Merging models and experts
Section snippets
Propositions
In this section I put forward four propositions, of which some are perhaps obvious, given the abundant evidence in the literature, while one or two propositions may seem surprising and lead to discussion. Some of the propositions are based on my consulting experience for the Netherlands Bureau of Economic Policy Analysis (CPB) and for the pharmaceutical company Organon. The CPB uses a many-equations model that covers the full economy in order to produce point forecasts for quarterly
Research agenda
To summarize, the above propositions clearly indicate that the interaction between a model and an expert is a very important one. It is doubtful whether macroeconomic or sales forecasts can be based either on experts' opinions only or on model output only, so the interaction is important, as it can be observed in everyday practice. Apparently, there can be something like an optimal interaction between the two. This means that modellers need to know more about what experts do with their
Acknowledgements
Many thanks are due to an anonymous Associate Editor for helpful suggestions.
References (10)
Improving the voluntary integration of statistical forecasts and judgement
International Journal of Forecasting
(2000)- et al.
Judgemental forecasting: A review of progress over the last 25 years
International Journal of Forecasting
(2006) - et al.
The M3-competition: results, conclusions and implications
International Journal of Forecasting
(2000) The future of macroeconomic forecasting: Understanding the forecasting process
International Journal of Forecasting
(2007)- et al.
Time series analysis and simultaneous equation econometric analysis
Journal of Econometrics
(1974)
Cited by (16)
Demand forecasting in the presence of systematic events: Cases in capturing sales promotions
2020, International Journal of Production EconomicsCitation Excerpt :Despite the broadly acknowledged human factor in forecasting, much of the research on forecasting methods that emerged prior to the 1990's advised against the use of judgment in forecasting (e.g., Armstrong, 1986; Hogarth and Makridakis, 1981). However, recent research advocates that statistical methods and human judgment should be integrated so that complementary benefits can be realized to mitigate the inherent weaknesses of each approach (Alvarado-Valencia et al., 2017; Baecke et al., 2017; Blattberg and Hoch, 1990; Fischer and Harvey, 1999; Franses, 2008; Marmier and Cheikhrouhou, 2010). One important benefit of integrating statistical methods and human judgment relates to their capability to handle different types of information.
Integrating human judgement into quantitative forecasting methods: A review
2019, Omega (United Kingdom)Citation Excerpt :The extent to which experts contribute to improving FSSs functioning is an unresolved point in the literature. Franses [47] argued that the forecasts produced by a smaller FSS are more likely to be adjusted by experts. This is because there is an expectation that the FSS has not considered all the aspects of the forecasting task.
How accurate are government forecasts of economic fundamentals? The case of Taiwan
2011, International Journal of ForecastingCitation Excerpt :Governments can, and do, provide both replicable and non-replicable forecasts. In virtually all cases, information on how a model-based forecast is translated into a non-replicable forecast is not recorded (see, for example, Franses, 2008, and Goodwin, 2000). The evaluation of model-based, or replicable, forecasts has drawn much attention in the applied econometrics and forecasting literature.
A collaborative demand forecasting process with event-based fuzzy judgements
2011, Computers and Industrial EngineeringCitation Excerpt :For instance, in defining product priorities in a forecasting process, Meunier Martins, Cheikhrouhou, and Glardon (2005) take into account contextual information on lead times, component commonalities and product criticalities. Franses (2008) considers the integration of contextual information in the downstream part of the forecasting process, where different combinations of judgemental and mathematical forecasts are possible. Furthermore, it has been proved that, in most cases, collaborative forecasting gives better results than a process with a single input (Aviv, 2007).
Data intelligence: Trends and challenges
2020, Xitong Gongcheng Lilun yu Shijian/System Engineering Theory and Practice