erc/metu
INTERNATIONAL CONFERENCE IN ECONOMICS  IV
September 13-16, 2000, Ankara

Statistical Properties of GARCH Models and Stylized Facts of Financial Time Series

Timo Teräsvirta (Stockholm School of Economics,Sweden)

Abstract

Financial series of sufficiently high frequency usually display rather high kurtosis. At the same time, the absolute and squared observations have an autocorrelation function that starts at rather low positive values and decays slowly with increasing lag length. Furthermore, it has been found that the autocorrelations of the absolute values themselves are higher than those of any other positive power of the absolute values. Generalized Autoregressive Heteroskedasticity (GARCH) models may be seen as a popular way of modelling these features in the data in a parsimonious way. In this presentation, the moment structure of some well-known GARCH models is considered and the potential of these models to explain the volatility clustering present in financial series assessed. Differences in the moment structure of the standard GARCH, the exponential GARCH models, and the Stochastic Volatility models are considered and their practical significance discussed.

Economic Research Center
Middle East Technical University
06531 Ankara Turkey
Phone: + 90 312 210 3044, 210 2003
Fax: +90 312 210 1244
e-mail: metuerc@metu.edu.tr