Bachelor Thesis
Do crude oil price fluctuations have an
effect on the exchange rate changes of the
Norwegian krone versus the Euro?
Investigating the effect and the potential improvement in
predictive capability of the exchange rate
Thao Hanh Le 10621407 University of Amsterdam Bachelor of Economics & Business, Specialization: Finance & Organization Supervisor: Dr. Philippe J.P.M. Versijp 31st May 2016Statement of Originality This document is written by Thao Hanh Le, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of contents
Abstract ... 3 1. Introduction ... 4 2. Literature review ... 8 3. Methodology ... 14 4. Data sample ... 18 5. Expected results ... 21 6. Empirical results ... 22 7. Conclusion ... 27 Appendix ... 29 References ... 40Abstract
The paper aims to investigate whether the oil price fluctuations have an influence on the movements of the exchange rate of the Norwegian krone (NOK) versus the Euro (EUR) and furthermore, whether the effect if it exists would be able to improve the predictability of the exchange rate. The paper model is built on the simple linear model used by Ferraro et al. (2015) to analyze the effect of commodity prices on the exchange rates of several currencies. In this analysis, Ferraro et al. ’s model is supplemented with other exchange rate prediction theories with the view to coming up with the least-biased results. The study found a significant effect of oil price variations on the NOK/EUR in all frequencies (daily, monthly, and quarterly) but the effect tends to increase in magnitude in the long run, which was not anticipated based on the empirical literature review. Additionally, the follow-up F-statistics test also shows significant improvement with the inclusion of oil price effect in the predictability model.
1. Introduction
Exchange rate predictability has always been an ongoing topic due to the critical impact of foreign exchange market on the daily basis economic activities whereas the predictive capability of the established and implemented models is still questionable. The standard models of exchange rate forecasting to be mentioned are the monetary models i.e. sticky-price models of Dornbusch and Frankel (1980) and the classic 1970s models i.e. the purchasing power parity and the real interest differential model. Despite being applied widely in the monetary and financial policies, these models in fact have not empirically proved their accuracy and consistency. Taking the random walk model as the benchmark, Cheung, Chinn, and Pascual (2005) conducted a systematic empirical examination of the predictability of exchange rate models and compared the performance of five popular and implementable exchange rate models in forecasting US dollar-based rates of several currencies. They could not find an exclusively successful performance of any model
since the forecasting performance turns out to be largely currency or horizon specific (2005, pp. 1164-1165). For instance, the paper stated that the purchasing power parity model was the best at predicting the movement direction of the exchange rate yet the predictability yields relatively high rate with the dollar-yen exchange rate but not with the dollar-pound rate (2005, p.1164). Additionally, the performance of the models also differs significantly with forecast series of one quarter ahead compared to four or twenty quarters ahead (2005, p. 1165).
Moreover, Cheung et al.’s analysis of the comparison of the above-mentioned structural models to the random walk model results in a superior outperformance of the random walk (2005, p.1161). This finding is supportive by many academic studies, from the milestone work of Meese and Rogoff in 1983 to recent work of Kilian and Taylor in 2003, which have confirmed the “hard-to-beat” trait of the naïve random walk prediction. On the other hand, Cheung et al. (2005) pointed out that most of exchange rate performance evaluations are based on one single approach instead of the combination of different models (p.1151). Inspired by this idea, this paper will combine different theories and ground models to build up the forecasting models with high predictability. Motivated by the classical and empirical theoretical frameworks, the factors determining the exchange rate movements to be considered are interest rates differential, inflation rates gap, current account deficit, national income and terms of trade.
In addition to the so-called classic elements of the predictive model of the exchange rate, commodity prices are one determinant among those under investigation. In the
World Economic Outlook: Adjusting to lower commodity prices in October 2015, the
International Monetary Fund (IMF) gave priority to evaluate the effect of the recent plunge of commodity prices on different aspects of the economy including the exchange rates. In the foreword, IMF pointed out that the effect comes not only directly through the devaluation of commodity exporters’ currencies but also indirectly through the decline in the financial inflows to the emerging markets, which results in a general depreciation against the US dollar, euro and Japanese yen (2015, p. xiv). Especially in case of countries with floating exchange rate regimes,
the drop in commodity prices has been reflected to a considerable extent in the movements of currency values (2015, p. 7).
This paper narrows down to investigate the effect of crude oil price, one of the most important commodities in the market with an enormous influence on the world economy, including the exchange rate fluctuations. The main purpose of the research does not limit to find out if this effect does exist or not but aims for figuring out whether the effect can improve the predictive capability of the exchange rate model. The improvement if found can be considered to be applied in monetary models ad policies with the hope to reduce the “random walk” characteristic of the exchange rates.
Beside, Norway is chosen as the country of interest driven by two main factors. Firstly, Norway is one of the largest crude oil exporters in the world despite the volume decline in recent years, still ranking 11th in the global oil exporters with
more than 1.2 million barrels per day (OPEC Annual Statistic Bulletin 2015). Crude oil exporting plays an important role in the country's trade flows maintaining a share of more than 25% in the total national exports value in last 15 years (Statbank Norway). Consequently, any oil price fluctuations affect significantly the national economy and financial markets. Moreover, the market share that Norway accounts for is not large enough to be the price-setter for the commodity. This is a critical point to avoid the occurrence of endogeneity in the research. In case of applying for major market players in oil exporting such as Saudi Arabia or Russia, it is more problematic to analyze the real effect of oil price on the exchange rate since the fluctuations in the value of Russian ruble or Saudi riyal potentially leads to an impact in the market oil price as well. Furthermore, since 1992 when the unilateral peg to the US Dollar and the Euro was abandoned, Norway has been pursuing the free-floating exchange rate regime with the target of retaining the inflation rate at 2.50% over years. Historically, Norges Bank also showed reluctance to intervene the foreign exchange market, as there were no interventions from the central bank since January 1999 (Norges Bank). From this respect, the Norwegian krone value can be considered to fluctuate not under the influence of monetary intrusions but in line with market determinants changes.
