Oracally Efficient Estimation of Vector Nonlinear Additive Autoregressive Models

• Activity Number: 607
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Joshua Patrick* and Jiaming Xie
• Companies: and UC Davis
• Keywords: vector autoregressive ; oracle ; splines ; local linear regression ; solar irradiance

Abstract

We estimate vector nonlinear autoregressive time series models by imposing an additive structure and using spline-backfitted local linear regression. This estimation method uses a spline procedure to pre-estimate the different components of the time series model and then backfitting to obtain pseudo-responses which are then estimated using local linear regression. The backfitting stage helps to reduce the curse of dimensionality while having the asymptotic properties of the local linear smoother. Through simulation, we show that the relative efficiency of the fit to the oracle estimate approaches one as the series length increases. We apply the estimation method to solar irradiance data obtained from the University of Oregon Solar Resource Monitoring Lab.