Version 0.2.10 (July 26, 2002)
This release introduces a few new features:
(a) Two new runtime options to generate reduced output (-r) and
allow the program to run in "quiet mode" (-q), suppressing
warning messages so that they do not appear on the screen
during runtime.
(b) A new argument (divpower) was added to the series
spec to allow the user to rescale the original series
by a power of ten.
(c) Two new input data formats have been added to the program which
mirror those used by X-11-ARIMA/2000. They are similar to the
"X-11" style formats used in previous version of X-11 and
X-11-ARIMA, but utilize 4 digit rather than 2 digit years.
Other program improvements include:
(a) Many subroutines were restructured to make the source code more
compatible with other Fortran compilers - the use of labelled END DO
statements was eliminated, and double quote marks were removed from
a few FORMAT statements.
(b) Some changes were made to the program output, such as revising
error messages to make them more helpful to the user and adding
the spectrum of the regARIMA residuals to the default printout.
(c) More tables are saved to the graphics directory when X-12-ARIMA
runs in graphics mode; also, more information is stored in the
diagnostics files stored in the graphics directory.
Several bugs have been fixed in this version of the program, including:
- initializing internal variables correctly,
- refining how errors are handled automatic model identification
procedure,
- checking the data beyond the user-specified model span before
applying data transformations;
- checksing for fatal errors before printing out diagnostics,
- checking if the henderson trend filter being applied is too
long for the series,
- checking if indirect seasonal factors can be computed within
the sliding spans procedure,
- handling the processing of outliers correctly within the revisions
history routines, and
- ensuring that prior trading day factors are not generated and
applied a second time if X-11 holiday adjustment is done.
Version 0.2.9 (January 22, 2002)
This release introduces several new features:
(a) The procedure for generating the default critical value
for the outlier identification procedure has been changed
to give more plausible values for smaller outlier spans.
(b) A new argument (reweight) was added to the x11regression
spec to reweight daily weights from the irregular
regression so that there are no negative daily weights.
(c) The program will now produce chi-squared statistics for
groups of related regressors (such as change of regime
variables or calendar regressors), as well as chi-squared
statistics for different types of user-defined regressors.
Other program improvements include:
(a) The program now uses the number of observations tested for
outlier (set by the span argument of the outlier spec) to
generate the default outlier critical value rather than
the model span (set by the modelspan argument of the
series spec).
(b) Additional subroutines were restructured to make the
source code compatible with the GNU Fortran compiler.
(c) Parts of the program output have been rewritten to make
them more helpful to the user (including several error
messages and text associated with change of regime
regressors).
(d) Revisions were made to the code to ensure that prior trading
day adjustments and the D 8.B table were generated correctly
for log additive seasonal adjustments.
Several bugs have been fixed in this version of the program,
including errors in assigning initial values to the regressors,
computing indirect seasonal factors, producing the final t-test
table for the outlier identification, starting dates for the
sliding spans analysis, and including AO outliers in the
printed D13 table when TC outliers were also present.
Version 0.2.8 (January 19, 2001)
This release introduces several new features:
(a) Additional seasonal adjustment diagnostics are saved when the -s
option is used. Spectral peak diagnostics are now saved,
as well as values from Tables D 9.A (Moving seasonality ratio)
and D 8.B (Final unmodified SI ratios, with labels for outliers
and extreme values) and hinge values from the revisions history
analysis.
(b) A new model diagnostics file (with the file extension .mdg)
is generated by the -s optionl. This file contains
such diagnostics as the regARIMA coefficient estimates,
likelihood statistics, outlier critical values, residual
normality test statistics, and chi square test statistics
for groups of regressors.
(c) The interaction between the irregular regression estimation and
the history and sliding spans statistics has been improved.
New arguments for the history and slidingspans specs control
whether the irregular regression coefficients are re-estimated
(fixx11reg) or whether AO outliers for the irregular
regression (x11outlier) are re-identified for each span of
data analyzed.
(d) The Hannan-Quinn statistic can now be saved to the log file via the
savelog argument of the estimate spec.
