Do currency exchange rates impact gold prices? New evidence from the ongoing COVID-19 period

Abstract

We apply the nonlinear autoregressive distributed lag method to examine the relationships between seven leading currency exchange rates and gold prices using daily data from January 2017 to April 2021. The results reveal that in the short term, while negative United States dollar (USD) to United Kingdom pound, negative USD to Canadian dollar, negative USD to Japanese yen, negative USD to Danish krone, and positive USD to euro exchange rates increase gold prices, a lagged positive USD to euro and lagged positive USD to Danish krone exchange rates decrease gold prices. A test of the pre-pandemic normal period reveals that the uneven and unpredictable impacts of six exchange rates on gold prices are particularly due to COVID-19. We find efficiency in the gold market, in line with the market efficiency hypothesis and random walk theory. Our findings indicate that gold acts as a safe-haven asset for investors during COVID-19.

1. Introduction

The COVID-19 pandemic is likely to continue having substantial financial, economic, and social consequences on financial markets and institutions (Corbet, Larkin, & Lucey, 2020; Goodell, 2020). Gold allows investors to diversify their portfolios to minimize macroeconomic and financial risks (Agyei-Ampomah, Gounopoulos, & Mazouz, 2014; Baur & Lucey, 2010; Baur & McDermott, 2010; Beckmann et al., 2015, Beckmann et al., 2019; Bilgin, Gozgor, Lau, & Sheng, 2018; Bouoiyour, Selmi, & Wohar, 2018; Gürgün & Ünalmiş, 2014; Reboredo, 2013). As such, in both stressed and unstressed economic conditions, gold is seen as a safe asset (Beckmann et al., 2019; Harris & Shen, 2017; Ji, Zhang, & Zhao, 2020; Mensi, Hammoudeh, Al-Jarrah, Sensoy, & Kang, 2017), spurring investors to buy gold as a hedge against exchange rate volatility (Singhal, Choudhary, & Biswal, 2019).

Over the past two decades, literature and research have reflected a significant interest in investment in gold (O'Connor, Lucey, Batten, & Baur, 2015). Investors prefer to invest in gold when the economy underperforms (Jain & Biswal, 2016) and during periods of financial uncertainty (Bouri, Jain, Biswal, & Roubaud, 2017). Following the global financial crisis (GFC), gold has become an immensely popular substitute hedge tool in portfolio diversification (Kirkulak Uludag & Lkhamazhapov, 2014).

A series of studies (Jain & Biswal, 2016; Pukthuanthong & Roll, 2011; Singhal et al., 2019; Sjaastad & Scacciavillani, 1996) have identified the relationship between exchange rates and the volatility of gold prices, suggesting that changes in exchange rates contribute to gold prices' volatility during normal times. Pukthuanthong and Roll (2011), in particular, have indicated that the United States dollar (USD) exchange rate generally relates negatively with gold prices in USD, meaning that when the USD depreciates in light of appreciations in other currencies, gold prices in USD increase. This suggests a link between appreciations in other currencies and gold prices in USD. Studying the period spanning January 2, 1971 to December 10, 2009, Pukthuanthong and Roll (2011) empirically demonstrate that fluctuations in USD, European Union euro, United Kingdom pound, and Japanese yen exchange rates affect gold prices. The investigation, however, treats the span of time as a single sitting, not differentiating crisis periods from normal periods and consequently, does not explain whether—and, if so, how—the relationship of exchange rates and gold prices differs in crisis periods.

Studies (Baur & McDermott, 2010; Nguyen, Bedoui, Majdoub, Guesmi, & Chevallier, 2020; Yang & Hamori, 2014) have also pointed out that historical financial crises impacted the relationship in terms of gold's role as a safe-haven asset to protect against the risk of exchange rate depreciation. Morales-Zumaquero and Sosvilla-Rivero (2014) have investigated exchange rate volatility in 80 countries ranging from 1970 to 2011 and revealed that GFCs cause structural breaks in exchange rate volatility. Furthermore, these crises required central banks to take macroeconomic measures through unconventional monetary policy, such as pushing the interest rate to zero bound and using quantitative easing to stabilize the exchange rates–gold prices relationship (Baur & McDermott, 2010; Nguyen et al., 2020; Yang & Hamori, 2014).

The COVID-19-induced financial crisis is more global and perilous in scale than other historical financial crises, and the COVID-19 crisis has substantially affected the exchange market (Shehzad, Xiaoxing, & Kazouz, 2020). An overview of the ongoing COVID-19 crisis highlights that this outbreak has brought intense and widespread uncertainty to the financial markets, unprecedented even by the GFC (Baker, Bloom, Davis, & Terry, 2020; Goodell, 2020). Recent studies (Ilzetzki, Reinhart, & Rogoff, 2020; Yilmazkuday, 2021) have indicated that many advanced and emerging economies have adopted unconventional macroeconomic measures in response to the COVID-19's impact on the exchange rate, promoting zero-bound interest rates to prevent disruption in the long-term downward trend in exchange rate volatility.

All the above considerations have motivated us to perform this study that examines the relationship between exchange rates and gold prices during the sudden and unexpected COVID-19 crisis. Our new research question is: Have exchange rates impacted gold prices during the ongoing COVID-19 crisis, and if so, how? In analyzing the nexus between exchange rates and gold prices during and prior to the crisis, this study addresses a topic of immense interest to scholars, policymakers, and investors.

