Has COVID-19 intensified the oil price-exchange rate nexus?

This paper extends the existing empirical literature by investigating whether the COVID-19 crisis has strengthened the dynamic relationships between oil price and exchange rate. We find significant breaks in the relationships wherein a common break is detected around the COVID-19 outbreak period. Of note, the interactions between the two markets intensified since the outbreak of the COVID-19 pandemic. Overall, our findings imply that the investors and policymakers are taking stock of the valuable information from the unanticipated occurrence of the COVID-19 pandemic. Thus, diversification in the form of portfolio switches towards foreign currency-denominated assets may be effective in the case of a depreciation of the domestic currency.

Introduction

The fluctuations in oil prices are a major driving force of variations in macroeconomic and financial variables such as current account, inflation, cross-border capital flows, business cycle movements, and investment. Besides, the exchange rates are a linking factor between oil prices and financial variables of an open economy (Amano and Van Norden, 1998a, Amano and Van Norden, 1998b, El Shazly, 1989, Chen and Rogoff, 2003, Coudert et al., 2015, Elder and Serletis, 2010, Golub, 1983, Hussain et al., 2017, Krugman, 1983, Narayan et al., 2008, Reboredo, 2012, Turhan et al., 2014, Yang et al., 2017). Theoretically, the relationship between oil prices and exchange rates can be analyzed through three channels – terms of trade channel1  (Amano and Van Norden, 1998a, Amano and Van Norden, 1998b), wealth channel2  (Krugman, 1983), and portfolio channel.3

Given the theoretical backdrop, a large number of empirical studies have been conducted till date. Majority of the empirical literature have focused on the returns of the two series (Amano and Van Norden, 1998b, Akram, 2009, Blomberg and Harris, 1995, Reboredo, 2012, Sadorsky, 2000, Turhan et al., 2014, Zhang et al., 2008), only a few have analyzed the ‘volatility spillover’ (Ding and Vo, 2012, Salisu and Mobolaji, 2013, Zhang et al., 2008) and ‘return-volatility’ interactions (Ahmad et al., 2020, Jawadi et al., 2016, Narayan et al., 2008, Ghosh, 2011) across these financial markets. Zhang et al. (2008) argued that since the US dollar is used as the major invoicing currency in international trade of crude oil hence there could be significant volatility spillovers within the two markets. Further, the rapid global financial integration necessitates the need to explore the linkages between returns and volatilities associated with these markets. The appropriate information regarding ‘return-volatility’ relationships not only improves the forecasts of these two variables but also benefit investors to form their optimal hedging strategies in presence of financial risk. For instance, if return and volatility spread from one market to another, portfolio investors and policymakers will need to adjust their investment and financial policy decisions in order prevent possible contagion risks during market crashes or financial crisis (Arouri et al., 2011) Therefore, investigating ‘volatility-spillover’ and ‘return-volatility’ interactions can offer new dimensions of both the oil and exchange rate markets.

Along these lines, we analyze the following relationships : (i) the bi-directional interactions between the volatilities of exchange rate and oil price, (ii) the effect of exchange rate volatility on exchange rate return, (iii) the effect of oil price volatility on exchange rate return, (iv) the impact of exchange rate return on oil price volatility, and (v) the impact of exchange rate return on exchange rate volatility.

Our study departs from previous works on two counts. The main contribution of our paper lies in testing the stability of the aforementioned relationships since the dynamic relationships, either in the form of return spillover or volatilty spillover are highly dependent on the state of the economy and the market situations. Sudden shocks in the economy, emanated from the factors like wars, terrorist attacks, political events and financial crisis seem to play a crucial role in affecting the predictability and thus changing the intensity of the spillover effect. Empirically, it has been obtained that the linkages between the return series are more profound and significant during the crisis periods than in normal situations (Brayek et al., 2015, Devpura et al., 2021, Reboredo, 2012, Turhan et al., 2014, Yin and Ma, 2018). It is worth mentioning that previous studies only deal with the stability of the ‘return-spillover’ between the oil and exchange rate markets. On the other hand, studies related to the stability of the ‘volatility spillover’ and the ‘return-volatility’ interactions are scarce. To the best of our knowledge only one study by Ding and Vo (2012) have examined the volatility interactions between oil and exchange rate markets in a structural break framework. However, it is unsettled how the impact of an unanticipated shock – as large as the current COVID-19 pandemic (Narayan et al., 2021, Phan and Narayan, 2020) – can affect the predictability of oil price and exchange rate. We try to fill this gap.

Second, we contribute to the vast literature on the impact of COVID-19 on oil and foreign exchange markets. However, we do so by putting to test the hypothesis: Has COVID-19 pandemic has affected the oil price—exchange rate relationships? Has the relationship strengthened or weakened due to the pandemic? To the best of our knowledge, we are the first to test the hypotheses within the COVID-19 context. We offer new insights as recent literature by Narayan et al. (2018) and Sharma et al. (2019) deliberates that sudden events contain crucial information which can improve exchange rate prediction. Since the sudden and dramatic outbreak of the COVID-19 pandemic is a typical case of an unanticipated event, we have an ideal context to test whether COVID-19 has influenced the strength of the relationships between oil price and exchange rate or not? Our findings also show that there are breaks detected during the initial phase of the outbreak and the COVID-19 occurrence has significantly affected the relationships.

To examine the relationships, our empirical strategies are as follows. (1) First, we gather daily dataset of Brent oil prices and nominal exchange rates of the four largest net oil-importing countries of Asia, viz. China, India, Japan, and Korea.4 (2) Next, we test for the unit root properties of oil price and exchange rates. (3) Then, we implement the ‘two-step’ estimation procedure of Grier and Perry (1998), which was further developed by Fountas et al. (2002).5 More specifically, we have taken a bivariate VAR- asymmetric GARCH (VAR-AGARCH) model at the first step to estimate conditional volatilities of exchange rate and oil price return. Then, in the second step, these estimates of volatilities and exchange rate return are included in a VAR set up to study different possible causal relationships. (4) Further, we have augmented the second step of the ‘two-step’ procedure by considering structural breaks in the relationships. To that end, we apply the Qu and Perron (2007) method to detect multiple structural breaks in the relationships. The Qu-Perron test is superior to other breakpoint tests that have been widely used in the literature, such as Bai and Perron, 1998, Bai and Perron, 2003, since these tests may not be appropriate in a simultaneous equation framework. (5) Finally, once the structural breaks are detected, we use the generalized impulse response functions (GIRFs) of Koop et al. (1996) and Pesaran and Shin (1998) to track the dynamic responses of exchange rate return, exchange rate volatility and oil price volatility to the shocks related to these variables. (6) As a robustness check, we tested for a structural break in the trend function of a time series before applying the formal analysis, especially when the break date is assumed to be unknown. It is worth mentioning that in the absence of a break in the trend function, the ADF test is superior to other alternative tests since it has the highest power, and thus making it the most appropriate test for testing the stationarity properties in a time series (Perron and Yabu, 2009). We check for robustness by implementing the Perron and Yabu (2009) test.