Other program improvements include:
(a) When the program will automatically switch to using
fixed model coefficients for the history analysis has now
been standardized. This happens when the start of the
revisions history analysis causes some truncated data
span to have fewer than sixty observations for regARIMA
model estimation. Irregular regression models from the
x11regression spec are treated similarly.
(b) Several subroutines were restructured to make the source code
compatible with the GNU Fortran compiler.
(c) The program will no longer indicate trading day frequencies for
spectra derived from series with less than 60 observations.
(d) Parts of the program output (including several error messages)
have been rewritten to make them more helpful to the user.
Also, larger numbers can now be displayed in the main output
tables.
Several bugs have been fixed in this version of the program,
including:
(a) an error in computing the mean of the Easter effect for the
x11easter adjustment (x11 spec) has been fixed;
(b) an error in specifying the 3-term moving average;
(c) a problem with the filtering done during the generation
of regARIMA forecasts;
(d) restructuring parts of the irregular regression estimation
procedure to ensure regARIMA model estimation would not
effect the irregular regression;
(e) ensured that length-of-month factors were only generated
in situations where the log transformation and/or
multiplicative seasonal adjustment is specified or implied;
(f) minor errors in the outlier, irregular regression, seasonal
factor end weight, and other routines.
Version 0.2.7 (May 19, 2000)
The most significant change from the previous release of X-12-ARIMA
concerns how forecasts are generated when the modelspan argument is
specified in the series spec and has a starting date later than the
starting date specified in the span argument (or the starting date
of the original series if the span argument is not used). In
previous versions of X-12-ARIMA, the program would apply the estimated
model to the entire span of data to generate the innovations needed
to derive forecasts specified by the user. Now the forecasts are
calculated exclusively from the data beginning at the starting date
of the model span. This is both theoretically better and, as we have
found, also better in practice. As a consequence of this change,
backcasts will no longer be generated when a modelspan argument with
a starting date after the start of the data span is specified.
Also in this release, the D 8.B table which highlights the effect of
regARIMA outliers and extreme values on the unmodified SI ratios can
now be saved in a separate file via the x11 spec (save=unmodsiox or
save=d8b). In future releases, we may replace table D 8 with table
D 8.B.
A few additional bugs have been fixed with this version of the program,
including a minor correction in the factor mean computation for the
x11easter option, overwriting certain types of saved output,
ensuring forecasts are saved with various types of factors when
appendfcst = yes.
Version 0.2.6 (February 11, 2000)
This release introduces three new features:
(a) A new table can be specified in the x11 spec which highlights the
effect of regARIMA outliers and extreme values on the unmodified
SI ratios. A character label is given next to each modified SI
ratio that specifies if the observation has been identified as
an extreme value or regARIMA outlier.
(b) A table containing the final t-tests for the regARIMA outlier
identification procedure has been added to the outlier spec.
(c) A true7term argument was added to the x11 spec to allow users to
specify that the asymmetric ends weights for the 7 term
Henderson filter are applied for observations at the end of the
series where a central Henderson filter cannot be applied.
By default, central and asymmetric weights from a 5 term Henderson
filter are applied.
Several bugs have been fixed in this version of the program,
including:
(a) ensured factors generated by user-specified prior trading day
weights (tdprior in the x11regression spec) are removed from
the seasonally adjusted series;
(b) corrected how errors in the history and sliding spans analysis
are handled by the composite adjustment procdure;
(c) changed how the outlier procedure keeps track of what outliers have
been tested, to provide more accurate listing for the table of
outliers that are close to the cutoff value;
(d) corrected errors in saving regARIMA model information.
Minor errors in the outlier identification (printed incorrect ending
date for outlier identification when AO outliers were used), forecasting
(moved code to close save files for transformed forecasts), and double
precision equality. Also changed how several character variables were
initialized and referenced to assist other developers with compatibility
issues for their Fortran compiler.
Other changes to the program include:
(a) The long-term means for the Easter regression variables used in the
regression and x11regression specs have been changed to correspond
to the first 400 year period of the Gregorian calendar, 1583-1982.
