Spillovers and connectedness in foreign exchange markets: The role of trade policy uncertainty

Abstract

This paper analyses the directional spillover effects and connectedness for both return and volatility of nine US dollar exchange rates of globally most traded currencies under the influence of trade policy uncertainty. We find two interesting results over the study period ranging from December 1993 to July 2019. First, there exists asymmetric spillovers and connectedness among the considered exchange rates when trade policy uncertainty is present. Second, the volatility spillover is stronger than the return connectedness between exchange rate and trade policy uncertainty. These findings are robust to the presence of economic policy uncertainty effects. Concomitantly, the trade policy uncertainty patterns are also found to be useful for predicting currency market dynamics. Our findings contribute to the debate on the impact of trade policy uncertainty on the global economy and financial sector.

1. Introduction

In the wake of the recent trade wars between the United States and China as well as between the United States and its neighbours and allies including Canada, Mexico, and European Union, the global economy has been experiencing high levels of uncertainty in terms of economic policies, trade negotiations, proposals, and trade deals. The obvious reason leading to the trade tension and trade war was the large US trade deficit, mainly with China, Germany and Japan, where there have been accusations of competitive devaluation of the exchange rate and unfair trade practices (Nasir & Jackson, 2019). The United States has also withdrawn from the Trans-Pacific Partnership (TPP), imposed tariffs and renegotiated the North Atlantic Free Trade Agreement (NAFTA) within the United States–Mexico–Canada Agreement (USMCA or also known as NAFTA 2.0) (Brox & Fader, 2004). According to Baker, Bloom, and Davis (2019), actions and events such as the US withdrawal from the TPP, the Brexit, and the US-China trade tensions raised import tariffs and escalation in trade disputes, which leads to a significant increase in the Trade Policy Uncertainty (TPU).

Previous studies have shown the important implications of uncertainty for the financial sector and the economy as a whole, following the seminal work of Knight (1921) which pioneered the scholarly endeavours to understand the causes and the consequences of uncertainty. It is now common that high levels of economic uncertainty can have significant impacts on, among others, the probability of recessions, bank loans’ pricing, cost of capital, corporate cash policy, corporate innovation, and currency exchange activities (e.g., Karnizova and Li, 2014, Berg and Mark, 2018, Ashraf and Shen, 2019, Li and Zhong, 2020, Xu, 2020). Economic agents and policymakers thus face a number of various economic challenges as economic uncertainty makes it difficult to forecast the future economic and financial outlook as well as to carry out effective forward-looking macroeconomics policies (Nasir & Morgan, 2018). The study of Bloom (2009) examines “the impact of uncertainty shocks” on the US economy using firm-level data and shows that rising uncertainty can push firms to halt their investment and recruitment, and further have a negative impact on productivity in the short-term. However, the productivity, employment, and output bounce back in the medium term as the uncertainty fades. A more recent study by Basu and Bundick (2017) documents the negative effects of uncertainty on output, consumption, investment, and hours worked. They further argue that uncertainty can cause price instability and hence have implications for the monetary policy. Nonetheless, it can also make the recessions worst, particularly in the zero lower bound regimes.

An important aspect of uncertainty is its implications for the effectiveness of the economic policy. Bernanke (2012) show that fiscal policy uncertainty has been acting as a headwind for the recovery of the US economy after the global financial crisis 2008–2009. Insights into the negative impact of policy uncertainty were also provided in Fernández-Villaverde, Guerrón-Quintana, and Juan Rubio-Ramírez (2015). In their analysis addressing the impact of fiscal policy uncertainty on the US economy through the use of a VAR and New Keynesian models, Fernández-Villaverde et al. (2015) report that the unexpected changes in the fiscal volatility shocks can have sizable adverse effects on economic activity.

Regarding the trade tensions and resulting trade policy uncertainty, the recent empirical evidence suggests adverse implications for various aspects of the global economy and financial sector such as tariffs, exports, market access, investments, economic activity, employment, and stock markets (Baker, Bloom, Davis, & Kost, 2019; Crowley, Meng, & Song, 2018; Caldara, Matteo, Patrick, Andrea, & Andrea, 2020; Osnago, Piermartini, & Rocha, 2015; Osnago, Piermartini, & Rocha, 2018). Specifically, the US-China trade war, as declared by the US President Donald Trump, is found to not only harm the US equity market (Burggraf, Fendel, & Huynh, 2019) but also detrimental for the global stock markets (Huynh & Burggraf, 2020; Thanh, Canh, & Doytch, 2020).

Our paper contributes to the existing literature on economic and trade uncertainty by examining the impact of the TPU on exchange rate markets. This investigation is crucial because trade uncertainty arising from international trade disputes is supposed to have a direct bearing on exchange rates dynamics. More precisely, we empirically assess the directional (return and volatility) spillover effects and connectedness among nine globally most-traded currencies against US dollar exchange rates under the TPU context. In terms of measurement of TPU, we follow the TPU index constructed by Baker, Bloom, Davis et al. (2019), whose effectiveness relies on the frequency of joint occurrences of trade policy and uncertainty terms across major newspapers. Accordingly, trade-policy news and anxieties are a potential source of uncertainty which may be transmitted to exchange rate volatility when the parties are embroiled in trade conflicts with major trading partners. The nine globally most-traded currencies against US dollar employed in this study are from the countries which are among the main trading partners of the United States. The statistics as of December 2019 from the US Census Bureau show, for example, that Canada, Japan, the United Kingdom account respectively for 14.8 %, 5.3 %, 3.2 % of the US total trade in goods, while the trade in goods between the United States and the European Union (19 out of 27 member countries use the euro currency) reached US$851.94 billion or 20.56 % of the US total trade in goods.1 To the extent that the TPU has political dimensions and expectations around the trade policy stance, our work draws on the theoretical framework put forward by Engel and West (2005) as well as Beckmann and Czudaj (2017), which explains the exchange rate volatility through expectations channels (Conrad & Lamla, 2010).