As a research method, this study uses the nonlinear autoregressive distributed lag (NARDL) approach to reveal whether gold prices are susceptible to fluctuating exchange rates during the ongoing COVID-19 pandemic and whether changes in exchange rates can predict future gold prices. The results suggest that the exchange rates–gold prices nexus is dynamic and nonlinear but not necessarily asymmetric. Exchange rates have had weak predictive power over gold prices during the ongoing COVID-19 period, a trend that is likely to continue at least until the end of the pandemic. Our study follows a suggestion by Harris and Shen (2017) that a price index free of global exchange rates would safeguard gold from changes in such rates. He, O'Connor, and Thijssen (2018) have further argued that the relationship between gold prices and other assets is constant and that gold is invariably a hedge rather than a safe haven. In examining the relationship between exchange rates and gold prices, however, our study contrasts this view, as we contend that gold will continue to serve as a safe haven. This research is expected to provide new insights into gold as an investment instrument and shed light on how the period following COVID-19 may also contribute to finance literature.

This study brings six-fold contributions to the literature. First, it demonstrates the asymmetric, dynamic, and nonlinear impacts of exchange rates on gold prices in the short term with no impact in the long term. Second, this study is the first to explore the effects of prevailing exchange rates on gold prices during the ongoing COVID-19 pandemic. Third, to some extent, this study suggests that leading exchange rates have unpredictably impacted gold prices on occasion during COVID-19. Fourth, the results reveal that exchange rates do not contribute evenly to gold prices. Fifth, to some degree, this study endorses the findings of Bredin, Conlon, and Potì (2015), who suggest that gold can serve as a hedging tool for 1 year. Sixth, we confirm that the recent uneven effects of exchange rates on gold prices are due to the COVID-19 crisis.

2. Data and research methods

2.1. Data

This study uses daily time-series data from multiple sources, outlined in Table 1 . The data cover a period from January 1, 2020 to April 19, 2021. Furthermore, we have analyzed the data for a normal period preceding the COVID-19 crisis, from January 1, 2017 to December 31, 2019, during which no crisis or structural break occurred. To control seasonality effects and gain a better understanding of the dynamics, we group the obtained data into 5-day-week clusters and estimate accordingly. We take seven exchange rates as independent variables. These are one United States dollar (USD) to one United Kingdom pound (USD → GBP), one USD to one euro (USD → EUR), one USD to one Canadian dollar (USD → CAD), one USD to one hundred Japanese yen (USD → JPY),1 one USD to one Swiss franc (USD → CHF), one USD to one Norwegian krone (USD → NOK), and one USD to one Danish krone (USD → DKK).

Table 1.

Variables.

Variables	Definitions	Notations	Primary source	Data source
Dependent variable:
Gold Prices	Prices of Gold Bullion, London Bullion Market in United States Dollar (USD), Per Metric Ton Ounce	GLDP	ICE Benchmark Administration	Refinitiv Eikon
Independent variable(s):
United States Dollar to United Kingdom Pound	USD to GBP - Exchange Rate	USD → GBP	Bank of England (BOE) Spot Rates
United States Dollar to Euro	USD to EURO - Exchange Rate	USD → EUR	European Central Bank (ECB)
United States Dollar to Canadian Dollar	USD to Canadian $ - Exchange Rate	USD → CAD	Refinitiv (GTIS - FTID/TR)
United States Dollar to 100 Japanese Yen	USD to 100 Japanese Yen - Exchange Rate	USD → JPY	Refinitiv (GTIS - FTID/TR)
United States Dollar to Swiss Franc	USD to Swiss Franc - Exchange Rate	USD → CHF	Refinitiv (GTIS - FTID/TR)
United States Dollar to Norwegian Krone	USD to Norwegian Krone - Exchange Rate	USD → NOK	Refinitiv (GTIS - FTID/TR)
United States Dollar to Danish Krone	USD to Danish Krone - Exchange Rate	USD → DKK	Refinitiv (GTIS - FTID/TR)

The rationale in selecting these leading exchange rates is that they are commonly used and thus can provide a rich understanding of the relationships between the exchange rates and gold prices. For example, Goldman (2000) uses USD → GBP; Pukthuanthong and Roll (2011), USD → EUR, USD → GBP, and USD → JPY; Bedoui, Braeik, Goutte, and Guesmi (2018), USD → GBP, USD → EUR, USD → CAD, USD → JPY, and USD → CHF; Mirkov, Pozdeev, and Söderlind (2019), USD → CHF; Dong, Chen, Lee, and Sriboonchitta (2019), USD → GBP and USD → EUR; Cumperayot and Kouwenberg (2020), USD → GBP, USD → CAD, USD → JPY, USD → NOK, and USD → DKK; Li, Lu, Jiang, and Petrova (2021) and Ding (2021), USD → GBP, USD → EUR, USD → CAD, and USD → JPY.

Table 1. Describes the list of variables in detail.

Both gold prices (y) and exchange rates (x) are transformed into their natural logarithmic forms. Fig. 1 presents gold prices and the aforementioned seven exchange rates during the ongoing COVID-19 period. During the pandemic, gold prices (Fig. 1i) have jumped from USD 1500 to over USD 2000, and from August 2020, the prices came down to float between USD 1950 and 1800 until the beginning of the third week of April 2021. A sharp decline also occurred between March 1 and April 1, 2020, when tension because of COVID-19 reached a high point. Fluctuations in all seven exchange rates, however, were intense. Similarly to gold prices, these exchange rates experienced an acute decline between March 1 and April 1, 2020, again highlighting the fear resulting from COVID-19. Although USD → JPY (Fig. 1v) is returning to equilibrium as of April 2021, the rest of the exchange rates (Fig. 1ii–iv and vi–viii) are experiencing a rising trend in general.

Fig. 1.

Exchange rates and gold prices during the ongoing COVID-19 period.

