Moving Average Forecasts
A Moving Average (MA) forecast works by taking the average of the last observed points and projecting this forward. It is one of the simplest forecasting approaches and is often used as a baseline model.
Limitations of Simple Moving Averages
- Equal weighting: Every value in the chosen window contributes equally, regardless of how recent it is. This means recent shifts in the data are not emphasized.
- No trend or seasonality: SMAs produce flat forecasts that cannot capture upward/downward movements or repeating seasonal patterns.
- Window size sensitivity: The choice of window length heavily influences the forecast. Too small a window may be noisy; too large a window may overly smooth the series.
- Forecast horizon problem: An SMA essentially fails to produce meaningful predictions beyond the averaging window, since it always outputs a constant extension.
Because of these drawbacks, SMA is rarely used for serious forecasting beyond benchmarking. More advanced methods, such as Exponential Smoothing, Holt’s, or Holt-Winters’, address these issues by giving more weight to recent data and modeling trend and seasonality explicitly.
Variants
Simple Moving Average (SMA)
- Uses the last values to compute a single average.
- Produces a flat forecast line into the future.
- Useful as a naive benchmark but not for dynamic patterns.
Rolling Moving Average
- Each new prediction is updated iteratively using the most recent window, which can include earlier forecasts.
- Produces a line smoother than Holt’s, but still limited in capturing complex dynamics.