key: cord-0846910-7ha9mngd
authors: Tanin, Tauhidul Islam; Sarker, Ashutosh; Brooks, Robert; Do, Hung Xuan
title: Does oil impact gold during COVID-19 and three other recent crises?
date: 2022-03-02
journal: Energy Econ
DOI: 10.1016/j.eneco.2022.105938
sha: 804f55fa8b3ea0791c057ec07bb7739787d730a3
doc_id: 846910
cord_uid: 7ha9mngd

The ongoing COVID-19 pandemic has inspired an examination of the oil–gold prices nexus during four recent crises: the COVID-19 pandemic, the gold market crash, the European sovereign debt crisis, and the global financial crisis. Using daily data from May 2007–August 2021, we employ the nonlinear autoregressive distributed lag method to reveal five novel findings. First, this study contrasts with much of the literature, which infers that the relationship between oil and gold prices is strongly positive. Second, we find no oil and gold price relationship in the long term during all the crisis periods. Third, oil prices have substantially lost their power to predict gold prices in recent times and the oil–gold price linkage is not functional across all crisis periods. Fourth, in the short term, only negative Brent and negative West Texas Intermediate price changes cause positive gold price changes during the pandemic and gold market crash, respectively. Fifth, Brent prices have shown no link to gold prices before COVID-19. We argue that gold prices are less sensitive to oil prices than ever, and the uncertainty resulting from the COVID-19 crisis has attracted investors to gold. Our main findings hold under robustness analyses using fractional cointegration/integration models.

The novelty of this study is sixfold. First, this study serves as the first examination of the effect of oil prices on gold prices during the period in question, including COVID-19, GMC, ESDC, and GMC, allowing for a unique investigation of the nexus of the price of oil, a major global energy source, and gold prices during periods of heightened financial uncertainty. Second, this study finds that while oil prices used to enhance gold prices, the relationship has weakened in recent years. Third, this study relates to the work of Antonakakis and Kizys (2015) , who suggest that the impacts of oil prices on gold prices are timeand event-dependent. Fourth, the results demonstrate that not all crude oil prices contribute evenly to gold prices; hence, using one crude oil price as a proxy for all could produce misleading results, validating the argument of Mann and Sephton (2016) that there is presently no global crude oil benchmark. Fifth, we find that different crisis periods show varying results; hence, examining each period separately is crucial.

Sixth, we argue that the current positive impact of Brent oil prices on gold prices is due to the widespread adversity and uncertainty of the ongoing COVID-19 crisis.

The remaining parts of this paper is organized as follows. We describe the data and research methods in section 2. Section 3 discusses the empirical results emerged from the NARDL approach and subsequently, presents the robustness analyses derived using the fractional cointegration/integration models. Finally, section 4 delivers concluding remarks. Table 1 provides an overview of dependent and independent variables. We transform oil prices (x) and gold prices (y) into natural logarithmic forms. visualizes the choice of these crisis periods. This approach of dividing data into different horizons helps us achieve a more granular and coherent view of the outputs. Fig. 1 (i-iv) presents gold prices and three oil prices during the horizons under investigation. During COVID-19, gold prices have so far fluctuated between USD 1475.03 and 2052.5; the highest value was on August 6, 2020 (Fig. 1i) . The gold prices sharply declined from USD 1692.84 to USD 1051.97 during the GMC with a downward trend, significantly increased from USD 868.1 to 1898.25 during the ESDC with an upward trend, and find fluctuations between USD 642.45 and 1011.6 during the GFC (Fig. 1i) .

Furthermore, all oil prices have significantly fluctuated during those four horizons while the lowest and highest oil prices were USD 14.24 and 141.07, and USD 20.66 and 140.56 for Brent, WTI, and Dubai oil prices, respectively . We can see that WTI has experienced negative WTI oil prices in history for the first time during COVID-19, precisely on April 20, 2020. The drops in oil prices were quite substantial and showed extreme volatility. 

