Digital resources in the Social Sciences and Humanities OpenEdition Our platforms OpenEdition Books OpenEdition Journals Hypotheses Calenda Libraries OpenEdition Freemium Follow us

Risques climatiques : la donnée et l’intelligence artificielle ne remplaceront pas la solidarité

Notre tribune, Risques climatiques : la donnée et l’intelligence artificielle ne remplaceront pas la solidarité, écrite avec Laurence Barry, a été publiée par L’Argus de l’Assurance.

À mesure que les risques climatiques s’intensifient, l’essor des données fines et de l’intelligence artificielle promet une meilleure connaissance des expositions. Mais derrière cette avancée technique se joue un choix fondamental : préserver la logique de solidarité qui fonde l’assurance ou glisser vers une individualisation accrue des risques, au risque d’exclure les plus vulnérables ?

EGU’26, When Climate Hazard Granularity Challenges Risk Pooling: A Spatial Perspective

Tomorrow, Raphaël Dalbarade will attend the European Geosciences Union (EGU) Annual Meeting, in Vienna, to present “When Climate Hazard Granularity Challenges Risk Pooling: A Spatial Perspective“, a joint work with Laurence Barry, Caroline Hillairet, Hamza El Hassani, Azeddine Bamansor, Quentin Hénaff, and Simon Blaquière. More very soon…

The French “CatNat” regime provides mandatory natural disaster coverage based on national solidarity, using a uniform rate for all homeowners. However, the increasing availability of high-resolution geoscience data challenges this uniformity. This is notably the case for Clay Shrink-Swell (CSS) risk, which has become a primary cost driver in the last years. Does the shift from national pooling to granular risk segmentation threaten the viability of such solidarity regimes?

To answer this question, we combine empirical analysis with theoretical modeling. First, utilizing a large-scale collection of insurance quotes, we identify a fragmented market where insurers leveraging granular hazard maps coexist with traditional “pooling” actors. Second, to capture the long-term dynamics of this fragmentation, we develop a game-theoretic model of market equilibrium. This model allows us to explicitly simulate how risk selection strategies impact affordability and access to coverage. Our findings suggest that while granular segmentation improves pricing accuracy, it risks creating “insurance deserts” for vulnerable areas. Finally, this technical evolution undermines the regime’s solidarity principle, potentially reducing the socio-economic resilience of communities facing increasing climate geohazards.

Insurance, Biases, Discrimination and Fairness

Two years ago, my book “Insurance, Biases, Discrimination and Fairness” was published in the Springer Actuarial series.

Discrimination in insurance is a difficult topic because, in a very specific sense, insurers are expected to discriminate: they classify risks, build risk pools, and differentiate premiums. This is the logic of risk-based pricing and actuarial fairness. But insurance is not only about pricing risks accurately (accuracy is overrated). It is also about mutualization, risk sharing, and solidarity. The real question is therefore not simply whether insurers should differentiate, but which differences should matter, which should not, where the limits should be drawn, and how to navigate a complex world in which several notions and metrics of fairness coexist, sometimes in tension with one another.

In the book, I tried to connect actuarial pricing, statistical discrimination, legal constraints, algorithmic fairness, explainability, mitigation techniques, and the limits of “fairness through unawareness”. I also discuss a dimension that is often overlooked: causality. Correlation may be useful for prediction, but prevention, explanation, and fairness often require asking what mechanism lies behind an observed association.

For those interested in the mathematics, I have also made lecture notes freely available.

Couvrir les risques de catastrophes naturelles au Japon

Un rapport présentant l’assurance des catastrophes naturelles au Japon est en ligne dans la collection des documents de la Chaire PARI,

Cette note montre que le Japon a construit, face à une exposition exceptionnelle aux séismes, tsunamis, volcans, inondations et glissements de terrain, un système de couverture des catastrophes naturelles original, qui ne repose ni sur un régime unifié de type français ni sur le seul marché privé. Après avoir rappelé la géographie physique et administrative du pays, puis les spécificités du marché immobilier japonais (abondance relative du parc, forte dualité entre propriété et location, faible valorisation du bâti ancien, importance des logements vacants), elle explique que l’assurance habitation japonaise est structurée autour d’un socle de fire insurance couvrant les périls ordinaires et climatiques, auquel s’ajoute un régime spécifique d’assurance séisme, facultatif mais adossé au contrat incendie, partiellement indemnitaire et massivement réassuré par l’État. L’étude montre ensuite, risque par risque, que le séisme constitue le cœur du modèle japonais, avec un partenariat public-privé très institutionnalisé, tandis que l’inondation relève surtout du marché privé complété par des aides publiques ex post ; le tsunami et le volcan, eux, sont couverts juridiquement par le régime séisme mais gérés administrativement par des dispositifs autonomes de surveillance, d’alerte, d’évacuation et de cartographie. Au total, cette note défend l’idée que la “couverture” japonaise ne se réduit jamais à l’assurance : elle combine indemnisation privée, réassurance publique, réserves prudentielles, soutien budgétaire, hazard maps et gouvernance territoriale, dans le but moins de compenser intégralement les pertes que de stabiliser rapidement les conditions de vie et d’organiser la reconstruction après catastrophe.

