Back to results
US 8,285,582 B2Granted
Automatic calculation of forecast response factor
AssigneeTeradata US, Inc.
InventorsArash Bateni, Edward Kim, Philippe Hamel, Stephen Szu Chang
Life of this patent
Estimating the term…
Abstract
A forecast response factor (RF) determines how quickly product demand forecasts should react to recent changes in demand. When a product sales pattern changes (e.g., a sudden increase in product demand), RF is adjusted accordingly to adjust the forecast responsiveness. The present subject matter provides automatic calculation of the RF, based at least in part on the nature of the product sales (autocorrelation) and the status of recent forecasts (bias).
Claims18
A machine-implemented method residing in a non-transitory computer-readable medium and executed by a machine, comprising:
- determining, by the machine, an auto-correlation coefficient (AC) using historical demand data, the AC measures a correlation of demand at each week with respect to that of a previous week, demand values are present in the historical demand data for each week for a given product or service;
- determining, by the machine, a bias of last forecast (B) using a previous forecast and historical demand data, the bias measures a dimensionless deviation of a last week sales forecast for the given product or service from an actual de-seasonalized demand, the bias is calculated as the absolute value of the last weeks sales forecast minus a last weeks de-seasonalized demand, the absolute value is then divided by the max of the last weeks sales forecast and the last weeks de-seasonalized demand;
- determining, by the machine, a number of consecutive over/under forecasts (n) using previous forecasts, n measures a specific number of times that sales forecasts falls over or falls under actual de-seasonalized demand;
- and
- automatically calculating, by the machine, a forecast response factor (RF) using the auto-correlation coefficient, the bias of last forecast and the number of consecutive over/under forecasts, wherein automatically calculating the RF is calculated as a sum of the auto-correlation coefficient added to a product of the number of consecutive over/under forecasts and the bias of the last forecast, the sum is then divided by 2.
2 more elements
A machine-implemented method, comprising:
- accessing a database populated with data indicative of historical demand and demand forecasts for at least one product;
- determining an auto-correlation coefficient (AC), a bias of last forecast (B), and a number of consecutive over/under forecasts (n) based on the data, the AC measures a correlation of demand at each week with respect to that of a previous week, demand values are present in the historical demand for each week for the at least one product, the bias measures a dimensionless deviation of a last week sales forecast for the product from an actual de-seasonalized demand, the bias is calculated as the absolute value of the last weeks sales forecast minus a last weeks de-seasonalized demand, the absolute value is then divided by the max of the last weeks sales forecast and the last weeks de-seasonalized demand, the number measures a specific number of times that sales forecasts falls over or falls under actual de-seasonalized demand;
- using a processor to automatically adjust a forecast response factor (RF) using the coefficient, bias and number, wherein automatically adjusting the RF includes automatically calculating the RF as a sum of the auto-correlation coefficient added to a product of the number of consecutive over/under forecasts and the bias of the last forecast, the sum is then divided by 2;
- and
- forecasting product demand based at least in part on the adjusted RF.
2 more elements
A system, comprising:
- a database comprising a plurality of entries corresponding to historical demand data and forecast data for a product;
- a forecast response factor module configured to access the database and calculate a forecast response factor (RF) based on at least one of the entries, wherein the forecast response factor includes an indication of responsiveness of a forecast to changes in demand and wherein the calculated RF is calculated as a sum of an auto-correlation coefficient added to a product of a number of consecutive over/under forecasts and a bias of the last forecast, the sum is then divided by 2;
- and
- a demand forecast generator configured to access the database and receive the forecast response factor, the demand forecast generator further configured to generate demand forecasts using the calculated forecast response factor.
1 more element
Dependencies are inferred from the claim text.