Value and efficiency of judgmental forecasts
In many forecasting situations statistical model based forecasts do not capture key features of the problem, for example, a simple extrapolative model (often used in supply chain forecasting) cannot include promotion or weather effects (Fildes and Goodwin, 2007).
This important area of forecasting practice has been neglected by previous researchers until Fildes et al. (2009). This EPSRC funded study showed the importance of judgmental adjustments in affecting accuracy and how adjustments suffered from systematic biases, an effect leading to inefficient, unduly inaccurate forecasts. The effects on the businesses are poorer service or too high stock holding.
The problem has been examined in the context of intermittent demand (Syntetos et al., 2009) and in an attempt to clarify conflicting results; Davydenko and Fildes (2013) have shown that the choice of how you evaluate the evidence through the choice of error metric is crucial. Trapero et al. (2013b) investigated the impact of judgemental interventions for promotional forecasts and found them to have inconsistent performance and should be used with care. The issue of understanding the ‘value-added’ arising from judgmental interventions in the forecasting process is important to organizational forecasting practice.
Two current issues are being researched: (i) how to combing judgmental information and statistical data into a joint model (Davydenko and Fildes, 2012); (ii) the effect of positive and negative information on judgmental forecasts (Goodwin, 2012).
A full list of publications can be found here.