A cross-country analysis of the relationship between a falling oil price and central bank reserves from oil-exporting countries under a pegged exchange rate to the U.S. dollar

(1)

University of Amsterdam

Faculty of Economics and Business

A cross-country analysis of the relationship between a falling oil

price and central bank reserves from oil-exporting countries

under a pegged exchange rate to the U.S. dollar

Author: F.R.M. Kempen (10364587) Supervisor: R.E.F. van Maurik MSc Study program: BSc Economics and Business Date: January 25, 2016

(2)

1

Abstract

This paper empirically shows the relationship between a falling oil price and central bank reserves from oil-exporting countries under a pegged exchange rate to the U.S. dollar. The expectation is, given economic theory and literature, that a pegged exchange rate to the U.S. dollar has a significant effect on central bank reserves from oil-exporting countries given a falling oil price. To determine this effect, a unique panel dataset from 1980 up to 2001 for fifteen significant oil-exporting countries was created and converted into time series. A time series analysis with country dummies is performed to look at the significant effect of the dummy variable d that was created in case of a pegged exchange rate to the U.S. dollar. This paper shows that, by rejecting the null hypothesis at a significance level of 10 percent, the effect of a pegged exchange rate to the U.S. dollar on an oil-exporting country’s reserves, given a falling oil price, is significant.

(3)

2

1 Introduction

On October the 17th 2015 The Economist stated that sinking might be a better description than floating when it comes to many of the world’s currencies. A force weakening many currencies has been the ongoing slump in commodity prices since mid-2014, which had hit producers of natural resources hard. A fall in commodity prices lowered the value of exports and causing their currencies to decrease, too. The current account of commodity-exporting countries will be influenced by the lower value of exports, resulting in a relative deficit on the balance of payments. In case there is a relative deficit on the balance of payments, the

domestic currency will depreciate.

Specifically the weak oil price has undermined the current account position of oil-exporting countries. As reported by The Economist Intelligence Unit, many currencies have followed the oil price down. For instance, the Norwegian krone, the Brazilian real and the Russian ruble declined since June 2014 respectively 26 percent, 40 percent and 45 percent against the U.S. dollar. According to economic theory a competitive exchange rate will boost economic growth. A country with a depreciating currency will have an increase in foreign demand that positively affects the economy.

However, countries with freely adjustable currencies are not that common. The International Monetary Fund (IMF) stated in its 2014 annual report about exchange arrangements and exchange restrictions that only 34 percent of member countries let their currencies float. In particular, only 15.2 percent of these member countries intervened rarely enough for the IMF to classify them as ‘free floating’. The other 18.8 percent of member countries just let their currencies ‘float’.

On the other hand, a pegged exchange rate to the U.S. dollar for an oil-exporting country makes sense because the oil price is priced in that currency. The value of exports can then remain its (higher) value under a pegged exchange rate to the dollar, but depends on its trading partners.

Recently The Economist wrote that on January 12th West Texas Intermediate (WTI) crude oil price, America’s benchmark, briefly dipped below $30 a barrel, its lowest level since 2003. According to the data of MacroTrends.net1

, a site that records economic charts and historical data, WTI crude oil (in U.S. dollars) lost from June 2014 until October 2015 approximately more than half of its value.

(5)

4

The recent fall in oil price (in U.S. dollars) puts pressure on the currency of an oil-exporting country to lose value. As a consequence, in case of a peg with the U.S. dollar, the central bank of an oil-exporting country must act to keep the exchange rate constant. This can be done through buying their own domestic currency by selling official reserves, called an official intervention. An official intervention will increase the demand for the domestic currency, which will offset the pressure on the currency to lose value. A thorough background of the relationship between the balance of payments, the current account and exchange rate regimes is given in section 2.

As stated above, a falling oil price has large consequences for oil-exporting countries, specifically on their exchange rates. Fluctuating oil prices have a strong effect on an oil-exporting country’s reserves in case the U.S. dollar is an exchange rate anchor. In this paper there is a particular interest in the relationship between a falling oil price and central bank reserves from oil-exporting countries under a pegged exchange rate to the U.S. dollar. In case there is pressure on the domestic currency to lose value, oil-exporting countries have to burn through central bank reserves as a short-term solution to defend their currencies.

Oil-exporting countries can have different exchange rate regimes. Therefore, the sort of exchange arrangement of the oil-exporting country in question will be taken into account. In this paper a conventional peg like the U.S. dollar as exchange rate anchor will be used, compared with other exchange rate arrangements. The research question of this paper will be: is there a significant difference in central bank reserves from oil-exporting countries with a pegged exchange rate to the U.S. dollar compared to other exchange rate regimes given a falling oil price? Until now there is no literature available about the direct relationship between central bank reserves and a falling oil price given a particular exchange rate regime. Only economic theory is known about the relationship between the depreciation of a currency and central bank reserves. Due to the lack of literature about this subject this research is interesting to perform. This paper will continue with a literature review in part 2, data description and summary statistics in part 3, methodology in part 4, results in part 5 and the conclusion and final remarks in part 6.

(6)

5

2 Literature review

In this section there will be a thorough background given about the relationship between the balance of payments, the current account and exchange rate regimes. After analyzing

economic theory economic literature will be discussed. Both form the theoretical background of this paper.

2.1 Economic theory

2.1.1 Balance of payments

According to Pilbeam (2013, p. 31) the balance of payments is a statistical record of all the economic transactions between residents of the reporting country and residents of the rest of the world during a given time period. It reveals how many goods and services the country has been exporting and importing and whether the country has been borrowing from or lending money to the rest of the world. Reported figures are normally in de domestic currency of the reporting country (2013, p. 32).

Pilbeam (2013, pp. 32-35) states that the main accounts of the balance of payments are the current account balance and the capital and financial account balance.