Our empirical findings indicate a high level of risk in both the exchange rate and oil markets during the sample period. We also find that the asymmetric GARCH (AGARCH) model is appropriate for all four countries to estimate the volatilities. Further, we find multiple structural breaks in the relationships in the case of all four countries. Specifically, we detect a break during the COVID-19 period for all countries, coinciding with the rising uncertainty about the pandemic as well as the plunge in oil prices. Overall, we find significant relationships between oil price volatility, exchange rate volatility and exchange rate returns. Specifically, the relationships strengthened during the COVID-19 outbreak period.

The rest of the paper proceeds with a critical overview of the empirical literature. In Section 3, we propose the empirical framework to test the hypotheses. The data properties and results are discussed in Section 4. Section 5 concludes with a summary.

Review of literature

This section reviews the main empirical literature on the oil price – exchange rate nexus. The review has been organized around the relationships explored, data utilized, econometric methodologies, and findings, including the latest literature on the COVID-19  pandemic (cf. Table 1).

Table 1

Review of empirical literature.

Study	Countries/Currencies	Data	Method	Main findings
Amano and Van Norden (1998a)	Germany, Japan and the US	Monthly data from January 1973 to June 1993	Johansen cointegration and causality tests	Negative relationship between oil prices and exchange rates. Causality runs from oil prices to exchange rates
Amano and Van Norden (1998b)	US real effective exchange rates in terms of 15 industrial currencies	Monthly data from February 1972 to January 1993	Johansen cointegration and causality tests	Positive relationship between oil prices and USD and causality runs from real oil prices to real effective exchange rates
Chen and Chen (2007)	G7 countries	Monthly data from January 1972 to October 2005	Panel cointegration	Negative relationship between real oil prices and real exchange rates
Bénassy-Quéré et al. (2007)	US REER and real oil prices	Monthly data from January 1974 to November 2004	Johansen cointegration and causality tests	Negative relationship between oil prices and USD. Causality runs from real oil prices to real effective exchange rates
Narayan et al. (2008)	FJD vis-à-vis USD	Daily data from November 29, 2000 to September 15, 2006	GARCH, EGARCH	Positive relation between oil prices and Fijian dollar.
Zhang et al. (2008)	EUR vis-à-vis USD	Daily data from January 4, 2000 to May 31, 2005	VAR, Granger causality and TGARCH	Long-run relationship between oil prices and exchange rates but no significant volatility spillover.
Ghosh (2011)	INR vis-à-vis USD	Daily data from July 2, 2007 to November 28, 2008	GARCH and EGARCH	Negative relationship between oil prices and exchange rate.
Ding and Vo (2012)	CAD, NOK, EUR, INR, JPY, SGC, BZR, MXP, vis-à-vis USDX	Daily data from July 28, 2004 to October 28, 2009	MSV, CCC-MGARCH, DCC-MGARCH	Before the crisis, oil prices and exchange rates respond simultaneously. Bidirectional volatility interaction during the crisis period.
Reboredo (2012)	8 major currencies EUR, AUD, CAD, GBP, JPY, MXP, NOK vis-à-vis USD	Daily data from 4th January 2000 to 15th June 2010	Correlation and Copula models	Co-movement weak during the pre-crisis period but strong in the post-crisis period.
Wu et al. (2012)	USDX	Weekly data from January 2, 1990 to December 28, 2009	Copula-based GARCH models	Negative relationship between oil prices and USDX. Dependence structure is negative and decreasing after 2003. Short-run volatility is smaller than the long-run volatility
Aloui et al. (2013)	5 major currencies: EUR, CAD, GBP, AUD, CHF, JPY vis-à-vis USD	Daily data from January 4, 2000 to February 17, 2011	Correlation and Copula-GARCH models	Positive comovement except for Yen. Dependence increased during the crisis period.
Salisu and Mobolaji (2013)	NGN vis-à-vis USD	Daily data from January 02, 2002 to March 20, 2012	VAR-GARCH model	Significant bidirectional return and volatility spillovers between oil prices and exchange rate
Turhan et al. (2014)	G20 countries	Daily data from January 2, 2000 to April 17, 2013	DCC-GARCH and GJR-GARCH	Negative comovement and the relationship strengthened during and post-crisis period.
Brayek et al. (2015)	Seven major currencies: EUR, CAD, AUD, JPY, MXP, NOK, GBP vis-à-vis USD	Daily data from January 3, 2000 to April 17, 2014	Breakpoint test, Granger-causality, Copula model, DCC-GARCH model	Causality running from oil prices to exchange rates during and post-crisis period. The dependency increased during the crisis period with oil prices as the lead variable.
Aloui and Aïssa (2016)	USDX	Daily data from January 4, 2000 to May 31, 2013	Copula-based GARCH model	Negative relationship between oil prices and USDX
Chen et al. (2016)	OECD countries	Monthly data from January 1990 to December 2014	Cointegration and Structural VAR	Oil demand shock have significant impact on exchange rates
Jawadi et al. (2016)	USD vis-à-vis EUR	Intraday data from August 2104 to January 2016.	GARCH model	Negative relationship between oil prices and USD/Euro exchange rate. Significant volatility spillovers from exchange rate to oil prices
Hussain et al. (2017)	12 Asian economies: RMB, HKD, INR, IDR, JPY, KRW, MYR, PKR, PHP, SGD, LKR, and TWD vis-à-vis USD	Daily data from 21st May 2006 to 18th May 2016	Granger causality	Unidirectional causality runs from oil prices to exchange rates
Li et al. (2017)	RUB, CAD, JPY, and EUR vis-à-vis USD	Weekly data from January 1, 1999 to December 4, 2015	Stochastic volatility with correlated jumps	Jump spillover from oil market to foreign exchange market
Seyyedi (2017)	India	Daily data from 12th January 2004 to 30th April, 2015	Cointegration and causality	No evidence of cointegration but bidirectional causality between oil prices and exchange rate
Yan et al. (2017)	Oil-exporting countries (BZR, CAD, MXP, RUB) and oil-importing countries (EUR, INR, JPY, KRW)	Daily data from January 1, 1999 to December 31, 2014	Wavelet coherence	Negative relationship between oil prices and exchange rates. Interdependence is higher for oil-exporting countries as compared to oil-importing countries
Yang et al. (2018)	Four major currencies: EUR, CAD, JPY, GBP vis-à-vis USD	Daily and monthly data from January 3, 1994 to December 31, 2015	DCC-MIDAS model	Negative long-run correlations between oil prices and exchange rates except for Japan.
Mollick and Sakaki (2018)	8 industrial countries (CHF, JPY, GBP, AUD, NZD, EUR, CAD, NOK) and 6 emerging countries (KRW, INR, ZAR, BZR, RUB, MXP) vis-à-vis USD	Weekly data from January 1, 1999 to July 1, 2017.	VAR, GARCH	Positive oil price shocks lead to an appreciation of the domestic currency. Significant negative relationship between oil prices and exchange rates.
Wen et al. (2018)	USD	Weekly data from January 7, 2000 to July 25, 2014	Causality, SVAR, DCC-GARCH	Linear causality runs from exchange rate to oil prices but nonlinear causality runs from oil prices to exchange rate. High dependence between 2002 and 2012.
Yin and Ma (2018)	9 major currencies: AUD, EUR, NZD, GBP, CAD, RMB, JPY, SEK, CHF.	Monthly data from January 1994 to July 2017	Bayesian graphical VAR	Time-varying dependence insignificant before 2008 crisis but significant after the crisis.
Kisswani et al. (2019)	ASEAN-5 countries: IDR, PHP, MYR, SGD, THB	Quarterly data from 1970 to 2016	NARDL	Evidence of both symmetric and asymmetric effect of oil prices on exchange rates.
Xu et al. (2019)	14 countries: AUD, CAD, CHF, EUR, GBP, INR, JPY, KRW, NOK, NZD, BZR, SEK, DKK, ZAR vis-à-vis USD	Daily and monthly data from January 1995 to December 2016	Normal Mixture Model	Weak negative correlation but dependency enhanced during the crisis. Oil demand shock affects the dependency more than oil supply shocks.

