The Modeling of Space Risks

I.The Paradox of a Risk One Cannot Model

An infrastructure that has become critical

The space economy is undergoing an unprecedented expansion. The number of orbital launches now exceeds two hundred a year, low-orbit constellations deploy by the thousands of satellites, and whole swathes of human activity, from communications to navigation and Earth observation, depend on assets placed a few hundred kilometers above our heads. In this context, insurance should play a leading role, as it does for any critical infrastructure. Yet it is the opposite that is observed. Of around thirteen thousand active satellites, barely three hundred are actually insured, and most losses linked to a collision are expressly excluded from policies¹. The space premium market remains modest, on the order of six hundred million dollars a year, for in-orbit assets whose value runs into the tens of billions².

A specialized and fragile market

This market further presents a structural fragility, for it rests on a narrow circle of insurers with the required technical expertise, among them a few major names such as AXA XL, Munich Re, Swiss Re, Hiscox and Global Aerospace². This concentration of supply makes the market vulnerable, since the withdrawal of a few players suffices to drain the available capacity. The loss experience is by nature jagged, made of calm years interrupted by total losses of great severity, a destroyed satellite often representing hundreds of millions of dollars. This combination of a low frequency and an extreme severity, added to the smallness of the portfolio, produces a volatility that insurers struggle to smooth, and partly explains their caution toward a risk they know from experience to be capable of tipping over abruptly.

Modeling, not the peril, as the obstacle

This paradox calls for an explanation, and it lies not in how dangerous space is. The real obstacle to space insurability is the impossibility of modeling the risk reliably, more than its magnitude. Where the risk cannot be quantified, the insurer withdraws, raises its premiums or excludes cover, for want of being able to price what it cannot estimate¹. Space thus illustrates a more general truth of the discipline, namely that the frontier of insurability coincides with the frontier of modeling. A risk is insurable only insofar as it is measurable, and it is precisely measurement that is lacking in orbit. To understand space insurance therefore requires examining not the danger itself, but the reasons why this danger escapes the actuary's instruments.

II.The Scarcity of Data and the Geostationary Legacy

Too few events for a law of large numbers

The first difficulty is that of data scarcity. Insurance rests on the observation of a large number of similar events, which allows the frequency and severity of losses to be estimated. The space domain offers exactly the opposite, a small number of highly heterogeneous assets, rare but extremely severe losses, and a history too short and too unstable to support reliable statistics. Each satellite is also largely unique, by its mission, its orbit and its technology, which forbids treating the portfolio as a homogeneous population. The law of large numbers, the foundation of any pricing, here has neither the number nor the homogeneity it needs, and the actuary is forced to extrapolate from a few dozen cases what would elsewhere be estimated over millions.

Models inherited from another space

To this shortage is added a cumbersome legacy. The sector's actuarial frameworks were forged in the era of the large geostationary satellites, expensive objects, few in number and placed on high, clear orbits, for which the principal risk was launch failure. These models rest on probability distributions per isolated asset and on the idea that the bulk of the danger is concentrated in the first minutes of flight³. Yet the new space, that of the low-orbit mega-constellations, contradicts each of these assumptions, since it involves thousands of cheap objects, heavily concentrated on congested corridors, whose dominant risk is no longer launch but in-orbit collision. The growth in premiums was not accompanied by a corresponding overhaul of the models, so that the market prices a new risk with tools designed for the old.

This gap is not merely a matter of technical refinement, it touches the very validity of the extrapolations. When one prices a fleet of thousands of small short-lived satellites from a history made of a few dozen large long-lived ones, one transposes frequencies and severities observed in one regime into an entirely different one. Yet nothing guarantees that the lessons of the old world remain valid in the new, where orbital crowding, the rapid renewal of fleets and the standardization of craft create unprecedented dynamics. The space actuary is thus confronted with a formidable problem, that of having to price a future that does not resemble its past, and for which the accumulated data risk being less a guide than a mirage.

