Dynamic Systems Estimation (Time Series Package)

Tools for multivariate, linear, time-invariant, time series models. This includes ARMA and state-space representations, and methods for converting between them. It also includes simulation methods and several estimation functions. The package has functions for looking at model roots, stability, and forecasts at different horizons. The ARMA model representation is general, so that VAR, VARX, ARIMA, ARMAX, ARIMAX can all be considered to be special cases. Kalman filter and smoother estimates can be obtained from the state space model, and state-space model reduction techniques are implemented. An introduction and User's Guide is available in a vignette.


Known Bugs

o Handling of initial conditions for the state space filter and smoother still needs work.

Changes in dse version 2015.12-1

o Fixed bug passing xlab in tfplot methods for TSdata and TSestModel.

o Added default package imports as now required by CRAN.

Changes in dse version 2014.11-1

o Removed directory src-c with C code which has not been used recently.

o Adjusted tests to write postscript files to a tempdir and remove it.

Changes in dse version 2014.4-1

o Removed capability to set DUP=FALSE in .Fortran calls, since that is now deprecated, and may be disabled in future versions of R.

Changes in dse version 2013.11-1

o Adjusted Imports and Depends.

o Minor changes so setRNG could be moved from Depends to Imports.

Changes in dse version 2013.4-1

o Removed old email address from vignette.

Changes in dse version 2013.3-2

o Fixed typo in DESCRIPTION caught by new CRAN checks.

o To reduce load on the CRAN check servers, the example for estBlackBox() was changed to use max.lag=3 rather than the default 12.

Changes in dse version 2013.3-1

o Setting of TREND in was not working properly for ARMA models in estMaxLike. (Bug identified by Joseph Francois Tagne.)

o Adjusted a Guide example simulation to set tframe so a line graph is produced in place of points.

o Setting of inputSeriesNames and outputSeries names was not working properly. (Bug identified by Judith Riedler.)

Changes in dse version 2012.6-1

o Function stability modified to return roots and their moduli as an attribute of the logical result. (Suggestion by Alexander Braumann.)

Changes in dse version 2012.4-1

o Added tfplot to Depends.

Changes in dse version 2011.11-1

o Updated maintainer email address.

o Changed package.description() to packageDescription() in DSEversion().

Changes in dse version 2011.3-1

o Changed periods() and tfperiods() to Tobs() to be consistent with changes in tframe.

o switched .onLoad to .onAttach and removed require("tframe") as that should be handled by Depends in DESCRIPTION file.

Changes in dse version 2009.12-1

o fixed reference to ts in error messages (now stats).

o Additional documentation clean-up.

Changes in dse version 2009.10-2

o Standardized NEWS format for new function news().

o Fixed a few documentation problems found by R-2.10.0 (beta) quality checks.

Changes in dse version 2009.10-1

o The dse bundle has been unbundled and the packages dse1 and dse2 renamed to dse and EvalEst respectively. Some additional reorganization has been done to give a slightly different focus to the second package. Stub packages dse1 and dse2 have been created to facilitate conversion for users' existing code. These stub packages simply require dse and EvalEst respectively.

o The renaming has not been done in the NEWS files below.

Changes in dse version 2009.7-1

o minor documentation corrections as found by a new R-devel.

Changes in dse version 2008.10-1

o Changed the way a default graphics device is opened in a test, because of a change in R .

o Some documentation cleanup related to previous changes.

Changes in dse version 2007.11-1

o GPL-2 license option was added explicitely.

o Function mineStepwise and method plot.mineStepwise were removed. It used stepwise and probably has not worked in a long time.

Changes in dse version 2007.7-1

o fixed a bug passing data to dse1 fortran arma code when the input data was longer than the output data. This especially affected forecast() in dse2.

o fixed a bug passing data to dse1 fortran kf code as with arma.

o fixed a bug in l.ARMA which caused an error when predictT exceeded sampleT and compiled=FALSE.

o the ... arg to forecast() is now passed along to l(). Among other things, this allows compiled=FALSE to be used with forecast() which helps testing.

Changes in dse version 2007.5-2

o dseplus is no longer maintained as a bundle.

o constants was added as argument in setArrays.

Changes in dse version 2007.5-1

o tframe was removed from the dse bundle and released as a stand alone package.

o Calls to function freeze (which made an R copy of data from a database based on a database descriptor) where removed from dse1 and dse2, and a few other places, since this functionality was rarely used by functions in these packages. This has simplified other package dependencies.

o warnings in optimization in the estMaxLik function evaluation was switched to FALSE so that warnings about working on a subspace should not happen. The warning is still generated at the end if the problem persists in the final model.

o A number of small fixes caught by R-2.5.0 were incorporated

o The syskern package in dseplus is defunct and the few remaining functions were moved to the monitor package, which was the only place they were being used. That package is being revised and may not be released for awhile.

Changes in dse and dseplus version 2006.10-1

o Namespace imports were fixed as need for R-2.4.0.

