Business Cycle Analysis and VARMA models

An important question in empirical macroeconomics is whether structural vector autoregressions (SVARs) can reliably discriminate between competing DSGE models. Several recent papers have suggested that one reason SVARs may fail to do so is because they are finite-order approximations to infinite-order processes. In this context, we investigate the performance of models that do not suffer from this type of misspecification. We estimate VARMA and state space models using simulated data from a standard economic model and compare true with estimated impulse responses. For our examples, we find that one cannot gain much by using algorithms based on a VARMA representation. However, algorithms that are based on the state space representation do outperform VARs. Unfortunately, these alternative estimates remain heavily biased and very imprecise. The findings of this paper suggest that the reason SVARs perform weakly in these types of simulation studies is not because they are simple finite-order approximations. Given the properties of the generated data, their failure seems almost entirely due to the use of small samples.