First order exponential smoothing or simple exponential smoothing is suitable with constant variance and no seasonality. The approach is usually recommended to make short-term forecast. Chapter 2, Understanding Time-series data, has introduced the naïve method for the forecasting where prediction in horizon h is defined as value of t (or the last observation):
xt+h = xt
The approach is extended by simple moving average, which extends the naïve approach using the mean of multiple historical points:
The approach assumes equal weight to all historical observations, as shown in the following figure:
Figure 3.4: Weight assigned to observation with increasing window size
As the window size for moving average increases, the weights assigned to each observation become smaller. The first order exponential extends this current approach by providing exponential to historical data points where weights decrease exponentially from the most recent data point to the oldest....