|
The forecasting bundle of R
packages provides new forecasting methods, and graphical tools for
displaying and analysing forecasts. It comprises the following
packages
Downloads
Current version: 1.12.
Last updated: 22 April 2008.
Pre-release of version: 1.13.
Last updated: 24 April 2008.
Help and examples
History of changes
Version 1.13 (not yet released)
- Bug in plot.ets() fixed so that plots of non-seasonal models for seasonal data now work.
- Warning added to ets() if the time series contains very large numbers (which can cause numerical problems). Anything up to 1,000,000 should be ok, but any larger and it is best to scale the series first.
Version 1.12 (22 April 2008)
- Objects are now coerced to class ts in ets(). This allows it to work with zoo objects.
- A new function dm.test() has been added. This implements the Diebold-Mariano test
for predictive accuracy.
- Yet more bug-fixes for auto.arima().
Version 1.11 (8 February 2008)
- Modifications to auto.arima() in the case where ML estimation does not work for the chosen model. Previously this would return no model. Now it returns the model estimated using CSS.
- AIC values reported in auto.arima() when trace=TRUE and approximation=TRUE are now comparable to the final AIC values.
- Addition of the expsmooth package.
Version 1.10 (21 January 2008)
- Fixed bug in seasadj() so it allows multiple seasonality
- Fixed another bug in print.Arima()
- Bug fixes in auto.arima(). It was sometimes returning a non-optimal model, and occasionally no model at all. Also, additional stationarity and invertibility testing is now done.
Version 1.09 (11 December 2007)
- A new argument "restrict" has been added to ets() with default TRUE. If set to FALSE, then the unstable ETS models are also allowed.
- A bug in the print.Arima() function was fixed.
Version 1.08 (21 November 2007)
- AICc and BIC corrected. Previously I had not taken account
of the
sigma^2 parameter when computing the number of parameters.
- arima() function changed to Arima() to avoid the clash
with the arima() function in the stats package.
- auto.arima now uses an approximation to the likelihood
when selecting a model if the series is more than 100 observations or
the seasonal period is greater than 12. This behaviour can be
over-ridden via the approximation argument.
- A new function plot.ets() provides a
decomposition plot of an ETS model.
- predict() is now an alias for forecast()
wherever there is not an existing predict() method.
- The argument conf has been changed to level
in all forecasting methods to be consistent with other R functions.
- The functions gof() and forecasterrors()
have been replaced by accuracy() which handles
in-sample and out-of-sample forecast accuracy.
- The initialization method used for a non-seasonal ETS model
applied to
seasonal data was changed slightly.
- The following methods for ets objects were added: summary,
coef and logLik.
- The following methods for Arima objects were added:
summary.
Version 1.07 (25 July 2007)
- Bug fix in summary of in-sample errors. For ets models with
multiplicative errors, the reported in-sample values of MSE, MAPE,
MASE, etc., in summary() and gof() were incorrect.
- ARIMA models with frequency greater than 49 now allowed.
But there is no unit-root testing if the frequency is 50 or more, so be
careful!
- Improvements in documentation.
Version 1.06 (15 June 2007)
- Bug fix in auto.arima(). It would not always respect the
stated values of max.p, max.q, max.P and max.Q.
- The tseries package is now installed automatically along
with the forecasting bundle, whereas previously it was only suggested.
Version 1.05 (28 May 2007)
- Introduced auto.arima() to provide a stepwise approach to
ARIMA modelling. This is much faster than the old
best.arima().
- The old grid-search method used by best.arima() is still
available by using stepwise=FALSE when calling auto.arima().
- Automated choice of seasonal differences introduced in
auto.arima().
- Some small changes to the starting values of ets() models.
- Fixed a bug in applying ets() to new data using a
previously fitted model.
Version 1.04 (30 January 2007)
- Added include.drift to arima()
- Fixed bug in seasonal forecasting with ets()
Version 1.03 (20 October 2006)
- Fixed some DOS line feed problems that were bothering unix
users.
Version 1.02 (12 October 2006)
- Added AICc option to ets() and best.arima().
- Corrected bug in calculation of fitted values in ets models
with multiplicative errors.
Version 1.01 (25 September 2006)
- Modified ndiffs() so that the maximum number of differences
allowed is 2.
Version 1.0 (31 August 2006)
- Added MASE to gof().
- croston() now returns fitted values and residuals.
- arima() no longer allows linear trend + ARMA errors by
default. Also, drift in non-stationary models can be turned off.
- This version is the first to be uploaded to CRAN.
Version 0.99992 (8 August 2006)
- Corrections to help files. No changes to functionality.
