Abstract

This paper examines the relationship between oil prices, exchange rates, and BRVM-C stock prices within the context of the COVID-19 pandemic. We identified a cointegration relationship among the variables, allowing us to estimate the long-run elasticities. Our findings indicate a statistically significant positive effect of oil prices on stock prices prior to the onset of the pandemic. However, during the COVID-19 period, this relationship lost significance, while a positive and significant effect was observed in the post-COVID resilience period. In terms of exchange rates, we found a statistically significant negative effect on stock prices before the pandemic. This relationship shifted to a positive and significant effect during the spread of the virus, followed by a return to a negative and significant effect in the resilience period. Additionally, our short-run elasticity estimates reveal that oil prices had a negative and significant impact on stock returns before the pandemic, a negligible effect during the pandemic, and a non-significant effect thereafter. Conversely, exchange rates exhibited a positive but non-significant effect before and during the pandemic, transitioning to a negative and significant effect only in the period following the pandemic.

1. Introduction

Oil prices play a crucial role in the functioning of socioeconomic systems. Since the 1970s, fluctuations in oil prices have garnered significant attention. If oil prices influence economic activities, it stands to reason that they will also impact the stock market ¹. Theoretically, oil price shocks can affect stock market returns or prices through their impact on expected earnings, cash flows, and/or discount rates ². Additionally, a substantial body of research on the effects of oil is based on macroeconomic variables such as inflation, business cycles, exchange rates, and growth rates ³.

Since 2019, the global economy has been significantly impacted by the COVID-19 pandemic. Financial analysts frequently draw comparisons between the effects of COVID-19 and those of the global financial crisis of 2008. However, ⁴ emphasizes a distinction between these two crises, referring to the current pandemic-related challenges as the “Great Compression” ⁵. Building on the foundational research by ⁶ and ⁷, a substantial body of literature has emerged demonstrating that fluctuations in oil prices have a pronounced influence on various economic indicators ^{7, 8, 9, 10}. ⁷ discovered that any increase in oil prices tends to have a detrimental effect on economic growth in the United States, although it remains unclear if economic growth accelerates during periods of low oil prices.

While some research supports a nonlinear relationship between oil prices and macroeconomic variables ^{11, 12, 13}, recent attention has primarily focused on the connection between changes in oil prices and stock market performance ^{14, 15, 16, 17, 18, 19, 20}. These diverse studies reveal that there is no consensus regarding the impact of oil price shocks on the stock market. Some research indicates a negative correlation; for instance, ¹⁴ utilized variance decomposition analysis to establish that oil price shocks adversely affect stock market returns in Greece during the initial four months. Similarly, ¹⁹ examined the relationship between oil prices and stock returns in the United States, concluding that an increase in oil prices typically leads to a decline in short-term stock returns.

Several studies have identified a positive correlation between oil price shocks and stock market returns. Notably, research by ²⁰ and ²¹ highlight that oil price shocks have a favorable impact on stock returns in various markets, including China. Specifically, ²¹ examined the influence of oil price fluctuations and volatility on the Chinese stock market index and five sector returns, concluding that these shocks positively affect Chinese stock returns. Additionally, ²⁰ reported significant positive effects of oil prices on stock prices in Vietnam.

Conversely, some studies have reported an insignificant response of stock markets to changes in oil prices. ²², for instance, analyzed the relationship between oil price shocks and stock market behavior in Nigeria and discovered a negligible positive response at first, which transitioned to a negative effect over time, influenced by the nature of the oil price shocks.

Furthermore, recent literature has examined the effects of oil prices on various industrial sectors within the stock market. Research by ^{23, 24, 25}, and ²⁶ has contributed to this area of study. In his 2011 research, Arouri analyzed the impact of oil prices on stock returns for 12 European sectoral stock indices using weekly data. His findings indicate that oil price variations positively influence certain sector stock returns while adversely affecting others. Specifically, indices related to oil and gas, basic materials, and consumer services sectors exhibit a positive relationship with oil price changes, whereas sectors such as financials, food and beverages, technology, telecommunications, healthcare, transportation, and manufacturing show a negative response to oil price fluctuations.

