You are running the Meridian portfolio through Touchstone for the first time. Location MGA-068, the Santiago, Chile office tower at Av. Apoquindo 3600, has good data: confirmed RC Shear Wall construction, 2015 build year, precise street-level coordinates. You hit run.
The model assigns an Annual Average Loss of USD 84,000 to this location. The building is worth USD 6.4M. That is an AAL rate of 1.31%, which, combined with the policy structure, will drive the pricing recommendation your underwriter acts on.
But where did that USD 84,000 come from? Before the model touched the exposure data or applied any financial conditions, something happened first: the model looked at the coordinates and asked, "What is the hazard at this location?" It reached into a database of simulated earthquake events, found every event in the catalogue that would produce significant shaking at this precise point in Santiago, calculated what the ground motion intensity would be for each one, and assigned a probability to each scenario.
The hazard module answers one question for every location in portfolio: given where this property sits, what physical intensity could each possible catastrophe event produce here? It does not ask whether the building is insured, how much it is worth, or how well it is built. It asks only about the physical environment, the seismic activity in the region, the wind patterns of tropical cyclones, the flood hydrology of the river basin, and translates that physical environment into a probability distribution of intensities at a specific geographic point.
The hazard module's output is a hazard surface, a spatial grid of intensity values. For earthquake, the intensity metric is typically Peak Ground Acceleration (PGA) or Spectral Acceleration (SA), the rate at which the ground accelerates during shaking, measured in fractions of gravitational acceleration. For hurricane, it is 3-second gust wind speed in mph or m/s. For flood, it is inundation depth in metres. For wildfire, it is a composite fire weather and vegetation intensity index. Each metric is calculated at specific geographic grid points across the entire model domain.
When you give the model a coordinate pair for the Santiago tower, it places that point on the hazard grid and reads off the intensity value for every simulated event in the catalogue that would affect that location. Those intensity values, hundreds or thousands of them, each with an associated annual probability, are what the vulnerability module then converts into damage ratios, and what the financial module converts into insured losses. The hazard module's output is the starting point of the entire loss estimation chain.
If the hazard module assigns the wrong intensity to a location, everything downstream is wrong, regardless of how good the exposure data or vulnerability functions are. A building placed in the wrong grid cell gets the wrong hazard score. The wrong hazard score produces the wrong damage ratio. The wrong damage ratio, multiplied by even a perfectly accurate TIV, produces a wrong loss. This is why geocoding quality, the accuracy of the coordinate that places the location on the hazard grid, is the most critical data quality issue in the entire cat modelling workflow. A perfect address in the wrong location on the grid is as useless as no address at all.
The stochastic event catalogue is the core of the hazard module. It is a large synthetic database of simulated catastrophe events, each with a defined physical description (location, magnitude, track, intensity) and an assigned annual rate (the probability of that event occurring in any given year). A typical modern cat model catalogue contains between 10,000 and 100,000 individual events per peril, representing the full distribution of possible events, from frequent, moderate-severity events to rare, extreme events well beyond anything in the modern historical record.
The catalogue is not a list of historical events. It is a synthetic set of plausible events, generated by physical simulation calibrated to historical observations. For hurricane, this means running thousands of years of simulated Atlantic hurricane seasons through a track model calibrated to historical hurricane behaviour, generating synthetic storm tracks, wind speed profiles, and landfall probabilities. For earthquake, it means constructing a probabilistic seismic hazard model from the geological record of fault locations, slip rates, and rupture characteristics, and using that model to generate synthetic earthquake scenarios across the full range of possible magnitudes and locations.
Bar length = relative hazard intensity at MGA-068. Rate = expected occurrences per year. A rate of 0.007/yr = 1-in-143 year return period.
The annual rate assigned to each synthetic event is what transforms a list of scenarios into a probability distribution. The rate is derived from the physical science of each hazard, from geological fault slip rates, from the historical frequency of tropical cyclone formation, from the observed recurrence interval of major flood events in a river basin. A rate of 0.01 per year means an expected frequency of one occurrence every 100 years, a 1-in-100 year return period. A rate of 0.002 per year means one occurrence every 500 years.
These rates are not simply read from the historical record. The historical record for most locations is too short, spanning perhaps 50–150 years of reliable observations, to provide stable frequency estimates for events with return periods of 200, 500, or 1,000 years. For these rare events, rates are derived from the physical model: fault geology, geodetic measurements of plate motion, palaeoseismological evidence of past earthquakes preserved in geological strata. The model's ability to estimate rates for events beyond the historical record is precisely its analytical value, and its primary source of uncertainty.
