The Nexus Between Oil Price Shock and the Exchange Rate in Bangladesh

We examine the nexus between oil price and exchange rate for Bangladesh economy by using annual data covering from 1980 to 2018. Given the stationarity properties, the Johansen cointegration and the ARDL bounds cointegration tests find a long-run cointegrating relationship between the variables. We reveal that oil price granger causes exchange rate in the long-run but not in the short-run. According to DOLS and DARDL methods, an increase in oil price appreciates exchange rate by 0.40% and 0.30%. We argue that the central bank’s proper monitoring mechanism is necessary to avoid oil price’s adverse effects on the exchange rate.


INTRODUCTION
As one of the most commuted goods, oil plays a crucial role in sustaining any nation's welfare, economic growth, and social modernisation. . Amin (2015) argues that the consequences of the oil price shocks became more prominent after the oil crisis of 1973. Since then, many researchers have analysed the effect of this oil price shock on the global and local economies and found that oil price fluctuation is associated with major world development and gives rise to economic inflation or recession (Hussin et al., 2012). Figure 1 shows how frequently international oil price fluctuates over time from 1980 to 2018.
Oil price shock can affect macroeconomic stability through demandside and supply-side channels (Amin and Marsiliani, 2015;Fueki et al., 2020). Bangladesh consumes a high volume of imported oil, mainly to support the transport and energy sector; hence, the effect of the fluctuation of oil price is very crucial and demands rigorous attention. The exchange rate, one of the essential macroeconomic indicators, can also affect the economy through different channels. The exchange rate and other macroeconomic indicators like inflation, interest rate, current account deficit, public debt, and trade terms can influence any country's macroeconomic settings. The exchange rate can play a vital role in those countries which are heavily dependent on international trade. Tokuo and Hayato (2016) argued that the shifts in the exchange rate regime in many countries in the 1970s caused a wave of empirical literature on the nexus between the exchange rate and macroeconomy. Figure 2 shows that the real exchange rate scenario of Bangladesh from 1980 to 2018.
international price of crude oil. The price of global oil was $42 per barrel in 2005, which increased up to $147 per barrel in July 2008 (Volkov and Yuhn, 2016).
Theoretically, higher oil price increases the cost of imported oil, puts pressure on the exchange rate, and leads to the depreciation of the local currency for the oil-importing country (Berument et al., 2010). If the oil price increases, consumers' purchasing power will decrease, causing a decrease in the demand for non-tradable goods. As the demand for non-tradable goods decreases, it reduces the country's price, which eventually depreciates the currency (Kin and Courage, 2014). These all imply that oil price acts as a crucial variable to determine the currency's strength and variation on currency price.
Many researchers have empirically discussed the linkage between oil price and exchange rates globally; however, it yields inconclusive results. As a transitional country, Bangladesh stiffly depends on oil for the development of the economy. So, it is imperative to know how Bangladesh's exchange rate is affected by oil price shocks. To our knowledge, the relationship between these two variables in Bangladesh's context has not been studied yet. So in our paper, we are going to address the following research questions: 1. Whether there is any long-run and short-run relationships between oil price and exchange rate in Bangladesh? 2. Is there any causality between these two variables? 3. How can oil prices affect the exchange rate of Bangladesh?
We run the Augmented Dickey Fuller (ADF) unit root test and the Dickey Fuller-Generalised Least Squares (DF-GLS) unit root test to check the variables' stationarity. Johansen cointegration and the Auto Regressive Distributed Lag (ARDL) Bounds cointegration tests are done to check the cointegrating relationship between the variables. The Granger causality and the Vector Error Correction Model (VECM) tests are also conducted to detect both the short-run and long-run causal relationship. The Dynamic Ordinary Least Squares (DOLS) and Dynamic Auto Regressive Distributed Lag (DARDL) estimation have been conducted to estimate the long-run coefficients of the variables. Finally, DARDL is also conducted to simulate the counterfactual effects of regressors on the response variables.
We reveal that variables are stationary at first differenced form and cointegrated in the long-run. We also find a long-run causal relationship from oil price to the real exchange rate in Bangladesh. However, in the short-run, no causal relationship is observed. According to DOLS and DARDL results, a 1% increase in oil price decreases the real exchange rate by 0.40 and 0.30%, respectively.
The rest of the paper is organised as follows. Section 2 reviews the literature review; Section 3 discusses the methodology and data. Section 4 highlights the results of the study. Finally, section 5 focuses on conclusion and policy recommendations.