This table presents the vital studies conducted on exploring the relationship between oil prices and exchange rates. For countries’ currencies, standard abbreviations have been used for brevity. These are: Australian Dollar (AUD), Brazilian real (BZR), British Pound Sterling (GBP), Canadian Dollar (CAD), Chinese Renminbi (RMB), Danish Krone (DKK), Euro (EUR), Fijian Dollar (FJD), Hong Kong Dollar (HKD), Indian Rupee (INR), Indonesian Rupiah (IDR), Japanese Yen (JPY), Malaysian Ringgit (MYR), Mexican peso (MXP), New Zealand Dollar (NZD), Nigerian Naira (NGN), Norwegian Krone (NOK),  Pakistan Rupee (PKR), Philippines Peso (PHP), Russian Ruble (RUB), Singapore Dollar (SGD), South African Ran (ZAR), South Korean Won (KRW), Sri Lanka Rupee (LKR), Swedish Krona (SEK), Swiss franc (CHF), Taiwanese new dollar (TWD), Thailand Baht (THB), and U.S. dollar index (USDX).

The first set of studies rely on cointegration and causality techniques between oil prices and exchange rates, and utilizes monthly and quarterly data (Amano and Van Norden, 1998b, Ahmad and Hernandez, 2013, Chen and Chen, 2007, Bénassy-Quéré et al., 2007, Chen et al., 2016, Yin and Ma, 2018, Kisswani et al., 2019) with a few exceptions of daily data (Seyyedi, 2017, Hussain et al., 2017) that utilizes nominal oil prices and nominal exchange rates.

Another set of studies have focused on the time-varying relationship and comovement between daily oil price returns and exchange rate returns by employing a variety of methodologies, such as DCC-GARCH, Copula, Wavelets, etc. (Reboredo, 2012, Aloui et al., 2013, Turhan et al., 2014, Brayek et al., 2015, Yang et al., 2017, Yang et al., 2018, Xu et al., 2019), along with Wen et al. (2018) using weekly data. While the literature has mostly focused on the returns relationship, only very recently volatility relationships have received some attention. Of these, most studies use daily data (Bénassy-Quéré et al., 2007, Narayan et al., 2008, Zhang et al., 2008, Ghosh, 2011, Ding and Vo, 2012, Salisu and Mobolaji, 2013, Jawadi et al., 2016, Tian and Hamori, 2016, Aloui and Aïssa, 2016, Le, 2017), with a few exceptions on weekly data (Wu et al., 2012, Li et al., 2017, Mollick and Sakaki, 2019).

In terms of empirical findings, a number of studies found positive relationship between oil prices and USD (Amano and Van Norden, 1998a, Amano and Van Norden, 1998b, Chen and Chen, 2007, Ghosh, 2011, Yang et al., 2017, Mollick and Sakaki, 2019). On the other hand, few studies found a negative relationship, implying a rise in oil prices lead to a depreciation of the USD (Bénassy-Quéré et al., 2007, Narayan et al., 2008, Wu et al., 2012, Aloui and Aïssa, 2016, Jawadi et al., 2016, Le, 2017). It is interesting to note that negative relationship is found in studies mostly using daily and weekly data. With reference to the time-varying dependence between oil and exchange rate markets, notably most of the studies found weak dependence before the 2008 GFC however the dependence strengthened during the crisis period (Ding and Vo, 2012, Reboredo, 2012, Aloui et al., 2013, Turhan et al., 2014, Brayek et al., 2015, Yin and Ma, 2018, Wen et al., 2018, Xu et al., 2019).

In summary, very few studies have explored the bidirectional volatility spillover between oil prices and exchange rates. Among them, Zhang et al. (2008) and Ding and Vo (2012) found no evidence of significant volatility spillovers in the pre-crisis period. On the other hand, Salisu and Mobolaji (2013) and Jawadi et al. (2016) found evidence of bidirectional volatility spillover and unidirectional spillover from exchange rate to oil market, respectively, in the case of U.S. Ahmad et al. (2020) find evidence of negative impact of oil jumps on exchange rate volatility but an asymmetric impact of exchange rate volatility on oil jumps. Interestingly, Narayan et al. (2008) and Ghosh (2011) examined the impact of exchange rate volatility on exchange rate return but found no significant impact. We summarize, in the following Table 1, the main relevant literature on the topic, as a matter of providing a comparative perspective for readers.

COVID-19 impact on oil and foreign exchange markets

With regards to the literature on energy markets in time of COVID-19 pandemic, we classify the existing studies into multiple strands.6 First strand examines the impact of the pandemic on the oil prices (Devpura and Narayan, 2020, Gil-Alana and Monge, 2020, Liu et al., 2020, Narayan, 2020c, Salisu and Adediran, 2020). Devpura and Narayan (2020) explores the evolution of hourly oil price volatility. They use multiple measures of volatility and finds that volatility of oil prices increased following the COVID-19 outbreak. Similarly, Liu et al. (2020) finds a negative connection between crude oil returns and stock returns in the US. Narayan (2020c) finds that not only oil related news but also the COVID-19 cases influences oil prices. Gil-Alana and Monge (2020) examined the persistence of the COVID-19 shock on energy markets. They find that before the public health crisis there was market efficiency, however the oil market became inefficient after the onset of the crisis. They further argue that even though the shock is temporary but it could have long-lasting effects on the oil market. In another study, Salisu and Adediran (2020) highlighted the importance of volatility due to infectious diseases in prediction of energy market volatility.

The second strand focuses on the impact of the pandemic on oil related investment (Fu and Shen, 2020, Huang and Zheng, 2020, Prabheesh et al., 2020a, Prabheesh et al., 2020b). Prabheesh et al., 2020a, Prabheesh et al., 2020b examines the dynamic dependence between oil prices and stock indices of oil-importing and oil-exporting countries, respectively. They found that the initial phase of the outbreak intensified the dynamic dependence between the two markets. Fu and Shen (2020) finds that COVID-19 negatively affects performance of energy companies. Huang and Zheng (2020) examine the relationship between investor sentiment and crude oil futures price and find evidence of a cointegrating relationship between them. Another strand of literature focuses on the impact of the pandemic on financial and foreign exchange markets (Bouri et al., 2021, Corbet et al., 2020, Garg and Prabheesh, 2021, Lyócsa et al., 2020, Mishra et al., 2020, Narayan, 2020b, Narayan et al., 2020, Prabheesh, 2020, Rai and Garg, 2022, Wei et al., 2020, Zaremba et al., 2020)

With reference to the financial literature, few studies have been conducted on the impact of the pandemic on foreign exchange markets (Devpura, 2020, Mishra et al., 2020, Narayan, 2020a, Narayan et al., 2021, Rai and Garg, 2021, Salisu and Sikiru, 2003) and stock markets (Ali et al., 2020, Baig et al., 2021, Haroon and Rizvi, 2020a, Haroon and Rizvi, 2020b, He et al., 2020, Gil-Alana and Claudio-Quiroga, 2020, Prabheesh, 2020, Sharma, 2020). The overall conclusion from the findings of the above studies indicate that the pandemic has adversely affected the energy and foreign exchange markets. The oil prices, both WTI and BRENT, have witnessed record lows and the volatility increased during the lockdown periods. Of note, none of these studies explore different linkages between oil and exchange rate markets as mentioned in the Introduction section in the light of multiple structural breaks. Further, they do not consider the impact of COVID-19 on these relationships. Our study fills these gaps.

