Time Varying Impact of Oil Prices on Stock Returns: Evidence from Developing Markets

Chen et al. [16] are among the first to study the impact of macroeconomic variables on the stock market. They found that there is no impact of oil prices on stock prices. Similarly, Hamao [17] and Huang et al. [18] and Jones and Kaul [19] are among the earliest studies which explored the relationship between oil prices and the stock market. Hamao [17] and Huang et al. [18] also did not find the evidence of linkage between the two. However, Jones and Kaul [19] considering Producer price index as oil price, found the relationship between oil prices and stock market of Canada and the USA. In the case of Japan and the United Kingdom, the relationship was found to be weaker. Karim et al. [8] also found a weaker relationship between S&P 500 and future oil prices.

Several studies have analyzed the association between oil prices and exchange rates. Krugman [20] is one of the earliest studies based on the relationship between oil prices and exchange rate. In the study, the author developed a theoretical model to examine the effect of change in oil prices on the exchange rate. The study revealed that the short-run effect and long-run effect are in opposite direction. In the short run, increase in oil price would lead to the appreciation of the dollar, but in long-run, the dollar would depreciate [21]. Further, Amano and Van Norden [22] examined the long-term relationship between the domestic price of oil and real effective exchange rate. The study applied Granger causality and found that there is a robust relationship between the real domestic price of oil and real effective exchange rate. Other studies which have explored the relationship are [15, 23, 24].

Enormous literature is available based on the relationship between oil prices and the stock market. Basher and Sadorsky [24] investigated the impact of oil prices on the stock markets in emerging economies. The study used international multi-factor model which allows conditional and unconditional risk factors to assess the impact of oil prices on stock market. The study found a strong relationship between the variables. Maghyereh and Al-Kandari [25] applied rank tests of cointegration to analyse the impact of oil prices on stock markets of the Gulf Cooperation Council (GCC). The study found that there is a non-linear impact of oil prices on the stock market. One more similar study is by Narayan and Narayan [26]. The study considered the impact of oil prices on the stock market of Vietnam from 2000 to 2008. The study found the cointegrating relationship between the variables. Sukcharoen et al. [27] analysed the relationship between oil price and the stock market indices, applying copula-based interdependence on data ranging from 1982 to 2007. The study found that there is largely weak dependence between the oil prices and the stock indices, excluding the USA and Canada. Stock indices of the countries have shown strong dependence with oil prices.

There are other studies also which have considered the same connection, such as Odusami [28], Cong et al. [29], and Miller and Ratti [14], investigated the relationship between the stock markets of OECD and oil price shocks by applying panel threshold cointegration approach on data from January 1995 to 2009. They found a cointegrating relationship between the variables. Bidirectional causality was also found between oil price and stock market indices of OECD and non-OECD countries [14]. Aydogan and Berk [15] analyzed the similar relationship in Turkey. The study applied a variance decomposition model on data ranging from 1990 to 2011. The study found little evidence of any relationship between the oil prices and stock market of Turkey [15]. However, Tursoy and Faisal [30] reported positive relationship between crude oil and stock prices in Turkkey [30].

Apart from linkages between crude oil prices and the stock market, few other financial and macroeconomic factors have also been included in the recent studies. Jain and Ghosh [31] studied the dynamic linkages between crude oil prices, exchange rate and stock market index in India. The findings indicated that a decrease in oil price leads to a decrease in the value of the Indian rupee and benchmark index (Sensex). Shinghal et al. [32] found that in the long-run oil prices affect the exchange rate in Mexico.

As discussed above, there are studies on the relationship between oil prices and stock markets. But the studies based on the linkages in the context of financial crises are limited. Wen et al. [33] examined the contagion effect between oil prices and stock markets of China and the USA. Based on the time-varying copula, the study found that dependence increased between the two. Similarly, Du and He [34] analysed the relationship between WTI and the S&P stock index. The study utilized Granger causality and Kernel-based value at risk measure. The study found an increased bidirectional correlation between the oil and stock market owing to the crisis. In the same way, Bouri [35] analysed the linkage between the crude oil price and Lebanon stock market by applying VAR-GARCH model on weekly data ranging from 1998 to 2014. This study also found that the crisis intensified the relationship. Mensah et al. [36] assessed the long-term relationship between Crude oil prices and the exchange rate of major oil-dependent economies in the pre and post period of the global financial crisis of 2008-2009. It was observed that there is a long-term relationship between oil prices and the exchange rates of major oil-exporting countries, particularly in the post-crisis period.

