The Impact of Oil Prices on Income in Azerbaijan

Oil prices in the global market have been fluctuating sharply since the end of the XX century. As oil is a predominant component that affects global indicators, it is necessary to assess by economists. Because GDP per capita, exchange rate, and total trade turnover depend on the effects of oil price shocks. This article researched the impact oil fluctuations on income from a short and long-term perspective. For the purpose of the research, Azerbaijan has been exemplified as an oil-producing country for the period of 2001m01-2021m06. Applying the empirical method, we achieved that world oil shocks affected income (in dollar) significantly in a short perspective. The recent declining oil prices make this topic more valuable to research. Moreover, as many empirical methods show asymmetric dependency, the research focused mainly on negative shocks of oil prices.


INTRODUCTION
The historical records of the world oil prices reveal that the USA power as an oil producer (the ability to influence prices, volume and supply of oil) has weakened since 1960 (ECB, 2010). Simultaneously, OPEC embarked on influencing the market and as a result, it caused the oil price shock in 1973 and 1979. After 1970, crude oil has been the driving force for many developed countries (Berg and Gröttheim, 1992). Although between 1971-2014 total primary energy supply grew on average 1.0% yearly (Gasimzade, 2015). Blasting of oil price twice in 1973-1974 and thrice in 1979-1980 led to macroeconomic changes and emerged new notions such as "expenditure inflation" and "stagflation." In other words, the surge in oil prices caused inflation and unemployment (Hamilton, 1983;Bruno and Saks, 1985;Blanchard and Gali, 2007).
Although there is wide range of researches dedicated to the development of alternative energy sources, the world economy is still significantly depending on the crude oil and its derivative products (Imanov and Hasanli, 2014). Researches revealed that there was a general consensus over the importance of oil price on the socio-economic development and well-being of the country. In general, oil prices are used to predict unemployment and shortterm progress. However, there had no influence of oil price on the well-being and income of the people (Stiglitz et al., 2010). The reason why we accept this indicator as a significant factor is to have the power to influence all people. Secondly, it is flexible and important for businesses and governments in the short and medium term. Thirdly, analysts forecast stronger shocks in the future thanks to the advancement of technology and geological research (Beneş et al., 2012). It is firmly believed that the surge of oil price leads to the reduction of oil-importers' income in a short term . Contrary, it has positive effects on export and state revenue of the oil-producing countries. Although it is sometimes hard to forecast whether importers or exporters suffer a lot, arguably, oil importers are affected more quickly.
The world oil prices decreased because of several reasons: the declining pace of the world economy, slow progress in Europe and Japan, the increase of oil production in non-OPEC countries. OPEC announced to cut 5 million barrels of oil daily in order to stop oil prices to go down. It was somewhat useful. Since 2009, prices jumped to recover and resisted 75 dollars per barrel. The Economist reportedly revealed that there had been a rising interest to dig out this topic comprehensively (Broadstock and Zhang, 2014). Although, between 1971 and 2014 total primary energy supply grew on average 1.0% yearly (Gasimzade, 2015). The prices constantly increased from 2010 to 2014. However, this growth was related to the volume of demand and changes in crude oil prices. During 4 years, the oil price has started to decline since 2014 (from 105 dollars) and remained flat at the low level for a long period. This has been the third time that the prices declined during 35 years. This is a significant but unexpected case.
Simultaneously, there were some significant decisions taken: increasing oil production by non-OPEC countries, refusing to stick to target prices, and the period of a surge of oil production in 1985-1986.
The oil price has diminished sharply since the financial crisis in 2008. There are several factors caused the reduction of oil price in 2014-2016: • The unexpected growth of oil production for several years • The reduction in demand • The significant change in OPEC policy • The weakening of some geopolitical risks • The strengthening of the US dollar exchange rate. In fact, OPEC refused price support and increased oil production by unconventional sources. It gradually affected the prices and has played a crucial role since 2014. However, the driving force for price drop was generated by supply. The researchers are not enough to explain whether the factors impacted on dropping oil prices in 2008 and in 2014-2016 were the same.
The falling of oil price will negatively affect oil-exporting countries while the world economy is intended to improve in the middle term Mikayilov et al., 2020;Kopytin et al., 2021). Because reduction in revenues influences budget and the exchange rate will slow down as a result of weak economic development. The changes of oil prices strengthen the changes in the financial and currency market (Mukhtarov et al., 2017;Musayev and Aliyev, 2017;Muradov et al., 2019). Furthermore, investment in the oil industry not only decreases in oil-exporting countries but also diminishes in oil-importing countries that possess great potential for oil production.
Concomitantly, the price of agricultural production will parallelly drop as oil prices fall (Paladines Amaiquema, 2015). As a result, people benefit from the low prices of raw materials. Contrary, the rocket of oil prices under equal terms cause prices to increase. Also, oil shocks have different impacts on oil exporters and importers: Oil shocks influence to oil exporters faster than oil importers.

