Answers are grounded in this paper and cite their sources.
This article analyzes the evolution and current state of forecast combination methods since the publication of the seminal work by Bates and Granger in 1969. Based on an extensive review of both Russian and international research, it examines the major directions in the development of forecast combination approaches – from classical statistical methods and simple averaging to modern Bayesian and neural network models. Special attention is paid to the challenges of constructing and interpreting combination weighting coefficients, the role of negative weighting coefficients and the constant term, and the use of nonlinear combination forms. The main comparative Table 1 of methods for constructing weighting coefficients is updated, and a comprehensive bibliography in the field of forecast combination is provided. A systematic analysis of the strengths and limitations of existing approaches is conducted, and the case is made for a shift toward adaptive and hybrid forecasting methods that integrate statistical, expert, and machine learning techniques. The paper surveys more than half a century of progress in forecast combination theory and identifies avenues for further improvement in the face of growing data complexity and volatility.
No takes yet. Share an insight, caveat, or question.
Frenkel et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: