknown.’ However, what can hurt a portfolio lies outside the body of the distribution: a subset of properties with poor building conditions can cause very high pain.  

A case study in data quality for hurricane risk modeling

For our second example, let’s use a portfolio of about 100,000 randomly-selected residential properties in Florida.

We will first analyze the portfolio using only the known primary characteristics, and then analyze it again using Exposure Enrichment attributes (i.e., also adding all secondary modifiers). Comparing location-level Average Annual Losses (AAL), we see some very interesting results (see Figure 2 below).

• On average, the portfolio-level AAL changes by only -0.7%, which is almost negligible. However, the surprise is that only around 13% of locations present a change within +-5%.
• About 37% of locations show a general increase in AAL (Figure 2; B), and 5.5% of locations show an AAL at least 50% higher than the case with unknown secondary modifiers.
• On the other hand, the remaining half of the portfolio shows a decrease in AAL (Figure 2; A); a large proportion of locations (21.6%) have AAL decreases ranging between 15% to 30%. The remaining 13.4% of locations show a decrease in AAL greater than 30%.

The first portfolio example assumed that all primary characteristics were known; of course, there are many examples of portfolios where only a subset of primary characteristics is known, sometimes only the occupancy is reported.

In such cases, when using the Exposure Enrichment data, modeling results can show greater variations than presented above, even at the portfolio level. In such cases, better data can really help to spot major biases.

How can this level of insight help? Well, based on our customers' feedback, this is incredibly useful!

While reinsurers explore how best to balance risk-based pricing and mutualization, improving risk estimation can help inform how to create mutuality and risk sharing in line with the company’s principles.  

For example, insurers can calibrate their policy review to incentivize increased mitigation measures for at least a subset of locations with a higher propensity to generate more frequent/higher claims (Figure 2; B). At the same time, an insurer could offer better rates to homeowners who apply mitigation measures and hence contribute less to portfolio losses (Figure 2; A).

Brokers could provide improved advice to their clients, for example, toward a better portfolio composition and risk-transfer strategy.  

And finally, reinsurers can also certainly benefit from a clearer understanding of the quality of exposure and underwriting strategies across cedants, and take adequate measures. 

Figure 2. Location-level windstorm AAL changes before and after the application of Enhanced Exposure Enrichment, across about 100,000 residential properties randomly selected in Florida.

In a time when carriers are very concerned about increasing losses and subsequent impact on earnings, having both an objective characterization of exposure characteristics and a measure of its impact on technical price provides a major competitive advantage.

By explicitly linking building characteristics and mitigation measures to modeled outcomes, reinsurers can better assess the impact of actions such as roof upgrades, vegetation management, structural improvements, or alternative policy terms such as higher or lower deductibles and limits.

All this supports more transparent pricing, more credible mitigation credits, and clearer communication across the risk transfer chain, and with regulators—grounded in objective data rather than assumptions.

Delivering on a longstanding market vision

Across perils such as North America severe convective storms, North America wildfire, and North Atlantic hurricane, Moody’s continues to invest in advancing catastrophe science while ensuring continuity and reliability for model users.

With state-of-the-art models and Exposure Enrichment working together—and with Moody’s as the only authorized provider of the mapping between enriched exposure data and Moody’s RMS catastrophe models—clients now have access to a more complete, consistent, and realistic foundation for understanding catastrophe risk.