Leitura do Dia - Fitting non-Gaussian persistent data

Fitting non-Gaussian persistent data

Wilfredo Palma and Mauricio Zevallos

Applied Stochastic Models in Business and Industry

Special Issue: Fourth Brazilian Conference on Statistical Modelling in Insurance and Finance

Volume 27, Issue 1, pages 23–36, January/February 2011

Keywords:

* ARFIMA models;

* conditional variance;

* long-range dependence;

* persistence;

* quasi-maximum likelihood;

* prediction

Abstract

This paper discusses a new methodology for modeling non-Gaussian time series with long-range dependence. The class of models proposed admits continuous or discrete data and considers the conditional variance as a function of the conditional mean. These types of models are motivated by empirical properties exhibited by some time series. The proposed methodology is illustrated with the analysis of two real-life persistent time series. The first application is concerned with the modeling of stock market daily trading volumes, whereas the second application consists of a study of mineral deposit measurements. Copyright © 2010 John Wiley & Sons, Ltd.

Uma generalição muito valiosa dos modelos ARFIMA para modelos condicionais com distribuições não Gaussianas. Em especial acho a aplicação para dados de Trading Volume muito importante, já que estas séries são ainda pouco exploradas em finanças, em parte devido as dificuldades na sua modelagem. E seguindo a tradição do livro do Wilfredo Palma e das aulas do Maurício, uma apresentação muito clara.