²⁷ and ^{28, 29} contribute to the understanding of the relationship between changes in oil prices and their impact on the economy and financial markets by distinguishing the origins of oil price shocks. Following ²⁹, the real oil price can be classified into three categories: crude oil supply shocks, shocks to the global demand for all industrial commodities, and demand shocks specific to the global crude oil market ³⁰. A growing body of literature suggests that supply-side oil price shocks no longer significantly affect the economy or financial markets ^{29, 31, 32, 33, 34}.

While numerous studies have examined the relationship between oil price fluctuations and stock market returns, a consensus on a uniform relationship has yet to be established. Furthermore, most of these studies have focused on developed countries, with only a few addressing developing regions such as ECOWAS and WAEMU (West African Economic and Monetary Union). This paper aims to reassess the impact of oil price changes on the BRVM Composite stock returns in the context of the COVID-19 pandemic. Specifically, it seeks to analyze how fluctuations in oil prices and the ongoing pandemic have influenced BRVM Composite stock returns.

This study contributes to existing literature in two significant ways. First, it employs the ³⁵’s test to explore the cointegration relationship while accounting for potential endogenous structural breaks. In this research, we utilize the last three models proposed by Gregory-Hansen. The second contribution of this paper involves considering the impact of the COVID-19 pandemic in the analysis of how emerging stock markets respond to fluctuations in oil prices and exchange rates.

The structure of this paper is organized as follows: the next section will present the data and methodology employed in the study. Section 3 will discuss the estimation process, and the results obtained. Finally, Section 4 will provide concluding remarks.

2. Data and Methodology

We utilized daily data from the INVESTIN.COM database covering the period from May 1, 2000, to September 6, 2021. The data series include oil prices (BR Brent), the BRVM-C stock price index, and the nominal exchange rate. The exchange rate analyzed is the euro-to-dollar rate, as the CFA franc and euro maintain a fixed exchange rate (1€ = 655.96 CFA francs). The dataset is categorized into three distinct periods: the first period (pre-COVID-19) spans from May 1, 2000, to February 28, 2020; the second period (during COVID-19) occurs from March 2, 2020, to June 8, 2020; and the third period (post-COVID-19), which corresponds to the timeframe following the last phase of confinement in the region, starts from June 9, 2020. The BRVM serves as the stock market for member countries of the West African Economic and Monetary Union (WAEMU). Observations of data plots indicate that all series exhibit a trend (refer to Figures 1-24 in the appendix), with oil prices and exchange rate growth displaying greater volatility compared to BRVM-C stock returns. This finding is inconsistent with the results presented by ³⁶. Additionally, return series are easier to analyze than raw series, as returns adhere more closely to statistical properties ³⁷.

We consider the following long-run model (Narayan and Narayan, 2010):

(1)

Where is the BRVM-C stock price returns (growth), is oil prices and is the exchange rate. There are many types of returns but those interest us is the continuous component returns because it includes all the proprieties of the others method of calculation like this:

(2)

We will present here augmented Dickey-Fuller (ADF), ADF-GLS and Philip-perron (PP) unit root tests.

Augmented Dickey-Fuller test (ADF, 1981)

³⁸ unit root test supposes that the series is an autoregressive process . Higher-order lags of series correlation can violate the assumption of white noise in disturbances. To address this issue, the Augmented Dickey-Fuller (ADF) test is employed to adjust for higher-order correlations. This assume that the series contain an process with lagged difference terms of in the right hand side.

(3)

Where are exogenous regressors which can be considered like optional (a constant, or a constant and trend). , and are parameters to be estimated.

Dickey-Fuller test with GLS detrending (DFGLS)

In the above test, one can include a constant, or a constant and a time trend in the ADF test. These situations lead ³⁹ to propose a modification of the ADF tests. It consists to remove (detrended data) explanatory variables before running the test regression like this:

(4)

Where the quasi-difference of depends on the value of the specific point alternative against which on wish to test the null. Finally, the DFGLS test involves estimating the standard ADF test (3) after substituting the GLS detrend the GLS detrended for the original as follows:

(5)

The DFGLS t-ratio follows a Dickey-Fuller distribution only in the case of constant, but the asymptotic distribution differs when including a constant and a time trend ( ³⁹ ERS, 1996).