The current account balance is the sum of the visible trade balance and the invisible trade balance (2013, pp. 33-34). The visible trade balance represents the difference between receipts of exports of goods and expenditure on imports of goods which can be visibly seen crossing frontiers (2013, p. 33). The invisible balance shows the difference between revenue received for exports of services and payments made for imports of services. In addition, receipts and payments of interest, dividends and profits are recorded in the invisible balance just like unilateral receipts and payments (2013, p. 34). The capital and financial account records transactions concerning the movement of capital into and out the country (2013, p. 34). Capital inflows are, in effect, a decrease in the country’s holding of foreign assets or an increase in liabilities to foreigners (2013, p. 34). Capital outflows are, in effect, an increase in the country’s holding of foreign assets or a decrease in liabilities to foreigners (2013, p. 34). At last the settlements balance shows transactions (if any occur) undertaken by the central bank and records two key items: (i) rises and falls in foreign exchange reserves;

(7)

6

The official settlements balance equals the current account balance plus the balance on capital and financial account and the statistical error (2013, p. 41). The table of Pilbeam (2013, table 2.4 on p. 41) summarizes the key balance of payments concepts.

Table 1 Trade Balance + Exports of goods - Imports of goods = Trade balance Current Account Trade balance + Exports of services

+ Interest, dividends and profits received + Unilateral receipts

- Imports of services

- Interest, dividends and profits paid - Unilateral payments abroad = Current account balance

Basis Balance

Current account balance

+ Balance on long-term capital account = Basic balance

Official Settlements Balance

Current account balance

+ Balance on capital and financial account + Statistical error

= Settlements balance (opposite in sign to sum of above 3)

According to Pilbeam (2013, p. 37) when economists are talking about a balance of payments deficit or surplus they are really saying that a subset of items in the balance of the payments is in surplus or deficit. In the case of a fall in the oil price the trade balance will decrease which undermines the current account. For instance, The Economist wrote on December 5th 2015 that oil prices are about half what they were a year ago. In 2013 the Gulf countries had a huge combined current account surplus of 21.6 percent of GDP. But the IMF expects this to shrink to a deficit of 2.5 percent of GDP next year, thanks to the plunge in the value of their main export (2015, p. 67). A current account relative deficit means that the country as a whole is relatively spending more than it is earning and therefore reducing its net claims on the rest of the world or is increasing its indebtedness (2013, p. 37). The official settlements balance focuses on the operations that the monetary authorities have to undertake to finance any combined imbalance in the sum of the current account and financial account (2013, p. 38). Pilbeam (2013, p. 38) states that if the sum of the current and capital accounts are negative, the country can be regarded as being in deficit as this has to be financed by the authorities drawing on their reserves of foreign currency, borrowing from foreign monetary

(8)

7

authorities or the IMF. From now on, in case ‘(official) reserves’ is used, this actually implies the ‘official settlements balance’.

2.1.2 Exchange rate regimes

The International Monetary Fund (IMF) classifies exchange rate arrangements as a hard peg, soft peg, floating exchange rate and as residual (other managed arrangements) in their 2014 annual report on exchange arrangements and exchange restrictions. Within each exchange arrangement classification there are several sub-arrangements, for example within the floating exchange rate arrangement (34 percent) there are member countries that let their currencies float (18.8 percent) and member countries (15.2 percent) that intervened rarely enough to be classified as free floating. In this paper the emphasis will lay on oil-exporting countries that have a conventional peg like the U.S. dollar as exchange rate anchor and oil-exporting countries that have other exchange rate regimes.

According to the IMF (2014), 23 percent of member countries use a conventional peg as exchange rate arrangement and 34 percent of member countries apply a sort of floating exchange rate regime. Some of member countries that use a conventional peg anchor their exchange rate to the U.S. dollar; many of them are large oil-exporting countries. In a working paper of the National Bureau of Economic Research (NBER) written by Reinhart and Rogoff (2002, pp. 54-104) the historical classifications of the sort of exchange rate arrangement per country are given. This working paper will be used to classify the sort of exchange rate arrangement that significant oil-exporting countries use during a certain period of time. More about this working paper will be given in section 2.2. In case of a conventional peg, the central bank will act to keep the exchange rate unchanged when there is pressure on the domestic currency to lose or gain value. When a country faces a free floating exchange rate the central bank will not intervene on the foreign exchange market to change its currency’s value. In table 2 the percentages of the IMF’s member countries per exchange rate

arrangement in the period 2008-2014 are summarized. This table was obtained from the IMF’s annual report on exchange rate arrangements and exchange restrictions (2014, p. 19). As seen in the table, (free) floating exchange rate regimes are not that common and still a lot IMF member countries use a pegged exchange rate regime.

(9)

8

Table 2

Exchange rate arrangements, 2008-14 (percent of IMF members as of April 30)

2.1.3 Effects on official reserves

According to Pilbeam (2013, pp. 38-39) the official settlements concept of a surplus or deficit is not relevant to countries that have floating exchange rates as it is to those with fixed

exchange rates. This is because if exchange rates are left to float freely the official settlements balance will tend to zero because the central authorities neither purchase nor sell their

currency and so there will be no changes in their reserves.

The settlements concept is, however, very important under fixed exchange rates because it shows the amount of pressure on the authorities to devalue or revalue the currency (2013, p. 39). Under a fixed exchange rate system a country that is running an official settlements deficit will find that sales of its currency exceed purchases, and to avert a devaluation of the currency authorities have to sell reserves of foreign currency to purchase the home currency (2013, p. 39).

Summarizing, under a floating exchange rate the home currency is left to appreciate or depreciate and with no intervention the official reserves will be zero. Under a fixed exchange rate, however, the domestic currency will not appreciate or depreciate due to the official intervention of the monetary authorities by selling official reserves. For example, The Economist wrote on last December 5th that in October alone, Saudi Arabia’s central bank spent $7 billion of foreign reserves financing the kingdom’s deficit (2015, p. 67). This current account deficit is a result of a plunge in the value of their main export product (oil). To

maintain a pegged exchange rate to the dollar, the central bank of Saudi Arabia has to sell official reserves.

