Geocode match level, explained for cat modelling

Suburban homes along the edge of a neighborhood, overlooking an open field under a partly cloudy sky.

What "match level" tells you

Every geocoded location in a submission carries a match level: rooftop, street/parcel interpolation, ZIP+4, census block, ZIP centroid, or worse. It's a flag your geocoding engine attaches when it tries to resolve an address string to a lat/long, and it tells you how it got there, not how close it got.

Rooftop match means the engine found the structure itself, usually from a parcel database or address point file, and placed the pin on (or very near) the building. Street-level match means it found the street segment and interpolated a position along it based on the address range, which is fine on a straight suburban block and much rougher on a long rural route or a cul-de-sac. Below that, ZIP+4 and ZIP centroid matches are placing the location at the middle of an area that can run from a few blocks to several square miles, which is a real problem when you're pricing cat peril that varies sharply over short distances: flood depth, distance to coast, brush interface, hail swath edges.

The match level sits on the submission as metadata. It doesn't tell you whether the rooftop match landed on the right rooftop, whether the structure is still standing, or whether something got built on the lot since the address file was last refreshed.

Rooftop vs street-level: what changes in the loss estimate

For single-location, high-TIV risks, the gap between rooftop and street-level match is where a chunk of basis risk lives. A street-interpolated point can sit a parcel or two off from the real structure. For wind and quake that's usually close enough that it washes out in the portfolio. For flood, wildfire WUI, and storm surge, a few hundred feet can move a location in or out of a hazard zone entirely, which flips the loss cost for that risk and, at scale, shifts the tail of the distribution for the whole treaty.

This is also why match level correlates with model accuracy without being the same thing as it. A portfolio that's 95% rooftop-matched still has geocoding resolution problems if the underlying address points are five years stale, because the match level describes precision of placement, not currency of what's standing on the ground. You can have a perfect rooftop hit on a parcel that was a single-family home when the address database was built and is now a six-unit infill rebuild. The pin is exact. The exposure behind it is wrong.

Where the schedule runs out of answers

Treaty submissions are built from self-reported schedules, and schedules are only as good as what the cedant's underwriting file captured at binding, updated at whatever cadence renewals happen to run on. New construction, rebuilds after a prior event, and informal additions rarely make it back into that file before the next renewal cycle. None of that shows up as a geocoding problem. It shows up as a gap between the match level on the record and what's standing on the ground, and no amount of re-geocoding the address string fixes it, because the address string was never wrong. The building behind it changed.

Exposure Mapping is built to sit under that geocoded schedule as a second layer: a current, image-derived read of built footprints across the territory, refreshed on an annual cadence from high-resolution satellite imagery, so the structure count and rough form your model works from match what's on the ground this year, not what was on file at last renewal.

A few things to check before you trust a match-level summary

  • Ask what "rooftop" means in your vendor's pipeline. Some engines call a tight street interpolation "rooftop" if it falls within a small radius of a parcel centroid. The label isn't standardized across providers.
  • Look at match level distribution, not just the average. A portfolio can be 90% rooftop-matched and still have its worst basis risk sitting in the remaining 10%, concentrated in exactly the rural or newly developed areas where hazard data is thinnest.
  • Treat match level and data currency as separate questions. One tells you where the pin sits. The other tells you whether the thing under the pin is still the thing the schedule describes.

If you're staring down a renewal and want a structure-level read of what's actually built across the treaty before you lean on the schedule alone, that's the gap this site exists to close.

Get started

← Back to the blog