Accounting for emerging market countries' international reserves: Are Pacific Rim countries different?

Accepted Manuscript Accounting for Emerging Market Countries’ International Reserves: Are Pacific Rim Countries Different? Atish R. Ghosh , Jonathan D. Ostry , Charalambos G. Tsangarides PII:

S0261-5606(14)00083-7

DOI:

10.1016/j.jimonfin.2014.05.006

Reference:

JIMF 1428

To appear in:

Journal of International Money and Finance

Please cite this article as: Ghosh, A.R., Ostry, J.D., Tsangarides, C.G., Accounting for Emerging Market Countries’ International Reserves: Are Pacific Rim Countries Different?, Journal of International Money and Finance (2014), doi: 10.1016/j.jimonfin.2014.05.006. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

April 17, 2014

M AN U

Abstract

SC

Atish R. Ghosh* Jonathan D. Ostry Charalambos G. Tsangarides

RI PT

Accounting for Emerging Market Countries’ International Reserves: Are Pacific Rim Countries Different?

EP

TE D

Popular perception is that emerging market economies (EMEs), and Asian Pacific Rim countries— China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam (RIMs)—in particular, have been rapidly accumulating reserves, perhaps beyond what is justified by precautionary motives. This paper compares and contrasts the determinants of the demand for international reserves in the RIM countries with other EMEs over the last three decades, based on current and capital account risks, mercantilism, and other motives. Our findings suggest shifting motives for holding reserves from insurance against current account shocks, insurance against capital account shocks, and as the by-product of possible mercantilism. We also find some differences between country groups: RIM countries tend to hold more reserves against current account vulnerabilities and fewer reserves against capital account vulnerabilities, but more reserves overall. There is also greater evidence of mercantilist motives being at play for RIM countries, though this motive accounts for only a small fraction of the rise in reserve holdings in recent years, peaking in 2004 and declining thereafter.

AC C

JEL classification: E58, F15, F31, and F43. Keywords: International Reserves, Precautionary Demand, Mercantilism, Quantile Regression.

_____________________ * Corresponding author. Tel.: +1 202 623 6288; fax: +1 202 589 6288; [email protected]

ACCEPTED MANUSCRIPT 2 1. Introduction

SC

RI PT

Over the past few decades, despite somewhat greater exchange rate flexibility, and some draw down during the global financial crisis, emerging market economies (EMEs) have been accumulating large stocks of international reserves. Reserve holdings, which averaged about 5% of GDP in the 1980s, have been doubling every decade since, reaching some 25% of GDP by 2010. While foreign exchange reserves may provide useful insurance in the face of current or capital account shocks, there is often a perception that reserves are not being accumulated for precautionary purposes but rather as the by-product of deliberate mercantilism. Beyond the possibility of unfair trade practices and exchange rate manipulation, such mercantilism may perpetuate global imbalances and ultimately undermine the stability of the international monetary system (Ghosh et al., 2010).

TE D

M AN U

Asian countries, notably those on the Pacific Rim (RIM), are often singled out for purportedly pursing export-led growth strategies by keeping their currencies undervalued, in turn resulting in excessive reserve accumulation. For instance, writing in 2010, Paul Krugman claimed Today [2010], China is adding more than $30 billion a month to its $2.4 trillion hoard of reserves. This is the most distortionary exchange rate policy any major nation has ever followed.1 And as recently as 2013, Dani Rodrik has noted that: Although China phased out many of its explicit export subsidies… mercantilism’s support system remains largely in place. In particular, the government has managed the exchange rate to maintain manufacturers’ profitability, resulting in a sizable trade surplus.2 But does this view have any merit? In this paper, we take up that question with specific reference to the Asian Pacific Rim countries— China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam.

AC C

EP

There are potentially two approaches to addressing this question. The first is to try to establish an absolute norm for how much reserves are required for precautionary purposes, with any further accumulation that is undertaken in the context of an undervalued currency being deemed “mercantilism”. The main difficulty with this approach is that “how much is enough” is an evolving standard: what was sufficient for current account shocks in the 1980s was clearly inadequate for the EME capital account crises of the 1990s. Likewise, prior to the global financial crisis, many commentators thought that Russia’s US$600 billion of reserves were ample, and perhaps even excessive, yet that view was rapidly revised when the central bank spent more than one-third of its reserve stock in the space of a couple of months. Moreover, existing reserve adequacy metrics generally have wide margins—the recently developed IMF methodology for 1

Paul Krugman (14 March 2010). “Taking on China and its currency” http://www.nytimes.com/2010/03/15/opinion/15krugman.html. 2 Dani Rodrik 2013, The New Mercantilist challenge, http://www.project-syndicate.org/commentary/the-return-ofmercantilism-by-dani-rodrik.

ACCEPTED MANUSCRIPT 3 assessing reserve adequacy, for instance, suggests a range between 100% and 150% of its Reserve Adequacy Metric as being appropriate, but for the typical EME, that range translates into 10% of GDP.

SC

RI PT

In this paper, therefore, we adopt a more pragmatic tack. Rather than try to assess whether Asian RIM countries are stockpiling excessive reserves against some absolute standard, we compare the behavior of RIM countries to that of other emerging market economies. This allows for evolving notions of reserve adequacy while still assessing whether RIM countries are exceptional in their reserves accumulation behavior. Specifically, we seek to determine whether RIM countries hold more reserves than non-RIM countries, controlling for various current and capital account vulnerabilities; whether they react differently in terms of their reserve holdings to such vulnerabilities; and whether a larger proportion of their reserve accumulation over the past twenty years can be accounted for by mercantilist motives.

EP

TE D

M AN U

Building on our earlier work (Ghosh, Ostry, and Tsangarides (2012), henceforth GOT), we define mercantilism as reserve accumulation that is unrelated to current or capital account vulnerabilities and that takes place in the context of an undervalued currency. Our empirical strategy proceeds in three steps. We begin by estimating a reserve demand function for a large sample of (RIM and non-RIM) EMEs over the period 1980–2010 relating reserve holdings (expressed in percent of GDP) to precautionary motives (current and capital account vulnerabilities) and mercantilism. Our estimates allow for the possibility that the motives for reserve accumulation differ over time, or according to the level of reserve holdings. Our first test is simply whether RIM countries hold significantly more reserves than non-RIM countries, controlling for their current and capital account vulnerabilities. Next, we allow for the possibility that RIM countries’ reserve holdings react systematically differently to the various vulnerabilities (as well as to currency undervaluation and the other determinants) by introducing RIM interaction terms. Finally, we decompose countries’ reserve accumulation since 1990 into the various motives to examine their relative importance in the RIM and non-RIM samples.

AC C

Our findings may be summarized briefly. First, for the EME sample as whole, the analysis suggest shifting motives for reserve accumulation—in the 1980s (and at low levels of reserves), it was mostly insurance against current account shocks; in the 1990s (especially post-Asian crisis), insurance against capital account shocks became more important; in the 2000s (and at higher levels of reserves), there is greater evidence of mercantilism. Second, there is some evidence that RIM countries hold more reserves than would be expected on the basis of their current and capital account vulnerabilities. Allowing for different responses across country groups suggests that, while RIM countries hold more reserves in general, they hold fewer reserves against capital account shocks—the exception being the period immediately following the Asian crisis—but hold more reserves against current account vulnerabilities. Third, decomposing the factors behind the buildup of EME reserves since the 1ate 1990s suggests that (the by-product of) mercantilism has played a more important role in the RIM sample than in the non-RIM sample. Finally, our results survive a series of robustness tests including model specification, period of analysis, alternative proxies for

ACCEPTED MANUSCRIPT 4 the mercantilist motive, and address potential reverse causality.