LITERATURE REVIEW 1
Many studies have analysed the nexus between oil prices and real exchange rates around the world. This section aims to briefly go through the existing body of literature sequencing to country and region-specific studies. Djebbouri (2018) studies the effect of oil price shock on Algeria's exchange rate because it controls Algeria's economy. He finds that oil price shock harms the Algerian exchange rate in Algeria remarkably. Crude oil prices shock can explain 26.25% of the variation in Algerian Dinar. So, it is recommended to make variations in Algeria's economy through higher direct investment in prime sectors of the economy where oil is not produced.

Country-specific Studies
Trung and Vinh (2011) study the oil price and exchange rate relationship for Vietnam using monthly data from 1995 to 2009. They find a positive effect of oil price devaluation and economic activity. They recommend that by preventing real appreciation of Vietnam's currency, the government can support competitiveness and ensure economic growth. Lastly, they emphasise that it is helpful for the activities of the economy if inflation occurs modestly. Hussin et al. (2012) analyses the dynamic effects of changes in oil price and macroeconomic variables on Malaysia's Islamic stock market. They use monthly data from January 2007 to December 2011 and reveal that the exchange rate is positively affected by Islamic stock prices. Besides, Islamic stock returns also influence oil prices in Malaysia. Also, oil price also affects the Islamic stock price in the short and long-run, but not vice-versa. Kilian and Zhou (2019) find that the price of imported crude oil becomes lower if there is an external devaluation of U.S. currency 1 For methodological details, See  related to its main partners with whom it trades. They also find the shocks of external real exchange rate manage the real value of U.S. currency. Babatunde (2015) reveals that the effect of positive or negative oil price shock on the exchange rate is not symmetric in Nigeria's case. He recommends that structural transformation would save Nigeria's economy from the damage of higher external transmit of affluence during a long period of oil price shocks due to the vast importation of crude oil. Sharma et al. (2017) study the impact of oil price fluctuation in India's exchange rate. By collecting the daily time-series data from 16 th February 2015 to 1 st February 2018, they show a unidirectional relation between oil price and exchange rate in India's case. They recommend that India should include variety to its trade of oil to maintain a stable exchange rate.
Kin and Courage (2014) examine how oil prices can influence the nominal exchange rate of South Africa from 1994 to 2012. They find that a 1% increase in oil price is the reason for the 0.12% depreciation of the Rand exchange rate of South Africa. They argue that the reason behind the vulnerability of the South African economy to the oil price changes is their flexible exchange rate system. They recommend that the central bank can obtain a feasible Rand exchange rate by using the interest rate to reduce the exchange rate fluctuation. Baboli et al. (2018) explore how the sudden fluctuations in the exchange rate and oil price can affect Iran's inflation using annual data covering from 1991 to 2016. The main findings of their study assert that the speedy growth of inflation is taking place in Iran, and one of the most crucial reasons behind this condition is the exchange rate heavily depends on the foreign exchange earnings of oil exports. It is suggested that the government manage oil incomes soundly and prevent the swift change of exchange incomes, which aggravates inflation. Fratzscher et al. (2014) examine the nexus between the U.S. dollar and oil prices and find that there is a bidirectional relationship between the oil price and the U.S. dollar. They also argue that the equity market returns and risks also influence the oil prices and the U.S. dollar.