III.The Blind Spot, a Partly Invisible Environment

The danger one does not see

The second difficulty is more fundamental still, for it touches the very observation of the danger. To model a risk supposes at minimum being able to observe the elements that compose it, yet the orbital environment is only partially visible. According to the European Space Agency's report, there are around fifty-four thousand objects larger than ten centimeters in orbit, of which only some forty-three thousand are actually tracked, but also nearly 1.2 million fragments between one and ten centimeters, and more than one hundred and forty million objects smaller than a centimeter, all undetectable by current surveillance systems⁵. The problem is that these invisible fragments are perfectly lethal, for they travel at an average relative speed of around twenty-eight thousand kilometers per hour, enough to destroy a satellite on impact⁶.

Modeling a population one does not know

This partial invisibility vitiates the modeling at its root. The insurer must estimate the probability that a satellite encounters a dangerous piece of debris, but it knows only a fraction of the population of that debris, the large tracked objects. The most numerous part, and largely as destructive, remains beyond the reach of sensors and enters the models only by statistical inference, that is, by hypothesis. The real density of the danger at a given point of the orbit is thus unknown, and the calculated collision probability is only an approximation resting on an estimated rather than observed population. One asks the actuary to price the encounter with a threat most of which is invisible to it, an exercise that is less a matter of measurement than of informed conjecture.

IV.Non-Linearity and Correlation, the Specter of Kessler

A danger that feeds itself

The third difficulty breaks a cardinal assumption of insurance, that of the independence of losses. As early as 1978, the scientists Donald Kessler and Burton Cour-Palais described a scenario in which a collision in orbit generates a cloud of debris, which causes new collisions, in a self-sustaining chain reaction⁷. The fearsome feature of this Kessler syndrome is its non-linearity, for the number of collisions does not increase in proportion to the number of objects, but exponentially beyond a certain density threshold. Such a runaway could render inoperative around 20% of low-orbit satellites within six months of its trigger, and durably turn certain orbits into unusable zones. The space danger is therefore not a stable hazard observed once and for all, but a dynamic process liable to tip over.

Losses that are no longer independent

It is this tipping over that ruins classical modeling. A Kessler-type event does not produce independent losses, dispersed in time and space, but a single correlated loss event striking thousands of assets simultaneously⁹. According to an OECD model, such a scenario could inflict 191 billion dollars in immediate losses and weigh in the long term close to 2% of global gross domestic product, invalidating thousands of in-orbit policies at a stroke and exceeding the available reinsurance capacity⁸. Correlation in orbit here joins exactly the problem already encountered for pandemics, namely that a risk striking the whole portfolio at the same time destroys the diversification on which insurance lives. Insurers are beginning to develop models specific to constellations, incorporating these correlated events, but the exercise remains in its infancy in the face of a phenomenon whose trigger threshold no one knows precisely.

V.The Reflexivity of Space Risk

The risk that the insured activity itself creates

The fourth difficulty is doubtless the most disconcerting, for it lies in the reflexivity of space risk. In most lines, the danger is external and independent of the insured activity, a hurricane does not strengthen because more houses are insured. In orbit, on the contrary, it is the activity itself that creates the danger, since each launch, each satellite and each abandoned piece of debris increases the orbital density and therefore the probability of a future collision. The risk environment degrades as the activity the insurer covers develops, and a model calibrated on last year's orbit already underestimates this year's. The model must then predict the evolution of a system of which the insured, and the insurer through it, are a part, which introduces a feedback loop foreign to traditional actuarial science.