SC

RI PT

Overall, we conclude that there is some merit to the view that reserve accumulation by RIM countries has been different from other EMEs, including because mercantilist motives have been at play. Yet it is important not to exaggerate the importance of mercantilist motives in accounting for the greater reserve accumulation. Our estimates imply that, at its peak (in 2004), mercantilist motives accounted for an average of 3½% of GDP of the stock of reserves accumulated by RIM countries (in other words, absent such mercantilism, RIM countries’ average reserve holdings in 2004 would have been 3½% of GDP lower). Moreover, this contribution to the stock of reserves subsequently shrank, to around 2.3% of GDP by 2007, remaining roughly constant thereafter. Hence, we find little evidence that RIM countries on average continue to accumulate reserves due to mercantilist motives.

2. Empirical strategy 2.1. Determinants of demand for reserves

M AN U

The rest of the paper is organized as follows. Section II reviews the literature and lays out our empirical strategy. Section III presents our main OLS results. Section IV turns to quantile regressions. Section V concludes.

AC C

EP

TE D

Traditional explanations—and policy prescriptions—for holding foreign exchange reserves centered on the “insurance” that reserves provide. The original “three months of imports” rule, for instance, was designed to help insulate developing countries against current account shocks: shortfalls in export earnings or domestic shocks—such as natural disasters—that might necessitate exceptionally large imports. As developing and emerging market countries became more financially integrated, insulation against capital account shocks gained importance—as amply demonstrated by the emerging market crises of the 1990s. By providing confidence to investors, reserves can reduce the likelihood of a sudden stop and rush for the exit or—if they nevertheless materialize—cushion the economy from their consequences (Ben-Bassat and Gottlieb (1992)). Rather than three months of imports, insurance against capital account shocks calls for reserves to be held against short-term debt (or other skittish foreign liabilities) and possibly against M2, in case domestic investors lose confidence in the banking system and the currency—as happened during the Mexican (1994) and Indonesian (1997) crises—and rush for the exit as well. While in principle reserve demand should be a function of the exchange rate regime, in practice few EMEs can be truly indifferent to the level of the exchange rate, and most studies find few differences in reserve holdings according to the country’s exchange rate regime.3

3

If the central bank was truly indifferent to what happened to the exchange rate, it would not need to hold any reserves. But even EMEs with more flexible exchange rates may want to insure against extreme events and hold reserves accordingly.

ACCEPTED MANUSCRIPT 5

SC

RI PT

Recent cost-benefit calibration models cast reserve accumulation as an explicit optimization problem, with reserves chosen to provide the optimal insurance against a sudden drop in consumption given risk aversion and the costs of holding reserves (see Caballero and Panageas (2008), Jeanne (2007), and Jeanne and Rancière (2006)). Empirical contributions confirm the importance of precautionary determinants in explaining reserve holdings in EMEs (see, for example, Bastourre, Carrera and Ibarlucia (2009), Obstfeld, Shambaugh and Taylor (2010), Cheung and Qian (2009), and de Beaufort Wijnholds and Kapteyn (2001)). For East Asia in particular, Aizenman and Lee (2007) and Aizenman and Marion (2003) provide some evidence for precautionary motives in post-Asian crisis reserve accumulation.4 Cheung and Ito (2009) find that determinants of reserve demand are significantly different between developed and developing economies and that it changes over time, with the former holding less reserves that the latter (particularly in the most recent period).5

TE D

M AN U

A quite different explanation for observed reserve holdings is that they are the unintentional byproduct of modern mercantilism. In this view, emerging market economies—perhaps following the examples of Europe and Japan during the Bretton Woods period (Dooley et al., 2005)—often seek to maintain deliberately undervalued currencies through foreign exchange intervention (often supplemented by controls on capital inflows) in order to promote exports as part of an export-led growth strategy. Ghosh and Kim (2008) consider an economy in which the government has an incentive to maintain an undervalued exchange rate because it is equivalent to an export subsidy (the cost of sterilized intervention being the analog of the subsidy cost) in an economy where there are positive productivity spillovers, external to the firm, of output in the tradable sector. Aizenman and Lee (2008) show that, in a two-country game, such mercantilism can lead to the inefficient accumulation of reserves as each country engages in beggar-thy-neighbor competitive depreciations.

4

AC C

EP

Since it is not possible to observe the intentions of the monetary authorities directly, it is very difficult to establish whether reserves are being accumulated as a by-product of mercantilism or for some other reason. It is quite possible, for instance, that the central bank is trying to maintain an undervalued currency not to promote exports or gain some unfair trade advantage, but because it needs to accumulate reserves for precautionary purposes.6 To address this fundamental identification problem, here we follow GOT to define the mercantilist motive as reserve accumulation that is uncorrelated with proxies of current or capital account vulnerabilities, and that Aizenman and Marion (2003) use the volatility of export receipts as their measure of current account shocks, and implicitly include external debt and broad money in their analysis, but do not account for possible mercantilism. Aizenman and Lee (2007) and Delatte and Fouquau (2012) try to capture possible mercantilist motives, but neither includes banking system liabilities, and both are constrained to rather crude PPP-based measures of undervaluation, which they do not find to be robust determinants of reserves. 5 They also find limited evidence that East Asian economies (including China and Japan) are accumulating an excessive amount of international reserves. 6 Of course the current account is not the only way that the central bank can accumulate reserves: it can also do so by borrowing through the capital account.

ACCEPTED MANUSCRIPT 6 takes place in the context of a substantially undervalued (an estimated misalignment of at least 10%) currency. 2.2. Empirical specification

RI PT

2.2.1. Full sample and sub-periods For our baseline regression, we follow GOT to include various proxies of current and capital account vulnerabilities as well as currency undervaluation (to capture the mercantilist motive), augmenting their specification to differentiate between RIM and non-RIM countries: ln Rit = β 0 + βCACAit + β KA KAit + β M Mercantilistit + β R Regimeit + β O Otherit

where

CAit = {ln Mit , XVolit ,VolExtDemit } , KAit = {ln BMit , KOpenit , Debtit } , Regimeit = { NeerVolit , Pegit } , Mercantilistit = {Undervalit } , and

TE D

Otherit = {OppCostit ,ln GDPpcit ,ln Popit } .

