A0405
Title: Reconstructing partially observed functional data via factor models of increasing rank
Authors: Maximilian Ofner - Graz University of Technology (Austria) [presenting]
Siegfried Hoermann - Graz University of Technology (Austria)
Abstract: In functional data analysis, applied researchers often face the problem of missing fragments. To recover the missing information from the observed parts, linear reconstruction operators have been introduced in the literature. In this setting, we present a new approach for estimating linear reconstructions using approximate factor models of increasing rank. The proposed methodology aims at discretely sampled functional data with additive noise and avoids restrictive smoothness conditions. Under a triple asymptotic, we establish uniform convergence rates of our estimator. Furthermore, we discuss a simple and effective method for constructing simultaneous prediction bands. Finite sample properties of the proposed procedures are then examined in a simulation study. The methodology is finally illustrated by a set of incompletely observed temperature data.
