ETD

• causal inference
• independent component analysis
• recursiveness
• structural VARs
• uncertainty shocks

Recursive identification is one of the most widely used strategies for identifying structural shocks in structural vector autoregressions, largely because of its simplicity and transparency. However, its validity depends on a strong assumption, namely that the contemporaneous causal structure of the economic system can be represented by a recursive ordering of the variables, an assumption whose epistemological status is often left unexamined. This thesis develops a procedure to assess that assumption empirically rather than imposing it a priori. The proposed methodology combines independent component analysis with bootstrap inference. The structural impact matrix is first estimated without recursive zero restrictions, exploiting non-Gaussianity and statistical independence of the shocks. Bootstrap inference is then used to evaluate whether the estimated contemporaneous structure is statistically compatible with recursiveness. A complementary diagnostic based on Cholesky orthogonalisation and independence testing is also introduced to further evaluate whether some specific recursive ordering is supported by the data. The procedure is applied to an empirical analysis of uncertainty shocks in a trivariate VAR. The results indicate a sparse contemporaneous structure, broadly compatible with recursiveness, in which uncertainty appears largely isolated from industrial production and inflation. More importantly, recursiveness emerges from the estimated significance pattern of the data rather than being imposed at the identification stage. Overall, this thesis argues that the plausibility of recursive identification should itself become an object of empirical evaluation in SVAR analysis.