Phillips-Perron (1988) unit root test

A nonparametric method that allows to control for the serial correlation when we test for unit root is proposed by ⁴⁰. This test is based on the statistics:

(6)

Where is the coefficient standard error, the estimated, s the standard error of the test regression, the consistent estimate of the error variance and is an estimator of the residual spectrum at frequency zero.

The following sub-section presents Johansen and Gregory-Hansen tests of cointegration.

Johansen test

Pour qu’une relation de cointégration existe entre des variables, deux conditions doivent être réunies. D’abord, les variables doivent être non stationnaires et intégrées du même ordre. Ensuite, leurs tendances stochastiques doivent être liées. Il existe donc une relation de long terme entre les variables. L'approche initiale de Johansen repose sur l’estimation d’un VAR

(7)

Où est un vecteur de variables, un vecteur des variables exogènes incluant une tendance et des variables indicatrices, et le vecteur des perturbations suivant une loi normale.

Cependant puisque certaines variables économiques sont I (1), l’estimation du VAR en différence peut conduire à une perte d’information si les séries sont effectivement cointégrées. Pour tenir compte de cette dimension potentiellement cointégrée des variables économiques, Johansen (1988) réécrit (10) sous la forme Vectorielle à Correction d’Erreurs (VECM)

(8)

Selon l’auteur, si on étudie variables avec , on peut avoir jusqu’à () relations de cointégration. Ainsi, ce test repose sur l’hypothèse nulle qu’il existe au plus relations de cointégration. Les statistiques du test sont la suivante

(9)

(10)

est la valeur propre maximale estimée.

La première statistique teste l’hypothèse nulle de cointégration de rang contre l’alternative de la stationnarité . Cette statistique est appelée statistique de la trace. La seconde statistique teste contre. Elle porte le nom de statistique de la valeur propre maximale.

Structural break cointegration test (Gregory-Hansen, 1996)

Following the Johansen test, we proceeded with additional cointegration tests that account for structural breaks: the ³⁵ and its extension by ⁴¹, which incorporates two endogenous structural breaks. The first test addresses one endogenous structural break, while the second accommodates two. The ³⁵’s methodology offers several models that facilitate the analysis of cointegration with structural changes, including standard cointegration, level shift, level shift with trend, and regime shift.

Model 1: Standard cointegration

(11)

Where is and is . In this model, and describe the m-dimensional hyperplane towards which the vector process tends over time (see ³⁵. If model 1 is to capture a long-run relationship, one will want to consider and as time invariant.

Model 2:

(12)

where and represents the intercept before the shift, and represents the change in the intercept at the time of the shift. Adding a time trend into the level shift model we obtain model 3.

Model 3: (13)

Another possible structural change allows the slope vector to shift as well. This permits the equilibrium relation to rotate as well as shift parallel. It is called the regime shift model.

Model 4: (14)

Whith and are as in the level shift model, denotes the cointegrating slope coefficient before the regime the regime shift, and denotes the change in the slope coefficients. ³⁵ propose a suite of tests to analyze the cointegration between and with structural change. One will be interested in the smallest values of the test statistics. These statistics are commonly used ADF statistics and extensions of the and test statistics of ⁴³. These statistics are:

(15)

(16)

(17)

All these tests have the same null hypothesis of no cointegration.

⁴² and ³⁵’s tests confirm the presence of long-run relationship between the set of variables. In this case we should use a dynamic error correction model in order to catch the short-run dynamic as shown by ⁴⁴. The short-run regression model dynamic is the following:

(18)

All variables in equation (9) are defined in section 3.1 and 3.2, is the lag length and is the lagged error correction term.

We apply the parameter stability test proposed by ⁴⁵ accounting for one or more unknown structural breakpoints.