(10)

9

2.2 Economic literature

There is no economic literature available, as far as we know, about the direct relationship between a falling oil price and the official reserves from oil-exporting countries under different exchange rate regimes. However, there are some papers available that can be used for this research.

First of all there is a working paper of the National Bureau of Economic Research (NBER) written by Reinhart and Rogoff in 2002. This paper is about a reinterpretation of the modern history of exchange rate arrangements. In this working paper the country histories of exchange rate arrangements are presented. These results of Reinhart and Rogoff (2002, pp. 54-104) form the backbone of the determination of the sort of exchange rate regime per country during the last decennia. More about this working paper can be found in section 3. To support the economic theory in this section and the assumptions made in section 1 some literature is found.

Cheung and Ito (2007, p. 448) conducted an extensive empirical analysis of the determinants of international reserve holdings. From economic theory it is known that the current account has an effect on the official reserves. Besides the current account, the average propensity of import (i.e. imports-to-GDP ratio) determines, other things being equal, the level of official reserves (2007, p. 452). The imports-to-GDP ratio measures trade openness and, therefore, should have a positive effect on the demand for international reserves because of the precautionary holding to accommodate external shocks through trade channels (2007, p. 452).

Tufail and Qurat-ul-Ain (2013, p. 561) researched the role of the exchange rate for improving the current account balance of oil-exporting countries. As far as the effect of oil prices on exchange rate and current account is concerned, increase in oil price improves current account balance for all importing countries in short run and deteriorates it in long run except Bangladesh (2013, p. 561). On the other hand, all oil exporting countries experience deterioration of current account in response to oil price shock both in short and long run except Malaysia whose current account improves in long run, says Tufail and Qurat-ul-Ain (2013, p. 561). As stated by Tufail and Qurat-ul-Ain (2013, p. 561), given the results, it is recommended that oil-exporting countries should diversify their exports to overcome the recourse curse problem. The paper of Tufail and Qurat-ul-Ain confirmed that oil price shocks have a large consequence for the current account of oil-exporting countries and thus on their

(11)

10

exchange rates and reserves.

About the relationship between exchanges rate arrangements and a country’s reserves Baek and Choi (2008, p. 105) published a research in the Korean Economic Review. They concluded that the degree of exchange-rate flexibility has an inverted-U relationship with the country’s reserve holdings. Exchange rate regimes with intermediate flexibility need more reserves than polar regimes (hard pegs and freely floating). Second, reserve holdings are smaller under hard pegs than under freely floating (2008, p. 105). The empirical study of Baek and Choi is in contrast with the economic theory given in section 2.1. Pilbeam (2013, p. 39) stated that under a floating exchange rate the home currency is left to appreciate or depreciate and with no intervention the official reserves will be zero. Under a fixed exchange rate, however, the domestic currency will not appreciate or depreciate due to the official intervention of the monetary authorities by selling official reserves. In this paper the

assumptions of the economic theory will be followed, thus the results of Baek and Choi will have no effect on the way this research is conducted.

Finally, Neely (2000, p. 28) stated that changes in reserves are correlated positively with intervention activity, but may not be correlated strongly. It is difficult to say whether changes in reserves are an adequate proxy for intervention because the answer to that question may depend on the issue being researched, says Neely (2000, p. 29). That there is a certain correlation between official intervention and a country’s reserves corresponds with the assumptions and economic theory in sections 1 and 2.

(12)

11

3 Data

3.1 Data sources

A unique dataset is constructed with data from the World Bank, the IMF and

MacroTrends.net. Classifications in the exchange rate arrangements are obtained from a working paper from the National Bureau of Economic Research (NBER) written by Reinhart and Rogoff (2002, pp. 54-104).

From the databank of the World Bank the yearly total reserves minus gold (current US$) of member countries are obtained. According to the World Bank2

the total reserves minus gold comprise special drawing rights, reserves of IMF members held by the IMF, and holdings of foreign exchange under the control of monetary authorities. Besides the total reserves minus gold the total GDP (current US$) per country is collected. The World Bank states3

that GDP at purchaser's prices is the sum of gross value added by all resident producers in the economy plus any product taxes and minus any subsidies not included in the value of the products. Dollar figures for GDP are converted from domestic currencies using single year official exchange rates. At last the data of imports of goods and services (% of GDP) is

obtained from the World Bank. Imports of goods and services represent the value of all goods and other market services received from the rest of the world4

. All World Bank data is available from 1960 until 2014 but depends per country.

The International Monetary Fund (IMF) provided the data for the current account balances (in % of GDP) of the chosen oil-exporting countries. The data is from their World Economic Outlook (October 2015) and has from 1980 until now the yearly data. From economic theory and from the paper of Cheung and Ito (2007, p. 448) it is known that the current account balance and the average propensity to import (imports-to-GDP ratio) both influence the official reserves. The imports-to-GDP ratio is calculated with data from the World Bank. From the IMF the current account balances (as percentage of GDP) are obtained for the period 1981-2001.

From the site of MacroTrends the monthly WTI crude oil prices per barrel back to 1948 (in U.S. dollars) are obtained.

2_{Source: site of the World Bank – Data/Total reserves minus gold (current US$).}

3_{Source: site of the World Bank – Data/GDP (current US$).}

(13)

12

In June 2002 Reinhart and Rogoff wrote a NBER working paper about a reinterpretation of the modern history of exchange rate arrangements. In this working paper the country histories of exchange rate arrangements are presented. These results of Reinhart and Rogoff (2002, pp. 54-104) form the backbone of the determination of the sort of exchange rate regime per country during the last decennia.

