About AFS

The principal author of Autobox is David Reilly.  The program began to be developed in 1966 at the University of Wisconsin Statistics Department under the direction of George Box under the title "The Pack Program" with David J. Pack as the principle statistician/programmer. Beginning in 1968, David Reilly, then at the University of Pennsylvania, developed a rule-based heuristics system that is today the Autobox approach to ARIMA modeling. The main thrust was to provide a mechanized or automated option to the Box-Jenkins (ARIMA) approach to modeling a time series process.  SAS and BMDP were early source code customers in 1972 for the Pack Program.

 The first release of AUTOBOX (then called AUTOBJ) occurred in November 1969 on an IBM 360 located at The American Stock Exchange.  The Interactive Data Corporation and Chase Econometric Forecasting Associates provided AUTOBJ in the context of XSIM as a time sharing package.   It was also provided in a time sharing environment by Compuserve and Computer Science. AUTOBJ was converted to the PC beginning in 1982, and first released as a DOS application later that year.  It was converted to Windows beginning in 1990 and first released as a Windows application in 1991.  The name was changed to AUTOBOX in January 1988.  It is also available as dll and an object library for Unix/Linux platforms and has been successfully integrated (OEM) into a number of more general software packages.

Automatic Forecasting Systems (AFS) was founded in 1975 to launch and market radical improvements in forecasting software.  By incorporating causal factors into the forecast such as sales of related products (e.g., shampoo and conditioner), cannibalization effects and even the price of competitors products.  These systems must also account for events like advertising and promotions, including lead (sales arising ahead of holidays) and lag effects (e.g., reduced sales the week after a promotion ends).

  The preset or "pick best" models found in many forecasting systems today (e.g., SAP, SAS, ORACLE) produce sub-optimal forecasts. To produce more accurate forecasts, AUTOBOX automatically tailors the model to the problem at hand including selecting the best lead and lag structures for each input series and the best weightings.  It corrects for omitted variables (e.g., holidays or price changes that have affected the historical data, but that the system has no knowledge of) by identifying pulses, seasonal pulses, level shifts and local time trends, and then adding the needed structure through surrogate variables.  Conversely, it also eliminates unneeded structure (e.g., a statistically unimportant causal variable) to keep the model manageable.  It performs all these functions as part of its normal routine without human intervention.  It also reports the statistical tests used to determine the model parameters, and let users manipulate the coefficients and model structure if they want.

In the twenty-five plus years since its launch AFS continues to focus on that core goal while adding statistical features that are hot out of the statistical journals. AFS's products, across platforms and interfaces, continue to be based on a forecast engine that has been programmed and tuned to do what only the best of forecasters can do; build Box-Jenkins ARIMA and Transfer Function models.

AFS continually looks to the future of forecasting. This includes both methodologies and in applications. As advances in the field are made AFS has the depth of statistical knowledge and the practical programming experience to integrate them into our packages. We are also constantly working to expand our products breadth of application into exciting areas such as production planning and inventory control and real time systems.