At 8:46am on 11 September 2001, American Airlines Flight 11 struck the North Tower of the World Trade Center. Seventeen minutes later, United Airlines Flight 175 struck the South Tower. Two further aircraft hit the Pentagon and a field in Pennsylvania. The attacks killed 2,977 people and caused insured losses of approximately USD 47 billion — at the time the largest insured loss event in history, rivalling the most destructive natural catastrophes ever recorded.
Within weeks of the attacks, reinsurers began withdrawing terrorism coverage from renewal treaties. Primary insurers followed. Standard commercial property policies — which had included terrorism coverage without any separate mention because no underwriter had thought to exclude it — suddenly represented an enormous unpriced, uncapped liability that no one in the market knew how to quantify. The insurance market had written billions of dollars of terrorism coverage without ever building a model to price it. Governments in the US, UK, France, and elsewhere were forced to step in as reinsurers of last resort to prevent the complete withdrawal of commercial property terrorism coverage from the market.
Terrorism occupies a unique position in catastrophe modelling because the hazard is driven by human intent, not physical law. Earthquake, hurricane, and flood follow the same physical processes regardless of what buildings sit in their path. Terrorists actively observe defensive measures, intelligence operations, and prior attacks — and adapt their tactics, targets, and methods in response. A security measure that hardens one target redirects attention to another. A successful attack method that is widely publicised invites imitation. A method that leads to effective countermeasures is abandoned in favour of a new approach.
This adaptive adversary problem means that historical attack data provides an inherently incomplete guide to future risk. The Global Terrorism Database documents over 200,000 attacks since 1970 — a rich historical record. But the attacks of 1985 reflect the threat environment of 1985: different geopolitics, different technology, different security infrastructure, different targets. The calibration challenge that earthquake models face — sparse data for extreme tail events — is compounded in terrorism by the fact that the underlying hazard process is itself changing in response to human decisions on both sides.
Before examining how terrorism is modelled, the legal definition matters — because it determines whether a given event triggers terrorism coverage or a different policy provision. Terrorism definitions in insurance policies typically require several elements simultaneously:
The boundary between terrorism and other political violence is legally contested and practically significant. The same explosion in a city centre may be terrorism, organised crime, or an industrial accident — and the insurance response differs substantially depending on classification. Policy wordings and coverage disputes over classification are a persistent challenge in terrorism insurance, particularly when events occur at the ambiguous boundary between terrorism and civil commotion or political violence.
Terrorism cat models must represent a range of attack modes, each with different physical characteristics and damage patterns. The physical consequence model — the equivalent of a GMPE in earthquake modelling or a wind field model in hurricane modelling — depends entirely on which attack type is being assessed.
By far the most common attack type globally. Damage follows predictable physics: a blast wave propagates outward from the detonation point, with pressure decreasing rapidly with distance. The scaled distance relationship means doubling the standoff distance between the device and the target reduces blast pressure dramatically — physical security barriers (bollards, vehicle exclusion zones) are the single most effective mitigation. Building components fail sequentially: glazing first, cladding second, structural elements at high yields or short standoff only. Progressive collapse — cascading structural failure from a single column loss — is the primary catastrophic loss mechanism for very large devices.
The extreme tail of the terrorism risk spectrum. Chemical attacks (Tokyo subway sarin, 1995) use toxic agents that require atmospheric dispersion modelling. Biological attacks introduce epidemiological uncertainty — disease spread modelling adds extraordinary complexity. Radiological "dirty bombs" cause primarily economic loss from decontamination and stigmatisation rather than direct casualties. Nuclear devices in an urban setting would produce losses exceeding the entire insurance industry's capacity. CBRN attacks are almost universally excluded from terrorism insurance policies — the unlimited loss potential and pricing impossibility make them uninsurable in private markets.
Since 2010, low-complexity attacks — mass shootings and vehicle ramming attacks — have become the most frequent attack type in Western markets. The primary insurance losses are business interruption (venue closure, investigation access denial), liability claims, and event cancellation — not structural property damage. Las Vegas 2017, Nice 2016, Christchurch 2019 illustrate the pattern: significant economic loss, limited direct property damage. These attacks drive liability and specialty lines exposure more than property cat exposure.