LITERATURE REVIEW
Having reviewed short analysis of references, macroeconomic activity has always been the main indicator in all oil-related publishings (Mork et al., 1994;Jimenez-Rodriguez and Sanchez, 2003;Ozturk, 2010;Elder and Serletis, 2010;Jiménez-Rodríguez and Sánchez, 2005;Sonmez, 2016;Herrera et al., 2015;Byström, 2020). Mork et al. (1994) researched GDP growth and macroeconomic outcomes in OECD countries as a result of the drop-in oil prices. The results change from country to country. However, researches show that the change of oil prices heavily affects oil-dependent countries. Although there are plenty of publications referring to economic activity, there is almost no research on the effect of it on income and consumption of people. Some economists Mehra and Peterson (2005), Blanchard and Gali (2010) and Kilian (2008) researched the impact of oil fluctuations on consumption. However, each holds different views on it. For example, using the "lifecycle" consumption model popularized by Mehra and Peterson (2005) Modigliani and Brumberq (1954) empirically researched the impact of oil price on consumption in the USA. Zhang and Brodstock (2014) referring to Mehra and Peterson (2005) enlarged their researches for some ASEAN and East Asian countries.
However, Blanchard and Gali (2010) and Kilian (2008) prefer different approaches. They design a model that oil can be used as a raw material for production and the rest for consumption. They empirically indicate that oil price fluctuation influences people through food price. Because importing oil is expensive and the food price depends on it.

Developed countries
Research on the well-being of the people is mainly focused on oilimporting countries. As for the USA, Staniford (2007) researched average well-being indicators based on the real income generated from oil fluctuations in 1976-2006. According to his researches, there is no relation between social well-being and the growth of real income. The reason behind it is that the USA economy is so big that the dependence of the economy and people's income on oil prices cannot be noticed. Besides, Graham and Chattopadhyay (2010:M01-2009:M12) used daily petrol prices and the well-being of the people for the USA. They revealed that while the petrol prices increased in the third quarter of 2008, there was a negative dependence between gas price and the well-being. However, it has turned to the positive later. If we take the whole picture, there is a negative dependence at the end (there have been some differences between different income groups). Graham and Chattopadhyay (2010) reveal that oil price has an enormous impact on the well-being. Indeed, while the petrol prices increase 1 dollar, the well-being will parallelly increases 530 dollars in order to stabilize it. The extensive research on the relationship between well-being and oil prices belonged to Boyd-Swan and Herbst (2012). This research was devoted to the changes between the standard of living and the petrol prices in the USA in . Surprisingly, the impact of high oil prices on the standard of living was stronger in less populated areas. But this is insignificant for those who live in the city. Besides, the negative impact of higher oil prices is short. Thus, the first negative impact of oil prices is compensated with the upcoming positive impact of oil prices. Later Edelstein and Kilian (2009), using VAR model, indicated that energy prices have always been the main indicator for the increase of real consumption yet are not a dominant factor.

