Exchange rate response to economic policy uncertainty: evidence beyond asymmetry

The study investigates how economic policy uncertainty (EPU) influences exchange rates, motivated by theoretical channels whereby policy uncertainty affects FDI, trade flows, and interest rate decisions, which in turn alter currency demand and exchange rate dynamics. Existing empirical work shows EPU’s relevance for macroeconomic and financial variables, and some evidence links EPU to exchange rates and their volatility. However, prior research rarely distinguishes between positive versus negative EPU changes or between minor and major shocks. Given evidence of nonlinearity and asymmetry in macro–financial relationships, the authors aim to test whether exchange rates in G7 economies respond asymmetrically to varying magnitudes and signs of EPU shocks, and whether global EPU (GEPU) also matters. The paper seeks to provide policy-relevant insights for central banks and policymakers in highly influential G7 economies that account for large shares of global GDP and wealth.

Two strands of literature underpin the study. First, EPU has documented effects on macro variables (inflation, investment, consumption, unemployment) and asset classes (commodities, gold, bonds, equities), often exhibiting nonlinearities (e.g., Bloom 2009; Brogaard and Detzel 2015; Aastveit et al. 2017; Caggiano et al. 2017). Second, studies on exchange rates show EPU impacts volatility and levels: Krol (2014) finds EPU more relevant than general uncertainty for exchange rate volatility; Kido (2016, 2018) documents spillovers from US EPU to global FX and financial markets; Bartsch (2019) shows stronger daily-frequency effects; Yin et al. (2017) and Chen et al. (2019) find heterogeneous/quantile-dependent relationships. Recent works employ nonlinear and quantile methods and consider other risk factors (oil prices, geopolitical risk), but typically do not separate minor versus major shocks of differing sign. This gap motivates the current focus on multiple asymmetric thresholds in EPU and GEPU effects on exchange rates.

Data: Monthly data for G7 countries (Canada, France, Germany, Italy, Japan, UK, US) from January 1998 to January 2021. Dependent variable: real effective exchange rate (REER). Key explanatory variables: country-specific EPU (Baker et al., 2016) and, for robustness, global EPU adjusted for PPP (GEPU-PPP). Controls: industrial production index (IPI) and consumer price index (CPI). REER, IPI, CPI are from IMF’s IFS; EPU and GEPU from policyuncertainty.com. All series are in natural logs. Preliminary analysis includes descriptive statistics and unit root testing (ADF, KPSS) to ensure variables are I(0)/I(1) and not I(2). Models: (1) Nonlinear ARDL (NARDL; Shin et al., 2014) to capture asymmetry by decomposing EPU into positive and negative partial sums, allowing short- and long-run asymmetries. Cointegration is tested via bounds testing (Pesaran et al., 2001). Wald tests assess short- and long-run symmetry restrictions. (2) Multiple Asymmetric Threshold NARDL (MATNARDL; Uche et al., 2022a), extending Pal and Mitra’s MTNARDL, decomposes EPU into three positive and three negative partial sum series using 30th and 70th percentile thresholds to distinguish minor, moderate, and major shocks of each sign. Bounds testing evaluates cointegration; Wald tests examine asymmetries across thresholds. (3) Granger Causality in Quantiles (GCQ; Troster, 2018) assesses directionality of effects across the conditional distribution of REER and EPU/GEPU, capturing quantile-dependent causality. Diagnostics: Model specification (Ramsey RESET), serial correlation (LM), stability (CUSUM, CUSUMQ), cointegration via an error-correction term (ECM), and goodness-of-fit (adjusted R-squared). Lag selection uses AIC.