Attacks on industrial control systems of critical infrastructure — power grids, water treatment, financial systems — that cause physical damage or disruption without deploying any conventional weapon. The 2015–2016 Ukrainian power grid attacks and the 2021 Colonial Pipeline ransomware attack demonstrate the physical consequence potential of cyber means. Coverage sits in an unresolved gap between standard property policies (which typically require physical damage) and cyber policies (which may not cover downstream physical consequences). Pool Re in the UK extended coverage to cyber-enabled terrorism in 2018 in explicit recognition of this gap.
Standard probabilistic cat modelling — calibrate a stochastic catalogue from historical data, run it against exposure, produce an EP curve — does not transfer cleanly to terrorism. The adaptive adversary problem, the sparse loss record, and the strategic rather than physical nature of the hazard require a different approach.
The hazard component of a terrorism model assesses the probability that specific attack types will occur at specific locations. Unlike an earthquake GMPE — a physical equation derivable from first principles and empirical data — the terrorism threat assessment draws on intelligence analysis, historical attack databases, and target attractiveness models. Key inputs include: historical attack frequency and severity by type and geography; government threat level assessments (the UK's CONTEST system, the US Homeland Threat Assessment); and target attractiveness models that recognise that iconic buildings, high-density gatherings, transport hubs, and symbolic institutions attract disproportionate terrorist attention.
Once a scenario is defined, physical models estimate the damage footprint. For explosive attacks, blast wave propagation models (empirical and computational fluid dynamics based) estimate peak overpressure at each location around the device — which damage functions convert into glazing, cladding, and structural loss. For chemical or radiological attacks, atmospheric dispersion models estimate agent concentration at downwind locations. The physical models for terrorism are well-developed; the challenge is not consequence estimation but the specification of the scenario in the first place.
The worst-case terrorism scenario is not geographically dispersed — it is maximally concentrated. An attack on a financial district block in New York, London, or Singapore could simultaneously affect the headquarters of multiple large insurers, banks, and corporations. An insurer's terrorism exposure in a single postcode could represent an enormous proportion of their total portfolio value. Accumulation management — mapping all policies with terrorism coverage to a geographic grid and identifying concentration hotspots — is therefore the most operationally important output of terrorism cat modelling. This is why terrorism PML analysis focuses on specific city-block scenarios rather than broad geographic loss distributions.
One of the most distinctive features of the terrorism insurance market globally is the role of government as reinsurer of last resort. The near-complete market withdrawal after 9/11 demonstrated that private markets alone cannot provide reliable terrorism coverage at scale — the combination of unlimited potential loss, pricing uncertainty, and the systemic correlation of losses within a single attack event overwhelms standard insurance market structures. Government backstops restore market function by absorbing losses above thresholds that the private market cannot bear.
| Market | Scheme | Structure | Key Feature |
|---|---|---|---|
| United States | TRIA (2002, extended) | Government backstop above industry retention threshold. Insurers must make terrorism coverage available. Co-participation between private market and federal government above the trigger. | Requires government certification of a "certified act of terrorism." Covers commercial lines only — not personal lines. Has been extended multiple times since 2002; market would face major disruption without it. |
| United Kingdom | Pool Re (est. 1993) | Mutual reinsurance pool — member insurers cede terrorism risk to Pool Re at agreed premium rates. Pool Re is reinsured by the UK government as ultimate backstop. Covers both commercial and some residential risks. | Established after IRA City of London bombing (1992). Has expanded coverage to include cyber-enabled terrorism and CBRN since inception. One of the oldest and most developed terrorism pool structures globally. |
| France | GAREAT (est. 2002) | Mandatory pool for large commercial property terrorism risks above EUR 6M TIV. Backed by CCR (Caisse Centrale de Réassurance), a state-owned reinsurer with government guarantee. | Participation mandatory for insurers writing commercial property above the TIV threshold. Smaller risks covered by voluntary individual insurer arrangements or through GAREAT at lower layers. |
| Australia | ARPC (est. 2003) | Government-backed reinsurance facility. Eligible insurers can reinsure terrorism risk with ARPC. Government provides unlimited backstop above ARPC's own reserves. | Covers declared terrorist incidents in Australia. Less frequently tested than UK or US schemes. Extends to cover group life and workers' compensation terrorism exposures alongside property. |
Government backstop arrangements fundamentally reshape the loss distribution faced by private insurers. Understanding the specific terms — the trigger threshold, the co-participation rate, the definition of a covered event, the scope of covered lines — is essential for accurately modelling net terrorism losses. A gross loss model that does not incorporate the backstop structure will overstate the insurer's net loss for events above the backstop threshold. For US portfolios under TRIA, the government certification requirement is particularly important: an attack that is not certified — or certified after a delay — may produce gross losses that fall entirely on the private market rather than triggering the government co-participation.