The main model applied in this paper is built upon the model used in the research of Ferraro, Rogoff and Rossi (2015) on the similar topic focusing on the Canadian dollar versus the US dollar. However, one potential counterargument for the study is that the effect found comes from the dependence on the value of US dollar as the oil price is also denominated in US dollar. This paper’s result will broaden the validity of the model as the value of the home currency is held against Euro instead of US dollar whereas oil price is still denominated in US dollar. In addition, the model used by Ferraro et al., is a very simple linear model, therefore, with the purpose to provide least-biased answer, the research’s models will include other factors supported by different theoretical frameworks to have influence on the exchange rate movements. The analysis will be conducted for the time period since Jan 1999, when the euro was officially launched to the most recent data available date. In case of effect found, a follow-up test will be implemented to conclude whether including the oil price effect can actually improve the predictive capability on the exchange rate variations. All in all, the research aims to provide a reliable answer to the question: Do crude oil price fluctuations have an effect on the value of the Norwegian
krone against Euro?
2. Literature review:
a. Norway and the Eurozone
The exchange rate between two currencies measures, first of all, the value of one currency compared to that of the other. Yet it also reflects much of the competitiveness and trading relation between the two currency owners. Therefore, the first sector of the literature review is projected to present an overview of the economic and trading relationship of Norway and the Eurozone.
Despite not officially joining the European Union, Norway is one of the key trading partners with the European Union including 19 members of the Eurozone. Norway is the 5th import partner and 7th export partner of the European Union (European
Commission). On the other hand, 80% of Norway’s exports go to the EU and 60% of its imports come from the EU as well (Norwegian Ministry of Foreign Affair). This strong relation is greatly supported by the European Economic Area Agreement coming into force in January 1994. The agreement gives Norway the equal status and full access to all the trading freedoms of a European Union member state, which has been significantly boosting the trading flows between the two counterparties. This equal term at the same time provides a favorable environment for the exchange rate analysis due to the freedom of investment flows. The free movement of capital in combination with the floating rate regimes allows the exchange rate of the NOK versus the EUR to be applicable under the demand – supply model. Thus, any increases in competitiveness of one party can lead to a movement of capital and foreign investment in or out the country, resulting in the fluctuation of the demand for local or foreign currency and the adjustment of the exchange rate accordingly. Additionally, consisting of 19 members, among which are the advanced economies such as Germany, the Netherlands or France, the Eurozone certainly is one of the country groups with such a high level of energy demand including crude oil. For countries as the Netherlands, France or Germany, where not only the domestic demand for oil is relatively high but also the oil production has the tendency of declining in the recent years, Norway has been playing the role of one of major oil supplies for years, accounting for more than 10% total oil imports volume into EU (European Commission Statistics). From the supply and demand perspective, the
Eurozone is heavily dependent on oil importing with more than 50% of gross inland consumption coming from imported sources (Eurostat). Meanwhile, Norway is a net crude oil exporter with more than two thirds of annual oil production volume used for exports (Statistics Norway). Based on this fact and the oil trading volume between the Eurozone and Norway, intuitively it does support the hypothesis of a considerable effect of oil price on the competitiveness of the two parties, potentially reflected in the value of one currency against the other.
b. Oil price and exchange rate
The possibility of including commodity prices in the exchange rate forecasting model has been investigated in many theoretical and empirical papers. From a theoretical intuition, an increase in commodity prices indicates a growth in demand for the commodity and consequently, a potential growth in trade revenues for the commodity exporter and vice versa. This positive or negative signal will initiate a change in the currency value of the commodity exporter in either upward or downward trend. Moreover, Chen and Rogoff (2003) also provided empirical evidence regarding the relationship between the exchange rate and the commodity prices applying in developed countries with rich resources such as Australia, Canada and New Zealand. The analysis found a strong correlation between the exchange rate and the commodity price index whereas other classic factors as interest rate differentials or output differentials, in fact, did not show significant connections. The similar result is established in the recent paper of Ferraro, Rogoff, and Rossi in 2015. However, although most of papers supported the robust relation between the commodity prices and the exchange rate, this relationship has not been yet empirically confirmed as the correlation seemed to last only in a short-term horizon (daily data) and diminish in longer time frames i.e. monthly and quarterly frequencies. This is one of motives to conduct the follow research with the hope to provide a least-biased empirical evidence of the effect of the commodity prices, particularly crude oil price, on the exchange rate.
One critical point to be mentioned when carrying out the research is that to efficiently analyze the relationship between the commodity price and the exchange
rate, the subject country should be an open economy with floating exchange rate regime where both internal and external markets encounter little intervention. From this respect, Norway with a long history of flexible exchange rate policy since 1982 can be considered as a nearly perfect candidate. Empirically, the relationship between the oil price and Norwegian exchange rate still remains ambiguous due to disuniting evidence. While Akram and Holter (1996) found an insignificant and weak relation, Akram (2000) suggested a statistically significant negative correlation between the oil price and the Norwegian krone versus ECU exchange rate and a substantial improvement in the predictability of the exchange rate with the model including oil price. This paper supports the conclusion and anticipates the similar outcome applied for the exchange rate of Norwegian krone versus Euro as Akram (2000) taking into account the historical movements of the two variables. Figure 2a shows the graph plotting the annual movements of Brent crude oil price index and NOK/EUR exchange rate since 1999 when Euro is officially launched. The figure dedicates a contradictory trend with increasing magnitude of the two variables throughout time and especially at time where the oil price made such striking moves as in 2009, 2012 and potentially 2015 – 2016. This observation corresponds with the theory of the effect of oil price shocks on the exchange rate. When the oil price rises, the effect stemming from the transfer of wealth from the oil importing countries to the oil exporters, will possibly drive the currency of the oil exporters to appreciate. On the other hand, the fall in oil price results in major revenues cut for the exporters while due to cost commitments, spending still continues to go up. The negative impact leads to a decline in investment supply, current account deficit and so on, which would subsequently reflect in an increase in the exchange rate of the oil exporters (Moshiri, 2015, p.223).