The NARDL model allows us to study the asymmetric relationship between gold prices ( ) and oil prices ( ), expecting asymmetric adjustments throughout the business cycle (De Long and Summers, 1986; Falk, 1986; Neftci, 1984) , given that Apergis and Eleftheriou (2016) have recently shown that business cycles asymmetrically affect gold prices. The NARDL method (Shin et al., 2014 ) is a nonlinear, dynamic, and asymmetric model that can differentiate between short-and long-term effects. The short-term estimation measures the immediate impacts of changes in the exogenous variable on the dependent variable, while the long-term estimate assesses the reaction time and speed of adjustment toward a level of equilibrium.

Although the nonlinear threshold vector error correction model can capture these features, it exhibits a convergence issue when the parameters proliferate. The NARDL model, on the other hand, is free from such a problem and relaxes the typical restrictions in that it does not require the same order of integration for a particular time series of variables (Apergis and Cooray, 2015) . Additionally, a bounds testing approach, such as 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:

where is a vector of multiple regressors defined in such a way that is the autoregressive parameter, and are the asymmetrically distributed lag parameters, and is an i.i.d.

process with zero mean and constant variance . First, the long-term equation is the following:

where (gold prices) and (oil prices) are scalar variables. Here, is decomposed as , where and are partial sum processes of the positive and negative changes 4 in :

The symmetric short-term coefficients are examined using the Wald test, following an asymptotic χ 2 distribution. To test short-term dynamic asymmetries in the response of oil prices to a fall in gold prices, we indirectly impose the long-term symmetry restrictions , which can be simplified as

Short-term symmetry constraints can take two forms: (i) for all or (ii) ∑ ∑ . When allowing such restrictions in the existence of an asymmetric long-term relationship, we obtain the following:

Following the principles of the NARDL model, we limit insignificant lags of the first-differenced terms to formulate the final NARDL. With that, the most restricted model is attained when assuming nonlinearity of the long-term relationship in combination with short-term asymmetric adjustments (Shin et al., 2014) :

Finally, to graphically illustrate asymmetry, we visually represent the cumulative dynamic multiplier effects of a change in and to expose the nexus between asymmetric gold prices ( and oil prices ( . The asymmetric and cumulative dynamic multiplier effects of and on are as follows:

where and are the asymmetric long-term coefficients, , is the lag order, and denotes the horizon. The original Shin et al. (2014) approach that accommodates stepwise regression determines the optimal lag order for this study. 4 Notably, since our focus is on the effect of the oil prices on gold prices and whether that effect is asymmetric in the essence of negative and positive oil price changes, we chose a threshold of zero to distinguish these two zones. Economically, these changes can be considered negative and positive oil shocks, respectively. These two zones are a common interest in analyzing oil markets' asymmetric effect on other markets (see, e.g., Akinsola and Odhiambo, 2020; Charfeddine and Barkat, 2020; Hashmi et al., 2021; Narayan and Gupta, 2015) . In other words, we based our choice of a zero threshold on economic intuition rather than statistical reasoning. This was also the motivation of Shin et al. (2014) in choosing a zero threshold for the original NARDL model.

We followed four steps in conducting the NARDL estimation, following Tanin et al. (2022 Tanin et al. ( , 2021b Tanin et al. ( , 2021c . First, we performed unit root tests to confirm whether our variables were I(1). Second, although the classic OLS was the first estimation point, we followed a general-to-specific procedure (Sukmana and Ibrahim, 2017) to limit insignificant lags from our model and achieve the final specification (presented in   Tables 4-7) . Third, we conducted NARDL cointegration and asymmetry tests by analyzing the f-statistic (f PSS ) (Shin et al., 2014) and the t-statistics (t BDM ) (Banerjee et al., 1998) to determine whether a long-term relationship exists between gold prices and oil prices. Finally, we depicted the cumulative dynamic multiplier effects of a 1% change in and to vividly determine the asymmetric relationship between (oil prices) and (gold prices) (Shin et al., 2014) .

