2025-09-01-IMF-新凯恩斯宏观模型的符号限制_准不可知_识别过程的结果（英）页_20页_1mb

Sign Restrictions with a New-Keynesian Macro Model: Results from a Quasi-Agnostic Identification Procedure

Authors: Gregorio Impavido
Year: 2025
Working Paper: IMF WP/25/162

Abstract

This paper introduces a "quasi-agnostic" sign restriction method for identifying structural shocks in SVAR models. It argues that low acceptance rates in traditional agnostic procedures often stem from inefficiency in utilizing ex-ante priors on macro variable responses, not necessarily model misspecification. The proposed procedure improves acceptance rates by progressively refining structural parameters with prior sign knowledge.

Background and Objectives

• Structural VAR (SVAR) models are used to identify shocks, often via parametric (e.g., zero contemporaneous restrictions) or sign-based methods.
• Parametric restrictions are replaced here with sign restrictions, which specify direction but not magnitude of shock effects.
• The study uses a three-variable New-Keynesian SVAR model (US output gap, inflation, policy rate) to address issues like the "price puzzle" where monetary shocks positively affect inflation.
• Objectives: Improve shock identification through a novel identification procedure and highlight implications for model specification.

Baseline Model and Findings

• A recursively identified SVAR with a lag order selected via information criteria.
• Key findings: Shocks exhibit persistence (e.g., demand and supply shocks last up to 20 periods); the model reveals a price puzzle, with monetary shocks positively impacting inflation.
• Traditional parametric restrictions resolve puzzles but are criticized for rigid assumptions; sign restrictions offer flexibility but face low acceptance rates.

Agnostic Sign Restrictions

• Based on Ouliaris and Pagan (2016) SRC procedure, which randomly assigns signs to coefficients.
• Results: Very low acceptance rates (e.g., <1%), indicating that this method may not fully leverage economic priors, potentially misleading conclusions about model consistency.

Quasi-Agnostic Identification Procedure

• An adaptive grid search approach that progressively refines structural parameters using ex-ante sign priors.
• Method: Combines insights from multiple horizon restrictions to narrow plausible parameter ranges, increasing acceptance rates.
• Benefits: Achieves higher acceptance rates by focusing on economically plausible structures, addressing low rates from agnostic methods.

Conclusions and Implications

• The quasi-agnostic procedure enhances SVAR identification, allowing broader exploration consistent with sign restrictions.
• Future research needed: Extension to higher-dimensional models with ex-ante parameter restrictions.
• Caveats: While improving rates, the method still may not fully resolve issues of model misspecification.