Figure 2a. Annual movements of Brent crude oil index and the exchange rate NOK versus EUR (1999 – 2015) c. Exchange rate determinants
Intuitively, the exchange rate fluctuations are influenced by main factors of the economy other than the oil price, which also needs taking consideration for any analyses. First of all, monetary policy can directly affect the exchange rate via the foreign currency reserves and local currency rejections. However, in case of Norway, Norges Bank has been very reluctant to make any interventions to the foreign exchange markets and as of the records, no intervening actions from the central bank have been conducted since 1999. Hence, this fiscal influence can be disregarded within the extent of this research. The second factor to be considered is the interest rate differential. According to the Uncovered Interest Parity (UIP) hypothesis, a higher interest rate in the home country would attract foreign investment inflows, which triggers an increase in demand for the home currency in 0 1 2 3 4 5 6 7 8 9 10 11 0 10 20 30 40 50 60 70 80 90 100 110 120 NOK/EUR USD/barrel Oil price NOK/EUR
the market and consequently, leads to an appreciation of the home currency (2013, pp. 149-150). Therefore, if UIP holds, the changes in the interest rate gaps should have an effect on the exchange rate variations as well.
Moreover, in accordance with one of the classic exchange rate theory – Purchasing Power Parity (PPP) theory, the value of one currency compared to another also reflects the relation between domestic and foreign prices or the inflation differential between the two countries (2013, pp. 126-128). A country with a lower inflation rate would exhibit rising in purchasing power and an increase in the currency value against the higher-inflation-rate trading partner. Another determinant of the exchange rate movements to be mentioned is the current account deficit. In the light of the balance of payment monetary policy, a current account deficit indicates a larger outflow for foreign trade than inflow for domestic earnings. In other words, as the country is spending more on importing goods than it is earning from the sales of exports, the demand for foreign currency to make the payment for import flows would raise following by a depreciation of the domestic currency compared to the foreign one. Beside the above-mentioned factors, economic performance and terms of trade are also considered to have a considerable influence on the changes of the exchange rate. It is noticeable that changes in demand and supply on currency market depend largely on the investor’s confidence about the national economy in general. Especially in the case of Norway with GDP per capita ranking 7th globally by 2015
estimate, the stable and strong growing economic past performance will bring about much confidence on the value of the currency as well (World Economic Outlook Database, April 2016, IMF). Additionally, terms of trade are also a very important determinant of the exchange rate variations. Terms of trade are calculated by the ratio of relative export prices to import prices or in other words, the quantity of import goods can be purchased by one unit of export goods. For the oil exporters i.e. Norway, an upward climbing in the oil price will result in a favorably improved terms of trade, which also indicates an increase in the export revenues for Norway and subsequently an appreciation of the local currency krone. In this way, it seems like terms of trade and current account deficit affect the exchange rate in a rather
similar method. Nevertheless, it is vital to keep in mind terms of trade captures directly the effect of the increase in export prices with respect to import prices whereas with balance of payment, it is indistinguishable between the effect of changes in volumes and prices of either export or import goods.
Last but not least, since the paper narrows down to the value of the Norwegian krone against the Euro while the oil price is denominated in US dollar, it is also crucial to take into account of the influence of the fluctuations of the USD dollar against the Euro to eliminate part of oil price changes simply due to the volatility of either the US dollar or the Euro (Zhang, 2013, pp. 341-344). The investigation of all above factors is with the purpose to include as many essential factors as possible in the following exchange rate predictive model to hopefully provide a least-biased outcome. Nonetheless, it is well noted that these inclusions at this point are entirely based on theoretical intuition and whether these effects empirically exist is still to be determined from the statistical perspective.
3. Methodology
Based on the above literature review, the methodology of the analysis will be discussed step by step in the next section.
a. Testing for unit roots
The first important point to take into account is that the paper would conduct a time series analysis and therefore, most of the statistical forecasting methods are based on the main assumption of stationarity, which means that the time series has a probability distribution that is constant over time (Stock and Watson, 2015, p.587). A non-stationary or stochastic trend time series leads to a variety of problems, the most blatant of which is that OLS estimator of its coefficient and t-statistic can have a non-normal distribution even in large samples (Stock and Watson, 2015, pp.600-601). Additionally, the regression of two random-walk variables would very likely result in a significant correlation but this is simply due to the fact they both vary and move along a line but not because they are in fact correlated. Among the financial and economic time series, the exchange rate and commodity price are very typical examples of time series with non-stationary behavior in the mean. Consequently, to determine the appropriate form of the trend that should be used in the data, the unit root tests are implemented to investigate the stationarity of these two time series: NOK/EUR and Brent crude oil price. In order to test the stationary trait of the time series, two unit-root tests will be implemented given their relatively low power, namely the Augmented Dickey-Fuller (ADF) test and Kwiatkowski–Phillips– Schmidt–Shin (KPSS) test. The ADF test for a unit root autoregressive root tests for the null hypothesis of a stochastic trend versus the one-sided hypothesis of the stationarity (Stock and Watson, 2015, p. 605). Meanwhile, the KPSS test uses a one-sided right-tailed test with the null hypothesis of stationarity against the alternative hypothesis of a unit root (Kwiatkowski et al., 1992, p.160). From the results of the ADF test, the stationarity of variable Oil price is denied with the test statistics far larger than critical value of all 1%, 5% and 10% significance levels so the null hypothesis is not rejected (see Appendix, Table 3a). The stochastic trait of Oil price time series is confirmed by the KPSS test results as well. In all lags
from 0 to 11, the test statistics is larger than the critical values at 1% significance (see Appendix, Table 3b). As a result, the null hypothesis of stationarity is rejected. The same outcomes are found when applying both tests on time series of the NOK/EUR exchange rate. With a p-value of 0.1530, the null hypothesis of stationarity cannot be rejected at 5% significance level in ADF test (see Appendix, Table 3c). Meanwhile, with the test statistics in every lag between 0 and 11 higher than the critical value of 1% significance, KPSS test for the NOK/EUR exchange rate also confirms the stochastic trend of the time series (see Appendix, Table 3d). Based the above support of two tests for unit root, the dependent and independent variables are proved to be non-stationary and to avoid the validity issues in further tests, the first difference of both variables will be used in the model instead of the level form. b. Predictive model
