Title: Cointegration and Representation of Cointegrated Autoregressive Processes in Banach Spaces

Abstract: We extend the notion of cointegration for time series taking values in a potentially infinite dimensional Banach space. Examples of such time series include stochastic processes in C[0,1] equipped with the supremum distance, those in a finite dimensional vector space equipped with a non-Euclidean distance, etc. We then provide a suitable generalization of the Granger-Johansen representation theorem for such time series. To achieve our goal, we first note that an autoregressive law of motion is characterized by a simple linear operator pencil, so study its spectral properties to obtain a geometric condition that is necessary and sufficient for the inverse to have a pole of order 1 or 2 around one. Each of these conditions forms a fundamental basis for our I(1) or I(2) representation theorem for autoregressive processes. The representation theorems in this paper are derived under much weaker assumptions than the existing versions. Moreover, they provide a complete characterization of I(1) or I(2) solutions: except for the terms depending on initial values, all the other components are explicitly given in closed forms.