Longitudinal data analysis is common in biomedical research area. Generalized estimating equations (GEE) approach is widely used for longitudinal marginal models. The GEE method is known to provide consistent regression parameter estimates regardless of the choice of working correlation structure, provided the square root of n consistent nuisance parameters are used. However, it is important to use the appropriate working correlation structure in small samples, since it improves the statistical efficiency of β estimate. Several working correlation selection criteria have been proposed (Rotnitzky and Jewell, 1990; Pan, 2001; Hin and Wang, 2009; Shults et. al, 2009). However, these selection criteria have the same limitation in that they perform poorly when over-parameterized structures are considered as candidates. In this dissertation, new working correlation selection criteria are developed based on generalized eigenvalues. A set of generalized eigenvalues is used to measure the disparity between the bias-corrected sandwich variance estimator under the hypothesized working correlation matrix and the model-based variance estimator under a working independence assumption. A summary measure based on the set of the generalized eigenvalues provides an indication of the disparity between the true correlation structure and the misspecified working correlation structure. Motivated by the test statistics in MANOVA, three working correlation selection criteria are proposed: PT (Pillai's trace type criterion),WR (Wilks' ratio type criterion) and RMR (Roy's Maximum Root type criterion). The relationship between these generalized eigenvalues and the CIC measure is revealed. In addition, this dissertation proposes a method to penalize for the over-parameterized working correlation structures. The over-parameterized structure converges to the true correlation structure, using extra parameters. Thus, the true correlation structure and the over-parameterized structure tend to provide similar variance estimate of the estimated β and similar working correlation selection criterion values. However, the over-parameterized structure is more likely to be chosen as the best working correlation structure by "the smaller the better" rule for criterion values. This is because the over-parameterization leads to the negatively biased sandwich variance estimator, hence smaller selection criterion value. In this dissertation, the over-parameterized structure is penalized through cluster detection and an optimization function. In order to find the group ("cluster") of the working correlation structures that are similar to each other, a cluster detection method is developed, based on spacings of the order statistics of the selection criterion measures. Once a cluster is found, the optimization function considering the trade-off between bias and variability provides the choice of the "best" approximating working correlation structure. The performance of our proposed criterion measures relative to other relevant criteria (QIC, RJ and CIC) is examined in a series of simulation studies.

<p><br></p><p dir="ltr">The complexity of modern financial markets poses substantial challenges for analyzing time-series data, as traditional diffusion models often fail to capture the intricate dynamics of real-world market behavior. This dissertation develops data-driven, non-Markovian methodologies to overcome these limitations, enhancing the analysis of dependencies in financial and insurance data. The research employs advanced stochastic models and machine learning techniques to address critical phenomena in finance and insurance. Notable findings include the following: a Hawkes process framework is introduced to model cascading health transitions, capturing how past events amplify the likelihood of future occurrences (Chapter 3); a discrete-time Hawkes process is used to quantify time-varying lead-lag relationships between intraday and overnight returns, uncovering predictive dynamics in asset price movements (Chapter 4); and attention-based models are applied to high-dimensional, spatiotemporal limit order book (LOB) data, enabling robust analysis and forecasting of its complex structure and behavior (Chapter 5). These findings highlight the limitations of traditional Markovian models, particularly in representing memory-dependent systems, high-frequency data, and multi-state processes. By advancing non-Markovian methods, this dissertation provides practical tools for analyzing momentum effects, cascading health transitions, and intricate market microstructures. These contributions establish a robust analytical foundation for understanding memory-dependent dynamics in finance and insurance, addressing key limitations of Markovian assumptions and opening new avenues for research and application.</p>

Generalized estimating equations (GEE) are popularly utilized for the marginal analysis of longitudinal data. In order to obtain consistent regression parameter estimates, these estimating equations must be unbiased. However, when certain types of time-dependent covariates are presented, these equations can be biased unless an independence working correlation structure is employed. Moreover, in this case regression parameter estimation can be very inefficient because not all valid moment conditions are incorporated within the corresponding estimating equations. Therefore, approaches using the generalized method of moments or quadratic inference functions have been proposed for utilizing all valid moment conditions. However, we have found that such methods will not always provide valid inference and can also be improved upon in terms of finite-sample regression parameter estimation. Therefore, we propose a modified GEE approach and a selection method that will both ensure the validity of inference and improve regression parameter estimation. In addition, these modified approaches assume the data analyst knows the type of time-dependent covariate, although this likely is not the case in practice. Whereas hypothesis testing has been used to determine covariate type, we propose a novel strategy to select a working covariate type in order to avoid potentially high type II error rates with these hypothesis testing procedures. Parameter estimates resulting from our proposed method are consistent and have overall improved mean squared error relative to hypothesis testing approaches. Finally, for some real-world examples the use of mean regression models may be sensitive to skewness and outliers in the data. Therefore, we extend our approaches from their use with marginal quantile regression to modeling the conditional quantiles of the response variable. Existing and proposed methods are compared in simulation studies and application examples.