Using the generalized VAR, directional spillovers and connectedness measures developed respectively by Diebold and Yilmaz (2012), and Diebold and Yılmaz (2014), our study leads to several important findings over the period from December 1993 to July 2019. Firstly, the degree of directional volatility and return spillovers as well as the associated dynamic connectedness are not alike across the considered exchange rates under the influence of trade policy uncertainty. Secondly, the volatility connectedness between exchange rates and trade policy uncertainty is found to be stronger than the return connectedness. Finally, the role of the TPU regarding return and volatility transmissions in exchange rate markets remain robust when the so-called economic policy uncertainty is introduced into the model. These findings thus imply that trade policy uncertainty patterns are important information for economic agents and useful in predicting currency market dynamics.

The remainder of the paper proceeds as follows. Section 2 briefly reviews the existing literature. Section 3 introduces our methodological framework and data used. Section 3 reports and discusses the empirical results. Section 4 provides some concluding remarks and policy implications.

2. A brief literature review

Since the global financial crisis 2008–2009, a number of studies have focused on the measurement of uncertainty in its various forms, including the TPU. In their seminal paper on economic policy uncertainty, its measurement and impact, Baker, Bloom, and Davis (2016) reported that the policy uncertainty leads to increased volatility in the stock markets and adverse impact on investment. Furthermore, it also causes an increase in unemployment and a reduction in the output in the United States as well as in other major economies. A later study by Baker, Bloom, Davis et al. (2019) empirically establishes that tariff hikes, tariff threats, and tariff retaliation are the crucial sources of anxiety and the TPU can have an adverse impact on the US economy. Moreover, the increase in TPU contributes to exacerbating the equity market volatility proxied by the implied volatility index (VIX). It is worth noting that one of the pivotal points regarding the construction of the TPU index is the selection of trade policy related news in American newspapers instead of the general perspectives in the whole economy. The TPU thus corresponds to a specific category of the overall economic policy uncertainty and reflects the uncertainty and intensity of trade policy discussions.

Other studies have also documented that trade tensions, which resulted in trade wars and increased uncertainty, have non-trivial adverse impacts on both the US economy and its main trading partners. For instance, Caldara et al. (2020) employ the US firm-level and macroeconomic data to investigate the economic effects of trade policy uncertainty. Their obtained results show that the rise in the TPU increases markup and reduces investments and economic activity. The authors associated this decline in investment and economic activity with the increasing uncertainty about the future higher tariffs. On a note published on 10 September 2019 by Bank of England, Wise and Turnbull (2019) show that the recent economic slowdown in both advanced and emerging economies comes from a wide range of factors but coincided with an escalation in trade tensions and trade policy uncertainty. They also point out the larger size of the indirect effects of trade tensions on the global economy (e.g., the sentiment of economic agents) relatively to their direct effects. For example, if firms are uncertain about future trading policies across countries, they will tend to reduce their investments. On this aspect, Handley and Limão (2017) argue that the Chinese accession to WTO and trade with China had reduced the US threat of trade war and had led to increased exports. The reduced policy uncertainty has further induced a reduction in inflation and an increase in income in the United States. These authors thus emphasised the importance of agreement and reducing trade tensions which can drive up uncertainty. A later study by Crowley et al. (2018) focuses on the impact of trade (tariff) policy uncertainty on Chinese firms and reports that it negatively affected the likelihood of Chinese firms to enter into a foreign market and can even cause exits from existing markets. Similarly, Steinberg (2019) focuses on the Brexit and the UK economy and shows evidence that there could be significant consumption equivalent welfare costs associated with the trade policy uncertainty during the post-Brexit period. Osnago et al. (2015) investigate the margins-of-trade effects of TPU for a sample of 149 exporting countries, where the trade policy uncertainty is measured as “the gap between binding tariff commitments under trade agreements and applied tariffs”. They report the adverse effects of TPU on exports with the negative impact of TPU being higher in the countries whose institutional quality is low. Their obtained results suggest that the negative impact of TPU ranges from 1.7 to 8.7 % of tariffs. In a more recent study, Osnago et al. (2018) reach similar conclusions in that TPU constitutes an important obstacle to exports for 65 exporters and that its effects are heterogeneous across exporters. The probability of exporting and trade volume would increase by 6 % and 13 % respectively if all tariff gaps are eliminated.

