Working Paper: CEPR ID: DP8515
Authors: Lutz Kilian
Abstract: Structural vector autoregressive (VAR) models were introduced in 1980 as an alternative to traditional large-scale macroeconometric models when the theoretical and empirical support for these models became increasingly doubtful. Initial applications of the structural VAR methodology often were atheoretical in that users paid insufficient attention to the conditions required for identifying causal effects in the data. In response to ongoing questions about the validity of widely used identifying assumptions the structural VAR literature has continuously evolved since the 1980s. This survey traces the evolution of this literature. It focuses on alternative approaches to the identification of structural shocks within the framework of a reduced-form VAR model, highlighting the conditions under which each approach is valid and discussing potential limitations of commonly employed methods.
Keywords: Identification; Structural Model; VAR
JEL Codes: C32; C51
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| imposed restrictions on the model (C24) | ability to interpret the resulting shocks as structural (C50) |
| structural shocks (E32) | understanding economic phenomena such as recessions or energy price spikes (E32) |
| identification of structural shocks through short-run restrictions (C22) | accurate impulse responses (C58) |
| valid identification (Y50) | accurate reflection of causal relationships (C32) |
| non-recursive models (C52) | ambiguous causal interpretations (C20) |