[Paper Review] Efficient Estimation of State-Space Mixed-Frequency VARs: A Precision-Based Approach

Joshua C. C. Chan, Aubrey Poon+1 more|arXiv (Cornell University)|Dec 21, 2021

Monetary Policy and Economic Impact4 citations

TL;DR

This paper proposes two precision-based samplers for efficiently estimating missing low-frequency observations in state-space mixed-frequency VARs, leveraging block-banded precision matrices to improve numerical accuracy and computational speed over standard Kalman filtering. The method enables fast, joint drawing of missing values with exact or approximate linear constraints, significantly enhancing estimation efficiency in large-scale macroeconomic models.

ABSTRACT

State-space mixed-frequency vector autoregressions are now widely used for nowcasting. Despite their popularity, estimating such models can be computationally intensive, especially for large systems with stochastic volatility. To tackle the computational challenges, we propose two novel precision-based samplers to draw the missing observations of the low-frequency variables in these models, building on recent advances in the band and sparse matrix algorithms for state-space models. We show via a simulation study that the proposed methods are more numerically accurate and computationally efficient compared to standard Kalman-filter based methods. We demonstrate how the proposed method can be applied in two empirical macroeconomic applications: estimating the monthly output gap and studying the response of GDP to a monetary policy shock at the monthly frequency. Results from these two empirical applications highlight the importance of incorporating high-frequency indicators in macroeconomic models.

Motivation & Objective

• To address the high computational cost of estimating large state-space mixed-frequency VARs with stochastic volatility.
• To develop a numerically accurate and computationally efficient alternative to standard Kalman-filter-based methods for imputing missing low-frequency observations.
• To enable the incorporation of high-frequency indicators into macroeconomic models for improved nowcasting and policy analysis.
• To extend precision-based sampling techniques to handle missing observations with linear inter-temporal constraints, crucial for mixed-frequency modeling.

Proposed method

• Derives the joint Gaussian conditional distribution of missing low-frequency observations given high-frequency data and model parameters.
• Exploits the block-banded structure of the precision matrix to enable efficient sampling using band matrix algorithms.
• Implements a single-step precision-based sampler that draws all missing values simultaneously, avoiding iterative filtering steps.
• Incorporates both hard and soft linear constraints to ensure interpolated low-frequency values align with observed quarterly totals.
• Applies the method within a Bayesian framework using Gibbs sampling, with the precision-based sampler replacing standard Kalman filtering for missing data imputation.
• Utilizes recent advances in sparse and band matrix algorithms to maintain computational efficiency even in high-dimensional systems.

Experimental results

Research questions

• RQ1 How can the computational burden of estimating large state-space mixed-frequency VARs be reduced without sacrificing numerical accuracy?
• RQ2 Can precision-based sampling methods outperform standard Kalman filtering in terms of speed and accuracy when imputing missing low-frequency observations?
• RQ3 To what extent can linear inter-temporal constraints be effectively imposed during the imputation of missing observations in mixed-frequency models?
• RQ4 How does the proposed method perform in real-world macroeconomic applications such as output gap estimation and monetary policy shock analysis?

Key findings

• The proposed precision-based samplers are significantly more computationally efficient than standard Kalman-filter-based methods, especially in models with complex lag structures.
• The simulation study confirms that the new method achieves higher numerical accuracy in estimating missing low-frequency values compared to traditional filtering approaches.
• In the empirical application on the output gap, the model successfully produces monthly estimates of real GDP using high-frequency indicators, demonstrating improved nowcasting performance.
• The monetary policy shock analysis reveals a contemporaneous response of approximately five basis points in the federal funds rate to a one-standard-deviation surprise in real GDP, highlighting the model's policy relevance.
• The cumulative response of real GDP to a monetary policy shock is zero, consistent with the classical dichotomy, validating the model's structural consistency.
• The inclusion of real GDP rather than industrial production as a real economic indicator yields more economically meaningful policy responses, underscoring the importance of high-frequency data integration.