IFoA AI Ethics Governance & Risk Management forum

I was invited to give a talk at the “IFoA AI Ethics Governance & Risk Management forum” at the end of the week. I uploaded some slides to start the discussion.

I will not give a highly technical lecture, but rather to propose a way of thinking about fairness in insurance AI that may be useful for actuaries, for governance people, and for anyone involved in model risk or oversight. The main message is simple. In insurance, fairness cannot be reduced to predictive accuracy. And once we move to modern AI and large data environments, simply removing a sensitive variable is no longer enough to prevent discrimination. So the real question becomes: how should we govern the trade-offs between fairness and accuracy, and different notions of fairness?

Let me start with a very simple example, illustrating the fact that simply removing a sensitive variable is not enough to prevent discrimination. Here I take a French motor insurance dataset. In the raw data, average claim frequencies are about 8.94% for men and 8.20% for women. Now suppose that I estimate a logistic model for annual claim frequency, but I explicitly exclude gender from the explanatory variables. At first sight, this looks like the fair thing to do. I am not using gender, so maybe I am safe. But what happens?

As I add more and more variables to the model, the predicted frequencies for men and women move closer and closer to the empirical frequencies in the data. With enough explanatory variables, I am basically reconstructing the original gap. So the model is not using gender explicitly, but it is learning from variables that are correlated with gender. In other words, the information has not disappeared. It has simply been redistributed across other variables.

That is why I think the old defence of ‘we do not use the sensitive variable, therefore the model is fair’ has become much weaker in a big-data context (and here, it’s not big, it’s only 15 basic explanatory variables). Modern predictive models are very good at finding statistical associations. If protected traits leave traces in the data, the model will usually pick them up.

And this is not only a technical issue. It is also a philosophical one. The model tends to reproduce what is in the data, that’s what we call ‘generalization’ in machine learning. If historical inequalities are present in the data, the model will often learn them very efficiently. This is why I mention Hume here, and the old ‘is-ought’ problem. From the fact that a disparity exists in the data, it does not follow that it should be reproduced in pricing. So this slide is really the starting point for the rest of the talk: fairness through unawareness is often not fairness at all; it is frequently a way of hiding the channel through which unfairness reappears.

So once we accept that unawareness is not enough, we need to ask a more difficult question: what exactly do we mean by fairness?

Before going to the recent actuarial papers, I want to spend one minute on a classical dilemma. This is not new. In legal and philosophical debates, there has long been a tension between two positions.

One position says: if you want to correct unequal outcomes, you may need to take account of the protected characteristic. That is the logic behind many discussions of affirmative action or corrective treatment. If the world is already unequal, pretending not to see the relevant characteristic may simply reproduce the inequality. The opposite position is the so-called colorblind view: the way to stop discrimination is to stop using the sensitive characteristic altogether. That sounds very appealing, because it seems neutral and simple.

The trouble is that in insurance, and especially in AI-based pricing, both positions run into difficulties. If you ignore the sensitive variable, you may still reproduce its effects through proxies, that’s what we’ve seen before. But if you use it for corrective purposes, you are entering a difficult space involving direct differential treatment, redistribution, and legal or ethical contestability.

So what I want to emphasize here is that the fairness debate is not just about technical metrics. It is rooted in a much older normative tension: should fairness mean treating everyone under the same formal rule, or should it sometimes mean treating people differently in order to offset structural inequalities? I think that this dilemma is exactly why actuarial fairness debates cannot be settled by model performance alone.

In the contexte of insurance, the recent SOA report discussed in a previous meeting stresses this important tension fairness is not one thing.

The report is useful because it is very practical and very clear. Its main point is that fairness in life insurance is not a single concept. There is no universal fairness metric that would solve the problem once and for all. Instead, the paper distinguishes between different families of fairness criteria.