(i) GLDP, (ii) USD → GBP, (iii) USD → EUR, (iv) USD → CAD, (v) USD → JPY, (vi) USD → CHF, (vii) USD → NOK, (viii) USD → DKK. Notes: GLDP is the gold price, USD → GBP is one United States dollar (USD) to one United Kingdom pound, USD → EUR is one USD to one euro, USD → CAD is one USD to one Canadian dollar, USD → JPY is one USD to one hundred Japanese yen, USD → CHF is one USD to one Swiss franc, USD → NOK is one USD to one Norwegian krone, and USD → DKK is one USD to one Danish krone.

2.2. Model specification

The NARDL approach, as proposed by Shin, Yu, and Greenwood-Nimmo (2014), is used as it is a dynamic and asymmetric model that is able to differentiate between long- and short-term effects. The short-term estimation assesses the immediate impact of changes in the exogenous variable on the dependent variable, while a long-term estimate evaluates the reaction time and speed of adjustment toward a level of equilibrium. Although a nonlinear threshold vector error correction model could capture these facts, there would be a convergence issue when the parameters proliferate. In contrast, the NARDL model is free from such a problem as it relaxes this restriction by not requiring the same sequence of integration for a particular time-series of variables (Apergis & Cooray, 2015). Furthermore, the NARDL model automatically chooses the best lag order to resolve multicollinearity issues (Shin et al., 2014). It is worth mentioning that a bounds testing approach (e.g., NARDL) offers robust empirical results (Narayan, 2005).

The equations for the NARDL (p, q) model, as suggested by Shin et al. (2014), are as follows:

$$y_t=\sum _{j=1}^p{\varnothing }_j{y}_{t-1}+\sum _{j=0}^q\left({\theta }_j^{+\prime }{x}_{t-j}^{+}+{\theta }_j^{-\prime }{x}_{t-j}^{-}\right)+{\epsilon }_t,$$ (1)

Where x _t is a k x 1 vector of multiple regressors, defined in such a way that x _t = x ₀ + x _t ⁺ + x _t ⁻, ϕ _j is the autoregressive parameter; θ _j ⁺ and θ _j ⁻ are the asymmetrically distributed lag parameters; and ε _t is an i.i.d. process with zero mean and constant variance σ _ε ². First, the long-term equation is:

$$y_t={\beta }^{+}{x}_t^{+}+{\beta }^{-}{x}_t^{-}+u_t$$ (2)

$$∆x_t=v_t$$ (3)

where y _t (gold prices) and x _t (exchange rates) are scalar I(1) variables, x _t is decomposed as x _t = x ₀ + x _t ⁺ + x _t ⁻, where x _t ⁺ and x _t ⁻ are partial sum processes of the positive and negative changes in x _t:

$$x_t^{+}=\sum _{j=1}^t∆x_j^{+}=\sum _{j=1}^{t}max\left(∆x_j,0\right),x_t^{-}=\sum _{j=1}^t∆x_j^{-}=\sum _{j=1}^{t}min\left(∆x_j,0\right)$$ (4)

The symmetric short-term coefficients are tested using Wald's statistic, following an asymptotic χ ² distribution. To examine short-term dynamic asymmetries in the response of exchange rates to a fall in gold price, we indirectly impose long-term symmetry restrictions θ ⁺ = θ ⁻ = θ, which can be simplified as:

$$∆y_t={\mathit{\rho y}}_{t-1}+{\mathit{\theta x}}_{t-1}+\sum _{i=1}^{p-1}{\gamma }_i∆y_{t-1}+\sum _{i=0}^{q-1}\left({\pi }_i^{+}∆x_{t-i}^{+}+{\pi }_i^{-}∆x_{t-i}^{-}\right)+e_t.$$ (5)

Short-term symmetry constraints can take two forms: (i) π _i ⁺ = π _i ⁻ for all i = 0, ………, q − 1, or (ii) $\sum _{i=0}^{q-1}{\pi }_i^{+}=\sum _{i=0}^{q-1}{\pi }_i^{-}$. When allowing for such restrictions in the existence of a long-term asymmetric relationship, we obtain:

$$∆y_t={\mathit{\rho y}}_{t-1}+{\theta }^{+}{x}_{t-1}^{+}+{\theta }^{-}{x}_{t-1}^{-}+\sum _{i=1}^{p-1}{\gamma }_i∆y_{t-i}+\sum _{i=0}^{q-1}{\pi }_i∆x_{t-i}+e_t$$ (6)

We should restrict the insignificant lags of the first-differenced terms in order to formulate the final/restricted NARDL model according to the principles of the NARDL method. Finally, the most constrained model is attained when assuming nonlinearity in the long-term relationship, as well as short-term asymmetric adjustments (Shin et al., 2014):

$$∆y_t={\mathit{\rho y}}_{t-1}+\theta x_{t-1}+\sum _{i=1}^{\mathrm{p}-1}{\gamma }_i∆y_{t-i}+\sum _{i=0}^{q-1}{\pi }_i∆x_{t-i}+e_t.$$ (7)

We then visually represent the asymmetric, cumulative, and dynamic multiplier effects of a change in x _t ⁺ and x _t ⁻ to graphically reveal the relationship between asymmetric gold price (y _t) and exchange rates (x _t). The cumulative dynamic multiplier effects of x _t ⁺ and x _t ⁻ on y _t is as follows:

$$m_h^{+}=\sum _{j=0}^h\frac{\partial y_{t+j}}{\partial x_t^{+}}=\sum _{j=0}^h{\lambda }_j^{+},m_h^{-}=\sum _{j=0}^h\frac{\partial y_{t+j}}{\partial x_t^{+}}=\sum _{j=0}^h{\lambda }_j^{-},h=0,1,2\dots$$ (8)

$$h\to \infty ,m_h^{+}\to {\beta }^{+}\mathit{and}{m}_h^{-}\to {\beta }^{-},$$

where β ⁺ =  − θ ⁺/ρ and β ⁻ =  − θ ⁻/ρ are the asymmetric long-term coefficients, p, q is the lag order, and h denotes the horizon. One of the advantages of the NARDL method is that it automatically chooses the appropriate lag order for estimation. Because of its inbuilt programming mechanism, the method automatically selected a lag order of three.

