Consideramos expressões analíticas para as funções de verossimilhanças no modelo linear com erros nas variáveis e no modelo com efeitos aleatórios. Dada a complexidade da função de verossimilhança consideramos um algorítimo do tipo EM para a estimação dos parâmetros do modelo.

We propose a novel estimator for the stochastic discount factor (SDF) in a panel-data context. Under general conditions it depends exclusively on appropriate averages of asset returns, and its computation is a direct exercise, as long as one has enough observations to fit our asymptotic results. We identify the SDF using the fact that it is the “common feature” in every asset return of the economy. Moreover, it does not depend on any assumptions about preferences, or on consumption data, which allows testing directly different preference specifications, as well as the existence of the equity-premium puzzle. Preliminary results are encouraging.

Nas últimas décadas a tomografia computadorizada tem permitido avanços surpreendentes em diversas áreas de medicina. Do ponto de vista de desenvolvimento científico isso se deve a um tremendo esforço de integração interdisciplinar. Nesta palestra pretendo inicialmente fazer um breve apanhado das ferramentas matemáticas que permitiram a tomografia se tornar uma realidade, bem como alguns dos problemas matemáticos suscitados pela mesma. A seguir discutirei alguns dos desafios matemáticos encontrados pelo imageamento de tecidos com infra-vermelho e pela microscopia eletrônica tri-dimensional de macro-moléculas.

Planos de Contribuição Definida geram benefícios de aposentadoria, em regime individual, que dependem do montante obtido na aplicação das contribuições vertidas por seus participantes. A técnica usualmente adotada para a projeção das rendas de aposentadoria consiste na aplicação de modelos determinísticos. Após a concessão do benefício, a responsabilidade das entidades é dimensionada exclusivamente em função do valor esperado do fluxo de pagamentos. Como alternativa propõe-se a aplicação de modelos estocásticos no contexto de planos de contribuição definida em duas etapas: a) acumulação de recursos; e b) pagamento de benefícios. Na primeira etapa é apresentado um modelo de acumulação estocástico, orientado por uma política de macro alocação atrelada à meta sob o princípio de otimalidade de Bellman. Na segunda, a distribuição dos compromissos da entidade é simulada a partir da distribuição dos compromissos individuais de cada um de seus participantes que estejam em gozo de benefício de renda de aposentadoria.

In this talk we will present an empirical comparison of filter and wrapper procedures for supervised classification. The filter methods considered are: Relief, Las Vegas Filter (LVF), and FINCO, a new method introduced by us. Among the wrappers considered are: The sequential forward selection (SFS) and the sequential floating forward selection (SFFS). The classifiers used by the wrappers are: The linear discriminant analysis (LDA), The k-nearest neighbors (KNN) classifier and a decision tree classifier based on recursive partition (Rpart). Both type of feature selection procedures are compared on the percentage of selected features as well as on their effect on the misclassification rate of the classifiers. Finally, we compare our results with feature selection using Random Forest. The experiments are carried out in twelve datasets coming from the Machine Learning Database Repository available at the University of California, Irvine.

“Uncertainty”, like its complementary cousin, “information”, is a much used but not very well defined concept despite its intrinsic role in statistics. (Indeed, that latter is often described as the “science of uncertainty”.) In this talk, I will explore some of the meanings provided in a paper, written with Constance van Eeden, that are ascribed to that term and readily discover that seemingly natural questions can have answers that are either elusive or counter-intuitive. For example, surprisingly (in answer to one of those questions), the level of uncertainty (according to one defintion) can actually increase rather than decrease as the amount of information increases. For other definitions we have not been able to give general answers to that question. I will also address the issue of combining information to reduce uncertainty. Specifically, I will survey some recent work I have done with Constance van Eeden and others on the use of the weighted likelihood in conjunction with samples from populations different from, but similar to that under study. That resemblence can lead to very effective trade-offs of bias for precision when it derives from structural relations among the various population parameters, for example, when the difference in the population means may be bounded by a fixed constant.

