Tests for Joint Normality in Time Series.
Abstract
The well-known methods for analysis of time series, whether in the time domain or in the frequency domain -- for fitting parametric structures, for regression, for forecasting -- all involve second-moment statistics. If all variables are jointly normally distributed in stationary sequences, simple first and second moments contain all the information. If not, there is the possibility that some of the needed information is not contained in the statistics used. When a random sequence is other than stationary and jointly normal, it may sometimes equally well be described and thought of as stationary but not jointly normal (which is the terminology used here) or as nonstationary.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1980
- Accession Number
- ADA085885
Entities
People
- Francis J. Anscombe
- Ho-len H. Chang
Organizations
- Yale University