Show simple item record

dc.contributor.authorYuan, Miaoen_US
dc.date.accessioned2018-01-24T07:00:32Z
dc.date.available2018-01-24T07:00:32Z
dc.date.issued2016-08-01en_US
dc.identifier.othervt_gsexam:8496en_US
dc.identifier.urihttp://hdl.handle.net/10919/81905
dc.description.abstractRegarding quantifying uncertainties in prediction, two projects with different perspectives and application backgrounds are presented in this dissertation. The goal of the first project is to predict the corporate default risks based on large-scale time-to-event and covariate data in the context of controlling credit risks. Specifically, we propose a competing risks model to incorporate exits of companies due to default and other reasons. Because of the stochastic and dynamic nature of the corporate risks, we incorporate both company-level and market-level covariate processes into the event intensities. We propose a parsimonious Markovian time series model and a dynamic factor model (DFM) to efficiently capture the mean and correlation structure of the high-dimensional covariate dynamics. For estimating parameters in the DFM, we derive an expectation maximization (EM) algorithm in explicit forms under necessary constraints. For multi-period default risks, we consider both the corporate-level and the market-level predictions. We also develop prediction interval (PI) procedures that synthetically take uncertainties in the future observation, parameter estimation, and the future covariate processes into account. In the second project, to quantify the uncertainties in the maximum likelihood (ML) estimators and compute the exact tolerance interval (TI) factors regarding the nominal confidence level, we propose algorithms for two-sided control-the-center and control-both-tails TI for complete or Type II censored data following the (log)-location-scale family of distributions. Our approaches are based on pivotal properties of ML estimators of parameters for the (log)-location-scale family and utilize the Monte-Carlo simulations. While for Type I censored data, only approximate pivotal quantities exist. An adjusted procedure is developed to compute the approximate factors. The observed CP is shown to be asymptotically accurate by our simulation study. Our proposed methods are illustrated using real-data examples.en_US
dc.format.mediumETDen_US
dc.publisherVirginia Techen_US
dc.rightsThis Item is protected by copyright and/or related rights. Some uses of this Item may be deemed fair and permitted by law even without permission from the rights holder(s), or the rights holder(s) may have licensed the work for use under certain conditions. For other uses you need to obtain permission from the rights holder(s).en_US
dc.subjectDefault Risken_US
dc.subjectDynamic Factor Modelen_US
dc.subjectHigh-Dimensional Time Seriesen_US
dc.subject(Log)-Location-Scale Distributionsen_US
dc.subjectTolerance Interval.en_US
dc.titleCorporate Default Predictions and Methods for Uncertainty Quantificationsen_US
dc.typeDissertationen_US
dc.contributor.departmentStatisticsen_US
dc.description.degreePh. D.en_US
thesis.degree.namePh. D.en_US
thesis.degree.leveldoctoralen_US
thesis.degree.grantorVirginia Polytechnic Institute and State Universityen_US
thesis.degree.disciplineStatisticsen_US
dc.contributor.committeechairHong, Yilien_US
dc.contributor.committeememberSands, Lauraen_US
dc.contributor.committeememberZhu, Hongxiaoen_US
dc.contributor.committeememberKim, Inyoungen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record