Analysis of Count Time Series
Likelihood-based methods for model fitting and assessment, prediction and intervention analysis of count time series following generalized linear models are provided. Models with the identity and with the logarithmic link function are allowed. The conditional distribution can be Poisson or Negative Binomial.
Changes since version 1.3.0
- Renamed some functions in order to avoid confusion with S3 methods. For very few users the renaming of 'mean.fit' to 'tsglm.meanfit' might be relevant.
Changes since version 1.0.0
- The structure of the package has beed simplified with respect to the different link functions to avoid duplicated code (this does not affect the usage of the package).
- Functions 'se' and 'summary.tsglm' now return confidence intervals.
- New function 'QIC' computes a quasi information criterion.
- Some auxiliary functions for count data distributions have been exported now (see 'help(countdistr)').
- Covariates with negative values are no longer tolerated when fitting models with the identity link function but produce an error.
- S3 method 'predict' for class "tsglm" was extended. It now allows to choose the construction principle (argument 'type'), the computation method (argument 'method'), if the estimation error is accounted for (argument 'estim' and additional arguments 'B_estim' and 'coefs_given') and whether the coverage rate should hold globally (argument 'global').
- Functions 'marcal', 'pit' and 'scoring' now have a default S3 method which makes them more generally usable than just for objects of class "tsglm".
- Function 'scoring' does now allow to return the individual scores and not only the mean (argument 'individual').