Forecast skill shares with prediction intervals

Fits the chosen model to each series and returns the forecast median share and the 80% and 95% prediction intervals on the share scale.

Usage

forecast_share(tsb, model = "ETS", horizon = 12L)

Arguments

tsb

A tsibble from prepare_share_tsibble().

model

A single model name ("SNAIVE", "DRIFT", "ETS", "ARIMA") used for all series, or a data.frame with the key columns and modello, e.g. the serie element of select_share_model(). Series without a model are skipped.

horizon

Forecast horizon in months. Default 12.

Value

A data.table with the key columns, mese (Date, first day of the month), quota, lo80, hi80, lo95, hi95 and modello.

Details

Models are fitted to logit_quota as in backtest_share_models(); the forecast quantiles (0.5, 0.1, 0.9, 0.025, 0.975) are mapped back with the logistic function, so lo95 <= lo80 <= quota <= hi80 <= hi95 and all values lie in \([0, 1]\). quota is the forecast median. Failed fits give NA values.

Examples

if (requireNamespace("fable", quietly = TRUE)) {
  set.seed(8)
  panel <- data.table::CJ(skill_id = "a", mese_idx = 1:24)
  panel[, n := 2000L]
  panel[, x := rbinom(.N, n, plogis(-3 + 0.02 * mese_idx))]
  tsb <- prepare_share_tsibble(panel, "skill_id", origin = "2023-01")
  forecast_share(tsb, model = "DRIFT", horizon = 3)
}
#>    skill_id       mese      quota       lo80       hi80       lo95       hi95
#>      <char>     <Date>      <num>      <num>      <num>      <num>      <num>
#> 1:        a 2025-01-01 0.07366612 0.06179791 0.08760079 0.05626190 0.09590707
#> 2:        a 2025-02-01 0.07463007 0.05788942 0.09571997 0.05052009 0.10892618
#> 3:        a 2025-03-01 0.07560560 0.05503082 0.10303408 0.04639495 0.12087629
#>    modello
#>     <char>
#> 1:   DRIFT
#> 2:   DRIFT
#> 3:   DRIFT