To determine which countries are large oil-exporting countries data is used from the U.S. Energy Information Administration (EIA), a principal agency of the U.S. Federal Statistical System. The EIA provides this energy data and analysis since the last three decennia.

The above data sources resulted in a unique dataset from 1981 up to 2001 for fifteen significant oil-exporting countries during that period: Algeria, Australia, Canada, Colombia, Ecuador, Egypt, Indonesia, Kuwait, Malaysia, Mexico, Nigeria, Norway, Saudi Arabia, United Kingdom and Venezuela. For all these countries all the described data was available in the period of 1981 to 2001. That’s the reason why large oil-exporting countries such as Iran, Russia (Soviet Union), United Arab Emirates and Iraq are not taken into account in this research; not all significant oil-exporting countries have data available from 1981 to 2001. All these countries have different kinds of exchange rate regimes, from a pegged exchange rate to the U.S. dollar to other exchange rate arrangements.

3.2 Dataset explanation

Data is obtained from the data sources described in section 3.1 and with this collected data a unique dataset for fifteen significant oil-exporting countries for the period of 1981 to 2001 was created. With the data from the World Bank the logarithmic differences of the reserves-to-GDP ratios and the WTI crude oil prices (in US$) are calculated. The reason for this is that stationary data is necessary to conduct an Ordinary Least Squares (OLS) regression. More about the calculations to obtain stationary data can be found in appendix 1.

A dummy variable is created per year per country. The value 1 is given when an oil-exporting country has a pegged exchange rate to the U.S. dollar. The value 0 is given in case an oil-exporting country has no conventional peg to the U.S. dollar and thus does not anchors their exchange rate to the U.S. dollar. With the NBER working paper of Reinhart and Rogoff (2002, pp. 54-104) the exchange rate arrangement per country per year is determined. Only a peg to the U.S. dollar receives the value 1, soft pegs, floating exchange rates and residual exchange rate arrangements (other managed arrangements) all receive a 0.

(14)

13

In appendix 1 more can be found about the imports-to-GDP ratios and current account balances (as % of GDP). A summary of the panel dataset can be found there too.

3.3 Summary statistics

The data of fifteen significant oil-exporting countries in the period 1981-2001 resulted in the created panel dataset as seen in appendix 1. In appendix 2 the summary statistics are given per variable for fifteen significant oil-exporting countries during the period 1981-2001. As stated in appendix 1 Kuwait has large outliers in the data in 1991 and 1992. These outliers are not excluded from the summary statistics. However, when the regression model will be discussed, these outliers will be excluded from the dataset. A thorough background of how to deal with large outliers in a dataset to perform a regression will be given in appendix 3. The summary statistics of the panel dataset with Kuwait dropped from the dataset can be found in appendix 4. More background information about the regression will be given in the next section.

(15)

14

4 Methodology

4.1 OLS regression

The relationship between central bank reserves from significant oil-exporting countries and a falling oil price is determined with a regression model. Specifically, there will be an interest in the significant difference in reserves from oil-exporting countries that peg their exchange rate to the U.S. dollar compared with other exchange rate arrangements. The method of Ordinary Least Squares (OLS) will be used. The panel dataset of fifteen significant oil-exporting countries during the period 1981-2001 will form the basis of this OLS regression model. The different kinds of variables in the panel dataset that will be used for the OLS regression are summarized in appendix 2. The panel dataset is converted into time series to conduct the OLS regression.

4.2 Time series regression model with country dummies

To look at the significant differences in reserves from oil-exporting countries that use a pegged exchange rate to the U.S. dollar compared to other exchange rate arrangements given a falling oil price, country dummies of the remaining fourteen oil-exporting countries (after dropping Kuwait from the dataset) are created for the time series regression. With this regression method there will be controlled for other countries during the regression. In appendix 5 the output of the time series regression with country dummies is seen. In the output only thirteen exporting countries are seen while the observations of fourteen oil-exporting countries are taken into account. The reason for this is that in case all binary country dummies in the regression are used along with a constant, there will be perfect multicollinearity. In appendix 5 the ‘dummy variable trap’ is explained and how this is avoided in the conducted time series regression model. Below the regression model is given.

(16)

15

4.3 Hypothesis test

To test the validity of the claim that a pegged exchange rate to the U.S. dollar has a significant effect on the amount of an oil-exporting country’s reserves given a falling oil price, a

hypothesis test is conducted. A hypothesis test is used to determine the significance of the regression results. Specifically, the p-value of the dummy variable d in the time series regression model is used to evaluate if the effect of a pegged exchange rate to the U.S. dollar is significant on the dependent variable, the logarithmic differences in reserves minus gold as percentage of GDP. The null hypothesis is that the sort of exchange rate arrangement has no significant effect on reserves from an oil-exporting country given a falling oil price. The alternative hypothesis however indicates that the sort of exchange rate arrangement does have a significant effect on reserves from an oil-exporting country given a falling oil price. As a result, by stating in the alternative hypothesis that the sort of exchange rate arrangement does have a significant effect on reserves, this actually means that a pegged exchange rate to the U.S. dollar has a significant effect on reserves given a falling oil price. In the time series regression model with country dummies a significance level of 0.10 or 10 percent (α = 0.10) will be used. In appendix 6 the conducted hypothesis test can be found as the motivation for the used significance level of 10 percent.

4.4 Other regression techniques and tests

In this section more techniques will be discussed to analyze the panel dataset. The conducted OLS time series regression with country dummies will be compared to other regression methods to analyze the constructed panel data. Besides that, a few tests will be performed to evaluate the conducted time series regression with country dummies.

4.4.1 Random-effects model

In section 4.2 a time series regression model with country dummies is discussed. Specifically, with this regression the Ordinary Least Squares (OLS) method is used. To analyze panel data, two other methods are common: the random-effects model and the fixed-effects model.

