The forecasting model is based on the traditional sales-data decomposition approach:
•The first component of the model is the sales level. This level can be estimated by three different methods: either by a naïve forecast consisting in using the last observed sale, either by a moving average on a given rolling horizon, or by exponential smoothing.
•The second component is the trend and has to be considered for sales which exhibits a regular increase (or decrease) for several successive time periods. This trend adjustment is said to be additive if the increase (or the decrease) is linear over time, or multiplicative, if the increase (or the decrease) is exponential over time.
•The third component is the seasonality and has to be considered for sales which exhibits periodic fluctuations. This seasonality adjustment is said to be additive when it is done in absolute value, with respect to the basic forecast without trend adjustment. It is said to be multiplicative when it is done in relative value, with respect to the basic forecast without trend adjustment.
Level estimation
This level can be estimated by a naïve forecast, by a moving average on a rolling horizon, or by exponential smoothing,
Naïve forecast
The naïve forecast consists in using the last observed sale as level forecast.
Moving average
The moving average consists in estimating the arithmetic average for n last periods, where n is the horizon defined in the cell labeled Number of periods.
Exponential smoothing
The exponential smoothing consists in a weighted average, whereby each data is given a weight which decreases according to how hold the data is. One has to define the Smoothing coefficient, with a value between 0 and 1 (bounds excluded).
Trend adjustment
In a set of historical data, the trend reflects the fact that the data are increasing (or decreasing) over time. The forecasting model integrates this mechanism via a trend adjustment,
This adjustment is computed by means of an exponential smoothing of the observed trends between successive data. The observed trend can be
•additive : the trend adjustment is computed as the difference between successive data. This method has to be used when the data increase (or decrease) is mainly linear over time.
•multiplicative : the trend adjustment is computed as the ratio between successive data. This method has to be used when increase (or decrease) of data is exponential (or more generally non-linear) over time.
One has to define the Smoothing coefficient for the trend adjustment procedure, with a value between 0 and 1 (bounds excluded).
Seasonal adjustment
When historical data exhibit periodical fluctuations, one has to consider seasonal adjustment,
First, one has to enter the number of periods in a given periodical cycle, in the corresponding cell. Then, a seasonal adjustment is computed for each period of the cycle. This adjustment can be estimated in an absolute or a relative manner. One has thus to choose between:
•additive seasonal adjustments: in this case, the adjustments are estimated from the difference between the successive data and the average of the data on the associated cycle,
•multiplicative seasonal adjustments: in this case, the adjustments are estimated from the ratio between the successive data and the average of the data on the associated cycle,
One has to define the Smoothing coefficient for the seasonal adjustment procedure, with a value between 0 and 1 (bounds excluded).