As previously mentioned, we first conducted unit root tests. We confirm that our dependent variable (gold prices) and independent variables (oil prices) are non-stationary, except for the WTI prices during the Notably, the stationarity of WTI prices during the COVID-19 period is supported by ADF-GLS test but rejected by the ADF and the PP test. We also conducted the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test and found that the KPSS test rejects the stationarity of the WTI prices during the COVID-19. Besides, the estimated fractional degree of WTI prices during the COVID-19 is around 0.8 (see Table 9 , subsection 3.5.1), which is greater than 0.5 and quite close to 1, indicating that the series's dynamic is closer to the boundary of nonstationarity. Regarding the ESDC period, there is no strong evidence of stationarity for the oil prices. While the ADF tests support the stationarity of the oil prices (see Table 2A -III), the PP test and ADF-GLS test consistently show that these oil prices are nonstationary (see Table 2A-III and Table 2B ). In addition, the estimated fractional degrees for these oil prices are very close to 1 (see Table 9 in subsection 3.5.1), indicating that these oil prices are likely to be nonstationary during the ESDC period. The KPSS test results are available upon request. 6 We are indebted to the reviewers for the useful suggestion that we incorporate the ADF-GLS test in this study.

J o u r n a l P r e -p r o o f 

To proceed with the estimation, we test for asymmetric cointegration relationships among the variables.

The tests for cointegration (t BDM ) and asymmetry (f PSS ) are presented in Table 4 . The t-statistics of the BDM test (t BDM ) suggests that gold prices and oil prices are not cointegrated at the 5% level, but NARDL estimation-a much superior model-may suggest otherwise. The f-statistics of the PSS test (f PSS )

suggests the non-existence of a long-term asymmetric relationship or an asymmetric cointegration between gold prices and oil prices, except for the GMC period (bold). These results infer that the cointegration and asymmetry between gold prices and oil prices could be weak. We seek to substantiate this view through NARDL estimation. J o u r n a l P r e -p r o o f

Using Brent, WTI, and Dubai oil prices, which are all spot prices, we examined both the short-and longterm impacts of oil prices on gold prices. We first run an estimation for COVID-19 followed by other crisis periods.

During COVID-19, we find no long-term relationships between oil and gold prices ( (2017), and Lee and Lin (2012) , who argue that oil prices determine gold prices. The authors did not clarify whether this oil-gold price nexus is in short-or long-term considerations. Instead, we find that the oil-gold price nexus may exist only in the short term if not in the long term; all oil price changes do not, however, influence shortterm gold price changes.

This finding is somewhat in line with those of Apergis and Eleftheriou (2016) , who argue that gold prices increase more during the recessionary phase of the business cycle (because of oil prices, consumer prices, and macroeconomic factors) than during the boom phase. The COVID-19 crisis has caused oil prices to fall and has brought about uncertainty, volatility, and economic and financial consequences on a global level, potentially affecting the decisions of investors. In addition, the ongoing COVID-19 period may be considered a recession or may usher in a recession. However, we find that gold prices already indicate a recession, being positively affected by negative Brent prices at the time of the current pandemic.

We see that gold prices are not dependent on their own lags, in the short and long terms. By examining results over the other time horizons studied, we thus seek to confirm whether the ongoing pandemic is the exclusive cause of this scenario.

J o u r n a l P r e -p r o o f 

During the GMC, we again find no connection between gold prices and the three oil prices in the long term (Table 5 ). In the short term, negative WTI price changes cause positive gold price changes. During COVID-19, we find that negative Brent price changes result in positive gold price changes, which is true for negative WTI price changes during the GMC. This result infers that oil prices weakened and/or failed to determine gold prices during COVID-19.

We further note that gold prices depend on their own lags, and lagged gold prices negatively affect gold prices in the long term. This is valid for all oil prices. The long-term effects of lagged gold prices signify the presence of long memory, indicating that past oil prices could curb future oil prices over the long term (Kirkulak Uludag and Lkhamazhapov, 2014) . This hypothesis violates the efficient market assumption and is contrary to a martingale or random walk behavior (Fama, 1970 ), yet we find no such issue during COVID-19.

J o u r n a l P r e -p r o o f J o u r n a l P r e -p r o o f

Like the COVID-19 and GMC periods, we find no long-term impacts of three oil prices on gold prices during the ESDC (Table 6) . Nevertheless, while positive WTI and Dubai oil price changes cause positive gold price changes, lagged negative WTI price changes cause negative gold prices in the short term.

