Les modèles d'équations structurelles à variables latentes permettent de modéliser des relations entre des variables observables et non observables. Les deux paradigmes actuels d'estimation de ces modèles sont les méthodes de moindres carrés partiels sur composantes et l'analyse de la structure de covariance. Dans ce travail, après avoir décrit les deux principales méthodes d'estimation que sont PLS et LISREL, nous proposons une approche d'estimation fondée sur la maximisation par algorithme EM de la vraisemblance globale d'un modèle à facteurs latents et à une équation structurelle. Nous en étudions les performances sur des données simulées et nous montrons, via une application sur des données réelles environnementales, comment construire pratiquement un modèle et en évaluer la qualité. Enfin, nous appliquons l'approche développée dans le contexte d'un essai clinique en cancérologie pour l'étude de données longitudinales de qualité de vie. Nous montrons que par la réduction efficace de la dimension des données, l'approche EM simplifie l'analyse longitudinale de la qualité de vie en évitant les tests multiples. Ainsi, elle contribue à faciliter l'évaluation du bénéfice clinique d'un traitement. / Structural equation models enable the modeling of interactions between observed variables and latent ones. The two leading estimation methods are partial least squares on components and covariance-structure analysis. In this work, we first describe the PLS and LISREL methods and, then, we propose an estimation method using the EM algorithm in order to maximize the likelihood of a structural equation model with latent factors. Through a simulation study, we investigate how fast and accurate the method is, and thanks to an application to real environmental data, we show how one can handly construct a model or evaluate its quality. Finally, in the context of oncology, we apply the EM approach on health-related quality-of-life data. We show that it simplifies the longitudinal analysis of quality-of-life and helps evaluating the clinical benefit of a treatment.

Submitted by Daniel Bittar (danielbittar82@yahoo.com.br) on 2010-08-04T14:14:00Z No. of bitstreams: 1 Dissertação Daniel Bittar Definitiva.pdf: 202059 bytes, checksum: 48a1d002cecf496323f3b8b3e1eb3e31 (MD5) / Approved for entry into archive by Vitor Souza(vitor.souza@fgv.br) on 2010-08-04T14:28:33Z (GMT) No. of bitstreams: 1 Dissertação Daniel Bittar Definitiva.pdf: 202059 bytes, checksum: 48a1d002cecf496323f3b8b3e1eb3e31 (MD5) / Made available in DSpace on 2010-08-04T18:03:43Z (GMT). No. of bitstreams: 1 Dissertação Daniel Bittar Definitiva.pdf: 202059 bytes, checksum: 48a1d002cecf496323f3b8b3e1eb3e31 (MD5) Previous issue date: 2010-05-31 / This paper proposes through the principles of Corporate Finance and Asset Pricing measure the impact of the level of liquidity of the companies on the expected return of brazilian equities. The basic assumption of this relationship is that the cash position represents a type of risk not captured by other variables. To estimate this risk will be used a multifactor model. The basic model used is the three-factor model of Fama and French, adapted for the inclusion of the cash variable. From the database, it will attempt to estimate the sensitivity of the expected return of brazilian stocks to cash factor. / Este trabalho busca, através dos princípios de Finanças Corporativas e de Apreçamento de Ativos, mensurar o impacto do nível de liquidez das companhias na expectativa de retorno das ações no mercado acionário brasileiro. O pressuposto básico dessa relação é que a posição de caixa representa um tipo de risco não capturado por outras variáveis. Para mensurar esse risco, será utilizada a modelagem de fatores para apreçamento de ativos. O modelo básico utilizado será o de três fatores de Fama e French, adaptado para a inclusão da variável caixa. A partir da base de dados, se tentará estimar a sensibilidade do retorno esperado das ações brasileiras ao fator caixa.