Methodology

With reference to the volatility modelling of oil price returns and exchange rate returns, previous research has utilized multivariate GARCH-type models such as BEKK, DCC, etc to capture volatility effects (Ding and Vo, 2012, Salisu and Mobolaji, 2013, Algieri, 2014, Turhan et al., 2014, Mollick and Sakaki, 2019). However, we go a step ahead and test the relationships between exchange rate returns and volatilities in exchange rates and oil prices. Thus, we apply the ‘two-step’ procedure as discussed in Fountas et al. (2002). In the first step, a bivariate GARCH model is used to generate time-varying conditional volatilities of exchange rate and oil price and then in the second step, a VAR model consisting of exchange rate return, exchange rate volatility and oil price volatility has been employed to detect the nature of the relationships. It is important to note that we have augmented the second step of the ‘two-step’ procedure by giving due consideration to multiple structural breaks. In the following sub-sections, we first state the model to estimate exchange rate and oil price volatilities, and then briefly describe the Qu and Perron (2007) test for multiple structural breaks in the system of equations.

Model

In line with the previous literature we have employed bivariate BEKK-GARCH specification to estimate the conditional variances of exchange rate and oil price returns. However, it is worth mentioning that the bivariate GARCH model suffers from the limitation that it provides symmetric responses against positive and negative shocks of equal magnitudes. Further, it is evident that the economic effect of ‘negative shocks’ on exchange rate volatility is more than that of ‘positive shocks’ (Narayan et al., 2008, Ghosh, 2011). Therefore, while modelling, ignoring the sign of the past shocks may inadequately estimate the conditional volatility of the exchange rate. Glosten et al. (1993) have introduced a threshold GARCH model in a univariate framework to capture the asymmetric effects of positive and negative shocks on the volatility. Grier et al. (2004) have generalized the univariate threshold GARCH model in a bivariate framework. They have introduced an asymmetric version of the bivariate BEKK-GARCH (AGARCH) model to capture the effects of ‘good’ and ‘bad’ news on the variance-covarince processes.7 Following Grier et al. (2004), we have taken bivariate AGARCH model to compute the conditional volatilities of exchange rate and oil price returns. Eqs. (1), (2) constitute the VAR(p)-AGARCH(1,1) model involving exchange rate return  and oil price return .  where , , , , , , , , and  represents the information set at time point t.

In the matrix, , the diagonal elements,  and , are the conditional exchange rate volatility and oil price volatility, repsectively, and the off-diagonal elements  (or ) are the conditional covariance. In Eq. (2), , is a 2 × 2 matrix of coefficients associated with  where , and , , where  is an indicator function that takes the value 1 if  and 0 otherwise for exchange rate. Likewise, for oil price i.e.,  takes the value 1 if  and 0 otherwise.

The BEKK model here is extended by relaxing the symmetry assumption and thus allowing for positive and negative shocks in the conditional variance–covariance matrix. Therefore, the Eq. (2) is the AGARCH model wherein we put to test the following condition:  for  Then, we ontain the maximum likelihood estimates by employing the Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm.

Qu-Perron methodology

Econometric literature on structural breaks is voluminous (Perron et al., 2006). Following the seminal work of Perron (1989), one strand of literature focuses on the testing of stationarity/non-stationarity of a time series with due consideration to structural break(s) in the trend function. Most often, the well-known ADF test is used to identify the breaks. However, one major limitation of the ADF test is that the deterministic trend function in the estimating equation is assumed to be constant all throughout the sample period. However, Perron (1989) showed that the autocorrelation process of a random walk model is almost the same as that of a deterministic trend model with a break in the trend function. Thus, the ADF test may produce unreliable results and suggest a unit root when, in fact, there is no unit root, and the true data generating process is stationary with a structural change in the deterministic trend function. Perron (1989) proposed a unit root test under three different types of deterministic trend function (viz., (i) changes in the intercept, (ii) changes in the slope, (iii) changes in both intercept and slope coefficients) wherein the (single) change point was assumed to be known a priori. This latter assumption being restrictive, Zivot and Andrews (1992) and Vogelsang and Perron (1998) relaxed this, and the change/break point was assumed to be unknown and determined endogenously. Later, the unit root test have been generalized by including two (Lumsdaine and Papell, 1997, Lee and Strazicich, 2003, Narayan and Popp, 2010)8 or more break dates (Carrion-i Silvestre et al., 2009) in the trend function.

The other branch of structural break literature, following the seminal work of Bai and Perron, 1998, Bai and Perron, 2003 deals with the multiple structural breaks in the relationships between dependent and independent variables. However, the Bai-Perron break test is based on a single equation framework. Research work on structural changes in the context of a system of equations is indeed very recent. There are only very few studies on this, and the important ones are Hansen (2003) and Qu and Perron (2007). The distinct advantage of the test procedure proposed by Qu and Perron (2007) is that it gives a detailed treatment with regard to the issues of estimation, inference and computation in presence of multiple structural changes in vector autoregressive (VAR) models. The Qu-Perron test generalizes the Bai-Perron test in case of system of equations. Since, our study focuses on the ‘volatility-spillover’ and ‘return-volatility’ interactions between the oil price and exchange rate markets, the Qu and Perron (2007) test is more appropriate to study the stability of the relationships.

The general model considered by Qu and Perron (2007) is  where the vector  refers to observations for  dependent variables at th time point. Here,  is the identity matrix of order . It is assumed that there are  structural changes in the system and the break dates are denoted by an  vector , taking into account that  and . The subscript  indexes a sub-period , the subscript  indicates a temporal observation , and the subscript  indexes the equation  to which the scalar dependent variable  is associated. Further,  is the number of regressors and  is the set which includes the regressors from all the equations i.e.,  and  is the set of parameters in the th sub-period. Furthermore, the error  is assumed to have mean 0 and covariance matrix  for . The selection matrix is denoted by  in the above equation and it is of dimension , which involves elements which take the values 0 and 1. When we estimate a VAR model,  is then , which contains the lagged dependent variables, and hence  will be an identity matrix. The quasi-maximum likelihood (QML) method has been used to obtain the estimates of the coefficients along with the break dates. Qu and Perron (2007) have also derived the limiting distribution of the coefficients.