Lagged variables could represent investors' sentiments and future expectations and hence may justify the latter result.

Unlike the COVID-19 and GMC periods, during the ESDC, we see that lagged gold price changes caused negative gold price changes in the short term, highlighting the presence of short memory in the gold market. Furthermore, we find that lagged gold prices decrease gold prices during the GMC in the long term (Table 5 ), but we find otherwise (they increase gold prices) during the ESDC. These results again validate the presence of long memory in the gold market. Both short and long memory, however, violate the efficient market hypothesis and are contrary to a martingale or random walk behavior (Fama, 1970) . 

From Table 7 , we find no long-term impacts of three oil prices on gold prices, which is consistent with the COVID-19 and GMC periods. Like the ESDC, we see that positive WTI and (lagged) Dubai oil price changes cause positive gold price changes in the short term during the GFC. While negative Dubai oil price changes cause positive gold price changes, negative WTI also shows the same. The former result is consistent with the ESDC (Table 6 ), but the latter contrasts with that period. The dissimilar observation during the ESDC could be an exception to the general findings of our study. Our results may confirm that oil price changes similarly impact gold price changes, discarding the only exception during the ESDC.

We could conclude this with greater certainty through analyzing the findings from robustness tests in the FCVAR model. During the GFC, we only find short memory in the gold market and only on one occasion, where Dubai experienced a lagged distortion of gold prices, violating the efficient market hypothesis and acting contrary to a martingale or random walk behavior (Fama, 1970) . Nevertheless, this result is in line with the findings of Kirkulak-Uludag and Lkhamazhapov (2014), highlighting distrust in the weak form efficiency of the gold market, although they observe this for the Turkish gold market. The following robustness tests better illustrate this finding.

3.3.5 Summary of the results and further discussion In the short term, Brent price changes cause gold price changes only during COVID-19, signifying that COVID-19 may be the exclusive cause of this impact. Clearly, Brent prices have a weak affiliation with, or no effect on, gold prices, except for during the ongoing COVID-19 crisis. Hence, a study of only one or two oil prices, or else of an entire sample in one setting (i.e., without examining sub-samples or different periods separately), for example, as in Sephton and Mann (2018) , may produce misleading results.

We further find that WTI oil price changes impacted gold price changes only during the GMC (Jan 2013- periods only on two and one occasion(s), respectively. The short-and long-term impacts of lagged gold prices during the GMC, the ESDC, and the GFC are attributable to those crises. These findings highlight that the presence of short and long memory could be spurious, rather than being a property of the data.

During the GMC and the ESDC, and partially during the GFC, our results sharply contrast with the hypothesis of random walk behavior (Fama, 1970 ) and indicate weak form efficiency within the gold market.

Our results confirm that oil-gold prices nexus has weakened over time. Although oil prices significantly and positively impacted gold prices in the past, except for lagged negative WTI prices over the ESDC period 7 , that is no longer the case in recent times. During GMC, we see almost no bond between oil and gold prices. During COVID-19, oil prices appear to enhance gold prices only on one occasion. The COVID-19 result, as caused by Brent oil prices for the first time, could be explained by the chaos, tension, and fear created by the ongoing COVID-19 pandemic. The results suggest that oil prices have had minimal to no ability to predict gold prices in recent times. The above might explain why the BDM test in Table 3 failed to find cointegration between oil and gold prices.

Studies claiming that oil prices determine gold prices have relied mostly on a backdated sample period and used typical (but not necessarily robust) methods. For example, Lee and Lin (2012) Nevertheless, after ESDC, the dynamics change drastically.

Based on our robust and consistent estimation, we deduce that not all oil prices contribute evenly to gold prices, and not all sub-samples or crisis periods behave identically. This supports the finding of Mann and Sephton (2016) that there is presently no global crude oil benchmark. Studies using one crude oil price as a proxy for all and investigating only the entire sample, discarding the attention to the sub-samples, could produce misleading results. In sum, results based only on a full sample may be less accurate. For example, only during COVID-19, negative Brent price changes cause positive gold price changes. If we were analyzing the full sample, we could get biased results. Therefore, analyzing sub-samples is crucial for more robust findings. 

