Thursday 7 November 2019, 1.00PM to - 2.00 pm
Speaker(s): Mike Thornton (DERS)
Abstract: Given the growing availability of large datasets, we propose the spatio-temporal autoregressive distributed lag (STARDL) model which allows spatial and temporal coefficients to differ jointly across the spatial units. Our model encompasses the widely used spatial dynamic panel data models as well as the heterogeneous spatial autoregressive model recently proposed by Aquaro, Bailey and Pesaran (2015), the only paper in considering heterogeneous spatial parameters but without any dynamics. To deal with the simultaneity arising from spatial-lagged dependent variables, we develop both QML-based and control function-based STARDL estimators, which are shown to be consistent and asymptotically normally distributed when the time dimension is large, irrespective of whether the number of the spatial units is large or not. Furthermore, by deriving the system dynamic spatial panel data representation, we can develop the dynamic and the diffusion multipliers that can capture dynamic adjustments as well as network connectedness from initial to new equilibrium following an economic perturbation in a flexible manner. The utility of our proposed STARDL models is demonstrated by Monte Carlo studies as well as the empirical application to the Iraqi war casualties during 2003-2010.
Location: Staff Room - A/EC202
Admission: Staff and PhD