2008年-ECB欧洲央行_New_procedure_for_constructing_ECB_staff_projection_ranges_3页_189kb

ECB Staff Projection Ranges: New Procedure Summary

A. Introduction

The European Central Bank (ECB) and Eurosystem staff macroeconomic projections are presented as ranges to reflect the inherent uncertainty in forecasting. Since their initial publication in December 2000, these ranges have been based on historical projection errors, specifically using a width equal to twice the mean absolute projection error. However, this method has not been updated since then.

To address this, the ECB introduced a new procedure for constructing projection ranges starting from the September 2008 staff projection. This change is purely technical and aims to enhance the accuracy and robustness of the projection ranges by incorporating both historical data and more current economic information. The format of the projections in the ECB's Monthly Bulletin remains unchanged, continuing to present them as ranges to communicate uncertainty effectively.

B. Description of the New Procedure

The new procedure utilizes Bayesian Vector Auto-Regression (BVAR) models to generate projection ranges. These models include:

• One model that forecasts real GDP growth and HICP inflation.
• Another model that forecasts the growth of GDP components.

The BVAR models are estimated using standard Bayesian techniques and are conditional on the same exogenous variables used in the ECB staff projections, which include:

• The price of oil.
• The nominal effective exchange rate.
• Short and long-term interest rates.
• Foreign demand.

The process involves generating a forecast distribution (predictive density) for each variable and then "twisting" this distribution to align with the ECB staff projection. This method, described in Robertson et al. (2005), re-weights the original predictive density to produce a new density that is consistent with the projection outcome while maintaining a close resemblance to the original in a certain metric.

From this adjusted density, a 75% probability interval is derived. This interval is calibrated to maintain continuity with the previous projection-error-based ranges. The new BVAR-based ranges are:

• Time-dependent: They vary across different projection exercises.
• Horizon-dependent: The uncertainty increases with longer time horizons, resulting in wider ranges.
• Volatility-sensitive: Variables with higher volatility, such as the growth of gross fixed capital formation, tend to have wider ranges due to more dispersed predictive densities.

C. Key Features of the New Procedure

• Technical Update: The new procedure is a technical improvement over the previous method.
• Model-Based: It is grounded in BVAR models, which provide a more rigorous probabilistic interpretation.
• Consistency with Previous Ranges: The new ranges are calibrated to be similar in width to the previous ones, ensuring continuity.
• Improved Accuracy: The use of predictive densities and re-weighting techniques leads to more accurate and reliable ranges, especially for one-year-ahead HICP inflation projections.

D. Visual Comparison

The ECB has provided charts comparing the 75% BVAR-based ranges with the previous projection-error-based ranges for HICP inflation and real GDP growth across all projection exercises from December 2003 to June 2008. These visual comparisons show that the new ranges are stable over time, even though they are computed in real time and updated sequentially. For the one-year-ahead HICP inflation projections, the new ranges are narrower than the previous ones, which is consistent with empirical forecasting literature and the experience of other forecasters.

E. Conclusion

The new procedure for constructing ECB staff projection ranges enhances the method by integrating advanced statistical techniques and more current data. It maintains the clarity and continuity of the previous ranges while offering a more robust and probabilistically sound approach to conveying uncertainty in macroeconomic forecasts.