[2005] [2006] [2007] [2008] Seminar Homepage
In order to access Webinars please select the Webinar Title.
Speaker: Rob Brown, University of Waterloo
Stat & Act. Sci. Seminar Title: A Fine Balance - Safe Pensions, Affordable Plans, Fair Rules: The Report of the Ontario Expert Commission on Pensions
Abstract: In November of 2006, the then Ontario Minister of Finance, Greg
Sorbara, announced the formation of the Ontario Expert Commission on Pensions.
Its mandate was to look into the health of Employer-Sponsored Defined Benefit
(DB) pension plans and the environment in which they exist. The Commission
was to report back to the Minister by the summer of 2008.
In fact, the Commission
reported back by August 31, 2008. On November 20, the 222-page report was made
public. The new Minister of Finance, Dwight Duncan, has defined a comment period
to end February 27, 2009 for stakeholders to respond to the Commissions 142
Recommendations.
Rob Brown, of our Department, was the Director of Research
for the Commission. In this Seminar, Rob will report on the workings of the
Commission and highlight some of its major recommendations.
Time: 4:00 pm.
Place: DC 1304 (top)
Speaker: An Chen, University of Bonn
Stat & Act. Sci. Seminar Title: On the Regulation Under Solvency II -- A Structural Approach
Abstract: In this paper, we provide a new insight to the previous work of Briys and de Varenne [1994], Grosen and Jorgensen [2002] and Chen and Suchanecki [2007]. We first study how the regulator establishes regulation intervention levels in order to control for instance the default probability of the insurance company (under the real world probability measure). This part of the analysis is based on a constant volatility and there exists a one-to-one relation between the optimal regulation level and the volatility. Given that the insurance company is informed of the regulatory rules, we study how results can be significantly different when the insurance company follows a risk management strategy with non-constant volatilities. We show that insurers under the regulation of Solvency II can meet solvency requirements without a great sacrifice to the expected return either of themselves or their policyholders and thus that the regulation through Solvency II is not very "costly".
Time: 3:00 pm.
Place: MC 5136 (top)
Speaker: Dr. Thomas Post,
Humboldt-UniversitŠt zu Berlin, School of Business and Economics, Chair for Insurance and Risk Management
Stat & Act. Sci. Seminar Title: The Impact of Individual Investment Behavior for Retirement Welfare:
Evidence from the United States and Germany
Abstract: Much of the industrialized world is undergoing a significant
demographic shift, placing strain on public pension systems. Policymakers are
responding with pension system reforms that put more weight on privately managed retirement funds. One concern with these changes is the effect on individual welfare
if households invest suboptimally. Using micro-level data from the United States and
Germany, we compare the optimal expected lifetime utility computed using a
realistically calibrated model with the actual utility as reflected in empirical asset
allocation choices. Through this analysis, we are able to identify the population
subgroups with relatively large potential for welfare gains. Our results should be
helpful to public policymakers in designing programs to improve the performance of
privately organized retirement systems.
Keywords Asset Allocation, Retirement Welfare, Pension Reform
Time: 4:00 pm.
Place: DC 1304 (top)
Speaker: Eric Cheung, Department of Statistics and Actuarial Science, University of Waterloo
Stat & Act. Sci. Seminar Title: Generalizations of Gerber-Shiu functions in risk models involving dependence
Abstract: The seminal paper by Gerber and Shiu (1998) gave a huge boost to risk theory by incorporating the time of ruin, the surplus prior to ruin and the deficit at ruin into a single quantity currently known as the Gerber-Shiu function. Motivated by the fact that the surplus prior to ruin and the deficit at ruin are both quantities defined at the time of ruin, we propose to generalize the Gerber-Shiu function by further incorporating other random variables defined before the time of ruin. These include the surplus level after the second last claim before ruin, the maximum and the minimum surplus levels before ruin. Such generalizations of the Gerber-Shiu function allow us to study some new quantities such as the last interclaim time before ruin and the ladder height causing ruin. Applications of the newly proposed Gerber-Shiu functions in obtaining stochastic orderings and joint densities of ruin-related quantities are demonstrated in various risk models involving dependence.
Time: 4:00 pm.
