Title: Microeconometric dynamic panel data methods: Model specification and selection issues
Authors: Jan Kiviet - University of Amsterdam (Netherlands) [presenting]
Abstract: A strategy is presented to implement the GMM (generalized method of moments) tools provided by Stata package Xtabond2 such that an adequate model specification and matching set of instrumental variables is extracted from a panel data set to establish a single microeconometric structural behavioral presumably dynamic relationship. In the suggested specification search three comprehensive goals are pursued. Firstly including all the relevant appropriately transformed possibly lagged regressors, as well as any interactions between these if it is required to relax the strict homogeneity restrictions on the dynamic impacts of explanatories in standard linear panel models. Secondly correctly classifying all regressors as either endogenous, predetermined or exogenous, as well as either effect-stationary or effect-nonstationary, together implying the required differencing to obtain valid and relatively strong instruments from lagged internal variables. Thirdly enhancing the inference accuracy by both imposing valid coefficient restrictions and reducing the space spanned by the set of instruments through eliminating weak instruments. For the tests which trigger the decisions to be made considerations are spelled out to interpret their $p$-values. Complexities to identify the dynamic impact patterns are also unraveled. Finally, the strategy is applied to a classic data set and is shown to yield some new insights.