Mensi [37] explored the co-movement between oil prices and the stock market of Saudi Arabia during the Global Financial Crisis of 2008-2009(GFC). The author applied the value at risk and Wavelet method to study the impact. Findings indicate increased co-movement between the oil prices and the stock market. Ajmi et al. [38] applied the non-linear and asymmetrical causality test to explore the impact of oil prices and the stock market of MENA counties during the financial crisis of 2008-2009. The authors found that the interaction is non-linear between stock markets and oil prices. Fayyad and Daly [39] compared the impact of oil prices on the stock markets of GCC and UK and USA. The authors applied the Vector autoregression on the daily data ranging from 2005 to 2010. Authors found that predictive power and the impulsive response of oil prices increased during the Global financial crisis [40, 41].

In the Context of the COVID-19 crisis, Albulescu et al. [42] investigated the relationship between oil prices and stock markets of the USA during the COVID-19 crisis. The authors applied the wavelet approach to the data from February 2020 to August 2020. Results show that there was a lead-lag relationship between the oil prices and stock markets. Moreover, results show that during the early phase of COVID-19, March, and April 2020, co-movement was found even at the low scales of 3-5 days. Similarly, Sharif et al. [43] analyzed the relationship in the time and frequency domain by applying the wavelet-based granger causality. Authors found the impact of oil price shocks on the US stock market, where the COVID-19 outbreak was found to have a greater effect than the geopolitical and economic uncertainty.

The period also witnessed high volatility of exchange rate. An inverse relationship between the oil prices and the exchange rate was also [44-46]. Li and Wei [47] conducted a study in China. Time-varying based copula indicated that crisis increases the dependence between the oil prices and the stock market and long-run dependency increases more significantly than the short run. The study scrutinized the long-term association between crude oil price and the China stock market during the subprime crisis. The study used a nonlinear threshold cointegration method within a multivariate framework. Three macroeconomic factors (foreign exchange market, domestic economic development, and total foreign trade volume in China) were used as transmission channels between oil price and stock price. The study found that, among the three macroeconomic factors, the exchange rate plays the most significant role in transmitting the impact of oil prices on the stock market particularly after the financial crisis [21].

The impulse response function is a widely applied tool to study the impact of one variable on the other hand [24] applied the variance decomposition, Granger causality, and Impulse response function to study the relationship between the oil prices and exchange rate. The study utilizes variance decomposition as it gives information about random innovations in the system. Similarly, the Impulse response function illustrates the dynamic behavior of one variable on the current and future innovations of another variable. Sahu et al. [48] applied the Variance decomposition and Impulse response, as causality tests do not capture the dynamic nature and degree of strength of the relationship.

Thus, the review of previous studies discloses that enormous literature is available on the association between crude oil prices and the stock market. Studies are available which have considered other financial factors also, such as, exchange rate. The linkage or interdependence has been studied during the crises also. Still, the studies based on the linkages in the context of the crises are limited. The studies based on the impact of COVID 19 are even rarer. The present study utilises panel Vector Autoregressive (pVAR) model with vector of three endogenous variables. The variables are the returns series of crude oil price, exchange rate of ten countries and stock price of benchmark index of ten countries. The analysis is separately carried out for the pre and post COVID-19 announcement. The purpose is to assess transmission of the shock owing to COVID-19.

To examine the relationship between crude oil, exchange rate and stock returns of the selected emerging markets, we employ panel vector autoregression first introduced by Holtz-Eakin et al. [52] and later implemented in GMM framework by Love and Zicchino [53]. Recently, Salisu et al. [23] have also employed panel VAR to examine the impact of crude oil on index returns.

For setting up the panel VAR, we employ a vector of three endogenous variables i.e., $\mathrm{Y}_{\mathrm{t}}^{\prime}=\left[\begin{array}{lll}C_{t} & X_{t} & R_{t}\end{array}\right]^{\prime}$  where $C_{t}, X_{t}$ and $R_{t}$  represent returns on crude oil, exchange rate and stock index respectively. A panel VAR with K endogenous variables and L exogenous variables can be represented as follows [54].