Each peril uses a different physical metric to express hazard intensity. These metrics are chosen because they correlate well with building damage, they are the physical quantities that most directly drive structural response. Understanding what each metric measures is important because it explains why certain exposure data fields matter more for some perils than others.
| Peril | Primary Intensity Metric | Why This Metric | Secondary Factors |
|---|---|---|---|
| Earthquake | Peak Ground Acceleration (PGA) · Spectral Acceleration (SA) | PGA drives damage to short-period structures (masonry, low-rise). SA at the building's natural period drives damage to taller structures. Both are direct inputs to structural engineering damage functions. | Soil type (amplifies or attenuates shaking) · Depth of earthquake · Distance from fault · Local site conditions (Vs30) |
| Hurricane / Tropical Cyclone | 3-second peak gust wind speed (mph or m/s) | Structural wind loading is proportional to the square of wind speed. Peak gust governs cladding and roofing damage. Sustained wind speed governs structural frame loading. Most vulnerability functions use gust speed as the primary input. | Storm forward speed · Angle of approach · Terrain roughness · Building height · Proximity to coast (surge zone) |
| Inland Flood | Inundation depth (metres above floor level) | Flood damage to building contents and fabric increases steeply with inundation depth. Depth is the most directly measurable and correlatable metric for residential and commercial flood damage functions. | Flow velocity · Flood duration · Sediment load · Salinity (marine flood) · Warning time (affects contents removal) |
| European Windstorm | Maximum 3-second gust wind speed | Same physical basis as hurricane wind damage. European extratropical windstorms produce a different track and wind field structure than tropical cyclones but the structural loading mechanism is identical. | Storm duration · Multiple wind directions · Regional building stock characteristics |
| Wildfire | Fire Weather Index (FWI) composite · Flame length · Radiant heat flux | Structure ignition probability is driven by the combination of ember exposure, radiant heat from the fire front, and direct flame contact. Multiple metrics are combined in wildfire vulnerability models. | Vegetation type and fuel load · Wind speed and direction · Structural construction materials · Defensible space |
| Cyber | Attack vector type · Propagation mechanism · Network reach | Cyber hazard has no single physical intensity metric equivalent. Severity is modelled through the combination of attack type (ransomware vs wiper vs DDoS), propagation mechanism (network vs supply chain vs cloud), and the breadth of affected systems. | Patch management maturity · Vendor market share · Backup availability · Response time |
The intensity metric determines which exposure fields are most important for a given peril. For earthquake, soil type (Vs30) can amplify or reduce PGA by a factor of 3-5, making it a critical secondary modifier. For hurricane, building height affects vulnerability because taller buildings experience higher wind speeds at elevation and have different structural periods. For flood, first floor height above grade directly determines the depth of inundation at which damage begins. When you see a field in the SOV labelled "first floor height" or "Vs30" or "number of storeys," you are looking at the exposure data that allows the vulnerability module to apply the right damage function for the hazard intensity the hazard module has calculated.
Return periods are frequently misunderstood, even by experienced practitioners. They are a way of expressing the probability of a hazard exceeding a given intensity threshold in any given year. A 1-in-100 year return period earthquake does not mean an earthquake that occurs once per century. It means an earthquake whose intensity is exceeded with a probability of 1% in any given year, which is a subtly but importantly different statement.
The distinction between a return period and a scheduled recurrence matters operationally. If a 1-in-100 year event occurs this year, the probability of it occurring again next year is still 1%, not less. The event's occurrence does not "reset" the probability or mean the region is now "safe" for 99 years. Over any 100-year period, the probability of experiencing at least one 1-in-100 year event is approximately 63%, not 100%. Two 1-in-100 year events in the same decade is not impossible, it is unlikely but mathematically consistent with the stated probability.
For the hazard module specifically, return periods describe hazard intensity thresholds, not event frequency. A "1-in-100 year wind speed" at a specific location means the wind gust speed that is exceeded with 1% annual probability at that point. The hazard module calculates this threshold by summing the contributions of all events in the stochastic catalogue that affect that location, weighting each by its annual rate. The output, the hazard return period curve, is what the vulnerability module uses to calculate damage at each severity level.
Every hazard module is built on a set of assumptions about the physical world. These assumptions are necessary, a model that made no assumptions would be indistinguishable from the physical world itself, which is what we are trying to approximate. But the assumptions introduce systematic uncertainty, and understanding them is what allows a practitioner to know when the model is likely to be wrong.
Most hazard modules are calibrated to historical observation data and assume that the frequency and severity distribution of events is the same in the future as it was in the past. A catalogue built on 150 years of Atlantic hurricane data assumes the next 150 years will have similar statistics.
