International Reserve Holdings by Emerging Economies: Evidence From 1995 – 2009

(1)

International Reserve Holdings by Emerging

Economies:

Evidence From 1995 – 2009

Master Thesis

NADIA N. STEFANOVA

Student number 1667246

Supervisor

Dr. D.J. Bezemer

Rijks University Groningen

Faculty of Economics and Business

(2)

Abstract

This paper combines the precautionary and mercantilist view of holding international reserves by

emerging market economies into a single specification model. Using panel data from 35 emerging

economies over period of fifteen-year, 1995-2009, I use fixed effects estimation technique to determine

the factors contributing to international reserve accumulation. I find statistical support for the

opportunity cost measure and GDP per capita robust to both heteroskedasticity and autocorrelation.

The other factors trade openness, financial development, financial openness, indebtedness and

population do not seem to be relevant in the regression. Using linear regression absorbing indicators to

estimate the fixed effects, I find statistical significance for export, opportunity cost, financial debt,

capital account openness, short-term debt and GDP per capita. No statistical support is found for

import levels, private capital flow and population.

Key words

(3)

I. Introduction

The topic of this research is the stock of international reserves held by the central banks in emerging economies. During the past decade the emerging markets central banks have accumulated unprecedented amount of foreign reserves and shifted the merits of the international financial architecture. In 1995 the world total reserves amounted to 1,389,801 million US dollars, 67 % of which was held by the advanced economies. At the end of 2010 the International Monetary Fund in its COFER database publish that the total world‟s reserve is 9,258,179 million US dollars. 1_{The striking fact is not the amount by which the total reserve has increased but the fact that now 67 % is}

held by the emerging and developing economies and only 33 % by the advanced economies. With advanced countries holding about 4% of their gross domestic products as foreign reserve and the emerging countries with sometimes over 28%, the question one might ask is what is driving that accumulation.

The importance of the topic lies in the fact that holding reserves is opposite of investment and consumption and results in decrease in aggregate demand. As the 2008 global crisis induced worldwide liquidity crunch, the countries holding two thirds of the world liquidity take on the leading role in the global recovery path by the ability to increase world demand. Instead, the emerging nations seem to accumulate even more. This phenomena troubled many economists in the past decade but no clear consensus on the topic has been reached.

The reason for central banks to hold reserves is straightforward. Central banks hold reserves in the same way banks hold reserves or firms hold inventories, e.g. as an insurance against sudden shocks and payment disturbances. In addition, central banks hold reserves to interfere in the foreign exchange market to protect/manipulate their exchange rate or to use the reserves as collateral in international borrowing. As all economists agree that central banks should hold reserves the discussion moves to the optimal amount of reserves and the factors influencing that decision. The traditional theory of international reserve holdings identifies factors like level of imports, trade and financial mercantilism influencing reserve holdings either as an attempt to provide a buffer in case of an external shock or a way to promote growth. The latest development in the literature explains reserve holdings as a function of financial development, openness and exposure to the world financial system. The main idea is that reserves are held as a precaution against turbulence in the international financial markets or as leverage and country creditability.

The focus on this paper is to provide a systematic way to answer the question of what factors are driving the reserve accumulation in the emerging markets in the last 15 years. In addition, it is important to investigate which of the different factors play the biggest role and why. At last, one has to ask what could be the implications of the international reserves studies in the current economic environment.

To analyse the factors contributing to the increase in the stock of reserves I incorporate different theories and variables into a single model. The dependent variable is the natural logarithm of the international reserves as

(5)

percentage of GDP. The explanatory variables used are combination of mercantilist and precautionary views: trade openness, opportunity cost measure, financial openness, financial development, indebtedness. As control variables I use GDP per capita and population. I use a panel data of 35 emerging economies over fifteen year period 1995 – 2009 and data obtained from the World Bank‟s World Development Indicator and Global Development Finance databases. The main contribution of this research is that it focuses only on emerging countries and includes a combination of different theories into a single specification model. The period covered is also of great importance since it includes the latest developments. Any research on the topic of emerging countries done before 1995 may not be relevant because it is a period in which the developing countries had little or no role to play in the international capital market.

(6)

II. Theoretical background on international reserves

The common definition of international reserves is the overall liquidity of a country‟s central bank. It is the stock of all assets by central bank that can be converted with certainty to another financial medium. The international reserves comprise of all convertible foreign exchange, gold, special drawing rights (SDRs), reserve position in the International Monetary Fund (IMF), long-term foreign assets, and automatic drawing rights under bilateral credit arrangements.

1. Traditional Models

International reserves can be expressed in the form of the quantity theory of money, by replacing money supply for international reserves and income by imports. In this context, most economists use the term demand for international reserves in the way they use the demand for money balances. When discussing the terms supply and demand for international reserves one has to keep in mind that the supply side is the actual global supply of liquidity, whereas the demand is a country specific and considers individual country preferences for reserves.

1.1. Supply of international reserves

The supply side of the international reserve comprises of the foreign exchange, gold and SDRs and IMF position. The collapse of the gold standard (the Bretton-Woods agreement) is probably the main factor that changed the world supply of international reserves. The US dollar became the world reserve currency followed by the pound sterling, Japanese yen, Deutsche mark, Swiss francs and since 1999 the euro. Up to date, those currencies represent the main portion of the world foreign exchange. Figure 1 shows the division of foreign exchange among industrialised countries in one group and emerging and developing countries in the other. It is clear to see that starting in 2005 the emerging economies hold two thirds of the world foreign exchange.

Figure 1: World Foreign Exchange

World Foreign Exchange

0 2000000 4000000 6000000 8000000 10000000 1995 1997 1999 2001 2003 2005 2007 2009 Year Fo re x i n m il li on s of U S do ll a rs World total Advanced EM and Developing

***Source: IMF; Currency Composition of Official Foreign Exchange Reserves (COFER) database2

(7)

Immediately, one can ask what has triggered this sharp increase in foreign exchange, both in absolute value and in division between developed and emerging economies. Before elaborating further, it is wise to show the mechanism through which countries acquire foreign exchange. Countries receive foreign currencies through two main channels; call them goods and money. The goods channel represents the export of goods and services for which a country receives foreign currencies (US dollars, euros, yens, etc.). The money channel contains foreign direct investments, portfolio investments, financial aid, external borrowing (external debt) and remittances.

Considering Figure 1, the answer of the question above is that both world supply of reserves and demand for reserves has increased rapidly since 2005, especially in the emerging markets (EMs). On the supply side the explanation may be that USA, as originator of the reserve currency, incurs large current account deficit and continuous to cover this deficit by issuing debt.