Developing countries
The world oil prices and their impact on economic development have been researched in many articles but its impact on consumption has just commenced. Mehra and Petersen (2005) are the first researchers on this topic. They used similar methods as Broadstock and Zhang (2014) and researched economy of some Asian countries. They found some empiric evidence to have nonlinear asymmetric correlation between oil and the consumption prices. The leading economist of OECD Wang (2013) came to the same conclusion. Unlike previous authors, he used the logistics model in order to better research the relationships (Muhammad et al., 2017). Salim and Rafiq (2013) Andersen et al. (2004) using VAR methods researched the impact of oil fluctuations in six main Asian developing countries (China, India, Indonesia, Malaysia, Philippines and Thailand) (Rosnawintang et al., 2021 (ASEAN -5 Countries); Anuar et al., 2021 (ASEAN +3 Countries)). Oil fluctuations influence on production in a short term in China and Thailand while it influences on GDP in India, Philippines and Malaysia. By the support of Thailand Oil Fund, the negative impact of oil prices on macroeconomic indicators has weakened and affected economic indicators positively. Bouzid (2012) researched cause and effect relationships while studying oil prices and economic development in Tunisia in 1960-2009. Tunisia is not oil-producing country but importing one. Obviously, the rise of oil prices lowers economic development. The rise of oil prices negatively affected daily consumption. According to the results of the researches, Grencer's test -used to determine the relations between energy prices and economic development -revealed the relations between the real price of GDP and oil (Syaharuddin et al., 2021). Masih et al. (2010) researched the impact of oil price fluctuations on shares using VEC model including interest rate, economic activity, real share income, oil price in South Korea.

The Impact on Exporting Countries
The previous researches were limited by oil importing countries. The impact of oil prices on well-being in oil exporting countries seems complex, because positive influence can be compensated by rising exchange rate and the unsuitable side effects of rising oil prices (Corden and Neary, 1982;Corden, 1984;Frankel, 2010;Chauvet et al., 2012;Al Sabah, 2019). Although there were few macroeconomic researches in oil-exporting countries, the existing one reflects high variations from one country to another one. For example, Jimenez-Rodriguez and Sanchez (2004), using general model for oil-exporting and importing countries, revealed that in spite of the fact that oil prices had a negative impact in almost all oil-importing countries in a short term, the situation differs from country to country. Noteworthy to note that there is a positive relation with the change of GDP per person and standard of living Mohamad and Mohammad, 2021). The rise of unemployment impacts badly on people's standard of living (Di Tella et al., 2001;2003) . These researches were done in the example of developing countries (Jin, 2008;Graham, 2009;Easterlin, 2010) and those of having transition economy (Easterlin, 2010). However, the impact of oil price on people's income has been less researched. Some existing researches were dedicated to some countries (for example, Tiliouine et al., 2006). Some researches were done by using VAR method in 1980. However, the surprising part is not having enough researches dedicated to the second biggest oil exporters. Ito (2012 researched the impact of oil prices on macroeconomic indicators in Russia using VAR model. The research was done around for 15 years (1994: Q1-2009: Q3) and 63 observations. He found that not only did the rise of oil prices impact on GDP and exchange rate but also it influenced on the inflation rate.
Nigerian case was assessed by Apere and Ijeoma (2013) using exponential general autoregressive conditional heteroskedastic (EGARCH), impulse response function, and lag-augmented VAR (LA-VAR) models. Besides, Oriakhi and Osaze (2013) researched the relations between oil prices and the economic development of Nigeria based on the quarterly data in 1970-2010. They claim that oil price fluctuations impact real state expenditure, currency, and import. Thus, oil price fluctuation defines state expenditures. Englama et al. (2010) and Cini (2008)  Another economist Nusair (2016) researched short and long-term dependencies of oil prices on GDP in the Gulf countries using NARDL. The rise of world oil prices doesn't cause economic development. However, there has been a mutual relation between low prices and economic development in terms of significantly causing economic stagnation. As for Kuwait and Saudi Arabia, they revealed that low prices of oil didn't affect economic development significantly. The low prices of oil only lowered GDP in Kuwait and Qatar. To conclude, oil fluctuations had an effect on real GDP in the Gulf countries. Besides, Moshiri and Banihashem (2012), using VAR model researched the impact of oil fluctuation on the economic development in Kuwait and Saudi Arabia and six other OPEC countries in 1979-2009. Alkhathlan (2013 concluded with the positive and significant impact of oil revenues on GDP in a short and long term in 1970-2010. He found out that positive oil prices increased GDP growth and real oil shocks decreased real GDP. Emami and Adibpour (2012), using SAVR model, researched the relations between production, state expenditure, oil price shocks based on annual data in Iran. They think that production, state expenditure, the amount of money, and positive oil shocks influences positively on the economic development of Iran. Mehrara (2008) analysed the relations between oil prices and production growth in 13 oil exporting countries and a applied dynamic system of groups for calculating oil shock results. He revealed that negative shocks affect badly for a long term. Simultaneously, oil revenues generated from negative shocks are twice more than positive shocks. Another research analysed state budget determinants of oil exporting countries using constant and random selection methods in 1990-2015.

