Operational risk losses are heavy tailed and are likely to be asymmetric and extremely dependent among business lines/event types. The analysis of dependence via copula models has been focussed on the bivariate case mainly. In the vast majority of instances symmetric elliptical copulas are employed to model dependence for severities. This thesis proposes a new methodology to assess, in a multivariate way, the asymmetry and extreme dependence between severities, and to calculate the capital for operational risk. This methodology simultaneously uses (i) several parametric distributions and an alternative mixture distribution (the Lognormal for the body of losses and the generalised Pareto Distribution for the tail) using a technique from extreme value theory, (ii) the multivariate skew t-copula applied for the first time across severities and (iii) Bayesian theory. The former to model severities, I test simultaneously several parametric distributions and the mixture distribution for each business line. This procedure enables me to achieve multiple combinations of the severity distribution and to find which fits most closely. The second to effectively model asymmetry and extreme dependence in high dimensions. The third to estimate the copula model, given the high multivariate component (i.e. eight business lines and seven event types) and the incorporation of mixture distributions it is highly difficult to implement maximum likelihood. Therefore, I use a Bayesian inference framework and Markov chain Monte Carlo simulation to evaluate the posterior distribution to estimate and make inferences of the parameters of the skew t-copula model. The research analyses an updated operational loss data set, SAS® Operational Risk Global Data (SAS OpRisk Global Data), to model operational risk at international financial institutions. I then evaluate the impact of this multivariate, asymmetric and extreme dependence on estimating the total regulatory capital, among other established multivariate copulas. My empirical findings are consistent with other studies reporting thin and medium-tailed loss distributions. My approach substantially outperforms symmetric elliptical copulas, demonstrating that modelling dependence via the skew t-copula provides a more efficient allocation of capital charges of up to 56% smaller than that indicated by the standard Basel model.

abstract: In this work, I present a Bayesian inference computational framework for the analysis of widefield microscopy data that addresses three challenges: (1) counting and localizing stationary fluorescent molecules; (2) inferring a spatially-dependent effective fluorescence profile that describes the spatially-varying rate at which fluorescent molecules emit subsequently-detected photons (due to different illumination intensities or different local environments); and (3) inferring the camera gain. My general theoretical framework utilizes the Bayesian nonparametric Gaussian and beta-Bernoulli processes with a Markov chain Monte Carlo sampling scheme, which I further specify and implement for Total Internal Reflection Fluorescence (TIRF) microscopy data, benchmarking the method on synthetic data. These three frameworks are self-contained, and can be used concurrently so that the fluorescence profile and emitter locations are both considered unknown and, under some conditions, learned simultaneously. The framework I present is flexible and may be adapted to accommodate the inference of other parameters, such as emission photophysical kinetics and the trajectories of moving molecules. My TIRF-specific implementation may find use in the study of structures on cell membranes, or in studying local sample properties that affect fluorescent molecule photon emission rates. / Dissertation/Thesis / Masters Thesis Applied Mathematics 2019

Statistical analyses and modeling have contributed greatly to our understanding of the pathogenesis of HIV-1 infection; they also provide guidance for the treatment of AIDS patients and evaluation of antiretroviral (ARV) therapies. Various statistical methods, nonlinear mixed-effects models in particular, have been applied to model the CD4 and viral load trajectories. A common assumption in these methods is all patients come from a homogeneous population following one mean trajectories. This assumption unfortunately obscures important characteristic difference between subgroups of patients whose response to treatment and whose disease trajectories are biologically different. It also may lack the robustness against population heterogeneity resulting misleading or biased inference. Finite mixture models, also known as latent class models, are commonly used to model nonpredetermined heterogeneity in a population; they provide an empirical representation of heterogeneity by grouping the population into a finite number of latent classes and modeling the population through a mixture distribution. For each latent class, a finite mixture model allows individuals in each class to vary around their own mean trajectory, instead of a common one shared by all classes. Furthermore, a mixture model has ability to cluster and estimate class membership probabilities at both population and individual levels. This important feature may help physicians to better understand a particular patient disease progression and refine the therapeutical strategy in advance. In this research, we developed mixture dynamic model and related Bayesian inferences via Markov chain Monte Carlo (MCMC). One real data set from HIV/AIDS clinical management and another from clinical trial were used to illustrate the proposed models and methods. This dissertation explored three topics. First, we modeled the CD4 trajectories using a finite mixture model with four distinct components of which the mean functions are designed based on Michaelis-Menten function. Relevant covariates both baseline and time-varying were considered and model comparison and selection were based on such-criteria as Deviance Information Criteria (DIC). Class membership model was allowed to depend on covariates for prediction. Second, we explored disease status prediction HIV/AIDS using the latent class membership model. Third, we modeled viral load trajectories using a finite mixture model with three components of which the mean functions are designed based on published HIV dynamic systems. Although this research is motivated by HIV/AIDS studies, the basic concepts and methods developed here have much broader applications in management of other chronic diseases; they can also be applied to dynamic systems in other fields. Implementation of our methods using the publicly- vailable WinBUGS package suggest that our approach can be made quite accessible to practicing statisticians and data analysts.

