- Adds the option to generate forward simulations. See
`?predict.statespacer`

for details.

- Fixes the returning of NA when calling
`Cholesky()`

without a format.

- Removes the dependency on the YieldCurve package by including an
updated FedYieldCurve dataset in the statespacer package. See
`?FedYieldCurve`

for details.

- Introduces the simulation smoother for drawing random samples
conditional on the observed data. See
`?SimSmoother`

for details.

- Extraction of components now done a bit more efficiently.

- Fixed incrementation of initialisation_steps during initialisation when the dependent variable was missing.

- Kalman Filter and Smoother now fully written in c++.

- Computation of diagnostics now optional using
`diagnostics = TRUE`

. - Added full system matrices to the statespacer object.
- Added predicted, filtered, and forecasted
`P_star`

and`P_inf`

.

- Fixed bug in predicting using explanatory variables.
- Fixed edge case in adding explanatory variables. The bug occurred when there was only 1 explanatory variable and 1 dependent variable supplied, causing the Z system matrix to be a vector instead of an array.

- Patch for macOS. On macOS, assignment of nested lists is handled a
bit differently than on other platforms. For instance, if
`first <- list()`

, then assigning`first$second[[["third"]] <- 1`

returns on Windows (and other platforms) a list named`first`

, that contains another list named`second`

, that contains a named element`third`

equal to an unnamed length 1 numerical vector. On macOS though, it returns a list named`first`

, that contains a named element`second`

equal to a named (`third`

) length 1 numerical vector. So`second`

is a list on the other platforms, while being a named numerical vector on macOS. This caused a bug on macOS while computing standard errors.

API change by making use of S3 classes and methods:

Replaced

`StateSpaceFit()`

for`statespacer()`

.Replaced

`StateSpaceEval(...)`

for`statespacer(..., fit = FALSE)`

.`statespacer()`

returns a list object of class`statespacer`

.Replaced

`StateSpaceForecast()`

for the S3 method`predict.statespacer()`

.

Major:

- Calculation of loglikelihood now done using c++. Major performance improvement as the loglikelihood potentially gets calculated a lot of times by the optimisation procedure.

Minor:

Making use of crossprod and tcrossprod.

Improved efficiency of the computation of standard_errors.

y_temp outside of LogLikelihood function.

Easier to find proper initial values, reducing time spent on trial and error by the user.

Printing progress now optional using

`verbose = TRUE`

.Computation of standard_errors now optional using

`standard_errors = TRUE`

.

- Initial release to CRAN!