Hi Rumen,

yes, that is absolutely fine, and common practice actually. Researchers often use the log of an variable to account for non-linear relationships (log-linearizeation), and it's usually a good idea, unless you explicitly want to show a quadratic relationship or sth. You just have to be carefull with interpreting the parameter estimates. The usual interpretation of percentage change of y following from a 1% change in x doesn't work anymore. If you need this, you can calculate an estimator for this from your parameter estimates, since they give you the percentage change of y given a 1% change of your new indicator 1+inflation. But if you are just interested in the sign and significance of the parameter you can just use the paramter estimates as is.

Regards,

Tim

Edit: It seems you have significant outliers, you might consider dealing with those by excluding them or including an outlier dummy.

Thanks Tim,

In my model inflation is only one of 5 independent variables and it isn't significant, but when testing for violations it comes up that it has influential data entries, so I had to find out how to correct this issue. I tried it and it worked out pretty well.

It is better to construct the inflation series from CPI series as follows:

Inflation = 100 x ln (CPI_t / CPI_t-1).