Exchange rate response to economic policy uncertainty: evidence beyond asymmetry

Economic uncertainty negatively impacts economic activity, hindering investment decisions due to irreversible costs (Bernanke, 1983). Economic policy uncertainty (EPU) significantly affects international trade, economic sanctions, and macroeconomic variables. Following Bloom (2009), research quantifying EPU's effects on the aggregate economy has gained momentum. Existing literature explores EPU's relationship with macroeconomic variables (inflation, consumption, investment, etc.) and asset classes (commodities, stocks, etc.). However, research focusing specifically on the relationship between EPU and exchange rates remains limited. Theoretically, EPU impacts exchange rates through several channels: uncertainty in government policies affecting FDI (and thus foreign currency payments), EPU's influence on exports and imports (altering foreign currency demand), and policy changes affecting borrowing decisions in domestic versus foreign currencies. While a theoretical link exists, empirical evidence is scarce and often doesn't differentiate the effects of positive and negative EPU changes, or the impact of minor versus major shocks. This study aims to address this gap by utilizing more sophisticated methodologies to analyze the asymmetric impact of domestic and global EPU on exchange rates in G7 countries. The G7 countries were chosen due to their highly industrialized nature, significant contribution to global net worth and GDP, and substantial investment in research and development.

Previous studies have explored the relationship between EPU and exchange rate volatility (Krol, 2014), the spillover effects of US EPU on exchange rates (Kido, 2016, 2018), and the nonlinear relationship between EPU and macroeconomic variables (Makinayeri, 2019). Some studies have utilized quantile regression to investigate the heterogeneous impact of EPU on exchange rates (Yin et al., 2017; Chen et al., 2019). Other recent studies have examined the relationship between EPU and exchange rates using various methodologies such as GARCH models (Bartsch, 2019) and nonlinear ARDL models (Kisswani and Elian, 2021; Sohag et al., 2022; Song et al., 2022). However, a key limitation of this existing literature is the lack of differentiation between the effects of positive and negative changes in EPU, and the absence of distinction between the impact of minor and major shocks on exchange rates. This study addresses these limitations by employing advanced methodologies.

This study uses monthly data from January 1998 to January 2021 for G7 countries (Canada, France, Germany, Italy, Japan, UK, and US). The dependent variable is the real effective exchange rate (REER). The primary independent variable is EPU (Baker et al., 2016), with IPI and CPI as control variables. Global EPU (GEPU-PPP) is also included for robustness. The study uses three main methodologies: 1) Nonlinear ARDL (NARDL, Shin et al., 2014) to differentiate the effects of positive and negative EPU changes; 2) Multiple Asymmetric Threshold Nonlinear ARDL (MATNARDL, Uche et al., 2022a) to analyze the impact of minor to major positive and negative EPU shocks; and 3) Granger Causality in Quantiles (GCQ, Troster, 2018) to examine causality across different quantiles. The NARDL model decomposes EPU into positive and negative shocks. The MATNARDL further divides these shocks into three categories based on thresholds (30th and 70th quantiles). The GCQ test analyzes causality across quantiles of EPU and exchange rates. Diagnostic tests (Ramsey RESET, serial correlation test, CUSUM, CUSUMQ) assess model specification, stability, and goodness of fit. The study tests for cointegration using the F-test (Pesaran et al., 2001) and assesses asymmetry using Wald tests.

Descriptive statistics reveal that most variables are non-normally distributed. Unit root tests (ADF and KPSS) ensure that variables are I(0) or I(1). The bounds testing approach (F-test) reveals long-run cointegration between EPU and REER in Canada, Japan, and the UK using the NARDL model. However, the MATNARDL model shows long-run cointegration for all G7 countries. Wald tests indicate that NARDL reveals short-run and long-run asymmetry in Canada and Japan, and short-run asymmetry in the UK only. MATNARDL indicates short-run and long-run asymmetry in all G7 countries. The NARDL short-run estimates show that positive EPU shocks negatively impact the exchange rate in Canada, Japan, and the UK, while negative shocks have no significant impact. Long-run estimates show a positive effect of increased EPU on exchange rate in Canada and Japan only. The MATNARDL model reveals that most positive EPU shocks negatively affect the exchange rate in all G7 countries, while negative shocks are less significant. Robustness tests using GEPU yield consistent results. The GCQ test shows that the relationship between EPU and exchange rates varies across different quantiles.

The findings highlight the importance of using nonlinear models (MATNARDL) to capture the asymmetric and threshold effects of EPU on exchange rates. The standard NARDL model fails to capture the full picture, while the MATNARDL model reveals significant asymmetries across all G7 countries. This difference underscores the need for more nuanced models that account for varying magnitudes and directions of EPU shocks. The results suggest that positive EPU shocks (indicating increasing uncertainty) have a more substantial and consistent negative impact on exchange rates compared to negative shocks, possibly due to increased risk aversion among investors. The consistency of the results across both domestic EPU and GEPU supports the robustness of the findings. The varied effects across different quantiles emphasize the importance of considering the distribution of EPU and its heterogeneous effects on exchange rate dynamics.

This study contributes by demonstrating the asymmetric impact of EPU on exchange rates in G7 countries, particularly emphasizing the differential effects of minor versus major positive and negative shocks. The use of the MATNARDL model is crucial for revealing the full extent of these asymmetries, which were not fully captured by the traditional NARDL approach. Future research could explore the role of other macroeconomic variables or use panel data techniques for more comprehensive analysis. Further research should investigate the potential impact of the COVID-19 pandemic on these relationships.

The study relies on the assumption of purchasing power parity (PPP). While the results generally support this assumption, deviations might exist. Some diagnostic tests reveal model instability in certain cases (CUSUM in Japan, CUSUMQ in UK for NARDL, CUSUMQ in UK for MATNARDL). Future studies could incorporate additional variables, explore advanced panel data techniques, and account for the COVID-19 pandemic's influence. The interpretation of the findings should consider these limitations.