Menu
▾
▴
#109 Another X-12 Issue
In attempting to seasonally adjust labor force data from the Nassau-Suffolk metro area, I'm getting an output that does not look at all like it is seasonally adjusted. There is no error message, and data comes out. However, the original and "adjusted" series look almost identical.
The gretl file is attached.
gretl 1.8.1 on Windows XP
There's no bug here. The variation in your series is not seasonal.
Try a regression with a time trend, a dummy for around 1995, and
an interaction between dummy and trend (all this to capture the
initial fall and subsequent trend rise). Now add monthly dummies
and test them for significance: they're not in the least significant.
If you look at the spectrum of your series, detrended, you'll see the
peaks are at about 13 months, 6.5 months and 4.3 months. It's not
an annual pattern.
OK, my last comment. I suspect a bug in your data source. You have
what looks very much like a seasonal cycle (overlaid on a broken
trend), yet the cycle is not seasonal according to all the statistics:
the periodicity is wrong. Is it possible that that data source in fact
has thirteen values per year, one an annual average?
OK, I lied about that being my last comment. I have confirmed my
suspicion. Every 13th observation is an annual average. You have
to pay attention to this sort of thing! To fix this in gretl:
genr time
genr month = time%13
smpl month!=0 --restrict
setobs 12 1990:01 --time-series
Then save the corrected data.