On the other hand, there is an ample evidence to suggest that the economic fundamentals drive the exchange rate dynamics in the long term, although the policy announcements are considered to possess the short to long-term effects (e.g., Taylor, Peel, & Sarno, 2001; Sarno & Taylor, 2003). A study by Nasir and Morgan (2018) documents that the uncertainty preceding the Brexit had caused the GBP to depreciate significantly. They associate this sharp depreciation with increased uncertainty about the UK’s trade policy and future trading relationships with the European Union. Based on this line of reasoning, it implies that the uncertainty in general and the TPU in particularly may have implications for the exchange rate markets. However, the existing evidence does not provide much insights into the role of TPU in foreign exchange rate markets. To address this caveat in the existing literature, this paper analyses the implications of TPU for the exchange rate markets. We formally test the hypothesis that the increases in TPU will result in the rising spillover effects in the exchange rate markets. Our focus is on both return and volatility spillover effects as well as the robustness check of the TPU impacts when the economic policy uncertainty is also considered.

3. Empirical approach

In order to investigate the spillover effects and connectedness for both returns and volatility of nine US dollar exchange rates, we follow Diebold and Yilmaz (2012), and Diebold and Yılmaz (2014) to specify a generalized VAR model where exchange rate returns and volatility are alternatively used as dependent variables along with the TPU returns and volatility. This empirical approach allows us to generate the total spillover index based on the H-step ahead generalized forecast error variance decomposition (GFEVD) and to assess the degree of connectedness based on the directional spillover index (Baruník & Kočenda, 2019)2 because the approach of Baruník and Kočenda (2019) accounts for the spectral representation of variance decompositions for high-frequency data. Therefore, we also consider the aforementioned approach to estimate for the spillover effects and connectedness in this study.

3.1. Average monthly returns and realized volatilities

Before the generalized VAR model can be set up, we need to calculate the average monthly return and realised volatilities. Our realised variances (RV) are calculated using the approach proposed by Andersen and Bollerslev (1998), which can be specified as follows:

$$\text{R}\text{V}=\sum _{k=1}^{\text{n}}{\text{r}}_k^2$$ (1)

where $r_k=p_k-p_{k-1}$ with $p_k$ being defined as a continuous-time stochastic process of log-prices at k. This stochastic process with pure jump component is further modelled as in Eq. (2):

$$p_t={\int }_0^t{\mu }_{s}ds+{\int }_0^t{{\sigma }_s}^{2}d{W}_s+J_t$$ (2)

where $\mu$ means a locally bounded predictable drift process, and $\sigma$ denotes positive volatility process. $J_t$ indicates a jump part and all factors have common filtration. For the estimation of asymmetric realised variance, we drew on the approach proposed by Engle (2010).

Since our dataset contains daily time series, we transformed daily data into monthly data in order to match with the monthly TPU index. For this purpose, monthly average returns for exchange rates and TPU index are calculated from the daily returns at time k ($r_k$):

$$\text{A}\text{R}=\frac{1}{n}\sum _{\text{k}=1}^{\text{n}}{r}_k$$ (3)

Despite the availability of monthly exchange rate data, we choose to employ daily data to construct monthly exchange rate returns from daily observations because this method allows us to capture a rich and dynamic patterns of both exchange rate return and volatility. As it has been shown in previous studies (Gatheral, Jaisson, & Rosenbaum, 2018; Zhou, 1996), using the first and last monthly observation of exchange rate to compute the returns and volatility would lose a number of stylized facts in between observations.

3.2. GFEVD

Given the availability of monthly average returns and realized variances, we employ the generalized VAR and spillover index of Diebold and Yilmaz (2012), and Diebold and Yılmaz (2014) to estimate the directional return and volatility spillovers among exchange rates in the presence of TPU effects as well as between TPU and US exchange rates. This approach is advantageous in that it is invariant to the ordering of the variables. It also allows for the calculation of both the direction and strength of spillovers over time and among different variables. We build a VAR(p) process for the vector of realised volatilities (average returns) of both TPU index and exchange rates, ${\text{E}\text{X}}_{\text{t}}={\left({\text{E}\text{X}}_{1\text{t}},\dots ,{\text{E}\text{X}}_{\text{N}\text{t}}\right)}^{\text{'}}$, such as:

$${\text{E}\text{X}}_{\text{t}}=\sum _{\text{i}=1}^{\text{p}}{\text{Φ}}_{\text{i}}{\text{E}\text{X}}_{\text{t}-\text{i}}+{\text{ε}}_{\text{t}}\text{w}\text{h}\text{e}\text{r}\text{e}{\text{ε}}_{\text{t}}\sim \text{N}(0,{\sum }_{\epsilon })$$ (4)

The moving average representation of residual ${\text{ε}}_{\text{t}}$ in VAR(p) has the following form:

$${\text{E}\text{X}}_{\text{t}}=\sum _{\text{i}=1}^{\text{∞}}{\text{Ψ}}_{\text{i}}{\text{ε}}_{\text{t}-\text{i}}$$ (5)

where coefficients are in the matrix of ${\text{Ψ}}_{\text{i}}$. Eq. (6) briefly summarizes the total spillovers index by using H-step-ahead generalized forecast error variance decomposition matrix, having the following elements for H = 1,2…

$${\text{θ}}_{\text{j}\text{k}}^{\text{H}}=\frac{{\text{σ}}_{\text{k}\text{k}}^{-1}\sum _{\text{h}=0}^{\text{H}-1}{\left({\text{e}\text{'}}_{\text{j}}{\text{Ψ}}_{\text{h}}{\text{Σ}}_{\text{ε}}{\text{e}}_{\text{k}}\right)}^2}{{\sum }_{\text{h}=0}^{\text{H}-1}\left({\text{e}\text{'}}_{\text{j}}{\text{Ψ}}_{\text{h}}{\sum }_{\text{ε}}{\text{Ψ}\text{'}}_{\text{h}}{\text{e}}_{\text{k}}\right)},\text{j},\text{k}=1,\dots \text{N}$$ (6)