Test de stabilité des paramètres de Quandt-Andrews (1993)

The Quandt-Andrews test (1993) ( ⁴⁵) investigates one or more unknow structural breakpoints in the sample. This test is based on the Sup or Maximum statistic (or Likelihood Ratio F-statistic, Wald F-statistic) the Exp statistic and the Ave statistic with null hypothesis of “no breakpoints”. In this test, the trimming of 15% is used. The Maximum statistic is simply the maximum of the individual Chow F-statistics like follow:

(19)

The “Exp statistics” takes this form:

(20)

Finally, the “Ave statistics” is the simple average of the individual F-statistics:

(21)

3. Results and Discussion

The initial phase of our empirical analysis involves conducting unit root tests. We implement three specific tests: the Augmented Dickey-Fuller (ADF) test, the Dickey-Fuller generalized least squares (DF-GLS) test, and the Phillips-Perron (PP) test. These tests are applied to the BRVM-C Stock Price (BSP), the BRVM-C Growth Rate of Stock Price (BGSP), Crude Oil Price (OP), the Growth Rate of Crude Oil Price (OPG), the Nominal Exchange Rate (ER), and the Nominal Exchange Growth Rate. The results are summarized in Table 1. All tests share a common null hypothesis, which posits the presence of a unit root. Our findings indicate that the variables BSP, OP, and ER display a unit root. Conversely, we successfully reject the null hypothesis at the 1% level of significance for the variables BGSP, OPG, and ERG. These results facilitate our exploration of potential long-term relationships among these variables.

We examine the long-run relationship between BSPG, LOP, and LER (exchange rate). The methodology outlined by ⁴² and ⁴⁶ presents a multivariate long-run equilibrium framework based on vector autoregression (VAR), which enables us to identify statistically significant long-run relationships. The Johansen cointegration test allows for the possibility of multiple long-run equilibrium relationships. Prior to conducting this test, it is essential to determine the optimal lag length of the VAR model. The optimal lag lengths identified are 4, 1, and 1, corresponding to the periods before, during, and after the COVID-19 pandemic. This methodology involves two key statistics: trace statistics and the maximum eigenvalue statistics. Both the trace and maximum eigenvalue statistics indicate one (1) long-run relationship across all specified periods (refer to Table 2).

Table 3 reports the results of the three tests: the one-break, two-break and the three-break tests. When using ³⁵ Gregory-Hansen (1996) critical values at 5% level (see ³⁵ p.11 Table 1) the null of no cointegration is rejected for almost all these tests in presence of structural breaks and then the Johansen test above is consistent. There is a long-run relationship between the BRVM-C stock returns, crude oil price and exchange rate.

In Table 4, we report the long-run elasticities obtained from Ordinary Least Squares (OLS). All variables are statistically significant. Before the COVID-19 pandemic, an increase of 1% of oil prices increased the BRVM-C stock prices around 0.69% while an increase of 1% exchange rate led to a decrease of 0.24% of BRVM-C stock prices. This result is conformed to those of ⁴⁷ and ²⁰.

During the COVID-19 pandemic period, the effect of the exchange rate became positive on BSP. Thus, an increase in the exchange rate increases BSP of 1.25%. COVID-19 influenced BSP via the exchange rate comparatively to others sub periods (before and after). An increase in exchange corresponds to a depreciation of national current. This situation can lead to a surplus on balance of trade. But the oil prices are statistically not significant corresponding Babatunde and al. 2012 study in Nigeria.

After COVID-19, the rise of 1% of oil prices and exchange rates led respectively to an increase of 0.19% and a fall of 0.41% of BRVM-C stock prices. We note that before and after Covid-19 period response in the same way in contrast to the middle period which is during COVID-19. Following Narayan and Narayan, (2010), the relationship exist between exchanges rates and stock returns could be negative or positive. It depends on the type of country (import or export dominant country). According to ⁴⁸ the competitiveness of firms can be affected by reducing their earnings.

The long-run relationship is captured by the lagged value of error correction term reflecting the convergence of the long-run equilibrium system. The above equation considers the short and long-run relationship between BRVM-C stock returns, the oil price and exchange rate returns.

