This paper presents the nonlinear IV methodology as an effective inferential basis for nonstationary panels. The nonlinear IV method resolves the inferential difficulties in testing for unit roots arising from the intrinsic heterogeneities and cross-dependencies of panel models. Individual units are allowed to be dependent through correlations among innovations, interrelatedness of short-run dynamics and/or cross-sectional cointegrations. If based on the instrumental variables that are nonlinear transformations of the lagged levels, the usual IV estimation of the augmented Dickey-Fuller type regressions yields asymptotically normal unit root tests for panels with general dependencies and heterogeneities. Moreover, the nonlinear IV estimation allows for the use of covariates to further increase power, and order statistics to test for more flexible forms of hypotheses, which are especially important in heterogeneous panels.
