Eviews5 / EViews5 / Example Files / x12 / example2

30 October 1997

Example 2. The specs in this example demonstrate two strategies for
dealing with calendar month heteroskedasticity (i.e.,
data from some months have more statistical variability
than data from other months):
(1) Use different seasonal filter lengths for different
months. (Decisions on individual month's filter
lengths are usually based on the moving seasonality
ratios in Table D 9.A.)
(2) Use calendarsigma=select option. To see if different
calendar months have different variances, set calendarsigma=all
in the X11 spec and look at the standard deviations at
the bottom of Table C 17. Months to be treated differently
for the purpose of extreme value identification can then
be listed as values of the sigmavec= option used in
conjunction with the calendarsigma=select option. (See
the X-12-ARIMA reference manual for details.)

It is valuable to look at sliding spans and history diagnostics.


Midwest Single-Family Housing Starts --

Sliding spans improve when we use different filter lengths for
different months. 3x5 seasonal filters are used for December
and 3x9s in the rest.

The calendarsigma=select option improves the revision histories
for January and February. You can see this particularly in
percent change and seasonal adjustment value graphs in X-12-Graph.
(Use the -g execution flag so files are available for X-12-GRAPH.)



# Example 2: mw1fam0.spc

# Adjustment of Single-Family Housing Starts from the Midwest Region of the US.
# 3x9 seasonal filters only.

series{
period=12
title='MIDWEST Single Family Housing Starts'
file='example2.dat'
name='MW1FAM'
format='2L'
savelog=peaks
span=(1982.1, )
}
transform{function=log}
arima{model=(0 1 1 )(0 1 1)}
estimate{ }
check{print=all savelog=lbq}
x11{
seasonalma=(s3x9 )
sigmalim=(1.8, 2.8)
savelog=(m7 m10 m11 q q2 fd8 msf)
}
slidingspans{
fixmdl=yes
savelog=percent
cutseas=4.5
cutchng=4.5
# Nondefault thresholds added 10-19-98
}
#history{
# estimates=(sadj sadjchng)
# savelog=(asa ach)
# start=1992.1
#}



# Example 2: mw1fam1.spc

# Adjustment of Single-Family Housing Starts from the Midwest Region of the US.
# Uses both 3x5 and 3x9 seasonal filters to account for the moving seasonality
# in December. Notice improvements in the sliding spans.

series{
period=12
comptype=add
title='MW Single Family Starts (with CSD seasonal filters)'
file='example2.dat'
name='MW1FAM'
format='2L'
savelog=peaks
span=(1982.1, )
}
transform{function=log}
arima{model=(0 1 1 )(0 1 1)}
estimate{ }
check{print=all savelog=lbq}
x11{
seasonalma=(s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x5)
sigmalim=(1.8 2.8)
savelog=(m7 m10 m11 q2)
}
slidingspans{
fixmdl=yes
savelog=percent
cutseas=4.5
cutchng=4.5
# Nondefault thresholds added 10-19-98
}
#history{
# estimates=(sadj sadjchng)
# savelog=(asa ach)
# start=1992.1
#}



# Example 2: mw1fam.spc

# Adjustment of Single-Family Housing Starts from the Midwest Region of the US.
# Uses both 3x5 and 3x9 seasonal filters to account for the large variability in
# the Winter months. The calendarsigma option reduces revisions, particularly
# in January and February.

series{
period=12
title='MW Single Family Starts (with CSD seasonal filters and calendar sigma)'
file='example2.dat'
name='MW1FAM'
format='2L'
savelog=peaks
span=(1982.1, )
}
transform{function=log}
arima{model=(0 1 1 )(0 1 1)}
estimate{ }
check{print=all savelog=lbq}
x11{
calendarsigma=select
sigmavec=(jan feb)
seasonalma=(s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x9 s3x5)
sigmalim=(1.8 2.8)
savelog=(m7 m10 m11 q2)
}
slidingspans{
fixmdl=yes
savelog=percent
cutseas=4.5
cutchng=4.5
# Nondefault thresholds added 10-19-98
}
#history{
# estimates=(sadj sadjchng)
# savelog=(asa ach)
# start=1992.1
#}