More precisely, ${\text{Ψ}}_{\text{h}}$ is the matrix having the moving average coefficients, forecasted at time t while ${\text{Σ}}_{\text{ε}}$ denotes the variance matrix for the error vector ${\text{ε}}_t$. ${\sigma }_{kk}$ is the kth diagonal element of ${\text{Σ}}_{\text{ε}}$. In addition, ${\text{e}}_{\text{j}}$ and ${\text{e}}_{\text{k}}$ are selection errors with one as the jth and kth element and zeros otherwise. We follow the approach of Baruník and Kočenda (2019) to measure directional spillovers from exchange rate j to exchange rate k as follows:

$${\text{S}}_{\text{N},\text{j}\leftrightarrow \text{k}}^{\text{H}}=100\times \frac{1}{\text{N}}\sum _{\begin{array}{c}k=1\\ j\ne k\end{array}}^{\text{N}}{\stackrel{\sim }{\text{θ}}}_{\text{j}\text{k}}^{\text{H}}$$ (7)

The receiving effects are calculated when adding all numbers in rows j except the terms on a diagonal which correspond to own impacts. The sending effects are estimated by the sum of numbers in the column, excluding the numbers on the diagonal terms.

4. Findings and analysis

4.1. Data and descriptive statistics

For the empirical estimation of our model, we employ the monthly returns and realised volatility of the US TPU as well as those of the US exchange rates which are computed from daily exchange rates. Exchange rates are indirect quotations of nine most important currencies in international trade against the US dollar: British pound (GBP), Australian dollar (AUD), Canadian dollar (CAD), Swiss franc (CHF), Euro (EUR), Japanese yen (JPY), Norwegian krone (NOK), New Zealand dollar (NZD), and Swedish krona (SEK). Data are collected from Thomson Reuters database and span the period from December 1985 to July 2019. Fig. 1 shows the dynamics of TPU and the selected exchange rates, consisting of GBP and EUR, over the study period. This figure can be exemplified how the exchange rate returns could co-move the dynamics of Trade Policy Uncertainty.

Fig. 1.

The co-movement between TPU and selected exchange rate returns.

Table 1 summarizes the data characteristics along with stationarity and structural break testing.

Table 1.

Characteristics of return and realised volatility for exchange rates and TPU.

Variable	Mean	Std. Dev.	Skewness	Kurtosis	Zivot-Andrews	Break Time
TPU	4.239	0.864	−0.024	4.269	−4.714*	1995 March
GBP	0.001	0.001	−0.393	5.826	−9.735***	2002 February
AUD	0.002	0.001	0.465	4.780	−11.301***	2001 October
CAD	0.000	0.001	0.291	6.173	−9.735***	2002 April
CHF	−0.011	0.002	0.065	4.211	−21.258***	2001 July
EUR	−0.005	0.002	0.261	3.615	−20.282***	2001 July
JPY	−0.008	0.002	−0.903	6.405	−19.149***	1995 May
NOK	0.001	0.001	−0.005	3.841	−9.231***	2000 May
NZD	−0.004	0.001	0.345	5.085	−10.429***	2001 April
SEK	0.002	0.001	0.221	5.810	−16.887***	2001 October
GBP_V	0.082	0.001	4.980	38.263	−6.384***	2007 November
AUD_V	0.118	0.002	12.637	206.208	−13.003***	2007 July
CAD_V	0.048	0.001	6.456	64.527	−6.479***	2011 November
CHF_V	0.111	0.001	8.266	90.327	−18.575***	2008 February
EUR_V	0.084	0.001	2.536	12.962	−7.401***	2007 December
JPY_V	0.127	0.002	8.105	89.615	−7.917***	2007 August
NOK_V	0.069	0.001	4.255	26.987	−7.109***	2007 October
NZD_V	0.129	0.002	6.399	66.783	−5.671***	2007 June
SEK_V	0.071	0.001	5.314	41.930	−7.743***	2008 January

Notes: The total observation is 415. Zivot-Andrews represents the minimum t-statistic. The identified break dates are presented in the ‘year month’ format. In addition, the critical values for 1 %, 5 %, and 10 % significance level are -5.34, -4.80, and -4.58, respectively. ***, **, and * denotes rejection of null hypotheses at the 1 %, 5 %, and 10 % levels of significance. ‘_V’ denotes the realised volatility for each exchange rate pair. These exchange rates are indirect quotations of nine currencies against the USD from WM/Reuters. The snapshots are taken from the Reuters system around 16 pm and median rates are then selected for each currency from to construct this exchange. The mean values are in the percentage form, while the proxy of TPU is used as the natural logarithm to take into account for the uncertainty changes in unit instead of percentage changes. More noticeably, our TPU variable is stationary at 10 % significance level, which means that it is suitable for further statistical analysis.

