Departament d'Estadística i Investigació Operativa

Comparteix:

Programació del Seminari: Any 2002



A MONTH IN MEXICO AND A DAY IN VANCOUVER i RESEARCH METHODS
CONVIDAT:  Dr. Tony Greenfield, Industrial research consultant, Greenfield Research (U.K.)

IDIOMA: anglès

LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.

DATA: Divendres, 22 d'octubre de 2002, 12h30m

RESUM: The Seminar will have two parts and therefore two titles A month in Mexico and a day in Vancouver: Lots of pictures and a few observations that might stimulate further study. The Indian inch?. There is no mathematics worth mentioning. Just a simple calculation and a very simple and ingenious use of statistics for a first estimate of the Indian inch which is what I call the basic measurement unit of the ancient Maya. Research Methods: After interacting with lots of researcher of many fields I realized that a course covering the many different aspects of research methods and the many abilities that a researcher needs to put into practice to be effective was needed so I edited a book with the help of about 20 contributors. In the seminar I will comment briefly different aspects of it and give some tips.


IS STOCHASTIC VOLATILITY MORE FLEXIBLE THAN GARCH?
CONVIDAT: Prof. Esther Ruiz Ortega. Dep. de Estadística y Econometría, Universidad Carlos III de Madrid

IDIOMA: Castellà

LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.

DATA: Divendres, 22 de novembre, 12h30m

RESUM: This paper compares the ability of GARCH and ARSV models to represent adequately the main empirical properties usually observed in high frequency financial time series: high kurtosis, small first order autocorrelation of squared observations and slow decay towards zero of the autocorrelation coefficients of squared observations. We show that the ARSV(1) model is more flexible than the GARCH(1,1) model in the sense that it is able to generate series with higher kurtosis and smaller first order autocorrelation of squares for a wider variety of parameter specifications. Our results may help to clarify some puzzles raised in the empirical analysis of real financial time series.

RESUM EXTENS: clica aquí


BIVARIATE MARKER DEVELOPMENT WITH CENSORED VALUES AND INFORMATIVE DROPOUT

CONVIDAT: Ronald Geskus. Municipal Health Service,cluster of infectious diseases, Amsterdam. Dept. of Medical Statistics and Bioinformatic, Leiden University Medical Center. Leiden

IDIOMA: anglès

LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.

DATA: Dilluns, 2 de desembre de 2002, 12h30m

RESUM: A joint model is presented for the bivariate development of CD4 count and HIV-1 RNA load over time since seroconversion and the relation of the fitted marker values to the AIDS risk. Data from 400 homosexual men from the Amsterdam Cohort Study and the French SEROCO Study were used, all having a documented interval of seroconversion. For the marker development, a random effects model was used. The fitted values were included as time-varying covariates in a Cox proportional hazards model. A Bayesian approach was used, and models were fitted via MCMC using the WinBUGS package. The advantage of the Bayesian approach is that the integrals that appear in the likelihood play no role; moreover, left censored RNA load values are included without any major complication. However, convergence of the chains is a problem, and very much depends on the parametrisation used. The results from the joint model are compared with the results from an approach that fits the random effects model and the Cox model separately and with the results from an ordinary least squares model. The separate random effects model, which does not correct for random dropout, only gives a minor bias in the estimates of the population parameters. In the estimates of the individual random effects, results diverge for persons with incomplete follow-up. Joint models play an important role in prediction of residual time to AIDS based on previous marker development. Moreover, it allows to discriminate between direct effects and effects that are mediated through the markers. For both parts of the model, we investigated the effects of age and three genetic cofactors in this way.


THE SAR PROCEDURE: A DIAGNOSTIC ANALYSIS OF HETEROGENEOUS DATA

CONVIDAT: Daniel Peña. Dep. de Estadística y Econometría, Universidad Carlos III de Madrid

IDIOMA: Castellà

LLOC: Seminari 1, Edifici U, Facultat de Matematiques i Estadistica, C/ Pau Gargallo 5.

DATA: Divendres, 13 de desembre de 2002, 12h30m

RESUM: This paper presents a procedure for detecting heterogeneity in a sample with respect to a given class of models. It can be applied to determine if a sample of univariate or multivariate data has been generated by different distributions, or if a regression equation is really a mixture of different regression lines. The basic idea of the procedure is first to split the sample into small homogeneous subgroups and then recombine the observations in the subgroups to form homogeneous clusters. The splitting and recombining scheme forms the core of the SAR procedure that is implemented iteratively in three steps. First, outlier cleaning, an iterative process of model fitting and outlier detection is applied to the sample to eliminate isolated outliers. Second, splitting, the cleansed sample is split into more homogeneous subgroups by using a measure of association among the points in the sample given the model. The splitting is continued until groups of minimal size that cannot be split further are found. Third, recombining, a model is fitted to each homogeneous minimal group, and the rest of the observations in the sample are checked iteratively one by one for homogeneity and incorporation into the group. The final result of this entire interative process is a set of possible data configurations (PDC) for the sample. The proposed procedure is exploratory and can be applied to detect heterogeneity in any statistical modeling. The performance of the procedure is illustrated in univariate, multivarite and linear regression problems.