ECB introduces robust method for integrating external forecast data

The new 'parametric tilting' methodology, developed by ECB researchers, offers a significant advancement in econometric forecasting by providing a robust way to incorporate external information into model-based density forecasts.

Unlike its predecessor, entropic tilting, this novel approach leverages the flexibility of the skew-T distribution, which is adept at capturing complex macroeconomic time series features like asymmetry and heavy tails.

By minimizing the Kullback-Leibler divergence between the target and model-based distributions, parametric tilting ensures more reliable and numerically stable results.

This method effectively prevents the generation of unrealistic or unstable distributions, such as multimodal or degenerate outcomes, which have historically limited the practical applicability of traditional entropic tilting, especially when external data significantly deviates from model predictions or when original data is non-Gaussian.

The authors demonstrate that this approach consistently outperforms conventional methods, making it a valuable tool for policymakers and researchers seeking dependable forecasts in uncertain economic environments.

Traditional entropic tilting, a method introduced by Robertson, Tallman, and Whiteman in 2005, has been a standard approach for economists and policymakers to integrate external information into model-based predictions.

This technique modifies probability distributions to align with new constraints, such as high-frequency data or expert opinions, by minimizing Kullback-Leibler divergence.

While conceptually simple, entropic tilting frequently encounters significant practical challenges.

It can produce multi-modal or degenerate distributions, particularly when the external information substantially differs from the model's predictions or if the underlying data is non-Gaussian.

Such issues lead to unrealistic and unstable forecasts, as observed during the integration of COVID-19 period data, limiting its reliability for critical decision-making.

The inherent difficulties arise from the method's sensitivity to the divergence between the original and target distributions, often resulting in a low Effective Sample Size and reduced efficiency in reweighted samples.