567 – Computing with Time Series Data
Semiparametric Forecasting of Nonlinear Temporal Processes
Jane Harvill
Baylor University
Nalini Ravishanker
University of Connecticut
The curse of dimensionality has been problematic in the application of nonparametric and semiparametric regression techniques to high-dimensional time series data. Spline-backfitted local linear (SBLL) and spline-backfitted kernel (SBK) estimators have been successful in addressing this problem, and provide computationally efficient estimators. Moreover, under fairly weak conditions, the estimators are point-wise asymptotically normal. Little work has been conducted in investigating the properties of forecasts using models estimated via SBLL or SBK methods. We propose a method for SBLL and SBK forecasting, and investigate the properties of those forecasts. For illustration, we apply the forecasting methods to irradiance data collected from a solar power plant in Lanai, Hawaii provided by Sandia Research Laboratories.
