Panel vector autoregressions with binary data

We study identification of multivariate dynamic panel data logit models with unobserved fixed effects. We show that, in the pure VAR(1) case (without exogenous covariates), the parameters are identified with as few as four waves of observations and can be estimated consistently at rate n with an asymptotic normal distribution. Furthermore, we show that the identification strategy of Honoré and Kyriazidou (2000) carries over in the multivariate logit case when exogenous variables are included in the model. We also present an extension of the bivariate simultaneous logit model of Schmidt and Strauss (1975) to the panel case, allowing for contemporaneous cross-equation dependence both in a static and a dynamic framework. The results of this chapter are of particular interest for short panels, that is, for small T.