📈 economics

A Structural Matrix Autoregression Framework for International Spillovers

This paper introduces a Bayesian Structural Matrix Autoregression (BSMAR) framework that leverages the matrix structure of international macroeconomic data to parsimoniously model large multi-country systems, enabling the identification of heterogeneous cross-border shock transmissions where demand shocks are found to drive spillovers more significantly than supply shocks.

Original authors: Ignacio Moreira Lara, Jan Prüser, Christoph Hanck

Published 2026-08-04

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Ignacio Moreira Lara, Jan Prüser, Christoph Hanck

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the global economy as a massive, chaotic orchestra where every country is a musician playing their own instrument. Sometimes, a trumpet player in Paris sneezes, and suddenly the drummer in Tokyo loses the beat. Economists have long tried to figure out exactly how these musical hiccups travel from one musician to another. They use a special tool called a "Structural Vector Autoregression" (SVAR), which is like a super-advanced detective kit. This kit helps them listen to the music and ask: "Was that sneeze caused by a cold (a supply shock) or did the musician just get too excited (a demand shock)?" And more importantly, "How did that sneeze mess up the drummer's rhythm?"

The problem is that when you try to listen to 15 different countries at once, the math gets so huge and messy that even the smartest computers get dizzy. It's like trying to solve a puzzle where every piece is connected to every other piece in a billion different ways. Previous methods either had to ignore most of the countries or make so many guesses that the answers weren't very reliable. But what if there was a way to organize the puzzle so the pieces fit together neatly, revealing exactly who is sneezing on whom?

This paper introduces a clever new framework called the "Bayesian Structural Matrix Autoregression" (BSMAR). Think of it as a magic lens that turns a tangled ball of yarn into a neat, organized grid. Instead of treating the 15 countries and their economic data as one giant, confusing blob, the authors arrange the data into a matrix—a grid where one side lists the economic variables (like GDP and prices) and the other lists the countries. This simple trick drastically shrinks the size of the math problem, making it possible to solve without the computer crashing.

The authors used this new tool to look at quarterly data from 15 major economies, including the US, Germany, Japan, and others, covering the years 1998 to 2019. They specifically wanted to see how "demand shocks" (like people suddenly wanting to buy more stuff) and "supply shocks" (like factories suddenly making less stuff) spread across borders. Their findings suggest that demand shocks are the real troublemakers when it comes to international spillovers. When the US sneezes a demand shock, it doesn't just stay there; it travels fast and loud to other countries, often causing them to cough up inflation. In contrast, supply shocks tend to stay more local.

When they tested their model against the real-world chaos of the pandemic and the subsequent inflation surge (2020–2024), the results pointed a finger squarely at demand. The data suggests that the recent rise in inflation wasn't mostly because factories stopped working (supply issues), but because demand for goods and services went wild, and that excitement spread rapidly from the US to the rest of the world. The US emerged as the loudest "sneezer" in the orchestra, acting as a major transmitter of these shocks. While the model shows that every country reacts differently, the overall story is clear: in our interconnected world, when one big economy gets excited about buying things, the whole orchestra feels the rhythm change.

Technical Summary: A Structural Matrix Autoregression Framework for International Spillovers

Problem Statement
The paper addresses the fundamental challenge of extending Structural Vector Autoregressions (SVARs) to large, multi-country systems. While SVARs are standard tools for identifying macroeconomic shocks and their dynamic effects, applying them to international data faces a "curse of dimensionality." As the number of countries increases, the system's dimension and the number of unknown parameters grow quadratically, leading to severe overparameterization, prohibitive computational costs, and an impractical proliferation of identifying restrictions. Existing alternatives, such as Dynamic Factor Models (DFMs) and Global VARs (GVARs), offer computational tractability but impose restrictive assumptions: DFMs typically focus on common shocks rather than country-specific structural ones, while GVARs often assume uncorrelated contemporaneous cross-country innovations, limiting the identification of structural spillovers.

Methodology
The authors propose a Bayesian Structural Matrix Autoregression (BSMAR) framework that exploits the natural matrix structure of international macroeconomic data, where one dimension represents economic variables and the other represents countries.