Place: DC 1304 (top)
Speaker: Chengguo Weng, Department of Statistics and Actuarial Science, University of Waterloo
Stat & Act. Sci. Seminar Title: Optimal Reinsurance Designs: from an Insurer’s Perspective
Abstract: It is well-known that reinsurance can be an effective risk management technique for an insurer. An appropriate use of the reinsurance reduces the adverse risk exposure of an insurer and improves the overall viability of the underlying business. The use of reinsurance, on the other hand, incurs additional cost to the insurer in the form of reinsurance premium. This implies that an insurer is faced with a tradeoff between risk spreading and risk retaining. The thrust of my research is to propose some reinsurance designs that are optimal from the perspective of an insurer.
In this presentation I will summarize a series of results that I have obtained throughout my Ph.D. studies. In particular I will describe an intricate role between optimal reinsurance and some commonly used risk measures, including Value-at-Risk (VaR) and Tail Value-at-Risk (TVaR). I will also describe an empirical/nonparametric approach to optimal reinsurance designs. This new method has the advantage that it does not need to make any explicit assumptions on the distribution of the underling risk. More importantly, it provides a simple and yet practical way of obtaining the solution to a wide range of optimal reinsurance models.
Time: 11:00 am.
Place: MC 5158 (top)
Speaker: Matthew Till, Department of Statistics and Actuarial Science, University of Waterloo
Stat & Act. Sci. Seminar Title: A Stochastic Approach to Regime-Switching Model Validation
Abstract: Regime-switching models have become a popular tool in econometric time series modelling since Hamilton (1998). However, residual analysis for goodness-of-fit inference has not often been used, as the underlying state process presents difficulty in residual definition, and generally defined residuals are not normally distributed, even for models that have normal innovations within each regime. I present a filtering process that extracts residuals that are independent and N(0,1) distributed for a Markov regime-switching models with Gaussian innovations, and demonstrate the method using real world data.
In this presentation I will summarize a series of results that I have obtained throughout my Ph.D. studies. In particular I will describe an intricate role between optimal reinsurance and some commonly used risk measures, including Value-at-Risk (VaR) and Tail Value-at-Risk (TVaR). I will also describe an empirical/nonparametric approach to optimal reinsurance designs. This new method has the advantage that it does not need to make any explicit assumptions on the distribution of the underling risk. More importantly, it provides a simple and yet practical way of obtaining the solution to a wide range of optimal reinsurance models.
Time: 3:00 pm
Place: MC 5158 (top)
Speaker: Joe DiCesare,
Associate Director, Financial Engineering
The Bank of Nova Scotia – Scotia Capital Markets
Stat & Act. Sci. Seminar Title: Importance Sampling for Jump Diffusions with Applications to Finance
Abstract: Simulation of jump diffusions is rarely an exact exercise. In many financial applications one is often faced with the problem of calculating some expected value which is dependent on the path of a diffusion process. For this purpose the Radon-Nikodym derivative pertaining to probability measures induced by diffusion processes can be used to perform importance sampling. Unfortunately, it cannot be precisely evaluated using standard simulation methods. In this talk an unbiased estimator of the Radon-Nikodym derivative is proposed. This estimator is used to construct an importance sampling algorithm for diffusions. It is shown that a special case of this algorithm yields an exact rejection based simulation method for a large class of diffusions. The proposed methodology is applied to the pricing of barrier and lookback options under a Constant Elasticity of Variance (CEV) jump diffusion model.
Time: 4:00 pm.
Place: DC 1304 (top)
Speaker: Michelle Zhou, University of Waterloo
Stat & Act. Sci. Seminar Title: Information ratio test for model misspecification
Abstract: In a quasi-likelihood inference, correct specification of the first two moments leads to the information unbiasedness. This property essentially means that the Fisher information (the model-based covariance estimator) is equivalent to the Godambe information (the sandwich covariance estimator). In this talk, through the investigation of robustness property of the Godambe information, we will present an information ratio test for model misspecification of covariance structure under the correctly specified mean structure in the context of regression models, including linear, generalized linear and generalized estimating equation (GEE) models. The proposed test statistic is constructed on the validation of the information unbiasedness. We will discuss large sample properties. Moreover, through simulation studies we show that the proposed test statistics appear more powerful than the classical information matrix test proposed by White (1982).
Time: 4:00
Place: MC 5158 (top)
Speaker: Milorad Kovacevic , Statistics Canada (Sponsored by MITACS/NICDS Project Statistical Methods for Complex Survey Data)
Stat & Act. Sci. Seminar Title: Some Current Research Topics on Survey Methodology
Abstract: The presentation will give a general overview of the current research in the wide area of survey methodology and its importance for improving methods needed for assessments of the country's population, economy, and society, as well as for studies on human and social dynamics. More details will be given about the research on multi-level modelling of survey data, as well as about modelling of "synthetic" data obtained by record linkage.