M AN U

+ β RRIM regimeit × RIM it + β ORIM Otherit × RIM it + ε it

SC

+ β RIM RIM it + βCARIM CAit × RIM it + β KARIM KAit × RIM it + β MRIM Mercantilistit × RIM it (1)

lnRit is the natural log of reserves to GDP for country i at time t valued expressed as a ratio to GDP;

-

CAit includes proxies for current account precautionary motives, namely, the log of imports to GDP, the volatility of exports (measured as three-year standard deviation), and the volatility of external demand captured by the trading partners’ growth volatility (scaled by exports). Increases in the volatility of external demand are expected to result in more precautionary reserve accumulation as insulation from exogenous (partner) demand, while reserve accumulation is expected to be positively related to imports to GDP and export volatility;

-

KAit includes the log of M2 to GDP; a measure of the de jure openness of the capital account— the Chinn-Ito index, which is based on the IMF Annual Report on Exchange Arrangements and Restrictions; and short-term debt to GDP. All financial variables are expected to be positively related to reserve accumulation;

-

Regimeit includes the volatility of the nominal effective exchange rate, and a dummy variable for de facto pegged exchange rate regime. Both coefficients are expected to be negative;

AC C

EP

-

ACCEPTED MANUSCRIPT 7 Mercantilist is proxy for exchange rate undervaluation, as defined in GOT. Countries with undervalued exchange rates are expected to hold more reserves (in the target country’s currency) needed to maintain/reduce the value of their exchange rate; and

-

Other includes a proxy for the opportunity cost of holding reserves (measured as interest rate differential with the US), and two scaling variables, the log of population, and the log of real GDP per capita at purchasing power parity. Reserve accumulation is expected to be positively related to the scale variables, and negatively related to the opportunity cost. Finally, ε is a random error term.7

RI PT

-

M AN U

2.2.2. Quantile regressions and data

SC

To examine how the estimates obtained for (1) change when the sample is divided in several sub periods, namely (i) the 1980s, 1990–1997, 1998–2010; (ii) 1980–1997, and 1998–2010; and (iii) 1980–1997, 1998–2003, and 2004–2010. The cutoffs are chosen to coincide with the Asian crisis, and the post 2000 surge in reserves.

TE D

As discussed above, motives for holding reserves may have shifted over time, particularly as EMEs have become more financially integrated. Different motives may also apply at a given point in time, but at different points along the sample distribution of reserve holdings. For example, countries that hold low levels of reserves may do so because they are not very financially integrated and are mostly concerned about current account rather than capital account shocks. Conditional quantile functions may offer a more complete picture of the effect of the reserve demand determinants covariates on the location, scale and shape of the distribution of the reserve accumulation.8 Using quantile regressions, we allow the elasticities to vary across the various quantiles (q) of reserve accumulation. For a quantile q, equation (2) is modified as follows: q q ln Rit = β0q + βCA CAit + β KA KAit + β Mq Mercantilistit + β Rq regimeit + βOq Otherit

EP

q q q q + β RIM RIM it + βCARIM CAit × RIM it + β KARIM KAit × RIM it + β MRIM Mercantilistit × RIM it (2) q q regimeit × RIM it + βORIM Otherit × RIM it +ν it + β RRIM

7

AC C

For various values of q, we obtain estimated intercepts and slopes and conduct tests to examine differences across quantiles.

We do not include country fixed effects in equation (1) as that would imply identifying the effect of the variables solely through their time variation (and it is not very informative economically). However, we cluster standard errors at the country level. 8 Quantile regression techniques developed by Koenker and Bassett (1978) and Koenker and Hallock (2001). Quantile regressions allow the estimation of conditional quantile functions—models in which various quantiles (or percentiles) of the conditional distribution of reserve accumulation are expressed as functions of the determinants. While classical linear regression methods are based on minimizing sums of squared residuals and estimate models for conditional mean functions, quantile regression methods are based on minimizing asymmetrically weighted absolute residuals and can estimate conditional functions at any part of the dependent variable’s distribution.

ACCEPTED MANUSCRIPT 8

RI PT

Following GOT, we construct several estimates for misalignments based on application of three equilibrium exchange rate methodologies, namely, the macro balance (MB), the equilibrium real effective rate (ERER), and external sustainability (ES). We begin by constructing misalignment estimates based on each of the three methodologies and then combine estimates from the three methods to construct the average and median misalignments. Median misalignments are then translated into three-way classifications for aligned, overvalued and undervalued exchange rates. Given the uncertainty surrounding estimates, we consider percentage deviations of [-10, 10] as aligned, less than -10 (more than +10) undervalued (overvalued). Appendix A in GOT describes the methodologies used and the construction of each of the misalignment estimates in detail.

3. The story in simple averages

M AN U

SC

The remaining variables are constructed from the IMF’s WEO, IFS, and INS databases. The exchange rate regime classifications are derived from the IMF’s revised classification published in the 2009 Annual Report on Exchange Rate Arrangements and Exchange Restrictions. Using the IMF’s de facto exchange rate regime classification at the end of each period of the analysis, we classify exchange rate regimes into fixed and non-fixed, as well as separating the pegs into hard and soft pegs. Tables A1 and A2 in Appendix A provide details on the sample, the variables used, and summary statistics.

AC C

EP

TE D

Popular perception is that EMEs, and RIM countries in particular, have been rapidly stockpiling reserves, perhaps beyond what is justified by precautionary motives. A first glance at the data indeed shows not only that EMEs have increased their reserve holdings (in relation to GDP) markedly over the past three decades, but also that RIM countries have, on average, held more reserves (in relation to their GDP) than their non-RIM counterparts (Fig. 1). Over the full sample period, 1980–2010, RIM countries’ reserve holdings averaged 17% of GDP compared to 12% of GDP for non-RIM countries. By the eve of the global financial crisis, this difference was even more pronounced: at end-2007 reserves amounted to 32% of GDP in RIM countries and 18% of GDP in the non-RIM sample. Subsequently, however, while EME reserves have more than recovered from their use during the global crisis, the difference between RIM and non-RIM countries has narrowed to 9.7% of GDP. [Insert Fig. 1 here]

Why might RIM countries hold more reserves? As discussed above, traditional explanations would be that they face larger current or capital account vulnerabilities, and therefore have greater country insurance needs. Again, a first glance at the data suggests this explanation might have merit: over the full sample and various sub-periods (1980–97; 98–2004; 2005–10), RIM countries exhibit greater current and capital account vulnerabilities (Figs. 2). For instance, imports average 35% of GDP in RIM sample compared to less than 30% of GDP in the non-RIM sample, with the difference between them rising to some 15% of GDP by 2007. Although RIM countries tend to have lower short-term debt (especially in the latter part of the sample), their banking system

ACCEPTED MANUSCRIPT 9 monetary liabilities (M2/GDP) are almost twice as large as those in non-RIM countries. But the mercantilist story also gets some (potential) support. RIM countries’ are, on average, 7% more undervalued than non-RIM countries during the whole period of analysis, and about 22% more undervalued during the 2000s.9

RI PT

[Insert Fig. 2 here]

SC

Looking across the distribution of reserve holdings suggests that the relationships may be different at various levels of reserves (Fig. 3). Median reserve holdings have been increasing much more rapidly in RIMs compared to non-RIMs—in fact, the bottom 25th percentile of the RIM countries almost exceeds the top 75th percentile of the non-RIM countries. Unlike non-RIMs, the dispersion in reserve holdings across RIMs has also risen, with the difference between the top and bottom quartiles widening from 13% of GDP in 1980–1997 to 20% of GDP in 2005–10.

