January 31, 2013

Doctoral Candidate, Department of Statistics and Actuarial Science, University of Waterloo, Ontario Canada

Title: " The time to ruin: analysis of an insurer's solvency risk"

Abstract: Since the introduction of the discounted penalty function by Gerber and Shiu (1998), significant progress has been made on the analysis of various ruin-related quantities in risk theory. Indeed, the discounted penalty function approach not only brings a systematic methodology to jointly analyze the quantities of interest, but also provides the convenience to extract some specific pieces of information from the function. In this talk, we focus on the long-standing finite-time ruin problem. By utilizing the Gerber-Shiu type analysis, we derive explicit expressions for the distribution of the time to ruin in some Sparre Andersen risk models. We propose not only to unify previous methodology through the use of Lagrange's expansion theorem, but also to provide insight into the nature of the series expansions by identifying the probabilistic contribution of each term in the expansion through analysis involving the distribution of the number of claims until ruin.

February 5, 2013

Doctoral Candidate in Actuarial Science, University of Waterloo

Title: " Optimal allocation problems in insurance and finance with dependent risks"

Abstract:  Dependence is an important phenomenon in the field of insurance and finance. Although of great importance, models with general dependence are difficult to handle. In the literature, a lot of special dependence structures have been introduced to model insurance and finance risks, such as independence, comonotonicity, and common shock structures. In this talk, we develop several new dependence structures so as to incorporate most dependence structures that have been proposed. We then study some optimal allocation problems in insurance and finance with the generalized dependence structures. The study gives us insight on the solutions to some classical risk management problems, such as how to design optimal insurance policies under certain constraint, and how to allocate shares among different risky assets.

February 7, 2013

Assistant Professor, Department of Mathematical Sciences, Northern Illinois University

Title: " Seeking new copulas through tail orders"

Abstract: As more data become available, many new meaningful dependence patterns appear. Existing statistical models based on covariance matrices may not well capture those new dependence structures any more. To this end, copula provides a very useful tool. It is often the tail parts that distinguish a copula from a Gaussian dependence structure. Therefore, we propose tail orders of a copula for quantifying the degree of dependence in the tails. After establishing the relation between tail orders of a copula and the mixing random variable used to construct the copula, we can create new desirable copula families based on the knowledge of univariate distributions. By this way, one can gain great convenience in seeking copula families that are not only suitable for capturing new dependence patterns, but also computable and easy to be implemented in various applications.

February 14, 2013

Director, Clinical Trials Statistical and Data Management Center Professor, Department of Biostatistics, University of Iowa

Title: " An Overview of Current Issues in Adaptive Designs"

Abstract: The search by the clinical trial community for improved ways to conduct clinical research has led to considerable interest in adaptive clinical trial designs, which provide the flexibility to adjust trial characteristics on the basis of data reviewed at interim stages.  Statisticians and clinical investigators have proposed or implemented a wide variety of adaptations in clinical trials, but specific approaches have met with differing levels of support.  Currently, the pharmaceutical industry is well ahead of academic trialists with respect to addressing these barriers.  For example, a Drug Information Association (DIA) working group on adaptive designs has engaged regulatory agencies in discussions.  Many researchers working on publicly funded clinical trials, however, are not yet fully engaged in this discussion.  We organized the Scientific Advances in Adaptive Clinical Trial Designs Workshop to begin a conversation about using adaptive designs in publicly funded research.  Held in November of 2009, the 1½-day workshop brought together representatives from the National Institutes of Health (NIH), the Food and Drug Administration (FDA), the European Medicines Agency (EMA), the pharmaceutical industry, non-profit foundations, the patient advocacy community, and academia.  The workshop offered a forum for participants to address issues of adaptive designs that arise at the planning, design, and execution stages of clinical trials, and to hear the perspectives of influential members of the clinical trial community.  The participants also set forth recommendations for guiding action to promote the appropriate use of adaptive designs.  These recommendations have since been presented, discussed, and vetted in a number of venues including the University of Pennsylvania Conference on Statistical Issues in Clinical Trials and the Society for Clinical Trials annual meeting. The purpose of this presentation is to provide a brief overview of adaptive designs, describe the rationale behind conducting the workshop, and summarize the main recommendations that were produced as a result of this workshop.  Specifically, this talk will focus on a number of logistical barriers that will need to be addressed in order to make adaptive clinical trials more practical in the academic setting.

March 14, 2013

Title: " Reimagining the first year of undergraduate statistics: One statistician's experience"

Abstract: The Cornell College Department of Mathematics and Statistics offers an unusual 2-course introduction to the field of statistics.    The first course was updated this year from a more traditional intro course to use randomization based methods to introduce inference.  The second course, rather than being devoted to a specific area of statistics (e.g. regression, or experimental design), is a broad-based introduction to statistical modeling.  I will discuss both of these courses, including what we cover and why, available textbooks, and the software we use for the randomization-based methods.  And I will discuss why I think this 2-course sequence is successful at attracting new students to the field of statistics.

March 28, 2013

Erning Li, Ph.D.

Visiting Lecturer, University of Iowa Department of Statistics and Actuarial Science

Title: "An Alternative REML Estimation of Covariance Matrices in Linear Mixed Models"

Abstract: We propose a data-driven procedure for modeling covariance matrices in linear mixed-effects models with minimal distributional assumption on the random effects. It is based on elimination of the random effects using a transformation of the response variable.  The approach makes it possible for the first time to disentangle the covariance matrices and model them separately.  We rely on the modified Cholesky decomposition to formulate data-based and parsimonious models for the covariance matrices using covariates.  The performance of the proposed method is assessed via simulations and an application.  Properties and extensions of this approach will also be discussed.

April 4, 2013

Associate Professor, University of Iowa College of Engineering

Title: " Sensor system redundancy analysis and sensor fault diagnosis "

Abstract: In this talk, I will present two problems related to fault-tolerance of sensor systems. The first problem is to evaluate the redundancy degree of a sensor system. Computation of the redundancy degree of a sensor system is NP-hard. We propose a method to convert the problem to an equivalent problem of finding sparse solutions of an underdetermined linear system, which has received great attention recently in the area of compressed sensing and statistical variable selection. The second problem is to identify sensor faults in manufacturing processes. We propose a Bayesian variable selection method to solve this problem.

April 11, 2013

Visiting Lecturer, University of Iowa Department of Statistics and Actuarial Science

Title: "A goodness of fit test for the type II pareto distribution"

Abstract: The Type II Pareto Distribution (PD) has been used by many authors for modeling phenomena like: flood levels of rivers, earthquakes, droughts, major insurance claims and financial issues among others. We propose a goodness of fit test for the PD. The test is based on the mean residual life function, which is a linear function of a threshold for the PD. The test statistic is the sample correlation coefficient between the empirical mean residual life function and the sample order statistics, whose distribution has been obtained through simulation. The test does not require the estimation of the parameters of the PD, and the power is high (larger than 0.90) when data comes from a light tailed distribution and reasonable good when the data comes from a heavy tailed distribution. The test can be used as a first step in the analysis of extreme events.