Model-free Asymptotically Best Forecasting of Stationary Economic Time Series
- 1 September 1990
- journal article
- research article
- Published by Cambridge University Press (CUP) in Econometric Theory
- Vol. 6 (3) , 348-383
- https://doi.org/10.1017/s0266466600005314
Abstract
Given observations on a stationary economic vector time series process we show that the best h-step ahead forecast (best in the sense of having minimal mean square forecast error) of one of the variables can be consistently estimated by nonparametric regression on an ARMA memory index. Our approach is based on a combination of the ARMA memory index modeling approach of Bierens [7] with a modification to time series of the nonparametric kernel regression approach of Devroye and Wagner [16]. This approach is truly model-free, as no explicit specification of the distribution of the data generating process is needed.Keywords
This publication has 32 references indexed in Scilit:
- Automatic smoothing parameter selection: A surveyEmpirical Economics, 1988
- Armax model specification testing, with an application to unemployment in the NetherlandsJournal of Econometrics, 1987
- Kernel estimation and interpolation for time series containing missing observationsAnnals of the Institute of Statistical Mathematics, 1984
- Model specification testing of time series regressionsJournal of Econometrics, 1984
- Uniform Consistency of Kernel Estimators of a Regression Function Under Generalized ConditionsJournal of the American Statistical Association, 1983
- Unbiased determination of production technologiesJournal of Econometrics, 1982
- Projection Pursuit RegressionJournal of the American Statistical Association, 1981
- On the bias in flexible functional forms and an essentially unbiased formJournal of Econometrics, 1981
- Robust Methods and Asymptotic Theory in Nonlinear EconometricsPublished by Springer Nature ,1981
- Rational expectations and the dynamic structure of macroeconomic modelsJournal of Monetary Economics, 1978