Profit and Wealth Maximization
Accuracy and Bias
In general, a set of forecasts will be considered to be accurate if the forecast errors, that is, the set of et values which results from the forecasts, are sufficiently small. The next section presents statistics based on the forecast errors, which can be used to measure forecast accuracy. In thinking about forecast accuracy, it is important to bear in mind the distinction between error and noise. While related, they are not the same thing. Noise in the demand data is real and is uncontrollable and will cause error in the forecasts, because by our definition we cannot forecast the noise. On the other hand, we create the errors that we observe because we create the forecasts; better forecasts will have smaller errors.
In some cases demand forecasts are not merely inaccurate, but they also exhibit bias. Bias is the persistent tendency of the forecast to err in the same direction, that is, to consistently over-predict or under-predict demand. We generally seek forecasts which are as accurate and as unbiased as possible. Bias represents a pattern in the errors, suggesting that we have not found and exploited all of the pattern in the demand data. This in turn would suggest that the forecasting procedure being used is inappropriate. For example, suppose our forecasting system always gave us a forecast that was on average ten units below the actual demand for that period. If we always adjusted this forecast by adding ten units to it (thus correcting for the bias), the forecasts would become more accurate as well as more unbiased.
Logistics managers sometimes prefer to work with intentionally biased demand forecasts. In a situation where high levels of service are very important, some managers like to use forecasts that are "biased high" because they tend to build inventories and therefore reduce the incidence of stockouts. In a situation where there are severe penalties for holding too much inventory, managers sometimes prefer a forecast which is
In general, a set of forecasts will be considered to be accurate if the forecast errors, that is, the set of et values which results from the forecasts, are sufficiently small. The next section presents statistics based on the forecast errors, which can be used to measure forecast accuracy. In thinking about forecast accuracy, it is important to bear in mind the distinction between error and noise. While related, they are not the same thing. Noise in the demand data is real and is uncontrollable and will cause error in the forecasts, because by our definition we cannot forecast the noise. On the other hand, we create the errors that we observe because we create the forecasts; better forecasts will have smaller errors.
In some cases demand forecasts are not merely inaccurate, but they also exhibit bias. Bias is the persistent tendency of the forecast to err in the same direction, that is, to consistently over-predict or under-predict demand. We generally seek forecasts which are as accurate and as unbiased as possible. Bias represents a pattern in the errors, suggesting that we have not found and exploited all of the pattern in the demand data. This in turn would suggest that the forecasting procedure being used is inappropriate. For example, suppose our forecasting system always gave us a forecast that was on average ten units below the actual demand for that period. If we always adjusted this forecast by adding ten units to it (thus correcting for the bias), the forecasts would become more accurate as well as more unbiased.
Logistics managers sometimes prefer to work with intentionally biased demand forecasts. In a situation where high levels of service are very important, some managers like to use forecasts that are "biased high" because they tend to build inventories and therefore reduce the incidence of stockouts. In a situation where there are severe penalties for holding too much inventory, managers sometimes prefer a forecast which is