$\mathrm{Y}_{\mathrm{i}, \mathrm{t}}=\sum_{j=1}^{p} \mathrm{Y}_{i, t-j} \mathrm{~A}_{j}+X_{t} \mathrm{~B}+\mu_{i}+\varepsilon_{i, t}$      (1)

where, $Y_{i, t}$ and $Y_{i, t-j}$  are 1×K vectors of endogenous variables; $X_{t}$ is a 1×L vector of exogeneous covariates; $\mu_{i}$ is 1×K vector or country specific fixed effect and $\varepsilon_{i, t}$  is a 1×K vector of error terms. $A_{j}$  and Β are K×K and L×K matrices of coefficients. In our case, the PVAR is given as follows:

$R_{i t}=\sum_{j=1}^{p} \alpha_{j}^{(r)} C_{i, t-j}+\sum_{j=1}^{p} \beta_{j}^{(r)} X_{i, t-j}+\sum_{j=1}^{p} \gamma_{j}^{(r)} R_{i, t-j}$ $+\mu_{i}+\varepsilon_{i t}^{(r)}$        (2)

$X_{i t}=\sum_{j=1}^{p} \alpha_{j}^{(x)} C_{i, t-j}+\sum_{j=1}^{p} \beta_{j}^{(x)} X_{i, t-j}+\sum_{j=1}^{p} \gamma_{j}^{(x)} R_{i, t-j}$ $+v_{i}+\varepsilon_{i, t}^{(x)}$        (3)

$C_{i t}=\sum_{j=1}^{p} \alpha_{j}^{(c)} C_{i, t-j}+\sum_{j=1}^{p} \beta_{j}^{(c)} X_{i, t-j}+\sum_{j=1}^{p} \gamma_{j}^{(c)} R_{i, t-j}$ $+\omega_{i}+\varepsilon_{i t}^{(c)}$       (4)

where

R_(i,t), X_(i,t) and C_t are index returns, exchange rate returns and crude oil returns at time t for country i. α, β and γ are parameters to be estimated; μ_i, ν_i and ω_i are country specific effects in each of the equations; and, ε_(i,t), are error terms in the three equations.

The set of Eqns. (2)-(4) are estimated using GMM in the framework of panel vector autoregression for efficiency gains Holtz-Eakin et al. [52]. The set of equations is estimated separately for the periods before and after the announcement of COVID-19 as a pandemic.

In VAR models, the individual coefficients of lags of variables are usually not analysed. Rather the impact of lags of a variable on another variable is analysed with the help of and F-test. First, an unrestricted VAR is estimated with all the lags of all the variables, then, a restricted VAR is estimated by setting the coefficients of lags of a particular variable equal to zero. If the restriction is supported, then it is concluded that the lags of one variable are not helpful in predicting the other variable. This framework was first proposed by Granger (1969) and since then it is referred to as Granger causality. In our case, if all the α_j^((r) ) in Eq. (2) are jointly zero, then it would imply that changes in crude oil price do not impact stock returns. Or in other words, it will be said that oil returns do not Granger cause stock returns. Similarly, if in Eq. (2) all β_j^((r) ) are zero, then it would indicate that changes in exchange rate do not Granger cause stock returns. Granger causality test is helpful in determining whether history of a variable is helpful in predicting another variable. However, this test does not tell whether the variable has a positive or negative impact on another variable. For example, if Granger causality reveals that oil price changes Granger cause stock returns, then it implies that oil price changes are important for forecasting stock returns, but it does not reveal whether oil price changes have positive or negative impact on stock returns. In addition, Granger causality cannot reveal how long does it take for stock returns to fully absorb the impact of change in oil prices. Towards this purpose, impulse response function and variance decomposition are employed. Specifically, impulse response graph shows whether the impact of a unit shock to the error term in one equation has a positive or a negative impact and for how long this impact persists. variance error decomposition on the other hand tells the proportion of variation in forecast of a variable which is due to own shock versus shocks to the other variables. Because of the dynamic nature of VAR, the shocks to a variable not only affect that variable's future values, but also influence the future values of other variables. Error Variance decomposition reveals as to how much variation in h-step ahead forecast of a variable is explained by own versus other variables' shocks. We have used R and Stata-16 for carrying out the analyses.