On Monday of this week, I returned to Québec to defend my Ph.D. thesis, “Investigating High-Dimensional Problems in Actuarial Science, Dependence Modelling, and Quantitative Risk Management.”

This event marks the end of my 9-year journey at Université Laval, completing a bachelor’s degree, two master’s degrees and a Ph.D. within my hometown university.

Here are the slides for my defence.

I spent the month of May in Sydney, Australia for a research visit to UNSW and the International Congress of Actuaries. Thanks to Bernard Wong for inviting me to UNSW and Patrick Laub, Eric C.K. Cheung and Fei Huang for the research chats.

This week, I’m starting a research visit to ETH Zürich, where I am visiting Prof. Mario Wüthrich and RiskLab for two months. We will work on uncertainty-aware insurance pricing with deep neural networks. We aim to provide reliable uncertainty estimates for frequency and severity predictions from neural networks and provide tools for actuaries to know when a predictive model is confident or not in its predictions.

Last week was busy but exciting! I gave a seminar to the insurance group of KU Leuven. Then I went to London for CMStatistics, a huge statistics conference with almost 2000 participants. It was great to reunite with friends and meet new researchers. The slides for both talks are attached to this post.

I’m returning to the CREST (ENSAE Paris) for a four-month visit with Christian Y. Robert. I’m looking forward to spending the winter in France. We will be continuing our work on estimating high-dimensional FGM copulas.

Thanks to the Centre de recherches mathématiques for sponsoring this visit.

I just finished a visit to Europe for conferences and visits.

• It was great to meet the participants of EAJ 2022 in Tartu, Estonia.
• I returned to the CREST (ENSAE Paris, France) for a three-week visit.
• I went to Agistri, Greece for a workshop on dependence modelling.

Thanks to the Chaire d’actuariat de l’Université Laval and my co-supervisor Hélène Cossette for sponsoring this great opportunity.

I’m in Milano this week for the Insurance Data Science conference. I presented some ongoing research on using house images to predict insurance premiums. So many exciting talks about machine learning and actuarial science, it’s great to see the direction of our field.

Thanks to the project Analyse de données massives en assurance (IID, NSERC, Intact) for funding my conference visit.

My paper on the stochastic representation of Farlie-Gumbel-Morgenstern (FGM) copulas has just been accepted to Computational Statistics and Data Analysis. This is joint work with my PhD supervisors Hélène Cossette and Etienne Marceau.

The FGM copula is one of the oldest implicit copulas. While easy to analyze mathematically, this copula did not have a probabilistic interpretation, limiting its application in high-dimensional statistical applications. This paper proves a one-to-one correspondence between the class of FGM copulas and multivariate symmetric Bernoulli random variables. The representation allows us to derive new results on dependence ordering and measures of association. Further, we propose new subfamilies of parsimonious FGM copulas. Finally, we explore sampling procedures, which enable us to perform high-dimensional simulation for some subfamilies of FGM copulas.

This is the first paper in our series on FGM copulas, stay tuned for more contributions in applied probability, actuarial science and high-dimensional statistics.

Link: https://authors.elsevier.com/c/1e~Fl_3qPK302M

I’m currently visiting the Center for Research in Economics and Statistics (CREST) in Palaiseau, France (just south of Paris). I’ll be working with Christian-Yann Robert on high-dimensional copulas based on graphical models for the next two months.

Our paper on geographic ratemaking was recently published online in the ASTIN Bulletin. We present a method to leverage external and unstructured spatial data to improve spatial ratemaking.

We decompose the ratemaking process into two steps. First, we train spatial embeddings that capture spatial effects from the external data. The embeddings capture the spatial nature of the data. Then, we use the spatial embeddings as input variables to a generalized linear model.

The great advantage of our approach is that the regression model does not require a spatial component, since the spatial embeddings exhibit all desirable attributes of spatial models.

Link to paper

Here are the slides I prepared for the annual meeting of the Statistical Society of Canada, which earned the Actuarial Science Student Research Presentation Award.