The storm that created the modern catastrophe modelling industry. Andrew's catastrophic impact on South Florida, far exceeding what insurers had modelled, anticipated, or reserved for, exposed the complete inadequacy of pre-existing risk frameworks and forced an industry-wide reckoning that gave birth to commercial cat modelling as we know it today.
Hurricane Andrew holds a unique and foundational place in the history of catastrophe modelling. It is not merely a large hurricane loss event, it is the event that demonstrated, beyond any reasonable doubt, that the insurance industry had no reliable method of estimating the losses that a major hurricane could inflict on a densely developed coastal area. The failure was not marginal. Insurers had set aside reserves of approximately USD 4–8 billion to cover a worst-case Florida hurricane scenario. Industry predictions at the time were that it would cost insurers around USD 4 to 5 billion, but Andrew ended up costing the insurance industry USD 15.5 billion in Florida claims alone. The gap between expectation and reality was a factor of three to four.
The consequence was an industry crisis: the event resulted in the insolvency of 16 insurance companies, with many more coming perilously close to financial collapse. But Andrew's most enduring legacy was not the losses themselves, it was what those losses revealed about the inadequacy of the industry's existing risk assessment methods. Simple rules of thumb, "reserve twice the largest prior loss", had been catastrophically exposed. The commercial cat modelling industry was born in Andrew's aftermath, and the entire apparatus of probabilistic loss estimation that underpins global (re)insurance practice today traces its lineage directly to the reckoning that Andrew forced.
In the early morning hours of August 24, 1992, in a small office in downtown Boston, AIR Worldwide was working on its damage estimate. The AIR estimate of USD 13 billion came well before claims adjusters' final tallies. Although the figure was enormous and hard to believe, it helped mobilise necessary aid, even if there was some scepticism. As reality set in, that scepticism turned to intrigue. The cat modelling industry was born.
Andrew originated as a tropical wave that emerged from the west coast of Africa on August 14, 1992. For most of its early Atlantic life it struggled, persistent upper-level wind shear suppressed its organisation for nearly a week, and at one point the NHC nearly stopped issuing advisories as the system appeared likely to dissipate. It was not until August 22 that Andrew began to organise meaningfully.
Then, in the 24 hours before landfall, Andrew underwent rapid intensification that transformed it from a dangerous Category 4 storm into one of the most intense Atlantic hurricanes in the modern instrumental record. Upon its Florida landfall at 5 AM EDT on August 24, sustained winds reached 165 mph with a minimum central pressure of 922 mb, one of the lowest ever recorded for an Atlantic hurricane at landfall. A 2002 National Hurricane Center reanalysis confirmed Category 5 status, recognising that reconnaissance data had underestimated the peak winds.
Andrew's defining meteorological characteristic was its extraordinary compactness. Unlike Sandy's 1,000-mile wind field or the large, sprawling structure of many major hurricanes, Andrew was a tightly wound, small but extraordinarily intense system. The radius of maximum winds at landfall was only about 8 miles from the centre, meaning the most extreme winds were confined to an extremely narrow band. This compactness had two critical consequences:
Andrew made landfall near Biscayne Bay at the southern edge of the Miami metropolitan area, tracking almost due west across the southern Florida peninsula and emerging into the Gulf of Mexico approximately four hours after Florida landfall. The track was relatively fast, Andrew moved at around 18 mph, limiting the duration of extreme wind exposure at any given location but ensuring that the most intense eyewall passed through the most densely developed residential suburbs of South Florida. The storm then crossed the Gulf, weakened slightly, and made a second landfall in south-central Louisiana on August 26, causing significant but far less catastrophic losses.
Before Andrew, the largest insured loss from a U.S. hurricane was Hurricane Hugo in 1989, which caused approximately USD 4 billion in insured losses. Standard industry practice was to set catastrophe reserves at roughly twice the largest prior loss, approximately USD 8 billion for a worst-case scenario. Typical maximum reserves were on the order of USD 8 billion, twice the largest ever U.S. insured loss. But even that very generous reserve fell short of what would ultimately be needed.
The actual insured loss of USD 15.5 billion was nearly double the industry's most generous worst-case reserve assumption, and the gap was not a matter of statistical bad luck. The models and methods used to set those reserves were simply wrong. They did not account for the accumulation of insured value in South Florida over the preceding decades, they did not model the full range of plausible storm intensities and tracks, and they did not properly account for the vulnerability of the building stock to major hurricane winds.
