A1716
Title: Identification and estimation of network models using panel data analysis
Authors: George Kapetanios - Kings College London (United Kingdom)
Vasilis Sarafidis - Brunel University London (United Kingdom) [presenting]
Abstract: A novel methodology is proposed for identifying and estimating spatial and network models using large panel data. We address the challenge of estimating the spatial interactions between individual units by developing a boosting algorithm that relies on the statistical significance of individual neighbouring covariates tested one at a time. We refer to the proposed method as Boosting One Neighbour at a Time Multiple Testing (BONMT) procedure. Our approach allows for the flexible selection of neighboring units in the presence of high-dimensional networks, even in cases where the cross-sectional dimension of the panel is larger than the number of time series observations available. Furthermore, our procedure is robust to unequal weights, i.e., asymmetric network systems where some individuals influence their neighbors more strongly than others. Theoretical results are provided, demonstrating the algorithm's consistency and asymptotic properties under various network structures. The resulting IV estimator is shown to be consistent as N,T both grow large. Monte Carlo simulations illustrate the methods excellent finite-sample performance, confirm its robustness across different network configurations and levels of spatial correlation and show that it can outperform alternative methods.