To investigate the oil price changes effect on the exchange rate of Norway, the analysis is based on the linear regression model with dependent variable being relative change in the nominal exchange rate of the NOK and the EUR while the variable of interest is represented by the relative variation in oil price. Furthermore, in order to avoid omitting other explanatory factors of the exchange rate movements, several control variables as discussed in the literature review are included in the regression. However, as the model will be run on daily, monthly and quarterly basis, the control variables included also differ upon the frequency being investigated. For instance, while the inflation rate possibly varies from month to month and consequently, is likely to reflect its effect on the monthly basis exchange rate, current account or national income variations only come into effect annually. The best way to determine which appropriate frequency to include which factors is to correspond to the data availability frequency. This approach is supportive since at the same basis, the market receives the information on the fluctuations of those variables and then adjusts accordingly. As a consequence, only the relative change in USD/EUR rate and the differentials in interest rate are included in daily model and beside inflation rate gaps in the monthly model. For quarterly frequency, in addition
to three above-mentioned factors, current account deficit ratio, relative change in national income, and terms of trade are taken into account. The models for analyzing in daily, monthly and quarterly basis respectively are as follow: On daily basis: Δ𝑠!"#$% = 𝛽!+ 𝛽!Δ𝑝!"# + 𝛽!∆𝑠!"#$% !"# !"# + 𝛽 ! 𝑟 − 𝑟∗ + 𝜀! 1 On monthly basis: Δ𝑠!"#$!!" = 𝛽!+ 𝛽!Δ𝑝!"# + 𝛽!∆𝑠!"#$!!"!"#/!"#+ 𝛽! 𝑟 − 𝑟∗ + 𝛽! Δ𝑐𝑝𝑖 − Δ𝑐𝑝𝑖∗ + 𝜀! (2) On quarterly basis: Δ𝑠!"#$%&$'( = 𝛽!+ 𝛽!Δ𝑝!"# + 𝛽!∆𝑠!"#$%&$'(!"#/!"# + 𝛽! 𝑟 − 𝑟∗ + 𝛽 ! Δ𝑐𝑝𝑖 − Δ𝑐𝑝𝑖∗ + 𝛽!𝐶𝐴/𝐺𝐷𝑃 + 𝛽!𝐺𝐷𝑃 + 𝛽!𝑇𝑜𝑇 + 𝜀! (3) Where:
Δ𝑠!"#$%, Δ𝑠!"#$!!", Δ𝑠!"#$%&$'(: relative change in the nominal exchange rate
NOK/EUR in daily, monthly, quarterly frequency respectively.
Δ𝑝!"#: relative change in the Brent crude oil index in daily, monthly, quarterly frequency dependent on the respective model.
∆𝑠!"#$%!"#/!"#, ∆𝑠!"#$!!"!"#/!"#, ∆𝑠!"#$%&$'(!"#/!"#: relative change in the nominal exchange rate USD/EUR in daily, monthly, quarterly frequency respectively.
𝑟 − 𝑟∗ : spread between nominal sight deposit rate of Norway and the Eurozone
Δ𝑐𝑝𝑖 − Δ𝑐𝑝𝑖∗: spread between the relative change in Harmonized Consumer Price
Index of Norway and the trade-weighted Harmonized Consumer Price Index of the Eurozone.
𝐶𝐴/𝐺𝐷𝑃: current account deficit over gross domestic product (GDP) ratio of Norway 𝐺𝐷𝑃: gross domestic product of Norway 𝑇𝑜𝑇: terms of trade of Norway. The main coefficient of interest in each model is 𝛽! representing the oil price effect on the exchange rate of the NOK versus the EUR. Under the null hypothesis, oil price has no impact on the NOK/EUR rate or in other words, 𝛽! = 0. On the other hand,
under the alternative hypothesis that is expected to be found, oil price has a negative effect on the exchange rate (𝛽! < 0).
4. Data sample
From the above-mentioned models, the data sample is collected according to the appropriate frequency. The dependent variable – NOK/EUR exchange rate – is extracted from Datastream for the observations of daily, monthly and quarterly rate between 1999 and March 2016. The variable of interest – Brent crude oil spot price – is collected from EIA – U.S. Energy Information Administration Data Source. As the data is only available at daily, monthly and annual frequencies, the quarterly data is derived by taking the average of the monthly data of the corresponding quarter. After matching and filtering missing data, a sample of 4372 observations for the daily model, of 207 observations for the monthly model, and of 68 observations for the quarterly model is gathered. From the descriptive statistics of each sample (see Appendix, Table 4a), it is suggested that the period under investigation witnesses striking movements of the oil price considering the large standard deviation correspondent. This is also another motive to use the relative change instead of the first difference form of the variables in the regressions.
To capture the interest rate differentials effect, the spread between the nominal overnight deposit rate of the Norway and the Eurozone is used. The overnight rate is considered as the key policy rate determining interest rate in both countries and the most important monetary policy instrument for both central banks so this is the rate that reflects most accurately the competitiveness of investment return in two currency owners.
With the regards to the inflation rate, there are several methods and measurements of a nation’s inflation. With the view to staying consistent and comparable, the Harmonized Consumer Prices Index (HCPI) is used as the indicator of the price stability of each country. This index is reported to the European Central Bank and has been harmonized across European countries so any misalignments in measurement method can be eliminated. Due to data availability, the chosen base year is 2015 for the monthly and quarterly HCPI rates of Norway and Eurozone countries.
Furthermore, it is noteworthy that although all member states of the Eurozone share the common currency, they are very distinguishable with respect to the
economic elements. Therefore, in order to have an objective measurement, the HCPI rate of the Eurozone is calculated as a trade-weighted rate. The paper investigates two methods of assigning weights to each state member, namely the total exports value base or the oil exports value base. The total exports value in NOK between Norway and each country member of the Eurozone for the period between 1999 and 2015 is extracted from Statistics Norway. Based on this, the value share of each nation can be calculated for each year and subsequently, the average yearly value share will be the weight of the correspondent country in the total rate. In the second method where oil exports value is used as the base, the procedure is applied except for crude oil exports value being used instead of the total exports value. Table 4b presents the weights assigned to each state member calculated by both methods. It is noticeable that according to the oil export volume base method, there is some concentration in share value of certain state members namely the Netherlands (47%). Such a high volume of oil imports would raise the question of the re-export portion where the Netherlands is not indeed the final destination of the export flows. Nonetheless, from Energy Supply Security 2014, it can be derived that the crude oil imports in the Netherlands is used for domestic demand i.e. industry and transformation sectors. Hence, the crude oil imported in the Netherlands is either supplied for residential, industry and other segments or converted into other petroleum or chemicals products before leaving the country (Energy Supply Security 2014, p. 320). Thus, it is still acceptable to consider the respective weight of the Netherlands among the Eurozone in relation with Norway from the trade perspective. The total HCPI of the Eurozone will be then calculated as follow: 𝐻𝐶𝑃𝐼!" = 𝑤!𝐻𝐶𝑃𝐼! !" !!!