Version 0.99991 (2 August 2006)
- More bug fixes. ets now converges to a good model more
often.
Version 0.9999 (1 August 2006)
- Mostly bug fixes.
- A few data sets have been moved from fma
to forecast as they are not used in my
book.
- ets is now considerably slower but gives better results.
Full optimization is now the only option (which is what slows it down).
I had too many problems with poor models when partial optimization was
used. I'll work on speeding it up sometime, but this is not a high
priority. It is fast enough for most use. If you really need to
forecast 1000 series, run it overnight.
- In ets, I've experimented with new starting conditions for
optimization and it seems to be fairly robust now.
- Multiplicative error models can no longer be applied to
series containing zeros or negative values. However, the forecasts from
these models are not constrained to be positive.
Version 0.999 (27 July 2006)
- The package has been turned into three packages forming a
bundle. The functions and a few datasets are still in the forecast
package. The data from Makridakis, Wheelwright and Hyndman (1998) is
now in the fma package. The M-competition
data is now in the Mcomp package. Both fma
and Mcomp automatically load forecast.
- This is the first version available on all operating
systems (not just Windows).
- pegels has been replaced by ets. ets only fits the model;
it doesn't produce forecasts. To get forecasts, apply the forecast
function to the ets object.
- ets has been completely rewritten which makes it slower,
but much easier to maintain. Different boundary conditions are used and
a different optimizer is used, so don't expect the results to be
identical to what was done by the old pegels function. To get something
like the results from the old pegels function, use forecast(ets(...)).
- simulate.ets() added to simulate from an ets model.
- Changed name of cars to auto to avoid clash with the cars
data in the datasets package.
- arima2 functionality is now handled by arima() and pegels2
functionality is now handled by ets.
- best.arima now allows the option of BIC to be used for
model selection.
- Croston's method added in function croston().
- ts.display renamed as tsdisplay
- mean.f changed to meanf, theta.f changed to thetaf, rw.f
changed to rwf, seasonaldummy.f to seasonaldummyf, sindex.f to sindexf,
and spline.f to splinef. These changes are to avoid potential problems
if anyone introduces an "f" class.
Version 0.994 (4 October 2004)
- Fixed bug in arima which caused predict() to sometimes fail
when there was no xreg term.
- More bug fixes in handling regression terms in arima
models.
- New print.Arima function for more informative output.
Version 0.993 (20 July 2004)
- Added forecast function for structural time series models
obtained using StructTS().
- Changed default parameter space for pegels() to force
admissibility.
- Added option to pegels() to allow restriction to models
with finite forecast variance. This restriction is imposed by default.
- Fixed bug in arima.errors(). Changes made to arima() meant
arima.errors() was often returning an error message.
- Added a namespace to the package making fewer functions
visible to the user.
Version 0.99 (21 May 2004)
- Added automatic selection of order of differencing for
best.arima.
- Added possibility of linear trend in arima models.
- In pegels(), option added to allow parameters of an
exponential smoothing model to be in the "admissible" (or invertible)
region rather than within the usual (0,1) region.
- Fixed some bugs in pegels.
- Included all M1 and M3 data and some functions to subset
and plot them.
- Note: This package will only
work in R1.9 or later.
Version 0.98 (23 August 2003)
- Added facilities in pegels.
- It is now possible to specify particular values of the
smoothing parameters rather than always use the optimized values. If
none are specified, the optimal values are still estimated as before.
- It is also possible to specify upper and lower bounds
for each parameter separately.
- New function: theta.f. This implements the Theta method
which did very well in the M3 competition.
- A few minor problems with pegels fixed and a bug in
forecast.plot that meant it didn't work when the series contained
missing values.
Version 0.972 (11 July 2003)
- Small bug fix: pegels did not return correct model when
model was partially specified.
Version 0.971 (10 July 2003)
- Minor fixes to make sure the package will work with R
v1.6.x. No changes to functionality.
Version 0.97 (9 July 2003)
Version 0.96 (15 May 2003)
- Many functions rewritten to make use of methods and
classes. Consequently several functions have had their names changed
and many arguments have been altered. Please see the help files for
details.
- Added functions forecast.Arima and forecat.ar
- Added functions gof and seasadj
- Fixed bug in plot.forecast. The starting date for the plot
was sometimes incorrect.
- Added residuals components to rw.f and mean.f.
- Made several changes to ensure compatibility with Rv1.7.0.
- Removed a work-around to fix a bug in monthplot command
present in R v<=1.6.2.
- Fixed the motel data set (columns were swapped)
|