To decide between the random-effects model or fixed-effects model the Hausman test is conducted. The outcome of this test is statistically not significant at a 5 percent significance

(17)

16

level thus the null hypothesis cannot be rejected. The Hausman test shows that, with the used panel data, the random-effects model is most appropriate to use. Output of the conducted Hausman test can be found in appendix 7. In section 5 the results of the random-effects model will be compared with the results of the time series regression model with country dummies.

4.4.2 Tests for heteroskedasticity

To test heteroskedasticity in the time series regression a Breusch-Pagan/Cook-Weisberg (B-P/C-W) test for heteroskedasticity is conducted. The outcome of this test is statistically not significant at a 5 percent significance level thus the null hypothesis cannot be rejected. The standard errors in the panel regression are homoscedastic thus there will be no need for robust standard errors during the regression. In the conducted time series regression with country dummies the option ‘robust’ was not applied so with the B-P/C-W test this is confirmed that robust standard errors weren’t needed. Output of the B-P/C-W test can be found in appendix 9. The White’s test for heteroskedasticity confirms the outcome of the B-P/C-W test. At a significance level of 5 percent the null hypothesis of homoscedasticity cannot be rejected. Output of the White’s test can be found in appendix 9. Both tests for heteroskedasticity confirmed that the option ‘robust’ was not needed during the conducted time series regression with country dummies. The method of homoscedastic standard errors is thus the most

(18)

17

5 Results

5.1 Time series regression results

In this section the results of the time series regression model with country dummies are discussed and interpreted. In table 3 the relevant part of the output of the time series

regression model with country dummies is summarized. This table will be used to determine the significance of the regression results. The complete output can be found in appendix 5. Below the estimated regression model is given.

𝑟𝑒𝑠𝑒𝑟𝑣𝑒𝑠_! = 0.0497793 − 0.1347466 ∗ 𝑜𝑖𝑙𝑝𝑟𝑖𝑐𝑒_!− 0.051589 ∗ 𝑑𝑢𝑚𝑚𝑦𝑣𝑎𝑟_!+ 0.0152484

∗ 𝑐𝑢𝑟𝑟𝑒𝑛𝑡𝑎𝑐𝑐!+ 0.0088321 ∗ 𝑖𝑚𝑝𝑜𝑟𝑡𝑠𝑡𝑜𝐺𝐷𝑃!

Table 3

Summarized results of the time series regression model with country dummies

reserves Coef. Std. Err. t P > | t | [ 95% Conf. Interval ]

oilprice -.1347466* .0750238 -1.80 0.074 -.2824381 .0129449

dummyvar -.051589* .0292583 -1.76 0.079 -.1091868 .0060087

currentacc .0152484** .0022699 6.72 0.000 .0107799 .0197169

importstoGDP .0088321** .0028505 3.10 0.002 .0032205 .0144437

* Significant at 10 percent significance level ** Significant at 5 percent significance level

As seen in the summarized table of the time series regression model with country dummies, the p-value of the dummy variable d is 0.079. At a significance level of 0.10 or 10 percent (α = 0.10), the outcome of the effect of the dummy variable d on the dependent variable is said to be statistically significant. This implies that there is statistical evidence against the null hypothesis and thus that the null hypothesis can be rejected. The alternative hypothesis is the one believed if the null hypothesis is concluded to be untrue.

How can the outcome of the conducted hypothesis test be interpreted? By rejecting the null hypothesis the conclusion is that, at a significance level of 10 percent, the effect of a pegged exchange rate to the U.S. dollar on an oil-exporting country’s reserves given a falling oil price is significant. There is thus a significant difference in central bank reserves from oil-exporting countries with a pegged exchange rate to the U.S. dollar compared to other

exchange rate regimes given a falling oil price.

The results of the time series regression model with country dummies are in line with economic theory. Pilbeam (2013, p. 39) states that under a fixed exchange rate system a

(19)

18

country that is running an official settlements deficit will find that sales of its currency exceed purchases, and to avert a devaluation of the currency authorities have to sell reserves of foreign currency to purchase the home currency. A falling oil price results in an official settlements deficit and thus, to remain the pegged exchange rate to the U.S. dollar, oil-exporting countries have to sell reserves.

There are also some similarities between the findings in the economic literature and in this empirical research. Cheung and Ito (2007, p. 452) stated that the imports-to-GDP ratio measures trade openness and, therefore, should have a positive effect on the demand for international reserves because of the precautionary holding to accommodate external shocks through trade channels. The results of the time series regression model with country dummies is in line with the study of Cheung and Ito; the coefficient of importstoGDP on reserves is positive. Due to the lack of literature about this subject no other literature can be linked to this empirical research.

5.2 Random-effects model

Now the effects model will be used to analyze the panel data. Output of the random-effects model can be found in appendix 8. In the random-random-effects model the p-value of the dummy variable d is 0.236. This outcome is not statistically significant at a significance level of 10 percent. With this p-value the null hypothesis cannot be rejected. In this case the conclusion is that a pegged exchange rate to the U.S. dollar has no significant effect on reserves from oil-exporting countries given a falling oil price. However, when country dummies are added into the random-effects model the p-value of d is 0.078. This outcome is significant at a significance level of 10 percent. In this case the null hypothesis can be rejected; a pegged exchange rate to the U.S. dollar has a significant effect on reserves from oil-exporting countries given a falling oil price.

Looking at the output of the time series regression with country dummies and the random-effects model with country dummies there are many similar outcomes in the

regression results. In both models the dummy variable d is significant at a significance level of 10 percent. More important, the p-value of dummy variable d has in both regression methods almost the same value and is significant at the significance level of 10 percent. It is thus not a problem that the OLS time series regression with country dummies is performed because it generates the same outcome as the random-effects model (with country dummies).

