This paper derives the restrictions that apply to a transformed model obtained by a minimal I(2)-to-I(1) transformation. The relationship between the parameters of the I(2) vector autoregression and the transformed model is characterized, including the coefficients of polynomially cointegrating relationships. In a simulation experiment, unrestricted reduced rank regression in the transformed model, which is common in applied work, yields only a minor loss of efficiency compared to imposing the restrictions. A properly transformed vector autoregression thus provides a practical and effective means for inference on the parameters of the I(2) model.