For our study,  in Eq. (3) now consists of three variables, viz., exchange rate return, exchange rate volatility and oil price volatility, i.e., . In case of these three variables, the general model represented in Eq. (3) is nothing but a VAR model, which is being specified for lag 1 in Eq. (4).9 The coefficients in these three equations are being given different notations for the sake of meaningful easy understanding.  where  and  is an identity matrix of order 12. In the above equation, the main parameters of interest are ,  and . The parameters  indicate the effects of exchange rate volatility and oil price volatility on exchange rate return, respectively at the th sub-period. The coefficients  and  imply the effects of exchange rate return on exchange rate volatility and oil price volatility, respectively at the th regime. The coefficient  measures the impact of oil price volatility on exchange rate volatility at the th regime, whereas the opposite impact is indicated by . All the other coefficients in  have their usual meaning.

With reference to the testing, Qu and Perron (2007) proposed a number of test to identify multiple breakpoints in a system of  equations. Except some data points in the beginning and end of the sample that are lost due to the trimming factor, all data points are considered to be probable break points. The trimming factor is used to distinguish the first and last sub-periods. Qu and Perron (2007) proposed the following tests: (i) The  test, (ii) The  test: It is a sequential test wherein we test the null hypothesis of  breaks versus the alternative of  breaks. In line with Bai and Perron, 1998, Bai and Perron, 2003 and Qu and Perron (2007) suggest that first the  tests are used to check if one break is present. If the test statistic rejects the null of at least one break, then the sequential test of the  test is utilized for detecting the exact number of break points.

Once the structural breaks are detected, we use the estimation results for the last sub-periods to track the dynamic responses of exchange rate return, exchange rate volatility and oil price volatility with respect to the shocks related to these three variables. Thus, we apply the generalized impulse response functions (GIRF) that was developed by Koop et al. (1996) and Pesaran and Shin (1998). The GIRF is superior to the conventional impulse response functions since the former is not sensitive to the ordering of the VAR model and is not subject to the orthogonality issues.

Empirical results

Data and preliminary results

Our sample consists of daily oil prices and nominal exchange rates of four largest net oil-importing countries of Asia. We selected the major net oil-importing countries based on the list published in The World Factbook by the Central Intelligence Agency (CIA). Thus, our sample includes China, India, Japan, and Korea. The sample data are for the period from January 2, 2017 to August 10, 2020. We choose 2017 as the starting point due to two reasons. First, there are many studies already conducted that have analyzed the oil price spike during the GFC of 2008 and oil price shock of 2014. However, there is scant literature on analyzing the period since 2017 and more so, no studies have included the current pandemic outbreak period. Second, the oil prices in 2017 were in a stable range after the OPEC and non-OPEC allies agreed to cut production.

As a proxy for world oil prices, we consider Brent crude oil spot prices as it is widely used as a benchmark for the oil price in the literature, and highly co-move with other major crude oil prices such as WTI.10 For exchange rates, we use nominal exchange rates of domestic currency against the US dollar. Following Narayan (2008), we modelled the empirical framework with nominal variables since daily prices are unavailable to deflate the nominal variables into real variables. Daily returns of oil prices and exchange rates are calculated as:  where  is the returns on Brent oil prices and  is the returns of the respective country’s exchange rate, respectively. Table 2 presents the results of descriptive statistics.11 We find that in case of two developed markets, Japan and Korea, the average returns are negative, whereas for the two emerging markets, India and China, average returns are positive. The results suggest a higher level of risk in both the markets wherein oil prices are more volatile than exchange rates. The skewness and kurtosis statistics indicate a leptokurtic distirbution due to the rejection of the normality.

Table 2

Summary of descriptive statistics.

	Mean	Max.	Min.	SD	Skewness	Kurtosis	JB
EXR_India	0.011	1.374	−1.535	0.334	0.051	4.781	119.6319⁎
EXR_China	0.000	1.444	−0.913	0.261	0.304	6.433	447.244⁎
EXR_Japan	−0.012	2.072	−3.433	0.471	−0.573	7.902	950.4712⁎
EXR_Korea	−0.001	3.161	−3.096	0.489	−0.085	6.783	542.4550⁎
OILR	−0.025	41.202	−64.370	4.135	−3.166	88.065	272 855.6⁎

This table presents a summary of descriptive statistics of exchange rate returns (EXR) and oil returns (OILR), respectively. The exchange rate is defined as domestic currency per unit of US dollar while oil prices are Europe Brent Spot Prices measured as US dollars per Barrel, respectively. The sample consists of daily observations ranging from January 2, 2017 till August 10, 2020 and adjusted for weekend and public holidays.

Indicates significant at 1% level of significance.

Findings from the VAR(p)-AGARCH(1,1) model

We estimate the VAR()-AGARCH(1,1) model for exchange rate return (hereafter, ) and oil price return (hereafter, ), as specified in Eqs. (1), (2) to obtain the estimates of exchange rate volatility (hereafter, ) and oil price volatility (hereafter, ) for each of the four countries. The Schwarz information criterion (SIC) has been used to determine the value of p for the VAR(p)-AGARCH(1,1) model. To reduce the computational difficulties, we have restricted our search of optimal p between VAR(1)-AGARCH(1,1) to VAR(8)-AGARCH(1,1) models. The SIC values have been reported in Panel A of Table 3. The results indicate that the optimum value of  is 1 for all the countries.

Table 3

Results from VAR-AGARCH model.

	India	China	Japan	Korea
Panel A
Lag 1	5.073	4.629	5.759	5.792
Lag 2	5.098	4.657	5.782	5.823
Lag 3	5121	4.68	5.805	5.841
Lag 4	5.14	4.702	5.839	5.873
Lag 5	5.167	4.732	5.846	5.883
Lag 6	5.193	4.747	5.881	5.902
Lag 7	5.22	4.773	5.902	5.921
Lag 8	5.251	4.811	5.936	5.95
Panel B
Test for bivariate	14 665.72⁎	11 070.03⁎	11 017.56⁎	9566.21⁎
GARCH	[0.00]	[0.00]	[0.00]	[0.00]
Test for asymmetric	42.02⁎	43.69⁎	66.28⁎	21.89⁎
bivariate GARCH	[0.00]	[0.00]	[0.00]	[0.00]
Panel C
Squared residuals and squared standardized residuals for exchange rate return
Q(5)	3.57	8.32	5.46	3.64
Q(10)	4.15	10.76	13.02	12.45
Q2(5)	10.01	3.39	3.99	9.91
Q2(10)	11.38	6.2	13.56	11.75
Squared residuals and squared standardized residuals for oil price return
Q(5)	2.62	4.54	3.48	2.52
Q(10)	4.59	7.9	4.66	5.69
Q2(5)	1.92	5.53	8.06	6.33
Q2(10)	5.36	8.04	10.52	9.21

This table consist of three panels: A, B and C. Panel A represents the optimal lag length (in bold) based on the SIC. Panel B reports the test results from GARCH and AGARCH model. Panel C reports the results of the residual diagnostics tests. Value in the parenthesis [ ] are p-values.

Indicates significant at 1% level of significance.

Next, we test if the null hypothesis of ‘conditional homoskedasticity’ is rejected against the alternative of ‘conditional heteroskedasticity’. The empirical results suggest that the null hypothesis is rejected for all the four countries thus favouring heteroskedastic conditional variances (see Panel B, Table 2). Furthermore, the finding of joint significance of the elements of the D matrix i.e., the test statistic value for testing the null hypothesis  exceeds the critical value at 1% level of significance. Thus, we conclude that the covariance process is asymmetric for all the countries. Lastly, the Ljung–Box  statistic values based on the standardized residuals and squared standardized residuals of the VAR(1)-AGARCH(1,1) model suggest that the choices of the lag orders for the VAR process and for the asymmetric GARCH (AGARCH(1,1)) model are adequate for all the countries, and hence the model is well specified from this consideration.

