MIDAS RMSE

[miorinnovo]

Usually when I want RMSEs I code the program to take the first forecast and copy it to another series, then make the second forecast and copy to that same series and on and on so that I have a series of forecasts that I can use to calculate the RMSEs.

My quesitons is about MIDAS programs. How do I get the RMSE when I will have multiple forecasts from monthly data for a given quarter>

Thanks

[EViews Gareth]

Don't understand this question.

MIDAS produces a single forecast.

[miorinnovo]

Let me try to better explain my question.

In a non-mixed frequency situation its simple. Say I'm trying to explain y, a quarterly variable, with x, another quarterly variable. I get one forecast for every observation of y.

Now Midas uses mixed frequencies. So say I'm trying to explain y, a quarterly variable, with z, a monthly variable. I get one multiple forecasts for every observation of y. Since there are 3 months in a quarter I could get at least 3 forecasts for the same observation of y.

So I'm not sure the bets way to go able getting RMSEs for MIDAS forecasts.

[EViews Gareth]

No, with MIDAS you get one forecast for every observation of Y.

[miorinnovo]

You get one forecast for each observation of Y at any one point in time, but you get more than one forecast for Y in real time.

If I have January data I can forecast Q1, but then when I get February data I can do another forecast of Q1, same for March. So in a pseudo out of sample forecasting exercise you want to capture the RMSE at each point in the cycle.

This is what I'm trying to figure out.

[EViews Gareth]

I see what you're saying. But when you use the built in forecast from a MIDAS equation, it doesn't do the re-specification, estimation and forecast for the new data, it uses whatever the original specification was.

You'll need to roll the estimation, specification and forecast in a loop. I believe one of our guest blog posts does something similar:
http://blog.eviews.com/2018/12/nowcasti ... basis.html