CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Estimation of the spatial weighting matrix for spatiotemporal data with structural breaks Authors:  Philipp Otto - Leibniz University Hannover (Germany) [presenting]
Rick Steinert - European University Viadrina (Germany)
Abstract: A two-step penalized regression approach is proposed to estimate the entire spatial dependence structure of a spatiotemporal process under the presence of structural breaks in the mean. Simultaneously, we address an important problem in spatial econometrics. The classical approach is to replace the unknown spatial dependence structure by a linear combination of a scalar and a predefined, non-stochastic weighting matrix describing the dependence. In contrast to this classical approach, we estimate all entries of this weighting matrix by a penalized regression approach. In addition, we suppose that there might be an unknown number of structural breaks in the data. These breaks can occur at different time points for each location and they can be of different magnitude. For estimation of the model parameters, we propose a two-step estimation approach. In the first step, we determine a set of candidate change points assuming independent univariate time series. Consequently, the spatial dependence structure is ignored in this first step. This set is then passed to the full model to estimate the changes and the spatial dependence simultaneously by an adaptive lasso approach.