Converting monthly errors to quarterly

[raphael]

Dear Tom,

In my VAR model, I first interpolate GDP into monthly figures, but to chart the results, I'd like to convert it back to quarterly data. Two problems then: CALENDAR has trouble working in monthly and quarterly mode, and most important, how can I compute quarterly standard errors (so that I can make a fan chart of confidence bands).

Thanks a lot

[TomDoan]

The following will convert 20 quarters of monthly values into their quarterly averages:

Code: Select all

set xq 1 20 = xm(3*t-2)+xm(3*t-1)+xm(3*t)

Note that these will go into consecutive entries so, until you change the CALENDAR, they will look like 20 months.

Converting impulse responses is easier than other types of data, since they don't have any actual dates associated with their entries; they are just 1,...,H exactly as is assumed in this little code snippet.

[raphael]

Thanks, I would have thought it would be more complicated than this ? Aren't errors non-linear ?

[TomDoan]

That depends. What I was talking about was applying shocks at the monthly level and aggregating the results to quarterly. If you're talking about estimating the model with monthly data then applying shocks that have been aggregated to quarterly shocks, yes, that would probably be quite non-linear.