I am attempting to estimate a regression using the Mixed Data Sampling (MIDAS) model, as found in Kuzin, Marcellino and Schumacher (2009). Basically, the MIDAS model is a direct forecasting tool that makes use of data of different frequencies.
The problem I am facing is that I have two variables of different frequencies: a quarterly variable, y_tq, and a monthly variable, x_tm, and I want to construct a MIDAS model that uses the monthly variable, x_tm, to generate h-step ahead forecasts of the quarterly variable, y_tq. To be able to this, first I express the quarterly variable, y_tq, in monthly terms, y_tm, so that y_tq = y_tm for all tm = ..., Tm-6, Tm-3, Tm. (Where Tm is the last month at which y is available). So now my two monthly series look like:
.
.
.
Where M101 is the monthly indicator x_tm, and ldgdp is the quarterly variable expressed in monthly terms, y_tm.
Using these two series, the MIDAS model for forecast horizon h_m will have the form:
where,
The superscript on the x means that every third observation starting from the t_mth observation is included in the regressor. So,
Lags of the monthly indicator are also skip sampled accordingly. So,
To estimate the MIDAS I'm thinking of typing the equation as it appears in its above form into eviews, and estimating with non-linear least squares. What I am have trouble with is (1) how to type in the summation operator into eviews, and (2) how to skip-sample the variable x so that only every third observation is included in the regressor x and its lags.
Sorry for the long post! Any help on how to tackle these two problems will be much appreciated.
Thank you.