Fig. 2 (a-e: i-iii) graphically presents the findings by tracking asymmetric dynamic multipliers for the four crisis periods. These multipliers show how temporal adjustments to gold prices shifted to a new long-term equilibrium caused by any positive or negative shock to oil prices over the forty weeks.

We follow Shin et al. (2014) to calculate the multipliers using the constrained NARDL model specifications presented in section 3.3, finalized applying the stepwise regression approach. The dashed green and red lines represent positive and negative changes in exchange rates toward the adjustment of gold prices, respectively. The changes presented are based on 1,000 replications. The continuous blue line shows the difference between cumulative dynamic multipliers in terms of negative and positive shocks.

The shaded blue area shows the upper and lower bounds of a 95% bootstrapped confidence interval, providing an estimation of the significance of the asymmetry. 9

The figures show that the impacts of positive oil prices on gold prices are relatively more substantial than the effects of negative oil prices during COVID-19, GMC, ESDC, and GFC. These results validate the study of Kumar (2017) , who investigated for a period of April 1990-April 2016 and found that a positive shock in oil prices is more persistent (than a negative shock in oil prices) on gold prices. However, the opposite is also true for the GMC for the impact of Dubai prices on gold prices (Fig. 2b: iii) .

Meanwhile, the asymmetry is present for forty weeks for all cases, but the impact of those asymmetry seems weak and mostly negligible. What is the takeaway from these findings? The link between oil prices and gold prices seems weak, validating our findings of the PSS test in Table 3 . These graphs further confirm the results of our estimation (Tables 4, 5 , 6, and 7). 

The NARDL modeling framework we employ so far can only account for the integer degrees of integration such as I(1) in the gold and oil prices. Therefore, to ensure the consistency of our main J o u r n a l P r e -p r o o f findings in case the integration degrees of our variables are fractional, we employ the FCVAR model proposed by Johansen (2008) and then analyzed by Johansen and Nielsen (2012) . 10 The rank test within the FCVAR framework can help identify whether there is any long-term relationship among the endogenous variables in the system, considering the possibility that their degrees of integration are in fractional values. The bivariate FCVAR model can be specified as follows:

where is the vector of endogenous variables, collecting the gold prices ( ) and oil prices ( ),

. As mentioned earlier, to answer the question of whether there is any, and if yes, how many long-term relationships are, we perform the cointegration rank test based on the likelihood ratio test statistic within the FCVAR framework. The null hypothesis of, is tested against because there are two variables in each of our FCVAR systems. can be interpreted as the number of long-term relationships in the system. We perform 12 rank tests in total for four crisis periods and three oil products in each period. 11 We present the results in Table 9 . As seen, in all cases, the tests fail to reject the null hypotheses of rank 0, consistently indicating that there is no long-term relationship between the gold prices and oil prices. This result is highly consistent with our findings using the NARDL approach presented in sections 3.2 and 3.3 regarding the cointegration relationship between oil and gold prices. In other words, we find no long-term relationship between oil and gold prices during the four recent crisis periods (COVID-19, GMC, ESDC, and GFC) we investigate, as evidenced by NARDL and FCVAR models. --------Notes: (1) This table summarizes the results for cointegration rank tests between the gold and oil prices within the FCVAR framework of Johansen and Nielsen (2012) . (2) The lag order is chosen using the AIC criteria.

(3) Stars, if any, indicate the statistical significance such that, * p < 0.1, ** p < 0.05, *** p < 0.01, calculated based on the White heteroskedasticity-consistent standard errors. (4) GLDP denotes gold prices. Brent indicates the Brent crude oil, WTI denotes the West Texas Intermediate crude oil, and Dubai refers to the Dubai crude oil prices.

As there is no evidence on the long-term relationship between oil and gold prices, we now perform robustness analyses on the short-term relationship between oil and gold prices using the change in the prices, denoted as (for gold) and (for oil). To account for the asymmetric effects of oil price changes on gold price changes as well as the possible difference between their fractional degrees of integration, we estimate the following univariate fractionally integrated model:

where are the fractional degrees of the and , respectively. We employ the two-step estimation method in a similar spirit of Yip et al. (2017) and Do et al. (2014) , in which the fractional degrees are estimated in the first stage, and the remaining parameters are estimated in the second stage of the estimation procedure.