Time: 4:00 pm.
Place: DC 1304 (top)
Speaker: Steve Fan, University of Toronto
Stat & Act. Sci. Seminar Title: Local EM and EMS Algorithms
Abstract: The use of local likelihood methods (Tibshirani and Hastie, 1987; Loader, 1996) in the presence of area-censored or spatially aggregated data leads naturally to the consideration of EM-like strategies, or rather local EM algorithms. We begin by exploring a class of local EM algorithms for density estimation (Braun et al., 2005) and intensity estimation (Tolusso and Cook, 2007). We demonstrate that using a particular conditional distribution at the E-step results in the algorithm collapsing explicitly into an EMS algorithm of the type considered by Silverman et al. (1990). The advantages of identifying such a relationship between local likelihood and the EMS algorithm is that the former provides a natural context for the latter, while the latter provides a set of tools to inform the interpretation and implementation of local EM algorithms. For example, we establish a previously unknown connection between local EM algorithms and the penalized likelihood, a connection that is analogous to the more familiar pairing of EM and likelihood.
We then apply the proposed method to study the spatial risk that attributes to newly diagnosed lupus cases in Greater Toronto Area while using a generalized linear model to accommodate potential age-gender group and time effects. This analysis allows us to combine multiple disease maps with time-varying boundaries and identify the region with the highest risk while adjusting regional population compositions and time trend. (Joint work with Patrick Brown and Jamie Stafford)
Time: 2:00 pm
Place: MC 5158 (top)
Speaker: Xi Luo, Yale University
Stat & Act. Sci. Seminar Title: L1 Penalized Likelihood: Fast Algorithms and Risk Bounds
Abstract: Maximum likelihood estimation has been a major component in statistical modeling, and currently the variant with L1 penalty is a popular approach. In particular, L1 penalized regression (aka LASSO) has received considerable attention, and results range from computation to theoretical properties. To complement the existing research efforts, we present a general iterative algorithm to optimize the L1 penalized loglikelihood criterion. Attention is given to the class of optimizers using linear combinations of terms selected from a possibly very large library of candidate variables or functions. We prove that computation accuracy is bounded by a quantity over the number of iterations taken, and similar analysis could carry over to provide statistical risk bounds by showing variable-complexity covering properties. The L1 penalized regression problem is revisited as an example application, where we show adaptive risk bounds with proper choices of penalizing parameters as well as a modified LASSO criterion to adapt to unknown error variance. The computation and estimation performance are illustrated using numerical examples.
Time: 2:00 pm
Place: MC 5158 (top)
Speaker: Feng-Chang Lin, University of Wisconsin-Madison
Stat & Act. Sci. Seminar Title: Regression Modeling of Modulated Renewal Processes
Abstract: single realization of a modulated renewal process. The consistency and asymptotic normality of the estimators is proved under ergodicity conditions. Previous work has considered either parametric likelihood analysis or semiparametric multiplicative models using partial likelihood. The framework is generally applicable to semiparametric and parametric models. It facilitates a semiparametric extension of a popular parametric earthquake model. Simulations and empirical analyses of Taiwan earthquake sequences illustrate the methodology’s practical utility.
Further, a rate model is proposed for a modulated renewal process, where the covariate process may not capture the dependencies in the sequence. We consider partial likelihood based inferences under a semiparametric multiplicative rate model. Under an intensity model, gap times in a long single sequence may naively be used in the partial likelihood with variance estimation employing the observed information matrix. Under a rate model, the gap times cannot be treated as independent and studying the partial likelihood is much more challenging. Simulation studies demonstrate the improved performance of our inferences relative to intensity based analyses. An application to the spread of smallpox in 1967 in the town of Abakaliki, Nigeria, illustrates our approach.
Time: 4:00
Place: MC 5158 (top)
Speaker: Zhezhen Jin, Columbia University
Stat & Act. Sci. Seminar Title: Regression analysis of censored data
Abstract: For the analysis of right censored data, there has been extensive research on the estimation and inference procedures for regression models alternative to the Cox proportional hazards model. The common challenging feature in the use of the alternative models is that their corresponding estimating functions for regression parameters are nonregular,i.e., non-smooth and non-monotone. As a result, it is challenging to obtain the point estimation and its variance estimation. In this talk, I will review recently developed estimation methods, in particular, for the accelerated failure time (AFT) model, linear transformation models and general semiparametric regression models. Unsolved issues and problems will also be discussed.