We have estimated the VAR(1)-AGARCH(1,1) model for all the four countries and obtain the estimates of  and  which can be taken as the estimates of exchange rate volatility (EXV) and oil price volatility (OILV). Fig. 1(a)–(f) show the exchange rate return (EXR), EXV and OILV for the four countries.

Fig. 1

Plots of EXR, EXV and OILV. The figure shows the plot of exchange rate returns, exchange rate volatility, and oil price volatility. Figures (a)–(c) are for India; (d)–(f) for China; (g)–(i) for Japan; and (j)–(l) for Korea.

Results from multiple structural breaks

Once the exchange rate volatility and oil price volatility have been measured in the first step, in the second step, the Qu and Perron (2007) procedure is applied for finding the presence of multiple structural breaks in the model specified in Eqs. (4) involving EXR, EXV and OILV. The results on structural breaks along with the estimated break dates are reported in Table 4.

Table 4

Results from Qu and Perron (2007) multiple structural break tests.

Country	India	China	Japan	Korea
WDmax test
1 break	44.79⁎⁎	64.45⁎⁎	308.98⁎⁎	70.72⁎⁎
Sequential test
SupF_T(2|1)	22.48	26.38	92.26⁎⁎	40.62⁎⁎
	(38.64)	(38.64)	(38.64)	(38.64)
SupF_T(3|2)			65.30⁎⁎	69.43⁎⁎
			(39.99)	(39.99)
SupF_T(4|3)			44.20⁎⁎	16.45
			(40.88)	(40.88)
SupF_T(5|4)			32.97
			(41.60)
Estimated break dates
Tˆ_1	March 6, 2020	February 3, 2020	10th July, 2017	10th July, 2017
	[March 3, 2020 : March 9, 2020]	[January 14, 2020 : February 13, 2020]	[July 7, 2017 : July 13, 2017]	[July 7, 2017 : July 11, 2017]
Tˆ_2			November 1, 2018	April 4, 2018
			[October 31, 2018 : November 2, 2018]	[April 3, 2018 : April 5, 2018]
Tˆ_3			August 27, 2019	February 27, 2020
			[August 22, 2019 : August 30, 2019]	[February 12, 2020 : March 13, 2020]
Tˆ_4			February 27, 2020
			[February 21, 2020 : March 4, 2020]

, ,  and  give the estimates of the first, second, third and fourth break dates, respectively. Figures in ( ) are the critical values at the 5% significance levels while the values in [ ] are the 95% confidence intervals of the estimated break dates.

Indicates significance at 5% level of significance.

It is evident from Table 4 that all the tests results indicate the presence of at least one structural break in the relationship involving the three variables for each of the four countries. For instance, in case of Korea, the test statistic value for the  test is found to be 70.72 and significant at 1% level. Hence, the null hypothesis of ‘no break’ is strongly rejected in favour of the alternative of ‘up to one break’ in the system equations. To detect further if there is more than one structural break, the sequential break test is now carried out. By looking at the relevant entries of Table 4, we note that the test statistic  has the value 40.62 for Korea, which is highly significant. So, the test rejects the null hypothesis of ‘one break’ in favour of ‘two breaks’. Similarly, the test statistic value of  indicates there are three breaks in the relationships. However, the sequential test for detecting more than three breaks, i.e.,  cannot be rejected for Korea. The value of test statistic is found to be 16.45, which is insignificant at 1% level of significance and hence no further test is required for Korea. Finally, the three breakpoints for Korea have been estimated following the procedure of Qu-Perron, and these are found to be July 10, 2017, April 4, 2018, February 27, 2020. Applying the same procedure, we obtained one break each for India and China, four breaks in case of Japan. The estimated break dates are March 6, 2020 and February 3, 2020 for India and China, respectively. In case of Japan, break dates are July 10, 2017, November 1, 2018, August 27, 2019 and February 27, 2020.

In case of India and China, the breakpoints are detected in 2020, owing to the COVID-19 pandemic. For Japan and Korea, we find multiple breakpoints in spread over 2017–2019 and then in the first quarter of 2020. We find July 10, 2017 as the first break in case of both Japan and Korea. The break can be ascribed to the period when oil prices were underway on a rising trend after a couple of years of stable oil prices. The two break dates in 2018 is due to another turning point from an oil price decline to an upward trend. However, this upward trend was short-lived as the oil prices declined sharply in early November 2018 after the U.S. EIA announced that the U.S. is the largest producer of crude oil in the world.

Interestingly, all four countries witnessed a break in 2020 suggesting that the pandemic affected the developed and emerging markets alike. The first breakpoint of the year, February 3, 2020, can be attributed to the oil and foreign exchange markets reactions to the World Health Organization declaring the COVID-19 outbreak a “Global Public Health Emergency”. The second and third breakpoint during COVID-19 period, February 27 and March 6, coincides with the period of financial markets turmoil and the Russia-Saudi Arabia oil price war. This period saw sharp declines in stock markets in Asia-Pacific, Europe and the U.S. Specifically, oil price fell to their lowest level in over a year on February 27 while stock markets worldwide closed on March 6 while the oil prices saw the largest one-day decline, 9%, in about eleven years.

Sub-period wise relationships

With regards to the empirical evidence on the relationships between oil price volatility, exchange rate volatility and exchange rate returns, we find several interesting results.

In case of India, we find a significant positive impact of oil price volatility on exchange rate volatility in the period before COVID-19. A 10% increase in the oil price volatility leads to 0.02% increase in the exchange rate volatility. For other relationships, we did not find any significant results. Interestingly, in the sub-period II which coincides with the COVID-19 period, we find multiple significant relationships between the variables. Specifically, we find a significant positive bidirectional relationship between exchange rate volatility and exchange rate return. A 1% increase in EXV leads to an increase of 0.56% increase in EXR while a 10% increase in EXR leads to a 0.5% increase in EXV. We also find that there is asymmetric bidirectional relationship between OILV and EXR. While a 10% increase in OILV leads to a 0.01% decrease in EXR, a 1% increase in EXR leads to a 14.5% increase in OILV suggesting exchange rates as potentially strong drivers of oil prices in the short-run. These results are in line with the discussion in Fratzscher et al. (2014) and Beckmann et al. (2020) that a depreciation of the US dollar can lead to changes in oil prices as the latters is invoiced in US dollars. Finally, we find that a 10% increase in OILV leads to a 0.01% increase in EXV. Thus, it can be concluded that OILV was a determining factor in explaining EXV in the pre-pandemic period however the returns-volatility relationship strengthened after the outbreak. With regards to China, we find similar results. One significant structural break is detected during the COVID-19 period. In the sub-period before the pandemic, we find that there exists a significant positive impact of EXV on EXR whereas we find positive impact of EXR on EXV and OILV in the last sub-period. Specifically, a 1% increase in EXR leads to 0.05% increase in EXV and 31% increase in OILV.