Panel Studies
Using monthly data from 1998 to 2012, Volkov and Yahn (2016) find that in Russia, Brazil, and Mexico, the frequent fluctuation in the exchange rate due to oil price shocks is significant. However, the shock effect is very insignificant in Norway and Canada. It takes more time for the exchange rate of Russia, Brazil & Mexico than of Norway and Canada to achieve the equilibrium which occurs due to shock in oil price.
Tokuo and Hayato (2016) conduct a multi-country analysis for Australia, Canada, Japan, Norway, and the United Kingdom. They show that the structural shocks related to oil price fluctuations are important in explaining Australia's currency (19%) and Japan's currency (35%), while it is relatively insignificant in explaining the fluctuation of the Canadian Dollar (4%) and U.K. pound (2.7%). Norway stands in the middle with 15%. Ahmed et al. (2017) examine how oil price shocks can influence the key macroeconomic variables for India, Pakistan, Bangladesh, Sri Lanka, and Bhutan. They find that GDP and real exchange rates are cointegrated and cause each other's movement in the longrun. They also reveal that fluctuations in oil prices influence the exchange rate and macroeconomy in five SAARC nations both in the short-run and the long-run.
Narayan (2013) examines whether oil prices can play any role in forecasting the exchange rate returns in 14 Asian countries with a different exchange rate regime. He shows that the Vietnamese Dong experiences future depreciation due to the increased oil price. On the other hand, the scenario is quite the opposite in Bangladesh, Cambodia, and Hong Kong; higher oil prices result in future appreciation in these regions.

METHODOLOGY AND DATA
Following Babatunde (2015) and Narayan (2013), we use the real exchange rate as the dependent variable and oil price as an independent variable. The real exchange is used to avoid the effect of any change in the general price level on the exchange rate. We consider the Consumer Price Index (CPI) as a control variable. Hacker et al. (2014) and Branson (1983) point out that a rise in CPI can appreciate the real exchange rate or vice-versa. An increase in CPI can also increase the interest rate and make the interest-bearing assets more attractive. So, the foreign exchange market of an economy starts to observe an increase in demand for the local currency.
The model's functional form can be expressed by equation (1) as a log-linear equation for the time "t." The natural logarithmic transformation of the variables is advantageous because it not only reduces the high level of skewness from the dataset but also expresses the coefficients as elasticity. It is worth mentioning that elasticity measurement is important for policy implications (Amin & Khan, 2020l;Hasanov et al., 2016).
Where, LNRER t = log of the real exchange rate, LNOP t = log of oil price, LNCPI t =log of CPI, and ε t = error term. Data on oil price, CPI, and REER range from 1980 to 2018 and collected from the British Petroleum and the World Development Indicators (WDI), respectively.

Unit Root Test
In the time series analysis, unit-roots can cause unpredictable results. From the graphical and empirical combination of the data, it can be shown that if the mean and variance are both changing in the same manner, there is less possibility of having unit root among variables. However, suppose both the variance and mean are not changing in the same manner. In that case, there is a high possibility of having a unit root among those variables, leading to spurious regression results. We have applied the traditional ADF test to check the variables' stationarity properties to avoid distorted results. Besides, for robustness checks, we have also employed the DF-GLS test.

Tests for Cointegration
Cointegration tests are designed to inspect non-stationary time series procedures, which essentially contain a mean and a variance that changes with passing time . This mechanism makes room for estimating the long-run parameters or the equilibriums of such systems that carry variables with unit roots. The Johansen test is a general multivariate concept of the ADF test. This particular generalisation mainly investigates the linear combinations of all the incorporated variables with unitroots. It has become feasible to evaluate all of the cointegrating vectors due to the Johansen test's presence and proper estimation strategy. If "n" number variables, all with unit roots, are present in the system, then there will be maximum n-1 numbers of cointegrating vectors will be found.
On the contrary, the presence of n number of variables and n number of cointegrating vectors implies that the variables do not hold unit-roots. The reason behind this fact is the cointegrating vectors' being able to be written as the scalar multiples of every single variable alone. The Johansen cointegration test is widely used to test cointegration. This test determines how many independent linear combinations are present in the time series variables set, which generates a stationary process. This test can give the rank of cointegration. For applying this approach, we need to estimate an Unrestricted Vector Autoregression (VAR) as follows.
∆x t = α+θ 1 ∆x t-1 + θ 2 ∆x t-2 +...+θ k-1 ∆x t-k+1 + θ k ∆x t-k + u t This equation ∆ denotes the difference operator, x is the symbol of an (n-1) number of vectors of non-stationary variables in levels, and u also represents the (n-1) number of vectors of errors that are randomly occurred. The matrix θ holds all the necessary detailed information that is essential to illustrate the relationship between the variables. If the rank of θ appears to be 0, then it can be inferred that the variables are not cointegrated. If rank, which is denoted by r, is 0, then it can be claimed that there is only one cointegrating vector. Lastly, when the scenario is "1<r<n" then it is confirmed that multiple numbers of cointegrating vectors are present. However, as Zhou (2001) mentioned that size distortion in the dataset could lead to spurious Trace, and Eigen tests statics of the Johansen procedure, we have also applied the ARDL Bounds test for cointegration based on surface regressions.

