Introducing Nowcast as an observable


#1

Dear all,

Referring to the paper of Del Negro and Eusepi (Fitting Observed inflation expectations) but in a different context, I am facing an issue similar to the one reported here http://www.dynare.org/phpBB3/viewtopic.php?f=1&t=3830 in the context of a pseudo-real time forecasting exercise, but with different timing in the observed variables.

Del Negro and Eusebi introduce expectations of a variable at t+1 made at time t. I want use ‘’nowcasts’’ (expectations made during t for t) rather than ‘’expectations’’ (expectations for t+1 made at t) as an observable. This is because ‘’nowcasts’’ are available before the true data is released (since usually the data release occurs with a time lag), and I hope they improve the forecasting of the model.

How can I plug this information content into the model?

Thank you in advance


#2

Dear Fabio,

If you need the expectation of y_t given the information available at time t-1, you can use the EXPECTATION operator (see section 4.3.2 in the reference manual).

Best,
Stéphane.


#3

In that case, you need to think hard about the information structure of your model, i.e. what the information set is that agents and the econometrician have. Do that agents know y_t, but we as the econometricians do not observe it? Or are the agents in your model also not aware about the true states of the economy?
If it’s only about the former, then the nowcasts are simply that actual realization of y_t plus a measurement error.


#4

Dear Johannes,

Thanks for your prompt response.

Let us put it this way, the context is the one of a pseudo real time forecasting exercise. Where, as data are released with some time lag, the econometrician just observes today’s nowcast (that is a guess made today about today’s actual data realization still to be observed) .

Let’s assume today is t+1, real data are available until t, and the econometrician observes also t+1 nowcast. In a sense, abstracting from nowcast information, it would be like estimating the model in t, since agents knowledge is until t. (agents just don’t care about where the econometrician is located on the time span. ) Since the econometrician has some informative advantage (that is nowcast) , I would like to exploit it just by assuming that nowcast are the agents observed expectations, and thus using those expectations as observables. This might be time inconsistent from the agents perspective, but it is not from the econometrician point of view. After all the aim of this exercise is just forecasting, not theory.

What do you think? Does it make sense, or there is maybe some other better way to exploit this nowcast info?