(20)

19

5.3 Robustness check

As stated by Lu and White (2014, p. 194), a common exercise in empirical studies is a ‘robustness check’, where the researcher examines how certain ‘core’ regression coefficients estimates behave when the regression specification is modified by adding or removing regressors. If the coefficients are plausible and robust, this is commonly interpreted as evidence of structural validity, says Lu and White.

Cheung and Ito stated that per capita GDP and population are potential determinants of international reserves. Per capita GDP and population are included to capture the size effect on international reserve holding, says Cheung and Ito (2007, p. 452). To make the data of these two new variables stationary, the logarithmic differences are calculated in the same way as described in appendix 1.

To evaluate the robustness of the conducted time series regression model with country dummies these two extra explanatory variables are added to this model. According to Lu and White a formal robustness test is a Hausman-style specification test (2014, p. 205). The outcome of this test is statistically not significant at a 5 percent significance level thus the null hypothesis cannot be rejected. The conducted Hausman test showed that the conducted time series regression with country dummies is better than the model with two extra explanatory variables. Output of the conducted Hausman test can be found in Appendix 10.

Now there will be another check for robustness, modifying the regression specification by removing the data of some oil-exporting countries from the panel dataset. Removing data from the panel dataset will, probably, have a negative effect on the significance of the core regression coefficients. However, if these core regression coefficients are still significant these regression coefficients are robust. The data of two oil-exporting countries is removed from the panel dataset: Australia and Ecuador. There has been chosen for these two oil-exporting countries because: (i) Australia has no pegged exchange rate to the U.S. dollar; (ii) Ecuador has mainly a pegged exchange rate to the U.S. dollar. The coefficients of the time series regression model with country dummies did not, after modifying the regression specification, changed a lot. All regression coefficients are still significant at a 10 percent significance level, currentacc and importstoGDP even at a 5 percent significance level. The regression coefficients in the time series regression model with country dummies are thus robust. Output of the time series regression model with country dummies after dropping Australia and Ecuador from the dataset can be found in appendix 11.

(21)

20

6 Conclusion

This paper empirically shows the relationship between a falling oil price and central bank reserves from oil-exporting countries under a pegged exchange rate to the U.S. dollar. Until now, there is no literature available about the relationship between central bank reserves and a falling oil price, given a particular exchange rate regime. Only economic theory is available about the relationship between exchange rate regimes and central bank reserves.

A unique panel dataset is constructed with data from the IMF, the World Bank, MacroTrends.net and the U.S. Energy Information Administration. Classifications in the exchange rate arrangements are obtained from a working paper from the National Bureau of Economic Research written by Reinhart and Rogoff. This resulted in a unique dataset from 1981 up to 2001 for, at that time, fifteen significant oil-exporting countries.

To determine if a pegged exchange rate to the U.S. dollar has a significant effect on central bank reserves from an oil-exporting country, given a falling oil price, a time series regression with country dummies is performed. In this model the effect of a pegged exchange rate to the U.S. dollar is compared to other exchange rate regimes. A hypothesis test is

conducted to determine the significance of the regression results. Specifically, the p-value of the dummy variable d in the time series regression is used to evaluate if the effect of a pegged exchange rate to the U.S. dollar is significant on the dependent variable.

This paper shows that, by rejecting the null hypothesis at a significance level of 10 percent, the effect of a pegged exchange rate to the U.S. dollar on an oil-exporting country’s reserves, given a falling oil price, is significant. There is thus a significant difference in central bank reserves from oil-exporting countries with a pegged exchange rate to the U.S. dollar, compared to other exchange rate regimes given a falling oil price. These results are in line with economic theory and literature.

One of the limitations of this research is the use of a relatively high level of

significance (10 percent), at which the null hypothesis is rejected. The chance of a type I error is thus larger than in case a more conservative significance level is used. The constructed panel dataset is very limited with only a few independent variables. In future work, more independent variables can be added into the panel dataset to avoid a type I error and omitted- variable bias. Due to above mentioned limitations of the constructed model; a more profound research is needed to investigate the subject of this paper.

(22)

21

Appendices

A

Appendix 1

With the data from the World Bank, the total reserves minus gold (in current US$) as a percentage of GDP are calculated per year per country. The total reserves minus gold as a percentage of GDP is calculated because with this reserves-to-GDP ratio there can be easily a cross-country comparison. However, the yearly reserves-to-GDP ratios are not stationary, to obtain stationary data the logarithms of these ratios are calculated. To make the data of the logarithms of the yearly reserves-to-GDP ratios more stationary the differences of the yearly logarithms of the yearly reserves-to-GDP ratios are calculated. The later one is calculated by comparing the current yearly logarithm of the reserves-to-GDP ratio with the previous yearly value.

With data from MacroTrends.net the yearly WTI crude oil prices (in US$) from 1981 to 2001 are obtained and entered per year per country. The yearly oil price is measured on every 31st of December. Like the data of the reserves-to-GDP ratios the yearly oil prices are not stationary due to large fluctuations in the oil prices. The same method is used to make this data stationary; first the logarithms of the yearly oil prices are calculated. Then the differences of the yearly logarithms are calculated through a comparison with the previous year. In graph 1 and 2 the stationary data of the reserves minus gold (as % of GDP) and the yearly WTI crude oil prices per year per country are seen.

Graph 1 Graph 2

The yearly changes (compared to the previous year per country) in the imports-to-GDP ratios are calculated. This data is stationary which is seen in graph 4. Like the imports-to-GDP ratio, the yearly changes are taken from the current account balance (as % of GDP) per year per country, see graph 3 below. The data of the current account balances (as % of GDP) yearly changes is stationary. Two large outliers are visible in the data of the current account balance (as % of GDP) yearly changes and in the data of the imports-to-GDP ratio yearly changes. The outliers are in Kuwait in 1991 and 1992, possibly a result of the Gulf War (1990-1991). A thorough background of how to deal with large outliers in a dataset to perform a regression will be given in appendix 3. In table 4 and 5 the panel dataset is summarized.