Time: 1:30 pm.
Place: MC 5158 (top)
Speaker: Xi Luo, Department of Statistics, Yale University
Stat & Act. Sci. Seminar Title: L1 Penalized Likelihood: Fast Algorithms and Risk Bounds
Abstract: Maximum likelihood estimation has been a major component in statistical modeling, and currently the variant with L1 penalty is a popular approach. In particular, L1 penalized regression (aka LASSO) has received considerable attention, and results range from computation to theoretical properties. To complement the existing research efforts, we present a general iterative algorithm to optimize the L1 penalized loglikelihood criterion. Attention is given to the class of optimizers using linear combinations of terms selected from a possibly very large library of candidate variables or functions. We prove that computation accuracy is bounded by a quantity over the number of iterations taken, and similar analysis could carry over to provide statistical risk bounds by showing variable-complexity covering properties. The L1 penalized regression problem is revisited as an example application, where we show adaptive risk bounds with proper choices of penalizing parameters as well as a modified LASSO criterion to adapt to unknown error variance. The computation and estimation performance are illustrated using numerical examples.
Time: 2:00 pm.
Place: MC 5136 (top)
Speaker: Grace Chiu, University of Waterloo
Stat & Act. Sci. Seminar Title: Statistical Inference for Food Webs, Part I: Bayesian Melding
Abstract: Quantifying various aspects of dependence within a food web or ecological network has largely been descriptive or deterministic. Existing schemes that attempt to assess reliability of such quantification are often non-statistical in nature. This talk focuses on understanding ecological networks in the context of mass balance (as opposed to identifying trophic compartments). Our statistical inferential approach employs Bayesian melding, and embodies the traditional deterministic views on network relations; yet, it is developed for the estimation of physical quantities and assessment of the estimation, both from a statistical perspective. The practical advantage of incorporating natural measurement variability into deterministic beliefs about the relationships among measurements is a more honest representation of the true state of nature, and the ability to formally assess balance before data are passed on to later stages of an ecological network analysis (ENA). We also find that general Bayesian inference for ENA can yield new ecological insight that may not be available through standard classical inference.
Time: 4:00 pm.
Place: MC 6007 (top)
Speaker: Prof. Csaba Szepesvari, University of Alberta, Department of
Computing Science
Stat & Act. Sci. Seminar Title: Manifold-Adaptive Dimension Estimation
Abstract: Clever machine learning techniques attempt to exploit whatever favorable properties a learning task may have. They better do so, since learning under unfavorable conditions is known to be hard in an information theoretic sense. One supposedly favorable situation is when the data lies in a low-dimensional submanifold of a high-dimensional Euclidean space. The last couple of years have seen a flourish of algorithms that either directly or indirectly attempt to exploit this condition. Less is known, however, about the theoretical properties of these algorithms, in particular about their rate of convergence. A closer look at them reveals that many of these approaches require the knowledge of the dimension of the submanifold that the data lies on. Hence, their performance will be limited by the speed at which this dimension can be estimated from data. Therefore, in this talk I will look at the question of how efficiently can we estimate the dimension. In particular, I will propose a particular dimension estimation method, argue about and illustrate its efficiency. I will demonstrate that the rate of convergence of the dimension estimation method proposed is (essentially) independent of D, the dimension of the embedding space, which entails that the convergence rate of many learning algorithms will also scale (essentially) independently of D. This explains why cleverly tuned algorithms might be efficient even when the inputs are very high-dimensional, but the emphasis should be on `cleverly tuned'. At the end of the talk I will list a few interesting related open questions.
This talk is largely based on the ICML-07 paper 'Manifold-Adaptive Dimension Estimation' which is joint work with Amir massoud Farahmand and Jean-Yves Audibert.
Biography: Csaba Szepesvari (University of Alberta, Canada)
1999 PhD from "Jozsef Attila" University, Szeged, Hungary currently an Associate Professor at the Department of Computing Science of the University of Alberta and a principal investigator of the Alberta Ingenuity Center for Machine Learning. Previously held a senior researcher position at the Computer and Automation Research Institute of the Hungarian Academy of Sciences, where he headed the Machine Learning Group. Before that spent 5 years in the software industry. The coauthor of a book on nonlinear approximate adaptive controllers, published over 80 peer reviewed journal and conference papers, serves as the Associate Editor of IEEE Transactions on Adaptive Control and AI Communications, and as a member of the program committee at various machine learning and AI conferences. Areas of expertise include statistical machine learning, Markovian decision processes, reinforcement learning and nonlinear control.