A first distinction is between individual fairness and group fairness. Individual fairness is close to the actuarial intuition: similar risks should be treated similarly. Group fairness, by contrast, is about parity across groups. Those two perspectives can both sound reasonable, but they are often in tension. The paper also reviews several criteria. On the individual side, it discusses unawareness, awareness, and omitted-variable bias. On the group side, it discusses independence, sufficiency, and separation.

The most useful takeaway is not any particular metric. It is the methodological lesson: before choosing a metric, you must define what notion of fairness you are trying to serve. Otherwise you are optimizing a number without knowing what ethical or institutional objective it really represents. The SOA paper gives us a map. It tells us: do not look for the one true fairness metric. Clarify the objective first, and then be explicit about the trade-offs.”

Our own recent work tries to push that argument one step further, by asking what the core dimensions of fairness in insurance pricing really are…

In our recent paper on what we call, with Olivier Côté, and Marie-Pier Côté, “the fairness trilemma“, we recall the idea that insurance is not just a predictive exercise. It is also a social institution. Insurance has a double nature. On the one hand, it is about risk-based pricing: aligning premiums with expected losses. On the other hand, it is also about risk-sharing: spreading burdens, preserving access, and sometimes accepting cross-subsidies. From there, we argue that fairness in insurance pricing is governed by a trilemma between three principles.

  • The first is actuarial fairness. In simple terms, this means that premiums should reflect expected losses as accurately as possible.
  • The second is social solidarity. This means that we may accept departures from strict risk-based pricing in order to preserve access, affordability, or some broader social objective.
  • The third is causal legitimacy. This is extremely important in the AI context. Not every predictive variable is equally legitimate. A variable may improve predictive performance, but if it works mainly by reconstructing a protected characteristic, or if its causal meaning is questionable, then its legitimacy is fragile. So causal legitimacy asks not only whether a factor predicts, but whether it deserves to shape the price.

The key point is that these three principles are all attractive, but they cannot be fully satisfied at the same time. If I strengthen actuarial fairness, I may weaken solidarity. If I enforce more solidarity, I move away from pure risk adequacy. If I become very demanding about causal legitimacy, I may sacrifice some predictive performance and perhaps some segmentation logic. So the problem is not to find the perfectly fair model. The problem is to govern the trade-offs. That is why I insist on governance tools: scorecards, causal due diligence, comply-or-explain procedures. The point is to make choices explicit, reviewable, and contestable. In that sense, fairness is not just a modelling problem. It is a governance problem.

But of course, at this stage, someone in the room may say: all this sounds interesting, but how do we make it operational? How do we move from philosophy to something an actuary can actually measure?

That is exactly the purpose of the CAS paper. The starting point is again that removing a sensitive variable does not remove discrimination, because allowed variables may still act as proxies. So the challenge is to expose indirect discrimination in a way that is usable in actuarial practice. The objective of the paper is to make fairness operational and measurable in actuarial terms. It is not enough to speak in abstract statistical language. In insurance, we need to know what unfairness means in premiums, in dollars, and across segments of policyholders.

The paper is based on a real auto insurance case study from Québec, using credit score as the sensitive variable. And it again organizes the discussion around three dimensions: actuarial fairness, social solidarity, and causality legitimacy. This is important because it shows continuity with our more conceptual framework. The toolbox is not something separate. It is a way of translating those dimensions into measurable diagnostics. If the trilemma paper says fairness must be governed, the CAS paper says: here are some tools that help you see what you are governing.

One of the most interesting contributions of the toolbox is that it introduces local metrics such as risk spread, proxy vulnerability, fairness range, and parity cost. Let us focus on two of them.

The first is proxy vulnerability. Intuitively, proxy vulnerability measures how much a segment may be over-priced or under-priced because apparently neutral variables indirectly reconstruct the sensitive attribute. This is very valuable because it moves us away from vague concerns about bias and toward a concrete actuarial question: where, and by how much, might the pricing rule be unfair?

The second is parity cost. This is also very important because it forces us to be honest. If we want more parity across groups, that is not free. Someone has to bear the cost. In plain English, enforcing one notion of fairness may end up robbing Peter to pay Paul. And I do not say this as a criticism; I say it because redistribution should be explicit rather than hidden.

Another strength of the toolbox is that it helps actuaries identify local pockets of unfairness. A model may look acceptable on average across broad groups, but still be very problematic in specific vulnerable niches. So fairness should not be assessed only globally. Local diagnostics matter.