For estimation purposes, we applied four steps. First, we conducted a unit root test to confirm our variables are I(1). Second, although the regular OLS was the first point of estimation, we followed a general-to-specific procedure (Sukmana & Ibrahim, 2017) to limit insignificant lags from our model and get a final specification (Table 4). Third, cointegration and asymmetry tests were conducted to determine whether a long-term equilibrium and asymmetric relationship exist between gold and exchange rates. This is done by analyzing f _PSS (Shin et al., 2014) and t _BDM (Banerjee, Dolado, & Mestre, 1998) statistics. Finally, we visualized the cumulative and dynamic multiplier effects of a 1% change in ∆x _t ⁺ and ∆x _t ⁻ to graphically determine the asymmetric relationship between x and y (Shin et al., 2014).

Table 4.

NARDL estimation (restricted) (Exchange Rates → Gold Prices).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
x:	USD → GBP	USD → EUR	USD → CAD	USD → JPY	USD → CHF	USD → NOK	USD → DKK
yt−1	−0.004	0.008	0.005	−0.001	0.004	0.006	0.009
	[0.02]	[0.02]	[0.02]	[0.02]	[0.02]	[0.02]	[0.02]
xt−1+	0.036
	[0.26]
xt−1−		0.042
		[0.64]
∆yt−2	0.184	0.229⁎
	[0.13]	[0.13]
∆yt−3		0.236⁎⁎
		[0.12]
∆xt+	0.450	0.829⁎	−0.696	−0.083	0.564	0.435	0.676
	[0.47]	[0.48]	[0.68]	[0.45]	[0.60]	[0.43]	[0.68]
∆xt−1+		−1.541⁎⁎⁎	0.138			−0.020	−0.833⁎
		[0.51]	[0.55]			[0.25]	[0.46]
∆xt−2+		−1.141⁎					1.141
		[0.62]					[0.79]
∆xt−3+	−0.606	0.132		−0.781	−0.365		−0.573
	[0.51]	[0.53]		[0.61]	[0.49]		[0.48]
∆xt−	1.200⁎⁎⁎	0.563	1.125⁎	0.760⁎⁎	0.951	−0.130	1.740⁎⁎
	[0.43]	[0.51]	[0.59]	[0.34]	[0.76]	[0.29]	[0.68]
∆xt−1−		1.240	0.719	−0.006	0.409
		[0.86]	[0.50]	[0.42]	[0.49]
∆xt−2−					−0.538	0.008	−0.934
					[0.63]	[0.19]	[0.73]
∆xt−3−	0.276	−1.339	−0.802
	[0.35]	[0.91]	[0.59]
Constant	0.032	−0.057	−0.036	0.007	−0.028	−0.045	−0.066
	[0.14]	[0.14]	[0.14]	[0.15]	[0.15]	[0.15]	[0.14]
Statistics and diagnostics:
Obs. (weeks)	67	67	67	67	67	67	67
F-Statistic	2.696	2.612	1.466	1.348	1.033	0.233	2.519
RMSE	0.010	0.010	0.011	0.011	0.011	0.011	0.010
χ2 Serial Corr.	9.571	6.331	2.898	2.447	2.584	4.765	2.403
	(1.00)	(1.00)	(1.00)	(1.00)	(1.00)	(1.00)	(1.00)

Study Period: January 1, 2020 to April 19, 2021; y: Gold Price (GLDP).

Notes: (1) Standard errors are presented in brackets. (2) p-values are noted in parentheses. (3) ^⁎p < 0.1, ^⁎⁎p < 0.05, ^⁎⁎⁎p < 0.01. (4) k, the order of integration, is 1. (5) The superscripts + and − denote positive and negative variations, respectively. (6) The different indicators were individually sampled (not in a panel) and then compiled into the table. (7) STATA omitted insignificant coefficients because we have constrained those to zero. (8) GLDP is a dependent variable, whereas the rest (USD → GBP, USD → EUR, …, USD → DKK) are independent variables, and each independent variable is framed under different equations, in line with the dependent variable (GLDP). (9) GLDP is the gold price, USD→GBP is one United States dollar (USD) to one United Kingdom pound, USD→EUR is one USD to one euro, USD→CAD is one USD to one Canadian dollar, USD→JPY is one USD to one hundred Japanese yen, USD→CHF is one USD to one Swiss franc, USD→NOK is one USD to one Norwegian krone, and USD→DKK is one USD to one Danish krone.

3. Results and discussion

3.1. Unit root test

Both the augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) tests (Table 2 ) suggest that all variables are stationary in their first differenced forms, indicating that we can empirically examine exchange rate and gold price dynamics.

Table 2.

Unit root test.

