Holt-Winters with additive seasonality — morie_holt_winters

Level, trend and seasonal recursions with the season ADDED, which is the right choice when the seasonal swing is a fixed number of units regardless of the level.

Usage

morie_holt_winters_additive(
  y,
  period = 4,
  alpha = 0.3,
  beta = 0.1,
  gamma = 0.1,
  horizon = NULL
)

Arguments

y: series.
period: seasonal period.
alpha, beta, gamma: smoothing parameters in \([0, 1]\).
horizon: steps ahead; defaults to one full period.

Value

list with forecast, level, trend, seasonal, fitted, residuals, sse.

Details

Three smoothing parameters is a great deal to ask of a short series, and the method needs two full periods before it can even initialise. On anything under about four periods the seasonal estimates are mostly the initialisation.

References

Winters, P. R. (1960). Forecasting sales by exponentially weighted moving averages. Management Science, 6(3), 324-342.

Examples

y <- rep(c(10, 20, 30, 20), 5) + seq_len(20)
round(morie_holt_winters_additive(y)$forecast, 1)
#> [1] 29.5 41.0 52.6 44.3