More than 650,000 claims were filed. Eight insurers became insolvent immediately, with a further three driven into insolvency the following year, the largest wave of insurance company failures from a single natural event in U.S. history. Many of the affected companies were small Florida-focused personal lines insurers who had written large concentrations of homeowners policies in South Florida without any meaningful understanding of their aggregate catastrophe exposure. The absence of cat models meant they had no way of knowing how dangerously concentrated their book had become.
The survivors faced a reckoning of a different kind. Virtually every major property insurer operating in Florida announced plans to reduce their Florida exposure significantly after Andrew, triggering a market withdrawal crisis that ultimately required state intervention. To fill the immediate reinsurance gap after the market seized up post-Andrew, in 1993 the state created the Florida Hurricane Catastrophe Fund (FHCF) with a structure based on modelling results, including RMS models. The state-backed insurer Citizens Property Insurance was eventually created in 2002 as a market of last resort for Floridians who could not obtain coverage from the withdrawing private market.
Andrew's impact on the reinsurance market was equally severe. Reinsurers had provided catastrophe cover to Florida primary insurers based on their own inadequate models of Florida hurricane risk, and found themselves paying losses that exceeded their own expectations substantially. The reinsurance market hardened dramatically after Andrew, rates increased sharply, capacity contracted, and the terms and conditions of cat cover changed substantially. This hardening created the economic incentive for the development of new capital market alternatives to traditional reinsurance: cat bonds were first issued in 1994, largely in response to the capital markets' recognition that the reinsurance market was leaving natural catastrophe risk underserved at acceptable prices.
Post-Andrew engineering surveys revealed that a significant component of the destruction resulted not from winds exceeding any reasonable design standard, but from egregious failures of construction quality. Roof coverings were improperly nailed; structural connections between roof trusses and walls were inadequate; inspection and enforcement of the existing building code was found to have been grossly deficient. In some cases, homes built to code on paper were constructed with 20–30% fewer fasteners than required. Andrew exposed not just the inadequacy of the building code itself but systemic failures of enforcement across South Florida's construction industry.
Post-Andrew forensic engineering surveys produced findings that shocked both engineers and policymakers. A significant fraction of the destruction resulted from substandard construction that fell short even of the existing (inadequate) building codes. The findings prompted the most comprehensive overhaul of Florida's building code in the state's history, the post-Andrew Florida Building Code dramatically raised design wind speed requirements, strengthened roof connection standards, and substantially improved inspection and enforcement mechanisms. Post-code buildings have consistently outperformed pre-code construction in every subsequent Florida hurricane.
Before Andrew, the insurance industry operated without probabilistic cat models in any meaningful sense. The "models" in use were actuarial loss ratios, historical analogs, and simple rules of thumb. These methods shared a common flaw: they extrapolated from the historical record of relatively modest hurricane losses to estimate future risk, without accounting for the massive accumulation of insured value in Florida's coastal counties over the preceding decades, without modelling the full range of physically possible storm intensities and tracks, and without incorporating engineering-based vulnerability functions that captured how buildings actually fail in extreme winds.
The industry had a seven-year major hurricane drought preceding Andrew (1966–1992 with no Florida major hurricane landfalls). This drought created complacency, reserves that might have seemed generous in 1966 were wholly inadequate by 1992, because the insured value base in South Florida had grown enormously over those 26 years. No rule of thumb could capture this dynamic; only an exposure-based model could.
The single most consequential act in the history of catastrophe modelling took place in a small office in Boston on the morning of August 24, 1992. Four hours after Hurricane Andrew made landfall near Homestead, AIR issued a statement that insured losses could exceed USD 13 billion in Florida. Actual losses equalled USD 15.5 billion. AIR's estimate was not produced by simple rules of thumb or experience ratings, it was produced by running Andrew's observed meteorological parameters through a physics-based simulation model that estimated damage building by building across the South Florida exposure base.
The precision of that estimate, within 15% of the final outcome, issued just hours after landfall, was a proof of concept unlike anything the insurance industry had seen. Traditional insurers and actuaries had been telling themselves for decades that catastrophe risk could not be reliably quantified. AIR's four-hour estimate demonstrated that it could, and that the tools to do so were already built.
