Modeling the Density of a Distribution Containing a Jump Nonstationarity
Abstract
An algorithm for density estimation in the presence of a jump nonstationary is described. This approach is compared to an approach using a kernel estimator windowed to use a fixed number of data points. Assume a random variable X is distributed with probability density function (pdf) d sub 1 before time t sub J and with pdf d2 after time t sub j. Neither the densities nor t sub j are known. The data are viewed as arriving sequentially, with a requirement to report the pdf estimate after each point. A technique for density estimation under these conditions is proposed which models the density as a mixture of normal distributions. The nonstationarity of the data is accounted for through the use of a weighted window. This approach is compared to a windowed kernel estimator.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1991
- Accession Number
- ADA243852
Entities
People
- D. J. Marchette