We present the estimation results for 12 cases (four sub-samples and three oil products in each sample) in Table 10 . As the table shows, the results remain highly consistent with the main findings presented in This result is stronger than that obtained from the NARDL method, but it is consistent in the nature of the relationship.

(3) During the ESDC period, strong evidence indicates a short-term positive effect of oil prices on gold prices. More specifically, the fractional integration results suggest that negative WTI oil price changes positively impact gold price changes in four cases during the ESDC. This fractional integration result is much more consistent and more in line with the economic intuition than that obtained from the NARDL method. As a result, we lean on this fractional integration result and conclude that WTI oil prices positively impact gold prices in the short term during ESDC. (4) During the GFC period, we also find strong evidence that oil price changes positively impact gold price changes. Consistent with the NARDL approach's analyses, we find that not all oil prices evenly impact gold prices in all crisis periods.

J o u r n a l P r e -p r o o f In the main analysis, we employ the original approach of Shin et al. (2014) in choosing the final specifications for the NARDL models, using the stepwise regression procedure to determine the maximum lag length and dropped predictive variables. Following this, our NARDL models determine a maximum lag length of three, which may raise a concern that the included lag information may not sufficiently capture day-of-the-week effects. Moreover, as per the standard stepwise regression procedure, the standard errors have not yet been adjusted to be heteroskedasticity-consistent. In other words, if heteroskedasticity is present, the statistical inference of our chosen NARDL models can be biased. 12

To address the two abovementioned concerns, we reestimate our NARDL models with a lag length of five and, at the same time, use the White heteroskedasticity-consistent errors. We summarize the estimated outputs for all 12 cases (four subsamples and three oil products in each sample) in Table 11 . In general, most of the main conclusions we drew from the NARDL's results in section 3.3 and the fractional (co)integration approach in section 3.5.1 remain consistent. These include the following: (1) there is evidence that negative Brent oil price changes have a positive impact on the gold price changes;

(2) there is strong evidence that the oil price changes had a positive impact on gold price changes during the GFC and the ESDC periods; and (3) not all oil prices evenly affected gold prices in the four crisis periods. Furthermore, as is consistent with the main analysis, the overall result in this robustness analysis does not support a long-term impact of the oil prices on gold prices across the four crisis periods.

Nevertheless, we find two occasions showing statistical evidence of this relationship, including the effect of Brent (WTI) prices on gold price changes during COVID-19 (ESDC). Considering their economic significance, however, we find that these long-term effects were approximately six to ten times less than the corresponding short-term effects. For example, the long-term effect of a 1% increase in WTI oil prices during the ESDC periods was estimated to result in a gold return increase of approximately 0.01%, while the short-term effect was estimated to result in an increase of between 0.1% and 0.19%. Therefore, our analyses do not provide strong evidence that oil prices have had a long-term impact on gold prices across the four recent crisis periods. 12 We thank an anonymous referee for this suggestion.

J o u r n a l P r e -p r o o f 

We investigated the nexus between gold and three oil prices using time series data for the ongoing COVID-19 crisis, the gold market crash (GMC), the European sovereign debt crisis (ESDC), and the global financial crisis (GFC). We find no strong evidence to support a long-term impact of oil prices on gold prices during four recent crisis periods, in contrast to Sephton and Mann (2018) . Nevertheless, in the short term, while only negative Brent price changes cause positive gold price changes only during COVID-19, negative WTI price changes did the same during the GMC. While positive WTI and Dubai price changes cause positive gold price changes, negative WTI price changes cause gold prices changes during ESDC. Regardless of changes in WTI and Dubai prices, they cause positive gold price changes during GFC. All these eventualities happen in the short term. Except for the COVID-19 period, in all time horizons, Brent prices (or price changes) do not significantly impact gold prices (or price changes), either positively or negatively, and not all oil prices (or price changes) evenly impact gold prices (or price changes) in all crisis periods. It indicates that studying one or two proxies of crude oil prices and/or failing to examine sub-samples may present misleading evidence. Our robustness analyses using the fractionally integrated and fractionally cointegrated frameworks ensure the consistency of our main findings.

