Should consider per metric - some holidays might be really important e.g., for Calls but not for Responses. @TomMonks parsimony... if it's not doing anything, it should be removed, keep things simple. For all models but particularly ARIMA, where holidays are coded as a single dummy variable.
Should short-notice weather-related events be included as holidays?
Could ARIMA include holidays any better? e.g.,
- Not just single dummy variable.
- Change creation of holiday dummy variable for ARIMA so that it doesn't just use dates, but also uses lower_window and upper_window (already have function for this -
expand_holiday_windows() - just need to apply).
Potentially relevant: dynamic regression with lagged predictors
Should consider per metric - some holidays might be really important e.g., for Calls but not for Responses. @TomMonks parsimony... if it's not doing anything, it should be removed, keep things simple. For all models but particularly ARIMA, where holidays are coded as a single dummy variable.
Should short-notice weather-related events be included as holidays?
Could ARIMA include holidays any better? e.g.,
expand_holiday_windows()- just need to apply).Potentially relevant: dynamic regression with lagged predictors