M AN U

Overall, there is considerable time series and cross sectional variation in reserve holdings between RIM and non-RIM countries, as well as within the distribution of each sample. These results suggest that reserve demand determinants may be different between RIM and non-RIM countries, a point that reinforces the motivation of our paper. In addition, while the average behavior of RIM versus non-RIM countries is certainly informative, there may be a richer story taking account of differences between these groups of countries at various points in the reserves holdings distribution that bears examination.

4. Empirical results

TE D

[Insert Fig. 3 here]

4.1. Period analysis of reserve demand

AC C

EP

We estimate our full model (1) in steps, adding groups of variables (scale, regime, opportunity cost, current account, capital account, and mercantilist) sequentially. For the moment, we allow RIM and non-RIM countries to differ only in the intercept term by the inclusion of a dummy variable for RIM countries (Table 1). The regressions are estimated on centered variables so the exponentiated value of the constant gives the average reserve holdings for the non-RIM sample.10 [Insert Table 1 here]

9

These are the estimates of exchange rate misalignment in percent. Given the large margins surrounding these point estimates, in the empirical work below, we use a dummy variable for (more than) 10% undervaluation. 10 For example, a specification with only the scale variables (log(per capita income) and log(population)) included implies that average reserves holdings by non-RIM countries were exp(-2.453)=8.6% of GDP, compared to 20.4 (=exp(-2.453+0.864)) % of GDP in RIM countries.

ACCEPTED MANUSCRIPT 10

M AN U

SC

RI PT

Adding explanatory variables in sequence suggests that exchange rate regime, current and capital account precautionary variables, and mercantilist motives are important in explaining reserve accumulation in the broader sample of RIM and non-RIM countries. The full model including all explanatory variables is given in Table 1[7]. The model explains about 56% of the variation in reserve holdings (without the inclusion of annual or country fixed effects), with both current account and capital account precautionary motives as well as mercantilist motives statistically significant and with the expected signs. Countries with higher per capita income hold more reserves; although the fixed exchange rate regime dummy is insignificant, countries with more flexible exchange rate regimes (as measured by the volatility of the nominal effective exchange rate) do hold fewer reserves. All of the current account vulnerability variables are significant: a one standard deviation increase in the imports-to-GDP ratio increases reserve holdings by roughly 4%, while a similar one standard deviation increase in the volatility of exports or the volatility of partner country growth increases reserve holdings by roughly one quarter percent of GDP. Turning to capital account vulnerabilities, short-term debt is insignificant, but a one-standard deviation increase in broad money-to-GDP or financial openness, is associated with 1.5% of GDP higher reserves. Other things equal, reserves of a country with an undervalued currency are 2% of GDP greater than those of a country whose currency is not undervalued.

EP

TE D

Looking at the sub-periods (Table 1 [8-10]), suggests shifting motives for reserve holdings by EMEs. In the 1980s, it was insurance against current account shocks was the predominant motivation for holding reserves with capital account motives almost non-existent. Following the capital account crises of the 1990s, insurance against capital account shocks became more important. These effects are large: a one standard deviation increase in the short-term debt-to-GDP ratio (corresponding to an increase in short-term debt of 5% of GDP) is associated with about 1% of GDP higher reserves. As of the late 1990s, there is also greater evidence of mercantilism with the undervaluation dummy turning economically and statistically significant: other things equal, reserves of a country with an undervalued currency are 4-5% of GDP greater than those of a country whose currency is not undervalued. 4.1.1. Are RIM countries different?

AC C

Turning to our main question of interest, the regressions in Table 1 include a dummy variable for RIM countries. In specifications [1]-[3], which include only the “scale” variables (per capita income and population), the RIM dummy is statistically significant, and indicates that RIM countries hold some 10% to 12% of GDP more reserves than corresponding non-RIM countries. Once the current and capital account vulnerabilities are included in the regression (Table 1 [4]-[7]), the RIM dummy turns insignificant—suggesting that RIM countries are no different controlling for their vulnerabilities. While that may be partly true, it is also possible that multi-collinearity between the various explanatory variables is masking statistically significant differences. It is noteworthy, for instance, that population—which is insignificant in specifications [1]-[3]—suddenly turns positive and

ACCEPTED MANUSCRIPT 11

RI PT

statistically significant. To address this concern, Table 2 reports the coefficients of a similar regression, but where the various proxies of scale, current account vulnerabilities, and capital account vulnerabilities are replaced by their respective first principal components. In the full model (Table 2 [7]), the current account, capital account, and undervaluation proxies are all statistically significant. Moreover, the RIM dummy retains its statistical significance and implies that controlling for current and capital account vulnerabilities, RIM countries hold 4.5% of GDP more reserves than non-RIM countries (Table 2 [5]). Adding our proxy for mercantilism, the difference between RIM and non-RIM countries shrinks to 3.2% of GDP but remains statistically significant. Across sub-samples, the RIM dummy is significant until 2004, but turns insignificant for the period 2005-10.

SC

[Insert Table 2 here]

M AN U

The results are similar, but starker, for China. The China dummy is significant even in the full model with multiple proxies current and capital account vulnerabilities and mercantilism. The estimates imply that China holds 11% of GDP more reserves than would be expected on the basis of its current and capital account vulnerabilities (Table 2 [11]). The results for the RIM sample are not being driven by the inclusion of China within the group, however: excluding China leaves the RIM dummy statistically significant and of almost the same magnitude.

TE D

Overall, therefore, controlling for precautionary motives, there is a statistically significant difference between the reserve holdings of RIM and non-RIM countries, part—but only part—of which is explained by greater mercantilist behavior on the part of RIM countries. 4.1.2. Do RIM countries respond differently to vulnerabilities?

AC C

EP

The analysis above imposes the same slope coefficients for RIM countries as non-RIM countries. One reason for the statistically significant RIM dummy may be that RIM countries respond differently to the current and capital account vulnerabilities. For example, if RIM countries were more risk averse, they may wish to hold higher reserves against a given level of vulnerability. If so, imposing the same (lower) coefficient as the non-RIM sample would force the difference to be captured by the RIM dummy. To allow for this possibility, Table 3 repeats this analysis using the principal component proxies.11 The analysis suggests that RIM countries do respond to vulnerabilities somewhat differently from non-RIM countries. Specifically, RIM countries hold more reserves against current account 11

We do not interact the “scale” variables (population, per capita income) with the RIM dummy as that simply assigns a lot of weight on the population variable, given the high correlation between RIM countries and population (0.41; or 0.31 excluding China). We also re-estimated the regression allowing both the intercept and the slope coefficients to differ between RIM and non-RIM countries which we don’t report for brevity. Results of this analysis as well as that of Table 3 suggest that RIM countries do respond differently from non-RIM countries.