This is also, I think, very relevant for AI governance. Senior management and boards do not just need a statement saying ‘the model passed a fairness test’. They need to know where the model is fragile, which subpopulations are exposed, what the monetary magnitude is, and what the trade-offs are if a correction is introduced. So in practice, the toolbox helps turn fairness into something discussable in risk committees and governance processes.

To wrap up, first, the broader background for all this is developed in my recent textbook, which tries to connect legal, philosophical, statistical, and actuarial perspectives on discrimination in insurance.  Second, fairness should not be assessed only at the level of one insurer’s portfolio. A pricing rule may look fair on one portfolio and still be unfair at the market level, because portfolios are not representative of the insured population. This is a selection-bias problem. If different insurers attract different segments of the market, portfolio-specific fairness does not necessarily aggregate into market-wide fairness. So we also need to ask: fair for whom, and on which reference population?

Third, there is always a cost and a redistribution issue. If we enforce one notion of fairness, someone pays. That does not mean fairness is undesirable. It means that fairness choices are unavoidably political, institutional, and governance choices, not just technical ones. And fourth, there is a major practical difficulty when sensitive variables are unobserved. Very often, people say: let us not collect the sensitive variable. But then how do we audit fairness? How do we provide evidence? How do we challenge the model? If the protected attribute is precisely the thing we do not observe, fairness becomes not just a modelling issue, but a data, audit, and governance issue.

So my overall conclusion would be the following. AI did not create the fairness problem in insurance, but it has made it sharper. It has weakened the old comfort of unawareness. It has increased the power of proxies. And it has made the trade-offs more difficult to hide. For actuaries, I think this creates both a risk and an opportunity. The risk is to treat fairness as a box-ticking exercise or as a purely technical constraint. The opportunity is to contribute something distinctive: a language of prices, trade-offs, portfolios, and governance. So perhaps the right ambition is not to promise a perfectly fair model. It is to build institutions that are capable of identifying proxy discrimination, making normative choices explicit, and governing them responsibly.

Will Technology Save Us?

This post was initially written in French, La technologie nous sauvera-t-elle ?

I feel as though I keep hearing, more and more often, that climate change is above all a problem of innovation. Emissions continue to rise, targets keep slipping out of reach, yet we fill our collective imagination with carbon-capture machines, artificial intelligences supposedly able to optimize the transition, and even technologies designed to alter the climate itself. There is nothing absurd about such confidence in itself, and technology will very likely help. But it becomes politically suspect when it serves mainly to postpone difficult questions, beginning with this one: what are we willing to change, here and now, in the way we produce, consume, and govern? The literature on “mitigation deterrence” helps us understand how the promise of a future intervention can legitimize delaying present efforts. Nor is this mechanism unique to climate change. The COVID pandemic, it seems to me, offered a strikingly similar scene, in which fascination with the biomedical response sometimes pushed into the background the social, institutional, and political tools that the strongest research nevertheless regarded as indispensable.
Continue reading Will Technology Save Us?

La technologie nous sauvera-t-elle ?

J’ai l’impression d’entendre de plus en plus souvent dire que le climat est avant tout un problème d’innovation. Les émissions continuent d’augmenter, les objectifs se dérobent, mais on essaye de peupler notre imaginaire collectif de machines à capturer le carbone, d’intelligences artificielles capables d’optimiser la transition, voire de techniques de modification du climat. Cette confiance n’a rien d’absurde en elle-même, et il y a fort à parier que les technologies aideront. Mais elle devient politiquement suspecte lorsqu’elle sert avant tout à repousser des questions difficiles, à commencer par “que sommes-nous prêts à changer, ici et maintenant, dans nos manières de produire, de consommer et de gouverner ?” La littérature sur la “mitigation deterrence” a permet de mieux comprendre la promesse d’une intervention future légitimant de retarder l’effort présent. Ce mécanisme n’est pas propre au climat, il me semble que la pandémie de COVID nous a offert une scène assez proche, où la fascination pour la réponse biomédicale a parfois relégué au second plan les instruments sociaux, institutionnels et politiques pourtant jugés indispensables par les travaux les plus solides .
Continue reading La technologie nous sauvera-t-elle ?