Table 1 shows evidence of asymmetries and fat tails in the probability distributions of exchange rate returns, given the non-zero values of skewness coefficients and kurtosis coefficients superior to 3 in all cases. We also perform the Jarque-Bera test and find that all series depart from normality.3 It means that all variables are non-normal distribution. Four out of nine exchange rates under consideration exhibit the negative average return over the study period (USD/CHF, USD/EUR, USD/JPY, and USD/NZD). Furthermore, since we use the VAR model to derive return and volatility spillovers, it is necessary to make sure that our variables are stationary in order to avoid estimation biases. Following Chevallier, Nguyen, Siverskog, and Uddin (2018), we use the Zivot-Andrews test to examine the null hypothesis that our variables contain a unit root while assuming a potential exogenous structural change. The obtained results indicate that the null hypothesis is rejected for all variables at the 1 % level, which suggest that we can proceed with further quantitative estimations.

Fig. 1 displays the heat map correlation with regard to returns and realised volatility. We observe that the correlation of TPU with exchange rates is relatively higher in return than in volatility among our variables. Noticeably, the exchange rate of GBP has a high volatility correlation with the remaining variables, while the exchange rate of CHF tends to be immune to the volatility linear dependence. The return correlation is quite modest, except for the pair of CHF-EUR, NZD-AUD, and NOK-SEK. These currencies have a strong linear dependence. This finding can intuitively be understood in the light of the close economic and geographical association between these economies and hence their currencies.

4.2. Static exchange rate returns and volatility spillover effects

Table 2 summarises the results on connectedness among the 9 pairs of exchange rates and trade policy uncertainty. At first sight, we see that the return spillover effect from TPU to exchange rate returns is considerably marginal, as it does not exceed 1 % in all cases and its total given spillovers only sum up to 3.91 %. The top three currencies that received the most spillover effects from TPU include the Norwegian krone, the Swiss franc, and the New Zealand dollar. This could be explained by the fact that Norway and New Zealand are significantly exposed to trade activities with both the United States and China, while trade policy uncertainty can incite investors to trade more the Swiss franc given its safe haven property. On the other way around, the transmission of shocks affecting exchange rate returns to TPU is greater with a total spillover effect of 12 %. The most important givers of spillovers among the exchange rates are the Swiss franc (3.74 %), the Japanese yen (2.10 %), and the euro (1.28 %). The return spillovers among exchange rates are richer than between them and the TPU, even though their own variance shares are still high and range from 64.39 % (SEK) to 90.62 % (CHF). The SEK, JPY, CAD, and NZD currencies are the most vulnerable to return shocks of other exchange rates.

Table 2.

The static return spillover effects in the presence of TPU.

	TPU	GBP	AUD	CAD	CHF	EUR	JPY	NOK	NZD	SEK	Total	${SE}_{n⟵\bullet }$
TPU	88.00 %	0.10 %	1.23 %	0.73 %	3.74 %	1.28 %	2.10 %	1.02 %	0.84 %	0.95 %	100 %	12.0 %
GBP	0.14 %	86.84 %	1.30 %	0.18 %	6.05 %	2.32 %	0.16 %	1.85 %	0.81 %	0.34 %	100 %	13.2 %
AUD	0.07 %	6.82 %	82.39 %	0.09 %	6.07 %	0.56 %	0.55 %	1.97 %	1.13 %	0.35 %	100 %	17.6 %
CAD	0.14 %	1.80 %	0.83 %	73.12 %	14.00 %	6.50 %	1.28 %	0.91 %	0.24 %	1.17 %	100 %	26.9 %
CHF	0.87 %	2.05 %	1.90 %	0.20 %	90.62 %	2.82 %	0.02 %	0.51 %	0.77 %	0.24 %	100 %	9.4 %
EUR	0.39 %	5.71 %	2.09 %	0.08 %	3.77 %	85.01 %	0.43 %	0.68 %	1.51 %	0.35 %	100 %	15.0 %
JPY	0.39 %	10.17 %	2.57 %	0.39 %	7.43 %	7.09 %	67.35 %	3.78 %	0.49 %	0.34 %	100 %	32.7 %
NOK	0.98 %	3.42 %	0.62 %	0.14 %	1.35 %	0.71 %	0.27 %	90.39 %	2.11 %	0.01 %	100 %	9.6 %
NZD	0.57 %	10.81 %	3.68 %	0.18 %	5.28 %	0.87 %	0.07 %	1.98 %	74.54 %	2.02 %	100 %	25.5 %
SEK	0.36 %	17.45 %	0.97 %	0.42 %	2.87 %	7.72 %	1.47 %	0.82 %	3.53 %	64.39 %	100 %	35.6 %
Total	91.91 %	145.16 %	97.58 %	75.55 %	141.18 %	114.89 %	73.70 %	103.91 %	85.96 %	70.16 %	1000 %	197.4 %
${SE}_{n\to \bullet }$	3.91 %	58.32 %	15.19 %	2.43 %	50.56 %	29.88 %	6.35 %	13.52 %	11.42 %	5.77 %	197.4 %	19.74 %
Net	−8.09%	45.16 %	−2.42%	−24.45%	41.18 %	14.89 %	−26.30%	3.91 %	−14.40%	−29.84%

Notes: This table demonstrates the results of the return spillover effects performed with the generalized FEVD (GFEVD). Directional return spillovers (SE_mn) which correspond to the percentage share of error variance in exchange rate n (rows) contributed by shocks to exchange rate m (columns) including the existence of TPU. Total received spillover for exchange rate n is given by its row sums reported in the columns added to the right of the table, both including $\sum _{n=1}^N{SE}_{mn}=100)$ and excluding ${SE}_{n⟵\bullet }$ own variance share.

