Bayesian VAR & Institutional Shocks

Overview

This project studies the transmission of institutional and geopolitical uncertainty shocks using Bayesian vector autoregressive models.

The work combines geopolitical-risk indicators, policy-uncertainty measures and macroeconomic variables to analyze how institutional shocks propagate through economic systems.

The project should be read as an applied Bayesian econometrics prototype for scenario interpretation and institutional shock analysis, not as a production forecasting system.

Problem

Geopolitical events can affect policy uncertainty, macroeconomic expectations, investment decisions and institutional behavior. These effects are dynamic: they unfold over time and may differ across countries, variables and regimes.

The goal of the project was to use a Bayesian time-series framework to study how macroeconomic and institutional variables respond to uncertainty shocks.

Model Idea

The central model is a Bayesian VAR, where a vector of macroeconomic variables is modeled as a function of its own lags:

Y_t = A_1 Y_{t-1} + ... + A_p Y_{t-p} + ε_t

In a Bayesian VAR, prior assumptions are introduced to regularize the model, especially when the number of variables and lags increases relative to the available sample size.

Impulse response functions are then used to study how the system reacts after a institutional or geopolitical uncertainty shock.

Data and Variables

The project used macroeconomic time series and uncertainty indicators related to geopolitical events and economic policy.

• Geopolitical risk indicators.
• Economic policy uncertainty indicators.
• Macroeconomic variables such as output, prices or activity indicators.
• Institutional or policy-relevant indicators such as uncertainty and activity measures.
• Country-level datasets for comparative macro-financial analysis.

Example — Shock Transmission

A typical analysis asks how the system reacts after an increase in geopolitical risk.

Geopolitical-risk shock
        ↓
Bayesian VAR / structural VAR
        ↓
Impulse response functions
        ↓
Dynamic response of macro-financial variables

The focus is not only on whether a variable reacts, but also on the timing, persistence and sign of the response.

Technologies and Methods Used

• MATLAB for model implementation, estimation and impulse-response analysis.
• Bayesian VAR for regularized multivariate time-series modeling.
• Structural VAR logic for shock interpretation and dynamic transmission analysis.
• Minnesota-style prior for shrinkage and stabilization of VAR parameters.
• Impulse Response Functions for tracing the effects of geopolitical-risk shocks over time.
• Macro-financial time-series preprocessing for aligning variables, transformations and sample periods.
• Scenario interpretation for connecting model outputs to risk and policy narratives.

Implemented Elements

• Preparation and alignment of macro-financial time series.
• Construction of VAR input matrices.
• Bayesian VAR estimation with shrinkage priors.
• Shock-response analysis through impulse response functions.
• Comparison of responses across macroeconomic and financial variables.
• Graphical output of dynamic responses after geopolitical-risk shocks.
• Interpretation of results in a macro-financial risk framework.

Outputs

The project produced impulse-response figures and model outputs describing how selected variables respond to geopolitical-risk shocks over time.

These outputs are useful for scenario analysis, risk interpretation and macroeconomic discussion, but they depend on model specification, variable selection and identification assumptions.

Evaluation Limits

The project provides a structured econometric analysis of geopolitical-risk transmission, but several points require caution.

• Identification: impulse responses depend on the assumed shock-identification strategy.
• Sample sensitivity: results may vary across countries, periods and variable transformations.
• Prior sensitivity: Bayesian VAR estimates depend on prior specification and shrinkage strength.
• Model uncertainty: alternative lag lengths, variables or ordering assumptions may change the results.
• Forecasting: the project is primarily a shock-analysis and scenario tool, not a fully validated forecasting engine.

Methodological Note

The value of the project lies in combining Bayesian econometrics with risk interpretation.

Instead of treating geopolitical risk as a qualitative narrative only, the framework translates it into a measurable shock and studies its dynamic transmission through a macro-financial system.

This connects naturally to broader themes in risk management, stress testing, scenario analysis and decision-making under uncertainty.

Modern Extension

A modern version of the project would preserve the Bayesian VAR structure but extend the evaluation and comparison framework.

• Compare Bayesian VAR results with local projections.
• Test robustness across alternative geopolitical-risk and uncertainty indicators.
• Run prior-sensitivity analysis across shrinkage specifications.
• Estimate rolling-window or time-varying parameter versions of the model.
• Compare responses across countries and macro-financial regimes.
• Integrate narrative event analysis with quantitative shock transmission.

Resources

Technical report in preparation.

Code available upon request.

Technical Context

• Litterman (1986), Forecasting with Bayesian Vector Autoregressions — relevant to Minnesota-style shrinkage priors in Bayesian VAR models.
• Sims (1980), Macroeconomics and Reality — relevant as the foundational VAR framework for macroeconomic dynamics.
• Caldara & Iacoviello, Measuring Geopolitical Risk — relevant to the use of geopolitical-risk indicators in macro-financial analysis.
• Kilian & Lütkepohl, Structural Vector Autoregressive Analysis — relevant to structural shock interpretation and impulse-response analysis.