Exposé “Risque climatique, retrait des assureurs et granularité des tarifs” pour la Chaire PARI

Mercredi, je donnerai la première partie de l’exposé Risque climatique, retrait des assureurs et granularité des tarifs, organisé par la Chaire PARI. Je donnerai un point de vue un peu général sur le problème qui nous préoccupe, à savoir la modélisation d’un marché concurrentiel d’assurance, et la recherche de politiques optimales, pour un régulateur, pour que l’équilibre concurrentiel soit optimal (ou a minima améliore certains critères) pour le bien être global. Raphaël Dalbarade présentera ensuite ses travaux sur le sujet.

Faut-il socialiser les risques ou responsabiliser les territoires ?

Publication d’un court article, écrit avec Laurence Barry, en ligne sur le site de la Revue Banque,

Le dérèglement climatique, qui s’accompagne de l’intensification des phénomènes extrêmes, prend de l’ampleur à un moment où les données disponibles concernant ces événements se multiplient à une maille de plus en plus fine. De plus, dans certains pays, et notamment en France, des stress-tests climatiques mis en place ces dernières années ont contribué à une montée en capacité des compagnies d’assurance sur ces modèles.

(à suivre…)

On my way to the University of Hong Kong (HKU , 香港大學)

Today is a long travel day, going from Beijing to Hong Kong (a bit more than 2000km), by train. At least it is a direct train (a night train was mentioned online, but none was proposed). As a comparison, Beijing to Hong Kong by train, that’s like Montréal to New Orleans, or Oslo to Barcelona…

Friday morning, I will give a talk at HKU (slides are online)

On my way to Tsinghua (清华大学), Beijing

Next week, I will be at Tsinghua University in Beijing. On Tuesday, in the early afternoon, I will give three lectures for undergraduate students on the theme: ‘Three lectures on AI and its implications for actuarial (and/or financial) professions.

These lectures explore the relationship between artificial intelligence and insurance. They begin from the observation that insurance has long relied on prediction, classification, and decision-making under uncertainty, well before the recent rise of AI. AI therefore does not introduce these issues from scratch, but changes their scale, granularity, and practical consequences. The lectures review the insurance foundations of pricing and pooling, then examine the main challenges raised by AI, including personalization, selection, causality, bias, fairness, governance, and trust. They finally turn to the concrete uses of AI across the insurance value chain, emphasizing that a good system should not be judged by accuracy alone, but also by its calibration, its fairness, and its ability to support real decisions in practice.

In the evening, I will give a talk at the seminar, at Renmin University of China, on the theme: ‘Using optimal transport to mitigate unfair predictions and quantify counterfactual fairness.’ The first part will revisit topics that I presented in greater detail in the lectures notes of my course this autumn at Kyoto University, particularly the price to be paid in terms of accuracy in order to achieve fairness. The second part will discuss the paper ‘Sequential Transport for Causal Mediation Analysis,’ which was posted online a few days ago.

On Wednesday, I will have in-depth academic exchange session with students from the Tsinghua Actuarial Science Association, at Tsinghua University.

Fair regression under localized demographic parity constraints

Our paper, Fair regression under localized demographic parity constraints, with Christophe Denis, Romuald Elie, Mohamed Hebiri and François Hu, is now available online, on arXiv,

Demographic parity (DP) is a widely used group fairness criterion requiring predictive distributions to be invariant across sensitive groups. While natural in classification, full distributional DP is often overly restrictive in regression and can lead to substantial accuracy loss. We propose a relaxation of DP tailored to regression, enforcing parity only at a finite set of quantile levels and/or score thresholds. Concretely, we introduce a novel (\boldsymbol{\ell},\mathcal{Z})-fair predictor, which imposes groupwise CDF constraints of the form F_{f\mid S=s}(z_m)=\ell_m for prescribed pairs (\ell_m,z_m). For this setting, we derive closed-form characterizations of the optimal fair discretized predictor via a Lagrangian dual formulation and quantify the discretization cost, showing that the risk gap to the continuous optimum vanishes as the grid is refined. We further develop a model-agnostic post-processing algorithm based on two samples (labeled for learning a base regressor and unlabeled for calibration), and establish finite-sample guarantees on constraint violation and excess penalized risk. In addition, we introduce two alternative frameworks where we match group and marginal CDF values at selected score thresholds. In both settings, we provide closed-form solutions for the optimal fair discretized predictor. Experiments on synthetic and real datasets illustrate an interpretable fairness-accuracy trade-off, enabling targeted corrections at decision-relevant quantiles or thresholds while preserving predictive performance.

"sendo l'intento mio scrivere cosa utile a chi la intende…"