In terms of net spillover effects, the TPU stands as the fifth most important receiver of return spillovers (-8.09 %), behind the exchange rates of SEK (-29.84 %), JPY (-26.30 %), CAD (-24.45 %), and NZD (-14.40 %). By contrast, the exchange rates of GBP, CHF and NOK are the return spillover givers with a net transmission of 45.16 %, 41.18 %, and 3.91, respectively, suggesting that they can provide some valuable indications about the changes in other exchange rates and trade uncertainty.

Table 2 shows the static volatility spillover effects among the exchange rates and trade policy uncertainty. We can easily see an increase in the level of their volatility transmissions as compared to the return transmissions in Table 2 since the own variance share of TPU is now reduced to 80.7 % (88 % in the return spillovers) and it gives a total volatility spillover of 23.73 % to exchange rates (only 3.91 % in the case of return spillovers). The highest volatility spillover from TPU to exchange rates is observed for the case of the Canadian dollar (7.61 %), followed by the New Zealand dollar (4.27 %) and the Norwegian krone (4.07 %). The total contributions of exchange rates to the forecast error variance of the TPU’s volatility also increase and attain 19.93 %, compared with 12 % in the case of return spillovers. The most important giver of volatility spillovers to the TPU is the exchange rate of the Japanese yen (6.02 %), followed by the New Zealand dollar (4.81 %) and the Australian dollar (4.15 %). It is worth noting that the changes in the volatility of the CHF exchange rate do not exert the same importance as in the return spillover to the TPU (only 0.05 %). Surprisingly, the euro has very little volatility spillover effect with the TPU throughout the period from 1985 to 2019, despite the fact that the United States is the largest trading partner for EU exports of goods and the second-largest partner for EU imports of goods. However, the foremost finding of Table 3 is that the TPU is a net transmitter of volatility spillovers (3.80 %), suggesting that a higher degree of trade policy uncertainty could raise instabilities in the foreign exchange markets. Among the exchange rates, the GBP, AUD, and JPY currencies are the net sender of volatility transmission with 12.22 %, 47.74 %, and 24.80 %, respectively, whereas the remaining exchange rates are net receivers of external volatility shocks.

Table 3.

The static volatility spillover effects in the presence of TPU.

	TPU	GBP	AUD	CAD	CHF	EUR	JPY	NOK	NZD	SEK	Total	${SE}_{n⟵\bullet }$
TPU	80.07 %	1.88 %	4.15 %	1.86 %	0.05 %	0.51 %	6.02 %	0.46 %	4.81 %	0.20 %	100 %	19.93 %
GBP	0.34 %	76.68 %	8.25 %	2.79 %	0.04 %	0.70 %	9.81 %	0.23 %	1.11 %	0.05 %	100 %	23.32 %
AUD	2.68 %	1.93 %	88.24 %	0.66 %	0.05 %	3.35 %	0.34 %	0.30 %	2.32 %	0.11 %	100 %	11.76 %
CAD	7.61 %	3.91 %	1.89 %	81.53 %	0.15 %	0.34 %	1.20 %	0.26 %	3.00 %	0.11 %	100 %	18.47 %
CHF	0.63 %	0.12 %	5.39 %	1.12 %	87.51 %	2.70 %	0.08 %	1.51 %	0.38 %	0.55 %	100 %	12.49 %
EUR	0.79 %	0.59 %	10.55 %	0.49 %	0.07 %	83.08 %	0.31 %	0.30 %	2.67 %	1.14 %	100 %	16.92 %
JPY	1.50 %	5.84 %	1.37 %	2.54 %	0.14 %	3.04 %	83.51 %	1.17 %	0.62 %	0.26 %	100 %	16.49 %
NOK	4.07 %	10.98 %	4.38 %	2.34 %	0.37 %	6.02 %	13.57 %	55.18 %	0.61 %	2.50 %	100 %	44.82 %
NZD	4.27 %	1.14 %	10.80 %	4.25 %	0.09 %	2.17 %	0.97 %	0.17 %	76.07 %	0.08 %	100 %	23.93 %
SEK	1.84 %	9.15 %	12.72 %	3.39 %	0.54 %	5.13 %	8.97 %	7.62 %	3.06 %	47.58 %	100 %	52.42 %
Total	103.80 %	112.22 %	147.74 %	100.97 %	89.00 %	107.04 %	124.80 %	67.21 %	94.65 %	52.58 %	1000 %	240.6 %
${SE}_{n\to \bullet }$	23.73 %	35.54 %	59.50 %	19.44 %	1.49 %	23.96 %	41.29 %	12.03 %	18.58 %	5.00 %	240.6 %	24.06 %
Net	3.80 %	12.22 %	47.74 %	0.97 %	−11.00%	7.04 %	24.80 %	−32.79%	−5.35%	−47.42%

Notes: This table demonstrates the results of the volatility spillover effects performed with the generalized FEVD (GFEVD). Directional volatility spillovers (SE_mn) which correspond to the percentage share of error variance in exchange rate n (rows) contributed by shocks to exchange rate m (columns) including the existence of TPU. Total received spillover for exchange rate n is given by its row sums reported in the columns added to the right of the table, both including $\sum _{n=1}^N{SE}_{mn}=100)$ and excluding ${SE}_{n⟵\bullet }$ own variance share.