1.2. The role of the US dollar and the fall of the Bretton-Woods system

In the international monetary system countries agree on different exchange rate regimes and rules of conduct when intervening in the international capital market. Currently, most countries choose their own corner of the trilemma (choose two out of three options): enjoy monetary autonomy, free capital flow and exchange rate stability. A big part of this discussion is the inevitable necessity to hold reserves as countries need stable asset to be able to intervene in the foreign exchange market to support their currency. After the fall of the Bretton-Woods gold standard in 1973, the US dollar took the leading role as an international reserve currency and took the role of supplying the world with liquidity. The US dollar managed to stay stable due to the fact that USA had a deep financial market and free capital flows as well as very open economy. Supplying the world with liquidity, however, brings a mismatch between US monetary policy goals and the international monetary goals. This is known in the literature as the Triffin paradox or Triffin dilemma. It also applies an unequal distribution of reserves and current accounts surplus or deficits in the world as a whole, forcing surplus countries to accumulate reserves while US runs a constant deficit. In short, the US monetary factors have an unambiguous effect in explaining the increased in supply of international reserves.

1.3. Demand for international reserves

(8)

reserves and cost of adjustment. Bahmani-Oskooee and Brown (2002) argue that serving as a buffer reserves facilitate the achievement of domestic policy goals, such as employment and inflation, by allowing authorities the time to choose the desired internal or external policy measure. 3_{According to this buffer model, countries would}

hold reserves as an inventory against different payment problems that may arrive through trade or other shocks. Much of the Bretton-Woods literature era used the level of import as a measure of reserve adequacy, according to which a rule of thumb of three months worth of imports should be hold as reserves4_.

Another theory views the reserve accumulation as a by-product of mercantilist trade policies followed by the country‟s authorities. If a country decides to follow export-oriented growth path it may be reluctant to allow its undervalued currency to appreciate since that would hurt exports through the terms of trade effect. Such a country finds itself in current account surplus and needs to intervene in the foreign exchange market to neutralise the effect on the value of the exchange rate. In this view the main constrain on the reserve accumulation is the degree to which the intervention of the central bank can be sterilised (Bird and Mandilaras, 2011, p.265).

In another view, the central bank reserve demand is perceived as optimisation behaviour between incurring the cost of holding reserve and the benefit from holding them. The cost of holding reserve is the forgone opportunity of investing that amount in a higher return alternative. High opportunity cost, therefore, implies lower demand for reserves.

Aizenman and Lee (2008) introduce the term financial mercantilism to explain the large reserve holdings in East Asia. Accordingly, they explain the recent reserve accumulation as a result of East Asian growth strategies, which combined outward oriented growth with credit subsidisation for the industries involved. Another argument they use is that East Asian countries are involved in competitive hoarding leading to negative externalities in the region (follow-the-neighbour behaviour).

2. New Literature on Reserve Holdings

Main perspective on international reserve holdings applied by many economists is based on the trilemma framework (Obstfeld et al., 2008, Obstfeld, 2009, Bastourre et al., 2009, Aizenman, 2008). The trilemma is especially applicable to the emerging markets as globalisation in trade induces financial globalisation, in which reserve holdings play important role as countries choose between financial integration, exchange rate stability and monetary independence. The discussion of the fixed versus floating exchange rate in time of financial integration has been heavily documented in the recent literature for emerging market financial liberalisation. Adoption of fixed exchange rate regime implies tough fiscal discipline and reliable institutional framework. According to Obstfeld (2009) large economies under pegged rate sacrifice the shock absorption capacity of exchange rate flexibility when nominal prices and wages are sticky. It is also known that countries need to hold very high reserves to cope with large current account imbalances. In addition, the loss of monetary autonomy and inflation control also adds to the criticism of fixed exchange regimes.

3_{See appendix for full reference of the article, p.1211}

4_{The Bretton-Woods is the period after the end of The Second World War till 1971/73 characterised as a fixed exchange rate}

(9)

On the other end, emerging markets choosing floating exchange rate regimes are faced with other challenges. Most emerging economies can not borrow externally in their own currency and their debt is issued in foreign currency. In case of sharp depreciation of the home currency the value of the external debt will rise rapidly, increasing the pressure in the domestic banking system. The appreciation of the currency on the other hand usually hurts export manufacturing industries and may cause undesirable employment shifts. As a result many countries have chosen the manage-float their currencies, while focusing on inflation targeting and limiting the currency mismatch. There is no straight forward rule for the choice on the trilemma and many of the emerging economies have learned the lesson in the most painful way, namely through painful recessions and banking and currency crises.

In the aftermath of the Asian 1996/97 crisis and the later Latin American crises many economists shifted their attention to the precautionary or self insurance argument for reserve holdings. The late 90‟s crises waves have shaken the foundations of the financial system and economies in many emerging markets, characterised by sharp output drop, credit crunches and at some places full size banking crises. Those emerging economies went through painful adjustment processes on their way back to stability and growth. Different countries used different measures to deal with the problems, some attracted capital flows while other introduced currency boards (Argentina, Bulgaria)5_{. Dealing with capital accounts liberalisations and increased exposure to international financial shocks,}

the emerging economies accumulated reserves as a precaution and insurance against future sudden stops of capital flows or bank runs.

The precautionary view has monetary issues in its foundation, especially the role of the central bank in controlling the monetary base and the stock of money in the economy. The idea of the central bank controlling the money stock and through it also controlling the overall price level in the country was first introduced by Henry Thornton in 1802 in a book called “An Enquiry into the Nature and Effects of the Paper Credit of Great Britain”. Considered by many economists as the father of modern central banking, he is the first to state the importance of the role of central bank in holding reserves as a mean to meet demand for currency and as a mean to counteract an unfavourable balance of trade. The ideas of Thornton have received little attention during the gold era of Bretton-Woods. Since the abolition of the gold standard and the switch to a managed float exchange rate regime Thornton‟s ideas have gain again popularity in the latest research on the topic of international holdings.

According to Obstfeld et al. (2008), there are two main reasons for precautionary behaviour of the central banks. The first is the risk of an internal bank drain (bank run), which is a residents run from deposits to currency. The central idea is that central bank needs adequate amount of reserves to protect the national bank system from a speculative run. The second risk to the financial system is an external drain scenario, in which domestic residents run to foreign currency. The internal drain is equivalent to a banking crisis and the external drain to a currency crisis. In case both appear at the same time, the „twin crises‟ or „double-drain‟ scenario, the central bank may be forced to the role of lender of last resort. If this is the case, it would require high holdings of reserve to be able to tackle both crises.

