A Monte Carlo test for the identifying assumptions of the Blanchard and Quah (1989) model

Abstract

In their VAR model, Blanchard and Quah (BQ, 1989) employed uncorrelatedness between Aggregate Supply (AS) and Aggregate Demand (AD) shocks and the long-run output neutrality condition as identifying assumptions. This article conducts a simple Monte Carlo experiment to gauge how well the BQ procedure can approximate the true structure if the underlying assumptions of uncorrelatedness and long-run output neutrality are not supported by data.

Keywords:

• Blanchard and Quah
• identifying assumptions
• Monte Carlo simulation

JEL Classification:

• C32
• C52
• E32

Acknowledgements

I thank the referee for the helpful input and the Department of Economics at the University of Washington for their generous hospitality. This work was supported by the National Research Foundation of Korea Grant funded by the Korean Government (NRF-2010-013-B00007).

Notes

¹ The structural shocks are generally permitted to correlate with each other in the simultaneous equation macroeconometric model (e.g. the Cowles Commission approach). The dynamic structure is restricted, instead, for identification, typically by excluding contemporaneous and lagged regressors. See Jacobs and Wallis (2005) for a comparison between this traditional approach and VAR modelling.

² We do not experimentally investigate the case in which the assumptions of uncorrelatedness and long-run output neutrality are relaxed simultaneously. As the two assumptions interact with each other, such joint relaxation would make it difficult to trace out which assumption is responsible for the rejection of the Blanchard and Quah (BQ) model.

³We experimented with combinations in which and have different values, but they were outperformed by the case of  =  in the fitting of the true impulse responses.

⁴ All cases fulfilled the stability condition.

⁵ We considered up to 20 horizons, as most impulse response estimates completed their empirical convergence to the long-run values by that time.

⁶ While the long-run response of to is restricted to zero by the assumption, the actual response turns out to be a nonzero value in the true model. One reason is that the correlation between shocks causes a change in to influence , which has long-run effects on . This would not occur in the BQ procedure, as the shocks are assumed to be uncorrelated.

⁷ In an interesting paper, Taylor (2004) presents another problem pertaining to models that identify structural shocks using long-run recursive restrictions. He proves that such models are not uniquely identified, because for an N-dimensional VAR model, there will be 2 ^N distinct parameterizations satisfying all of the restrictions, but each with a different set of implied impulse response functions. This result arises from the fact that some of the restrictions involve nonlinear equations, which can produce more than one solution. We conjecture that most empirical investigations drawing on the BQ-type identification scheme are likely to be subject to the critique raised by Taylor. However, this article tackles different issues and the Monte Carlo set-up in use is not suitable for addressing the issues brought up by Taylor. Thus, we do not pursue the matter further, but we acknowledge it as a possible concern.