Data Descriptions
The research relies on time-series data of the World oil prices (WOP), Population income in manats (PIM), and Population income in dollars (PID) (2001m01-2021m06) ( Table 1). The data were generated from the National Bank of Azerbaijan. The level changes were given in the statistics ( Table 2).

Unit Root Test
It is quite important to test the stationary of variables through the Unit Root Test before assessing regression equations. Many methods require variables to be non-stationarity in order to have long-term or cointegration relations. Notably, if any time series are stationary with the real values of the variables, it is regarded as I(0). If the variable is not I(0), the first difference is calculated and the stationary is tested. If it was stationary, the variable would be regarded as I(1). In order to make the results of the stationarity test more reliable, three different Unit Root tests will be done and the analysis will be conducted with and without trends. The Unit Root tests are these: Augmented Dickey-Fuller, (ADF), (Dickey and Fuller, 1981), Phillips-Perron (PP), and Kwiatkowski-Phillips-Schmidt-Shin (KPSS).

Auto Regressive Distributed Lags Bounds Testing (ARDLBT)
It is worth to mention that ARDL method has more advanatages than other cointegration methods and it yields reliable results for small samples. The ARDL test models can be used as: L -is a logarithm function, ψ 0 -constant value, ε t -white noise error, PID-Population income in dollars, PIM-Population income in manats, WOP -World oil prices -barrel/dollars. ψ 1i , ψ 2i -shortterm coefficients, λ 1 , λ 2 -long term coefficients.
In order to assess short-term relations among variables, ARDL error correction model ( The ARDL bounds-testing cointegration method applies the Wald test (F -stat) on λ i to show the existence of long-term cointegration between selected variables. The null hypothesis called H 0 : λ 1 =λ 2 =0 is tested for the absence of cointegration (H 1 : λ 1 ≠λ 2 ≠0). However, the alternative hypothesis is having cointegration relations among variables. Pesaran et al. (2001) proposed two types of boundaries based on F-statistics (i.e., upper bound and lower bound).
If the estimated value of the F-criterion is below the lower bound, then there is no significant long-term relationship between the variables. In addition, if the predicted value of the F -criterion is higher than the upper limit, then there is a long-term relationship between the variables. However, if the calculated statistics of the F-test are within the limits, the results are uncertain.

Engel-Granger Co-Integration Test
The Engel-Granger (EG) cointegration test creates an opportunity not only to verify the existence of a long-term relationship, but also to determine the direction of these relationships between the variables, as well as to investigate the short-term relationship.
In the first stage of the assessment process through the EG cointegration test, the regression equation is evaluated for nonstationary variables in the original case, but are stationary if differentiated to the same level (usually I (1)). So, for the two variables in our example: LPIM t =ψ 0 +λ 1 LWOP t +ε t Here, ψ 0 and λ 1 represent the regression coefficients, LPID, LPIM, and LWOP -the dependent and independent variables, ε -the white noise error, and t -the time. After evaluating the regression equation, the next step is to check the stationarity of the white noise error. If ε t is stationary, there is a cointegration relationship  between these variables and it is not spurious. That is why these equations (5-6) are regarded as long-term equations. Finally, the ECM is assessed using the cause-and-effect relationship between the variables, in other words, stationary variables to check the strength and direction of the dependency and white noise error for a periodic delay (ECT t-1 ) (equation 3-4).
Here ψ 0 , ψ 1i , ψ 2i , φ 1 and φ 2 represent the coefficients, p is the optimal delay scale, and ε is the white noise error of the model.
To determine the optimal delay scale, the relationship between the variables is first evaluated in the Vector Autoregressive (VAR) model. Then, the equations are evaluated using the Least Squares (LS) method, taking into account the optimal delay scale. Engle and Granger reveal that if there is cointegration between variables, the evaluation would be done through ECM. If the cointegration relationship is stable, the coefficients of Error Correction Term (ECT), or ECT t-1 , φ 1 and φ 2 must be negative and statistically significant. In other words, this ratio is called the Adjustment Speed and is usually rated between "-1" and "0." Using equations 3-4, the following cause-and-effect relationships can be tested: The Granger cause-and-effect relationship for the short-run period is evaluated using F-statistical or X i -quadratic statistical values together with the coefficients of all delayed first-order differences (∆LWOP t-1 ) for each variable (H 0 :ψ 2i =0, i=1…p) in order to check the statisical significance of the coefficients. The rejection of : indicates that LWOP has an effect on LPID and LPIM in the short term.
In order to check the Granger cause-and-effect relationship for the long run, the statistical significance of ECT t-1 coefficient is tested.
It is important to test it: (H 0 : φ 1 =0 və φ 2 =0). If H 0 : is rejected, the long-run period indicates that the deviations from the equilibrium have an effect on the dependent variable and will return to the equilibrium state over time. A strong cause-and-effect relationship should be tested simultaneously both in a short-term and a longterm period. In other words, hypothesis (H 0 : ψ 2i =φ 1 =0, i=1…p,; H 0 : ψ 2i =φ 2 =0, i=1…p,) should be tested using the F-statistical or X i -square statistical values by the Wald test.

