Forecasting Methods

A causal model is the most sophisticated kind of forecasting tool.  It mathematically shows the relevant causal relationships, and may include pipeline considerations, such as inventories, and market survey information.  It may also incorporate the results of a time series analysis, and everything known of the dynamics of the flow system by utilizing predictions of related events such as employee strikes and promotions, and competitive decision making.  The model may even include factors for each location in a flow chart and connects these by equations to describe overall product flow if certain data is available.  Sometimes, assumptions can be made about some of the relationships when specific aspects of data are insufficient.  After this assumption, you can track what is going on and test your assumptions.  This causal model can be revised at any time as more information is made available.

Some types of causal methods include the econometric model, intention-to-buy and anticipation surveys, input-output model, economic input-output model, diffusion index, leading indicator, and life-cycle analysis.

The diffusion index is typically used to forecast sales by product class.  Data from several years is usually required for this causal model.  The diffusion index is the percentage of a group of economic indicators that are increasing or decreasing.  It is good at identifying turning points, but its accuracy is low to medium for short and medium term forecasts, and very low for long-term forecasting.

An econometric model is used to forecast sales by product classes, and to forecast margins.  Multiple years’ worth of quarterly history is needed to find significant relationships.  Mathematically, there has to be two more observations than independent variables.  It is a system of mutually dependent regression equations that defines a segment of economic sales or profit activity.  They predict turning points more accurately than a typical regression model because of the system of equations that better show the connections involved.  Although it is an expensive model to develop, costing between $5,000-$10,000, its accuracy is very high for short to long-term forecasting.