This provides a close approximation to the average calculated
over the much longer period of a complete cycle of the dates of
Easter. This will cause some change in the Easter regression
factors generated by the program, but practically no change to the
combined adjustment factors generated by the program.
(b) If a negative trend value is detected by the program for a
multiplicative seasonal adjustment, the program will now replace this
value with either the average of the nearest positive values before
and after the observation, or with the nearest positive value.
A warning message is produced, and the observations replaces are
noted in the tabular output.
(c) The version number of X-12-ARIMA is now displayed on the screen at
runtime.
(d) X-12-ARIMA no longer has a default specfile name (x12a.spc) that is
used when no arguments are specified by the user at runtime. If no
arguments are specified when running X-12-ARIMA, the program will
stop executing, with an error message directing the user to supply
either an input specification file or a metafile.
(e) The default of the eastermeans argument of the x11regression spec
was changed; the long-term Easter means are now removed from the
Easter regressor in the x11regression spec rather than the monthly
means calculated from the span of data used to calculate the Easter
regressor coefficient.
Version 0.2.5 (October 1, 1999)
This release introduces two new features:
(a) Sliding spans and revisions history analysis can now be done in the
same X-12-ARIMA run.
(b) A centeruser argument was added to the regression and x11regression
specs to allow users to remove a sample mean or seasonal means from
user-defined regression effects.
Other upgrades to the program include:
(a) The yr2000 argument was added to the composite spec.
(b) The regression coefficients estimated during the identify spec are
no longer printed out by default. Users who wish to see this output
should specify the regcoefficients table in the print argument of the
identify spec.
(c) The fixmissing argument has been removed from the history and
slidingspans specs. Missing values will be fixed during the
history and sliding spans analysis.
(d) The printby argument has been removed from the history spec.
The output of the history analysis will print out the revisions
for each year.
(e) The output generated by savelog=indtest in the composite spec has
been modified.
Several bugs have been fixed with this version of the program,
including correcting the spectrum of the seasonally adjusted series,
ensuring the correct slidingspans or history analysis is produced when
mode=trend is selected in the x11 spec, simplifing how model spans
interact with the history spec, correcting an error in the routines
that tests for temporary level shifts, correcting how the outlier
options are passed to the revisions history routines, ensuring that
length of month factors were used in sliding spans and history analysis.
Version 0.2.4 (June 4, 1999)
This release introduces two new features:
(a) A spectrumaxis argument was added to the x11 spec to allow users to
choose how the spectrum for the modified seasonally adjusted series
is scaled. The default scaling for this spectrum has been changed so
that it does not match the scaling of the spectrum of the original
series.
(b) A new table containing the final calendar factors has been added to the
seasonal adjustment output. This table can be controlled from the x11
spec.
Other upgrades to the program include:
(a) simplifying the computations for the x11regression spec when trading day
and Easter regressors are used together;
(b) using the combined calendar factors in the summary diagnostics table
(Table F2) rather than the trading day factors alone;
(c) produced forecasts from the end of the series by default, rather than
at the end of the model span provided in the series spec.
Several bugs have been fixed with this version of the program,
including faulty intialization in the routine that fixes regression
variables, ensuring that forecasts adjusted on the original scale
contained various holiday and prior adjustments, ensuring that the
proper residuals were used in the automatic outlier identification
procedure, and fixing problems with the irregular regression procedure.
An update to this release was made July 1, 1999. This was done
to fix minor bugs in the forecasting, prior adjustment, and automatic
transformation selection.
Version 0.2.3 (March 19, 1999)
This release introduces two new features:
(a) A trimzero argument was added to the series and transform specs to
allow users to choose if zeroes at the beginning or end of a time
series read in via the file argument are treated as series values.
A similar argument (umtrimzero) was added to the x11regression spec.
(b) A numspan argument was added to the slidingspans spec to allow users
to specify the number of spans to use in the sliding spans analysis.