FMOLS DOLS and CCR
Other evaluation methods used -FMOLS, DOLS, and CCR -and analysis of the results of Engle-Granger analysis are very useful in the research process. Because reviewing the results several times through the ARDLBT co-integration approach allows for a more reliable analysis. Note that the Engle-Granger and Philips-Ouliaris cointegration tests were used to test for all regression equations evaluated using FMOLS, DOLS, and CCR.

Diagnostics
This study checks the stability of ARDL model through sequential correlation test (Breusch-Godfrey LM test (H 0 : "no serial correlation")), heteroscedasticity test for both Breusch-Pagan-Godfrey (H 0 "no heteroskedasticity problem"), and Autoregressive Conditional Heteroscedasticity test (ARCH) as well as Ramsey RESET Test (statistical). In all cases, it is desirable not to reject the null hypothesis. The Jarque-Bera test will be used to check the normal distribution of white noise error. The tested H 0 : is a normal distribution in the white noise error." The CUSUM and CUSUMSQ tests are also used to research the stability of the ARDL model.

Descriptive Statistics of the Variables and Correlation Analysis
Before moving to empirical evaluation, descriptive statistics and correlation analysis should be taken. The results of the descriptive statistics are given in

Unit Root Tests Results
According to the ADF and PP tests, there is no variable I (0) in the "With Intercept only," "With Intercept & Trend" models and "No Intercept & No Trend" models. All variables are I (1). According to the KPSS test, all variables in the "With Intercept only" and "With Intercept & Trend" models are I (0). The ARDL and the ARDL boundary-test methods can be used to estimate short-term and longterm dependencies between variables which enable the assumption of single root test assessment results in ADF, PP, and KPSS (Table 4).

VAR Lag Order Selection Criteria
In order to determine optimal lag for ARDL model, VAR Lag Order Selection Criteria was employed and we got the below-mentioned results. The models selection criterion used is AIC. The results of models selection criteria are reported in (Tables 5 and 6).

Cointegration Testing Results
The results of the ARDL boundary test are given in Table 7.
In both ARDL-related equations (F LPIM =(LPIM⁄LWOP) və F LPID =(LPID⁄LWOP), F test result indicates the existence of cointegration between the variables.  oil prices (WOP) have a positive and significant impact on both Population income-manats (PIM) and Population income-dollars (PID) in the long run. Thus, a 1% increase in world oil prices (WOP) increases Population income-manats (PIM) by 1.38% and Population income-dollars (PID) by 1.59%.

ARDL Bounds Test, Long Run and Short Run Results
However, the situation is completely different in the short term. Thus, World oil prices (WOP) have a negative and significant impact on both Population income-manats (PIM) and Population income-dollars (PID). Table 9 presents the results of diagnostic tests of the ARDL model. The evaluation results of the Breusha -Godfrey (BG) method confirmed that our ARDL model had no problems with sequential correlation. The results of the Breusha-Pagan-Godfrey (BFG) method confirmed that heteroscedasticity was not a problem. The results of the ARCH method indicate a heteroscedastic problem. Normality Test (Jarque -Bera) JB is not desirable. In Model 1, the graph of the total amount of recursive balances (CUSUM) is unstable and the graph of the squares of recursive balances (CUSUM) is fixed. In Model 2, the graph of the total amount of recursive balances (CUSUM) and the graph of the squares of recursive balances (CUSUM) are fixed.

