Forecasting Methods

Expert Selection takes the guesswork out of forecasting. The built-in expert system analyzes your data, selects the appropriate forecasting technique, builds the model and calculates the forecasts–it even explains its reasoning in ordinary English!

With Forecast Pro’s automatic machine learning (ML) models you no longer need to be a data scientist to build accurate ML-based forecasts. Forecast Pro uses Extreme Gradient Boosting Decision Trees and includes a predefined set of features allowing you to build models completely automatically. You can also customize the model by listing additional features (e.g., explanatory variables, event schedules, etc.)

Twelve different Holt-Winters exponential smoothing models are provided to accommodate a wide range of data characteristics. The robustness of exponential smoothing makes it ideal when there are no leading indicators, and when the data are too short or volatile for Box–Jenkins. You can select the model and set the parameters yourself or let Forecast Pro do it automatically.

For stable data sets, Forecast Pro supports a multiplicative seasonal Box-Jenkins model. The model can be built completely automatically or interactively using a full range of screen-oriented diagnostics.

If there are important leading indicators, use Forecast Pro’s dynamic regression. You can include independent variables, lagged or transformed variables and build generalized Cochrane–Orcutt models. Using Forecast Pro’s self–interpreting diagnostics, you can build and compare alternative models with a few clicks of the mouse.

Event models extend exponential smoothing by providing adjustments for special events such as promotions, strikes or other irregular occurrences. You can adjust for events of several different types such as promotions of varying sizes or types, or movable holidays like Easter and Rosh Hashanah. Creating, adjusting and maintaining events is easy using Forecast Pro’s intuitive Event Manager.

Multiple-level models allow you to aggregate data into groups that can be reconciled using a top–down or bottom-up approach to produce consistent forecasts at all levels of aggregation. Seasonal and event indexes can be extracted from the higher–level aggregates and applied to lower-level data.

Forecast Pro includes several methods for forecasting new products, including forecasting by analogy, item supersession and the Bass diffusion model.

This is a useful technique if you are forecasting data with more than 12 observations per year. Seasonal Simplification reduces the number of seasonal indexes used to model the data and often substantially improves forecast accuracy.

Croston’s intermittent demand model and discrete data models are provided to accommodate low volume and “sparse” data (i.e., data where the demand is often zero).

Curve fitting provides a quick and easy way to identify the general form of the curve which your data are following. Forecast Pro supports four types of curves–straight line, quadratic, exponential and growth (S–curve).

This set of “simple” models can be extremely useful. Moving average, “same as last year,” percentage growth and fixed forecast value models are included.