How to Test Block Exogeneity by the Marginal Likelihood
Guide 2 of 3 in the series VARs for a Small Open Economy
Based on Bayesian Macroeconometrics (Chan, 2027)
[ Full guide on GitHub ]
How to Estimate a VAR for a Small Open Economy treats Australia as the small open economy and the US as the foreign economy. Its VAR has a foreign block of US variables and a domestic block of Australian variables. That guide imposes block exogeneity: the lags of the Australian variables are excluded from the US equations. To assess the restriction, we compare that model with a VAR in which the Australian lags enter the US equations. The Bayes factor in favor of the restriction is the ratio of the restricted model's marginal likelihood to the unrestricted model's.
We compute this ratio in closed form. The prior hyperparameters are the same in both models. The Australian block is then identical in the two models, so its marginal likelihood cancels in the ratio. Only the two marginal likelihoods of the US block are needed. For the simulated sample of the estimation guide, the log Bayes factor is 17.66 in favor of block exogeneity.
The Bayes factor depends on the prior variances of the Australian lags. In that sample, the log Bayes factor is 1.41 when these variances are a hundredth of those in the estimation guide. It is 160.02 when they are a thousand times as large. With large enough prior variances, the test favors block exogeneity in any sample. This is the Bartlett-Lindley paradox (Lindley, 1957; Bartlett, 1957). The prior variances should therefore be defensible and not too extreme.
The steps
- Write the Bayes Factor with the US Block
- Compute the Two Marginal Likelihoods
- Vary the Prior Variances of the Australian Lags
Try it
In MATLAB, from the root of a copy of bvar-toolkit, run the script that prints every number of the guide:
run guides/soe_exogeneity_test/guide.m
In this series
Previous: How to Estimate a VAR for a Small Open Economy
Next: How to Trace a Foreign Shock through a Small Open Economy