Working Paper: CEPR ID: DP11041
Authors: Fabio Canova; Mehdi Hamidi Sahneh
Abstract: Non-fundamentalness arises when observables do not contain enough information to recover the vector of structural shocks. Using Granger causality tests, the literature suggested that many small scale VAR models are non-fundamental and thus not useful for business cycle analysis. We show that causality tests are problematic when VAR variables are cross sectionally aggregated or proxy for non-observables. We provide an alternative testing procedure, illustrate its properties with a Monte Carlo exercise, and reexamine the properties of two prototypical VAR models.
Keywords: Aggregation; Granger causality; Nonfundamentalness; Small scale VARs
JEL Codes: C32; C5; E5
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Granger causality tests (C22) | misleading conclusions (G41) |
| aggregation issues (C43) | spurious nonfundamentalness (E19) |
| aggregated data (C80) | unreliable evidence in business cycle analysis (E32) |
| alternative procedure (Y20) | reliable identification of structural shocks (C22) |
| Monte Carlo simulations (C15) | good size and power properties (C52) |
| aggregated measures (E10) | incorrect inferences about structural shocks (C22) |
| lack of information in VAR model (C32) | prevents recovery of technology shocks (O33) |