The terrorism threat landscape has changed substantially since 2001, with implications for accumulation management and model assumptions. The shift from centralised hierarchical organisations (Al-Qaeda pre-2001) toward decentralised franchise models (ISIS) and lone-actor attackers has increased attack frequency while reducing average severity — more events, smaller yields, harder to pre-empt. The hardening of traditional hard targets (government buildings, financial institutions with physical perimeters) has redirected attention toward soft targets: concert venues, transport hubs, shopping centres, sporting events — locations with large concentrations of people and minimal physical security. These soft targets may represent significant insured accumulations outside traditional financial district concentration zones and are often lower in the insurer's priority list for accumulation monitoring.
Domestic terrorism — right-wing extremism, eco-terrorism, and other forms of politically motivated violence by citizens of the country they attack — has grown as a proportion of total terrorism risk in Western markets. The geographic and target distribution of domestic terrorism differs substantially from international jihadist terrorism: less focus on financial districts, more focus on religious institutions, government facilities, and minority communities. Models calibrated primarily to international terrorism patterns may systematically understate domestic terrorism accumulation in specific locations.
The fundamental challenge in terrorism modelling: unlike physical hazards governed by fixed natural laws, terrorists observe and respond to defensive measures, security improvements, and prior attack outcomes — making historical data an inherently incomplete guide to future risk.
The distance between an explosive device and the target structure. The single most important physical parameter governing blast damage severity — because blast pressure decreases rapidly with distance, physical security barriers that increase standoff distance are the most effective mitigation against VBIED attacks.
Chemical, Biological, Radiological, Nuclear — the category of unconventional attack modes carrying potentially catastrophic and unlimited loss potential. Almost universally excluded from standard terrorism insurance policies. Specialist capacity available in very limited amounts through the London market.
Terrorism Risk Insurance Act — the US federal legislation (2002, extended multiple times) that requires insurers to make commercial terrorism coverage available and provides a government backstop above specified industry retention thresholds for certified terrorism events.
Pool Reinsurance Company — the UK government-backed terrorism reinsurance mutual established in 1993 following IRA City of London bombings. Member insurers cede terrorism risk to Pool Re; the UK government reinsures Pool Re as ultimate backstop. Has expanded to cover cyber-enabled terrorism and operates as one of the world's most developed terrorism pool structures.
A geographic concentration of terrorism-covered insured value — typically a city block or postal district — where a single event could produce losses exceeding the insurer's or reinsurer's risk tolerance. Identification of accumulation hotspots is the primary operational output of terrorism cat modelling for portfolio management purposes.
A publicly accessible location with high people density and limited physical security — shopping centres, transport hubs, entertainment venues, religious institutions. Increasingly the focus of terrorist attack planning following hardening of traditional hard targets (financial districts, government buildings). May represent significant insured accumulations outside standard terrorism monitoring priorities.
The cascading structural failure of a building triggered by the loss of a primary load-bearing element — typically a column — under blast loading. A critical loss mechanism for very large explosive devices at short standoff distances. The primary cause of catastrophic structural loss in major VBIED events such as the 1995 Oklahoma City bombing.