Model Structure: Instead of vectorizing the data into a single high-dimensional system, the model treats the data as an $n \times k$ matrix $Y_t$ (where $n$ is variables and $k$ is countries). The reduced-form dynamics follow a Matrix Autoregressive (MAR) process:
$Y_t = \sum_{\ell=1}^p A_\ell Y_{t-\ell} B'_\ell + U_t$
Here, $A_\ell$ captures dynamic interactions among economic variables, and $B_\ell$ captures dynamic interactions across countries. The error term $U_t$ follows a matrix normal distribution, separating the covariance structure into row ( $\Sigma_r$ ) and column ( $\Sigma_c$ ) components.
Structural Identification: The core innovation lies in the structural representation. The structural impact matrix $B_0$ is restricted to a Kronecker product decomposition:
$B_0 = B_c \otimes B_r$
- $B_r$ ( $n \times n$ ) captures contemporaneous relationships among variables within a unit (e.g., the supply/demand relationship between GDP and CPI).
- $B_c$ ( $k \times k$ ) captures contemporaneous transmission mechanisms across countries (spillovers).
  This decomposition reduces the number of structural parameters from $O((nk)^2)$ in an unrestricted SVAR to $O(n^2 + k^2)$ , drastically mitigating overparameterization.
Identification Strategy: The framework accommodates a broad class of identifying restrictions (zero, sign, magnitude/ranking, and soft restrictions) applied to the decomposed matrices.
- Hard Restrictions: Zero restrictions (e.g., small economies do not contemporaneously affect large ones) and sign restrictions are imposed directly.
- Soft Restrictions: The authors employ soft sign restrictions (Gaussian priors centered on a specific sign) to allow for deviations if the likelihood strongly contradicts the prior, avoiding the rigidity of hard constraints.
- Normalization: Scale indeterminacy in the Kronecker product is resolved by fixing specific diagonal elements to unity.
Bayesian Estimation: A Gibbs sampler is developed to draw from the joint posterior distribution.
- Autoregressive coefficients ( $A, B$ ) and shrinkage parameters ( $\kappa$ ) are updated via conjugate Gaussian and Generalized Inverse Gaussian (GIG) distributions.
- Structural parameters ( $\theta = \text{vec}(B_r), \text{vec}(B_c)$ ) are sampled using Elliptical Slice Sampling (ESS). This algorithm is tailored for targets combining a Gaussian prior with an arbitrary likelihood and handles the complex identified set defined by sign and magnitude restrictions without requiring Metropolis-Hastings tuning.

Key Contributions

Dimensionality Reduction: The paper provides a parsimonious representation that separates variable dependence from country dependence, making the estimation of large structural systems computationally feasible.
Flexible Identification: It extends the literature on matrix-valued time series by developing a structural framework that allows for the identification of contemporaneous structural relationships within countries and spillovers across countries simultaneously.
Algorithmic Innovation: The development of a Bayesian sampling algorithm that integrates Kronecker-structured identification with soft and hard restrictions (including magnitude dominance) offers a practical solution for high-dimensional structural inference.

Empirical Results
The framework is applied to quarterly data from 15 OECD economies (1998Q1–2019Q4) to identify 30 country-specific structural shocks (one supply and one demand shock per economy).

Shock Transmission: The results reveal substantial heterogeneity in international shock transmission. The United States emerges as a primary net transmitter of shocks, while smaller economies are generally net receivers.
Demand vs. Supply: Demand shocks play a more prominent role than supply shocks in generating cross-country spillovers.
Post-Pandemic Inflation: Out-of-sample historical decompositions for the 2020–2024 period indicate that the surge in inflation following the COVID-19 pandemic was primarily driven by demand shocks. In the US, domestic demand shocks were the key contributors initially, with foreign demand shocks becoming more prominent later. The UK exhibited a more varied pattern with significant foreign contributions.
Variance Decomposition: While domestic shocks explain the majority of forecast error variance (FEVD) in most countries, aggregate demand spillovers are confirmed as a critical channel of international transmission.

Significance and Claims
The paper claims to be the first study to identify contemporaneous structural spillovers among such a large group of countries within a unified structural framework. It argues that by exploiting the matrix structure of data, researchers can overcome the trade-off between computational feasibility and the ability to identify country-specific structural spillovers. The findings underscore the central role of demand shocks in driving international macroeconomic fluctuations and the post-pandemic inflation surge, offering a refined structural perspective on global economic interdependence. The authors note that while the current specification assumes separability (a single set of matrices for all countries), future extensions could relax these restrictions for smaller panels, though such flexibility comes at the cost of increased estimation uncertainty in high dimensions.

More like this