Nowcasting and "news" content of variables

[srlanalytics]

The gain_and_fit branch has been modified so that:

• if store_idx is specified, the forecast update contribution of each variable to the series specified by store_idx. The example us_gdp.R now includes an example of the contribution of each series to nowcast updates for a specified period. Updates to a specific series are stored as idx_update. If scale = T is specified (the default), forecast updates are unscaled before being returned. However, they remain in logs and/or differenced, so that these updates can be interpreted as percent change to the original variable.

• in every case, forecast updates to factors are stored in factor_update

• gain and prediction error are no longer returned as updates are a far more concise and useful way to save this information. Advanced users can still access gain and prediction error through Cppbdfm().

@christophsax --- idx_update and forecast_update are lists. Right now individual elements can only be accessed by index (i.e. est$idx_update[[460]]). If input data are ts_boxable() it would be nice to be able to call these results by date (i.e. what was the contribution of each series to the nowcast on 2018-03-31?). Any ideas on the best way to do this?

[christophsax]

m$idx_update is now a multiple time series and can be accessed, e.g., by:

window(m$idx_update, start = c(2018, 12), end = c(2018, 12))

Adjusted example in us_gdp.R.

[christophsax]

We should include a section on this in the vignette, this seems important.

[christophsax]

Has a section now, 'Forecast Updates', including a nice ggplot