I came across a few articles here and there that conclude:

When the data (say variables X, Y) for an impulse response function are on log level, the y-axis depicts the % response of Y to a 1% shock by X.

Broadly speaking, I think the shocks in an IRF are one SD Choleski shocks, where the y-axis reflects responses in the units of the dependent variable.

However, can someone explain (mathematically or not) why using logs automatically implies the interpretation of % impulses and % shocks? And: Why do some source say that log value responses are "approximately equal to" the percent change?

Thank you!


If you logarithmize the dependent variable, you will get approximately % changes, see Why is it that natural log changes are percentage changes? What is about logs that makes this so?

We know that the vcv matrix $\Omega$ of the error term can be factorized into $\Omega = ADA'$ with $D$ being a diagonal matrix (and $D$ is the vcv matrix of the "structual innovations"). Alternatively, we can factorize $\Omega$ by the Choleski decomposition into $\Omega = PP'$ where $P$ is lower triangular. So, we have $\Omega = ADA' = PP'$ and moreover, $$ ADA' = \underbrace{AD^{0.5}}_{=P}\underbrace{(D^{0.5})'A'}_{=P'} $$ so $P$ has the standard deviations of the structual innovations on its principal diagonal. Without loss of generalization, one can set the standard deviations of the structural innovations to 1 (i.e. $D = I$). Triangular factorization and Choleski Decomposition are then equal. That is, the interpretation of the Choleski Decomposition changes to: "If $y_{jt}$ is increased by "standard deviation of the $j$th structural innovation" units (but we just set the standard deviation to one and the units are given by %, see the link above), $y_{t+s}$ changes by $\Psi_sp_j$ units (but again, since we use logs the units are %)" ($p_j$ denotes the $j$th column of $P$)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.