In my field some papers are published with questionable econometric methodology.
The impulse responses are presented with 68% confidence bands only and the conclusion is that the effect is significant. I am wondering if there maybe is a statistical/econometric reason why the 68% interval can be used in the VAR context?
I don't know of a single other test which would accept a hypothesis not less than 0.9. Why is the VAR literature different?
68% roughly represents a confidence interval with half-length one standard deviation. I believe that the use of this in macroeconomics can be attributed to Christopher Sims and Tao Zha. Here is an excerpt from one of their papers:
Also, for characterizing likelihood shape, bands that correspond to 50% or 68% posterior probability are often more useful than 95% or 99% bands, and confidence intervals with such low coverage probabilities do not generally have posterior probabilities close to their coverage probabilities.
The paper "Unconventional Confidence Bands in the Literature on the Government Spending Multiplier" by Ryan Murphy (link here) discusses exactly this issue at length.