The aim of this thesis is to formulate and implement an algorithm that calculates the probability for two genes being orthologs, given a gene tree and a species tree. To do this, reconciliations between the gene tree and the species trees are used. A birth and death process is used to model the evolution, and used to calculate the orthology probability. The birth and death parameters are approximated with a Markov Chain Monte Carlo (MCMC). A MCMC framework for probability calculations of reconciliations written by Arvestad et al. (2003) is used. Rules for orthologous reconciliations are developed and implemented to calculate the probability for the reconciliations that have two genes as orthologs. The rules where integrated with the Arvestad et al. (2003) framework, and the algorithm was then validated and tested.

This thesis develops models and associated Bayesian inference methods for flexible univariate and multivariate conditional density estimation. The models are flexible in the sense that they can capture widely differing shapes of the data. The estimation methods are specifically designed to achieve flexibility while still avoiding overfitting. The models are flexible both for a given covariate value, but also across covariate space. A key contribution of this thesis is that it provides general approaches of density estimation with highly efficient Markov chain Monte Carlo methods. The methods are illustrated on several challenging non-linear and non-normal datasets. In the first paper, a general model is proposed for flexibly estimating the density of a continuous response variable conditional on a possibly high-dimensional set of covariates. The model is a finite mixture of asymmetric student-t densities with covariate-dependent mixture weights. The four parameters of the components, the mean, degrees of freedom, scale and skewness, are all modeled as functions of the covariates. The second paper explores how well a smooth mixture of symmetric components can capture skewed data. Simulations and applications on real data show that including covariate-dependent skewness in the components can lead to substantially improved performance on skewed data, often using a much smaller number of components. We also introduce smooth mixtures of gamma and log-normal components to model positively-valued response variables. In the third paper we propose a multivariate Gaussian surface regression model that combines both additive splines and interactive splines, and a highly efficient MCMC algorithm that updates all the multi-dimensional knot locations jointly. We use shrinkage priors to avoid overfitting with different estimated shrinkage factors for the additive and surface part of the model, and also different shrinkage parameters for the different response variables. In the last paper we present a general Bayesian approach for directly modeling dependencies between variables as function of explanatory variables in a flexible copula context. In particular, the Joe-Clayton copula is extended to have covariate-dependent tail dependence and correlations. Posterior inference is carried out using a novel and efficient simulation method. The appendix of the thesis documents the computational implementation details. / <p>At the time of the doctoral defense, the following papers were unpublished and had a status as follows: Paper 3: In press. Paper 4: Manuscript.</p>

Longitudinal studies, where data on study subjects are collected over time, is increasingly involving multivariate longitudinal responses. Frequently, the heterogeneity observed in a multivariate longitudinal response can be attributed to underlying unobserved disease states in addition to any between-subject differences. We propose modeling such disease states using a hidden Markov model (HMM) approach and expand upon previous work, which incorporated random effects into HMMs for the analysis of univariate longitudinal data, to the setting of a multivariate longitudinal response. Multivariate longitudinal data are modeled jointly using separate but correlated random effects between longitudinal responses of mixed data types in addition to a shared underlying hidden process. We use a computationally efficient Bayesian approach via Markov chain Monte Carlo (MCMC) to fit such models. We apply this methodology to bivariate longitudinal response data from a smoking cessation clinical trial. Under these models, we examine how to incorporate a treatment effect on the disease states, as well as develop methods to classify observations by disease state and to attempt to understand patient dropout. Simulation studies were performed to evaluate the properties of such models and their applications under a variety of realistic situations.

