Cross‐item learning for volatile demand forecasting: An intervention with predictive analytics

Abstract

Despite its importance to OM, demand forecasting has been perceived as a "problem-solving" exercise; most empirical work in the field has focused on explanatory models but neglected prediction problems that are part of empirical science. The present study, involving one of the leading electronics distributors in the world, aims to improve prediction accuracy under high demand volatility for procurement managers to make better inventory decisions. In response to requests for an integrated forecasting methodology, we undertook an iterative process based on three guiding principles-data pooling, theoryinformed feature engineering, and ensemble-based machine learning. The resulting framework managed to improve forecast accuracy significantly and is applicable to a broad range of situations. We present reflections and insights derived abductively through engagement with managers in this problem situation. This "problem-driven" process corresponds to intervention as a research strategy that can foster theoretical and methodological innovations in OM. Our contribution goes beyond the development of the prediction framework as it elucidates ways OM researchers could leverage theoretical foundations to inform feature derivation and model construction. We posit that this work points to a way forward to the combination of OM principles with the emerging innovations in data science and artificial intelligence.

Supervised machine learning for theory building and testing: Opportunities in operations management

“…A limitation is that most ML methods (including RF) are developed for independent observations and not fully calibrated for time series and panel data (Dominici et al, 2021). That said, empirical studies have shown efficacy of ML for panel data predictions (e.g., Chuang et al, 2021) and theoretical studies are thriving to show consistencies of ML estimators for longitudinal data (e.g., Medeiros & Mendes, 2016). Our hope is that our articulation of the upside of ML, abetted by clear examples meant to reduce barriers to adoption, will encourage OM scholars to apply ML methods in their research endeavors.…”

“…We labeled this category Supply Chain Management (SCM). We observe that ML exhibits considerable potential to generate innovative solutions to persistent SCM issues like forecast (Chuang et al, 2021), procurement (Ban et al, 2019;Mandl & Minner, 2020), and facility location (Glaeser et al, 2019). Besides traditional SCM topics, ML has been used to tackle environmental issues such as predicting violations of waste gas emission standards (Chang et al, 2021) and humanitarian logistics such as detecting sex trafficking locations (Keskin et al, 2021).…”

“…Laboratory work shows that shared forecast information may even be consciously misreported due to strategic concerns (Spiliotopoulou et al, 2016). Despite all the information-sharing technologies, recent evidence in semiconductor supply chains still shows that rolling forecasts do not necessarily become more accurate as they are updated (Chuang et al, 2021), in turn causing financial losses as the vendor builds capacity according to noisy forecast signals. That said, the average forecast level is an anchoring point for suppliers' sourcing/stocking decisions and should exhibit a positive association with inventory coverage.…”

Empirical research methods department: Mission, learnings, and future plans

“…Forecasting, machine learning models (Chuang et al, 2021;Ilk et al, 2020;Ketzenberg et al, 2020;Pak et al, 2020;Petropoulos et al, 2018) Review Reviewing the application of existing methods in OM research. An article in this class would begin with a comprehensive survey of published research, then provide a summary of issues identified, and follow up with recommendations for improvement.…”

“…Typically, they develop an empirical method to solve an OM problem that common techniques in the empirical toolbox do not fully address. Many studies utilizing prediction, forecasting, and machine learning models as part of their method development process naturally fit under this class (Chuang et al, 2021;Ilk et al, 2020;Ketzenberg et al, 2020;Pak et al, 2020;Petropoulos et al, 2018). Whether such research goes to a topical department or to the ERM department depends on the novelty of the method and the degree to which the contribution comes from the method rather than from the research project more generally.…”