2014年-BIS国际清算银行_A_parsimonious_approach_to_incorporating_economic_information_in_measures_of_potential_output_41页_516kb

Summary of "A Parsimonious Approach to Incorporating Economic Information in Measures of Potential Output"

Core Content

This working paper by Claudio Borio, Piti Disyatat, and Mikael Juselius examines the challenges of estimating potential output and output gaps using economic information, particularly through the lens of the Phillips curve. The authors argue that conventional methods of incorporating structural economic relationships into potential output estimation are opaque and highly sensitive to specification errors, which can lead to unreliable real-time estimates. They propose a more transparent and robust alternative approach.

Main Points

• Potential Output and Output Gaps: These are not directly observable and must be estimated from data. They represent the difference between actual output and the level of output that is sustainable given available resources.

• Conventional Estimation Methods: Most methods rely on the Hodrick-Prescott (HP) filter, which is a univariate approach that minimizes a loss function involving the variance of the output gap and the smoothness of the potential output series. However, this method is often criticized for its poor real-time performance and susceptibility to revisions.

• Structural Relationships: The most common structural relationship used is the Phillips curve, which links inflation to the output gap. The authors show that using the Phillips curve can either introduce a trend in the output gap or have little effect, depending on the weight assigned to the relationship.

• Limitations of Conventional Approaches:

  • The results are sensitive to auxiliary assumptions and specification choices.
  • The complexity increases with the number of equations, making the approach less transparent.
  • The "pile-up problem" arises when estimating the scaling factor, leading to over-smoothing and underestimation of mean reversion in the output gap.

• Alternative Approach: The paper introduces a parsimonious multivariate filter that directly evaluates the ability of observable economic variables to explain cyclical output fluctuations. This approach is:

  • Simpler: It avoids the need for complex system-based models.
  • More Transparent: It allows for clearer assessment of the contribution of conditioning variables.
  • Robust: It improves real-time performance by anchoring the output gap to variables with stable means.

• Key Variables for Estimation: The authors find that variables such as credit growth and unemployment rate are particularly useful in generating economically plausible and robust output gap estimates. Credit growth, in particular, is shown to be a strong proxy for the financial cycle.

• Real-Time Performance: The alternative approach significantly improves real-time performance by reducing the need for frequent revisions. It also avoids the end-point problem that plagues purely statistical methods.

Key Findings

• Phillips Curve Issues: The Phillips curve, while popular, is problematic when applied to US data due to the downward trend in inflation. If the curve is given too much weight, the output gap will inherit this trend, which is not economically meaningful. If given too little weight, it will have minimal impact on the estimates.

• Dynamic HP Filter: A dynamic version of the HP filter that includes an AR(1) process for the output gap shows a slight improvement in precision, but the main issue remains the sensitivity to scaling factors and specification choices.

• Multivariate HP Filter: This method, which incorporates additional economic relationships, introduces more complexity and sensitivity to auxiliary assumptions. It can lead to misleading results if not carefully specified.

• Parsimonious Filter: The proposed method is less vulnerable to misspecification and provides a clearer link between the output gap and economic variables. It allows for the inclusion of multiple conditioning variables without increasing the dimensionality of the system.

• Empirical Results: When applied to US data, the parsimonious filter generates output gaps that are both economically meaningful and perform well in real time. The use of financial cycle proxies and the raw unemployment rate is particularly effective.

Conclusion

The paper concludes that while incorporating economic information into potential output estimates is desirable, the conventional methods are often too complex and opaque. A more parsimonious, transparent, and robust approach is needed to improve the reliability and interpretability of output gap estimates. The proposed method offers a viable alternative that can be applied in real time with greater confidence in its results.

Key Information

• JEL Classification: E10, E40, E44, E47, E52, E60.
• Keywords: Potential output, output gap, Phillips curve, financial cycle.
• Methodology: The paper uses a state-space framework with Kalman filter and introduces a parsimonious multivariate filter.
• Data Used: US quarterly real GDP data from 1980q1 to 2012q4.
• Main Variables: Credit growth, property prices, and unemployment rate are identified as particularly useful proxies for the financial cycle and output gaps.

Structure of the Paper

1. Introduction

   • Discusses the importance of potential output and output gap estimation.
   • Highlights the limitations of conventional methods and introduces the alternative approach.

2. Univariate Kalman-filter estimates of potential output

   • Describes the static and dynamic versions of the HP filter.
   • Explains the role of the scaling factor and the pile-up problem.

3. Multivariate estimates: Adding structural economic relationships

   • Discusses the use of the Phillips curve as a structural relationship.
   • Highlights the issues with specification and sensitivity to auxiliary assumptions.

4. An alternative approach to embedding economic information

   • Introduces the parsimonious multivariate filter.
   • Explains the benefits of simplicity, transparency, and robustness.

5. Real-time performance

   • Compares the real-time performance of different methods.
   • Shows how the proposed approach improves performance and reduces revisions.

6. Conclusion

   • Summarizes the key findings and recommends the alternative approach for more reliable output gap estimation.