Where 𝑤! is the weight and 𝐻𝐶𝑃𝐼! is the HCPI rate of country i.
Table 4b. Exports-based weights of Eurozone country members Country Total exports based weight Crude oil exports based weight AT Austria 1% 0% BE Belgium 8% 2% CY Cyprus 0% 0% EE Estonia 0% 0% FI Finland 4% 4% FR France 17% 18% DE Germany 30% 18% GR Greece 1% 0% IE Ireland 3% 4% IT Italy 5% 4% LV Latvia 0% 0% LT Lithuania 0% 0% LU Luxembourg 0% 0% MT Malta 0% 0% NL Netherlands 26% 47% PT Portugal 1% 1% SK Slovakia 0% 0% SI Slovenia 0% 0% ES Spain 4% 2% Sum 100% 100% Other variables of Norway’s economic performance as current account deficit, gross domestic product (GDP), and terms of trade are extracted from Datastream at the quarterly frequency of the investigating period from 1999 to March 2016.
5. Expected results
Based on the literature view, it is predicted that oil price would have a significantly negative effect on the exchange rate fluctuations of the Norwegian krone and the Euro. The effect is also projected to diminish throughout time horizons according to the similar study of Ferraro et al. The authors found the existence of the oil price effect but the impact is short-lived and becomes insignificant in monthly and quarterly basis (Ferraro et al., 2015, p.117). This prediction is not only in the light of many empirical studies about the linear relationship of exchange rate and oil price but also followed by the overview of the historical trends. The below chart plots the empirical movements of oil price and exchange rate on daily, monthly, and quarterly basis. It is recognizable that the effect’s magnitude changes across the frequencies and tends to diminish in longer term. As a consequence, it is predicted that the coefficients of Oil price variable would decline across the models indicating a shrinking effect of oil price. Figure 5a. Plotting the movements of oil price and NOK/EUR rate in different frequencies
6. Empirical results
The next section will present results of the above linear regression models applied with daily, monthly and quarterly data basis. To ensure the statistical accuracy of the analysis, robust standard errors are used in all regressions. As mentioned above, to calculate the trade-weighted variables with respect to the Eurozone, two approaches have been applied using the total trade weights and using the oil exports volume weight. However, both methods show rather similar outcomes with regards to coefficients. Therefore, only the results from the total trade volume weights calculation are discussed in details and the outputs of models drawn from the other method will be presented in the Appendix.
a. Daily data model
Table 6a (Appendix) below shows the output from the robust linear regression explained in formula (1). As can be derived from the output, the oil price does show a negative effect on the movement from the exchange rate with the coefficient of ≈-.054 and this effect is significant even at 1% (p-value < 0.01). The result corresponds to the prediction following the literature review and historical study of the contradictory movements of the two variables. The exchange rate of the USD versus the EUR demonstrates a positive movement along with the rate of the NOK versus the EUR. Intuitively, this result does make sense since, for instance, the Euro depreciated against US dollar, it is either the case of a devaluation of the euro in general or with a global currency – US dollar. Both cases would lead to value downgrading of the euro with most of other currencies including the Norwegian krone.
Although the R-square stays at quite low level (less than 10%), it is an expected outcome as the models regarding daily data are most of the time rather noisy and most of the time, come out with relatively R-squared value.
What is also very interesting to observe is that the interest rate differentials show an extremely insignificant effect (p-value of 0.997) (see Appendix, Table 6a), which indicates that interest rate gaps may not have a considerable influence on the variations of the exchange rate. In order to confirm this observation, a second
regression is run without the variable RNOR – REUR and compared with the original
model. In fact, discarding the insignificant control variable of interest rate does not diverge the estimation error of the model and standard error of the independent variables actually decreases (see Appendix, Table 6b). This outcome, in fact, aligns with findings of Meese and Rogoff, which indicated that the relationship between the real exchange rate and the real interest rate is statistically insignificant (1988, p.943). They also concluded that adding the real interest rate differential to the random walk model does not improve the predictive capability substantially (1988, p.943). The results of the two model forms suggest the same conclusion and hence, it is recommended to leave the interest rate difference variable out of the prediction model. b. Monthly data model Applying the model in formula (2) mentioned in the Methodology, the similar robust regression is run with monthly data and an additional control variable, namely the inflation rate gaps ∆CPINOR - ∆CPIEU.
The linearity estimation once again confirms a significantly negative oil price effect on the exchange rate movement with a coefficient of ≈ -.068 (see Appendix, Table 6c). The effect appears to increase throughout the time zone as on daily basis, a 1% increase in the crude oil price is expected to result in a decrease in the NOK/EUR rate by 0.054%. Meanwhile, the same increase in the monthly crude oil price would lead to a drop of 0.068% in the exchange rate of the Norwegian krone versus the Euro. The inflation effect found in the model is positive (coef. ≈ 0.007) and statistically significant at 5%, which also corresponds to purchasing power parity theory discussed in the literature review.
The estimation also gives more confidence, as R-square is approximately 16.5%, which can be considered as acceptable for this type of analysis. Similarly to the previous model, interest rate shows insignificant correlation with the exchange rate fluctuation and the reduced regression without variable RNOR – REU does not show a
Therefore, it is recommendable to use the simplified regression model with only 2 control variables.
c. Quarterly data model:
The outputs of the regression applied for quarterly frequency presented in Table 6e (Appendix) do not entirely support the prediction derived from the empirical studies. The model establishes a significant effect of the oil price movements on the exchange rate at a very strong magnitude compared to the monthly model. It is suggested a countermovement of the oil price at 0.135% in the original model and 0.156% in the reduced model (excluding insignificant control variables) corresponds to an increase of 1% in the exchange rate of Norwegian krone versus Euro (see Appendix, Table 6e & 6f). The R-squared of the model of more than 35% does allow some comfort regarding the predictability of the model.
