Working Paper · January 2026

Global Risk, FX Volatility,
and the Real Economy

A Sign-Restricted SVAR Approach

How does global risk reach a small open economy? I isolate an FX-volatility-centered shock in Korea and trace its transmission to investment and cyclical output.

Korea1990Q2–2025Q3MATLABWorking paper

Request draft Request repository access

Impulse responses of FX volatility, investment growth, and the output-gap proxy to the identified shock — Baseline impulse responses with pointwise wild-bootstrap bands. FX volatility rises sharply on impact; real activity contracts in the short run.

At a glance

A compact empirical design built around identification, inference, and historical validation.

4variables in the baseline system

4 qtrsdirectional FXVOL significance horizon

1997 & 2008largest crisis-era shock realizations

Finding

The identified shock reaches its largest realizations in 1997Q4 and 2008Q4—linking the statistical object to Korea’s two defining external financial crises.

Empirical design

The shock is selected from admissible rotations rather than imposed through a fully recursive ordering.

System

Quarterly VAR

[log(VIX), log(1+FXVOL), dlog(INV), Y]

Y is the HP-filtered cyclical component of a pseudo log-level GDP index in the baseline specification.

Identification

Targeted sign restrictions

Short-run restrictions define the feasible set. A max-share criterion selects the rotation that concentrates forecast-error variance in FX volatility.

Selection discipline

Purity penalty

An auxiliary penalty discourages contemporaneous mixing with real-activity innovations; alternative weights are treated as robustness exercises.

Inference

Bootstrap re-identification

Each wild-bootstrap replication re-estimates the VAR and resolves the identification problem, carrying selection uncertainty into the confidence bands.

Historical validation

The identified series is quiet in ordinary periods and becomes economically consequential when global risk deteriorates abruptly.

Standardized identified shock series from 1990 to 2025 — The largest realizations occur around the 1997-98 Asian financial crisis and the 2008-09 global financial crisis.

Historical decomposition of investment and the output-gap proxy — Historical decompositions assign substantial contractionary contributions to the identified shock during the two principal crisis windows.

Replication package

The repository is private while the paper is under revision. Its structure is designed to reproduce figures and tables without redistributing third-party data.

Run

`run_main_analysis.m`

Set the quarterly CSV path, move to the repository root, and run the main MATLAB entry point. The implementation is designed not to require additional toolboxes.

Outputs

Figures, tables, and logs

outputs/figures/
outputs/tables/
outputs/logs/

Boundary

Data are not redistributed

Users obtain the underlying series from ECOS, FRED, and the EPU database, then construct the quarterly analysis file locally.

Replication detailsQuick start, required inputs, and measurement choices

Quick start

Obtain repository access.
Set MATLAB’s working directory to the repository root.
Point dataFile to the locally constructed quarterly CSV.
Run run_main_analysis.

Required quarterly inputs

VariableDefinition

VIXGlobal risk proxy, entered as log(VIX)

FXVOLQuarterly FX-volatility proxy, entered as log(1+FXVOL)

INVQuarterly investment series used to construct dlogINV

GDPgQuarterly GDP growth used to construct the cyclical output measure

FX volatility

Daily KRW/USD log returns are fitted with a univariate GARCH model. Conditional volatility is averaged within each quarter and entered as log(1+FXVOL).

Cyclical output

Quarterly GDP growth is cumulated into a pseudo log-level index. The baseline measure is its HP-filtered cyclical component with lambda = 1600; GDP growth is retained as a robustness alternative.