Business cycle analysis without much theory: A look at structural VARs (Q1377305)

VAR-models are plagued with an ambiguity of the results because these depend on the ordering of the variables selected by the researcher. This disadvantage has been overcome by switching to a new class of VAR-models, the structural VAR-models (SVAR). This class distinguishes from the atheoretical VAR-approach by imposing only weak theoretical restrictions, thus still allowing robust inferences. These restrictions are similar to the identification restrictions in standard econometrics but operate on the contemporaneous covariance matrix of the reduced form residuals instead on the structural coefficient matrices. Types of restrictions are normalization, exclusion restrictions and restrictions on certain elements of the contemporaneous variance-covariance matrix of the reduced form residuals. The latter type of restrictions allows one to choose among short-term or long-term restrictions adopted from economic theory. The number of restrictions imposed leads to a just identified system. Despite of these restrictions the SVAR-approach remains flexible enough to model the dynamics of the system data-based. Applications are to a large extent analyses of business cycle fluctuations. The authors cast doubt on the robustness of the conclusions drawn from SVAR estimates on the one hand, and they also demonstrate that SVAR-models fail to be identified despite of the restrictions imposed. This identification failure results from the just-identifying restrictions instead of over-identifying restrictions which can usually be tested. But imposing over-identifying restrictions in SVAR-models normally lead to serious estimation problems. SVAR models of business cycle fluctuations address the importance of demand and supply shocks on the one hand, and the transitory nature of nominal shocks as well the permanent nature of supply shocks on the other. The authors cast doubt on the conclusions drawn of the relative importance of both types of shocks. As they argue the empirical importance of technology and other shocks depends crucially on the just-identifying assumptions. Put in other words, using alternative just-identifying assumptions the results may be reversed. Nevertheless, SVAR models are able to reveal certain empirical regularities in the data and the results are robust with respect to the stylized fact of a hump shaped response of output to technology shocks.