It is also noticeable that most of factors expected to show some impact on the exchange rate do not indicate a statistical significance. Beside interest rate differentials with historical insignificant performance, the inflation gap that appear to be influential in the monthly model, loses its considerable effect in the quarterly despite being rather consistent with positive coefficient. Coefficients of the GDP
change and Terms of trade variables, in spite of not being statistically significant
with very high p-value, still give some insights of their relationship with the exchange rate. From the original model, the current account deficit ratio illustrates a negative effect that is significant at 10% but not at 5%. Therefore, to get a concrete answer whether this variable does matter or not, an additional regression is run without CA/GDP variable and the errors are compared between the two versions (see Table 6e & 6g). It is evident that eliminating current account deficit from the model worsens the predictive capability of the model with the root MSE increases substantially. d. Follow-up test As discussed at the beginning of the paper, the research intends to go beyond the confirmation of oil price effect and to investigate whether this effect can help to
improve the predictability of the exchange rate models. Based on the recommended models to predict exchange rate movements derived in the results analysis, the F-statistics tests are conducted to compare the models with and without oil price difference. All three additional F-tests result in better predictability of the model with oil price effect compared the model excluding this variable with very low p-value (smaller than 0.01) suggesting a very high statistical significance (see Appendix, Table 6b, 6d, 6f). As a consequence, it is implied that including oil price variations does improve the predictability of the exchange rate models.
Furthermore, it is suggested from many papers, some of which were discussed in the literature review i.e. Cheung et al., (2005), Meese and Rogoff (1988), that the random walk model empirically shows a superior outperformance compared to other popular structural models. The underlying idea of the random walk is that the value of tomorrow rate equals to today value plus an unpredictable change, which has a conditional mean of zero (Stock and Watson, 2015, pp. 598-599). The stochastic trend can be in the form of with drift or driftless depending on whether the movement falls for a consistent direction. As can be seen from Figure 5a aligned with intuition from empirical studies on the random walk model such as Cheung et a., (2005), and Rossi (2005), the exchange rate does not show a directional trend in movements, suggesting the implementation of the random walk model without drift. Consequently, the random walk model without drift can be expressed as follow: 𝑠!!!= 𝑠!+ 𝑢! Where: 𝑠!!!, 𝑠!: Exchange rate NOK/EUR at time t+1 and t 𝑢!: error term As discussed in the Empirical results, in fact, the predictability of the model at each frequency is very much differentiating, that leads to a question of which model indeed performs better Following the method of Diebold and Mariano (1995), the MSE of the models are calculated to compare the predictive capability of the proposed models at various frequencies. Table 6k shows the MSE values of the proposed model including oil price effect against the random walk model at respective frequencies (Appendix). As derived from the table, the proposed model
with oil price movement effect shows a superior outperformance with regards to MSE difference compared to the random walk model with rising effect in longer time frames. However, it is also notable that this outcome partially comes from the fact that random walk performance also worsens off as the estimation period increases.
7. Conclusion
In conclusion, the paper aims to investigate on daily, monthly, and quarterly level, the effect of crude oil price fluctuations on the variations of the exchange rate between the Norwegian krone and the Euro, using the linear regression of the relative first difference in the exchange rate on the relative change in the Brent oil price. The models also comprises other factors considered to be influential on the exchange rate in the light of different theoretical frameworks, such as interest rate differentials, inflation rate gaps, current account deficit, national income, and terms of trade. These factors play the role of control variables in each model, dependent on the examining frequency, with the view to increasing the forecasting capability of the models.
All in all, the regression results suggest a significant negative effect of oil price on the exchange rate on all three investigated frequencies. This outcome is in line with the empirical studies of the oil price in general and also for the specific case of Norwegian krone. What is rather interesting and new of the research’s finding is that the effect magnitude increases considerably from daily to monthly basis and even more substantially in quarterly basis. This influence extent was contradictory to the prediction prior the analysis when looking at the empirical literature and historical co-movements of the two variables. Although the relatively acceptable level of R-squared does give some comfort, the fact that most of control variables which are presumed to have an impact on the dependent variable turned out to be insignificant, also draws some attention. On one hand, it does raise a question of the existence of other exchange rate determinants that are not explored or taken into account in this research. On the other hand, one can also argue that given the “hard-to-beat” record of the random walk model, it does make sense that other variables do not show the significance. Nevertheless, the main result of a significant negative relationship between oil price and the exchange rate of Norway brings about another empirical evidence to confirm this so-called controversial effect.
Furthermore, the follow-up F-statistics test provides a confirmation of an improvement in the forecast capability of the exchange rate models with the inclusion of oil price effect variable. This outcome is a noteworthy remark to take into account the choice of adding oil price effect on exchange rate prediction models for the oil-exporters like Norway. It is also important that although the outcome in the second additional test seems quite extraordinary when all proposed models including oil price effect appear to be able to “beat the random walk”, it is still very ambiguous given the simplicity of the comparison criterion. With the view to having a more concrete conclusion of the superiority of the models, further researches should take into account different aspects of the performance under various of comparison tests.
In spite of the fact that the final results are in general aligned with theoretical frameworks, the paper does realize some limitations to be considered in further research. Firstly, the insignificance of most of the control variables in all models does suggest further investigation required to check if there are other determinants of the exchange rate not found or omitted. From the study of Ferraro et al., it is recommendable to apply the tests for other commodities as well. For instance, in case of Norway, natural gas can be a potential candidate. Secondly, the research is constrained to analyze the exchange rate of Norwegian krone versus Euro so it is very much crucial also to expand the investigation to the rate against other foreign currencies before coming up with a conclusion. Additionally, the follow-up test at the moment aims to test the probable improvement of oil price effect on the recommendable models. Nonetheless, it would be interesting to investigate the possible improvement of including the oil price effect in existent popular forecasting models such as purchasing power parity or monetary models.