Our work thus does not confirm the previous literature regarding the prominent role of the Canadian dollar in volatility transmission (Wen & Wang, 2020). This currency’s impact is also limited in terms of return spillover effects. The high sensitivity of the NOK and SEK currencies to external volatility shocks could potentially be related to their export-driven and resources-based economy structures which depend on international prices of commodities and economic outlooks.

4.3. Rolling spillover effects between exchange rate and trade policy uncertainty

Fig. 2 demonstrates the return connectedness among our variables of interest. The TPU established itself as the net spillover transmitter to the exchange rate returns over the study period from 1985 to 2019. There is a short period where it changes position from spreading return shocks to receiving return shocks, which corresponds to the episode of the 2008–2009 global financial crisis. The reversed trend started at the beginning of 2011. In addition, three currencies, namely the CAD, CHF, and NOK, have the asymmetric return connectedness, while the remaining currencies exhibit one side shock transmission only (either sender or receiver). More precisely, the GBP is the most important transmitter with the rolling spillover effects in return ranging from 9 % to 20 %. Another important finding is that the EUR acts as a return shock recipient, but it plays a more crucial role in volatility transmission as a sender of volatility shocks (Fig. 3 ).

Fig. 2.

Return connectedness.

Fig. 3.

Volatility connectedness.

Fig. 3 depicts the dynamic volatility spillover effects between TPU and exchange rates. There is prima facie evidence of asymmetric spillover effects for all currencies depending on time horizons, whereas the TPU transmits volatility shocks to exchange rates over the whole period, except for some quite short time points. Five currencies (CAN, JPY, NOK, NZD, and SEK) are, with an exception of some short periods, net receivers of volatility shocks, which seems to reflect country-specific factors or the influence of the 2008–2009 global financial crisis or the 2010–2011 European debt crisis. By contrast, the AUD and GBP are mainly transmitters of volatility shocks, except for some short periods. The evidence regarding the TPU’s volatility transmission thus implies that it can be a useful tool to predict exchange rate volatility.

4.4. The role of economic policy uncertainty in driving exchange rate volatility

This section examines the relevance of TPU in explaining the dynamics of exchange rate returns and volatility when the soc-called global economic policy uncertainty (GEPU) proposed by Baker et al. (2016) is introduced into the estimation models. To the extent that the data for the GEPU only started since 1997, we have to shorten our sample period and conduct our robustness analysis from January 1997 to July 2019.4 Due to its global coverage and economic-wide uncertainty nature, it is expected that the GEPU has a strong correlation with the TPU and drives shocks to the TPU. However, to our great surprise, the correlation between the GEPU and the TPU is only moderate (0.41), despite significant. This preliminary evidence suggests that the TPU brings incremental contributions to the explanation of exchange rate dynamics, even in the presence of the GEPU.

Table 4 shows the volatility spillover effects of the same system when the GEPU is added. What is important to note is the increase in the average spillover effects that goes from 24.06 % (Table 3) to 37.68 % (Table 4). This result indicates the benefit of having the GEPU in the model as it helps improve the interactions and explanations of exchange rate volatility. As expected, the GEPU transmits more volatility shocks to the TPU (6.04 %) than the other way around (1.94 %). This can be explained by the fact that the economic policy uncertainty, as a factor encompassing a broad economic spectrum, causes an unstable business environment which leads to more uncertainty in trading policies and activities. However, the TPU remain superior to the GEPU in driving exchange rate volatility with a total transmission of 37.36 %, compared to 31.32 % by the GEPU. It is also interesting to note that both uncertainty indexes receive volatility shocks from exchange rates (30.61 % for the GEPU vs. 40.66 % for the TPU), with the highest pressures from AUD and JPY currencies. In addition, as in Table 3, the currencies of high oil-dependent nations (SEK, NOK, CAN, EUR, and GBP) are the most affected by shocks to other variables in the system. Finally, the impact of the TPU on exchange rate volatility is very similar to the finding of Table 3 when the GEPU is not included. At the same time, the AUD, JPY and NZD are the currencies exhibiting the largest contagious effects to the TPU. Taken together, the TPU stands as an important factor for explaining the dynamics of exchange rate volatility, whether the GEPU is considered or not.

Table 4.

The static volatility spillover effects in the presence of both EPU and TPU.