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Rong Zhu, McMaster University
Stat & Act. Sci. Seminar Title: Model clustering and its application
Abstract: The classification of objects into groups where the objects within a group share a set of common traits is important in many areas of applications and particularly in environmental pollution studies. Consider the situation where variables are measured on different occasions for each object, and the objective is to classify these objects into groups according to some common characteristics. We propose a procedure called model clustering which groups objects according to the similarity of their underlying models. This procedure consists of two aspects: model fitting and clustering. The model fitting selects a family of models appropriate for the structure and nature of the available measurements, and then is performed for both individual and pooled data sets. The similarity of models is defined as the equality of their parameters of interest. Here we partition the parameter vector into two sub-vectors corresponding to the interested parameters and ancillary parameters. The clustering will group together objects that have common interested parameters while allowing the ancillary parameters to be object specifics. The p-value associated with the proposed model linking test is used as the similarity measure. Several grouping strategies are proposed like cluster peeling, pairwise combining, as well as a speeding technique called splitting-and-binding. The proposed model clustering is utilized in an environmental application where the interest is to classify Ecoli bacteria according to their responses to antibiotic treatments. The data were collected bi-weekly at several locations within three Canadian watersheds during 2005. Metric closeness in parameter space used by conventional method and likelihood closeness in model space employed by model clustering are discussed in this application.
Time: 3:00 pm.
Place: MC 5136 (top)
Speaker: Wayne Oldford, University of Waterloo
Stat & Act. Sci. Seminar Title: Graph theoretic structure of statistical graphics
Abstract: Every statistical graphic is a construct of display elements arranged in a spatial (and sometimes temporal) layout designed to communicate statistical information to the viewer. Spatial examples include arrangement of cells in scatterplot matrices, ordering of axes in parallel coordinate plots or of simultaneous confidence intervals in multiple comparison plots; temporal examples include any dynamically changing two dimensional image, such as a 2d scatterplot in a 3d rotating scatterplot, or in a projection tour of higher dimensional data. In each case, perception is always affected by layout.
Graph theory is a convenient and useful model for many such arrangement problems -- graph nodes represent display components and edges connections (temporal or spatial) between them. Edges can be directed or undirected, weighted or unweighted. Layout becomes graph traversal.
In this talk, some interesting graphs as well as traversals will be introduced/reviewed and applied to examples of statistical graphics. Pairwise comparisons and exploration of high dimensional space will be given particular attention to show how graphs can provide structure for graphics.
This is based on joint work with Catherine Hurley of the National University of Ireland, Maynooth.
Time: 4:00 pm.
Place: MC 5136 (top)
Speaker: Bonnie Jeanne MacDonald, University of Waterloo
Stat & Act. Sci. Seminar Title: The Cost of Basic Needs for the Canadian Elderly
Abstract: Our project determines the after-tax income required to finance the basic needs for Canadian seniors living in different circumstances in terms of age, gender, city of residence, household size, homeowner or renter, means of transportation and health status. Using 2001 as our base year, we price the typical expenses for food, shelter, medical, long-term care, transportation and miscellaneous basic living items for seniors living in Halifax, Montreal, Toronto, Calgary and Vancouver. Such information is important for seniors, prospective retirees, financial planners, policy makers and actuaries in assessing the minimum level of income required in retirement and the adequacy of savings and income security programs.
Our study is unique because it is the first Canadian study of basic living expenses tailored to seniors rather than simply to adults in general, prepared on an absolute rather than a relative basis. We also uniquely account for an individual's life circumstances, as listed above, rather than simply their city of residence and household size.
Time: 4:00 pm.
Place: DC 1304 (top)
Speaker: Abdel El-Shaarawi, Water Resources Institute and McMaster University
Stat & Act. Sci. Seminar Title: Matrix Inversion and Statistical Data Analysis
Abstract: Recent work on spatial temporal analysis of environmental data requires the inversion of large dimensional symmetric matrices. Although Gaussian elimination is the best-known general direct method for matrix inversion, special methods will often be useful when inverting matrices of special form. Indirect methods are chiefly iterative in nature and can be used in combination with direct methods to improve the accuracy of the computation. The objectives of this talk are to review these methods, suggest improvements, present some environmental applications, and link the computation of matrix inversion to statistical thinking.
Keywords: Matrix inversion, Gaussian Elimination, Choleski decomposition, Cross Validation
Time: 3:00 pm.
