Document Type
Lecture
Publication Date
11-12-2004
Abstract
We present three methods for forecasting using a functional coefficient autoregressive (FCAR) model for univariate and vector time series. The first method is a "naive forecast," and the form of the functional coefficient is determined using only the within-sample series values. The second method is a bootstrap predictor, and is a variation of the naïve forecast with predicted values computed using a bootstrap value of within-sample residuals from the fitted FCAR model. The final method is a multistage method, where the functional coefficients are updated at each step to incorporate the information from the time series in the predicted response. The three methods are applied to U.S. GNP and unemployment to compare performance and illustrate utility.
Relational Format
presentation
Recommended Citation
Harvill, Jane L., "Modeling and Prediction for Nonlinear Time Series" (2004). Probability & Statistics Seminar. 56.
https://egrove.olemiss.edu/math_statistics/56