	Variable	ADF				PP
		Z(t) t-stat.	5% C. V.	p-value	Result	Z(ρ) t-stat.	5% C. V.	Result
Level form	GLDP	−1.594	−2.879	0.4868	NS	−4.292	−14.000	NS
	USD → GBP	−0.964		0.7663	NS	−3.742		NS
	USD → EUR	0.360		0.9800	NS	0.232		NS
	USD → CAD	−0.913		0.7835	NS	−1.787		NS
	USD → JPY	−2.016		0.2794	NS	−10.888		NS
	USD → CHF	−0.865		0.7992	NS	−3.012		NS
	USD → NOK	−0.783		0.8239	NS	−3.076		NS
	USD → DKK	−0.221		0.9360	NS	−1.102		NS
1st difference form	dGLDP	−15.070	−2.883	0.0000	S	−217.485	−13.904	S
	dUSD→GBP	−13.015		0.0000	S	−176.794		S
	dUSD→EUR	−12.446		0.0000	S	−179.290		S
	dUSD→CAD	−14.515		0.0000	S	−185.183		S
	dUSD→JPY	−13.349		0.0000	S	−191.494		S
	dUSD→CHF	−13.235		0.0000	S	−165.782		S
	dUSD→NOK	−12.933		0.0000	S	−171.213		S
	dUSD→DKK	−12.736		0.0000	S	−160.789		S

Notes: (1) “NS” and “S” denote “non-stationary” and “stationary,” respectively. (2) “d” represents first differenced variables. (3) We use the MacKinnon approximate p-value for Z(t). (4) “C. V.” denotes “critical value.” (5) GLDP is the gold price, USD→GBP is one United States dollar (USD) to one United Kingdom pound, USD→EUR is one USD to one euro, USD→CAD is one USD to one Canadian dollar, USD→JPY is one USD to one hundred Japanese yen, USD→CHF is one USD to one Swiss franc, USD→NOK is one USD to one Norwegian krone, and USD→DKK is one USD to one Danish krone.

3.2. Asymmetry and cointegration relationships

The bound tests for cointegration and asymmetry are presented in Table 3 . The t _BDM confirms that exchange rates and gold prices are not cointegrated at the 5% level, denoting the unlikeliness of long-term equilibrium between them, yet the empirical evidence may suggest otherwise. Nevertheless, we expect at least a short-term nonlinear relationship. The f _PSS validates a lack of asymmetry between gold prices and exchange rates. These initial understandings underline the safe-haven properties of gold and the lack of impact of exchange rates on gold prices. Later sections confirm this view.

Table 3.

Bound tests for cointegration and asymmetry.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	5% critical value
x:	USD → GBP	USD → EUR	USD → CAD	USD → JPY	USD → CHF	USD → NOK	USD → DKK	Lower	Upper
tBDM	−0.2066	0.4216	0.2679	−0.0300	0.1842	0.2815	0.4778	−2.86	−3.22
fPSS	0.0298	0.0897	0.0718	−0.0300	0.0339	0.0792	0.2283	4.94	5.73

Notes: (1) Critical values are from Pesaran, Shin, and Smith (2001). (2) GLDP is a dependent variable (y), whereas the rest (USD → GBP, USD → EUR, …, USD → DKK) are independent variables (x). (3) GLDP is the gold price, USD→GBP is one United States dollar (USD) to one United Kingdom pound, USD→EUR is one USD to one euro, USD→CAD is one USD to one Canadian dollar, USD→JPY is one USD to one hundred Japanese yen, USD→CHF is one USD to one Swiss franc, USD→NOK is one USD to one Norwegian krone, and USD→DKK is one USD to one Danish krone.

3.3. Short-term and long-term relationships

Table 4 shows that the seven exchange rates studied do not impact gold prices in the long term, which highlights two findings. First, this result validates the t _BDM cointegration test as to whether a long-term relationship exists between variables (Table 3). Second, exchange rates have comparatively weak predictive power over gold prices, underscoring the status of gold as a safe haven. To confirm whether this result is due to the COVID-19 pandemic, we carry out further analysis of the normal period preceding the crisis.

In the short term, while negative (∆x _t ⁻) USD → GBP increases gold prices, so does positive (∆x _t ⁺) USD → EUR. Worth mentioning is that gold prices were dominated by the European currency unit (ECU)2 throughout the 1980s (Sjaastad & Scacciavillani, 1996). Can we assume the euro, the current version of the ECU and the United Kingdom pound (GBP) have a notable bond with gold prices? However, lagged (1- and 2-week-prior) positive USD → EUR decrease gold prices. Brexit, its consequences, and the rising tension resulting from COVID-19 could explain this result. An additional explanation is that financial markets generally take time to adjust and settle down following a shock.

Negative USD → CAD, USD → JPY, and USD → DKK increase gold prices, but lagged (1 week prior) positive USD → DKK appears to decrease gold prices in the short term. The former result may underscore the fact that COVID-19 has hit almost all financial markets, and the Canadian, Japanese, and Danish markets are no exception. The latter result, however, could be explained by the adjusting nature of the financial markets, as mentioned. Our later tests of the normal period preceding COVID-19 may support these explanations.

We find no evidence that lagged gold prices impact gold prices either positively or negatively. This result calls attention to the efficiency of the gold market, in line with the market efficiency hypothesis and random walk theory (Fama, 1970). Our finding is in agreement with Goldman (2000), who confirmed the efficiency of the gold market, investigating the USD → GBP exchange rate under the gold standard between 1890 and 1906. Our finding, however, contrasts the results that Kirkulak Uludag and Lkhamazhapov (2014) report in the Turkish gold markets and Narayan, Narayan, and Zheng (2010) in the gold-oil futures markets' nexus.

3.4. Additional test: pre-COVID-19 normal period

We validate our findings by examining the normal period preceding COVID-19, from January 1, 2017 to December 31, 2019, to determine whether the effects of seven exchange rates on gold prices during the ongoing COVID-19 crisis are identical to those observed before the crisis.

