Tag Archives: SCOR

SCOR Foundation – Scope and limits of Artificial intelligence

On May 15, 2024, the SCOR Foundation for Science hosted a webinar titled “Scope and limits of Artificial intelligence”, delivered by Arthur Charpentier. A professor in the Department of Mathematics at the University of Quebec in Montreal and a member of the Institute of Actuaries, Arthur Charpentier is an internationally recognized expert in actuarial science and the author of numerous academic articles published in the best actuarial academic journals worldwide.

During the webinar, Arthur Charpentier discussed the research project “Fairness of predictive models: an application to insurance markets”, which is supported by the SCOR Foundation for Science. This project addresses biases within the automatic artificial intelligence algorithms utilized to determine optimal pricing in individual policies. Its aim is to mitigate or eliminate such biases, which could lead to inequities or discriminatory practices based on factors such as gender, race, religion, or origin in the coverage provided by insurers or reinsurers to policyholders.

“Scope and limits of artificial intelligence” at the SCOR foundation monthly webinar

This morning, I will give a talk on “scope and limits of artificial intelligence” at the SCOR foundation monthly webinar. As discussed previsously, we currently have ongoing research on discrimination and fairness founded by the fondation (newsletter #1 is online).

Insurance (and further motivations)

Since we will talk about fairness, I will start with a couple of motivations. The first one is about COMPAS,

Interestingly, we have the data to analyse that one. In the original analysis, conditional on non-re-offending, proportions of being wrongly classified in the two protected groups are significantly different, so the algorithm is racist,

The answer was that actually, conditional on being classified as high risk, the probability of re-offense in the two protected groups are significantly similar, so the algorithm is not racist,

So clearly, we can start to see that it will not be so easy, since using the same data and the same models, two different conclusions can be obtained.

We will also disccuss legal aspects.

This idea of “determining actuarial factor” has been remove in Europe, but we can still find it in Québec

I can also mention some recent projects, in Colorado, where insurers are asked  to predict race and ethnicity (that specific topic is on our agenda for the summer)

And finally, I should stress that discrimination has not much to do with the intention of the statistician. This is the idea of indirect discrimination

I should also mention “redlining“. About 100 years ago, in the US, we started to see maps, created by HOLC (based on City Survey Files, 1935-1940). Those maps contained “red” areas and “green” areas. Bankers were supposed to avoid the red areas, because they were considered too risky.

As a sidenote, we see nowadays some blue-lining related to climate risks,

“Blue-lining,” from the consumer’s perspective, is when banks or mortgage lenders draw lines of risk around certain streets or neighborhoods, often without clear disclosure.

Finally, I just want to recall that algorithms just tend to reproduce what can be observed in data. If there is a difference between men and women, they will reproduce it.

A bit more on insurance

I should also stress an important problem (that could be related to a paper we wrote, in French, a few years ago). Classically, when modeling categorical variables, such as a binary variable y\in\{0,1\}, practitionners just care about getting the good category. On the left, we have pictures of cats and dogs to train a model, then we try on a new picture, that is either a cat or a dog. Somehow, there is a ground truth and it is possible to see if we are right or wrong. Same if we want to detect a disease on medical pictures. Now, if we move to the right. In the middle, we have a model that predicts if it will rain, or not. But here, maybe, what we care about is actually the probability to have rain. On the right, we have the actuarial problem of modeling claims frequencies. We do not want to predict who will claim a loss, but we want a good estimator of the probability to claim a loss. The challenge, clearly, is that we cannot observe that one. We cannot observe the latent risk factor. We only observe if people got an accident or not. But some people with a very small probability can still claim a loss. And very bad drivers can actually be very lucky, and got no accident one year.

Again, in insurance, we care more about the score, the estimation of the probability than the class \widehat{y}. So we can slightly modify standard fairness definitions, to be based not on predicted classes \widehat{y}, but on the score m(\boldsymbol{x},s). As we will discuss, there are usually three general definitions of so-called “group fairness”

Quantifying unfairness with optimal transport

Let us start with demographic parity. A weak version is that, on average, scores in the two groups should identical (or close). An alternative is the strong version, asking for equalities in distributions : for any set \mathcal{I}\subset[0,1], the probability that the score is in \mathcal{I} (e.g. between 40% and 60%) should be the same in the two groups.

Mathematically, we need a distance between the distributions of scores in the two groups. And a popular distance is Wasserstein distance, that is related to optimal transport.

The empirical version is perhaps easier to understand, and mapping is based on matching of individuals. xxx

As a cultural sidenote, a couple of slides to explain why it has to do with “optimal transport”, going back to Monge (1781)‘s problem. It’s all about transporting the sand, grain by grain, from the hole to the pile. Below, we have a (purely) random transport. Which is not efficient at all…

and then the optimal version (for a strictly convex cost function), he leftmost grain in the hole goes on the lefttmost part of the stack, etc.