(23)

22

Graph 3 Graph 4

Table 4

Significant oil-exporting countries in the period 1981-2001

Algeria Colombia Indonesia Mexico Saudi Arabia

Australia Ecuador Kuwait Nigeria United Kingdom

Canada Egypt Malaysia Norway Venezuela

Table 5

Layout of the constructed panel dataset

Country Year Log diff.

(reserves as % of GDP)

Log diff. (WTI crude oil price) Dummy variable* CA^ balance (as % of GDP) yearly change Imports (as % of GDP) yearly change Algeria 1981 -0,029063072 -0,061258046 1 -1,043 0,53928115 .. .. .. .. .. .. .. Algeria 2001 0,177546416 -0,174643424 0 -3,803 0,66261955 .. .. .. .. .. .. .. Venezuela 2001 -0,172094162 -0,174643424 1 -8,463 1,29263456 * 1 if the exchange rate is pegged to the U.S. dollar, 0 if not

^ CA is an abbreviation of current account

B

Appendix 2

Table 6

Variables in the panel dataset used for the regression

Dependent variable Y, reserves Independent variable X1, oilprice Dummy variable d, dummyvar Control variable 1 X3, currentacc Control variable 2 X4, importstoGDP Log diff.

(reserves minus gold as % of GDP)

Log diff.

(WTI crude oil price)

1 if FX rate is pegged to the US$, 0 if not

CA balance (as % of GDP) yearly change Imports (as % of GDP) yearly change

Y is the logarithmic difference of the yearly reserves minus gold (as % of GDP), X1 the logarithmic differences of the yearly WTI crude oil prices, d the dummy variable for the U.S. dollar peg,

X3 the yearly changes in the current account balances (as % of GDP) and X4 are the yearly changes in imports (as % of GDP).

(24)

23

Table 7

Summary statistics of the used panel dataset (Kuwait included in the dataset)

C

Appendix 3

Table 8

Outliers in the panel dataset for Kuwait in 1991 and 1992

Country Year CA balance

(as % of GDP) yearly change Imports (as % of GDP) yearly change Kuwait 1991 -262,524 67,63886412 Kuwait 1992 239,918 -71,49311652

No large outliers are one of the key assumptions for the regression method of Ordinary Least Squares (OLS). As stated by Stock and Watson (2012, p. 167) large outliers can make OLS regression

misleading. If an outlier is due to a data entry error, then the error can be corrected or, if that is

impossible, drop the observation from your dataset (2012, p. 168). The outliers for Kuwait in 1991 and 1992 are however no data entry errors. After checking the data sources (IMF and World Bank) in 1991 and 1992, the conclusion is that these outliers are just extreme data points and thus have to be dropped from the dataset. The reason for these large outliers is probably the Gulf War (1990-1991) that had a large impact on the economy of Kuwait.

(25)

24

D

Appendix 4

Table 9

Summary statistics of the used panel dataset (Kuwait dropped from the dataset)

E

Appendix 5

Table 10

Output of the time series regression model with country dummies

In the output of table 10 only thirteen oil-exporting countries are seen while there are observations of fourteen oil-exporting countries taken into account. The reason for this is that in case all binary country dummies in the regression are used along with a constant, there will be perfect

(26)

25

each observation falls into one and only one category, if there is an intercept in the regression, and if all G binary variables are included as regressors, then the regression will fail because of perfect multicollinearity. This situation is called the dummy variable trap. The usual way to avoid the dummy variable trap is to exclude one of the binary variables from the multiple regression, so only G – 1 of the G binary variables are included as regressors (2012, p. 243). In this case, the coefficients on the included binary variables represent the incremental effect of being in that category, relative to the base case of the omitted category, holding constant the other regressors (2012, p. 243). By excluding Kuwait the dummy variable trap is avoided. In the output of the time series regression model with country dummies thus thirteen oil-exporting countries are seen instead of fourteen. Kuwait is here the base case, in other words the ‘reference’ country in the time series regression.

F

Appendix 6

Table 11

The conducted hypothesis test (α = 0.10)

𝑯𝒐 = 𝒕𝒉𝒆 𝒔𝒐𝒓𝒕 𝒐𝒇 𝒆𝒙𝒄𝒉𝒂𝒏𝒈𝒆 𝒓𝒂𝒕𝒆 𝒂𝒓𝒓𝒂𝒏𝒈𝒆𝒎𝒆𝒏𝒕 𝒉𝒂𝒔 𝒏𝒐 𝒔𝒊𝒈𝒏𝒊𝒇𝒊𝒄𝒂𝒏𝒕 𝒆𝒇𝒇𝒆𝒄𝒕 𝒐𝒏 𝒓𝒆𝒔𝒆𝒓𝒗𝒆𝒔 𝑯𝒂 = 𝒕𝒉𝒆 𝒔𝒐𝒓𝒕 𝒐𝒇 𝒆𝒙𝒄𝒉𝒂𝒏𝒈𝒆 𝒓𝒂𝒕𝒆 𝒂𝒓𝒓𝒂𝒏𝒈𝒆𝒎𝒆𝒏𝒕 𝒅𝒐𝒆𝒔 𝒉𝒂𝒗𝒆 𝒂 𝒔𝒊𝒈𝒏𝒊𝒇𝒄𝒂𝒏𝒕 𝒆𝒇𝒇𝒆𝒄𝒕 𝒐𝒏 𝒓𝒆𝒔𝒆𝒓𝒗𝒆𝒔;

𝒂 𝒑𝒆𝒈𝒈𝒆𝒅 𝒆𝒙𝒄𝒉𝒂𝒏𝒈𝒆 𝒓𝒂𝒕𝒆 𝒕𝒐 𝒕𝒉𝒆 𝑼. 𝑺. 𝒅𝒐𝒍𝒍𝒂𝒓 𝒉𝒂𝒔 𝒂 𝒔𝒊𝒈𝒏𝒊𝒇𝒊𝒄𝒂𝒏𝒕 𝒆𝒇𝒇𝒆𝒄𝒕 𝒐𝒏 𝒓𝒆𝒔𝒆𝒓𝒗𝒆𝒔

The relatively high level of significance (α = 0.10), at which the null hypothesis is rejected, has a disadvantage. By testing many statistical hypotheses at the 10 percent significance level, the null hypothesis will be incorrectly rejected on average once in 10 cases. Stock and Watson (2012, p. 119) stated that this is a type I error, in which the null hypothesis is rejected when in fact it is true.