Granger Causality
The Granger causality test is used to check whether the time series variables provide meaningful insights (Amin and Hossain, 2017). Granger (1969Granger ( , 1980Granger ( , 1988  don't Granger cause each other, y causes x but not vice versa, x cause y but not y don't, and both x and y Granger cause each other.

Two sets of equation have been used for conducting this study
x t =α o +α 1 X t-1 + α 2 X t-2 +...+α i X t-i +β 1 Y t-1 +...+β i Y t-i +u t (3) In the case of all likelihood (x, y) series in the group, the F-statistics are the Wald statistics for the joint hypothesis, β 1 =β 2 =β 3 =·······=β 1 =0. Engle and Granger (1987) assert that VECM is an ideal approach to identify the variables' short-run and long-run magnitude. Suppose it is found that a set of variables have one or more cointegrating vectors. In that case, VECM will be a suitable estimation technique that adjusts the changes and variation from equilibrium (Khandker et al., 2018). Causality hypothesising in a multivariate framework can be estimated by the parameters of the following VECM equations: Here, Z˗1 is the error-correction term, which shows the variation of the variables from the long-run equilibrium condition. Error correction term avails Y and X's adjustment from the short-run towards their corresponding long-run equilibrium. Stock and Watson (1993) proposed the DOLS methods, a modified version of the OLS approach, to deal with a small sample size.

Dynamic OLS (DOLS)
It is a single equation method that is robust, and it also corrects the regressor internality by including lags and leads . As we have a relatively small sample size, we have applied the DOLS approach to avoid fake assessments. If Y t is the dependent variable with regressors X i , t i =1,2,3…, n then,

Dynamic Simulation from Dynamic ARDL (DARDL)
To incorporate the dynamic simulations in our study, we have employed the Dynamic ARDL as stated in Jordan & Phillips (2018). They proclaimed that the traditional ARDL, either ECM or any other forms, might sometimes appear difficult to interpret. Moreover, conventional ARDL fails to demonstrate the short-run, medium-run, and long-run dynamic changes .
Also, when the motive behind the model specification addresses several kinds of complications in the society, the problems associated with the traditional ARDL get severe. To obtain a satisfactory understanding of the fundamental scenario in a given situation, the dynamic simulation of the ARDL models can play a praiseworthy role than conventional VAR based IRFs. 2 variables are non-stationary at the level, however, at the first difference for both constant and constant and trend configurations. So, our variables are integrated of order one.

RESULTS AND DISCUSSIONS
The Johansen Cointegration test reveals a cointegrating relationship among the variables (Table 2). Both the Maximum Eigenvalue and the Trace test confirm that our variables have one cointegrating relationship. Table 3 shows the F-statistics is well above the upper bound critical values indicating a meaningful long-run cointegrating relationship among the variables. We then move on to the next step and run the Granger causality test to determine the long-run direction of the causality between the variables. Table 4 reveals a unidirectional causality from oil prices to the exchange rate in the long-run. No causal relationship is found in the longrun between the real exchange rate and CPI. We have also incorporated the VECM to determine the short-run direction of the causal relationship among interest variables. The VECM results can be depicted in 5. Table 5 gives a different kind of intuition from the long-run causal direction regarding the real exchange rate and oil price linkage. The VECM results assert no short-run causal relationship between oil price and exchange rate. One of the prime reasons in this regard can be that the effect of energy-related fluctuations 3 may not immediately be seen in the macroeconomic variables due to the time-lag effect . In    To understand the magnitude (i.e., elasticity) of the effect of oil price and CPI on the real exchange rate, we have applied DOLS and DARDL long-run estimation techniques ( Table 6). The DOLS estimation result shows that if the oil price increases by 1%, the real exchange rate will decrease by 0.40%. In DARDL, we also get a negative and significant coefficient, but with a higher significance level (5%), which is -0.30. It indicates that if the oil price rises by 1%, the exchange rate will drop by 0.30%. However, the coefficient of CPI is not significant at all. So, we highlight that any CPI changes will not change the real exchange rate significantly in Bangladesh.

