Computationally Intensive Spectrum Estimation Methods and Nonstationary

• Activity Number: 161
• Type: Contributed
• Date/Time: Monday, August 9, 2004 : 2:00 PM to 3:50 PM
• Sponsor: General Methodology
• Author(s): Juana Sanchez*+
• Companies: University of California, Los Angeles
• Keywords: bootstrap ; splines ; spectrum ; ensemble ; nonstationarity ; seasonality

Abstract

Considerable progress has been made in estimating the spectral distribution of a stationary time series. New computationally intensive methods such as log-spline, ensemble and bootstrap are now easy to implement. When a time series is nonstationary, however, the user is at the mercy of the controversy on the appropriate way to make the data stationary so that the spectrum can be estimated accurately. If, in addition to nonstationarity, the data present seasonal frequencies that are so dominant that it is very hard to discern the presence of other frequencies, one has to worry about what to do with those. The nonexistence of a unified way to approach the problem of making a nonstationary dataset stationary and the problem of dominant frequencies without losing information about the data make it hard to avoid that controversy. I build on the latest computationally intensive spectrum estimation methodology to explore ways to integrate the spectrum estimation and handling of nonstationarity in a unified framework that retains all the information about the data.