The internal drain is scenario concerned with the financial development of the banking system or the total amount of credit circulating in the economy. The argument is that central banks need to hold reserves proportionately to the level of credit in the economy in order to prevent bank crisis in case of bad credit collection and possible bank

(10)

runs. The external drain scenario relates to the exposure of the country to the international financial system volatility. Main issue here is the degree of financial openness of the country, meaning a more open country would be more likely to be exposed to volatility and sudden stops scenarios.

In another view, Cheung and Qian (2007), Bird and Mandilaras (2010) use the so called “Mrs. Machlup‟s wardrobe theory” to provide alternative explanation for the increasing reserve hoardings. The theory basically compares the stock of international reserves to the stock of clothes in the wardrobe of Mr. Muchlup‟s wife, arguing that no matter how big the stock is the countries want to stock more each year, just as Mr. Maclup‟s wife desires more and more clothes. Put in this way, the reserve accumulation is positively related with time. Naturally, a point would be reached at which accumulating more reserves is no longer reasonable; what that point or threshold might be is highly arguable.

Bird and Mandilaras (2011) suggest an entirely different explanation for reserve holdings in East Asia. 6_{They see}

holding reserves and IMF credit facility as imperfect substitutes during payment problem times. They argue that the conditionality of the IMF credit lines is a major put-off for many countries that once used the help of IMF. Another main put-off is that countries are afraid of the possibility of dealing with IMF in the future and its consequent implication on the country‟s reputation. Holding reserves therefore provides the benefit of zero conditionality and does not destroy the country‟s reputation internationally.

Supporting the same notion, Joseph E. Stiglitz expresses his opinion on the reserve holdings of the emerging markets in a book called “Free Fall”. He argues that the East Asian countries have learned their lesson in 1997 from the intervention of IMF and chose to hold more reserves than to deal with the fund in the future. Stiglitz criticises the involvement and policy decisions of the IMF top man Alan Greenspan and argues that the involvement increased the severity of the crises in the region. Furthermore, according to him, the oil exporting countries acquire more reserves because they do not expect the price of oil to stay stable for long. A third reason according to Stiglitz is the export-oriented growth model of some economies. After many of the traditional trade restrictions were abolished in the WTO, some emerging economies moved to policy of controlling exchange rates on competitive industries, forcing them to buy dollars and sell own currency and accumulate reserves. 7

The rest of the literature overview is discussed in the economic model that follows, providing each section with the relevant theory and empirics and main hypotheses.

3. The Economic Model

This section introduces all variables used in this research by providing discussion of the relevant literature and derives all hypotheses to be later used in the analysis.

(11)

3.1. Dependent variable

There are different ways in which international reserves are measured in the literature. The prevailing one is the log of the ratio of reserves (minus gold) to GDP. Cheung and Wong (2008) compare seven different measures of international reserves for 174 countries in the period 1957-2004. They take the ratio of reserves to imports, total foreign liabilities, short-term external debts, cumulative (gross) FDI inflows, population, nominal GDP in US dollars, and money supply M2 and compare them to five structural characteristics. Those five characteristics are: a) geographic region, b) income level, 3) level of indebtedness, 4) stage of development and 5) exchange rate regime. The seven measures of reserves are found to have highest variation across geographic classification. Under stage of development category, the developing economies hold more reserves than developed ones with the exception of reserves to population measure. The pattern of international reserve ratios dependency to the structural characteristics varies across time and across the different measures of reserves. The paper also proves that gauging the adequate reserve is difficult to assess considering the different ways in which it is measured. I follow the literature in calculating the dependent variable with some minor changes, which prove not to differ significantly. I take the overall reserves including gold measure to capture to total liquidity position of a central bank. As this measure is in current US dollars I take the GDP in current US dollars in the calculation of the reserves to GDP ratio. Another manipulation is that the percentage form is used since it proves to be more useful in comparison among the countries. Therefore the dependent variable used in the regression is the natural logarithm of the reserves as percentage of GDP in current US dollars.

3.2. Explanatory variables and hypotheses

This section discusses each variable used in the model and the different measures used in the literature. Each section ends with a derived hypothesis.

3.2.1. Trade openness and reserve holdings

The increase in trade openness raises the need for foreign reserves through increasing the vulnerability to low demand from abroad (for example, sudden drop in exports). In the empirical literature in the 60‟s and 70‟s, trade openness is mostly measured as the ratio of imports to GDP. Obstfeld, Shambaugh and Taylor (2008) use the log ratio of imports to GDP as well as the log of trade (import and export) to GDP in comparing the traditional versus monetary model. Bastourre, Carrera and Ibarlucia (2009) use the standard measure of export plus import to GDP as a proxy for trade openness (not the log term). Delatte and Fouquau (2009) use the export growth rate as a measure for trade openness, calculated as a three-year moving average of the growth rate of real export, lagged two years in the regression.

(12)

level of transactions in the economy with the outside world. Therefore, higher levels of import induce proportionately higher reserve holdings.

Hypothesis 1a: The coefficient for the first trade openness measure, import to GDP, is expected to have positive sign.

Using the trade argument as a proxy for the level of country transactions, opposite effect can be expected from the level of exports. As countries receive more foreign exchange from export the need to hold reserves should decrease. On the other hand, as foreign exchange flows into the financial system of a country (certain percentage flows into the central bank reserves), the issue moves to what is to be done with it and what is actually foreign exchange used for? Naturally, it can be used to pay outstanding debt issued in foreign currency. The authorities can use it to intervene in the foreign exchange (forex) market with open market operations (OMO) or buy foreign assets and natural resources abroad. Obviously, it can be used to pay for imports. Assuming there are imperfections in the international capital market or in the institutional and political system in the country, reserves may not be used optimally and therefore may just lay idle. If that is the case the increase in the level of export would imply higher reserves. Among the emerging markets, the East Asian countries have an export-driven growth, for which export is extremely important as well as trade competitiveness. Therefore, the central banks in those countries may find it successful policy making to protect their exchange rate from appreciation. Assuming perfect capital markets and instant reserve adjustment the level of export should influence reserves through its transaction effect e.g. have negative effect on it. Since this assumption is unrealistic I leave the data to provide the answer to the direction of that relationship.

Hypothesis 1b: The coefficient for the second trade openness measure, export to GDP, is expected to be different from zero. 3.2.2. Opportunity cost and reserve levels

The opportunity cost is a highly arguable factor in the literature. Most studies measure the opportunity cost as the interest differential between a country‟s interest rate and the US Treasury rate since it is extremely difficult to find a measure for best alternative rate of return. Thus, the higher the interest rate differential the lower the reserves should be. Other studies use GDP per capita as a proxy for opportunity costs. Sula (2008) uses a quantile regression at different points of the reserves distribution in a dataset of 96 developing nations over the period 1980-1996, and finds that at the 95th_{quantile the opportunity cost measure is significant and negative. In addition, the interest rate}

elasticity of the demand for reserves is not constant and rises with higher levels of reserves. One reason for that might be that the opportunity cost of holding reserves rises at an increasing rate as reserves increase.

