Title: Exponential smoothing with locality parameter for functional data in robust forecasting of economic phenomena
Authors: Daniel Kosiorowski - Cracow University of Economics (Poland)
Jerzy Rydlewski - AGH University of Science and Technology (Poland)
Przemyslaw Jasko - Cracow University of Economics (Poland) [presenting]
Abstract: A variety of economic phenomena may be described as random functional variable or as a functional time series, i.e., as a family of such variables indexed by time. A functional view on the phenomena is very often more natural than classical view as in the functional view we do not need to divide economic system into separate parts. Economic functional time series very often consist of functional outliers of various kinds. A development of effective robust methods of forecasting of the economic time series is an issue of a prime importance for theory of economics as well as for applied economics and empirical finance. Due to a temporal dependency between observations well-know strategies of dealing with contaminated datasets are not directly applicable in case of the functional time series. We propose a version of exponential smoothing predictor for functional data based on selected depths for functional data. We discuss its properties and compare it with other predictors known form the literature. Theoretical considerations are illustrated via results of simulation studies and empirical example related to an analysis of a process a development of a city based on satellite maps.