The Forecast Pro SDK is a totally seamless forecasting solution for your application. The Forecast Pro SDK generates accurate statistically-based forecasts using the same state-of-the-art methodology found in Forecast Pro.

The Forecast Pro SDK uses a Windows dynamic link library (dll) to expose several forecasting functions to a calling program. The calling program provides pointers to structures containing the time series to be forecasted, instructions indicating how the forecasts should be prepared and room for the output. The Forecast Pro SDK calculates the forecasts and writes out the results, including forecasts, confidence limits, safety stocks, model details and summary statistics.

The Forecast Pro SDK was written in Visual C++. Because it communicates via pointers, the Forecast Pro SDK can be called from virtually any Windows-based development platform including C, C++, VB, Java and .NET. The Forecast Pro SDK comes with detailed documentation describing how to call the library as well as sample calling programs written in VB, C++ and .NET and we also provide full technical support during the integration process.

The Models

The Forecast Pro SDK supports many of the models contained in Forecast Pro XE. The supported models include:

• Expert Selection: Expert selection uses a combination of rule-based logic and out-of-sample testing to automatically choose the appropriate forecasting method from among all of the supported time series models.
• Quick Expert Selection: Quick expert selection uses rule-based logic to automatically choose the appropriate forecasting method from among all of the supported models except Box-Jenkins.
• Simple Moving Averages: The calling program can specify the number of terms to use or allow the Forecast Pro SDK to determine the number automatically.
• Croston’s Intermittent Demand Model.
• Exponential Smoothing: Twelve different Holt-Winters models are supported (all combinations of Trend = none, linear, damped, exponential and Seasonality = none, additive, multiplicative). The specific smoothing model can be determined automatically by the Forecast Pro SDK or dictated by the calling program. Parameters can be optimized using a nonlinear search or dictated by the calling program.
• Box-Jenkins: The Forecast Pro SDK supports a multiplicative seasonal Box-Jenkins model. Model identification is automatic and parameters are estimated via unconditional least squares.
• Dynamic Regression: The Forecast Pro SDK supports dynamic regression models including the ability to use lagged dependent variables and build generalized Cochrane-Orcutt models.
• Event Models: Event models extend exponential smoothing by providing adjustments for special events like promotions, strikes or other irregular occurrences. The calling program passes a schedule of events covering the historical and forecast period and the Forecast Pro SDK calculates indices to adjust for different type of events.
• Weight Transformations: Any supported model can be used in conjunction with weight transformations. The calling program passes a set of weights covering the historical and forecast period and the Forecast Pro SDK divides the times series by the weights, forecasts the resultant de-weighted series and then reapplies the weights. Weight transformations can be used for many purposes including user-defined seasonal patterns, adjusting for 4-4-5 calendars, new product forecasting (analogy forecasting) and many others.
• Out-of-Sample Testing: You can specify a hold-out sample and the Forecast Pro SDK will calculate rolling out-of-sample statistics including MAPE, MAD and GMRAE.

Nearly Two Decades of Refinement

Business Forecast Systems has been the leader in forecasting software since 1986. With more than 25,000 installations worldwide, our software has forecasted literally billions of time series. Over the years, our clients have sent us hundreds of “oddball” time series that generated poor forecasts. The program’s author, Dr. Robert Goodrich, has carefully analyzed each of these series to determine the cause of the behavior and then modified the forecasting technique or expert selection algorithm to detect these exceptions and respond appropriately. Thus the Forecast Pro SDK is far more than just a handful of forecasting techniques—it is a robust forecasting tool that embodies the knowledge and experience of nearly two decades of working with business data. It recognizes and responds to numerous situations never written about in textbooks.

Proven Accurate

Forecast Pro recently outperformed all of the other software approaches as well as 18 out of 19 academic teams in the largest and most comprehensive empirical forecasting study ever performed. Sponsored by the International Journal of Forecasting, the Makridakis-3 study compared the accuracy of 26 different approaches used to prepare 3,003 forecasts based on historic demand data. Nineteen of the approaches were implemented by forecasting experts from academic institutions, including the Wharton School, Case Western Reserve, INSEAD, the University of Pennsylvania and the Imperial College (London). The remaining seven approaches were implemented using commercially available, fully automated forecasting packages, specifically Autobox (3 submissions), ForecastX, SmartForecasts, Autocast and Forecast Pro. Approaches included using techniques such as exponential smoothing models, Box-Jenkins models, neural networks and rule-based approaches. Human judgment and statistical expertise played a significant role in many of the approaches. The most striking result was the performance of Forecast Pro (using the automatic expert selection), which significantly outperformed all other software approaches as well as 18 out of 19 academic teams. The study’s results were published in two special issues of the International Journal of Forecasting (Volume 17, Numbers 3&4 (2001)).