It is also evident that gold prices depend on their own lags in some instances. During the GMC and in the cases of Brent, WTI, and Dubai prices, lagged gold prices negatively impacted gold prices in the long term. During the ESDC, lagged gold prices positively impact gold prices in the long term, while lagged gold price changes cause negative gold price changes in the short term. During the GFC, lagged gold price changes negatively impacted gold price changes in the short term, but only on one occasion. These results suggest the presence of long memory during GMC and ESDC and short memory during ESDC and GFC (only on one occasion), indicating weak form efficiency within the gold market only during the GMC and ESDC and partially during GFC. The cumulative dynamic multipliers support the finding that the relationship between oil prices and gold prices is somewhat weak, although the asymmetry was present for forty weeks in all cases.

This study suggests that asymmetric, dynamic, and nonlinear relationships exist between oil and gold prices. Although oil prices impacted gold prices positively in the past, the oil-gold prices nexus has weakened over time. During GMC and COVID-19, a very minimal connection exists between oil and gold prices. During the COVID-19, oil price changes enhanced gold price changes on one occasion.

Notably, Brent oil price changes impacted gold price changes for the first time. The chaos, tension, and fear spread by the present COVID-19 pandemic may explain this result. We therefore argue that in recent times, oil prices have had minimal to no ability to predict gold prices.

This study contrasts with the findings of Singh et al. (2019), Shahzad et al. (2019) , Kumar (2017) , and Lee and Lin (2012) that oil prices determine gold prices. We argue that while oil prices had a functional link with gold prices before the GMC period, the relationship has proven nearly inexistent in recent times, J o u r n a l P r e -p r o o f except for one case of Brent prices only during COVID-19 and WTI prices during the GMC-all of which have positively contributed to gold prices. We, therefore, conclude that changes in gold prices caused by oil prices are time-and event-dependent, supporting the arguments of Antonakakis and Kizys (2015) . During the present period of financial volatility and uncertainty, investors must take this oil-gold prices nexus into account. Furthermore, we find that gold exhibits safe-haven characteristics during COVID-19 consistent with Tanin et al. (2021b) and Salisu et al. (2020) , although Disli et al. (2021) discover otherwise.

This study is constrained to an analysis of WTI, Dubai, and Brent prices owing to inadequate availability of oil price data (i.e., proxies) in the records. Future research could be extended to include Maya and Refiner Acquisition Cost oil prices.

We contend that gold prices are now less sensitive to oil prices, and, in line with the proposition of Selmi et al. (2018), we suggest that investors are likely to park their investments in gold during turbulent periods such as the current COVID-19 crisis. These new and robust findings are useful for investors, policymakers, and researchers seeking to explore links between oil and gold prices during economic downturns and periods of financial uncertainty such as those caused by the ongoing COVID-19 crisis, the gold market crash, the European sovereign debt crisis, and the global financial crisis. Considering policy implications, if there are no long-term relationships between gold and oil prices, national financial and energy policies can be planned independently (Chang et al., 2013; Seyyedi, 2017) . Our robust results suggest there are no long-term relationships between gold and oil prices, and thus, we argue that national energy and financial policies can be created independently. Consistent with Mann and Sephton (2016) , we find that no global crude oil benchmark exists. Thus, we infer that central bank officials, dealers, traders, policymakers, and other stakeholders should closely watch all three spot crude oil prices.

Does gold offer a better protection against losses in sovereign debt bonds than other metals?