Place: MC 5158 (top)
Speaker: Sunny Wang, University of Waterloo
Stat & Act. Sci. Seminar Title: Parallel Computing in R
Abstract: Large data sets usually make computation very intensive. In most of implementation of the algorithms, parallel computing is employed to speed up computations. A parallel computer is a kind of computer with multiple processors acting to achieve some common goal. Some statistical problems are easily parallelized, such as Monte Carlo simulation, bootstrap, cross-validation and optimization with multiple restarts. However, some problems are hard to parallelize: EM algorithm, Optimization, MCMC, Clustering and most supervised learning algorithms. In this talk, we are going to demonstrate how to program parallel R code for bootstrap, cross-validation, EM algorithm and K-means.
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Alain Vandal, McGill University
Stat & Act. Sci. Seminar Title: Marginal Modelling of Capture-Recapture Data
Abstract: In the absence of individual covariate information, the usual representation for capture-recapture data from k lists is that of an incomplete 2^k-1 contingency table, possibly subject to stratification. The unobserved cell in the table is the number of unobserved individuals in the population of interest. Log-linear models form a natural and popular class of models to predict the value in the unobserved cell, loosely speaking. Such models usually take the cells of the contingency table to be distributed multinomially and model the cell expectations or probabilities: this is joint log-linear modelling.
An alternative is to model the expected number of individuals in the intersection of lists in each subset of the set of all lists, an option called marginal log-linear modelling. Using the Mˆbius inverse of a simple measure of list dependence, we show how to set up and fit marginal log-linear models in such a way that the parameters of interest correspond to a so-called dependence dividend (DD). The DDs quantify the contribution of each subset of a set Q of lists to the overall dependence of the lists in Q. The DD interpretation of the marginal model coefficient make every such model interpretable in a sense, and provide a common ground to compare and interpret all joint and marginal log-linear models, including non-hierarchical ones. Further, as DD estimates are available nonparametrically, they provide a way to assess a form of model self-consistency. We illustrate these possibilities on toy and real data. This is joint work with Elizabeth L. Turner, London School of Hygiene & Tropical Medicine, and Russell Steele, McGill University.
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker:Bonnie Jeanne MacDonald, University of Waterloo
Stat & Act. Sci. Seminar Title: Population Microsimulation Modeling 101 Statistcs Canada LifePaths
Abstract: This presentation will provide a general overview of Statistics Canada LifePaths - a microsimulation model designed to simulate life histories of Canadians, taking account of birth, death, immigration status, inter-provincial migration, marital history (including common-law unions), educational history, employment history and the birth and presence of children at home.
The beauty and purpose of the LifePaths model is that it enables us to pose and answer O grave ; what if O acute ; questions relating to government policies having an essentially longitudinal component and whose nature requires evaluation at the individual or family level, such as post-secondary education costs and benefits or public pension sustainability. It can also be used to explore a variety of societal issues of a longitudinal nature such as intergenerational equity or time allocation over entire lifetimes.
LifePaths is free, modifiable and publicly available on the Statistics Canada website. This highly-advanced and sophisticated tool is an exciting initiative since it provides the foundation upon which a researcher can leverage their effort to tackle many more and interesting questions and topics. This talk will be introductory and suitable for all.
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Jason Nielsen, Carleton University Department of Mathematics
Stat & Act. Sci. Seminar Title: Quantile Functions Distributed over Space and Time
Abstract: The quantile function maps a probability into events that will not be exceeded with that probability. It is substantially more useful in applications where interest is focused primarily on extreme outcomes, such as risk management, than its functional inverse the cumulative distribution function. We develop a partial differential equation for a quantile function varying over space and/or time, and use methods recently developed for parameter estimation for ordinary differential equation systems in this context. Spatially and seasonally varying distributions of rainfall in the Canadian prairies are estimated in order to assist food producers with crop management.
(This is joint work with Jim Ramsay, McGill University)
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Dr. Krzysztof M. Ostaszewski, Actuarial Program Director and Professor of Mathematics, Department of Mathematics, Illinois State University
Stat & Act. Sci. Seminar Title: Demand Uncertainty Reduces Market Power and Enhances Welfare
Abstract: Classical welfare economics assumes that the demand function, or consumers' utility, is known with certainty. Probabilistic microeconomics generalizes it by maximizing expected utility, or by optimizing under a specific constraint. Existing research has provided only limited insight into the welfare effects of demand uncertainty, and that limited insight suggests welfare reduction as a result of demand uncertainty. In contrast with previous works, we do not prescribe the form of demand uncertainty, but rather derive it from individual consumers' choices. We then analyze monopolist optimization problem, first constrained by a "Safety-First" type condition imposed on the coefficient of variation, and then by considering risk-adjusted profit measure. Our results indicate that the Marshallian welfare measure, when compared with the deterministic model, increases with uncertainty of the demand function.