As shown in Table 5 , the seven exchange rates studied have no impact on gold prices in the long term, as in the COVID-19 period (Table 4). In the short term, positive and lagged positive (1 week prior) USD → JPY, positive USD → CHF, positive USD → NOK, and positive USD → DKK exchange rates have been shown to increase gold prices, although we see no impact from such positive exchange rates with the exception of USD → DKK, during the ongoing pandemic. This casts doubt on the role of exchange rates during the pandemic. However, 1-week-prior and positive USD → EUR and USD → DKK decrease gold prices, a finding which matches the trends observed during the COVID-19-period. This could again be explained by the adjusting nature of the financial markets.

Table 5.

Additional test (restricted NARDL estimation) (Exchange Rates → Gold Prices).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
x:	USD → GBP	USD → EUR	USD → CAD	USD → JPY	USD → CHF	USD → NOK	USD → DKK
yt−1	−0.004	−0.005	−0.002	−0.003	−0.004	−0.003	−0.003
	[0.01]	[0.01]	[0.01]	[0.01]	[0.01]	[0.01]	[0.01]
∆yt−1				−0.131⁎⁎
				[0.06]
∆xt+	0.062	−0.059	0.140	0.931⁎⁎⁎	1.029⁎⁎⁎	0.376⁎⁎	0.536⁎⁎⁎
	[0.13]	[0.21]	[0.19]	[0.20]	[0.22]	[0.18]	[0.20]
∆xt−1+	−0.037	−0.429⁎⁎	0.061	0.454⁎⁎⁎			−0.487⁎⁎
	[0.18]	[0.20]	[0.26]	[0.16]			[0.22]
∆xt−2+		0.228			0.028	−0.022
		[0.25]			[0.19]	[0.16]
∆xt−3+		0.119		−0.188			0.399⁎
		[0.21]		[0.20]			[0.22]
∆xt−	0.426⁎⁎⁎	0.398⁎	0.632⁎⁎⁎	0.785⁎⁎⁎	0.200	0.299	0.323
	[0.16]	[0.21]	[0.23]	[0.24]	[0.28]	[0.21]	[0.25]
∆xt−1−	−0.272	−0.135	0.040			−0.026	0.346⁎
	[0.19]	[0.27]	[0.28]			[0.16]	[0.20]
∆xt−2−	0.489⁎⁎
	[0.20]
∆xt−3−				0.058	−0.058		−0.191
				[0.19]	[0.22]		[0.26]
Constant	0.032	0.038	0.018	0.017	0.031	0.019	0.025
	[0.06]	[0.06]	[0.06]	[0.05]	[0.05]	[0.06]	[0.06]
Statistics and diagnostics:
Obs. (weeks)	156	156	156	156	156	156	156
F-Statistic	3.153	1.719	2.405	8.458	6.670	2.465	4.025
RMSE	0.006	0.006	0.006	0.006	0.006	0.006	0.006
χ2 Serial Corr.	0.401	4.161	0.784	4.112	1.749	1.783	2.008
	1.000	1.000	1.000	1.000	1.000	1.000	1.000

Study Period: January 1, 2017 to December 31, 2019; y: Gold Price (GLDP).

Notes: (1) Standard errors are presented in brackets. (2) p-values are noted in parentheses. (3) ^⁎p < 0.1, ^⁎⁎p < 0.05, ^⁎⁎⁎p < 0.01. (4) k, the order of integration, is 1. (5) The superscripts + and − denote positive and negative variations, respectively. (6) The different indicators were individually sampled (not in a panel) and then compiled into the table. (7) STATA omitted insignificant coefficients because we have constrained those to zero. (8) GLDP is a dependent variable, whereas the rest (USD → GBP, USD → EUR, …, USD → DKK) are independent variables, and each independent variable is framed under different equations, in line with the dependent variable (GLDP). (9) GLDP is the gold price, USD→GBP is one United States dollar (USD) to one United Kingdom pound, USD→EUR is one USD to one euro, USD→CAD is one USD to one Canadian dollar, USD→JPY is one USD to one hundred Japanese yen, USD→CHF is one USD to one Swiss franc, USD→NOK is one USD to one Norwegian krone, and USD→DKK is one USD to one Danish krone.

While we find that short-term negative USD → GBP, USD → EUR, USD → CAD, and USD → JPY, 2-week-prior negative USD → GBP, and 1-week-prior negative USD → DKK increased gold prices in the normal period preceding COVID-19, our findings from the ongoing COVID-19 period present mixed results, with exchange rates having both positive and no impacts on gold prices. In the COVID-19 period, short-term negative USD → GBP, USD → CAD, USD → JPY, and USD → DKK have increased gold prices, but negative USD → EUR, USD → CHF, and USD → NOK have had no impacts on gold prices (Table 4). We thus propose that the uneven and unpredictable effects of exchange rates on gold prices have been due to the stressful situation of the ongoing COVID-19 period.

In sum, we argue that the COVID-19 crisis has contributed to the disharmony in the nexus between exchange rates and gold prices. To some extent, our findings from the normal period preceding COVID-19 contradict our findings from the ongoing COVID-19 period. Thus, we contend that the pandemic is the exclusive cause of those contradictions.

Our finding is in agreement with reports in the literature. Bedoui et al. (2018) show that the co-movement between exchange rates and gold prices differs in normal period than that seen in crisis periods–(1) the Asian crisis of 1997–1998 and the crisis following the Russian debt default in 1998 and (2) the GFC of 2007–2009. In particular, their results suggest that the dependence between USD → GBP, USD → EUR, USD → CAD, USD → JPY exchange rates and gold prices during two crisis periods (i.e., 1 and 2) was stronger than that during normal times. They further find the unusual movement of USD during the GFC of 2007–2009 spurred on the co-movement between those exchange rates and gold prices. Dong et al. (2019) also similarly reveal results for the nexus between exchange rates (USD → GBP and USD → EUR) and gold prices during the GFC differ from those in other periods they investigate. Furthermore, Dong et al. (2019) document that the correlation coefficients between USD → GBP and USD → EUR exchange rates and gold prices have higher absolute values during the GFC than those in other periods they investigate.