(13)

This being told, there is a problem with the interest rate differential as Sula (2002) mentions there might be a bias from endogeneity. There could also be a feedback effect between interest rate and reserves. On one hand, higher reserves could reduce the chance of a currency crisis and may reduce the risk premium on domestic assets that can lead to lower domestic interest rate. This relationship has, however negative, the opposite direction that I am interested in. On the other hand, very high interest rate attracts large capital inflows, which could put presure on the currency and could induce interference from the central bank to protect the currency from appreciation. That means the central bank buys foreign currency e.g. accumulates reserves and sells domestic currency to keep the exchange rate at a desired level. In such a case there should be positive relationship between the real interest rate differential and reserves. This said, according to Mundell-Fleming model this can happen if the country has a fixed exchange rate. Under floating exchange rate regime holding the LM curve constant, capital flows in depressing net export and shifting the IS curve to the left.

Ball and Reyes (2009) argue that the endogeneity between the domestic interest rate and reserves exists under fixed exchange rate regime. Any change in the currency‟s level or expected rate of depreciation would require foreign exchange sales and purchase of domestic currency, therefore changing the reserve levels. This implies changes in the domestic currency held in private hands and alters the domestic interest rate. Under flexible exchange rate the domestic or foreign interest movements influence the nominal exchange rate but not the level of reserves. Controlling for the endogeneity under fixed exchange regimes using 2-stage least squares estimation, they find the opportunity cost to be significant.

A positive interest rate differential between the domestic real interest and the US real interest rates, assuming perfect capital mobility, should encourage capital flow into the country all other things being equal. Using this logic positive interest rate differential is associated with high reserve levels. Most countries in the sample have floating or managed-floating regimes; therefore, it is safe to expect negative relationship between the levels of reserve and the opportunity cost.

Hypothesis 2: The coefficient for opportunity cost measured by the interest differential between the domestic and US real interest rates is expected to be significant and negative.

3.2.3. Financial development and reserve levels

The amount of credit circulating the economy is a measure of financial depth8_{. If the repayment of those credits}

turns out to be difficult, due to bad credit policy or other economic factors, the country‟s financial system will come under pressure. This may cause severe distress in the banking sector and cause the central bank to interfere by using reserves.

Bastourre, Carrera and Ibarlucia (2009) use the quadratic term of GDP per capita to capture the degree of financial development. They argue that low income countries are not actively participating in the international capital market and thereby are not influenced or vulnerable to capital movements. High income countries on the other hand have well developed and deep financial system and lender of last resort (LLR) options and are less exposed to capital flow disruptions. That leaves the middle income countries, the ones that are in a process of opening up and

(14)

liberalising their capital accounts and have no own LLR and therefore are exposed to sudden capital flows disruptions. Obstfeld, Shambaugh and Taylor (2008) use the money base M2 as a measure for financial depth, e.g. the log of the ratio of M2 to GDP and find it to be significant and with positive sign. All of the above can be hypothesised as follows:

Hypothesis 3: The sign on the coefficient for financial depth, measured by the log form of the percentage of M2 relative to GDP, is expected to be positive.

3.2.4. Financial openness and reserve levels

Other studies argue that the openness of the capital account, specially the amount of capital inflow relative to GDP can be a good measure of a possible sudden stop in inflows, thus, possible risk to the financial system9_{. Using}

dynamic specification of the reserve demand and the system GMM estimator, Bastourre, Carrera and Ibarlucia (2009) measure financial openness as the volume of capital inflow to GDP ratio and find it to be significant and with the positive sign. Bird and Mandilaras (2011) use measure of financial openness (KAOPEN) created by Chinn and Ito (2002), based on the IMF‟s Annual Report on Exchange Arrangements and Exchange Restrictions, which is measures the degree of capital account openness. Obstfeld, et al. (2008) measures financial openness using index developed by Sebastian Edwards (2007), scaled between 0 and 1. Delatte and Fouquau (2009) use capital liberalisation index constructed by Lane and Milese-Ferretti (2006).

One of difficulties in measuring financial openness is the quantifying of capital controls. The conventional measures may fail to capture the intensity of capital controls. In addition, capital controls can differ depending on the direction of the flow or on the type of financial transaction targeted (Chinn and Ito, 2007).

It is difficult to argue which measure is the right one. The critique of the Edwards index is that it does not show whether a country is on its way to liberalising or restricting. The index of Chinn and Ito also fails to provide any information on the intensity or the level of capital flow throughout the year or as years pass. What if a country has many restrictions on the capital account but has high level of trade?

Intuitively, the openness measure could be constructed in the way trade openness is constructed to capture the overall openness and thus vulnerability to capital movements. I choose to measure financial openness using the amount of total private flows as percentage of GDP. Private capital flows consist of net foreign direct investment and portfolio investment. This can be seen as a measure for the total flow of foreign currencies to and out of the country through the years. The measure is a good proxy for the foreign exchange available to the country, therefore, high positive value implies high reserve levels.

Hypothesis 4a: The sign of the coefficient for financial openness, measured by the total private capital flows as percentage of GDP, is expected to be positive.

The Chinn and Ito index (2009) then can be added as a de jure measure of the degree of the capital account openness. It measures the restrictions on cross-border capital movements. Theoretically, the index could range

(15)

between -4 to 4 with high values representing perfectly unrestricted capital account. 10_{Providing that the countries}

of concern are emerging economies I would expect that they have not yet liberalised their capital accounts as much as the OECD countries have. Therefore, I would expect a negative sign for the KAOPEN index.

Hypothesis 4b: The expected sign of the KAOPEN coefficient is negative. 3.2.5. Indebtedness and reserve levels

Aizenman and Lee (2007) compare the precautionary and mercantilist views in a single specification model. Of the precautionary factors they include the degree of capital account liberalisation and a dummy variable capturing the balance of payment (BoP) crises, the Mexican and the Asian crises. They include the dummy to test whether emerging economies have accumulated reserves in a reaction to the past crises and protected against future BoP crises.

Delatte and Fouquau (2009) expand Aizenman-Lee model by including log of the ratio of short term external debt to GDP. Based on the argument of Bird and Rajan (2003) that capital account growth and liberalisation have brought more balance of payment instability, Delatte and Fouguau argue that countries with large stock of short term debt and an open account hold high levels of reserves to smooth the adjustment process in case of sudden stops. They find no support for the precautionary view and find that the estimator of debt decreases or explains less reserves build-up. However, they find enough support for the mercantilist view, particularly that the elasticity between reserves and export growth increases along the capital account openness.

