Is US trade policy uncertainty powerful enough to predict global output volatility?

ABSTRACT

Trade policy uncertainty is at an all-time high in the United States and continues to escalate. This paper empirically examined the ability of US trade policy uncertainty to predict global output volatility. To this end, a battery of econometric tests was employed—Toda and Yamamoto linear Granger causality test, nonparametric test for nonlinear causality, and nonlinear Granger causality test in frequency domain. Findings based on standard linear Granger causality tests suggested that US trade policy uncertainty is not a significant predictor of global output volatility. Further tests, however, showed that due to the presence of nonlinearities in the US trade policy uncertainty–global output volatility nexus, the linear Granger causality framework initially relied upon might have led to misspecification. Consequently, a nonparametric causality test was further conducted. The test results showed that in fact the US trade policy uncertainty is a significant predictor of global output volatility. To further verify the findings, the powerful frequency domain-based Granger causality test which is able to detect causality at short, medium and longer horizons was conducted. The test findings again confirmed that trade policy uncertainty emanating from the United States is a significant predictor of global output volatility.

KEYWORDS:

• Global trade agreement
• Granger causality
• time series analysis
• trade policy

JEL CLASSIFICATIONS:

• F13
• F41
• F42

Acknowledgments

The author is highly indebted to Professor David Giles and two anonymous referees for the insightful comments and suggestions provided on the earlier draft of this paper. Of course, all remaining errors are the authors.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The trend of GOV shown in Figure 5 is obtained from the original value (not absolute value) of the percentage change in the deviation of global output (x_t) from the trend component. The rationale for this is so that positive spikes can be viewed separately from negative spikes.