Changes in dse and dseplus version 2006.4-1

o 00Intro.dse1.Rd was renamed 00.dse.Intro.Rd and 00Intro.tframe.Rd was renamed 00.tframe.Intro.Rd. Also, aliases dse, dse.Intro, and dse-package were added, and similarly for tframe. With luck, this will mean users can find the general package description.

o A bug in setting non- zero or one constants in ARMA models was fixed. (The constants were still treated as parameters in estMaxLik().)

o A bug in setting non- zero or one constants in ARMA models was fixed.

o A bug printing ARMA models with no lags in the AR part was fixed.

o Some of the general documentation was cleaned up (a bit).

o percentChange, ytoypc, and diffLog were moved from dse1 to tframe, and all were made generic.

o pre-pending of info to names (e.g. %ch ) in above was changed so that it can be turned off using a global option setting ModSeriesNames=FALSE.

o All methods for tfplot were changed so that adding the Title can be turned off using a global option setting PlotTitles=FALSE.

o A tframe method for aggregate was added to tframe.

o The function standardize was removed. (It has been deprecated in favour of scale for a long time).

o The function dseclass was removed (since it was only used to compensate for differences with Splus, which is not longer being tested).

o The functions dsescan and read.int were removed. (dsescan was replaced with R's scan and read.int was moved to a local function in programs that used it.)

o Several Splus test values where commented out (since Splus is not longer being tested).

Changes in dse and dseplus version 2006.1-1

o \reference rather than \references in the documentation was fixed in several places and the references were cleaned up. (The mistake seems to have resulted in some references being omitted from earlier documentation.)

o The argument SIGMA in simulate methods and makeTSnoise was changed to Cov to avoid confusion in forthcoming package tsfa.

o Splus "For" loops where removed, as these have not been tested recently, and some other variables reported as having no binding were cleaned up.

o The function findg was removed (it has never really worked).

o A recursion too deep problem printing some TSdata was fixed.

o The simulate generic in dse1 was modified to avoid conflict with the simulate generic in stats (in R-2.2.0).

o Some further clean up of the Make system. Parallel makes (make -j -k) seem to be working properly now.

Changes in dse and dseplus version 2005.9-1

o An experimental class TSrestrictedmodel has been added, which allows a very general specification of cross-coefficient restrictions,etc. See ?TSrestrictedmodel for more details.

o smoother and l.SS where fixed to use solve(A,B) constructs rather B %*% solve(A) in some instances where this was not done already. This was done in both the S and fortran versions of the code, so that results for fortran and S versions are nearly identical even in the case of "big k" state tracking initialization, which is known to be numerically problematic. (In more usual cases the results were previously nearly identical.)

o a bug was fixed in the structure of additional information passed back from estMaxLik.

o a bug was fixed in the structure of some information passed back from the state space smoother.

o smoother.default was changed to smoother.TSmodel and gives an error message if not called with an appropriate SS model. (There is now no default method.)

o the setting of constants was added to constuctors ARMA and SS and a bug was fixed in setTSmodelParameters, so that these are not now reset to parameters after estimation.

o an argument rootP0 (default NULL) was added to state space (SS) models so that an initial state covariance (P0) can be specified in a parsimonious form. This may not be a permanent feature. See also the new TSrestrictedModel class.

o an argument eps (default 1e-15) was added to the stability function and roots within eps of 1.0 are now considered unstable (unstationary) models.

o some more argument checking was added to tfplot

o Depends lines where added to the bundle description file so that installation should now properly look after dependencies.

o The makefile system was revised so that tests are run individually rather than as a single block.

Changes in dse and dseplus version 2005.6-1

o the matrix inversion routine was switched to use Lapack routines distributed with R. These work better in some circumstances

o fixed a bug in fortran code for the state space smoother (SMOOTH) when the output dimension exceeded the state dimension. The matrix inversion was also switched (see above) which gives improved results.

o fixed a bug in the state space smoother when the model has input data u_t.

o a bug in TREND for ARMA models was fixed again.(Some details where missed in fix in the previous release.

o several minor documention errors were fixed.

o added additional tests of the smoother.

o added test to check that setting TREND works.

Changes in dse and dseplus version 2005.4-1

o package guides have been renamed from "pac-guide.pdf" to "pac.pdf" (e.g. dse1-guide.pdf is now dse1.pdf) so they can be displayed with e.g. print(vignette("dse1")) and the vignette code loaded in an editor with e.g. edit(vignette("dse1")). The combined guide is still dse-guide.pdf.

o more arguments have been added to tfplot to pass to par() and matplot. These give better control over plot line colours and styles.

o a bug constructing a TSdata format with univariate ts is fixed.

o a bug in TREND for ARMA models was fixed.

o acfM was removed. This was a utility for some code that was previously removed. (Use acf to calculate autocorrelations.)

o more testing of TSdata constructors and methods has been added.