Markov chain Monte Carlo (MCMC) allows statisticians to sample from a wide variety of multidimensional probability distributions. Unfortunately, MCMC is often difficult to use when components of the target distribution are highly correlated or have disparate variances. This thesis presents three results that attempt to address this problem. First, it demonstrates a means for graphical comparison of MCMC methods, which allows researchers to compare the behavior of a variety of samplers on a variety of distributions. Second, it presents a collection of new slice-sampling MCMC methods. These methods either adapt globally or use the adaptive crumb framework for sampling with multivariate steps. They perform well with minimal tuning on distributions when popular methods do not. Methods in the first group learn an approximation to the covariance of the target distribution and use its eigendecomposition to take non-axis-aligned steps. Methods in the second group use the gradients at rejected proposed moves to approximate the local shape of the target distribution so that subsequent proposals move more efficiently through the state space. Finally, this thesis explores the scaling of slice sampling with multivariate steps with respect to dimension, resulting in a formula for optimally choosing scale tuning parameters. It shows that the scaling of untransformed methods can sometimes be improved by alternating steps from those methods with radial steps based on those of the polar slice sampler.

Markov chain Monte Carlo (MCMC) allows statisticians to sample from a wide variety of multidimensional probability distributions. Unfortunately, MCMC is often difficult to use when components of the target distribution are highly correlated or have disparate variances. This thesis presents three results that attempt to address this problem. First, it demonstrates a means for graphical comparison of MCMC methods, which allows researchers to compare the behavior of a variety of samplers on a variety of distributions. Second, it presents a collection of new slice-sampling MCMC methods. These methods either adapt globally or use the adaptive crumb framework for sampling with multivariate steps. They perform well with minimal tuning on distributions when popular methods do not. Methods in the first group learn an approximation to the covariance of the target distribution and use its eigendecomposition to take non-axis-aligned steps. Methods in the second group use the gradients at rejected proposed moves to approximate the local shape of the target distribution so that subsequent proposals move more efficiently through the state space. Finally, this thesis explores the scaling of slice sampling with multivariate steps with respect to dimension, resulting in a formula for optimally choosing scale tuning parameters. It shows that the scaling of untransformed methods can sometimes be improved by alternating steps from those methods with radial steps based on those of the polar slice sampler.

Model discrimination deals with situations where there are several candidate models available to represent a system. The objective is to find the “best” model among rival models with respect to prediction of system behavior. Empirical and mechanistic models are two important categories of models. Mechanistic models are developed based on physical mechanisms. These types of models can be applied for prediction purposes, but they are also developed to gain improved understanding of the underlying physical mechanism or to estimate physico-chemical parameters of interest. When model discrimination is applied to mechanistic models, the main goal is typically to determine the “correct” underlying physical mechanism. This study focuses on mechanistic models and presents a model discrimination procedure which is applicable to mechanistic models for the purpose of studying the underlying physical mechanism. Obtaining the data needed from the real system is one of the challenges particularly in applications where experiments are expensive or time consuming. Therefore, it is beneficial to get the maximum information possible from the real system using the least possible number of experiments. In this research a new approach to model discrimination is presented that takes advantage of Monte Carlo (MC) methods. It combines a design of experiments (DOE) method with an adaptation of MC model selection methods to obtain a sequential Bayesian Markov Chain Monte Carlo model discrimination framework which is general and usable for a wide range of model discrimination problems. The procedure has been applied to chemical engineering case studies and the promising results have been discussed. Four case studies, order of reaction, rate of FeIII formation, copolymerization, and RAFT polymerization, are presented in this study. The first three benchmark problems allowed us to refine the proposed approach. Moreover, applying the Sequential Bayesian Monte Carlo model discrimination framework in the RAFT problem made a contribution to the polymer community by recommending analysis an approach to selecting the correct mechanism.

This paper presents a statistical framework for estimating output-specific efficiencies for the 2-output case based upon a DEA frontier estimate. The key to the approach is the concept of target output-mix. Being usually unobserved, target output-mixes of firms are modelled as missing data. Using this concept the relevant data generating process can be formulated. The resulting likelihood function is analytically intractable, so a data augmented Bayesian approach is proposed for estimation purposes. This technique is adapted to the present purpose. Some implementation issues are discussed leading to an empirical Bayes setup with data informed priors. A prove of scale invariance is provided. (author's abstract) / Series: Department of Economics Working Paper Series