In line with the findings during the COVID-19 period, we have found no evidence of short memory, meaning lagged gold prices impacting gold prices, supporting our view that the gold market is efficient and validating the market efficiency hypothesis and random walk theory (Fama, 1970). As stated earlier, our finding is in line with Goldman (2000), who investigates USD → GBP exchange rate under the gold standard between 1890 and 1906. Our finding, however, contrasts Kirkulak Uludag and Lkhamazhapov's (2014) findings in the Turkish gold market and Narayan et al.'s (2010) in the gold-oil futures markets' nexus.

3.5. Diagnostic test

The diagnostic test on data during and before COVID-19 periods (Table 4, Table 5) suggests that the models are free from the serial correlation problem. With that, it confirms the reliability of this paper's findings in formulating strategies.

3.6. Cumulative dynamic multiplier effects

These multipliers show an array of gold price adjustments toward a new long-term equilibrium caused by any positive or negative shock to exchange rates over a 40-week forecast period. The multipliers are generated using the best possible estimation offered by the constrained NARDL method. Fig. 2 (ii and iv) suggests that a negative change only in USD → EUR and USD → JPY would supersede any positive changes over the following 40-week period, whereas a positive change in USD → GBP, USD → CAD, USD → CHF, USD → NOK, or USD → DKK (Fig. 2i, iii, v–vii) would supersede any negative changes. In all cases, the outcome is close to zero.

Fig. 2.

Cumulative dynamic multipliers (Exchange Rates → Gold Prices).

(i) USD → GBP, (ii) USD → EUR, (iii) USD → CAD, (iv) USD → JPY, (v) USD → CHF, (vi) USD → NOK, and (vii) USD → DKK on GLDP.

Notes: (1) 95% bootstrap CI is based on 1000 replications. (2) GLDP is a dependent variable, whereas the rest (e.g., USD → GBP, USD → EUR, …, USD → DKK) are independent variables. (3) The time period is measured in weeks (i.e., 40 weeks). (4) GLDP is the gold price, USD → GBP is one United States dollar (USD) to one United Kingdom pound, USD → EUR is one USD to one euro, USD → CAD is one USD to one Canadian dollar, USD → JPY is one USD to one hundred Japanese yen, USD → CHF is one USD to one Swiss franc, USD → NOK is one USD to one Norwegian krone, and USD → DKK is one USD to one Danish krone.

These graphs confirm a weak asymmetric link and prior findings presented in Table 4. Fig. 2 also confirms why the f _PSS test (Table 3) failed to capture asymmetry between gold prices and exchange rates.

4. Conclusions

This study has asked and answered the central research question of whether and how exchange rates have impacted gold prices during the ongoing COVID-19 crisis. The analysis reveals that none of the leading seven exchange rates has demonstrated a long-term impact on gold prices during or before the crisis. During the crisis period, changes in exchange rates have mostly increased gold prices regardless of the degree and direction of change in the short term. However, exchange rates have, as a result of the stress of the pandemic, sometimes exhibited unpredictable behavior compared to that in the normal period preceding COVID-19. Over the 40-week forecasted horizon, all exchange rates show a weak asymmetric link with gold prices, and the positive impacts of exchange rates most likely outweigh the negative impacts. Although gold prices may have occasionally decreased during the ongoing COVID-19 pandemic and during the normal period prior to the pandemic in the short term, gold's safe-haven properties are most likely to remain intact over time.

This research has the potential to benefit scholars, researchers, investors, and policymakers in this field. For instance, our findings highlight the efficiency of the gold market, in agreement with the market efficiency hypothesis and random walk theory. We endorse the idea that gold can serve as a hedging instrument for nearly 1 year and as a safe-haven asset. Investors may consider adding gold to their portfolios, despite the fact that this study confirms the recent uneven effects of exchange rates on gold prices have been due to COVID-19. Our study follows a suggestion by Harris and Shen (2017) that a price index free of global exchange rates would safeguard gold from changes in such rates. Out of the seven exchange rates, USD to Swiss franc and USD to Norwegian krone have no impact on gold prices. Thus, investors may add these two exchange rates to their investment portfolios to diversify risk in the short term. A dedicated study could confirm this view.

Future research ought to consider an analysis of the nexus between exchange rates and gold prices from a different perspective or by applying a different approach. Researchers also may wish to investigate whether other exchange rates, including those concerning G20 member countries not studied here, could serve as a hedging instrument. If COVID-19 were to continue for years, researchers could test whether gold remains a hedging tool for global equity and debt markets for a year or more in light of this pandemic.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of interest statement

None.

Appendix

A.1. Descriptive statistics/summary of raw data

Variable	Obs.	Mean	Std. Dev.	Min.	Max.	Pearson correlation^	p-value
GLDP	339	1775.256	128.156	1475.030	2052.500	1.0000
USD → GBP	339	1.305	0.054	1.149	1.413	0.3028⁎⁎⁎	(0.0000)
USD → EUR	339	1.156	0.047	1.071	1.234	0.7116⁎⁎⁎	(0.0000)
USD → CAD	339	0.757	0.027	0.689	0.806	0.3048⁎⁎⁎	(0.0000)
USD → JPY	339	0.937	0.019	0.892	0.977	0.7275⁎⁎⁎	(0.0000)
USD → CHF	339	1.074	0.035	1.014	1.138	0.7925⁎⁎⁎	(0.0000)
USD → NOK	339	0.109	0.007	0.085	0.121	0.3676⁎⁎⁎	(0.0000)
USD → DKK	339	0.155	0.007	0.143	0.166	0.7180⁎⁎⁎	(0.0000)

