Forecast Manager

5. Special events

Time series forecasting models allow projecting values in the future according to a given array of existing actual data. These models, however, do not allow integrating side information, so to make the whole forecasting exercise more close to reality. If our company wants to put in place a promotional campaign for increasing consumption, for instance, this may lead consumers to react in one of three different ways:

1. They may prefer our product due to the temporary benefit, otherwise they would have not (preference of non-users);
2. They may tend to buy more product then usual due to the appealing offer, which implies an increased stock level (buyers increased stock);
3. Some buyers anticipate their purchase in order to profit from the special offer, which implies again an increased stock level.

Therefore, special events may be characterized by two phases:

1. An increase in sales during the campaign;
2. A decrease in sales in the following periods.

The target reaction to a promotional campaign depends on three main factors:

1. Nature of the campaign. It may be a price reduction, an increase in quantity, a free gift, etc.
2. Intensity of the event(s). A temporary reduction in price of 5% generates different reactions than a reduction of 10%.
3. Coverage of the campaign. A quick and local campaign may produce different results than a prolonged national campaign.

In general, however, in order to produce effective results with promotional actions the campaign should be short enough to push the reaction of consumers. A too long time in action mitigates the effect of the temporary benefit and results tend to lose their appeal.

The Special event summary table below was made with Forecast Manager. The first column on the left lists the time periods when the user flagged the occurrence of special events (with the special events range in Input Data page) and on its right is shown the label the user assigned to each event. The smoothed value is found with one of the formulae described under Special Events in the chapter Technicalities. The Event Effect is found subtracting the smoothed value to the actual one. Finally, Event Coefficients are found with one of the methods described, again, under Special Events in the chapter Technicalities. Read Lewandowski (pg 196) for more details.

The Event coefficients, one for each kind of event, are then put on top of the forecast value with the formula below and at the desired forecast periods as required by the user:

It must be, however, said that there are other source of abnormality that can impair the forecasting exercise:

1. Changes in the series characteristics, such as average, trend, seasonality, etc.
2. Transitory exceptional events, such as holidays, strikes, etc.
3. Permanent exceptional events, such as exit or entry of a major competitor, political and regulatory changes, etc.

From case to case one may realize that the features offered by Forecast Manager for handling special events may also fit the case when working with other sources of abnormality.