The result reported in
The long-run estimation result is consistent with Narayan (2013). As Bangladesh is an oil importing country, in the long-run, any increase in the oil price decreases (appreciates) the real exchange rate. 5 We argue that when there is an upsurge in the international oil price, Bangladesh's current account 6 observes a large distortion as net export earnings decreases from the stable (anticipated) value. The net export earning decreases due to higher spending (such as providing huge import subsidy) for importing fossil fuel. Therefore, the real exchange rate needs to appreciate to reduce foreign reserve outflow and improve the non-oil trading scenario.
As DARDL allows us to simulate the counterfactual effects of regressors on the response variable, we have conducted a simulation analysis where counterfactual shocks are given in the oil price to observe the real exchange rate's behaviour. Figure 3 shows that although a negative 1 standard deviation shock increases (depreciates) the real exchange rate, a positive 1 standard deviation shock decreases (appreciates) the real exchange rate. The effects from both of the shocks tend to remain till t=10. From t=11, the effects start to stabilise in the economy.
Novel CUSUM test results of each variable used in the proposed model of the analysis can be seen from Figure 4. The CUSUM test 4 An exchange rate determination approach, where a legislative institution fixes the rate around a certain value. However, the approach still allows small fluctuations, usually within determined ranges, resulting small flexibility in the currency exchange market. 5 Real exchange is calculated as ( ).

PL NER PL
Therefore, negative sign in the regression refers to the appreciation of Taka against USD. For more details, please see Narayan (2013); Javaid (2011); Edwards (1989); World Bank (2019)  6 Current account=Net earnings+Net investment+Cash transfer results of each variable show that the plots stay within the boundary of 5 percent critical value, indicating that the model is stable both in terms of systematic and sudden movements. The estimated

CONCLUSION
Since the mid-1950s, oil has been started to be considered as the essential source of energy, and it has become a stimulus for the rapid growth of the industrialised nations across the whole world. The frequent fluctuation of oil prices can affect the economy through various macroeconomic channels, and arguably, the exchange rate is the most crucial of all those channels. There is a growing interest in finding out how the oil price can affect the exchange rate and, ultimately, the whole economy.
The motive behind conducting this study has been to explore the linkage between oil prices and the exchange rate in Bangladesh.
In this regard, we have considered the time series annual data for 1980 to 2018 and applied recent robust econometric techniques. Upon analysing empirical findings, we argue that an upward trend in the oil price leads to a decrease (appreciation) in Bangladesh's exchange rate in the long-run. On the other hand, oil price has no impact on the real exchange rate in the short-run.
We recommend that the government should adopt optimal policies to minimise oil price shocks' adverse effects to achieve the Sustainable Development Goals (SDGs) in Bangladesh. More focus should be given on minimising the negative consequences of oil price fluctuation towards the macroeconomic variables like the exchange rate. The Bangladesh Bank should closely monitor the oil price so that if there is any positive shock in the oil price, necessary steps can be taken to dissolve the adverse effect on the exchange rate quickly.
Amidst the devastating outbreak of Covid-19, it is forecasted that there is no probability of experiencing any upward movement in the oil price till 2024 (OECD, 2020). 7 Moreover, the oil price has dropped below the average level of the last 5 years due to worldwide lockdown to reduce the spread of COVID-19. Therefore, a portion of the money that is allocated for importing oil will be saved for the next few years. The saved money can be invested in other priority sectors of Bangladesh's economy like health, education, and, most importantly, the social security to improve the living standards as well as achieve overall stability during this pandemic 8 .
We like to extend the analysis by including more control variables in this study and conduct a sector-specific analysis for a more robust result. A region-wise comparison can also be made by employing a regression for other South Asian countries.

ACKNOWLEDGEMENT
We want to thank the editor and the anonymous referees for their valuable comments to improve the paper's quality. The corresponding author also acknowledges the research support from his research assistants, Farhan Khan, and Foqoruddin Al Kabir.