With reference to Japan and Korea, we find multiple structural breaks. Specifically, in case of Japan, we find four structural breaks occurring in 2017, 2018, 2019 and 2020. In all the sub-periods except sub-period IV, we find a significant negative relationship between EXR and EXV. A 10% increase in EXR results in about 0.9%, 0.6%, and 1.8% decrease in EXV in the sub-period I, II, and III. However, the relationship strengthened during the pandemic period wherein a 10% increase in EXR leads to a 5% decrease in EXV. We also find evidence of positive impact of EXV on EXR in the sub-period II and III. But, we do not find any significant relationships in the sub-period IV. Similar to India and China, we find multiple significant relationships that also strengthened during the crisis period. We find that EXV positively affects EXR and OILV while EXR negatively affects EXV and OILV. Notably, a 1% increase in EXR leads to about 16% decrease in OILV whereas 1% increase in EXV leads to 8% increase in OILV. Finally, in case of Korea, we detect three structural break in 2017, 2018 and 2020. For the period before the pandemic, we do not find any significant relationships however we find evidence in favour of significant positive impact of EXR on EXV and OILV. Specifically, a 1% increase in EXR results in 0.15% and 14.5% increase in EXV and OILV, respectively.

It must be noted that we find that both EXV and OILV affects each other. In contrast to the existing literature that focuses on EXR as the dependent variable and OILV as the explanatory variable, we find that not only EXR and OILV affects each other, but also there are significant linkages between the two volatilities, EXV and OILV. It can also be seen that the relationships take different directions in different countries. For instance, we find positive effect of EXR on OILV in case of India, China, and Korea but EXR negatively impacts OILV in case of Japan. Likewise, we find that OILV negatively impacts EXR in case of India. The relationship between the two volatilities is symmetric wherein we find positive impact of OILV on EXV in case of India and EXV positively affects OILV in case of Japan. Interestingly, the results indicate that the two markets take account of the information about the pandemic as the two volatilities exhibit significant linkages only in the last sub-period.

In summary, one thing we find that there is a break detected during the COVID-period in case of all four countries, indicating the major impact of the pandemic on the two markets. Second, we find multiple significant relationships in the last-sub period as compared to the pre-pandemic sub-periods. This further justifies the importance of an unanticipated event in improving the predictive power. Third, the empirical evidence show that EXR significantly impacts OILV only in the pandemic sub-period. Even then, the relationship intensified during this time where a unit change in EXR leads to manifold change in the OILV (see Table 5).

Table 5

Sub-period wise results.

	India	China	Japan	Korea
Sub-period I
λ_2,1	0.096	0.593⁎	−0.416	0.061
λ_3,1	−0.003	0.002	−0.019	−0.031
λ_4,1	0.004	−0.002	−0.094⁎	−0.018
λ_6,1	0.002⁎⁎	0.000	0.002	0.009
λ_7,1	0.158	0.234	0.325	0.202
λ_8,1	1.006	0.574	−1.971	0.248
Sub-period II
λ_2,2	0.561⁎⁎⁎	0.011	1.078⁎⁎⁎	0.925
λ_3,2	−0.001⁎⁎	0.000	−0.037	−0.061
λ_4,2	0.053⁎⁎⁎	0.053⁎	−0.064⁎⁎	−0.011
λ_6,2	0.001⁎	−0.000	0.001	0.001
λ_7,2	14.534⁎⁎⁎	30.927⁎	0.12	0.069
λ_8,2	−34.931	−75.393	−0.942	3.609
Sub-period III
λ_2,3			0.522⁎⁎⁎	0.223
λ_3,3			−0.007	−0.006
λ_4,3			−0.182⁎	0.006
λ_6,3			−0.003	0.001
λ_7,3			−0.615	0.219
λ_8,3			0.850	3.540
Sub-period IV
λ_2,4			−0.039	−0.043
λ_3,4			0.005	0.000
λ_4,4			−0.041	0.148⁎⁎
λ_6,4			0.001	0.000
λ_7,4			−0.651	14.488⁎
λ_8,4			−5.735	−0.38
Sub-period V
λ_2,5			0.142⁎⁎⁎
λ_3,5			−0.000
λ_4,5			−0.504⁎
λ_6,5			−0.000
λ_7,5			−15.718⁎
λ_8,5			8.257⁎

The parameter  indicates the effect of EXV on EXR at  sub-period of the respective country. The parameter  indicates the effect of OILV on EXR at the  sub-period of the respective country. The coefficients  and  imply the effects of EXR on EXV and OIlV, respectively at the th regime. The coefficient  and  measure the impact of OILV on EXV and EXV on OILV, respectively at the  sub-period of the respective country.

Indicate significance at 1% level of significance.

Indicate significance at 5% level of significance.

Indicate significance at 10% level of significance.

Impulse response functions

Given the coefficients of the VAR model for the last regime of each of the four countries, we have obtained the generalized impulse response functions (GIRFs).

Fig. 2(a) and (b) depict the impulse responses of exchange rate return to the exchange rate volatility and oil price volatility shocks, respectively for India. It is evident from Fig. 2(a) that the Indian Rupee initially depreciates after the impact of the EXV shock, reaches to 0.025 percentage point. Then, within one week, the exchange rate becomes insignificant. However, the Indian Rupee appreciates after the initial impact of OILV shock, which reaches to the downward peak at −0.025% and takes two weeks to reach zero. The EXR innovation increases EXV initially, then it decreases (Fig. 2(c)). On the other hand, OILV shock decreases EXV (Fig. 2(d)). Fig. 2(e) depicts that the shock in EXR initially raises OILV and then it declines and takes two to three weeks to dissipate. From Fig. 2(f), it is evident that EXV innovation reduces OILV.

Fig. 2

Impulse responses to generalized one standard deviation. The figure shows the plot of impulse responses to generalized one standard deviation innovation. Figures (a)–(f) are for India; (g)–(l) are for China; (m)–(r) are for Japan; and (s)–(x) are for Korea.

In case of China, the EXV shock has a very little impact on exchange rate return (c.f. Fig. 2(g)). On the other hand, shock in OILV has a significant effect on exchange rate return for China (Fig. 2(h)). The return declines after the impact of OILV shock. Then, there is a stimulus in the return, which takes approximately two weeks to dissipate. One unit shock in EXR increases EXV initially, then after reaching a peak it declines (Fig. 2(i)). OILV innovation has a negative effect on EXV (Fig. 2(j)). Initial response of OILV to EXR Innovation is positive, reaches at the peak of 12 percentage point (Fig. 2(k)). Thereafter, OILV declines and takes two weeks to become insignificant. It can be noted from Fig. 2(l) that the EXV shock reduces OILV. In case of Japan, similar kinds of behavior have been observed in the return series to the shocks in EXV and OILV (see Fig. 2(m) and (n)). In both the cases, the return series decreases after the impact of shocks. The Japanese Yen declines after the impact of the EXV shock and within two weeks it reaches to zero. The unit shock in OILV initially increases the exchange rate return, which reaches to 0.065%, then it declines and within two weeks it becomes zero. It is worth noting that the response of EXV to the shock in EXR is negative (reaches to the downward peak −0.03%) and then the effect dissolves within one week (Fig. 2(o)). On the other hand, OILV innovation has a very short term impact on EXV (Fig. 2(p)). The effect of EXR innovation to OILV (c.f. Fig. 2(q)) is negative and it fully dissipates within three weeks. It is evident from Fig. 2(r) that EXV innovation reduces OILV.