Notes: (1) ^Dependent variable (GLDP) versus independent variable (USD → GBP, USD → EUR, …, USD → DKK). (2) p-values (of the Pearson correlation) are noted in parentheses. (3) ^⁎p < 0.1, ^⁎⁎p < 0.05, ^⁎⁎⁎p < 0.01. (4) All variables used in this study have 224 observations/data points. (5) GLDP is the gold price, USD→GBP is one United States dollar (USD) to one United Kingdom pound, USD→EUR is one USD to one euro, USD→CAD is one USD to one Canadian dollar, USD→JPY is one USD to one hundred Japanese yen, USD→CHF is one USD to one Swiss franc, USD→NOK is one USD to one Norwegian krone, and USD→DKK is one USD to one Danish krone.

A.2. Unrestricted NARDL estimation (Exchange Rates → Gold Prices)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
x:	USD → GBP	USD → EUR	USD → CAD	USD → JPY	USD → CHF	USD → NOK	USD → DKK
yt−1	0.001	0.009	0.002	−0.009	−0.009	0.000	0.000
	[0.02]	[0.02]	[0.02]	[0.02]	[0.02]	[0.02]	[0.02]
xt−1+	−0.909⁎⁎	0.117	0.114	−0.855	−0.464	0.028	0.028
	[0.45]	[0.54]	[0.63]	[0.74]	[0.59]	[0.35]	[0.35]
xt−1−	0.551	−1.276⁎⁎	−0.623	−0.076	0.956	0.009	0.009
	[0.43]	[0.63]	[0.71]	[0.53]	[0.57]	[0.25]	[0.25]
∆yt−1	−0.096	0.044	−0.061	−0.064	−0.001	−0.004	−0.004
	[0.14]	[0.14]	[0.15]	[0.16]	[0.14]	[0.16]	[0.16]
∆yt−2	0.252⁎	0.238⁎	0.179	0.111	0.028	0.188	0.188
	[0.14]	[0.13]	[0.16]	[0.15]	[0.16]	[0.16]	[0.16]
∆yt−3	0.143	0.234⁎	0.058	0.138	0.152	0.134	0.134
	[0.13]	[0.12]	[0.13]	[0.14]	[0.13]	[0.14]	[0.14]
∆xt+	0.330	0.846⁎	−0.400	−0.093	1.047	0.514	0.514
	[0.51]	[0.49]	[0.83]	[0.57]	[0.68]	[0.46]	[0.46]
∆xt−1+	1.119	−1.672⁎⁎		0.640	−1.017
	[0.71]	[0.76]		[1.20]	[0.89]
∆xt−2+	0.975	−1.300⁎	−0.279	1.777	1.468	−0.149	−0.149
	[0.62]	[0.76]	[0.90]	[1.20]	[1.00]	[0.54]	[0.54]
∆xt−3+			−0.448			−0.279	−0.279
			[0.83]			[0.57]	[0.57]
∆xt−	1.235⁎⁎⁎	0.480	1.135⁎	0.873	0.940	−0.088	−0.088
	[0.46]	[0.58]	[0.65]	[0.57]	[0.80]	[0.33]	[0.33]
∆xt−1−	−0.497	2.516⁎⁎	1.580⁎			0.375	0.375
	[0.70]	[0.99]	[0.87]			[0.34]	[0.34]
∆xt−2−	−1.126	1.260	0.794	0.050	−2.223⁎⁎
	[0.75]	[0.95]	[0.93]	[0.92]	[1.02]
∆xt−3−				0.366	−0.629	−0.120	−0.120
				[0.85]	[0.85]	[0.48]	[0.48]
Constant	−0.003	−0.066	−0.008	0.070	0.066	−0.002	−0.002
	[0.15]	[0.15]	[0.15]	[0.17]	[0.15]	[0.16]	[0.16]
Obs. (weeks)	67	67	67	67	67	67	67
tbdm	0.0457	0.4710	0.0748	−0.4140	−0.4141	0.0189	0.0189
fpss	1.4493	1.6713	0.2572	0.5306	1.1840	0.0023	0.0023
F-Statistic	1.839	2.363	0.931	0.897	1.458	0.651	0.651
RMSE	0.010	0.010	0.011	0.011	0.011	0.011	0.011
χ2 Serial Corr.	6.247	6.543	4.366	3.750	2.462	5.078	5.078
	(1.00)	(1.00)	(1.00)	(1.00)	(1.00)	(1.00)	(1.00)

Study Period: January 1, 2020 to April 19, 2021; y: Gold Price (GLDP).

Notes: (1) Standard errors are presented in brackets. (2) p-values are noted in parentheses. (3) ^⁎p < 0.1, ^⁎⁎p < 0.05, ^⁎⁎⁎p < 0.01 . (4) (4) k, the order of integration, is 1. (5) The superscripts + and − denote positive and negative variations, respectively. (6) The different indicators were individually sampled (not in a panel) and then compiled into the table. (7) STATA omitted insignificant coefficients because we have constrained those to zero. (8) GLDP is a dependent variable, whereas the rest (USD → GBP, USD → EUR, …, USD → DKK) are independent variables, and each independent variable is framed under different equations, in line with the dependent variable (GLDP). (9) GLDP is the gold price, USD→GBP is one United States dollar (USD) to one United Kingdom pound, USD→EUR is one USD to one euro, USD→CAD is one USD to one Canadian dollar, USD→JPY is one USD to one hundred Japanese yen, USD→CHF is one USD to one Swiss franc, USD→NOK is one USD to one Norwegian krone, and USD→DKK is one USD to one Danish krone.