In other studies, the judgmental adjustments led to improved accuracy of forecasts, providing support for the premise that combining statistical and judgmental methods improves accuracy. In a series of three studies, Matthews and Diamantopoulos (1986, 1989, 1990) investigated the effects of judgmental adjustments on the accuracy of statistical forecasts using data obtained from a health care company based in the United Kingdom. The company produced over 900 individual repeat purchase items within 18 categories. The study included 281 items representing all 18 product categories. The statistical forecasts for these items were adjusted by product managers who had domain knowledge about the product, the market and the customers. In their cross-sectional and longitudinal analyses, Matthews and Diamantopoulos reported an increase in accuracy of forecasts that were judgmentally adjusted by product managers.
A study that supports the above results was conducted by Huss (1985) where he examined load forecasting among the largest electric utility companies in the United States. He gathered sales forecasts and actual sales data from companies as well as the forecasting methods used and compared the effectiveness of several forecasting methods over various time horizons. The results showed that smoothing techniques coupled with judgmental adjustments outperformed econometric models.
The effects of judgmental adjustments of macroeconomic forecasts were examined by McNees (1990). According to the author, the psychology literature
concentrates disproportionately on the drawbacks of human judgment and hence discourages the use of judgmental adjustments lest it harms the accuracy of statistical forecasts. The macroeconomic literature, on the other hand, views judgmental
adjustments as a symptom of poor modeling. In other words, if a researcher specifies a model correctly and the model functions well, there would be no need for judgmental adjustments of the model output. In this study, macroeconomic forecasts of three models were compared to those of four forecasters who adjusted the output of their models judgmentally. In most of the cases, judgmentally adjusted forecasts were more accurate than forecasts that were publicized without any judgmental adjustments. Next, the author compared the original and revised forecasts of four different forecasters. Again, he reported that judgmental adjustments helped improve forecast accuracy. He concluded that it is unwise to view judgmental adjustments in an all-or-nothing mindset but rather judgmental adjustments should be used if and when appropriate. In fact, it is impossible to explain the macroeconomic behavior of a country by a single model and predict its future performance. Although a model may represent the fundamental relationships between major players in an economy, there are many factors that occasionally affect the economy such as regulatory changes, strikes, and natural disasters. The author suggests that one can improve forecast accuracy by judgmentally adjusting macroeconomic forecasts for the above mentioned events.
In a review comparing mixed results on relative efficacy of judgmental versus statistical forecasts, Bunn and Wright (1991) make three important points. First, they maintain that the studies that emphasize the fallacies of human judgment and favor statistical forecasts underestimate the effectiveness of human judgment in real life since
these studies have serious methodological limitations. In fact, most of these studies rely on undergraduate students performing simple paper and pencil tasks in an experimental setting. In an actual business setting, however, a judgmental forecast is generated under much more favorable conditions than in a lab study; that is, the forecaster tends to be well-trained, well-motivated, familiar with the forecasting task and have business knowledge and experience. The discrepancies between lab experiments and field studies will be explained in more detail in Chapter 4 (see Section 4.1).
Second, they caution against generalizing drawbacks of human judgment to judgmental forecasts because judgment is generated by different mechanisms in these cases. The studies about human judgment mainly focus on past events. However, judgmental forecasting involves future events and the underlying cognitive processes may be different. Studies have shown that significant differences exists when individuals judge the correctness of their answers related to past versus future events (Wright 1982;
Wright and Wisudha 1982; Wright and Ayton 1986, 1988 and 1989). In general, people tend to be more confident about their judgments when they related to past or present information such as when they are guessing the length of the Suez Canal. However, when forecasting future events, people tend to be better calibrated in terms of subjective
probability judgments. Hence, human judgment related to future events such as
judgmental forecasting and judgmental adjustments should be studied in their own right.
Finally, they recognize the comparative advantages and disadvantages of judgmental and statistical forecasts and recommend that statistical and judgmental forecasts be combined to improve the accuracy of the final forecast.