In an experiment to understand colon carcinogenesis, all animals were exposed to a carcinogen while half the animals were also exposed to radiation. Spatially, we measured the existence of aberrant crypt foci (ACF), namely morphologically changed colonic crypts that are known to be precursors of colon cancer development. The biological question of interest is whether the locations of these ACFs are spatially correlated: if so, this indicates that damage to the colon due to carcinogens and radiation is localized. Statistically, the data take the form of binary outcomes (corresponding to the existence of an ACF) on a regular grid. We develop score??type methods based upon the Matern and conditionally autoregression (CAR) correlation models to test for the spatial correlation in such data, while allowing for nonstationarity. Because of a technical peculiarity of the score??type test, we also develop robust versions of the method. The methods are compared to a generalization of Moran??s test for continuous outcomes, and are shown via simulation to have the potential for increased power. When applied to our data, the methods indicate the existence of spatial correlation, and hence indicate localization of damage. Assuming that there are correlations in the locations of the ACF, the questions are how great are these correlations, and whether the correlation structures di?er when an animal is exposed to radiation. To understand the extent of the correlation, we cast the problem as a spatial binary regression, where binary responses arise from an underlying Gaussian latent process. We model these marginal probabilities of ACF semiparametrically, using ?xed-knot penalized regression splines and single-index models. We ?t the models using pairwise pseudolikelihood methods. Assuming that the underlying latent process is strongly mixing, known to be the case for many Gaussian processes, we prove asymptotic normality of the methods. The penalized regression splines have penalty parameters that must converge to zero asymptotically: we derive rates for these parameters that do and do not lead to an asymptotic bias, and we derive the optimal rate of convergence for them. Finally, we apply the methods to the data from our experiment.

In an experiment to understand colon carcinogenesis, all animals were exposed to a carcinogen while half the animals were also exposed to radiation. Spatially, we measured the existence of aberrant crypt foci (ACF), namely morphologically changed colonic crypts that are known to be precursors of colon cancer development. The biological question of interest is whether the locations of these ACFs are spatially correlated: if so, this indicates that damage to the colon due to carcinogens and radiation is localized. Statistically, the data take the form of binary outcomes (corresponding to the existence of an ACF) on a regular grid. We develop score??type methods based upon the Matern and conditionally autoregression (CAR) correlation models to test for the spatial correlation in such data, while allowing for nonstationarity. Because of a technical peculiarity of the score??type test, we also develop robust versions of the method. The methods are compared to a generalization of Moran??s test for continuous outcomes, and are shown via simulation to have the potential for increased power. When applied to our data, the methods indicate the existence of spatial correlation, and hence indicate localization of damage. Assuming that there are correlations in the locations of the ACF, the questions are how great are these correlations, and whether the correlation structures di?er when an animal is exposed to radiation. To understand the extent of the correlation, we cast the problem as a spatial binary regression, where binary responses arise from an underlying Gaussian latent process. We model these marginal probabilities of ACF semiparametrically, using ?xed-knot penalized regression splines and single-index models. We ?t the models using pairwise pseudolikelihood methods. Assuming that the underlying latent process is strongly mixing, known to be the case for many Gaussian processes, we prove asymptotic normality of the methods. The penalized regression splines have penalty parameters that must converge to zero asymptotically: we derive rates for these parameters that do and do not lead to an asymptotic bias, and we derive the optimal rate of convergence for them. Finally, we apply the methods to the data from our experiment.

The class of models which is studied in this paper, multi-unit longitudinal models, combines both the cross-sectional and the longitudinal aspects of observations. Many empirical investigations involve the analysis of data structures that are both cross-sectional (observations are taken on several units at a specific time period or at a specific location) and longitudinal (observations on the same unit are taken over time or space). Multi-unit longitudinal data structures arise in economics and business where panels of subjects are studied over time, biostatistics where groups of patients on different treatments are observed over time, and in situations where data are taken over time and space. Modelling issues in multi-unit longitudinal models with random coefficients and patterned correlation structure are illustrated in the context of two data sets. The first data set deals with short time series data on annual death rates and alcohol consumption for twenty-five European countries. The second data set deals with glaceologic time series data on snow temperature at 14 different locations within a small glacier in the Austrian Alps. A practical model building approach, consisting of model specification, estimation, and diagnostic checking, is outlined. (author's abstract) / Series: Forschungsberichte / Institut für Statistik

