Back-Transform Forecasts

Converts forecasts produced on a transformed scale back to the original scale.

Usage

ts_back_transform(forecast_tbl, params, bias_adj = TRUE)

Arguments

forecast_tbl

Forecast tibble from ts_forecast().

params

The params element from ts_transform().

bias_adj

Logical; if TRUE (default), the point-forecast column is back-transformed with a bias adjustment so that forecasts on the log/Box-Cox scale represent means rather than medians. The forecast variance required by the adjustment is derived per horizon from the widest stored prediction interval, so it applies only to genuine forecast rows: when the tibble has no lo_*/hi_* interval columns the plain transform is used (historical/fitted values round-trip unchanged). Interval columns are quantiles and are always transformed with the plain transform.

Value

The forecast tibble with numeric forecast columns restored.

Examples

x = as_ts_df(AirPassengers)
tr = ts_transform(x, method = "log")
fc = tibble::tibble(step = 1:2, forecast = log(c(500, 600)))
ts_back_transform(fc, tr$params)
#> # A tibble: 2 × 2
#>    step forecast
#>   <int>    <dbl>
#> 1     1      500
#> 2     2      600