Appendix Table 3a. Output from the Augmented Dickey-Fuller (ADF) test for Oilprice variable Dickey-Fuller test for unit root Number of obs 6296 Interpolated Dickey-Fuller
Test statistics 1% Critical Value 5% Critical Value 10% Critical Value
Z(t) -1.63 -3.43 -2.86 -2.57 MacKinnon approximate p-value for Z(t) = 0.4674 Table 3b. Output from Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test for Oilprice variable KPSS test Oilprice Maxlag = 11 chosen by Schwert criterion Autocovariances weighted by Bartlett kernel Critical values for H0: Oilprice is level stationary 10%: 0.347 5%: 0.463 2.5%: 0.574 1%: 0.739 Lag order Test statistics 0 405 1 202 2 135 3 101 4 81 5 67.5 6 57.9 7 50.7 8 45 9 40.6 10 36.9 11 33.8
Table 3c. Output from the Augmented Dickey-Fuller (ADF) test for NOK/EUR variable Dickey-Fuller test for unit root Number of obs 6296 Interpolated Dickey-Fuller
Test statistics 1% Critical Value 5% Critical Value 10% Critical Value
Z(t) -2.361 -3.43 -2.86 -2.57 MacKinnon approximate p-value for Z(t) = 0.1530 Table 3d. Output from Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test for NOK/EUR variable KPSS test for NOK/EUR Maxlag = 11 chosen by Schwert criterion Autocovariances weighted by Bartlett kernel Critical values for H0: NOK/EUR is level stationary 10%: 0.347 5%: 0.463 2.5%: 0.574 1%: 0.739 Lag order Test statistics 0 41.08 1 20.9 2 14 3 10.5 4 8.41 5 7.01 6 6.02 7 5.27 8 4.69 9 4.23 10 3.85 11 3.53
Table 4a. Descriptive statistics Daily sample
Variable Obs Mean Std. Dev. Min Max
Oil price 4372 62,67222 33,67764 9,77 143,95
NOK/EUR 4372 8,127615
0,463695
2 7,228 10,1075
Monthly sample
Variable Obs Mean Std. Dev. Min Max
Oil price 208 62,84802 33,76687 10,27 132,72
NOK/EUR 208 8,122987 0,4607763 7,2625 9,73325
Quarterly sample
Variable Obs Mean Std. Dev. Min Max
Oil price 68 63,27672 33,72067 11,29667 121,3967 NOK/EUR 68 8,125000 0,4639576 7,27 9,73325 Table 6a. Output of the original daily regression using total trade volume weighted variables Number of obs 4.371 F(3,4367) 69,66 Prob > F 0,0000 R-squared 0,0703 Root MSE 0,00457 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oilprice -0,0537491 0,003948 -13,61 0,000 -0,0614892 -0,046008
USD/EUR 0,0821111 0,0155167 5,29 0,000 0,0516905 0,1125318
RNOR - REUR 0,0000177 0,0046179 0,00 0,997 -0,0090357 0,0090712
_cons 0,0000532 0,0001295 0,41 0,681 -0,0002007 0,0003071
Table 6b. Output of the reduced daily regression using total trade volume weighted variables Number of obs 4.371 F(2,4368) 104,27 Prob > F 0,0000 R-squared 0,0703 Root MSE 0,00457 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oilprice -0,0537489 0,003946 -13,62 0,000 -0,0614851 -0,0460127 USD/EUR 0,082111 0,0155138 5,29 0,000 0,0516961 0,1125259 _cons 0,0000535 0,0000695 0,77 0,441 -0,0000827 0,0001897 • test Oilprice = 0 (1) Oilprice = 0 F (1, 4368) = 185.53 Prob > F = 0.0000
Table 6c. Output of the original monthly regression using total trade volume weighted variables Number of obs 206 F(4,201) 9,19 Prob > F 0,0000 R-squared 0,1651 Root MSE 0,01828 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oilprice -0,0680557 0,0143186 -4,75 0,000 -0,0962897 -0,0398218
USD/EUR 0,1541011 0,0643222 2,4 0,018 0,0272683 0,2809339
RNOR - REU -0,0669015 0,1061163 -0,63 0,529 -0,2761455 0,1423424
∆CPINOR - ∆CPIEU 0,0066258 0,0029889 2,22 0,028 0,0007321 0,0125195
_cons 0,0024293 0,0025615 0,95 0,344 -0,0026216 0,0074802
Table 6d. Output of the reduced monthly regression using total trade volume weighted variables Number of obs 206 F(3,202) 12,32 Prob > F 0,0000 R-squared 0,1633 Root MSE 0,01826 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oilprice -0,0687971 0,014188 -4,85 0,000 -0,0967726 -0,0408215
USD/EUR 0,1552868 0,0646383 2,4 0,017 0,0278344 0,2827393
∆CPINOR - ∆CPIEU 0,006647 0,0029774 2,23 0,027 0,0007763 0,0125178
_cons 0,0010419 0,001291 0,81 0,421 -0,0015037 0,0035875 • test Oilprice = 0 (1) Oilprice = 0 F (1, 202) = 23.51 Prob > F = 0.0000
Table 6e. Output of the original quarterly regression using total trade volume weighted variables Number of obs 67 F(7,59) 4,86 Prob > F 0,0002 R-squared 0,3918 Root MSE 0,03314 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oilprice -0,1353805 0,0549458 -2,46 0,017 -0,2453267 -0,0254342
USD/EUR 0,1920642 0,0731393 2,63 0,011 0,0457127 0,3384157
RNOR – REU -0,3788384 0,3896127 -0,97 0,335 -1,158452 0,4007748
∆CPINOR - ∆CPIEU 0,0009107 0,0078681 0,12 0,908 -0,148334 0,0166548
CA/GDP -0,231675 0,129495 -1,79 0,079 -0,490794 0,0274439
GDP change -0,1487698 0,4578379 -0,32 0,746 -1,064901 0,763618
Terms of trade -0,0004586 0,0007257 -0,63 0,53 -0,0019108 0,0009936
_cons 0,0928403 0,0860716 1,08 0,285 -0,0793885 0,2650692
Table 6f. Output of the reduced monthly regression using total trade volume weighted variables Number of obs 67 F(3,63) 10,51 Prob > F 0,0000 R-squared 0,3799 Root MSE 0,03238 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oilprice -0,1560157 0,0371147 -4,2 0,000 -0,2301835 -0,0818479 USD/EUR 0,1937826 0,0703024 2,76 0,008 0,0532946 0,3342707 CA/GDP -0,2758968 0,1645605 -1,68 0,099 -0,6047448 0,0529512 _cons 0,0411197 0,02375 1,73 0,088 -0,0063409 0,885803 • test Oilprice = 0 (1) Oilprice = 0 F (1, 63) = 17.67 Prob > F = 0.0001
