La modélisation de la dynamique des populations de thons et grands pélagiques pour l'évaluation des stocks est confrontée à deux enjeux majeurs. (1) L'hypothèse forte de proportionnalité entre Captures Par Unité d'Effort de pêche (CPUE) et l'abondance des stocks. Les CPUE des pêcheries commerciales sont en effet les seules mesures relatives de biomasse utilisées pour l'évaluations des stocks de thons et grands pélagiques, malgré leur manque de représentativité de l'abondance de ces populations. (2) Le manque de données informatives pour modéliser la relation Stock-Recrutement (SR) ce qui conduit à utiliser des contraintes sur la "steepness" de cette fonction. Nous examinons comment l'introduction d'informations indépendantes des pêcheries commerciales dans les modèles pour l'évaluation des stocks thoniers permet de lever l'hypothèse de capturabilité constante et de mieux justifier le choix de la steepness de la relation SR. Le cadre statistique bayésien autorise la prise en compte d'informations supplémentaires via des distributions a priori informatives (priors). Cette thèse examine donc les possibilités d'élicitation de priors informatifs pour des paramètres démographiques et des paramètres liés à la capturabilité des engins de pêche, ainsi que l'utilisation de ces priors dans un modèle global. Les cas d'études sont les stocks de thon rouge (Thunnus thynnus) et d'albacore (Thunnus albacares) de l'Atlantique. La grande variabilité naturelle des taux de mortalités pré-recrutement pose des limites à l'utilisation des seuls traits d'histoire de vie pour l'élicitation de priors pour des paramètres démographiques. Par ailleurs, la relation SR pour les thonidés est remise en question par une valeur de steepness proche de 1. Il apparait que des priors informatifs sur la capturabilité dans un modèle hiérarchique global permettent de réduire les incertitudes dans le diagnostic sur l'état d'un stock thonier. Nous montrons ainsi que le diagnostic sur le stock Atlantique d'albacore est plus pessimiste qu'attendu la tendance à la hausse des capturabilités des principaux engins de pêche est prise en compte. L'élicitation de priors présente donc un fort intérêt pour utiliser des informations supplémentaires et extérieures aux CPUE et améliorer la perception de l'état des stock thoniers. / Modelisation of the population dynamics of tunas and tuna like species for stock assessment is facing two issues. (1) The hypothesis of proportionality between Catch Per Unit Effort (CPUE) and abundance (constant catchability). CPUEs from commercial fisheries appear to be the only relative measure of abundance in spite of their lack of representativity of the abundances of the populations. (2) The lack of informative data for the modelisation of the Stock-Recruit (SR) relationship, which leads to constraint this function on its steepness. The introduction of fisheries-independent sources of information is investigated in order to relax the assumption of constant catchability and to provide better justification of steepness choice for the SR relationship. The Bayesian statistical framework allows the consideration of additional information a priori via informative distributions (priors). This work investigate the elicitation of informative priors for demographic parameters and parameters related to the catchability of fishing gear, as well as the use of these priors into a surplus production model. The cases of the Atlantic bluefin tuna (Thunnus thynnus) and of the yellowfin tuna (Thunnus albacares}) were taken as examples. The large natural variability of pre-recruits mortality rates limits the use of life history traits for eliciting priors for demographic parameters. In addition, the SR relationship for tuna is challenged by a steepness value close to 1. It appears that informative priors on catchability parameters, in a hierarchical surplus production model, reduce uncertainties in the diagnosis on the status of tuna stocks. We show that the status of the Atlantic yellowfin tuna stock is more critical taking into account upward trends in the main fishing gears catchabilities. We conclude that prior elicitation is a reliable tool to take into account additionnal information and to improve tunas stock assessment.

<p dir="ltr">In this thesis, we delve into the exploration of several nonparametric and semiparametric econometric models within the Bayesian framework, highlighting their applicability across a broad spectrum of microeconomic and macroeconomic issues. Positioned in the big data era, where data collection and storage expand at an unprecedented rate, the complexity of economic questions we aim to address is similarly escalating. This dual challenge ne- cessitates leveraging increasingly large datasets, thereby underscoring the critical need for designing flexible Bayesian priors and developing scalable, efficient algorithms tailored for high-dimensional datasets.</p><p dir="ltr">The initial two chapters, Chapter 2 and 3, are dedicated to crafting Bayesian priors suited for environments laden with a vast array of variables. These priors, alongside their corresponding algorithms, are optimized for computational efficiency, scalability to extensive datasets, and, ideally, distributability. We aim for these priors to accommodate varying levels of dataset sparsity. Chapter 2 assesses nonparametric additive models, employing a smoothing prior alongside a band matrix for each additive component. Utilizing the Bayesian backfitting algorithm significantly alleviates the computational load. In Chapter 3, we address multiple linear regression settings by adopting a flexible scale mixture of normal priors for coefficient parameters, thus allowing data-driven determination of the necessary amount of shrinkage. The use of a conjugate prior enables a closed-form solution for the posterior, markedly enhancing computational speed.</p><p dir="ltr">The subsequent chapters, Chapter 4 and 5, pivot towards time series dataset model- ing and Bayesian algorithms. A semiparametric modeling approach dissects the stochastic volatility in macro time series into persistent and transitory components, the latter addi- tional component addressing outliers. Utilizing a Dirichlet process mixture prior for the transitory part and a collapsed Gibbs sampling algorithm, we devise a method capable of efficiently processing over 10,000 observations and 200 variables. Chapter 4 introduces a simple univariate model, while Chapter 5 presents comprehensive Bayesian VARs. Our al- gorithms, more efficient and effective in managing outliers than existing ones, are adept at handling extensive macro datasets with hundreds of variables.</p>