FORESIGHT Issue 69

Description

2023:Q2

Special Feature ~ Is It Time to Retire the MAPE?

Mean Absolute Percent Error has long been derided by forecasting experts due to its many recognized problems, yet MAPE is still commonly relied upon by business forecasters. Malte Tichy leads off our first special feature by illustrating the potential danger of a MAPE-optimized forecast driving replenishment, and by calling for MAPE’s retirement. Stephan Kolassa and Flavio von Rickenbach provide sympathetic commentaries.

Special Feature ~ When and What Not to Forecast

The second special feature is based upon a webinar produced by the Lancaster Centre for Marketing Analytics and Forecasting. Webinar guests Paul Goodwin and Stephan Kolassa reopen the discussion, with additional commentary by Fotios Petropoulos.

Articles

Commentaries on 2023:Q1 Special Feature Does Forecast Accuracy Even Matter? Our previous issue featured seven wide-ranging approaches to this question. After a brief recap of these seven articles, three new commentaries by Argiris Mokios, Nico Sprotti, and Stefan de Kok add to the debate.

Practitioner’s Corner We’ve all had a chance to judge the work of our predecessors – and we’ll all have our own work judged by our successors. With the continuing progress of knowledge, we’re likely to look as foolish to our successors as our predecessors sometimes look to us. In his column, Patrick Bower makes an appeal for understanding and kindness when judging the work of those who came before you.

Financial Forecasting In something of a complement to the 2023:Q1 special feature, Marcin Klucznik and Jakub Rybacki examine whether policy predictability matters, focusing on the case of “forward guidance” provided by central banks.

Long-range Forecasting Hua Xie and Claudia Ringler from the International Food Policy Research Institute describe efforts toward the long-term projection of water supply and demand.

Machine Learning & AI In the first of a recurring series of practitioner articles adapted from research published in the International Journal of Forecasting, Jente Van Belle, Ruben Crevits, and Wouter Verbeke show how to reduce forecast instability with global deep learning models without necessarily harming forecast accuracy. This is an important takeaway for business forecasters, since more stable demand forecasts lead to fewer (and smaller) supply chain plan changes, and thus lower supply chain costs.

Spotlights Elaine Deschamps is Foresight’s Government & Public Policy Editor, and chair of IIF’s new Practitioner Section. Niles Perera is a Senior Lecturer at University of Moratuwa in Sri Lanka, with main research interests in judgmental forecasting.

Opinion-Editorial

• Stephan Kolassa provides the 10 steps he shares for increasing forecast accuracy.
• Spyros Makridakis, Fotios Petropoulos, and Yanfei Kang examine the impact of large language models (like ChatGPT) on forecasting.
• Gordon Reikard considers the issue of test design in comparing AI and nonlinear regression models.

Minitutorials We pull from Steve Morlidge’s The Little (Illustrated) Book of Operational Forecasting for our two minitutorials. The style of Steve’s book – concise sections of text focused on a single topic, and always with an accompanying illustration – was an inspiration for the Minitutorial concept.