In case of Korea, the GIRFs for a unit shock in EXV and OILV looks very similar to each other. The Korean Won initially appreciates after the impact of both the shocks, then quickly it reverts to a positive value (see Fig. 2(s) and (t)). Ultimately, the impact of shock dissolves within two weeks. It is observed from Fig. 2(u) that the initial response of EXV to EXR Innovation is positive. However, after reaches at a peak EXV quickly reverts to zero within 10 days. On the other hand, OILV shock reduces EXV (Fig. 2(v)). The initial impact of EXR innovation to OILV is positive which reaches to a peak of around 20 percent point and then it declines and reverts to zero within three weeks (Fig. 2(w)). Finally, from Fig. 2(x) it is noted that the innovation in EXV declines OILV.

Robustness check

As a robustness check, we tested for the presence or absence of break in the trend function of a time series to corroborate our empirical analysis. Perron and Yabu (2009) argue that knowing if a structural break is present in the trend function of a given time series is very crucial before applying a unit root test, especially when the break points are unknown. Thus, knowledge about the beaks in the trend function is important in deciding which unit root test is more appropriate. For instance, One can implement (Narayan and Popp, 2010) in case of two unknown breaks or Carrion-i Silvestre et al. (2009) in case of multiple breaks. However, tests for structural breaks in terms of intercept and/or slope of a deterministic trend function suggest that the performance of these tests depend on whether the time series under study is stationary or non-stationary having unit roots, thus leading to a ‘circularity’ problem. To ameliorate upon this drawback, Perron and Yabu (2009) proposed a test to detect structural changes in the trend function of a time series. This test can be without any prior knowledge of the stationarity properties of the noise component.

We perform the Perron and Yabu (2009) test to detect the presence of structural breaks in a time series. The results are reported in Table 6. The rejection of the null hypothesis indicates presence of break/s in the time series while a nonrejection of the null hypothesis indicates an absence of structural breaks. The results suggests that the null of no structural breaks statistic for exchange rates returns (EXR) in case of all four countries could not be rejected. Likewise, we did not find any significant structural breaks with regards to oil price returns (OILR). Hence, we proceeded with the ADF test, as discussed earlier, to check for stationarity. Overall, our results are robust since we could not find any structural breaks in the trend function of time series and validate our empirical analysis.

Table 6

Robustness check.

Variables	Test statistic
EXR_India	2.582
EXR_China	3.023
EXR_Japan	−0.137
EXR_Korea	0.298
OILR	−0.241

This table reports the Perron and Yabu (2009) test results. The test is implemented to detect the presence of structural break in the trend function of a time series. The rejection of the null hypothesis indicates a break in the time series while a nonrejection of the null indicates an absence of structural breaks.

Conclusion and policy implications

In this paper, we examine whether the sudden occurrence of the COVID-19 has strengthened the relationship between the oil prices and exchange rates. Further, we extend the existing literature by investigating the stability of the relationships between the two markets by accounting for breaks in the relationship under the assumption that the dynamic interactions may not not be permanent, rather temporary. We choose four largest net oil-importing countries in Asia, viz. India, China, Japan and Korea and utilized daily data from January 2, 2017 to August 10, 2020 that covers the COVID-19 outbreak period. First, we apply the ‘two-step’ procedure and thus a VAR-AGARCH model is run to estimate the conditional volatilities. Second, we augmented the second step of the ’two-step’ procedure by giving due consideration to structural breaks and thus testing the stability in the relationships. Finally, we conduct a sub-period wise analysis.

We report the following key findings. First, we find that there is high level of risk in both the exchange rate and oil markets due to the rising uncertainty during the COVID-19 crisis. Second, we find that the impact of COVID-19 on oil price-exchange rate relationships is heterogeneous across four countries. Third, we find significant structural breaks in the relationships and time-varying. Specifically, we detect a break during the COVID-19 period for all countries, coinciding with the rising uncertainty about the pandemic as well as the plunge in oil prices. Overall, we find significant interactions between the two markets and the relationship between the duo intensified in the outbreak period. Thus, we conclude that COVID-19 pandemic has strengthened the predictability of variations in oil prices and exchange rates.

The findings from our empirical investigation have important implications. In case of India, our findings imply that the interrelations intensified during the pandemic period. While we could only find positive volatility spillovers from oil prices to exchange rates in the pre-pandemic period, we find significant impact of exchange rate returns and volatility on oil price volatility during the pandemic. Specifically, an increase in oil price volatility increases the exchange rate volatility. This result is not surprising since the oil price volatility caused by a demand-driven shock, albeit transitory in nature, led to reduction in current account deficits. The results are particularly relevant for investors who look to diversify their portfolio in the face of a depreciation of the Indian rupee. The investors can shift towards foreign assets in case the returns on rupee-denominated domestic assets fall. For China, we find only exchange rate return-volatility relation whereby an exchange rate volatility shock appreciates the Chinese Yuan by 0.59%. In contrast, we find that during the pandemic period, a 1% appreciation of the Chinese Yuan leads to a rise in both exchange rate volatility and oil price volatility by 0.05% and 30.92%, respectively. Interestingly, we did not find any significant impact of oil price volatility on exchange rate. This finding can be attributed to the fact that Chinese authorities actively supported the Yuan through monetary stimulus in the form of reverse repo cuts and reduction of reserve requirements for banks, and relaxing capital flow management measures during the pandemic.

For Japan, we find results similar to China for the pandemic period. We find that the exchange rate volatility shock appreciates the Japanese Yen but the impact is opposite when the direction of causal relationship is reversed. Likewise, for the pandemic period, the relationships are strengthened which indicates better predictability of oil prices and exchange rates in face of high uncertainties. In particular, we find significant positive impact of exchange rate volatility on its returns while a 1% appreciation of the Yen leads to a decline in exchange rate and oil price volatilities by 0.5% and 15.71%, respectively. Finally, in case of Korea, we do not find any significant relations between exchange rate returns and the oil price and exchange rate volatilities. However, we find significant positive return-volatility spillover wherein a 1% appreciation of the Korean Won leads to a 14.88% increase in oil price volatility and 0.15% rise in exchange rate volatility. These results are in line with the findings from India and China.

Overall, our findings imply that the interactions strengthened in the wake of an unanticipated even such as the COVID-19 crisis and that there are significant spillovers when the markets are both integrated and volatile. The investors in the two markets are taking stock of valuable information from such a dramatic shock. Thus, portfolio switches between domestic currency-denominated assets and foreign currency-denominated assets in case of a depreciation/appreciation of the currency can be an avenue for diversification. For macroeconomic policymakers, oil prices have always been a major concern of rising domestic inflation and current account deficits in large net oil-importing countries. However, our finding that exchange rate can also affect the oil price volatility in the short-run, since the invoicing is done in US dollars, imply that an appreciation of importing countries’ currencies against the US dollar can lead to oil price movements. The empirical findings analysis in this paper is based on oil-importing countries and opens a debate on the possible impact of a shock – as large as the COVID-19 – on the interrelations between oil prices and exchange rates. Thus, future research warrants an investigation examine the impact of the pandemic shock in oil-exporting countries.