Hazard models are calibrated to events in the instrumental record, typically 50–150 years. The model assumes this period is long enough to have sampled the full distribution, including the most severe events.
Most cat models assume that events in the stochastic catalogue are statistically independent, the occurrence of one event does not change the probability of another in the same year or the following year.
The model assumes its hazard grid covers the full geographic domain of the exposure and that intensity can be reliably estimated at every point on the grid.
For earthquake specifically, the model uses Ground Motion Prediction Equations (GMPEs) to estimate how seismic energy attenuates with distance from the fault rupture. These equations are derived from recorded ground motion data.
The hazard module assigns each grid cell a local site amplification factor based on soil type. Soft soils amplify seismic waves significantly relative to bedrock, by a factor of 3-5 or more in extreme cases.
You do not need to understand the physics of Ground Motion Prediction Equations to be an effective cat modelling analyst. But you do need to know which hazard assumptions are most likely to be violated for a specific risk. For a portfolio with significant exposure in Turkey, the stationarity assumption is challenged by ongoing seismic activity along the North Anatolian Fault. For a US wildfire portfolio, the stationarity assumption is challenged by changing vegetation and drought conditions. For a Singapore portfolio with substantial Southeast Asian exposure, the completeness assumption is challenged by sparse historical loss data for Indonesian earthquake. Identifying which assumption is under stress for a given geography is the analytical judgement that separates a practitioner from a technician.
Magnitude is a measure of the total energy released by an earthquake, not the ground shaking at any specific location. A magnitude 8.0 earthquake 500km away may cause less shaking at a specific site than a magnitude 6.5 earthquake 10km away. The hazard module calculates local intensity at each location, which depends on magnitude, depth, distance, source characteristics, and local site conditions simultaneously.
The catalogue is synthetic, constructed by simulation, not by listing historical events. Historical events are used to calibrate the physical models, but the catalogue itself contains tens of thousands of hypothetical events that have never occurred. Including events that have never occurred is precisely the point: it is what allows the model to estimate probabilities at return periods beyond the historical record.
Return period is a probability statement, not a scenario definition. RMS's 1-in-100 year California earthquake and AIR's 1-in-100 year California earthquake may correspond to very different physical events, different magnitudes, different locations, different intensity distributions across the state, even if both are calibrated to the same 1% annual exceedance probability. The hazard catalogues behind the headline return period numbers can diverge significantly.
High hazard is necessary but not sufficient for high loss. A vacant lot in a high-seismic zone has zero insured loss regardless of the PGA. A well-engineered, well-constructed modern building in the same zone has much lower expected loss than an unreinforced masonry building nearby with the same TIV. Hazard tells you the physical intensity. Exposure and vulnerability, the next two components, determine how that intensity translates into financial loss.
The first of the four cat model components. Takes a geographic location and returns the probability distribution of physical intensities that could occur there from catastrophe events. The foundation on which all downstream loss calculations are built.
A large synthetic database of simulated catastrophe events, each with a defined physical description and an assigned annual rate. The core of the hazard module. Typically contains 10,000–100,000 events per peril, representing the full distribution of possible events.
The geographic distribution of intensity from a single event in the stochastic catalogue. For a hurricane, this is the wind field, the pattern of wind speeds across the affected region. For an earthquake, it is the ground motion pattern. Every event in the catalogue has a unique footprint.
The expected number of times a specific synthetic event would occur per year if the world ran forward indefinitely. A rate of 0.01 per year corresponds to a 1-in-100 year return period. Rates are derived from physical science rather than historical event counting for rare events.
The primary intensity metric for earthquake hazard. Measured in fractions of gravitational acceleration (g). PGA drives damage to short-period, low-rise structures. Taller buildings respond more to spectral acceleration at their natural period.
The spatial database of hazard intensities at specific geographic points that makes up the model's hazard surface. When a location is geocoded, the model places it on the hazard grid and reads the intensity values from the relevant grid cells.
The increase in ground motion intensity caused by soft or unconsolidated soils relative to bedrock. Can amplify earthquake intensity by a factor of 3–10. A critical secondary hazard parameter captured, or not, in the hazard module's site classification system.
The condition where the statistical distribution of hazard is changing over time. Climate change introduces non-stationarity into hurricane intensity, wildfire weather, and extreme heat distributions. When present, historical calibration systematically underestimates future risk.
Ground Motion Prediction Equation. The mathematical relationship used to estimate how earthquake ground motion intensity decreases (attenuates) with distance from the fault. A primary source of epistemic uncertainty in earthquake hazard models and a major driver of vendor model divergence.
Five questions · 4 of 5 correct (80%) to pass