Using a rolling training approach to improve judgmental extrapolations elicited from forecasters with technical knowledge

详细信息    查看全文

• 作者：Fotios Petropoulos^a ; ^{PetropoulosF@cardiff.ac.uk}Author Vitae ; Paul Goodwin^bAuthor Vitae ; Robert Fildes^cAuthor Vitae
• 关键词：Judgmental forecasting ; Unaided judgments ; Rolling training ; Feedback ; Time series ; Expert knowledge elicitation
• 刊名：International Journal of Forecasting
• 出版年：2017
• 出版时间：January-March 2017
• 年：2017
• 卷：33
• 期：1
• 页码：314-324
• 全文大小：915 K
• 卷排序：33

文摘

There are several biases and inefficiencies that are commonly associated with the judgmental extrapolation of time series, even when the forecasters have technical knowledge about forecasting. This study examines the effectiveness of using a rolling training approach, based on feedback, to improve the accuracy of forecasts elicited from people with such knowledge. In an experiment, forecasters were asked to make multiple judgmental extrapolations for a set of time series from different time origins. For each series in turn, the participants were either unaided or provided with feedback. In the latter case, the true outcomes and performance feedback were provided following the submission of each set of forecasts. The objective was to provide a training scheme that would enable forecasters to understand the underlying pattern of the data better by learning from their forecast errors directly. An analysis of the results indicated that this rolling training approach is an effective method for enhancing the judgmental extrapolations elicited from people with technical knowledge, especially when bias feedback is provided. As such, it could be a valuable element in the design of software systems that are intended to support expert knowledge elicitation (EKE) in forecasting.