ACCEPTED MANUSCRIPT 12

SC

RI PT

vulnerabilities than non-RIM countries: a one standard deviation increase in the current account proxy raises reserve holdings by 2.5% of GDP in non-RIM countries but by 3.7% of GDP in RIM countries (Table 3 [7]). This is partly offset, however, by lower reserve holdings against capital account shocks: a one standard deviation increase in the capital account proxy raises reserve holdings by 3% of GDP in non-RIM countries but by only 2% of GDP in RIM countries (Table 3 [7]). The negative interactive RIM dummy on capital account vulnerabilities is significant in all sub-periods except 1998–2004 when, presumably in response to the Asian crisis, RIM countries sought to insure themselves against capital account shocks to the same degree as other EMEs who were learning from the financial crises of the 1990s. There are a few other differences between RIM and non-RIM countries, with some evidence that RIM countries are more sensitive to the costs of holding reserves (especially in the latter part of the sample, 2005–10), and in some specifications and sub-periods, the RIM interaction term on the undervaluation variable turns negative and statistically significant (Table 3 [8]-[9]).

[Insert Table 3 here]

TE D

4.2. Quantile analysis of reserve demand

M AN U

With (largely) offsetting differences in RIM countries’ responses to current account and capital account vulnerabilities compared to non-RIM countries, the story above remains unchanged: even allowing for the different responses to these vulnerabilities, over the full sample period, RIM countries hold 9.9% of GDP more reserves than non-RIM countries—though this differential shrinks over time, falling to 4.7% of GDP and turning insignificant by 2005–10.

AC C

EP

Thus far, our discussion has been based on the average behavior of RIM and non-RIM countries. But as noted in Section II, there is also substantial variation in reserve holdings both within and across the two groups of countries. We therefore turn to quantile regressions, as specified in (2), which allow the relationship between reserve holdings and the various explanatory variables to vary by the level of reserve holdings. As before, we first estimate the full model (Table 4) and then collapse the various proxies to their first principal component also allowing for the slope coefficients to differ between RIM and non-RIM countries (Table 5). [Insert Table 4 here]

The quantile regressions show that, at relatively low levels (in percent of GDP) of reserve holdings, precautionary motives against current account shocks dominate (Table 4[12-15]).12 In part, this echoes the story above: reserves were low during the early part of the sample (the 1980s), and precautionary motives against current account shocks were the main driving force behind reserve 12

The 25th, 50th, 75th, 90th, and 99th percentiles correspond to reserves of 5.6%, 11.0%, 18.4%, 27.3%, and 54.0% of GDP, respectively. Of the 175 RIM country observations, 119 (i.e., 68%) are above the sample median, and 56 (i.e., 32%) are below the median.

ACCEPTED MANUSCRIPT 13

RI PT

holdings. But while the low-reserves observations are predominantly in the early part of the sample, they are not exclusively so, thus low-reserve holders in general (regardless of the time period) tend hold them against current account shocks. Perhaps more surprising is that the mercantilist motive also seems to be important for the low-reserve observations; closer examination, however, shows that in many of these cases, the country was losing reserves—that is, the “undervaluation” reflected collapsed exchange rates (mainly during crises) as the country exhausted its reserves, rather than deliberate undervaluation for mercantilism.13

SC

The picture is more mixed for capital account vulnerabilities. For high reserves holders, capital account vulnerabilities (particularly broad money) are more important, but financial openness and short-term debt are more important at the lower end of the distribution—while being smaller and/or insignificant for high reserves holders (this is intuitive inasmuch as many high reserve holders have very little short-term debt, and certainly ample reserves to cover their debt).

TE D

M AN U

As before, to ensure that results are not influenced by possible co-linearity of the proxies, we repeat the analysis using their principal components.14 As in the case of the period analysis, collapsing the vulnerability proxies improves the precision of the estimation. The results seem to suggest that RIM countries differ from non-RIM countries only in the lower quartiles of the reserves distribution. For countries with below-median reserves, RIM countries hold some 3% to 4% of GDP more reserves than would be expected given their vulnerabilities (and assuming the same response to these vulnerabilities as other EMEs).15 For countries with above-median reserves, RIM countries do not appear to hold more reserves than their peers.

AC C

EP

The latter conclusion, however, is heavily dependent on the homogeneity assumption. In Table 5, we allow the slope coefficients to differ across country groups. Doing so suggests that (above the 25th percentile) RIM countries hold more reserves against current account shocks than do non-RIM countries, but (across the reserves holding distribution), fewer reserves against capital account shocks than their non-RIM counterparts. The net result is that the RIM dummy becomes significant across the distribution, with the implication that, controlling for current and capital account vulnerabilities (and the RIM countries response to them), RIM countries hold some 6% to 9% of GDP more reserves than would be expected. [Insert Table 5 here]

13

Recall that our definition of mercantilism requires that the country be accumulating reserves in the context of an undervalued currency (and that such accumulation be uncorrelated with precautionary needs). 14 For brevity, these results are not included but are available upon request. 15 The estimated coefficient for the RIM countries dummy for the 25th and 50th percentiles (constant) is 0.506*** (2.725***) and 0.283*** (-2.300***), respectively.

ACCEPTED MANUSCRIPT 14 5. Robustness and extensions 5.1. Robustness

SC

RI PT

The empirical analysis presented thus far suggests that different motives for reserve accumulation apply at different points in time and different level of reserves, and that these motives are different for RIM compared to non-RIM countries. In this section we discuss a series of extensions and various robustness analyses including those related to model specification, period of analysis, alternative proxies for the mercantilist motive, and addressing potential reverse causality. We present these results separately for the OLS and quantile regressions baseline specifications (Table 3 [7] and Table 5 [1-4]) in Tables 6 and 7, respectively. 5.1.1. Results using OLS

M AN U

The first column of Table 6 repeats the baseline specification of Table 3 [7] for the full period of analysis. We start by examining whether changes in reserve accumulation behavior during the global financial crisis that started in 2008 might be driving our results. Ending the sample in 2008 (Table 6 [2]) instead of 2010 makes virtually no difference to the results. Similarly, excluding China or China and Korea from the sample Table 6 [3-4] leaves the results unchanged, suggesting that the difference between RIM and non-RIM countries is not driven by the inclusion of these two countries in the sample.

TE D

[Insert Table 6 here]

AC C

EP

Next, we supplement our mercantilism proxy with a second proxy—the difference between actual exports and those predicted by a gravity model of trade which we label “excess exports” following GOT. A higher value of excess exports suggests that a country that actually exports more than “it should” is implementing a policy to promote exports and boost export competitiveness by preventing or slowing exchange rate appreciation. Replacing our mercantilist variable with excess exports (Table 6 [5]) yields a positive and statistically significant coefficient, suggesting that our findings above on mercantilist motives are not driven by our specific choice of proxy. As four of the RIM countries in our sample we directly affected by the Asian crisis, we augment our specification with a dummy variable for the Asian crisis (essentially an indicator that takes the value of 1 after 1997) and its interaction with RIM countries. As expected, the estimated coefficient of the Asian crisis dummy variable is significant and positive, suggesting that countries generally increased their reserve accumulation past 1997 but RIM countries did not behave differently from non-RIM countries as the interaction term is not statistically significant. Going beyond the traditional measures of demand determinants, recent empirical research has investigated the role of regional imitation (or the “keeping up with the Joneses” motive) in reserve accumulation (see Cheung and Qian (2009), and Bastourre et al. (2009)). The idea is that countries may try to mimic the actions of neighbors of similar characteristics based on strategic behavior

ACCEPTED MANUSCRIPT 15

RI PT

(e.g., having more reserves then regional neighbors makes the country less risky by credit-rating agencies, investors, and speculators). We construct a series of variables to capture this motive and add them sequentially to the baseline specification (Table 6 [7-9]), namely, the percentage of regional partners that increase their reserves in the previous period; the average reserves of regional partners in the previous period; and the average change in reserves of regional partners in the previous period. 16 All three variables have a statistically significant and positive effect suggesting that there is evidence that countries are indeed trying to imitate neighbors/regional partners’ reserve accumulation; overall, this effect seems to be less important in RIM countries compared to non-rim countries, albeit not always statistically significant.