Obstfeld, et al. (2008) relate to the IMF rule of thumb for adequate reserve holdings based on short-term foreign-currency debt. It is also known as the Guidotti-Greenspan rule after the Argentinean policymaker Pablo Guidotti and Federal Reserve chairman Alan Greenspan. The rule emerged from the Asian crises period due to fear of „sudden stops‟ of capital inflows. Therefore, Greenspan proposed that emerging markets should have reserves to cover up to one year without foreign credit. Obstfeld, Shambaugh and Taylor (2008) found no statistical support for the short-term effect. The period after 1995 has been very interesting from empirical point of view since it allows testing old and new variables on set of countries that are liberalising their capital accounts and opening up their borders. The late 90‟s and early 2000 was a period with intense bank and currency crises through the emerging markets, main examples of which are the Asian crisis, the Russia, Bulgaria, Argentina and Mexico crises. The failure to find support for short-term indebtedness may lie in the argument that not short-term but overall level of indebtedness is the relevant factor. Long-term foreign direct investment can as easily suffer from a „sudden stop‟ scenario and triggers the same pattern of problems facing the economy. Therefore, I propose to check both measures to see which one receives support from the data.

Hypothesis 5a: The expected coefficient for the indebtedness is positive; measured as the short-term external debt stock as percentage of the GDP.

Hypotheses 5b: The coefficient for total external indebtedness, measured as the total external debt stock as percentage of gross national income, is expected to have positive sign.

(16)

3.3. Other factors influencing reserves

Gupta and Agarwal (2004) include economic size as a determinant for international reserves. The argument is that as the size of the economy increases, so does the level of transactions implying increase in the reserve levels as population and GDP grow. Many others, for example Bird and Mandilaras (2011), Obstfeld et al. (2008), use GDP per capita and population as control variables in the analysis.

Hypothesis 6a: The expected coefficient of the control variable population is positive and significant. Hypothesis 6b: The expected coefficient of the control variable GDP per capita is positive and significant.

Other variables used often in the empirical literature are exchange rate regime, current account as a fraction of GDP, trade and financial volatility etc. However due to time and data constrains I do not include those in the analysis.

3.4. Theoretical model

The above discussion can be summarised in a single specification model that is the basis for the analytical framework (equation 1):

Int. Reserves = β0 + β1 (Trade Openness1) - β1’ (Trade Openness2) - β2 (Opportunity Cost) + β3 (Fin. Development) + β4 (Fin. Openness) + β4’ (KAOPEN) + β5 (Indebtedness1) + β5’ (Indebtedness2) + β6 (Population) + β6’ (GDP per capita) + e [Eq.1]11

where the dependent variable is international reserves; β0 is the intercept of the regression. The explanatory variables are trade openness (import and export respectively), opportunity costs of holding reserves, financial development, financial openness, capital account openness (KAOPEN) and indebtedness. 12_{The control variables}

used in the model are population and GDP per capita and e is the random error term. The next section provides detailed information on the methodology used in the research.

11_{β1’ represents the level of export; β5’ is total external debt stock;}

12_{I use two measures of indebtedness: short-term debt stock and total external debt stock. Those do not appear in the same}

(17)

III. Data and Methods

3.1. Model Specification

Obstfeld, Shambaugh and Taylor (2008) use a sample of 134 countries, over 26 years (1980–2004), 2671 country-year observations, and find the log of imports to GDP to be significant and robust for both developed and emerging markets. Sula (2008) uses a quantile regression approach in a dataset of 96 developing nations over the period 1980-1996 and finds statistical support for level of import in all quantiles of the regression. Bird and Mandilaras (2011) use least squares dummy variable regression (LSDV) to analyse their model. The advantage of the LSDV is that it allows for country fixed effects and errors that are robust to cross-country correlation. Using ordinary least squares and generalized least squares (OLS and GLS) fixed effects is a common practice in the empirical papers. However, the formulation is static and implicitly assumes instantaneous adjustments of reserve to their equilibrium level when there is a shock to the reserves. This assumption may not be the most realistic one as the adjustment in reality is far from instantaneous. Bird and Mandilaras (2011) as well as Bastourre, Carrera and Ibarlucia (2009) employ general methods of moments (GMM) estimation to tackle the problems with dynamic specification and endogeneity.

Sula (2008) argues that the least squares regressions fail because of the constant elasticity assumption that all central banks have the same reserve elasticity. In reality, the central banks have different policy choice preferences and level of risk aversion. Sula therefore, uses a quantile regression approach to allow for the different elasticity. The model used here incorporates the variables discussed in mostly in the literature and explained in the previous section in a single specification model (equation 2):

Ln (Reserve%GDP) it = a + β1 ln (Import) it+ β1’ ln (Export) it - β2 (Opportunity Cost)it + β3 ln (Fin. Development)it + β4 (FinOpenness1)it + β4’(KAOPEN)it + β5 ln (Indebtedness1)it + β5’ ln (Indebtedness2) it + β6 ln (Population) it + β6’ ln (GDP per capita) it + e it [Eq.2]

where the transcript i represents an individual observation, which in this case is a single country i in time t; and β5’ represents the second measure for indebtedness, namely the total external debt stock as percentage of gross national income. KAOPEN is the Chinn and Ito (2009) index of capital account openness.

This is the general regression form, in which 35 individual country observations are used over 15 years of time (1995-2009). The total number of observations, missing values not included, includes 35 countries in 15 year period and 9 observations in each period. This number is satisfactory even in the presence of missing values.

(18)

All data is taken from two of the World Bank e.g. World Development Indicators and Global Development Finance databases. The population considered includes emerging and developing countries, since those hold two-thirds of world‟s international reserves. Of this population, the emerging markets are selected by pure choice since they are good on their way to reform their economies and liberalise their capital accounts. Those economies are the actual driver behind the vast reserve accumulation seen in the past 15 years.

It is strongly balanced panel dataset but not a random sampling procedure as it focuses on pre-chosen specific group of countries, namely the emerging market economies. The countries are taken from the Dow Jones emerging market index (Total Stock Market Index), which comprises of 35 emerging countries from different parts of the world. In alphabetical order, those countries are: Argentina, Bahrain, Brazil, Bulgaria, Chile, China, Colombia, Czech Republic, Egypt, Estonia, Hungary, India, Indonesia, Jordan, Kuwait, Latvia, Lithuania, Malaysia, Mauritius, Mexico, Morocco, Oman, Pakistan, Peru, Philippines, Poland, Qatar, Romania, Russia, Slovakia, South Africa, Sri Lanka, Thailand, Turkey, and United Arabic Emirates (UAE). 13

3.2.1. Dependent variable

The dependent variable in this research is the level of international reserves, measured as the natural logarithm of the ratio of reserves (including gold) to GDP measured both in current US dollars. 14_{As it is common in the}

literature I take the log value to diminish the effect of the outliers. When using the logarithm term one has to keep in mind that the variable should be positive for the logarithm to exist. In the regression I use the natural logarithm for all variables that have positive values. Other measures of international reserve used in the literature are reserves to import ratio, reserves to broad money (M2), and reserves to short-term debt. All data (unless otherwise mentioned), including the dependent variable, are obtained from the World Development Indicator (WDI) database of the World Bank (WB).

