How much data is enough for generating monthly sales forecasts?
We are regularly asked how much historical data I need to improve my forecasting accuracy using sales forecasting software?
Here is a general guide, to reliably fit and test statistical models, you need a time series of 48-60 months. To pick up seasonal patterns correctly you will need at least 3 seasonal cycles to improve forecasting accuracy for a forecasting sales or purchasing, stock seasonality model.
With 48 months, for example, the first 36 can be used to fit the model and the last 12 to test the model's forecasting accuracy. For short-term forecasting, going back more than 60 months is unlikely to be helpful, since most statistical methods apply more weight to the more recent than to the distant past.
And when there is no data available and a forecast is none the less required, it is still possible to develop a forecast. One easy way is to enforce the history from another item and literally cut and paste it into the forecast making manual adjustments. Top-Down Planning is another option when using Forecast Pro. This uses the pattern from the product group to drive the forecast. This has the benefit of being automated and derives its sales and seasonality from the other like products at the group level. The results can then be reviewed and adjusted at the granular level using modifiers by the planners.
A little user experience in Forecast PRO, natural curiosity and training will greatly improve forecasting accuracy, speed and response times.
Contact MXI Software