When there are missing values in a time series, there are two ways this can be handled:
A. Maintain a complete calendar (consecutive dates relative to the frequency of the
series) and insert “NA” (or some symbol defined as indicating “missing”).
B. Record only non-missing values, or in other words just skip dates for which there’s
no valid observation.
It’s my impression that dbnomics has mostly followed method A, but I recently came
across ECB/BSI/M.U2.N.R.LRE.X.1.A1.3000.Z01.E which uses method B. To be precise,
it uses method B in JSON output, though if you download CSV the calendar is complete
with NAs inserted for 21 missing monthly observations.
With gretl’s JSON-oriented dbnomics reader in mind, I’d really like to know if method B
is considered legitimate by dbnomics, or if it’s an anomaly that should be fixed. Gretl could
in principle handle JSON with skipped observations, but it’s more complicated than method
A. Moreover, method A – with its explicit NAs – gives the user greater confidence that
the data are actually correct, and not broken in some obscure way.