Chapter 10: Monitoring and Revising Forecasts
Introduction: Four Primary Topics
1. Forecast guidelines
2. Identification of forecast error
3. Graphical and statistical techniques to track forecast results
4. Rules for auditing forecasts and comparing methodologies
Forecast Guidelines and Controls:
1. Statement of desired level of forecasting accuracy (quantitative or qualitative)
2. Acceptable error rates are higher for longer range forecasts
3. Sectors subject to rapid technological change expect higher errors
4. .Reflection of past unique historical patterns, such as numerous past turning points, increase expected errors. Past average percentage error is a guideline.
Sources of Forecast Error
1. Changing external factors (tax laws, global market influences, unusual weather patterns, etc.)
Dummy variables or responses to changing competitive conditions (market share influences) may be introduced into model.
2. Internal decisions and policies, such as advertising or pricing policies, will influence forecasts and necessitate greater communication with management.
3. Forecast methodology may be due to flawed data, wrong transformations, definition changes in what the data measures, and the level of aggregation. Will past historical influences continue into the future as assumed by forecasting models? Should new explanatory variables be introduced in regression models? If a conditional forecast is used is the future estimate of the independent variable reliable?
Tracking Techniques
1. There is no "best" way to track forecasted values over consecutive time periods.
Graphical Techniques
1. One graphical technique is the ladder chart that shows (a) the average value for each month, (b) the highest value for the month, (c) the lowest value for the month, and (d) the forecast value for the month.
2. A modified turning point error diagram rotates the turning point diagram until the 45 degree line of perfect forecast is horizontal. Turning point error occur when the predicted change and actual change are in opposite directions. Overestimated growth or decline and underestimated growth and decline occur if errors are consistently above or below the line of perfect forecast. See page 494.
3. A third graphical techniques plots actual versus predicted magnitudes for each time period.
4. Structural problems in the model may be shown by sequential forecasts versus actual values at different selected periods of time to see if wide variation take place at different time intervals.
5. Finally, the forecast value and lower and upper confidence limits for the error terms can be plotted against time. These error plots may be either on each value or on cumulative values.
Quantitative Tracking Techniques
1. Trigg's tracking signal is the ratio of the exponentially smoothed error term to the exponential smoothed mean absolute deviation. The smoothing constant is usually between .1 and .2. If the Triggs value exceeds 0.5 of alpha of .1 or 0.74 for alpha of .2 then the errors are nonrandom.
2. Other tests of the randomness of the error terms are the correlogram of the error terms and the Durbin-Watson statistic.
The Forecasting Audit and Control Cycle
1. The stages of the audit and control cycle depend upon the industry and firm under study. They begin with new information (see figure 10-10) either external or internal. When combined with monitoring and tracking procedures the forecaster is able to evaluate the effectiveness of the present forecasting methodology.
2. New information may suggest revisions of forecast methodology
3. The cycle is ongoing and used to monitor forecast revisions that also incorporates an estimate of the costs associated with the forecasting effort.