For mitigation (once we have observe that there was discrimination, as discussed previously) heuristically, we want to be somewhere in between the two distributions in the two subgroups,

Being “in between” can be interpreted locally: for someone in group A, it should be between (weights are related to proportions in the two groups) the prediction, as someone in group A, and then some sort of counterfactual in the other group, namely the prediction that person would have obtained if she had been in group B, based on the same probability level,

For the other group it is the opposite

Beyond demographic parity

If we get back to our COMPAS examples, demographic parity, in the standard classification-based definition, would be translated as

If we get back to the original motivation we gave, it had nothing to do with demographic parity, the first slide had to do with separation, or equalized odds, while the second one had to do with sufficiency, or calibration.

More generally, if we consider a weak version of the independence criterias, we have moments equality, within each protected subgroup,

Let us mention a bit more calibration. Calibration is deeply related to the interpretation of “probabilities” as returned by models as “real probabilities”. In machine learning, it is hard to define properly what those “probabilities” are.

Calibration is related to the following idea, discussed above: if we consider all cases where the predicted probability was 40% (or say, close to 40%), then the proportion on 1’s should be close to 40%.

To conclude that disgression, I can mention the following example highlighting that we should be concerned by probabilities returned by machine learning algorithms. Consider some pictures, generated by some algorithm, and more precisely, some flow of pictures, from a woman to a man

Below, we can see probabilities given by some online appplication, that returns probabilities to be a woman, given a picture. Can’t we agree that it is surprising that those probabilities (of beeing a woman) do not decrease continuous, from the picture in the top left corner and the one in the bottom right one ?

Finally, I can also mention “individual fairness”, or “counterfactual fairness”. Here also, optimal transport can be used, to quantify counterfactual unfairness. But I won’t be too long here.

Finally, an opening for next year’s agenda, with interpretability. Interpretability is a very important issue in actuarial science, which is not as objective as people might think, and the popular

let the data speak for itself

In insurance, interpretation is very important, probably more important than model assumptions

Interpretation become a key concept when dealing with multiple sensitive attributes

To conclude, just a final reminder that dealing with mitigation is a complex philosophical problem….

Tomorrow, we will discuss further at our workshop, in Québec city

Fondation SCOR, Fairness of predictive models: an application to insurance markets

The Scientific Council of the SCOR Foundation has decided to fund the research project “Fairness of predictive models: an application to insurance markets” until its anticipated completion date in three years (2023-2025). The project will be led by the University of Quebec and directed by Arthur Charpentier, professor in the mathematics department of the University of Quebec in Montreal. This project aims to propose corrections to the automatic artificial intelligence algorithms that can be used to determine the optimal pricing of individual policies in order to remove or limit the biases likely to generate inequities or even discrimination based on gender, race, religion, origin, etc. in the coverage offered by insurers or reinsurers to policyholders. The subject is of both theoretical (better control of black boxes constituted by models based on artificial intelligence algorithms) and practical (reduction of the risks of discrimination and inequity) interest. From this point of view, it is very topical for insurers and reinsurers facing major reputational challenges in the context of the growing importance of social networks. In addition to his role at the University of Quebec, Arthur Charpentier is a member of the Institute of Actuaries, internationally recognized expert in actuarial science, author of numerous academic articles published in renowned academic actuarial journals in both nationally and internationally.

Assurance collaborative, théorie des graphes et actuariat

Mardi prochain, j’interviendrais (en visio) au colloque SCOR sur le thème “Actuariat, effets réseaux et théorie des graphes

Mes slides (on m’a demandé de parler sur le thème assurance collaborative, théorie des graphes et actuariat) sont en ligne, je présenterai notre papier Collaborative Insurance Sustainability and Network Structure, mais je peux en profiter pour mentionner d’autres articles, dont modéliser la contagion, ou les réseaux pour réinventer l’assurance?

Modeling analogies in life and nonlife insurance

On Wednesday afternoon I will be giving a talk at the SCOR Reserving Seminar. The talk will be on modeling analogies in life and nonlife insurance. We will start by discussing data analogies, based on the Lexis diagram in life insurance and in nonlife (when modeling claims dynamics),

This will induce similarities in datasets used in life models, and in nonlife reserving

Further, in the two cases, logPoisson models are usually used, either to model the number of deaths, or the amount of payment. The main difference is that in nonlife insurance, forecasting future payments is rather simple,

But in life models, unfortunately, we need to forecast the behavior of year based parameters.

Note that this is also the case in nonlife insurance when an inflation factor is introduced.
To go further, the slides are available here.

