Implications of Partial Information for Econometric Modeling of Macroeconomic Systems

Representative models of the macroeconomy (RMs), such as DSGE models, frequently
contain unobserved variables. A finite-order VAR representation in the observed
variables may not exist, and therefore the impulse responses of the RMs and SVAR
models may differ. We demonstrate this divergence often is: (i) not substantial; (ii)
reflects the omission of stock variables from the VAR; and (iii) when the RM features I
(1) variables can be ameliorated by estimating a latent-variable VECM. We show that
DSGE models utilize identifying restrictions stemming from common factor dynamics
reflecting statistical, not economic, assumptions. We analyze the use of measurement
error, and demonstrate that it may result in unintended consequences, particularly in
models featuring I (1) variables.