Other upgrades to the program include:
(a) improving the savelog output for AIC tests performed in the regression
and x11regression specs;
(b) Easter means generated for the regression procedure are now computed
only for complete years when eastermeans=no;
(c) implemented a different algorithm for computing the periodogram when
spectrumtype=periodogram in the series and composite spec; and
(d) several routines were revised to improve the structure and
performance of the code.
Several bugs have been fixed with this version of the program, including
conflicts between the automatic transformation selection procedure and the
sliding spans analysis, ensuring that spectral peaks for spectrums calculated
for indirect adjustment are stored into the log file with the savelog option,
an error in saving the B11 and C11 tables, problems with the critical argument
in x11regression, improper labels for histograms in the sliding spans analysis,
and fixing the values of regARIMA regressors.
Version 0.2.2 (December 11, 1998)
This is primarily a bug-fix release, to fix problems with the
application of regARIMA factors, specifying starting values for
user-defined regressors, the application of user defined regressors
within the irregular regression estimation procedure, seasonal
factor generation with the calendarsigma option of the x11 spec,
and the application of the aicdiff argument.
There are five major changes in program options:
(a) The aicdiff argument in the transform spec was changed to
conform with the aicdiff arguments in the regression and
x11regression spec. The default is now -2, and values less
than zero bias the selection in favor of the log
transformation. The documentation has been changed to
reflect this change.
(b) The F-test for stable seasonality performed after the B1
table has been changed to conform with that of
X-11-ARIMA/88.
(c) The augmentusertd argument has been removed from the
regression spec. Now, any length-of-month, length-of-quarter,
or leap year regressor or prior adjustment factor is assumed
to be part of the trading day effect.
(d) Table and plots of the sample partial autocorrelation function
(PACF) of the regARIMA residuals can be produced by the
check spec. The PACF tables and plots are not part of the
default output.
(e) The frequency at 0.304 will no longer be used for testing
monthly series for residual trading day effects.
Version 0.2.1 (October 16, 1998)
This is primarily a bug-fix release, to fix problems with title
generation in the composite analysis and ensure that the proper
output is stored via the savelog argument.
There are two changes in the savelog option:
(a) A new option for savelog is included in the check spec.
Setting savelog=ljungboxq (or savelog=lbq) will generate a
printout of lags from the residual ACF with significant
Ljung-Box Q statistics.
(b) The seven options for storing the percentage of observations
flagged as unstable in the sliding spans has been collapsed to
one. Setting savelog=percent (or savelog=pct) in the
slidingspans spec will store the pecentage of observations
flagged for all estimates tested.
Version 0.2 (October 2, 1998)
Version 0.2 of X-12-ARIMA introduces the following new features:
(a) Users can now set starting values for regression coefficients
in the regression and x11regression specs via the b argument.
Users can also fix the value of specified regression
coefficients using the same argument.
(b) The fixreg argument in the sliding spans and history specs
allows users to fix selected types of regressors specified
in the regression and x11regression specs (trading day,
holiday, outlier and user-defined regressors) during
sliding spans and revisions history analysis.
(c) The default outlier identification threshold now depends
on the length of the series; the longer the series, the
higher the default of the critval argument becomes.
(d) A "default" model choice is now available for the
automatic model identification procedure. When no models
selected, this model will be used for testing for and
removing regression effects before the time series is
seasonally adjusted. Forecast extension will not be
performed for the "default" model. This default model is
specified in the automatic model file (x12a.mdl by default)
by placing a "*" instead of an "X" at the end of the record.
(e) The AIC corrected for the length of the series (AICC) will
be used in the AIC tests performed in the transform,
regression, and x11regression specs. This was done to
improve preformance of these tests for short series.
(f) An aicdiff argument is included in the transform, regression,
and x11regression specs to allow users to set an acceptance
threshold for AIC tests performed in these specs.
(g) New modeling diagnostics have been incorporated into this
version of X-12-ARIMA, including normality tests of the
regARIMA model residuals, an ACF plot of the squared
residuals, and a spectral plot of the regARIMA model
residuals.