Les généralités sont généralement fausses

Suite à un commentaire sur un vieux billet (ici), concomitante avec une demande d’intervention pour la formation ERM proposée par l’Institut des Actuaires, je me suis replongé dans les calculs de SCR (solvency capital requierement). Plus je lis la documentation technique, plus je suis surpris par l’importance de l’hypothèse de normalité ! J’ai repris mon Loss Model, qui est le livre de base pour l’examen d’actuariat non-vie dans les pays anglo-saxons, et j’ai vu des dizaines de lois bizarres… mais rien sur la loi normale…

  • de l’hypothèse de normalité dans la formule d’agrégation

Comme je le disais dans un vieux billet (ici), on ne peut montrer la validité de la formule de calcul du SCR que dans le cas Gaussien (ou plus précisément dans le cas elliptique, voire ici). En effet, la preuve est assez simple. Je renvois vers un papier tout chaud de Laurent Devineau et Stéphane Loisel sur le sujet, ici.

  • où l’on continue à entrevoir l’hypothèse de normalité

En fait l’hypothèse de normalité est vraiment partout dans les calculs de SCR. Mais bien sûr, sans le dire. Par exemple dans un papier de la SCOR,


Comme je le disais l’autre jour (ici), le problème des documents “professionels” (par opposition à “académiques“), c’est que comme on veut que le CEO le lise, on évite de parler d’hypothèses, ou pire, de mettre un peu de maths…. Bref, on apprend que “generally” (terme d’une grande rigueur scientifique, que je recaserai un jour dans mes démonstrations), “the expected shortfall[…] at the 99 % level corresponds quite closely to the[…] value-at-risk at a 99.6% level “. Mais on retrouve la même phrase dans un rapport de l’edhec


Là aussi, aucune source… Damned ! En creusant un peu, Actuaris m’a mis sur la piste (ici), car on apprend que c’est l'”autorité suisse” qui a énoncé ce résultat que je qualifierais de surprenant1.


Effectivement, on peut retrouver le document originel,


J’avoue avoir essayé – sans grand succès – de lire l’annexe D. Je n’ai pas tout compris, mais j’y ai lu un “assuming that the distribution functions follows a normal law”,


Donc si j’ai bien tout compris, “generally”, “the distribution functions follows a normal law” ! Bon, je crois que je vais pouvoir envoyer un mail à Stuart, Harry et Gordon pour leur expliquer qu’ils peuvent virer des pages à leur livre ! Ou peut-être Jean-Luc Besson (pour lequel j’ai une immense estime !) qui présente beaucoup de lois utiles en assurance dans l’ouvrage qu’il a coécrit avec Christian Partrat…
Mais peut-être a-t-il raison…? On peut se dire qu’en actuariat, on avait l’habitude de présenter l’approximation “normal power“, et peut être que ce résultat est vrai pour d’autres lois….
Regardons vite-fait sur quelques lois….. Par exemple, pour la loi lognormale, si on regarde en fonction de l’écart-type de la loi normale sous-jacente,

Effectivement, on colle au quantile à 99.6%. L’ordonnée est en échelle logarithmique pour que le dessin soit présentable, avec en pointillé la VaR (formule exacte, à 99.6% en rouge, et à 99.8% en bleu), et les ronds sont des estimations de la TVaR à 99%. Par contre, par exemple pour la loi lognormale, si on regarde en fonction de l’écart-type de la loi normale sous-jacente, plus la volatilité est grande, plus l’approximation est mauvaise.

Bon, essayons une loi Gamma (classique en assurance, car c’est une loi de la famille exponentielle, et souvent comparée à la loi lognormale).

Visiblement, là aussi l’approximation semble marcher….  mais rappelons que la loi Gamma est une loi à queue fine… Allons-y franchement, comparons avec le cas d’une loi de Pareto…

cette fois, l’abscisse se lit dans l’autre sens: à gauche, c’est le cas le plus risqué (avec une variance infinie entre 1 et 2). On notera que pour les risques les plus importants, là aussi on sous-estime la TVaR en considérant la VaR (à un seuil plus important). On retrouve d’ailleurs exactement le même type de graphique avec une loi de Student,

Bref, on me dira que je pinaille, n’empêche qu’à mon avis, les compangies sous-estiment “généralement” la TVaR (à supposer que ça soit la mesure de risque qui les intéresse) en prenant une hypothèse de normalité…

1 en tant qu’enseignant, ce genre de conclusion me gêne toujours, car je me fatigue à faire des cours sur la mesure de risque, j’essaye de présenter des notions aussi proprement que possible, en expliquant que les hypothèses c’est important, qu’il faut définir différentes notions de mesures de risques (entre les convexes, les cohérentes, les dynamiques, etc), et là tout d’un coup, on nous apprend que “generally” c’est kif kif….