Seasonality may be due to various reasons such as company policies of the buyers, preferences of the consumers, government policies which may be periodic, and seasonal pattern due to the climate. The corrections for such seasonal peaks and valleys in demand can be made by comparing these peaks/valleys with the general average demand during the non-seasonal periods. Comparative indices could be formulated for the seasons where the demand is over shooting or under shooting the average. The multiplication of the corresponding seasonal indices with the forecast average should give the forecast for the different seasonal periods.

Often, when forecasts are made for the total demand during the next year, the demands in the various months are expressed seasonally in terms of the fractions of the annual demand observed generally. For instance, if the March and September demand is observed to be 0.25 fraction of the total annual demand and the rest of the months are equally distributed in terms of the demand, the forecast for next year’s annual demand of one lakh items can be forecasted as 25,000 items each for the months of March and September and 5,000 items each for the rest of the months. Often such a simple forecast serves adequately. But sometimes a more rigorous forecasting model is necessary. In such cases we further correct our earlier exponential forecast model for seasonality.

Procedure for using Exponential Smoothing:

1. The demand for the past year is noted, say in terms of monthly demands in the past year.

2. If significant seasonal variation is observed, then a base series is formed. This series could be the demand for the last year repeated verbatim, or if the seasonal periods themselves are slightly fluctuating, (say, if it is undecided as to whether the seasonal peak comes during February, March or April) a centered moving average is found for all the past data (with the number of periods being 3, in the case mentioned here).

3. After finding the base series, the ratio of the current month’s demand and the corresponding base series demand is calculated. This is called the demand ratio.

4. The demand ratio is now forecast for the next period. This forecast is called the forecast ratio.

5. A forecast for the next month’s demand ratio is made by noting the pervious month’s forecast ratio, the alpha factor and the current month’s demand ratio. This is similar to what we did earlier for non seasonal data. The only difference is, here is the data has been processed in terms of demand ratios which need to be smoothened for the random component and a trend component has to be incorporated.

6. The forecast ratio now is corrected for the trend component by first finding the trend factor from the pervious observations of the demand ratio. So, the corrected forecast ratio for the next period (month) is given as follows:

RFRt+1 = FRt+1 + {1 – α / α} Tt+1

7. The corrected forecast ratio has now been rectified for any random component as well as for trend made cyclical factors. You may note that the seasonality has already been taken into account because of the ratio. Therefore, the next step is to get the forecast of the demand (not the demand ratio) by multiplying the forecasted demand ratio by the base series demand observed for the corresponding month. The demand history up to August 2001 is given. The forecast for the net period i.e. September 2001 is made giving trend and seasonality corrections.

—