# Unemployment forecasting model specification

I am building a model to forecast unemployment using GDP, the CPI index and the industrial production index (INDPRO) as covariates. Since I am looking to use stationary time series, I gathered the percentage changes of each of these quantities such that the model now looks something like this:

%Unemployment = f(%GDP, %CPI, %INDPRO)

Is this an econometrically-sound model specification? Or should I use other data transformations since CPI and INDPRO are already expressed in percentages?

• To see the $%$ change in unemployment due to a $%$ chang in GDP or some other covariate one usually uses a log-log specification.
– tdm
Nov 11, 2021 at 6:42
• @tdm I see, thanks! Should other covariates (like for instance the federal funds rate) be included as logs as well or are they usually kept in their original format? Nov 11, 2021 at 10:20
• These are already in percentages so I guess they should be kept in levels (but I'm not a macro-economist).
– tdm
Nov 12, 2021 at 8:44