Asymmetric effect of oil price on economic growth: panel analysis of low-income oil-importing countries

Co-movement measure of information transmission on international equity markets

Dynamic spillovers between commodity and currency markets

Asymmetric interest rate pass-through in the US, the UK and Australia: New evidence from selected individual banks

Gold returns: Do business cycle asymmetries matter? Evidence from an international country sample

Does uncertainty move the gold price? New evidence from a nonparametric causality-in-quantiles test

Error-correction mechanism tests for cointegration in a singleequation framework

Is gold a hedge or a safe haven? An analysis of stocks, bonds and gold

Is gold a safe haven? International evidence

Gold price dynamics and the role of uncertainty

Does gold act as a hedge or a safe haven for stocks? A smooth transition approach

The effects of uncertainty measures on the price of gold

Measuring the response of gold prices to uncertainty: An analysis beyond the mean

Cointegration and nonlinear causality amongst gold, oil, and the Indian stock market: Evidence from implied volatility indices

Interactive relationships between crude oil prices, gold prices, and the NT-US dollar exchange rate-A Taiwan study

Short-and long-run asymmetric effect of oil prices and oil and gas revenues on the real GDP and economic diversification in oil-dependent economy

The changing cyclical variability of economic activity in the United States

In search of safe haven assets during COVID-19 pandemic: An empirical analysis of different investor types

Is 'oil and gas' industry of ASEAN5 countries integrated with the US counterpart?

How does trading volume affect financial return distributions?

Efficient tests for an autoregressive unit root

Further evidence on the asymmetric behavior of economic time series over the business cycle

Efficient capital narkets: A review of theory and empirical work

Stock markets and gold suffer a June to forget

The bubble contagion effect of COVID-19 outbreak: Evidence from crude oil and gold markets

COVID-19 and finance: Agendas for future research

Is gold a safe haven against equity market investment in emerging and developing countries?

The intrinsic value of gold: An exchange rate-free price index

Asymmetric effect of oil prices on stock market prices: New evidence from oil-exporting and oil-importing countries

Directional spillover effects and time-frequency nexus between oil, gold and stock markets: Evidence from pre and during COVID-19 outbreak

Dynamic linkages among oil price, gold price, exchange rate, and stock market in India

Searching for safe-haven assets during the COVID-19 pandemic

A representation theory for a class of vector autoregressive models for fractional processes

Likelihood inference for a fractionally cointegrated vector autoregressive model

Long memory and structural breaks in the returns and volatility of gold: Evidence from Turkey

On the nonlinear relation between crude oil and gold

Threshold effects in the relationships between USD and gold futures by panel smooth transition approach

Global relationships across crude oil benchmarks

Dynamic risk spillovers between gold, oil prices and conventional, sustainability and Islamic equity aggregates and sectors with portfolio implications

Impact of COVID-19 outbreak on asymmetric multifractality of gold and oil prices

Gold price and stock markets nexus under mixed-copulas

A Matlab program and user's guide for the fractionally cointegrated VAR model (No. 1330), Queen's Economics Department Working Paper

The Contagion of Covid-19 Pandemic on The Volatilities of International Crude Oil Prices, Gold, Exchange Rates and Bitcoin

Bounds testing approaches to the analysis of level relationships

Is gold a safe haven or a hedge for the US dollar? Implications for risk management

Hedging oil price risk with gold during COVID-19 pandemic

Is Bitcoin a hedge, a safe haven or a diversifier for oil price movements? A comparison with gold

Gold and crude oil prices after the great moderation

Analysis of the interactive linkages between gold prices, oil prices, and exchange rate in India. Glob

Spillovers from oil to precious metals: Quantile approaches

COVID-19 pandemic, oil prices, stock market, geopolitical risk and policy uncertainty nexus in the US economy: Fresh evidence from the wavelet-based approach

Modelling asymmetric cointegration and dynamic multipliers in a nonlinear ARDL framework

Feedback spillover dynamics of crude oil and global assets indicators: A system-wide network perspective

Return and volatility linkages among international crude oil price, gold price, exchange rate and stock markets: Evidence from Mexico

How Islamic are Islamic banks? A non-linear assessment of Islamic rate conventional rate relations

Shariah-compliant equities and Shariah screening: Need for convergence of ethical screening of stocks with Shariah screening

Have the Covid-19 Pandemic and Other Crises Similarly Determined the Nexus between Currency Exchange Rates and Gold Prices?

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

Do volatility indices diminish gold's appeal as a safe haven to investors before and during the COVID-19 pandemic?

Dynamic spillover between commodities and commodity currencies during United States QE

We sincerely thank editors R.S.J. Tol and R. Smyth and two anonymous reviewers for their insightful comments, which have helped us enhance the quality of this manuscript.