The paper analyzes identification issues raised by general dynamic specifications of the trend, as opposed to the conventional random walk, in both univariate and multivariate models of permanent-transitory components. General trends are interpreted as representing processes of diffusion of technical change. In this framework, we prove that a I91) process admits permanent-transitory decompositions where the trend and the covariance between trend and stationary components are arbitrary. Moreover, we show that ARMA approximations to diffusion processes can generate non-fundamental moving average representations, thereby posing a non-standard identification problem of the relevant impulse-response functions.
Download is not available