We point out that uncertainty characterizes markets that lie between the pure monopoly model, and perfect competition model. We believe that our model of demand uncertainty is a realistic one, very much like observed behavior of markets. Most importantly, our work suggests that transition from monopolistic market structure to competitive one may be explained better by demand uncertainty than by mere presence of competitors.
Finally, we show how a demand can be efficiently estimated from simple consumer surveys (admitting its random structure).
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: John Manistre, Vice-President Risk Research, AEGON NV, Baltimore, Maryland, USA.
WatRISQ Webinar Title: A Cost of Capital Approach to Credit and Liquidity Spreads
(*Please Note: Link to Webinar will be live one hour prior to showing)
Abstract: The Cost of Capital method is an approach to valuing non-hedgeable, e.g. underwriting, risk that the European Chief Risk Officers Forum has endorsed for market consistent financial reporting. When the method is applied analyze hedgeable, i.e. market, risk the differences between model and market prices give us insight into both the risk and the valuation methodology. This paper builds a cost of capital model to value credit and liquidity risk issues that comes close enough to explaining observed credit and liquidity spreads to be useful. Potential applications of the model consist of filling "holes" in observed market data and justifying the use of credit and liquidity spreads to value insurance liabilities.
Time: 4:00 p.m.
Location: The Davis Centre, Room 1302 (*Please Note: Location not the usual room.) top
This presentation is jointly sponsored by WatRISQ and the Department of Statistics and Actuarial Science.
Speaker: Andrew Morton, alumnus of UW (BA in Math) and is a co-developer of the Heath, Jarrow and Morton interest rate model
WatRISQ Seminar Title: Quantitative Modeling on Wall Street, Pre and Post Credit Crisis
(*Please Note: Link to Webinar will be live one hour prior to showing)
Abstract: Quantitative techniques have been in heavy use on Wall Street since the development of Black-Scholes. The failure of some of these models has been listed by some as a significant factor in the onset of the credit crisis. We discuss these issues and how practitioners are deploying mathematical techniques in the post crisis world.
Time: 4:00 p.m.
Location: The Davis Centre, Room 1304 top
This presentation is jointly sponsored by WatRISQ and the Department of Statistics and Actuarial Science.
Speaker: Martin Crowder, Department of Mathematics Imperial College London, UK
Stat & Act. Sci. Seminar Title: Estimating functions for repeated measures with incidental parameters
Abstract: Repeated measures, or longitudinal data, are considered. The statistical characteristics for each individual case are supposed to be governed by a structural parameter, common to all, and an incidental parameter, specific to the individual. This terminology was used by Neyman and Scott (1948), who studied the properties of estimators in a likelihood framework. In this talk the model specification is taken to be more limited, not sufficient to construct a proper likelihood function. The proposal here is to seek an estimating function, based on the data and the structural parameter alone, whose maximum has an identifiable limit as the sample size grows. Then a transformation of the maximum is sought so that the modified version is a consistent estimator. Some examples are worked through and asymptotic distributions of the resulting consistent estimators are outlined to enable tests and confidence regions to be derived. Relative efficiency of competing estimators is also considered.
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Nalini Ravishanker, Department of Statistics University of Connecticut- Cancelled
Stat & Act. Sci. Seminar Title: TBA
Abstract:TBA
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Ali Ghodsi, Department of Statistics University of Waterloo
Stat & Act. Sci. Seminar Title:Visualization and Classification on subspace
Abstract: In this talk I will introduce a novel algorithm and its applications in visualization, classification and variable selection.
Conventional regression and classification methods fail to produce satisfactory results when applied to high-dimensional data sets. This is partially due to the problem of curse of dimensionality. In order to tackle this problem, one approach is to reduce the dimensionality of the covariate data by projecting it on to a subspace. Principal components analysis (PCA) is the most prominent subspace method.
However, the subspace modeled by PCA captures only the maximum variability in the data and therefore principal components are independent from response variable. In this talk, A new algorithm will be introduced that we call `Supervised Principal Component Analysis'. This algorithm estimates a sequence of principal components with maximum dependence to the response variable. We will show that conventional PCA is a special form of this more general frame work. Similar to PCA, supervised PCA can be solved in closed-form but unlike PCA it is well suited for classification and regression problems. We derive a dual form of Supervised PCA that significantly reduces the computational complexity of problems in which the number of predictors greatly exceeds the number of observations (such as DNA microarrays). Furthermore, the approach will be extended by kernelizing it, allowing for estimating non-linear dimensionality reduction.
