Demand forecasting is and has been discussed quite a lot in supply chain management. Most professionals seem to have an opinion about the best way to forecast demand and how forecasting accuracy should be measured. There are lots of opinions and I have found that different situations need different approaches to forecasting. However, some basic issues are always true in forecasting. I thought it might be worthwhile to revisit them.
The indisputable facts of forecasting are:
1) A forecast always contains errors
- If you get it just right you’ve probably ordered only the amount forecast and have run out of goods – or then it is a coincidence
2) Aggregation increases forecast accuracy
- Aggregation effect works with time – e.g. week-level forecasts are more accurate than day-level forecasts
- Works with product hierarchies – e.g. product-group-level forecasts are more accurate than product-level forecasts
- Works with location hierarchies – e.g. retail-chain-level forecasts are more accurate than store-level forecasts
3) Increased time horizons reduce forecast accuracy
- You can forecast the near future more accurately than the more distant future, because you have fresher input data
- As a side note this holds normally true, but with a highly seasonal assortment the accuracy does not increase much before the beginning of the next season as you do not get any new sales data
The beauty of these three facts is that they are always true. You can also use them in many different ways in your supply chain management. In my next post I will share some examples on how our customers have done it.