"Simple rules of thumb that insurers used for estimating loss from catastrophic events, take the largest historical loss and double it, had met their match. The cat modelling industry was born."
— The Actuary Magazine, on the legacy of Hurricane AndrewAndrew's precursor disturbance exits the African coast but is suppressed by strong wind shear over the central Atlantic. For nearly a week, the system shows little sign of becoming a significant threat. NHC nearly stops issuing advisories.
As Andrew passes over warm western Atlantic waters north of the Bahamas, wind shear relaxes and the storm undergoes dramatic intensification. Within 24 hours it surges to 165 mph sustained winds and a pressure of 922 mb, one of the lowest on record for an Atlantic landfall.
Andrew's eye crosses the coast just north of Homestead at 5:05 AM EDT. The narrow eyewall, only 8 miles wide, concentrates maximum destruction in a tight corridor through suburban South Florida. Homestead is effectively destroyed. The National Hurricane Center, 25 miles to the north, records only Category 1 conditions.
In a small Boston office, AIR releases an estimate that insured losses could exceed USD 13 billion, a figure regarded with wide disbelief by industry veterans accustomed to thinking in single-digit billions. The estimate will prove to be within 15% of the final figure.
Andrew makes landfall in Louisiana approximately 100 miles southwest of New Orleans as a Category 3 hurricane, causing significant but far less catastrophic damage than in Florida. The storm dissipates over the south-central United States in the following days.
As loss estimates climb above USD 15 billion, the industry crisis deepens. Eight insurers declare insolvency immediately. Major national insurers announce plans to withdraw from Florida. AIR, RMS, and EQECAT see dramatic acceleration in interest from insurers who had previously ignored their products.
Florida creates the FHCF backed by cat model outputs. The world's first catastrophe bonds are issued, driven by recognition that traditional reinsurance capacity is insufficient for the Florida risk. Florida begins a multi-year building code reform process. Cat modelling firms grow from curiosities to indispensable industry infrastructure.
Andrew demonstrated conclusively that probabilistic simulation models, already built by AIR, RMS, and EQECAT, produced more accurate loss estimates than any industry rule of thumb. Within five years of Andrew, cat models had become standard tools for every major insurer and reinsurer globally. The industry would never return to pre-Andrew methods.
The post-Andrew engineering surveys produced the first systematic, large-scale dataset of hurricane wind damage to residential buildings, enabling the development of vulnerability functions grounded in empirical damage data rather than actuarial loss ratios. This shifted the entire paradigm of vulnerability modelling from statistical to engineering-based.
Andrew triggered the most comprehensive building code overhaul in Florida history. The post-Andrew Florida Building Code dramatically raised design wind speed requirements, strengthened roof connection standards, and substantially improved inspection and enforcement. Post-code buildings have consistently outperformed pre-code buildings in every subsequent Florida hurricane.
The FHCF was created as a state-backed reinsurance facility to fill the gap left by reinsurance market withdrawal after Andrew. Its structure was explicitly based on cat model outputs, the first time a government insurance programme was designed using probabilistic loss models. It remains a central pillar of Florida's insurance system today.
The reinsurance market's failure to adequately price and provide capacity for Florida hurricane risk after Andrew directly motivated the development of catastrophe bonds and the broader insurance-linked securities market. The first cat bond was issued in 1994, directly tracing its origins to the capital market's recognition of the opportunity created by Andrew.
The "20 miles that saved Miami" counterfactual permanently embedded track sensitivity into cat modelling practice. Models now generate large stochastic event catalogs precisely because the precise track of a storm relative to exposure concentrations, not just storm intensity, determines the loss outcome. Andrew is why cat models have 10,000+ simulated storms rather than relying on historical analogs.
This event is directly referenced in Exposure Data Preparation for Catastrophe Modelling, the Intermediate course on The Return Period - in Lesson 3.4 (Interpreting Model Output), where the origins of probabilistic hurricane modelling and AIR's four-hour estimate are used to explain why cat models were built the way they are; and in Lesson 4.1 (Model Uncertainty), where Andrew's reserve failure is the primary illustration of why deterministic worst-case scenarios systematically underestimate tail risk and why the industry moved to full probabilistic EP curves.