Differential Geometrical Methods in Time Series.
Abstract
Inference for time series processes was investigated using differential geometrical methods as well as sampling based Bayesian methods. For univariate autoregressive moving average (ARMA) processes and fractionally integrated ARMA processes analytical forms of asymptotic properties of inference such as bias in parameter estimates and improved test statistics were obtained from geometrical quantities. These terms provide collections useful when the sample size is moderate or small. Markov chain Monte Carlo procedures facilitated modeling of univariate and multivariate ARMA and fractionally integrated ARMA processes in the Bayesian framework. This approach uses the exact likelihood function and is accurate even with small sample sizes. Outlier analysis, prediction and model selection were addressed. (AN)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1995
- Accession Number
- ADA304006
Entities
People
- Nalini Ravishanker
Organizations
- University of Connecticut