I extend existing spatial panel data models to incorporate time variation in unknown network structures by means of structural breaks. For optimization I propose a two-step Adaptive Lasso algorithm to sequentially detect multiple unknown breaks in the spatial dependence and fully recover the underlying network. The use of MLE plus regularization enables to estimate the high-dimensional nonlinear model. Via an extensive simulation study I showcase the accuracy and robustness of the algorithm. Moreover, the accuracy maintains when either break detection or network recovery fails. Finally, I employ the algorithm to the regional housing prices in The Netherlands. The results illustrate the time-varying inter-connectivity, for which the possibility of a contiguity-based underlying network is rejected.

Spatial weighting matrix, structural breaks, time-varying spatial dependence, maximum likelihood estimation, housing market.
Sun, Y.
Erasmus School of Economics

Gruisen, A.R.J. (2020, May 28). Recovering Time-Varying Network Structures from Panel Data using Break Detection. Econometrie. Retrieved from http://hdl.handle.net/2105/52213