The concentration of mega-constellations

This reflexivity is amplified by the extreme concentration of the activity in a few hands. The Starlink constellation exceeded six thousand operational satellites at the start of 2025, with filings for a deployment that may reach forty-two thousand craft, while Amazon Kuiper had its first 3,236 satellites approved and OneWeb, now under Eutelsat's wing, targets a fleet of several hundred units¹⁰. This concentration creates an unprecedented accumulation risk, for a single operator can alone represent a colossal exposure, and a single incident can affect a large number of assets sharing the same orbit and the same architecture. The fragmentation of the Long March 6A rocket stage in 2024 and the debris fallout observed on the ground recalled that these objects, far from abstract, break apart, scatter and fall back. To model so concentrated a fleet supposes abandoning the assumption of independent assets in order to think in terms of system failures, an exercise for which the data are almost entirely lacking.

The political and legal unknown

To this physical reflexivity is added a political and legal indeterminacy that further aggravates the uncertainty. Several powers have conducted anti-satellite weapon tests, Russia in 2021, India in 2019, China in 2007 and the United States as early as 1985, each of which generated debris clouds and demonstrated a capacity for deliberate destruction hard to model, and most often excluded from policies under war risk¹¹. There is no binding international treaty defining liability or the allocation of fault in a collision involving several operators. The space liability insurer must therefore price an exposure whose legal regime is itself indeterminate, which adds to the physical uncertainty a normative one, and makes the modeling of liability risk more conjectural still than that of damage risk.

VI.Modeling the Unmodelable, or Withdrawing

Between new models and exclusions

Faced with a risk thus data-poor, partly invisible, non-linear, correlated and reflexive, the market oscillates between two responses. The first consists in refining the models, by developing frameworks specific to constellations, by incorporating debris-tracking data, and by making cover conditional on compliance with mitigation measures, collision-avoidance protocols and end-of-life disposal plans. The second, more defensive, consists in withdrawing from the risk one cannot model, through exclusion or capacity restriction. The signs of this retreat are clear, since the underwriting crisis of 2018 and 2019 had already seen some insurers leave low orbit, and exclusions relating to collision with debris have since extended to a significant share of policies, while capacity and reinsurers' participation receded⁴.

The systemic share calls for collective action

There remains the heaviest share of the risk, that of the Kessler scenario, which appears beyond the reach of any purely insurance-based solution. A correlated loss on the scale of the orbit by nature exceeds the market's capacity and joins the category of systemic risks this series has already encountered for pandemic and war, those whose private pooling is impossible and which call for a collective response. This response lies less in indemnity than in prevention, that is, in the active removal of debris and the limitation of its production, as with the models estimating that five to ten large objects must be removed each year to stabilize low orbit, at a cost on the order of several hundred million dollars, and of the public zero-debris targets for the 2030 horizon¹². The modeling of space risk thus leads to a conclusion that goes beyond insurance, namely that the preservation of the orbit as a common good is a matter of collective governance before it is a matter of contract.

Cover stops where the model stops

At the end of this analysis, space risk appears as the textbook case of a truth this series has constantly crossed, namely that one insures only what one can model, and one models only what one can observe and estimate. Space brings together, in a single object, all the pathologies of modeling, the scarcity of data, the invisibility of the danger, non-linearity, correlation and reflexivity, which makes it the most demanding laboratory of contemporary insurability. Bringing it within reach of insurance will depend not on the boldness of underwriters, but on the progress of knowledge, from debris tracking to the modeling of cascades, for it is the capacity to measure that will open, or close, the capacity to cover. The frontier of space insurance runs through the models.

This primacy of knowledge over capacity nonetheless opens a constructive prospect. If the insurability of the orbit depends on the quality of observation and modeling, then investment in debris tracking, in data-sharing among operators and in research on cascades is the very condition of a viable market. The insurer thus has a direct interest in financing and requiring the transparency of orbital data, for every advance in measurement pushes back the frontier of the uninsurable. Space insurance could then cease to be a mere spectator of the risk to become an agent of its reduction, by making its cover conditional on sound practices and rewarding the most diligent operators, in a logic of prevention already encountered elsewhere in this series. The sky will become insurable only as it becomes legible.