Table 6g. Output of the reduced monthly regression without CA/GDP variable using total trade volume weighted variables Number of obs 67 F(2,63) 27,73 Prob > F 0,0000 R-squared 0,8103 Root MSE 0,05652 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oil price -0,0748384 0,0541705 -1,38 0,172 -0,1830243 0,0333474 USD/EUR 0,9357972 0,1257507 7,44 0,000 0,6846556 1,186939 _cons 0,0013171 0,0075766 0,17 0,863 -0,0138144 0,0164487 Table 6h. Output of the original monthly regression using oil-exporting volume weighted variable Number of obs 206 F(4,201) 8,74 Prob > F 0,0000 R-squared 0,1626 Root MSE 0,01831 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oilprice -0,0679601 0,0143129 -4,75 0,000 -0,0961828 -0,0397375
USD/EUR 0,1561271 0,0647228 2,41 0,017 0,0285043 0,2837498
RNOR - REU -0,0620607 0,1060569 -0,59 0,559 -0,2711876 0,1470663
∆CPINOR - ∆CPIEU 0,0061516 0,0030047 2,05 0,042 0,0002268 0,0120763
_cons 0,0023617 0,0025698 0,92 0,359 -0,0027055 0,0074289
Table 6i. Output of the reduced monthly regression using oil-exporting volume weighted variable Number of obs 206 F(3,202) 11,72 Prob > F 0,0000 R-squared 0,1611 Root MSE 0,01828 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oilprice -0,0686171 0,0141896 -4,84 0,000 -0,0965958 -0,0406383
USD/EUR 0,1572168 0,0650582 2,42 0,017 0,0289366 0,2854971
∆CPINOR - ∆CPIEU 0,0062177 0,0030001 2,07 0,039 0,0003021 0,0121332
_cons 0,0010739 0,0012928 0,83 0,407 -0,0014753 0,0036231
Table 6j. Output of the original quarterly regression using oil-exporting volume weighted variable Number of obs 67 F(7,59) 4,85 Prob > F 0,0002 R-squared 0,3918 Root MSE 0,03314 FX NOK/EUR Robust
Coef. Std. Err. t P>|t| [95% Conf. Interval]
Oil price -0,1352595 0,0544806 -2,48 0,016 -0,2442749 -0,0262441
USD/EUR 0,1920485 0,0731317 2,63 0,011 0,0457124 0,3383846
RNOR - REU -0,376377 0,3855195 -0,98 0,333 -1,1478 0,3950457
∆CPINOR - ∆CPIEU 0,0009015 0,0074164 0,12 0,904 -0,0139387 0,0157417
CA/GDP -0,2314801 0,1299297 -1,78 0,08 -0,4914689 0,0285087 GDP change -0,1503993 0,457543 -0,33 0,744 -1,065941 0,7651422 Terms of trade -0,0004584 0,0007262 -0,63 0,53 -0,019116 0,0009948 _cons 0,0927741 0,086192 1,08 0,286 -0,0796958 0,2652439 Table 6k. MSE comparison between random walk model and proposed model with oil price effect
Frequency Model Formula MSE Difference
Daily Random walk model 𝑠!!! = 𝑠!+ 𝑢! 0,001587715 0,001585 Proposed model Δ𝑠!"#$% = 𝛽!+ 𝛽!Δ𝑝!"# + 𝛽!∆𝑠!"#$%!"# !"# + 𝜀! 2,52084E-06 Monthly Random walk model 𝑠!!! = 𝑠!+ 𝑢! 0,027722671 0,027389 Proposed model Δ𝑠!"#$!!" = 𝛽!+ 𝛽!Δ𝑝!"# + 𝛽!∆𝑠!"#$!!"!"#/!"# + 𝛽! Δ𝑐𝑝𝑖 − Δ𝑐𝑝𝑖∗ + 𝜀 ! 0,000333428 Quarterly Random walk model 𝑠!!! = 𝑠!+ 𝑢! 0,110007555 0,108959 Proposed model Δs!"#$%&$'( = β! + β!Δp!"#+ β!∆s!"#$%&$'(!"#/!"# + β!CA/GDP + ε! 0,001048464
References Akram, Q.F. (2004). Oil prices and exchange rates: Norwegian evidence. The Econometrics Journal, 7(2), 476-504. Campell, J.Y., and Thompson S.B. (2008). Predict Excess Stock Returns out of sample: Can anything beat the Historical average?. The Review of Financial Studies, 21(4), 1509-1531. Cheung, Y.W., Chinn, M.D., and Pascual. A.G. (2005). Empirical exchange rate models of the nineties: Are there any fit to survive?. Journal of International Money and Finance, 24, 1150-1175. Diebold, F.X., and Mariano, R. (1995). Comparing Predictive Accuracy. Journal of Business and Economics Statistics, 13, 253-265. Ferraro, D., Rogoff, K., and Rossi, B. (2015). Can oil prices forecast exchange rates? An empirical analysis of the relationship between commodity prices and exchange rates. Journal of International Money and Finance, 54, 116-141.
International Energy Agency (2014). Energy Supply Security 201: Emergency
Response of IEA Countries, EA, Paris. DOI:
http://dx.doi.org/10.1787/9789264218420-en International Monetary Fund. (2015). World Economic Outlook Review October 2015: Adjusting to Lower Commodities Price. Washington (October). Killian, L., and Taylor, M.P. (2003). Why is it so difficult to beat the random walk forecast of exchange rates?. Journal of International Economics, 60, 85-107.
Kwiatkowski, D., Phillips, P.C.B., Schmidt, P., and Shin, Y. (1992). Testing the null hypothesis of stationarity against the alternative of a unit root. Journal of Econometrics, 54, 159-178. Meese, R. and Rogoff, K. (1988). Was it Real? The Exchange rate-Interest differential relation over the modern floating-rate period. Journal of Finance, 43(4), 933-948. Moshiri, S. (2015). Asymmetric effects of oil price shocks in oil-exporting countries: the role of institutions. OPEC Energy Review June 2015, 222-246. Stock, J. H. and Watson M. W. (2015). Introduction to Econometrics 3rd Edition. London, England: Pearson Education. Organization of the Petroleum Exporting Countries (2015). OPEC Annual Statistical Bulletin 50th Edition, Vienna. Pilbeam, K. (2013). International Finance 4th Edition. London, England: Palgrave Macmillian.
Zhang, Y. (2013). The Links between the Price of Oil and the Value of US Dollar.
International Journal of Energy Economics and Policy, 3(4), 341-351.