• Descriptive statistics show most series are positively skewed with kurtosis > 3; Jarque–Bera tests reject normality. Unit root tests (ADF, KPSS) confirm variables are I(0)/I(1), not I(2).
• Cointegration (bounds tests): Under NARDL, cointegration is detected for Canada (F ≈ 4.211), Japan (≈ 5.521), and UK (≈ 7.711), but not consistently for others. Under MATNARDL, cointegration holds across all G7 countries (e.g., US ≈ 12.12; Italy ≈ 12.01), indicating the extended model’s superiority in detecting long-run relationships.
• NARDL asymmetry (Wald tests): Asymmetry is found in Canada and Japan (both short- and long-run), and in the UK (short-run only). France, Germany, Italy, and the US show symmetry in NARDL.
• NARDL coefficients: In the short run, positive EPU shocks reduce REER (i.e., depreciate currency against the basket) significantly in Canada, Japan, and the UK; negative EPU shocks are generally insignificant. In the long run, positive EPU shocks significantly affect REER in Canada and Japan (negative sign in reported table entries for EPU+ on REER), while in the UK both positive and negative EPU shocks are significant and negative; effects in France, Germany, Italy, and the US are largely insignificant.
• MATNARDL results: Both short- and long-run asymmetries are supported across all G7. Long-run coefficients indicate that positive EPU shocks across thresholds (EPU+Q1, Q2, Q3) typically have significant negative effects on REER in most countries, while negative EPU shocks (EPU−Q1, Q2, Q3) are mostly insignificant. This indicates asymmetric sensitivity where increases in policy uncertainty exert stronger exchange rate impacts than decreases.
• Robustness with GEPU (Appendix): Bounds tests show NARDL cointegration in several countries and MATNARDL cointegration across all G7 when using GEPU. Symmetry tests and coefficient patterns broadly mirror country-level EPU results, reinforcing that global uncertainty shocks also matter for G7 exchange rates.
• GCQ: Granger causality is quantile-dependent. Bidirectional and unidirectional causal links between EPU/GEPU and REER vary across quantiles, with stronger causality often observed in mid-to-upper quantiles, indicating state-dependent dynamics. Overall, MATNARDL captures nuanced threshold-dependent asymmetries missed by standard NARDL, and positive EPU shocks are the primary drivers of exchange rate adjustments across G7 economies.

The research question—whether exchange rates respond asymmetrically to the sign and magnitude of EPU shocks—receives affirmative evidence. Standard NARDL detects asymmetry in only a subset of G7, while MATNARDL reveals pervasive asymmetry, indicating that modeling multiple thresholds is essential to uncover the full dynamics. The stronger and more prevalent impacts of positive (rising) EPU shocks relative to negative (declining) shocks align with risk-premium and real-option mechanisms: investors react more to heightened uncertainty, reducing cross-border investment and trade, which pressures exchange rates. Quantile-based causality results further show that these relationships are state-dependent, being more pronounced in certain parts of the REER distribution. For policy, the findings imply that central banks and fiscal authorities should account for the disproportionate and threshold-dependent influence of rising policy uncertainty on currencies, and design communication and stabilization tools accordingly. The superiority of MATNARDL suggests that overlooking threshold effects may underestimate true EPU impacts, potentially leading to suboptimal exchange rate management.

The paper advances the literature by: (1) demonstrating that exchange rates in G7 economies respond asymmetrically to EPU, especially to positive (rising) uncertainty; (2) introducing the MATNARDL framework in this context, which disaggregates shocks by sign and magnitude, revealing cointegration and asymmetries across all G7; (3) confirming robustness using global EPU (GEPU) and showing quantile-dependent causality via GCQ. Policy-wise, ignoring the differential effects of minor versus major and positive versus negative policy uncertainty shocks may lead to misguided interventions in FX markets. Future research could apply advanced panel methods to capture cross-country spillovers, extend samples to emerging markets, incorporate explicit COVID-19 pandemic dynamics, and explore alternative uncertainty measures and high-frequency data to refine temporal responses.

• Model stability: In NARDL, CUSUM indicates instability for Japan and CUSUMQ for the UK; in MATNARDL, CUSUMQ suggests instability for the UK. Although most diagnostics pass, these cases warrant caution in interpretation.
• Scope: Time-series focus on G7 limits generalizability; cross-sectional interdependencies are not modeled.
• Measures: EPU and GEPU are news-based indices and may not capture all facets of policy uncertainty; GEPU-PPP is used for robustness, but alternative constructions might yield differences.
• Data frequency and period: Monthly data (1998–2021) may miss intramonth dynamics; structural breaks (e.g., COVID-19) are not explicitly modeled beyond their presence in the sample.