Forecasting Process Summary
Forecast Function (Scoring)
For causal models that include controllable causal factors, the predictions can be influenced by the future decisions made by the organization that produces the forecasts. Changing the future values of the controllable causal factors changes the forecasts. Organizations want to make decisions that benefit themselves. To help organizations make better decisions, the future values of the controllable causal factors can be varied to their benefit. The future values of the causal factors can be varied for scenario analysis (what-if analysis), stochastic optimization, or goal-seeking to aid proper decision-making.
In scenario analysis, the organization sets the future values of the causal factors to specific values and then evaluates the effect on the forecasts. In stochastic optimization, the organization algorithmically varies the future values of the causal factors to find the optimum of an objective function (profit, revenue, or cost function) based on the forecasts. In goal seeking, the organization algorithmically varies the future values of the causal factors in order to determine the values that achieve a certain goal (profit, revenue, or cost goal) based on the forecasts.
For example, suppose the following:
An organization desires to predict the demand for a product or service.
The demand is influenced by its sales price and by its advertising expenditures.
These data are recorded over time.
The following types of analysis can be used to answer questions about the time series data:
Scenario analysis can help answer the question "What happens to demand if the organization increases the sales price and decreases the advertising expenditures?"
Stochastic optimization can help answer the question "What is the optimal sales price and advertising expenditure combination that maximizes profit?"
Goal-seeking can help answer the question "What are the combinations of sales price and advertising expenditures that achieve a specified sales target?"
The sales price and advertising expenditures for a given time period can influence demand in future time periods. Static regression ignores these dynamic effects, which often leads to poor predictions, which in turn leads to poor decisions. Dynamic regression captures these dynamic effects and provides better predictions, which in turn facilitates better decisions.
Forecast score files (or forecast functions) summarize the time series model’s parameter estimates and the final states (historical time series information). These files can be used to quickly generate the forecasts required for the iterative nature of scenario analysis, stochastic optimization, and goal-seeking computations. Since most of the computational effort associated with automatic forecasting is time series analysis, diagnostics, model selection, and parameter estimation, forecast scoring is relatively effortless. Therefore, forecast scoring makes the iterative nature of large scale decision-making more tractable.
The results of forecast scoring include the predictions, prediction standard errors, and the confidence limits. All of these results can be used in decision-making.