Open StaffanBetner opened 9 months ago
I think this is limitation in the x13 binary that is not as well documented as it could be. The earliest year for which this works for me is
seas(window(births_sweden, start=1942)
Issue #309, filed only weeks ago, touched the same topic and refers to the reference manual stating that 780 is the maximum number of observations. (And oddly length(window(births_sweden, start=1942))
is higher than that but that reference table 2.2 on program limits has some extra language.)
PS Same manual in its current version
Thanks! I looked into the source code and traced it to srslen.prm. pobs = pyr1*psp, pyr1 = maximum number of years in the series (set to 65), psp = maximum length of seasonal period (set to 12). The maximum number of years seems to me like a fairly arbitrary number.
It doesn't fully work to change. The maximum value (of years in the series) that I can compile x13binary with is 136, but the highest that actually works to run is 131. The longest span that actually works is seas(window(births_sweden, start=c(1895,5)))
, which is 1542 values, compared to 131*12 = 1572. I think I am satisfied for today with bending the capabilities of X-13ARIMA-SEATS.
Reprex:
Created on 2023-12-26 with reprex v2.0.2