Weighted Averages and Local Polynomial Estimation for Fractional Linear ARCH Processes
We consider local polynomial regression estimation for time series defined by Yi = μ(ti)+Ui where Ui is a stationary zero mean process with Wold decomposition Ui = C(B)Zi. The innovations Zi are assumed to be generated by a LARCH process, and may thus exhibit long-range correlations in volatility. The linear dependence structure, defined by the filter C(B) includes short memory, long memory and antipersistence. A central limit theorem for weighted averages, with triangular arrays of weights, and a limit theorem for local polynomial estimates of μ(ν)(i) are derived. The asymptotic distribution of μ(ν)(ti) turns out to be unaffected by long-range dependence in volatility. The question of optimal regression weights is also addressed.
AMS Subject Classification62G08 and 62M10
Keywordslong memory LARCH process volatility central limit theorem location estimation local polynomial estimation
Unable to display preview. Download preview PDF.
- Arcones, M.A., Bin Yu, 1994. Limit theorems for empirical processes under dependence. In: Chaos, Expansions, Multiple Wiener-Itô-Integrals and their Applications. CRC Press, London.Google Scholar
- Feng, Y., 2005. On the asymptotic variance in nonparametric regression with fractional time series errors. Preprint, Heriot-Watt University.Google Scholar
- Müller, H.G., 1987. Weighted local regression and kernel methods for nonparametric curve fitting. Journal of the American Statistical Association, 82 231–238.Google Scholar
- Rosenblatt, M., 1961. Independence and dependence. In: Proc. 4th Berkeley Symposium, J. Neyman (ed.), Univ. California, Berkeley Univ. Press, Berkeley, pp. 431–443.Google Scholar