	GEPU	TPU	GBP	AUD	CAD	CHF	EUR	JPY	NOK	NZD	SEK	Total	${SE}_{n⟵\bullet }$
GEPU	69.39 %	1.94 %	0.33 %	14.39 %	1.73 %	0.22 %	1.17 %	8.10 %	0.23 %	2.28 %	0.23 %	100 %	30.61 %
TPU	6.04 %	59.34 %	4.47 %	7.52 %	2.60 %	0.22 %	1.25 %	5.72 %	1.09 %	11.17 %	0.60 %	100 %	40.66 %
GBP	1.84 %	2.94 %	64.25 %	5.35 %	14.44 %	0.23 %	3.63 %	1.65 %	0.92 %	3.38 %	1.38 %	100 %	35.75 %
AUD	3.91 %	2.69 %	0.41 %	81.77 %	0.38 %	0.10 %	7.28 %	0.03 %	0.14 %	2.14 %	1.14 %	100 %	18.23 %
CAD	3.27 %	7.48 %	4.03 %	5.58 %	65.65 %	0.17 %	2.89 %	3.73 %	0.48 %	4.94 %	1.78 %	100 %	34.35 %
CHF	0.81 %	3.15 %	1.34 %	4.81 %	0.79 %	75.09 %	4.40 %	1.42 %	4.69 %	0.95 %	2.55 %	100 %	24.91 %
EUR	2.58 %	3.68 %	4.41 %	11.52 %	0.37 %	0.20 %	56.40 %	4.92 %	5.05 %	4.27 %	6.61 %	100 %	43.60 %
JPY	1.49 %	1.66 %	6.48 %	7.25 %	1.76 %	0.37 %	3.30 %	72.53 %	2.10 %	2.37 %	0.67 %	100 %	27.47 %
NOK	3.09 %	4.26 %	5.70 %	19.11 %	1.83 %	0.19 %	4.52 %	13.50 %	40.05 %	6.07 %	1.70 %	100 %	59.95 %
NZD	3.75 %	5.29 %	0.62 %	9.73 %	2.53 %	0.09 %	7.79 %	0.29 %	0.74 %	67.54 %	1.62 %	100 %	32.46 %
SEK	4.46 %	4.26 %	7.43 %	21.84 %	2.42 %	0.69 %	6.04 %	6.96 %	8.61 %	3.86 %	33.44 %	100 %	66.56 %
Total	100.61 %	96.70 %	99.48 %	188.86 %	94.51 %	77.56 %	98.66 %	118.86 %	64.09 %	108.95 %	51.71 %	1100 %	414.54 %
${SE}_{n\to \bullet }$	31.22 %	37.36 %	35.23 %	107.09 %	28.86 %	2.47 %	42.26 %	46.33 %	24.04 %	41.41 %	18.27 %	414.54 %	37.68 %
Net	0.61 %	−3.30%	−0.52%	88.86 %	−5.49%	−22.44%	−1.34%	18.86 %	−35.91%	8.95 %	−48.29%

Notes: This table demonstrates the results of the volatility spillover effects performed with the generalized FEVD (GFEVD). Directional volatility spillovers (SE_mn) which correspond to the percentage share of error variance in exchange rate n (rows) contributed by shocks to exchange rate m (columns) including the existence of Trade Policy Uncertainty and Economic Policy Uncertainty. Total received spillover for exchange rate n is given by its row sums reported in the columns added to the right of the table, both including $\sum _{n=1}^N{SE}_{mn}=100)$ and excluding ${SE}_{n⟵\bullet }$ own variance share.

In sum, the empirical evidence reported here points to the fact that the exchange rate returns and volatilities are associated with trade policy uncertainties, captured by the news-based estimation approach. It is also shown that not only the TPU strengthens volatility spillovers of exchange rates but unexpected shocks to exchange rates also account for 37.36 % of changes in TPU (Table 4), leading to their negative efects on international trade and economic growth in case of high exchange rate volatility. Therefore, stabilizing exchange rates would be a crucial challenge for policymakers and central bankers in times of rising trade policy uncertainty.

5. Conclusion

The contemporary global trade imbalances have caused trade tensions and resulted in trade wars. The latter has, in turn, caused a significant amount of economic and trade policy uncertainty. The current literature in financial economics is, however, very limited to the understanding of trade policy uncertainty effects. Concerns over international trade uncertainties thus intensified the policy debate and provided the rationale to research on their consequences for the global economy and financial sector including exchange rate markets.

Contextualising on the importance and underexplored of trade policy uncertainty, this study provides the first empirical investigation of the static and dynamic interactions between the US trade policy uncertainty and the US dollar exchange rates of the ten most globally traded currencies. Our analysis of volatility and return spillovers shows evidence of significant, asymmetric, and heterogeneous spillover effects between TPU and exchange rates, and among exchange rates themselves. Similar results are obtained when we consider the return and volatility connectedness based on a rolling approach. Moreover, we find that volatility connectedness between exchange rate and trade policy uncertainty is higher than their return connectedness. Finally, our findings regarding the role of TPU in explaining the exchange rate returns and volatility remain intact when the global economic policy uncertainty is introduced. While the existing literature focuses on equity markets (Antonakakis, Chatziantoniou, & Filis, 2013), fixed-income market (Wisniewski & Lambe, 2015), and the precious metals market (Huynh, 2020), our findings confirm the role of trade policy uncertainty as a driving force of the dynamics of exchange rate markets.

Our results suggest that policymakers, trading partners, and investors should pay attention to the changes in TPU and their effects on exchange rate movements due to bilateral spillover effects. Subject to high exchange rate volatility under important trade policy uncertainty, global investors may attempt to design internationally diversified currency portfolios that protect them from systematic risk in foreign exchange markets by considering empirical sensitivities of US dollar exchange rates to policy uncertainty. Furthermore, there is also evidence that the monetary policy uncertainty is the main driver of economic policy uncertainty, followed by the fiscal policy, currency and finally trade policy uncertainty Gupta, Ma, Risse, & Wohar, 2018). Therefore, at the macroeconomic level, the effectiveness of monetary policy and monetary policy communication might mitigate the severity of trade policy uncertainty. Future research can extend our study by considering alternative trade policy uncertainty such as the one developed by Caldara et al. (2020) based on firm-level data and by assessing the impact of TPU on other asset markets such as equity and commodities. It would be also interesting to examine how uncertainties related to the ongoing COVID-19 pandemic which has disrupted the global trade follows alter the results.

Declaration of Competing Interest

The authors report no declarations of interest.