Working Paper: CEPR ID: DP10765
Authors: David Meenagh; Patrick Minford; Michael R. Wickens; Yongdeng Xu
Abstract: Indirect Inference has been found to have much greater power than the Likelihood Ratio in small samples for testing DSGE models. We look at asymptotic and large sample properties of these tests to understand why this might be the case. We find that the power of the LR test is undermined when re-estimation of the error parameters is permitted; this offsets the effect of the falseness of structural parameters on the overall forecast error. Even when the two tests are done on a like-for-like basis Indirect Inference has more power because it uses the distribution restricted by the DSGE model being tested.
Keywords: DSGE model; error processes; indirect inference; likelihood ratio; structural parameters
JEL Codes: C12; C32; C52; E1
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Indirect Inference (C51) | Greater Power of Test (C12) |
| Likelihood Ratio Test (with reestimation) (C52) | Reduced Power of Test (C29) |
| Reestimation of Error Parameters (C51) | Obscured Effects of False Structural Parameters (C51) |
| Indirect Inference Test (C12) | Maintains Power (L94) |
| Indirect Inference Test (no reestimation) (C51) | Avoids Pitfalls of Likelihood Ratio Test (C52) |
| Indirect Inference Test (focus on specific model features) (C52) | Nuanced Understanding of Model Validity (C52) |
| Likelihood Ratio Test (under model falseness) (C52) | Diminished Power (D74) |