Table 5 reports the short-run dynamic results. As shown for all sub-periods, in the first there is a short-run relationship between OPG and BRVM-C stock returns. An increase of 1% of OPG led to a decrease in BSPG by around 0.01%. The adjustment rate of 0.7% is low. During and after COVID-19 pandemic periods, only the constant and the adjustment rate are statistically significant. The adjustment rate during pandemic period 13.1% is the largest of all but those after pandemic 4.3%. The insignificant short run during and after could be explained by time latency before seeing pandemic effect on the stock market returns. Finally, the residual diagnostic (Table 6) reveals that over the three periods, only those during pandemic have homoscedasticity and no serial correlation of residuals. This can also be explained by the fact that during this period most of the world economic and financial activities have slowed down.

The short-run diagnostic for before and after the COVID-19 exhibit ARCH effects. This situation could lead to bias in using the short-run OLS estimation. Hence, we estimate the short-run model using a GARCH (1, 1) specification. Before-COVID-19, we apply a GARCH (1, 1) model and thus the ARCH LM (1, 4816) test statistic 0.880 with 0.348 of probability value does not exhibit an ARCH effect.

The results are reported in Table 7. Before the pandemic, with 3374 breakpoints considered, the null hypothesis is not rejected for some tests. The during-COVID-19 period considered 45 breakpoints and the null hypothesis of no breakpoints cannot be rejected. After the pandemic, Quandt-Andrews unknown breakpoint test cannot also reject the null hypothesis with 175 number of breaks compared. These results are confirmed by CUSUM test below. This means that parameters are stable.

4. Concluding Remarks

This paper aims to analyze the relationship between oil prices, exchange rates, and BRVM-C stock prices within the context of the COVID-19 pandemic. The pandemic has been identified as a source of systematic risk, as noted by ⁴⁹, and warrants an examination of its financial implications. In response to the pandemic, many governments implemented stringent measures such as shutdowns, lockdowns, and restrictions on mobility, which have led to a decline in economic activity and financial instability. Our analysis utilizes daily data from three distinct periods: prior to the pandemic (from January 5, 2000, to February 28, 2020), during the pandemic (from the declaration of the first case on March 2, 2020, to the conclusion of restrictions on June 8, 2020), and post-pandemic (from June 9, 2020, to March 18, 2021, corresponding to the resilience period).

Our analysis indicates that stock prices on the BRVM-C, along with oil prices and nominal exchange rates, exhibit a cointegrated relationship, as confirmed by the Johansen test. Additionally, the Gregory-Hansen structural break cointegration test validates these findings, suggesting a long-term relationship among these variables. We proceeded to estimate the long-run elasticities and discovered a statistically significant positive impact of oil prices on stock prices prior to the COVID-19 pandemic. This relationship became statistically insignificant during the COVID-19 period, while re-establishing a significant positive correlation post-COVID, referred to as the resilience period.

Regarding exchange rates, we found that their impact on stock prices was statistically significant and negative before the onset of the pandemic. During the spread of the virus, this relationship shifted to a positive and significant correlation, followed by a negative and significant association during the resilience period. We have estimated the short-run elasticities and found that oil prices exert a negative and statistically significant effect on stock returns prior to the COVID-19 pandemic. During and after the pandemic, however, this effect was not significant. In contrast, exchange rates demonstrated a positive but non-significant effect on stock returns before and during the pandemic, transitioning to a negative and statistically significant impact only after the pandemic period. The lack of significant effects of oil prices and exchange rates on stock returns during the COVID-19 period may be attributed to the general slowdown in global economic and financial activities during this time.

APPENDIX

Before-COVID-19 series graphs

Before-COVID-19 returns graphs

During-COVID-19 series graphs

During-COVID-19: returns graphs

After-COVID-19 series graphs

After-COVID-19 returns graphs

CUSUM tests

Descriptive statistics

Before COVID-19 pandemic, the growth rate of all series pointed of smallest standard deviation comparing to crude (gross) series. This result is similar to during and after COVID-19 pandemic periods. The growth rate series have also a mean which is approximately equal to median value. When the skewness is negative and different to zero, the returns profitability responds more to a negative chock than a positive chock. During pandemic situation, oil prices series and the nominal exchange growth rate are normally distributed. It also the case of the nominal exchange growth rate after COVID-19 period.

References