Changes in dse and dseplus version 2005.1-1

o added NAMESPACES to all packages (except setRNG). Exporting generics but not specific methods means calling specific methods must be avoided. NAMESPACES in bundles seemed to require switching .DSECOMPILED and .DSEDUP global flags to "dynamic variables" using a local function.

o moved makeTSnoise from tframe to dse1 so tframe does not require setRNG.

o fortran code for state space models was changed to pass scratch arrays from R, so state and output dimension can now be arbitrarily large. Previously the state and output dimensions could not exceed a fortran parameter IS=100 without being recompiled, and an error was not always indicated. (There still seems to be a bug in maximum likelihood estimation when the state or output dimension exceeds 100.)

o fixed fortran code (SIMSS) to use w instead of e as measurement equation noise for non-innovations models (as in S version of code) and fixed related test.

o revise makefiles QC system to combine Rcode and codetest targets, which are now called $(CODETEST). This has a target that is the tested lib/package rather than a flag, which is more reliable and makes reading the makefiles much easier. The tarball target was changed in the same way.

o revise makefiles QC system to use configure so that QC can take place in a separate directory. Testing of different architectures and different R versions should now take place in a different directory, so the TMP and FLAGS naming convention was simplified.

o revise makefiles to test demos separately, and other makefile cleanup.

o reversed dependency between padi and dsepadi as generic is now in dsepadi (so that padi is not necessary for most things).

Changes in dse and dseplus version 2004.10-1

o The NEWS file for setRNG is now maintained separately in that package.

o Started using separate version numbers for setRNG and syskern, but still using a common one for all dse and dseplus packages.

o moved some more documentation from the code into the help files. There is still some clean up to do.

o fixed documentation mistakes, bugs and error checking in simulations with user supplied random numbers.

o TREND in ARMA models generalized and documented better and to indicate that this term in ARMA models handles more than just trends.

o added "state" extraction utility and examples in SS, l.SS and smoother documentation

o added a check in seriesNames<- that the number of names matched the number of series.

o added \concept{DSE} in the documentation.

o require(ts) and require(mva) changed to require(stats) for R 1.9.0

o continuing Makefile cleanup.

o moved bundle constuction to Bundle directory.

o moved NEWS to Bundle directory.

Changes in dse, dseplus and setRNG version 2004.4-1

o fixed window.default (tfwindow) warnings in vignettes

o fixed a bug in the constructor TSdata.default which prevented tframe and series name information from being properly used in the resulting TSdata object.

o fixed some version number tests which will not work correctly in R-2.0.

Changes in dse, dseplus and setRNG version 2004.3-1

o setRNG has been unbundled from dse and is distributed as a standalone package, but is required by dse.

o the tframe package was reworked to avoid the need to override generic functions (start, end, frequency, time, and diff) in base, as this is not allowed in 1.9.0. New generic functions (tfstart, ...) were added to look after some programming situations where the old generics do not work. Users typically should not need to use these methods at the command line (but may prefer to for programming). The method periods() was renamed tfperiods() to give a consistent naming convention.

o a tf (time frame) argument, which can replace start, end and frequency, was added to several functions.

o start. and end. where replaced by start and end in tfwindow, tfplot, and simulate methods, since the code can now distinguish the function start() from the variable start.

o the names of some method arguments were modified to match generic arguments (as caught be R 1.9.0 tools).

o makefiles were re-arranged and cleaned up.

o as.double() and double() calls to .Fortran changed slightly.

Changes in dse and dseplus version 2003.9-1

o fixed bug in fixConstants which prevented setting of arbitrary elements in SS models to constant (non-parameter) values other than 0 or 1. The documentation of fixConstants was also improved.

o fixed bug causing occassional seg faults in smoother.

o fixed bug preventing estMaxLik properly setting the estimated parameters in SS models (so they appeared to be unchanged).

o usage of random seed tightened up to correspond to R 1.8.0

o minor documentation errors caught by new tools() have been fixed.

o use of tools() in Makefiles has been been further refined.

o prcomp eliminated as version in R 1.8.0 now supports complex matrix.

Changes in dse and dseplus version 2003.6-1

o several more program names have been revised so as to avoid confusion with methods for generics with similar names. (That is, function foo.bar has been changed to fooBar, or to barFoo in some cases.)

o DSEversion (previously version.DSE) gives the version of each package.

o testing in Makefiles has been been further refined.

Changes in dse and dseplus version 2003.4-1

o dseplus which works with the previous dse changes is released.

o testing in Makefiles has been been further refined.

o test tolerances in dseplus have been adjusted for R 1.7.0 and Splus 3.3.

o very minor changes because of R 1.7.0 name spaces.

Changes in dse and dseplus version 2003.2-1

o several program names have been revised so as to avoid confusion with methods for generics with similar names. For this reason, the new Users' Guide should be used, despite its problems.

o the Users' guide is (being) broken down into pieces for each package and re-written using Sweave (with .Stex files in inst/doc). The Users' Guide (dse-guide.pdf) is still placed in dse1/inst/doc as the bundle structure does not have a place for documentation.

o the old package syskern has been divided into setRNG which is included with dse and a new package syskern which is moved to dseplus.

o most routines have been convert to use La.svd and La.eigen in place of svd and eigen (and test values were revised slightly in some cases).

o DSE.HOME has been removed and .path.package is used in its place.

o version.DSE now gives the version of each package.

o testing in Makefiles has been revised so code (Rcode+tests) is the only inter-package dependency. Examples, demos, and guides do not affect other packages.

o reliance on bundle structure is largely removed, and Makefiles have been re-organized to reflect this.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.