Sometimes a more conservative significance level might be in order. Being conservative, in the sense of using a very low significance level, has a cost: the smaller the significance level, the larger the critical value and the more difficult it becomes to reject the null when the null is false (2012, p. 120). In other words, the lower the significance level, the lower the power of the test. Given the economic application in this paper, the power of the test is considered more important than a type I error. In this paper the, not too conservative significance level of 10 percent is seen as a reasonable compromise.

(27)

26

G

Appendix 7

The Hausman test tests whether the unique errors are correlated with the regressors. Under the null hypothesis, the random-effects model is most appropriate, while the fixed-effects model is best under the alternative hypothesis. The chi-squared outcome of this test is 1.93. This outcome is statistically not significant at a 5 percent significance level thus the null hypothesis cannot be rejected.

Table 12

The Hausman test for random-effects vs. fixed-effects model

H

Appendix 8

Table 13

(28)

27

Table 14

Output of the random-effects model with country dummies

I

Appendix 9

The Breusch-Pagan/Cook-Weisberg (B-P/C-W) test tests the null hypothesis that the error variances

are all equal versus the alternative that the error variances are a multiplicative function of one or more variables. Under the null hypothesis there are constant variances in the panel data thus there is

homoscedasticity. Under the alternative hypothesis there is heteroskedasticity. The chi-squared outcome of the B-P/C-W test is 6.89. This outcome is statistically not significant at a 5 percent significance level thus the null hypothesis cannot be rejected. This test for heteroskedasticity shows that there are constant variances in the panel dataset.

(29)

28

Table 15

Output of the Breusch-Pagan/Cook-Weisberg test for heteroskedasticity

The chi-squared outcome of the White’s test is 90.10. At a significance level of 5 percent the null hypothesis of homoskedasticity cannot be rejected.

Table 16

(30)

29

J

Appendix 10

Under the null hypothesis, the original time series regression model with country dummies is most appropriate, while the new time series regression model with country dummies with two newly added variables is best under the alternative hypothesis. The chi-squared outcome of the conducted Hausman test is 19.66. This outcome is statistically not significant at a 5 percent significance level thus the null hypothesis cannot be rejected.

Table 17

(31)

30

K

Appendix 11

Table 18

(32)

31

Bibliography

Baek, SG & Choi, C. (2008). Exchange-Rate Regimes and International Reserves.

The Korean Economic Review, 24, 105-129

Cheung, YW & Ito, H. (2007). A Cross-Country Empirical Analysis of International Reserves. International Economic Journal, 23 (4), 447-481

International Monetary Fund. (2014). Annual Report on Exchange Arrangements and

Exchange Restrictions 2014. Retrieved from https://www.imf.org/external/pubs

/nft/2014/areaers/ar2014.pdf

International Monetary Fund. (2015). World Economic Outlook October 2015.

Retrieved from http://www.imf.org/external/pubs/ft/weo/2015/02/weodata/download.a spx

Lu, X. & White, H. (2014). Robustness checks and robustness tests in applied econometrics.

Journal of Econometrics, 178 (1), 194-206.

MacroTrends. (2015). Crude Oil Price History Chart. Data retrieved from http://www. macrotrends.net/1369/crude-oil-price-history-chart

Neely, C. (2000). Are Changes in Foreign Exchange Reserves Well Correlated with Official Intervention? Federal Reserve Bank of St. Louis, 28-29

Pilbeam, K. (2013). International Finance, fourth edition. Basingstoke, United Kingdom: Palgrave Macmillan

Reinhart, C. & Rogoff, K. (2002). A Modern History of Exchange Rate Arrangements: A Reinterpretation (No. w8963). National Bureau of Economic Research, 54-104 Stock, J.H. & Watson, M.W. (2012). Introduction to Econometrics, third edition.

Harlow, United Kingdom: Pearson Education Limited.

The Economist. (2015, December 5). Finance and economics section – Keeping it riyal: Despite jitters, the oil-rich countries of the Gulf are unlikely to devalue, 67 The Economist. (2015, October 17). Finance and economics section – Pegs under

pressure: Currency pegs are still in fashion, but some are creaking, 77-78

The Economist. (2016, January 17). Finance and economics section – $20 is the new $40: Why the oil price has plunged, 73-74

The U.S. Energy Information Administration. (2015). International – Exports of Crude Oil including Lease Condensates. Retrieved from http://www.eia.gov/beta/inter national/

(33)

32

The World Bank Group. (2015). Data – GDP (current US$). Data retrieved from http:/ /data.worldbank.org/indicator/NY.GDP.MKTP.CD

The World Bank Group. (2015). Data – Imports of goods and services (% of GDP). Data retrieved from http://data.worldbank.org/indicator/NE.IMP.GNFS.ZS The World Bank Group. (2015). Data – Total reserves minus gold (current US$).

Data retrieved from http://data.worldbank.org/indicator/FI.RES.XGLD.CD Tufail, S. & Qurat-ul-Ain, S. (2013). The Effect of Oil Price Innovations on the

Dynamic Relationship between Current Account and Exchange Rate:

A cross-country analysis of the relationship between a falling oil price and central bank reserves from oil-exporting countries under a pegged exchange rate to the U.S. dollar