AUTOBOX generates forecasts using historical data and causal data. Current forecasting systems vary in their ability to automatically detect seasonality and general trends in historical data and to deal with unusual values. It is extremely important to be able to quickly identify changes in trends.
Demand forecasting is usually performed using some variety of moving average or exponential smoothing, probably with seasonal adjustment. There are many applications where these tools are cost-effective and satisfy the customers needs. Technically aggressive customers have found a need for more robust methods which have the additional benefit of optimally incorporating cause variables because the past never "causes" the future. If you ignore or simply don't include important causal variables then all you have to lean on is the sales/shipment history. The future is "caused" by marketing decisions, competitor action,holiday effects etc. Optimally combining information in these variables as well as the past sales/shipments can be quite effective. Both the simple traditional methods and these powerful combining methods can be blind-sided by outliers, seasonal pulses, level shifts and local time trends. AUTOBOX identifies these factors and includes them in the final model.
AUTOBOX is a tool that is useful in dealing with unusual values or quickly incorporating changes in level or trend or detecting changes in response to company policies. For example, a 25% drop in price may not have the same effect now as it had when the product was first introduced. Statisticians refer to this as time-varying parameters. Robust tools have become important as enterprises need to forecast at the SKU level and use point-of-sale (POS) data. AUTOBOX is an expert system which can incorporate causal variables by detecting lead, contemporaneous or lagged effects.
This fine art even extends to cannibalization effects and even the price of competitors products. AUTOBOX accounts for events like advertising and promotions, including lagged effects (e.g., reduced sales the week after a promotion ends) along with the best weighting.
AUTOBOX does not select a model from a user or system-defined set of models. To produce more-accurate forecasts, AUTOBOX automatically tailors the forecast model to each problem. and the best weightings. It corrects for omitted variables (e.g., holidays or price changes that have unknowingly affected the historical data) by identifying pulses, seasonal pulses, level shifts and local time trends, and then enhances the forecast model through dummy variables and/or autoregressive memory schemes.
At the same time AUTOBOX eliminates unneeded structure (e.g., a statistically unimportant causal variable) to keep the model manageable. These are all performed as part of its normal routine without human intervention. A wide variety of reports provide detailed information on the statistical tests used to determine the model parameters. Advanced users may manipulate the coefficients and model structure if they choose to do so.
CLICK HERE:Example of Predicting Daily Beer Demand