(h) Additional diagnostics are available for storage via the
savelog argument. Spectral peaks, the residual normality test
results, AICC, and the results of the AIC tests can now all
be stored in the log file.
(i) The maximum t-value of the outlier identification procedure
is printed out if no outliers are found.
(j) A table containing the correlation matrix for the regression
coefficients of the irregular component regression has
been added to the x11regression spec.
(l) A print1stpass argument was added to the x11 spec to cut down
on the amount of tables printed out when the irregular regression
and X-11 holiday procedures are used. Sometimes, a first
pass is performed to estimate an irregular regression or
X-11 holiday procedure; the additional output from this
first pass will be suppressed unless print1stpass=yes in the
x11 spec.
In addition to these new features, there are other changes to
existing features:
(a) The residual PACF table and plot have been removed from the
check spec.
(b) The revisions history analysis no longer removes level shift
outliers by default. Users can fix outlier regressors
throughout the history analysis using the fixreg argument
mentioned above.
(c) The default value for the rate of decay of temporary
change (TC) outliers has changed from 0.7 to 0.7*(period/12),
where period is the periodicity of the time series being
modeled (12 for monthly series, 4 for quarterly series).
This allows the same amount of decay over a calendar year
for monthly as well quarterly time series
(d) The default starting dates for the revisons history analysis
and spectral plots were changed for quarterly series to allow
more observations to be used for these procedures.
(e) Moved the fix and file arguments (used to read in a saved
model to X-12-ARIMA) from the arima spec to the estimate
spec.
(f) Users can now save the concurrent forecasts and standard
errors from the forecast revisions history of the history
spec, as well as the sample ACF and PACF values for
different orders of differencing in the identify spec.
(g) The trendadjma argument has been removed from the regression
spec, along with the corresponding tables.
Version 0.1.4 (June 18, 1998)
This version of the program contains no new features from the last
release, but some bugs were found and fixed in the irregular regression
trading day and holiday adjustment procedure.
Version 0.1.3 (June 4, 1998)
This version of the program contains no new features from the last
release, but some bugs were found and fixed in the irregular
regression trading day procedure, the saving of the sample
autocorrelation and partial autocorrelations, and the saving of
the roots of the ARMA polynomial.
Version 0.1.2 (May 1, 1998)
This version of the program contains no new features from the last
release, but some bugs were found and fixed in the automatic AIC
tests for holiday regressors and the printing and calculations of the
revisions history analysis for seasonally adjusted estimates.
The source code was also restructured to remove ENTRY statements.
Version 0.1.1 (March 25, 1998)
This version of the program contains no new features from the last release, but
several bugs were found and fixed in the computation of the final combined
holiday factors, the automatic AIC tests for trading day and holiday,
problems dealing with zeros at the end of data files, prior and outlier
adjustments to the final seasonally adjusted series, and other procedures.
An update to this version was made available on April 9, 1998; it
contained no new features from the last release, but a few bugs were
found and fixed.
Version 0.1 (February 10, 1998)
There are several new features and changes from the Beta version of X-12-ARIMA,
including:
(a) an OLS regression procedure for the irregular component which allows
users to simultaneously estimate effects for trading day, Easter,
additive outlier, and user-defined regressors;
(b) automatic identification procedures for user-defined regressors,
data transformations (multiplicative versus additive adjustment),
and temporary change outliers;
(c) implementation of three types of change of regime regressors;
(d) more extensive revisions history options for seasonal adjustments and
trends;
(e) a new option to estimate the trend of a non-seasonal series;
(f) an option which stores files for use with a recently developed SAS/GRAPH
module, X-12-Graph, which generates graphical output from X-12-ARIMA runs.
A document which gives a summary of the new features contained in the Final
version of X-12-ARIMA and other changes from the Beta version of X-12-ARIMA
(Beta2Fin.ps for PostScript, Beta2Fin.pdf for Adobe PDF) can be found in the
directories listed above.
An input file conversion utility that converts Beta X-12-ARIMA spec files into
spec files that can be used with Version 0.1 of Final X-12-ARIMA is
provided with the Icon tools.