Â鶹Íø

QUEEN’S BUSINESS SCHOOL INFORMATION TECHNOOGY, ANALYTICS & OPERATIONS (ITAO) SEMINAR SERIES

Friday 17^th April

11:30am

“Judgmental Forecasting, Herding, and a ML-optimised consensus prediction”

Durham University

Bio

Professor Konstantinos Nikolopoulos is Professor of Business Information Systems and Analytics at Durham University Business School and serves as Associate Dean (Internationalisation). He is Chair in Business Information Systems & Analytics, Co Director of the Institute for Hazard, Risk and Resilience (IHRR), and Director of the IHRR Forecasting Laboratory. He is also Co Editor in Chief of the Journal of the Operational Research Society. His research focuses on forecasting and predictive analytics, judgmental forecasting, herding behaviour, and machine learning optimised consensus prediction, with extensive work using large scale international forecasting datasets.

Abstract

Drawing on the mechanisms of economist behavior and the nuances surrounding the role of general and task-specific experience, we revisit the relationship between the experience of macroeconomic forecasters and their herding behavior and accuracy. We employ a novel set of over 2.3 million forecasts by 4,574 economists, across 80 countries and decompose each forecaster’s experience into general-prior and indicator-specific. We find that forecaster experience presents a non-linear relationship with their herding and accuracy levels. Perhaps more importantly, the interplay between general experience and indicator expertise is crucial: economists with limited indicator-specific expertise initially deviate from the consensus of their peers, then transition to herding, before starting to deviate again after several years. General-prior experience influences this behavior, by accelerating herding across all levels of expertise. With respect to forecast accuracy, while general-prior experience is not consistently beneficial, it can offset the relative absence of indicator-specific expertise. We propose a procedure which generates debiased consensus forecasts and find them to outperform the actual consensus for several key indicators.

QBS Student Hub, Executive Lounge 01.015

Teams

Meeting ID: 335 653 576 211 488

Passcode: TM69nW7P