Made available in DSpace on 2016-06-02T20:04:52Z (GMT). No. of bitstreams: 1 4552.pdf: 1449121 bytes, checksum: 5d02e70bde72ea9ef3c257c80ceed1dc (MD5) Previous issue date: 2012-05-03 / Financiadora de Estudos e Projetos / In this dissertation we propose new models for survival with cure fraction to describe the biological mechanism of the event of interest (cancer) in studies of carcinogenesis in the presence of competing causes latent independent or correlated. The formulation of new models is based on stochastic modeling of the occurrence of tumors through three stages: initiation of a tumor not detectable, promotion and progression of the tumor to a detectable cancer. These models allow a simple pattern of the dynamics of tumor growth, and incorporate into the analysis features of the stage of tumor progression that is not possible in most survival models with cure fraction commonly used. For the proposed models, the inferential process was discussed in terms of classical and Bayesian point of view. Simulations studies were conducted in order to analyze the asymptotical properties of the classical estimation procedure. Real data applications demonstrate of use of the models. / Neste trabalho propomos modelos de sobrevivência com fração de cura para descrever o mecanismo biológico da ocorrência do evento de interesse (câncer) em estudos da carcinogênese na presença de causas competitivas latentes independentes ou correlacionadas. A formulação dos novos modelos é baseada na modelagem estocástica da ocorrência dos tumores através de três estágios: iniciação de um tumor não detectável, promoção e a progressão do tumor até um câncer detectável. Estes modelos permitem um padrão simples da dinâmica de crescimento do tumor, além de incorporarem características do estágio de progressão do tumor, que não é possível na maioria dos modelos de sobrevivência com fração de cura comumente utilizados. Para os modelos propostos, discutimos o processo inferencial do ponto de vista clássico e bayesiano. Estudos de simulações foram feitos com o objetivo de analisar as propriedades assintóticas do processo de estimação clássico. Aplicações a conjuntos de dados reais mostraram a aplicabilidade dos modelos.

A literatura dispõe de métodos de diagnóstico para avaliar o ajuste de modelos lineares generalizados (MLGs) para medidas repetidas baseado em equações de estimação generalizada (EEG). No entanto, tais métodos não contemplam a distribuição binomial nem bancos de dados com observações faltantes. O presente trabalho generalizou os métodos já desenvolvidos para essas duas situações. Na construção de gráficos de probabilidade meio-normal com envelope simulado para a distribuição binomial, foi proposto um método para geração de variáveis aleatórias com distribuição marginal binomial correlacionadas, baseado na convolução de variáveis com distribuição de Poisson independentes. Os métodos de diagnóstico desenvolvidos foram aplicados em dados reais e simulados. / Literature provides diagnostic methods to assess the fit of generalized linear models (GLM) for repeated measures based on generalized estimating equations (GEE). Still, such methods do not include the binomial distribution or databases with missing observations. This work generalizes the methods already developed for these two situations. A method for generating random variables with correlated marginal binomial distributions based on convolution of independent Poisson random variables has been proposed for the construction of half-normal probability plots. The diagnostic methods developed were applied to real and simulated data.

Este trabalho tem como objetivo principal apresentar uma abordagem clássica e Bayesiana dos modelos lineares com observações censuradas, que é uma nova área de pesquisa com grandes possibilidades de aplicações. Aqui, substituimos o uso convencional da distribuição normal para os erros por uma família de distribuições mais flexíveis, o que nos permite lidar de forma mais adequada com observações censuradas na presença de outliers. Esta família é obtida através de um mecanismo de fácil construção e possui como casos especiais as distribuições t de Student, Pearson tipo VII, slash, normal contaminada e, obviamente, a normal. Para o caso de respostas correlacionadas e censuradas propomos um modelo de regressão linear robusto baseado na distribuição t de Student, desenvolvendo um algoritmo tipo EM que depende dos dois primeiros momentos da distribuição t de Student truncada. / This work aims to present a classical and Bayesian approach to linear models with censored observations, which is a new area of research with great potential for applications. Here, we replace the conventional use of the normal distribution for the errors of a more flexible family of distributions, which deal in more appropriately with censored observations in the presence of outliers. This family is obtained through a mechanism easy to construct and has as special cases the distributions Student t, Pearson type VII, slash, contaminated normal, and obviously normal. For the case of correlated and censored responses we propose a model of robust linear regression based on Student\'s t distribution and we developed an EM type algorithm based on the first two moments of the truncated Student\'s t distribution.

Este trabalho tem como objetivo principal apresentar uma abordagem clássica e Bayesiana dos modelos lineares com observações censuradas, que é uma nova área de pesquisa com grandes possibilidades de aplicações. Aqui, substituimos o uso convencional da distribuição normal para os erros por uma família de distribuições mais flexíveis, o que nos permite lidar de forma mais adequada com observações censuradas na presença de outliers. Esta família é obtida através de um mecanismo de fácil construção e possui como casos especiais as distribuições t de Student, Pearson tipo VII, slash, normal contaminada e, obviamente, a normal. Para o caso de respostas correlacionadas e censuradas propomos um modelo de regressão linear robusto baseado na distribuição t de Student, desenvolvendo um algoritmo tipo EM que depende dos dois primeiros momentos da distribuição t de Student truncada. / This work aims to present a classical and Bayesian approach to linear models with censored observations, which is a new area of research with great potential for applications. Here, we replace the conventional use of the normal distribution for the errors of a more flexible family of distributions, which deal in more appropriately with censored observations in the presence of outliers. This family is obtained through a mechanism easy to construct and has as special cases the distributions Student t, Pearson type VII, slash, contaminated normal, and obviously normal. For the case of correlated and censored responses we propose a model of robust linear regression based on Student\'s t distribution and we developed an EM type algorithm based on the first two moments of the truncated Student\'s t distribution.