3.2.2. Explanatory variables

For hypothesis 1 the trade openness is split into two categories and captures the overall vulnerability to trade shocks on the economy. The first one is the level of import (Hypothesis 1a) measured as total value of imports of goods and services as percentage of the GDP. The second measure of trade openness is constructed in the similar way; total values of export of goods and services as % of GDP. Both variables are used with the log form.

In hypothesis 2 the opportunity cost is defined as the interest differential between the real interest rate in the country and the US real interest rate. This measure takes on negative values and can not be used in log form. In the third hypothesis, the financial development is measured using a proxy of the money stock M2 as percentage of GDP (in log-form). It represents a proxy for the overall credit development in the country‟s financial system. The financial openness variable featuring in the Hypothesis 4 is measured using two proxy variables: the private capital flows (total private capital flows as % of GDP) and Chinn and Ito (2009) capital account openness index. The

(19)

total private capital flows includes net foreign direct investments and portfolio investments.15_{Both measures have}

negative values and can not be used in log-form.

The level of indebtedness in Hypothesis 5a is measured by the short-term external debt stock to GDP ratio. The variable is constructed using the total short-term external debt and the GDP PPP adjusted. The reason for the purchasing power parity value is that I want a consistent measure of the value of debt and most debt is issued in foreign currency.

The other measure for indebtedness (Hypothesis 5b), the total external debt is calculated as a percentage of the gross national income.

The two measures contain information including short term debt and therefore, should enter into two different regressions. Both measures are taken in their log term.

3.2.3. Control Variables

The size of the economy‟s transactions is controlled for using two variables: population and GDP PPP per capita. Again for both firm the natural logarithm is taken. Both control variables come from the WDI database of the World Bank cited above.

3.3. Methods and assumptions

Throughout the paper the main assumption regarding multiple regressions are applied unless otherwise stated. Since the data are not randomly selected, strict sample of pre-determined countries has been taken, the way to estimate this regression is using fixed effect estimation procedure. The time-series dataset used is strongly balanced. Providing the number of individuals (countries) is 35, a least square dummy variable model (LSDV) is not convenient. An alternative way is to use the fixed effect estimation technique.

Random effects regression is also performed together with the fixed effect regression to test whether the fixed model is indeed the better choice. For this purpose a Hausman test is calculated. The Hausman test controls for the assumption (MR 5) of zero covariance between the explanatory variables and the error term {cov (xi,ei)≠0}. The idea of the test is simple: if the random and fixed estimators are consistent then there is no correlation between the error term and the independent variables. If there is a correlation between them the random estimator is no longer consistent but the fixed effects estimator remains consistent and converges to the true parameter value in large samples.

The assumption MR 3, namely equal variance (homoskedasticity), is checked using residual plots or The White test. The assumption of uncorrelated error terms and/or uncorrelated independent variable (MR 4) is also tested for autocorrelation using residual correlograms and the Lagrange Multiplier test. The statistical package provides alternative to the test by estimating the regression with heteroskedastic and autocorrelated robust standard errors. The assumption of zero correlation between explanatory variables and the error term is possibly the worst of the worries as it creates endogeneity problem for the OLS regression. Endogeneity may arise in this study because of

(20)

omitted variables, there are certainly more factors that have an effect on the dependent variable which are not specified in the model.

The fixed effects (FE) model assumes that the error terms are independent and have zero mean and constant variance. If that assumption holds the Best Linear Unbiased Estimator (BLUE) of the equation model is the least squared estimator. In this example, the assumption states that each country has its own characteristics that are independent from the other countries. Each country is different therefore the country‟s error term and constant (which captures the individual characteristics) are not correlated with the rest of the countries. Therefore, the FE eliminates the endogeneity problem assuming the country‟s error terms are not correlated.

(21)

IV. Empirical Results

This section refers to all statistical outcomes from the research. Firstly, the most important facts and charts are presented to get the feeling of the data and overview of the main issues.

4.1. Facts in Figures

The total reserve holdings (including gold) in the sample emerging countries are 23,568,721,452,334 current US dollars. The total reserves without gold are 21,669,945,467,895 current US dollars. This means that about 1.9 trillion worth of gold is held additionally to the foreign exchange. There seem to be a consistency in the level of gold kept by most countries; the stack of gold seems to stay relatively constant over time. Four countries, including Bulgaria Kuwait, The Philippines and Romania, seem to have greater gold stack comparing the two measures than the rest of the countries (Figure 2).

Figure 2: Bulgaria and Philippines Reserve Holdings % GDP Bulgaria Reserve Holdings

0 10 20 30 40 50 199519961997199819992000200120022003200420052006200720082009 Year To ta l R e s e rv e s ResGold%GDP Reserves%GDP Philippines 0 5 10 15 20 25 30 199519961997199819992000200120022003200420052006200720082009 Year To ta l R e s e rv e s ResGold%GDP Reserves%GDP

***Data is taken from World Development Indicator (WDI) database

All other countries have smaller margin between the two measures. Another valuable figure shows the total foreign exchange held by emerging and developing countries and their currency composition (Figure 3).

Figure 3: Emerging and Developing Countries Total Foreign Exchange Composition

EM Foreign Exchange 0 1000000 2000000 3000000 4000000 5000000 6000000 7000000 1995 1997 1999 2001 2003 2005 2007 2009 Year Fo re ign E x c ha ng e

Total foreign exchange holdings

Claims in US dollars Claims in pounds sterling

Claims in euros

***Source: IMF, Currency Composition of Official Foreign Exchange Reserves (COFER) database

(22)

attributed to the US current account deficit and increased government debt. On the other side of the coin is China and its current account surplus and reserve accumulation. That is shown in Figure 4, in which reserves are divided by geographical region.

Figure 4: Total Reserves per Geographic Region

Reserves per region

0 1000000 2000000 3000000 4000000 5000000 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year To ta l re s e rv e s i n m il li on s of c urr e nt U S do ll a rs Other Oil Exporters East Europe China LA and Mexico East Asia

* East Asia is taken apart form China ***Source World Bank, WDI database

4.2. Descriptive statistics

Throughout of this section the descriptive statistics from the STATA output are presented.

