Ivan Svetunkov

Rewrite C++ code using RcppEigen package

1

It seems that this should decrease computation time due to more efficient matrix multiplication. But this means that almost all the parts of C++ code would need to be rewritten.

Modify existing models to cater for external variables (provided by user). This is especially important for simulated intervals in cases of mixed ETS models.

This is needed for ARIMAX models, where ARIMA states are backcasted and X is optimised.

Develop a method that would accept lm/glm/what-not and return alm class. This should simplify things in case alm output is needed for a non-alm model.

Method for SR to produce point forecasts and prediction intervals from the SR

Implement method for dynamic multipliers in alm()

1

This should take formula, calculate polynomials then generate multipliers for specific variables.

Create a function producing LOWESS smoothing for specific quantiles. This might help with regression diagnostics, showing, how the distribution of residuals changes.

This is done for normal and normal-based distributions and for Poisson. Can also be done for: 1. Poisson with AR(p), 2. Logit regression, 3. Other? Need to investigate.

Compatibility with data.table and tbl

3

Need to check all the functions for the compatibility with "data.table" and "tbl" classes, so that they are used most efficiently and the functions still work. This relates to: cramer,...

Distributions for ALM

25

1. Normal distribution, 2. F-distribution, 3. Weird one with the ratio of folded normal distributions,