Promotional or Base-line Forecasts - Which is Right for Me?

The challenge that inevitably comes with forecasting of any kind is the need to reduce error.  It can be reduced in several ways.  Increasing the amount of data used helps reduce error.  For instance, having four samples of Sunday sales is not as good as having twenty samples, not just because it will help detect patterns better, but also to help weed out bad data and minimize its impact.  This is also true when determining whether or not to provide promotional indicators along with the pricing and sales data.  The debate comes from the fact that this data is not typically readily available.  The amount of work to extract promotional data, if it exists, varies from retailer to retailer. 

With base-line forecasts, promoted and non-promoted sales are mixed together.  This makes it harder to isolate anomalies in the data because it will contain both a mixture of high moving and low moving sales data.  So, depending on the frequency of promotions, non-promoted item forecasts might appear a little high and promoted item forecasts might appear a bit low.  The basic rule of thumb is that some promotion indicator is better than no promotion indicator from a forecast perspective.