Autoregressive Forecast for Future Observations — forecast

Applies exponential smoothing state space model (ETS) for forecasting beyond the last observed value. The method is autoregressive and automatically selects optimal smoothing parameters via ets. Falls back to Holt's linear trend extrapolation if the forecast package is unavailable.

Any NA values in the input series are first interpolated using interp_pchip_flag.

Usage

forecast_autoregressive(y, h = 3, method = "ets")

Arguments

y: Numeric vector of observed values (may contain NAs which are interpolated first)
h: Integer number of periods to forecast (default: 3)
method: Character string: "ets" (default) for exponential smoothing via forecast package, or "holt_linear" for manual Holt's linear trend fallback

Value

List with four components:

forecast: Numeric vector of length h with point forecasts
lower: Numeric vector of length h with lower 80% confidence interval bounds
upper: Numeric vector of length h with upper 80% confidence interval bounds
method: Character string indicating which method was used

Details

When method = "ets", the function attempts to use ets to fit an exponential smoothing model with automatically selected components (error, trend, seasonal). If the forecast package is not available, it falls back to Holt's linear trend method.

Holt's linear trend method fits y_t = l_t + t_t where l_t is the level and t_t is the trend, using standard smoothing parameters (alpha = 0.3, beta = 0.1).

Examples

if (FALSE) { # \dontrun{
y <- c(10, 12, 14, 16, 18, 20)
result <- forecast_autoregressive(y, h = 3)
result$forecast
} # }