A Morte Súbita dos Citrus (MSC) é uma doença que atinge certas combinações de enxerto/porta-enxerto em árvores de citrus já afetando parte da principal região citrícola brasileira e com potencial de espalhamento nos próximos anos. A doença foi detectada pela primeira vez em 1999 e descrita em 2001. O agente causador ainda não está determinado embora evidências apontem para uma variação do vírus da Tristeza dos Citrus.
Dada a situação atual de alto risco e potencial prejuízo a doença tem sido foco de atenção por porte dos órgãos ligados a produção de citrus. Diversos levantamentos de dados e estudos tem sido realizados a fim de compreender melhor os padrões de dispersão espacial da doença tanto a pequena escala em talhões, quanto na larga escala, na região produtora.
Nesta apresentação descreve-se o problema e as questões de relevância científica e estatística associadas. Resultados obtidos até o momento serão apresentados e serão discutidas estratégias para futuras análises.

In this talk I shall describe current joint research with Martin Ridout and Diana Price (Statistics, Kent), Mick Tuite and Lee Byrne (Biosciences Kent) and Lloyd Ruddock (Biosciences, Finland). We are motivated by experiments at Kent, designed to estimate the mean number, m, of prions in yeast cells. The work has important consequences for understanding both prion behaviour and the basic mechanisms taking place within living cells. The seminar assumes no previous knowledge of the biological setting, and the essence of the seminar involves elementary probability theory and model-fitting. The yeast protein Sup35p exists in two stable forms, giving rise to phenotypes [PSI+] and [psi-], distinguished by a colour assay. Because it replicates in a manner similar to the mammalian prion PrP, Sup35p is called a yeast prion. When the chemical GuHCl is added to a culture medium growing [PSI+] cells, the proportion of [PSI+] cells gradually decreases over time, because of inhibition of prion replication: the number of prions present initially is gradually diluted through an expanding population of yeast cells. The experimental data provide estimates of the proportion of [PSI+] cells at hourly intervals. We develop stochastic models for this process, and obtain an interesting analytical solution when cell doubling time is assumed to be constant, but not synchronised between cells. We then use a more realistic model that allows for variation in doubling time. Our estimate of m ~ 100 is substantially larger than a previously published estimate. Other parameters in the model include the mean, n, of cell doubling time and a time lag parameter that allows for disruption to cell doubling at the start of the experiment. Our estimates of n agree closely with independent estimates from optical density (OD) measurements. OD data are also used to distinguish between alternative estimates of the time lag that give almost equivalent fits to the data. Model extensions which incorporate greater biological realism, will also be described.

No presente trabalho discutimos o problema da estimação via máxima verossimilhança para o parâmetro de assimetria da distribuição normal assimétrica univariada (Azzalini, 1985) e apresentamos alternativas Bayesianas bastante eficientes para estimação desse parâmetro. Uma aplicação em modelos de regressão linear com erros nas variáveis será discutida, enfatizando a estrutura hierárquica do modelo.

In this seminar we present a Bayesian analysis for estimating the disease prevalence, the sensitivity and specificity of several independent diagnostic tests in the presence of a covariate and latent gold standard. We use Metropolis-Hastings algorithm to obtain the posterior summaries of interest. The methodology can be easily extended to the case with two or more covariates or more than two tests. With only two diagnostic tests we can not perform a maximum likelihood procedure since the number of unknown parameters is bigger than the number of observations. Under the Bayesian approach we overcome this problem by introducing latent variables. An application to a real medical data illustrates the methodology.

Dependencies in DNA sequences are frequently modeled using Markov models. However, Markov chains cannot account for heterogeneity that may be present in different regions of the same DNA sequence. Hidden Markov models are more realistic than Markov models since they allow for the identification of heterogeneous regions of a DNA sequence. In this work we present an application of hidden Markov models to a subsequence of the Xylella fastidiosa DNA data. We found that a three-state model provides a good description for the data considered.