M AN U

SC

An issue that has not yet received much attention in the literature concerns potential endogeneity of the regressors. While some of the regressors, such as partner country growth, are likely to be largely exogenous to the country’s reserve holdings, others—especially broad money and exchange rate undervaluation—may not be. We address reverse causality and endogeneity concerns in columns 10–13 of Table 6. We begin by fitting a treatment-effects model that considers the effect of an endogenous binary treatment variable (undervaluation) on reserves, using first-step probit estimates of the treatment equation. Our baseline results are preserved, with the effect of undervaluation slightly higher than the baseline.

AC C

EP

TE D

Next, in column 11, we replace all regressors with their lagged values, which again has virtually no effect on the results. In column 12 we use instrumental variable estimation using the second and third lags of the regressors. Finally in column 13 we use constructed instruments and augment the instruments used in column 12 with two variables. Following GOT, we use predicted exports and trading partners’ misalignments as instruments. We posit that predicted exports may be correlated with misalignment/undervaluation because predicted exports may provide a benchmark to which the authorities may (implicitly or explicitly) strive to achieve through exchange rate undervaluation, but there is no obvious correlation between a gravity-model predicted exports and reserve levels. The second instrument is trading partners’ misalignment weighted by trade shares. This measure is also likely to be correlated with the country’s measure of misalignment—if competitors are undervalued there is an incentive to also be undervalued—and uncorrelated with reserve accumulation in the country. Overall, instrumental variable specifications in Table 6 [1213] preserve the results of the baseline estimation. Reported tests of over-identifying restrictions, instrument validity, and redundancy suggest that the instruments used are valid and informative in this context.17 Importantly, therefore, our results do not appear to be driven by reverse causality.

16

The first constructed variable is the “regional imitation” proxy constructed by Bastourre et al. (2009); the last two are variations of variables constructed by Cheung and Ito (2009). 17 The Hansen test for instrument validity p-value is 0.11 and 0.13 for columns 12 and 13, respectively.

ACCEPTED MANUSCRIPT 16 5.1.2. Results using quantile regressions

RI PT

Turning to the quantile regressions, Table 7 reports robustness tests mirroring the OLS robustness checks in Table 6. We begin by examining possible effects of the global financial crisis by ending the period of analysis in 2008 (Table 7 [5-8]). Compared to the baseline results are virtually unchanged. Excluding China and Korea from the analysis in columns 9-12 also reserves the baseline results.

M AN U

SC

Replacing the undervaluation measure with another proxy (excess exports) in Table 7 [13]-[16] broadly preserves the results despite the fact that compared to the baseline the mercantilist motive is now insignificant for the median and 75th percentile of reserve holders. Compared to the baseline the significance of current and capital account and mercantilist motives is preserved, despite some small changes in the coefficients. Adding a dummy for the Asian crisis in the specification (Table 7 [17]-[20]) and a variable to proxy for relative reserves (Table 7 [21]-[24]) suggest that the basic story is preserved, even though both these variables enter significantly in the specifications (with both variables more important for low reserve holders). [Insert Table 7 here]

TE D

Finally, using lagged values of the regressors does not change the baseline story substantially. With the exception of the opportunity cost variable (which now becomes important for the high reserve holders only) and the slightly higher marginal effects of the current account variables the results are robust to correcting for possible endogeneity bias. 5.2. Model fit and reserve accumulation trends 5.2.1. Model fit

AC C

EP

How well does the model account for RIM and non-RIM reserve accumulation? And what has been driving reserve accumulation in recent years by RIM and non-RIM countries? In this section we try to answer these questions by comparing the model’s fitted reserve holdings to actual reserve holdings, and by decomposing the cumulative change in the stock of reserves since 1990 into the various precautionary and mercantilist motives. Fig. 4 compares fitted and actual reserve holdings on the eve of the global financial crisis, where the slope coefficients on the regressors are constrained to be the same for RIM and non-RIM countries (i.e., Table 1 [7]). The picture reinforces the impression that RIM countries generally hold more reserves than would be expected on the basis of their vulnerabilities: except for Korea (and Indonesia, which is exactly on the line), all of the RIM countries held more reserves than the model would imply—with the difference, in many cases, exceeding 2 standard deviations of the fitted value; on average, RIM countries were holding some 5½% of GDP more reserves than their corresponding non-RIM peers.

ACCEPTED MANUSCRIPT 17 [Insert Fig. 4 here]

[Insert Fig. 5 here] 5.2.2. Accounting for reserve accumulation

SC

RI PT

Figure 5 repeats this exercise but now using the specification where the slope coefficients are allowed to vary between the RIM and non-RIM samples (Table 3 [7]). While this naturally reduces the gap between actual and fitted values, it is noteworthy that most RIM countries remain above the 45° line—with actual reserves of China, Malaysia, Thailand, and the Philippines in 2007 exceeding the model’s fitted value by more than two standard deviations. While the average residual for RIM countries shrinks from 5½% of GDP to about 4½% of GDP, it remains substantial—consistent with the results above that, even allowing for possible differences in responses to vulnerabilities, RIM countries hold higher reserves than would be expected.

EP

[Insert Fig. 6 and Fig. 7 here]

TE D

M AN U

What has been driving reserves accumulation by EMEs—and by RIM countries in particular— over the past few years? Fig. 6 decomposes the factors behind RIM countries’ accumulation of reserves since 1990 using the model specification with the RIM dummy and the RIM slope coefficients. Perhaps contrary to popular perception, and regardless of which specification is adopted, the contribution of mercantilism is quite limited. The contribution of mercantilism, at its peak in 2004, was around 3½% of GDP; that is, in the counterfactual in which no mercantilist motive was operative, RIM countries’ average reserves holding in 2004 would have been 3½% of GDP lower. Moreover, this 3½% of GDP contribution to the stock of reserves either shrinks or stays relatively constant thereafter. In other words, there is little evidence that, on average, RIM countries continue to accumulate reserves for mercantilist reasons (as defined here). At the same time, it should be acknowledged that for non-RIM countries, mercantilism never appears to have been a driving force behind their reserves accumulation (Fig. 7).

AC C

Constructing a similar decomposition plot for China (Fig. 8) points out both some similarities and some differences compared to the rest of the RIM countries. In terms of the relative contributions of capital account, current account and mercantilist motives, capital account motives seem to contribute more to the reserve accumulation in China compared to the rest of the RIM counties, (while current account contribute proportionately less). After some large negative residuals in the early 1990s (prior to the unification of the exchange market in 1994) China held more reserves than would be expected given their capital and current account vulnerabilities, though, again, the role of mercantilism is quite limited (4½% of GDP). [Insert Fig. 8 here] 6. Conclusions

ACCEPTED MANUSCRIPT 18

RI PT

Over the past three decades, emerging market economies have accumulated large stocks of reserves, prompting suspicions that more than precautionary motives are at play. Asian Pacific Rim countries, in particular, are often singled out for accumulating reserves as a by-product of modern mercantilism—deliberate undervaluation of the currency as part of an export-led growth strategy. In this paper, we seek to assess the validity of this view by comparing Asian RIM countries to their non-RIM emerging market counterparts.