Engle-Granger Analysis Results
Another feature that indicates a cointegration relationship between the variables is that the white noise errors obtained from the estimates are stationary. Table 9 shows the results of the stationary test by applying single root tests ADF on the white noise error of each long-run equation evaluated by FMOLS, DOLS and CCR (Table 10). In general, white noise errors are stationary, but it appears in the first 2 equations. Based on these results, the fact that white noise errors are stationary in all models and thus the existence of a cointegration relationship is once again confirmed. However, this result does not support the results of the Engle-Granger and Phillips-Ouliaris cointegration tests given above (Table 11).

DISCUSSION
As an important component of the production, oil supply and oil price influence GDP per person, inflation rate, exchange rate and macroeconomic indicators (Chai et al., 2016). The world oil fluctuations influence energy consumption through exchange rate and GDP (Chai et al., 2016). We also researched the impact of oil fluctuation on GDP per person. We focused on oil price, exchange rate and gross trade volume. The negative and positive impact of crude oil price fluctuation on its profitability also proved for Azerbaijan (Jiang et al., 2018). The dramatic reduction in oil prices paved the way to prepare for the new period -post-oil period and led to reforms (Hesami et al., 2020). Azerbaijan commenced reforms too.
Researches revealed that low oil prices influenced GDP per person, gross trade volume and exchange rate. As an oil exporting country, Azerbaijan should have a diverse economy in order not to face difficulties later. Researches also showed that oil revenues decreased as a result of low oil prices and our conclusion overlaps with the same results as other articles: "...sheds light on the vulnerability of oil-producing regions to the oil price volatility. Gross domestic product (GDP) and government revenues in many Gulf countries exhibit a strong dependence on oil, while more diversified economies improve resilience to oil price shocks" (Vandyck et al., 2018).

CONCLUSION AND POLICY IMPLICATIONS
Policy and reforms on economics have been prepared taking the opportunities of every country. Although these problems are similar, territory, demographics, wealth, economic situation differs in every country. That is why, recommendations should be like this: The outcomes of the research are important in terms of helping optimisation of GDP per person and insurance against oil fluctuation. The research is one of the first researches to reveal mutual relations of GDP per person and oil price in oil-exporting countries using advanced statistics methods. It assists to understand the impact of oil prices on GDP per person in oil-exporting countries for a long and short term. This article stresses the importance of removing the dependence of economy from oil prices and reveals the benefits of economic policy and sustainable development in Azerbaijan and similar states.
The outcomes of the research overalps with targets hypothesis which explains the impact of oil prices on GDP per person. So, we can conclude that the changes in GDP per person better explain the impact of oil prices and why it is a strong indicator.
We can suggest the following recommendations in order to eliminate the dependence of GDP per person from oil prices: • Since oil is a non-renewable resource, the more the oil and gas industry decreases, the more other fields should develop. Although some countries have large reserves, our carbohydrogen reserves might deplete. However, financing the non-oil sector depends on oil reserves. That is why, development of non-oil sector to provide sustainable development and employment should be a priority. Besides, Azerbaijan should preserve oil reserves for future generations • The excessive dependence on oil might exacerbate macroeconomic uncertainty. Low oil prices might require less state expenditure and this might lead to delay the develeopment of the non-oil sector, employment and the volume of GDP per person. • Diversification of economics can create an opportunity to provide new workplaces and sustainable development for future generations. It can extend the base for GDP per person and resistance against low oil prices • Regulation of tax-budget, money and credit policy, exchange rate can be a strong base for diversification of economy and removing the impact of oil fluctuation on GDP per person and national economy • Policy and strategies motivating dynamic trade sectors can accelerate economic activity and increase non-oil GDP. Azerbaijan's strategic road map will lure the production industry through horizontal and vertical diversification and form new sources of increasing GDP per person and the non-oil sector. ADF denotes the Augmented Dickey-Fuller single root system respectively. ***, ** and *indicate rejection of the null hypotheses at the 1%, 5% and 10% significance levels respectively.