The paper considers time series GMM models where a subset of the parameters are time varying. The magnitude of the time variation in the unstable parameters is such that efficient tests detect the instability with (possibly high) probability smaller than one, even in the limit. We show that for...