Hi,

I estimate a DSGE model using bayesian statistic and asked for forecasting the endogenous variables. The variables are defined in level, but some of them display negative forecasted paths - like taxes, governmental expenditures, and even the GNP. What could be the reason to this?

Best regards, Márcio.

There could be a lot of reasons: forecasts being in deviations from the mean, unhandled trends and constants, etc. Without the files, it is hard to tell.

Attached to the message are the script, data and some figures with convergence diagnóstics and forecasting. The data were detrended, using hp filter. Thanks for your help.

nk_stn2_3.mod (27.7 KB)

Attached to the message are the script, data and some figures with convergence diagnóstics and forecasting. The data were detrended, using hp filter. Thanks for your help.

nk_stn2_3.mod (27.7 KB)

data_and_figures.rar (748 KB)

Your data is not mean 0 and therefore surely not HP-filtered. In the model for the starting values Y has mean 2.3, while the data mean is 1.0005. The estimation cannot properly handle this (only by pushing to a unit root). You can see this in the

graphs as well

So, even if the model isn’t log-linearized, the data needs to have mean 0? But in this case, some of the observations will show negative values.

That is not what I am saying. But if you decide to leave the constant in the model and the data, they must be consistent. You cannot have data that has mean 100 and tell estimation that this data corresponds to a variable that has mean 1. Regarding the “negative” values: demeaned data have the interpretation as fluctuations around the mean/trend. Negative values then correspond to the data being below the mean and are by construction desired.

In a nutshell, your observation equation is wrong. See Pfeifer(2013): “A Guide to Specifying Observation Equations for the Estimation of DSGE Models” sites.google.com/site/pfeiferecon/Pfeifer_2013_Observation_Equations.pdf