4.2.1. International reserve holdings

The measure for international reserves is calculated as percentage of total reserves (including gold) to GDP in current US dollars. In the regressions the natural logarithm is used. For the sample of 35 countries over 15 year period, the percentage of reserve to GDP ranges from 1.63 to 54.64 (see Table 1). The mean reserve held by the countries is 17.73 %, which is more than double of what developed countries hold. Graph 1 in the Appendix 6 shows the reserves as % of GDP for all countries during the period. While most countries exhibit relatively high constant levels of reserves, fluctuating just around 20 %, few others appear to hold excessive reserves relative to GDP. Some countries that have more than 30 % of their GDP tied up are Bulgaria (38% in 2009), China (49 % in 2009), Hungary (34%in 2009), Jordan (52.6 % in 2003), Malaysia (54.64 % 2007), Russia (36.8 % in 2007), and Thailand (52.5 % in 2009). The lowest value belongs to South Africa in 1995 (1.63 %). Turkey and the Latina American countries seem to hold the least reserves as well.

(23)

99% 51.11285 54.64723 Kurtosis 4.890138 95% 37.21099 53.22151 Skewness 1.260734 90% 30.37874 52.94804 Variance 93.65018 75% 21.87254 52.61711 Largest Std. Dev. 9.677302 50% 15.68031 Mean 17.73336 25% 10.8757 2.613965 Sum of Wgt. 519 10% 7.402742 2.064845 Obs 519 5% 5.945472 2.057982 1% 2.874288 1.628736 Percentiles Smallest ResGold%GDP

4.2.2. Trade openness

For the measure of trade openness, countries seem to range between very open and very closed. This is true for both import and export. The minimum level of trade openness is 15 percent relative to GDP, 7 % for import and 8 % for export respectively. The highest value for trade openness is 220 %of GDP, with 121 % for export and 104 % of import. Table 2 and 3 in Appendix 2 show the values for import, export and trade openness as a whole. The average import is 42.75 % of the GDP with a standard deviation of 22.5. For export the mean is 43.48 % and the standard deviation is 23.6. Export exceeds import, meaning on average the countries in the sample are net exporters. Thorough look into the data reveals two blocks of countries classified as net exporters: oil exporting nations and net trade exporters. Under the oil exporting countries fall Qatar (45%), Kuwait (40%), Oman (28%), Bahrain (26%), UAE (23%), and The Russian Federation with 20 % of current account surplus. Representing the net trade exporters, the lead is taken by Malaysia with 25 %, followed by Thailand (16%), Chile (15%), India (15%), Indonesia (11%), and China (8%).

4.2.3. Opportunity cost

The measure for opportunity cost of holding reserves is the interest rate differential (IRD) between the domestic and US real interest rates. Table 2 in Appendix 2 shows that for 466 observations the mean value of the IRD is 2.5. However, the standard deviation is 12.55, showing that the IRD varies highly among the countries and the years. Another way to see this variation is the range between the lowest value of minus 78 and the highest of 91. Both values come from Bulgaria, which experienced its „third national crisis‟ in 13 centuries history with hyperinflation and total collapse of the banking sector leading to the IMF intervention and introduction of a currency board16_.

4.2.4. Financial Development

I measure financial development with the money stock variable M2 to GDP (%). The mean value is 55.42 % and a standard deviation of about 29. This measure shows the development of the banking sector in the economy. Again the data shows that the countries have different levels of credit development ranging from M2 being 14% of GDP to 159%. In the regression the natural logarithm of M2 to GDP is taken to correct for the huge variation in the variable.

16_{Personal note: the first and second Bulgarian national crises are considered to be the World War I and II. The country}

(24)

4.2.5. Financial openness

Private capital flow

The first measure for financial openness is the total private capital flow as percentage of GDP. For 493 observations the average private capital flow is 2% with a standard deviation of 8.12. The value ranges from minus 49 to plus 50%, but they both come from Bahrain. Up till 2007 Bahrain seems to be net creditor to the world, and 2008 and 2009 attracting 43 and 50% of GDP to their country. As the variable takes negative values the log-term cannot be used.

KAOPEN

The second measure of financial openness is the KAOPEN index created by Chinn and Ito (2009), representing the degree of Capital Account openness. High values represent more open capital accounts. The mean value of KAOPEN in the sample is 0.67 with standard deviation of 1.47 for 519 observations (see Table 2). The lowest value is minus 1.84 and the most open capital account is 2.477. Those values indicate the big difference among the countries and their level of liberalisation of the capital account. Four countries exhibit constant KAOPEN indexes: China and India have constant of minus 1.14817, while Qatar and UAE have 2.477618. The result is not surprising as China and India are well known for the capital restrictions. For Qatar and UAE the case is different since their economies are extremely dependent on oil export and have little other resources meaning they need to import great amount of goods.

4.2.6. Indebtedness

Short-term debt stock

11 countries do not report any data for their short-term debt to GDP: Bahrain, Czech Republic, Estonia, Hungary, Kuwait, Latvia, Oman, Poland, Qatar, Slovak Republic, and UAE. The total number of observations is therefore 360. The mean value is 0.0315 or about 3 percent of the GDP is short-term debt. The minimum reported value is 0 and the maximum is 17.65% of GDP (0.01765).

External debt stock as % of GNI

Out of the sample 25 countries have data and 10 no data on external debt. From 374 observations the average external debt is 42.76% of gross national income. The standard deviation is 24 showing large variation in the data. The range of the external debt varies between 7 and 168, suggesting that the sample of countries contains highly indebted economies and economies that have very good finance position.

4.2.7. Control variables

Population

(25)

Therefore, in the regression the natural logarithm of the population (ln) is used. The ln of population has mean 16 and standard deviation of 1.89, which is much better to work with in the regression.

GDP per capita, PPP

The emerging economies in the sample have average GDP per capita is 11708 in current international dollars, controlled for purchasing power parity. The lowest value is 1184 and the highest is 91712 current international dollars. 17_{It is useful to compare the income levels of the Emerging and Developing nations and their total reserve}

holdings. Figure 5 compares lower middle income (LMC), middle income (MIC) and high-income non-OECD economies and their total reserve holdings. The graph shows that starting in 2005 the MIC and High-income non-OECD countries increased their reserve holding substantially. The EMs in the sample fall in the MIC and High-income non-OECD, which is another prove of the fact that they hold the biggest piece of the liquidity pie.

