The Risks of Model-Based Forecasting: Modeling, Assessing, and Remodeling
This chapter highlights risks and challenges to model-based forecasting and also provides guidelines to minimize the forecast errors stemming from those risks. The nature of forecast risks differs depending on the duration of the forecast. We divide these risks into two broad categories: (1) risks related to short-term forecasting and (2) risks related to long-term forecasting.
In short-term forecasting, the chance of a permanent shift in the behavior of a series is low, but sometimes factors, such as a hurricane or a work-related strike, can affect the outcome. Including those factors in a model is difficult due to their unexpected and temporary nature. That said, adjustments can be made to the model's forecast to reduce forecast error.
For long-run forecasting, the event's nature tends to be permanent, and the chance of significant changes, or structural breaks, is very high. For instance, trying to forecast the path of home prices during the housing boom of the late 1990s through the mid-2000s would have been very difficult. From January 1997 to December 2005, the Standard & Poor's (S&P)/Case-Shiller House Price Index (HPI) followed a strong increasing trend—a trend that many analysts speculated would continue for the foreseeable future. However, the index peaked in April 2006, and a strong downward trend followed. The break in the prior trend was unexpected and therefore difficult to forecast. ...