Exponential Smoothing is a broad statistical method used in a forecast to eliminate random variations from historical data, in order to identify the trend and obtain a better estimate of the future demand. In practice it is a dynamic average (compared to the latter it has the advantage of not having to use all the historical n data every time, but only the last forecast made, in which, however, different weights are used (to factor the different historical data), the importance of which diminish exponentially the older the data becomes.
The calculation of ES may be expressed by the following formula:
Ft +1 = Ft + α × Dt − α × Ft
Which simplified becomes: Ft + 1 = (1 − α) × Ft + α × Dt
F = forecast at time t or at the successive time t+1
D = the last observation made
Alpha = smoothing factor
Alpha will be a value between 0 e 1, the lower the value of alpha, the more level and stable the series. On the contrary the higher the value of alpha the more the forecast will be influenced by recent data and random variations. The value is usually between 0.05 and 0.3. There is an automatic way of calculating it which is finding the value that minimizes the forecast error.
Regarding Exponential Smoothing there are further versions of it. Of note, there is Double Exponential Smoothing which in its equation takes into account the trend.