U.S. flag

An official website of the United States government

Skip Header

Notes on X-13ARIMA-SEATS Version 1.1, Build 57

Build 57 of Version 1.1

The major change for this release (Version 1.1 Build 57) is that the Windows version of X-13ARIMA-SEATS binary has been built for the 64-bit architecture. The 32-bit version has been discontinued. 

New options

  • Add robustsa argument to spectrum spec, to set whether or not spectrum generated for robust versions of series.
  • Generate nonparametric diagnostic for the seasonally adjusted series in spectrum spec.
  • Change the default of hpcycle to yes, so that in a SEATS adjustment the trend-cycle will by default be decomposed into the long-term trend and a cycle using the modified Hodrick-Prescott filter.
  • Add seasonaloverdiff argument to automdl spec, to set whether or not the automatic model identification procedure performs a test for seasonal overdifferencing.
  • Add hptarget and hprmls arguments to seats spec, to set the target for the Hodrick-Prescott filter and if level changes are removed from the series before the Hodrick-Prescott filter is applied, respectively.
  • Update test for Easter in the automatic model identification procedure to take into account models with more than one Easter regressor.
  • Ensure proper test of models with Easter regressors is done in the AIC testing procedure when more than one Easter regressor is specified.
  • Do not allow the aicdiff and pvaictest of the regression spec to be specified in the same spec file.
  • Include information about p-value for AIC testing in output if the pvaictest used.

Changes to the program output

  • Remove stable seasonal F statistic, M and Q statistics from default output.
  • Remove double mention of tdstock in error message.
  • Allow printing QS for indirect adjustment separately from other QS statistics.
  • Add ARIMA roots to savelog output.
  • Add space to footnote output for trading day.
  • Add seasonal autocorrelation function values to savelog output.

In HTML output:

  • Move foot tags to end of table.
  • Add index entry for outlier passes.
  • Add index entry for Chi-squared tests.
  • Correct headers for sliding spans range output.
  • Error check for '.html' rather than '.out'.
  • Do not print out names and links to "_err.html" files that do not exist.
  • Change blank space for modeling output (space after nonfixed coefficient, for example) to a nonbreaking space in the HTML output (&nbsp).
  • Use HTML generating subroutines to generate tabular output rather than formatted writes.
  • Generate index for table related to the business cycle.
  • Center some tabular output in the modeling section.
  • Add abbr tags for output of trading day and length of month/quarter or leap year regressors.
  • Add </div> tags where needed.
  • Create a list of outliers that generate singularity errors.

Changes to .udg output

  • Add ARIMA roots to .udg output.
  • Increase precision of some modeling variables saved to the .udg output.
  • Remove output related to D11 F test from .udg output.
  • Increase precision of some modeling variables saved to the .udg output.
  • Change key samode to samodeseats in .udg output to avoid repeating the samode key.

Internal program changes

  • Set very small numbers (absolute value < 10-100) to zero before printed out or saved.
  • Change parameter ALL to ALLFIL.
  • Initialize timer indicator variable.
  • Double maximum size of input file names.
  • If trading day group name changed for change of regime, change the length of month/leap year group name. Change handling of missing values in plotting routines.
  • Create additional parameter files for table names to print and save.
  • Create additional parameter files for table descriptions.
  • Save forecasts for unmodified SI ratios when appendfcst=yes.
  • Initialize variable for log file div tags.
  • In the automatic model identification procedure, initialize backup for convergence tolerance in all situations.
  • Move code to extract file name to start of the main driver routine.
  • In the ratneg subroutine, handle cases where values are very small.
  • Initialize new variables for generating unique div labels.
  • Generate the date time string without relying on whether the program is run in DOS, UNIX, or Linux.
  • In the addtd subroutine, move statement to return to top of routine.
  • Turn off seasonal overdifferencing test if SEATS seasonal adjustment is specified.
  • Save forecasts for modified SI rations (D9).

Fixes to defects in previous versions

  • Correct problem with printing change-of-regime regressors when saving model specs.
  • Correct setting start date when more than one prior adjustment factor is read.
  • Correct how the series used to generate QS diagnostics adjust for prior effects.
  • Fix output for the Chi squared and F test diagnostics.
  • Fix problems with index entries in HTML output.
  • Correct problems with index entries.
  • Use correct names for AO or LS sequence outliers.
  • Correct calls to inpter in the getssp, getsrs, rngbuf subroutines.
  • Add “#” to address label in makeSkip subroutine.
  • Ensure the program never takes the log of zero when converting double precision numbers to character strings.
  • Use correct beginning and ending pointers for quarterly series for quarterly QS diagnostic generated from monthly series.
  • In the final stages of the automatic model procedure, stop the reduction of AR orders at zero.
  • Ensure the program handles the situation of no identified model orders correctly.
  • Correct the error checking of the input files to what is expected.
  • Correct code to catch underflow errors occurring due to high precision calculations. Underflow errors caused unstable final results.
  • Change index of the HTML output file to say “Backward deletion pass <number>” instead of “Backward addition pass <number>”.
  • Correct problems in the automatic modeling procedure related to the Hannan-Rissanen (H-R) routine. There was no stability guarantee in the H-R routine used in automatic modeling. In one case this led to estimating explosive MA roots. As a result, different models were selected under 32-bit versus 64-bit architecture when the program ran automatic modeling.
  • In monthly series run with qcheck=yes, correct the label of the QS of the quarterly series obtained using the spectrum span (the last eight years, typically). It had taken the quarter number but printed the corresponding month instead, e.g., Jan, Feb, Mar, or Apr, not Jan, Apr, Jul, or Oct.
  • Truncate the series on either end so only months that make up full quarters are used to calculate qcheck.
  • Change QS for the quarterly-series calculated from a monthly series run with qcheck=yes to agree with the QS from the quarterly series run directly.
  • show the t value for outliers that are not part of the regression model in the final t statistic table for additive outliers; it was showing zero in some cases.
  • Show zero for outliers that are in the regression model in the final t statistic tables for additive or level shift outliers, regardless of the ARIMA model.
  • Correct the _tbs.html columns. When running the seats spec, some extra output files are created, including _tbs.html. Whenever the residuals were included in the table, the HTML version incorrectly included a column of zeroes, labeled "E-Resid", and shifted every other column right, resulting in everything to the right of the residuals being improperly labeled and an extra unnamed column.

We are interested in your feedback. Please email your questions and comments to x12@census.gov.

Page Last Revised - October 8, 2021
Is this page helpful?
Thumbs Up Image Yes Thumbs Down Image No
255 characters maximum 255 characters maximum reached
Thank you for your feedback.
Comments or suggestions?


Back to Header