Plans, forecasts, and budgets are based on predictions. For many situations, predictions are built from the ground up each time a prediction is required and historical data is not used. The study of variation from past predictions can lead to alternative ways to improve the predictions. There are six basic approaches to developing predictions:

• Base them on research when results cannot be seen for a long time (e.g., school curricula).
• Use leading indicators (e.g., use housing starts to predict demand for flooring material).
• Develop time series models to take advantage of auto-correlation in historical data.
• Make real-time updates of predictions as new information comes available.
• Use simple averages of historical data.
• Anticipate special causes (e.g., airlines change cutoffs for loads when a large group books a flight).