Figure 5: Income to Total Reserves

Income - Reserves(Gold) 0 2000000000000 4000000000000 6000000000000 8000000000000 10000000000000 12000000000000 14000000000000 1995 1997 1999 2001 2003 2005 2007 2009 Year To ta l R e s e rv e s i nc . go ld

High Income non-OECD Moddle Income Low Middle income

*LMC economies are those in which 2009 GNI per capita was between $996 and $3,945 **MIC economies are those in which 2009 GNI per capita was between $996 and $12,195

***High-income non-OECD economies are those in which 2009 GNI per capita was $12,196 or more. ****Data from World Bank, WDI database

4.1.8. Correlation matrix

Of extreme importance for the regression is whether the explanatory variables are correlated with the dependent variable and with each other. Therefore, correlation matrix is performed on the variables to test for possible correlation. Table 6 shows that reserve (as percentage to GDP) is correlated with the trade openness variables. The correlation with import is 0.59 and with export 0.6. Furthermore, the export and import are highly correlated with each other (0.92). The dependent variable is also correlated to the level of financial development, M2%GDP, with a correlation of 0.58. All other variables are not correlated with each other or the dependent variable.

Table 6: Correlation matrix

(26)

> 0.3821 -0.3983 1.0000 lnGDP 0.0814 0.1425 0.2166 0.0620 -0.2071 0.0752 0.2409 > -0.3381 1.0000 lnPop -0.1843 -0.5844 -0.4188 0.0603 0.1467 -0.2226 -0.4022 > 1.0000 lnSTD 0.2040 0.2792 0.2768 -0.0079 0.0360 0.1408 0.2965 KAOPEN 0.2169 0.2560 0.1613 -0.0155 -0.0607 0.2261 1.0000 PCF 0.1758 0.2186 0.0495 -0.0074 0.0036 1.0000 lnM2 0.5778 0.3880 0.4124 -0.0541 1.0000 IRD -0.2470 -0.3509 -0.3772 1.0000 lnExp 0.6015 0.9168 1.0000 lnImport 0.5893 1.0000 lnRgold 1.0000 > > lnSTD lnPop lnGDP

lnRgold lnImport lnExp IRD lnM2 PCF KAOPEN (obs=313)

. correlate lnRgold lnImport lnExp IRD lnM2 PCF KAOPEN lnSTD lnPop lnGDP

4.2. Model description and estimation

The fixed effects model takes the form in equation 3. Yit = β1Xit +…+ βkXkt + αi + eit [Eq.3]

where

– αi (i=1….n) is the unknown intercept for each country (n country-specific intercepts). – Yit is the dependent variable (Reserves) where i = country and t = time.

– Xit represents one independent variable (IV), – β1 is the coefficient for that IV,

– eit is the error term.

The easiest estimation technique is using the STATA option to estimate fixed effects, taken that the time and individual dimensions are correctly specified.

Regression 1: The Full Model

Regression one includes the transformed variables (the differentials) of reserves, trade openness (Import and Export), opportunity cost (IRD), financial development (M2%GDP), financial openness (private capital flow and KAOPEN), short-term debt, and the two control variables. The model is specified in equation 4.

Ln (Reserve%GDP) it = a + β1 ln (Import) it+ β1’ ln (Export) it - β2 (IRD)it + β3 ln (M2%GDP)it + β4 (PCF%GDP)it + β4’(KAOPEN)it + β5 ln (STDS/GDP)it + β6 ln (Population) it + β6’ ln (GDP per capita) it + e it [Eq.4]

This is the original model, which later would be checked for robustness.

(27)

This section shows the different techniques used to check the model robustness. The first test performed is the Hausman test for correlation between the error term and the explanatory variables. Appendix 3 (Tables 7, 8, 9) shows the STATA output for the Hausman test. The null Hypothesis H0 of no correlation is tested against the alternative H1 of correlation between the error term and the independent variables. If the Prob>chi2 is less than 0.05 the H0 is rejected. This is the case in regression 1 as the Prob>chi2 is 0.00, meaning there is correlation between the error and the explanatory variables, H0 is rejected and H1 accepted. Based on this result the fixed effects estimators are consistent and convert to the true parameter value in large sample.

Regression 1 using fixed effects

Table 10 shows the results of the regression using fixed effects. The results are not robust to heteroskedasticity or autocorrelation, therefore, the significance of the coefficients is questionable since they may not be unbiased.

Table 10: FE using xtreg command in STATA

F test that all u_i=0: F(22, 281) = 24.89 Prob > F = 0.0000 rho .80365757 (fraction of variance due to u_i)

sigma_e .25345725 sigma_u .51278246 _cons -6.196153 .7953451 -7.79 0.000 -7.761743 -4.630562 lnGDP .6385896 .0730382 8.74 0.000 .4948182 .7823609 lnPop .0562555 .0381072 1.48 0.141 -.0187562 .1312673 lnSTD .0602055 .0274615 2.19 0.029 .0061491 .1142618 KAOPEN -.0960223 .0211856 -4.53 0.000 -.1377249 -.0543198 PCF .0070656 .0047086 1.50 0.135 -.002203 .0163341 lnM2 .3961187 .1148692 3.45 0.001 .1700054 .622232 IRD -.0056943 .0015618 -3.65 0.000 -.0087687 -.0026199 lnExp .2951337 .1152584 2.56 0.011 .0682541 .5220132 lnImport .0406768 .1651188 0.25 0.806 -.2843499 .3657036 lnRgold Coef. Std. Err. t P>|t| [95% Conf. Interval] corr(u_i, Xb) = -0.4150 Prob > F = 0.0000 F(9,281) = 37.53 overall = 0.2583 max = 15 between = 0.1857 avg = 13.6 R-sq: within = 0.5459 Obs per group: min = 2 Group variable: country Number of groups = 23 Fixed-effects (within) regression Number of obs = 313 . xtreg lnRgold lnImport lnExp IRD lnM2 PCF KAOPEN lnSTD lnPop lnGDP, fe

Regression 1 using FE, robust

Table 11 in Appendix 3 shows the regression results using robust standard errors but without the variable import. Import is dropped as it is highly correlated with export but performs far worse. Table 12 shows the results performing the same regression but with standard errors that are robust to heteroskedasticity and autocorrelation (HAC errors). Using the HAC standard errors significantly reduces the coefficients performance. Without corrections, export, interest rate differential, M2% GDP, KAOPEN, short-term debt and GDP per capita are all significant. Using HAC errors, the only significant coefficients are the interest rate differential and GDP per capita, both with the expected signs. The only difference between Table 11 and 12 is that dropping import makes the coefficient for export significant and positive (0.3158). The coefficients of the other two significant variables remain almost the same with slight reduction in the t-value comparing Table 11 and 12. The interest rate differential decreases t-value drops from -2.8 to -2.78, and for the GDP per capita, from 4.2 to 4.13.

International Reserve Holdings by Emerging Economies: Evidence From 1995 – 2009