We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Kernel-based Multistep-ahead Predictions of the US Short-term Interest Rate.
- Authors
Gooijer, Jan G. De; Zerom, Dawit
- Abstract
Recently, there has been an increasing demand for tools and methods that can detect and summarize structures and relationships between time series variables without relying too much on specific assumptions on the structure of the mean responses, such as linearity, or on error distributions, such as normality. In this paper the author is concerned exclusively with the multistep-ahead non-parametric prediction of weekly observations of the short-term interest rate, that is, 90-day U.S. T-bill rate. In doing SO, he considers three kernel-based prediction methods, i.e. the conditional mean, the conditional median, and the conditional mode. By employing three kernel-based prediction methods, he showed that multistep-ahead predictions of U.S. weekly T-bill rate changes using the conditional mean and the conditional mode are superior-or at least equal-to predictions based on the Random Walk. Moreover, relative to the AR model these predictors also perform better, but marginally. The conditional mean is good at predicting upward directions of change, whereas the conditional mode is better at predicting downward changes in the relatively volatile time series of T-bill rate changes.
- Subjects
FORECASTING; INTEREST rates; PREDICTION models; STATISTICS; MEDIAN (Mathematics); MARKET volatility
- Publication
Journal of Forecasting, 2000, Vol 19, Issue 4, p335
- ISSN
0277-6693
- Publication type
Article
- DOI
10.1002/1099-131X(200007)19:4<335::AID-FOR777>3.0.CO;2-3