The Future of Demand Forecasting with Generative AI
Yue Li and Rachel Pedersen taking a deep look into the part that generative AI will play in the future of demand forecasting.
Integrating Forecasting and Inventory Decisions Using Machine Learning
Joost van der Haar, Yves Sagaert, and Robert Boute share their study of three Belgian companies in the food industry, and the findings that better forecasts do not necessarily lead to better inventory decisions.
Types of Forecast Errors and Their Implications
Forecast errors are inevitable, but not all errors are created equal. So goes the argument by Kolja Johannsen, who categorizes four types of forecast errors and provides strategies for responding to them.
Special Feature: Revisiting Symmetric MAPE
- Errors on Percentage Errors Forty years ago, the “asymmetry” of mean absolute percentage error was noted by Scott Armstrong. Forecasts that exceeded the actual were penalized more harshly by MAPE than forecasts below the actual, introducing a possible incentive for biasing forecasts to the low side. Rob Hyndman opens our special feature with a recap of that history.
- Sparse-Proof sMAPE The discussion continues with Slawek Smyl proposing a new metric he calls Sparse-Proof MAPE (msMAPE), designed to better handle large-valued as well as sparse (intermit-tent) time series when forecasts and actuals are non-negative.
- Know Your Errors! The special feature ends with a commentary by Stephan Kolassa on Smyl’s msMAPE and a call for using simulation to better understand what any error metric does in a variety of situations.
Explainability: A Requirement for Trust in Forecasts
Trevor Sidery argues that explainability is a requirement for trust in forecasts and categorizes four types of explainability requirements involving methods, components, drivers, and errors. In a pair of commentaries, Anne-Flore Elard looks at the distinction between explainabilty and explanations and notes that when models lack a direct mapping with business drivers, this creates a roadblock to their trust and adoption. Then Zabiulla Mohammed agrees that explainability is important for building trust, but not at the expense of predictive power or business value.
Special Feature: United Nations Sustainable Development Goals
- The Role of Forecasting in Ending Global Hunger, Lauren Davis
- Life Below Water, Leo Sadovy
Book Review
Book review contributor Ira Sohn examines David Spiegelhalter’s The Art of Uncertainty – How to Navigate Chance, Ignorance, Risk and Luck. Sohn finds Spiegelhalter to have a singular command of the technicalities of statistics and probability, along with a special talent for communication that exudes confidence and trust.
Opinion Editorial – Overcategorization of Continuous Data
Malte Tichy discusses the overcategorization of continuous data. In the most egregious cases, “category hacking” occurs when different category splits are tested until one happens to be statistically significant.
To receive this current issue, download here.
