Set initial value for forecast in estimation

Estimation
1

Dear Professor Pfeifer,
I carefully read the manual, but I could not find how we could set an initial value for a forecast inside the estimation commend. I would like to set the last observation as my initial value for the forecast. In addition, I would like to know what is the interpretation of values in "oo_.MeanForecast.Mean ". My observations are deviation from the steady state, are my forecast also deviation from the steady state?
Kind regards,
Leo

2
  1. What matters are not the observations, but the values of the state variables. Their values is typically unknown. When you estimate a model, Dynare will conduct forecasts starting from the smoothed states derived from the Kalman filter. That seems to be what you want.
  2. If your model variables are linearized ones, then the forecasts will be also be deviations from steady state.
3

Thank you so much for the reply. My model is not linearized, but my observations and the variables I would like to forecast are in the form : x/x_bar-1. I was wondering, to compare the actual data and my forecast (For example for 6 periods), if I should simply compare "oo_.MeanForecast.Mean " and the actual data (6 period after my last observation)?
I just take the opportunity and raise my second question (that might be not relevant). Attached is a figure of Iacoviello (2015), I was wondering if you know where (in oo.) I can find the the blue line? iac.pdf (225.1 KB)

4
  1. Yes, you should be able to compare the data observations and the corresponding model forecasts at the respective horizon.
  2. Smoothed variables are in oo_.SmoothedVariables
5

Mr. Feifer, could you kindly elaborate on point n.1? If dynare do not take into account the observed data, how could it provide an efficient forecasting?

6

What matters for forecasts are the endogenous and exogenous state variables. There is typically not a straightforwards one to one mapping from observable data to the underlying state variables. Hence, you need a filtering algorithm that allows you to extract the value of the state variables from the actual observables.