This is a joint work with Elnaz Barshan and Hadi Zarkoob
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Mathieu Sinn, Department of Computer Science University of Waterloo
Stat & Act. Sci. Seminar Title: Detecting Change-Points in Time Series by Kernel Mean Matching of Ordinal Pattern Distributions
Abstract: A novel approach to detecting change-points in time series is proposed. Being computationally fast and robust, it is well-suited for the exploration of high-resolution biophysical time series where the exact calibration of the measurement devices is unknown or varies within time. The basic idea is to look at the order of successive values, as formally represented by ordinal patterns. Comparing the distribution of ordinal patterns in different parts of a time series allows to detect change-points. By applying the recently introduced Kernel Mean Matching criterion, this procedure can be automatized.
For time series resulting from ergodic processes, this method is consistent. In the Gaussian case, further asymptotic properties can be derived. The performance is evaluated in simulation studies; furthermore, the application to the analysis of electroencephalography (EEG) and electrocardiography (ECG) data is demonstrated.
This is joint work with Ali Ghodsi (Department of Statistics and Actuarial Science, University of Waterloo) and Karsten Keller (Institute of Mathematics, University of Lübeck, Germany)
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Benjamin Avanzi, University of New South Wales in Australia,
Stat & Act. Sci. Seminar Title: On a mean reverting dividend strategy with Brownian motion
Abstract: In actuarial risk theory, the introduction of dividend pay-outs in surplus models goes back to Bruno de Finetti (1957). Dividends strategies that can be found in the literature often yield pay-out patterns that are inconsistent with actual practice. One issue is the high variability of the dividend payment rates over time. We aim at addressing that problem by specifying a dividend strategy that yields stable dividend pay-outs over time.
In this paper, we model the surplus of a company with a Brownian risk model. Dividends are paid at a constant rate g of the company's modified surplus (after distribution of dividends), which operates as a buffer reservoir to yield a regular flow of shareholders' income. The dividend payment rate reverts around the drift of the original process µ, whereas the modified surplus itself reverts around the level l=µ /g.
We determine the distribution of the present value of dividends when the surplus process is never absorbed. After introducing an absorbing barrier a (inferior to the initial surplus) and stating the Laplace transform of the time of absorption, we derive the expected present value of dividends until absorption. The calculation of the optimal value of the parameter l (and equivalently g) is discussed. We conclude by comparing both barrier and mean reverting dividend strategies.
Keywords: dividends, Brownian motion, Ornstein-Uhlenbeck process, Mean reverting
JEL codes: G35, G32, G22, C44
Full paper: available from SSRN: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1504401
DE FINETTI, B. (1957) Su un'impostazione alternativa della teoria collettiva del rischio. Transactions of the XVth International Congress of Actuaries, 2, 433--443.
Time: 4:00 pm.
Place: MC 5158 (top)
Speaker: Benjamin Avanzi, University of New South Wales in Australia,
Stat & Act. Sci. Seminar Title: Forced Savings and Annuitisation with Cross-Subsidies: A Mutation of the Beast
Abstract: We adopt a two-tiered economic approach to classify national retirement savings schemes. We extend the plain vanilla system of forced savings by allowing for annuitisation and cross-subsidies.
The first tier of our model is controlled by government, which mandates contribution rates, interest rates, and conversion into benefits. In contrast, agents make voluntary contributions in the second tier, which earn interest at a rate broadly reflecting market conditions and any cross-subsidy between both tiers. Cross-subsidies within the mandated tier and between both tiers allow for social redistribution as well as the creation of a liquid market of privately provided annuities.
We conclude with a discussion of the Swiss and Australian systems of retirement savings as seen through the lens of our model.
Keywords: mandated retirement savings, annuitisation, pensions, regulation
JEL codes: J26, H55, D91, E21
Full paper: available from SSRN: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1466386
Time: 4:00 pm.
Place: MC 5158 (top)
Statistics and Actuarial Science
Math and Computer Building
University of Waterloo
200 University Avenue West
Waterloo, Ontario, Canada N2L 3G1
Main Office: (519) 888-4567 x33550
FAX (519) 746-1875
|
| privacy statement
Last Modified: Monday 4 January 2010