TE D

M AN U

SC

Our analysis points to three main conclusions. First, the reasons why EMEs hold reserves has shifted over time, and depends on the level of reserves. In the 1980s, and at low reserve holdings, reserves were mainly intended as insurance against current account shocks. In the 1990s, and following the emerging market financial crises in particular, insurance against capital account shocks gained importance. Added to this, in the 2000s, is greater evidence of mercantilist motives—that is, reserve accumulation that is uncorrelated with vulnerabilities and that takes place in the context of a substantially undervalued currency. Second, RIM countries do differ from their non-RIM counterparts: they hold more reserves against current account vulnerabilities but fewer reserves against capital account vulnerabilities, and overall hold more reserves than would be expected on the basis of these vulnerabilities. Adding the mercantilist motive shrinks this difference, but only somewhat—so mercantilism provides, at best, only a partial explanation. Third, confirming this, decomposition of the reasons behind RIM countries’ reserve accumulation in recent decades suggests only a modest role for mercantilist motives—with very little evidence that RIM countries, on average, continue to accumulate reserves as a by-product of mercantilism.

AC C

EP

We are thus left with a puzzle: why do RIM countries consistently hold more reserves than precautionary (or even mercantilist) motives would imply? One reason may be that they want to insure against shocks that are not captured here. Another reason may be that reserves provide benefits beyond their insurance value. Yet a third possibility is that restrictions on capital outflows mean that the public sector holds external assets that would otherwise be accumulated by the private sector. Future research will need to explore these possibilities more fully. What is clear is that, when it comes to holding reserves, RIM countries do differ from their non-RIM emerging market counterparts—and simply blaming this on mercantilism will not suffice. Acknowledgements

We are grateful to Joshua Aizenman, Yin-Wong Cheung, Menzie Chinn, IMF colleagues, and participants at the City University of Hong Kong International Conference on Pacific Rim Countries and the Evolution of the International Financial Architecture (December 19-20, 2013) for helpful comments and suggestions, and to Hyeon Ji Lee and Chifundo Moya for excellent research assistance. The views expressed in this paper are those of the authors and should not be attributed to the IMF, its Executive Board, or its management.

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

19

ACCEPTED MANUSCRIPT 20 References Aizenman, J., Lee, J., 2008. Financial Versus Monetary Mercantilism Long-Run View of Large International Reserves Hoarding. The World Economy. 31, 593–611.

RI PT

———, 2007. International Reserves: Precautionary Versus Mercantilist Views, Theory and Evidence. Open Economies Review. 18, 191–214.

Aizenman, J., Marion, N.P., 2003. The High Demand for International Reserves in the Far East: What Is Going On? Journal of the Japanese and International Economies. 17, 370–400.

SC

Bastourre, D., Carrera, J., Ibarlucia, J., 2009. What is Driving Reserves Accumulation. Review of International Economics. 17, 861–877.

M AN U

Ben-Bassat, A., Gottlieb, D., 1992. Optimal International Reserves and Sovereign Risk. Journal of International Economics. 33, 345–362. De Beaufort Wijnholds, J.O., Kapteyn, A., 2001. Reserve Adequacy in Emerging Market Economies. IMF Working Paper 01/143 (Washington: International Monetary Fund).

TE D

Caballero, R.J., Panageas, S., 2008. Hedging Sudden Stops and Precautionary Contractions. Journal of Development Economics. 85, 28–57. Cheung, Y.W., Qian, X., 2009. Hoarding of International Reserves: Mrs. Machlup’s Wardrobe and the Joneses. Review of International Economics. 17, 824–43.

EP

Cheung, Y.W., Ito, H., 2009. A Cross-Country Empirical Analysis of International Reserves. International Economic Journal. 23, 447–481.

AC C

Delatte, A.L., Fouquau, J., 2012. What Drove the Massive Hoarding of International Reserves in Emerging Economies? A Time‐varying Approach, Review of International Economics. Wiley Blackwell. 20 (1) 164–176. Dooley, M.P., Folkerts-Landau, D., Garber, P., 2005. International Financial Stability (New York: Deutsche Bank). Ghosh, A. R., Kim, J.I., 2008. Export Subsidies, Undervalued Exchange Rates, and Consumption Taxes—Some Equivalence Results of Relevance to Bretton Woods II. Manuscript, International Monetary Fund. Washington, D.C. Ghosh, A.R., Ostry, J.D., Tsangarides, C., 2010. Exchange Rate Regimes and the Stability of the

ACCEPTED MANUSCRIPT 21 International Monetary System. IMF Occasional Paper No. 270 (Washington: International Monetary Fund).

RI PT

———, 2012. Shifting Motives: Explaining the Buildup in Official Reserves in Emerging Markets since the 1980s. IMF Working Paper 12/34 (Washington: International Monetary Fund). Jeanne, O., 2007. International Reserves in Emerging Market Countries: Too Much of a Good Thing? Brookings Papers on Economic Activity: 1, Brookings Institution.

SC

Jeanne, O., Rancière, R., 2006. The Optimal Level of International Reserves for Emerging Market Countries: Formulas and Applications. IMF Working Paper 06/229 (Washington: International Monetary Fund).

M AN U

Koenker, R., Hallock, K.F., 2001. Quantile Regression. Journal of Economic Perspectives. 15, 143–56. Koenker, R., Bassett, G.J., 1978. Regression Quantiles. Econometrica. 46, 33–50.

AC C

EP

TE D

Obstfeld, M., Shambaugh, J., Taylor, A.M., 2010. Financial Stability, the Trilemma, and International Reserves. American Economic Journal: Macroeconomics. 2, 57–94.

ACCEPTED MANUSCRIPT 22 Fig. 1. Reserves to GDP RIM, non-RIM, full sample.

RI PT

35

30

25

SC

20

M AN U

15

10

5

0

1980-1997

TE D

1980-2010

Full sample

RIM

1998-2004

non-RIM

AC C

EP

Source: World Economic Outlook and authors’ calculations.

2005-2010

ACCEPTED MANUSCRIPT 23 Fig. 2. Proxies for precautionary and mercantilist motives for RIM and non-RIM countries. 1980-2010

90

90

70

70

50

50

30

30

10

10

-10

-10

-30

-30

M2/GDP

ST Debt/GDP

Imports/GDP

Misalignment

M2/GDP

M AN U

Imports/GDP

SC

110

RI PT

1980-1997

110

1998-2004 110

ST Debt/GDP

Misalignment

2005-2010

110

90

90

70

70

50

TE D

50

30

10

-30

Imports/GDP

EP

-10

M2/GDP

ST Debt/GDP

30

10

-10

-30

Misalignment

Imports/GDP

RIM

AC C

Source: World Economic Outlook and authors’ calculations.

non-RIM

M2/GDP

ST Debt/GDP

Misalignment

ACCEPTED MANUSCRIPT 24

100

AC C

EP

80 60 40 20 0

RIM sample

TE D

120

Full sample non-RIM sample

M AN U

0

20

40

SC

60

80

RI PT

100

120

Fig. 3. Reserves Distribution, various samples and periods.

