why 68% bounds rather than 95% In Impulse response graphs, I've seen some papers report 68% confidence intervals rather than 95% bounds. I guess this makes the results look more significant. But other than that, is there any legitimate reasons to use these non standard bounds?
 A: There is a tradeoff between confidence level and power. If you are willing to accept a lower level of confidence (equivalent to raising the $\alpha$-level of a hypothesis test), then you can get additional power. For whatever reason, these researchers deem it more important to have high power than high confidence. Depending on how they value errors (maybe it’s much worse to miss an effect than to falsely identify an effect that isn’t really there), this can be defended.
(The cynic in me thinks that most readers are going to see the usual terminology about significance or confidence intervals and assume the typical levels, giving the authors the ability to claim a lot of results as significant without having the reader know that specificity has been traded in for sensitivity, a decision with which the reader might not agree. That is, cynic-Dave fears that the authors might be trying to pull a fast one.)
The particular number $68$ comes from $68\%$ of a normal density being within $\pm$ one standard deviation of the mean, rather than the usual two standard deviations to give $95\%$.
