James runs the Meridian portfolio through the model and gets an AAL of USD 4.2M. The underwriter asks what it means. James explains: "It's the expected annual loss across the whole portfolio, what you'd expect to pay out on average each year if you wrote this account indefinitely."
The underwriter nods. Then she asks: "How confident are you in it?"
James pauses. He knows the Tokyo and Santiago locations have confirmed building data and precise coordinates. He also knows that six of the Indian locations have approximate postcodes, two have unknown construction class, and one, the Mumbai warehouse, was geocoded to a point in central India rather than Mumbai.
"It depends on the exposure data," James says. "The number is only as good as what we put in."
Exposure data is the complete description of what is at risk. For a property cat model, it is a structured record of every insured location in a portfolio, where each location sits, what the building is, and what financial value is exposed. The cat model cannot assess risk without it. The hazard module knows the intensity of every possible event at every geographic point on earth. But it has no way of knowing whether there is a building at any given point, what that building is made of, or what it is worth. That information comes entirely from the exposure data.
The relationship is foundational, not additive. A cat model does not supplement your exposure data with its own knowledge. It processes exactly what you give it, applies vulnerability and financial parameters, and returns a loss estimate. If the exposure data is wrong, the loss estimate is wrong, in a way that is often impossible to detect by looking at the number itself. A country-centroid coordinate produces a plausible-looking loss estimate, silently, with no error flag.
A cat model is not a machine that corrects bad data. It is a machine that amplifies what you give it. The only protection against bad exposure data is knowing what good data looks like before you run the model. That knowledge is what the Intermediate course's entire first module builds. This lesson gives you the conceptual foundation to start.
Exposure data does not originate in the insurance industry. It begins with the insured, a company that owns the assets being covered. That company maintains a property register for its own purposes: asset management, depreciation, maintenance. The fields it contains reflect what the company's facilities team needs, not what a cat model needs. By the time the data reaches the cat modelling team, it has passed through at least two pairs of hands, each step an opportunity for gaps to appear.
Maintains a property register in an asset management system built for accounting, not cat modelling. Contains: property name, mailing address, book value, year acquired. Typically missing: construction class, year built, replacement cost, precise coordinates.
Extracts insurance-relevant fields and assembles the Schedule of Values (SOV). Fills gaps where possible, leaves blanks where not. Submits as part of the placement pack. Most data quality problems are introduced at this step, not from negligence, but because the information simply does not exist in the source system.
Receives the SOV, assesses data quality, applies geocoding, fills remaining gaps with model defaults, runs the model, and presents results with appropriate caveats. The analyst is the model's only quality control.
The Schedule of Values (SOV) is the structured list of insured locations that forms the model's primary input. In practice it is almost always a spreadsheet - one row per location, one column per field. A complete SOV contains four categories of information, each corresponding to one component of the cat model's calculation chain.
The Intermediate course opens with three full lessons on the SOV: what every field should contain, where common gaps come from, how to QA the data before running the model, and how to handle the gaps that cannot be fixed. This lesson gives you the conceptual picture of what the SOV is. Intermediate Module 1 builds the professional skill of working with one.
Once the hazard module calculates the physical intensity at a location, the vulnerability module converts that intensity into a damage ratio. The conversion depends on the physical characteristics of the building. These are summarised by the acronym COPE, and they are the most frequently missing or incorrect element in a real-world SOV. The Intermediate course covers COPE data collection in full at Lesson 1.2. Here, you need to understand what each letter means and why it matters.
The structural system and primary building material, RC Shear Wall, Steel Frame, Masonry, Wood Frame. The single most important vulnerability variable for earthquake and hurricane. A pre-1980 masonry building and a 2015 reinforced concrete building hit by the same intensity earthquake can produce damage ratios that differ by a factor of four or more. Construction class and year built together determine which damage function the vulnerability module applies.
What the building is used for, commercial office, retail, industrial warehouse, healthcare, hotel, residential. Occupancy affects both the contents at risk and the construction norms associated with the building type. A hospital and a warehouse with identical construction can have very different vulnerability profiles, hospitals contain high-value equipment and have stricter structural requirements; warehouses are typically lower-value and more tolerant of damage before closure.
Fire suppression, alarm systems, and physical loss-prevention measures. Protection is most relevant to fire risk modelling and, increasingly, wildfire vulnerability. For earthquake and wind cat modelling, protection fields are less directly influential, though flood barriers and defensible space around structures can be significant modifiers for flood and wildfire respectively.
The financial values at risk, replacement cost, contents value, business interruption exposure. The damage ratio from the vulnerability module is applied to TIV to produce the loss. Getting the financial value wrong is directly proportional to getting the loss wrong, a 20% TIV understatement produces a 20% loss understatement across every event in the catalogue.
Total Insured Value is the financial multiplier in the cat model's loss calculation. Every damage ratio the vulnerability module produces is multiplied by TIV to generate a financial loss. A 30% damage ratio applied to a USD 10M TIV produces a USD 3M loss. The same ratio applied to a USD 20M TIV produces USD 6M. TIV errors are proportional and directional, they scale the loss estimate uniformly up or down across all events in the catalogue.
TIV must reflect replacement cost, the cost to demolish a damaged structure and rebuild it to current standards at current labour and material prices. Two other figures are commonly submitted in SOVs and both are wrong:
Geocoding is the process of converting a text address into a coordinate pair that places the location on the hazard grid. The hazard module assigns intensity values by reading from a spatial grid. The coordinate pair is the link between the address in the SOV and the hazard score the model assigns. Every downstream calculation, damage ratio, financial loss, return period estimate, is built on that hazard score. A wrong coordinate assigns the wrong score to everything that follows.
The chain is simple and unforgiving: address → coordinate → hazard grid cell → intensity → damage ratio → loss. An error anywhere in that chain propagates forward silently. The model produces a plausible-looking number. There is no error flag.
Not all geocodes are equally useful. The industry uses a hierarchy from most to least precise:
India's country centroid is approximately 20.59°N, 78.96°E, in Madhya Pradesh, roughly 1,400 km from Mumbai. Australia's is 25.3°S, 133.8°E, in the middle of the outback, 2,700 km from Sydney. Any location assigned a country centroid must be identified and fixed before modelling. It is not a conservative approximation, it is a wrong answer presented as a right one.
The complete description of what is at risk, where, what, and how much. The primary input to the cat model and the primary determinant of output reliability. The model amplifies whatever it receives.
The structured spreadsheet of insured locations forming the model's primary input. One row per location, assembled by the broker from the insured's property register.
Construction, Occupancy, Protection, Exposure. The four physical building data categories that determine vulnerability. Construction class and year built are typically the most consequential for earthquake and hurricane.
The maximum financial value at a single location, the multiplier in the loss calculation. Must reflect replacement cost. A proportional error in TIV produces a proportional error in every loss estimate for that location.
The cost to rebuild a structure to current standards at current prices. The correct basis for TIV. Distinct from market value (location and demand) and book value (accounting depreciation).
Converting a text address into a coordinate pair for placement on the hazard grid. The accuracy of the coordinate determines the accuracy of every downstream calculation.
The geographic centre of a country, assigned as a fallback when no address information is available. Produces a meaningless hazard score and must be corrected before any model run.
The assumption a cat model applies when a required COPE field is blank. Designed to be representative, not conservative. Applies uncertainty in both directions, not a protective margin.
Five questions · 4 of 5 correct (80%) to pass