Identifying long-run behaviour with non-stationary data.

Results for the identification of non-linear models are used to support the traditional form of the order condition by sufficient conditions. The sufficient conditions reveal a two step procedure for firstly checking generic identification and then testing identifiability. This approach can be extended to sub-blocks of the system and it generalizes to non-linear restrictions. The procedure is applied to an empirical model of the exchange rate, which is identified by diagonalising the system.