[Paper Review] Corrected Forecast Combinations

Chu-An Liu, Andrey L. Vasnev|arXiv (Cornell University)|Jan 15, 2026

Monetary Policy and Economic Impact0 citations

TL;DR

The paper proposes correcting forecast combinations when original combined forecast errors are serially dependent, showing that a simple correction using a fraction of the previous error can substantially improve forecast accuracy; the approach is formalized within a conditional-risk framework and linked to GLS-based weight estimation.

ABSTRACT

This paper proposes corrected forecast combinations when the original combined forecast errors are serially dependent. Motivated by the classic Bates and Granger (1969) example, we show that combined forecast errors can be strongly autocorrelated and that a simple correction--adding a fraction of the previous combined error to the next-period combined forecast--can deliver sizable improvements in forecast accuracy, often exceeding the original gains from combining. We formalize the approach within the conditional risk framework of Gibbs and Vasnev (2024), in which the combined error decomposes into a predictable component (measurable at the forecast origin) and an innovation. We then link this correction to efficient estimation of combination weights under time-series dependence via GLS, allowing joint estimation of weights and an error-covariance structure. Using the U.S. Survey of Professional Forecasters for major macroeconomic indices across various subsamples (including pre and post-2000, GFC, and COVID), we find that a parsimonious correction of the mean forecast with a coefficient around 0.5 is a robust starting point and often yields material improvements in forecast accuracy. For optimal-weight forecasts, the correction substantially mitigates the forecast combination puzzle by turning poorly performing out-of-sample optimal-weight combinations into competitive forecasts.

Motivation & Objective

• Motivate and formalize the problem of serially dependent errors in forecast combinations.
• Introduce a simple mean-forecast correction and extend to optimal-weight forecasts.
• Embed the corrections within the conditional-risk framework of Gibbs and Vasnev (2024).
• Show empirical gains using the U.S. SPF data across multiple macro indicators and periods.

Proposed method

• Define corrected forecasts by adding a correction term b_T to the equal- or optimally-weighted forecast (e.g., f_CEW = f_EW + b_T_EW).
• Show that the optimal correction b_T equals the conditional mean of the forecast error, reducing both conditional and unconditional MSE.
• Derive a one-step GLS approach that jointly estimates weights and correction terms under time-series dependence.
• Generalize to a GLS framework with a known or estimated error covariance matrix Omega(γ).
• Demonstrate a two-step and a one-step (GLS) procedure for combining and correcting forecasts.

Figure 1 : Improvements from combination (EW) and correction (CEW) in the motivating example from Bates and Granger ( 1969 ) . Points ES and BJ correspond to the original Brown’s exponential smoothing and Box-Jenkins forecasts.

Experimental results

Research questions

• RQ1 Can a simple correction based on previous errors reduce the mean squared forecast error of combined forecasts with autocorrelated errors?
• RQ2 Does incorporating a correction term into optimal (BG) forecast weights improve out-of-sample forecast performance under serial dependence?
• RQ3 How does a GLS-based joint estimation of weights and correction factors perform relative to two-step procedures?
• RQ4 What are the empirical gains from corrected mean and corrected optimal forecasts using SPF macro forecasts across different periods (pre/post-2000, GFC, COVID)?

Key findings

• A simple correction of the mean forecast with a coefficient around 0.5 yields robust improvements in forecast accuracy across SPF macro indicators.
• Correcting the mean forecast can yield larger MSFE reductions than the original forecast combination in some settings.
• Corrected optimal forecasts with modest corrections (e.g., gamma around 0.5–0.7) outperform the uncorrected mean and often beat the two-step BG approach.
• Joint GLS estimation of weights and correction factors further improves performance and nearly matches the corrected mean forecast in some cases.
• Across SPF data and periods, a fixed correction around 0.5 provides strong, robust improvements, while historical or optimized corrections offer additional gains in certain windows.

(a) UNEMP