1980-1997

1998-2004 Full sample non-RIM sample

2005-2010 RIM sample

Source: World Economic Outlook and authors’ calculations.

ACCEPTED MANUSCRIPT 25 Fig. 4. Actual vs. predicted reserves in 2007 (no differentiation between RIM and non-RIM countries).

RI PT

60%

MYS

CHN JOR

Actual 2007

RUS

BGR

M AN U

40%

SC

50%

LBN

THA

VNM MAR

30%

BIH

PER IND

ARM

UKR ROM PHL TUN

TE D

20%

KOR

URY ARG KAZ

CRI JAM

GEOIDN BRA GTM TURVEN ZAF SLV PAK COL PAN

LVALTU HUN

POL

CHL MEX

EP

10%

HRV EGY

DOM ECU

AC C

0%

0%

10%

20%

30%

40%

50%

60%

Fitted 2007

Source. World Economic Outlook and authors’ calculations. Notes. Slope coefficients on the regressors are constrained to be the same for RIM and non-RIM countries. RIM countries are marked in squares. RIM countries whose actual reserves are above 2 standard deviations of the predicted reserves are marked in triangles.

ACCEPTED MANUSCRIPT 26 Fig. 5. Actual vs. predicted reserves in 2007 (differentiating RIM and non-RIM countries).

RI PT

60%

MYS

50%

LBN

SC

CHN JOR

BGR

Actual 2007

RUS

M AN U

40%

THA VNM

MAR

30%

BIH

PER IND

UKR ROM PHL TUN

TE D

20% ARM

KOR

HRV EGY

URY ARG KAZ

LTU LVA HUN

CRI JAM POL

GEO BRA IDNGTM TURVEN ZAF SLV CHL PAK COL PAN MEX

EP

10%

DOM ECU

AC C

0%

0%

10%

20%

30%

40%

50%

60%

Fitted 2007

Source. World Economic Outlook and authors’ calculations. Notes. Slope coefficients are allowed to vary between the RIM and non-RIM samples. RIM countries are marked in squares. RIM countries whose actual reserves are above 2 standard deviations of the predicted reserves are marked in triangles.

ACCEPTED MANUSCRIPT 27 Fig. 6. RIM countries cumulative differences decomposition 1990-2010 (specification with both dummy and interactions).

RI PT

1.2 1.1 1.0 0.9

SC

0.8 0.7

M AN U

0.6 0.5 0.4 0.3 0.2

0.0 ‐0.1 ‐0.2 1990

1992

1994

TE D

0.1

AC C

Source: Authors’ calculations.

1998

Current account

EP

Capital account

1996

2000

2002

Mercantilist

2004 Residual

2006

2008

Reserves‐scale

2010

ACCEPTED MANUSCRIPT 28 Fig. 7. Non-RIM countries cumulative differences decomposition 1990-2010 (specification with both dummy and interactions) 1.4

RI PT

1.3 1.2 1.1 1.0

SC

0.9 0.8

M AN U

0.7 0.6 0.5 0.4 0.3 0.2

0.0 ‐0.1 ‐0.2 1990

1992

1994

1996

AC C

Source: Authors’ calculations.

1998

Current account

EP

Capital account

TE D

0.1

2000

2002

Mercantilist

2004 Residual

2006

2008

Reserves‐scale

2010

ACCEPTED MANUSCRIPT 29 Fig. 8. China cumulative differences decomposition 1990-2010 1.4

RI PT

1.3 1.2 1.1 1.0 0.9

SC

0.8 0.7

M AN U

0.6 0.5 0.4 0.3 0.2 0.1

‐0.1 ‐0.2 1990

1992

1994

1996

AC C

Source: Authors’ calculations.

1998

Current account

EP

Capital account

TE D

0.0

2000

2002

Mercantilist

2004 Residual

2006

2008

Reserves‐scale

2010

ACCEPTED MANUSCRIPT 30 Table 1. Current account, capital account and mercantilist determinants of reserve demand. (1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13) China

(14)

(15)

scale regime opp. cost +mercantilist

full

80-97

98-04

05-10

full

80-97

98-04

05-10

full baseline

0.579*** 0.579*** (0.088) (0.090) -0.073 -0.061 (0.071) (0.067)

Log(population) Regime Hard and Soft peg dummy Volatility of neer

0.591*** (0.091) -0.048 (0.064)

0.515*** 0.409*** (0.087) (0.087) 0.135** 0.138** (0.062) (0.063)

0.117 (0.093) 0.093 (0.076)

0.054 (0.092) 0.045 (0.070)

0.385*** (0.077) 0.135** (0.055)

0.417*** (0.146) 0.086 (0.076)

0.113 (0.093) 0.104 (0.076)

0.047 0.378*** (0.095) (0.081) -0.001 0.148*** (0.070) (0.052)

-0.110 (0.100) -0.005 (0.006)

0.041 (0.119) -0.008* (0.005)

0.306* 0.065 (0.175) (0.137) -0.002 -0.019*** (0.006) (0.005)

-0.124 (0.095) -0.005 (0.006)

0.057 (0.119) -0.008* (0.005)

-0.110 (0.948)

-0.021 (0.078)

0.033 (0.121) -0.007 (0.005)

0.016 (0.143) -0.017*** (0.005)

0.056 (0.118) -0.008* (0.005)

0.308 0.057 (0.184) (0.120) -0.001 -0.019*** (0.006) (0.005)

-0.390*** (0.061)

-0.065 (0.071)

-0.039 (0.082)

-0.393*** (0.069)

-0.025 (0.080)

-0.017 (0.095)

-0.068 (0.843)

-0.342 (0.933)

-0.013 (0.079)

0.039 (0.102)

-0.079 (0.860)

0.770*** 0.648*** (0.150) (0.134) 0.170** 0.239** (0.078) (0.093) 0.311** 0.319** (0.141) (0.128)

0.677*** (0.133) 0.212*** (0.058) 0.233* (0.119)

0.532*** (0.153) 0.197** (0.079) 0.395** (0.192)

0.457** (0.169) 1.385 (1.070) 0.206 (0.255)

0.509*** (0.149) 0.459 (1.989) 0.269 (0.579)

0.691*** (0.121) 0.215*** (0.058) 0.215* (0.119)

0.690*** (0.149) 0.182** (0.072) 0.292* (0.156)

0.476*** (0.134) 1.371 (1.093) 0.231 (0.261)

0.426*** 0.690*** (0.138) (0.121) 0.386 0.215*** (1.898) (0.058) 0.163 0.231* (0.580) (0.118)

0.100** (0.048) 0.278*** (0.100) 0.227 (0.212)

0.113** (0.049) 0.258*** (0.093) 0.274 (0.203)

0.060 (0.071) 0.195* (0.104) -1.068 (0.693)

0.123** (0.056) 0.263** (0.118) 0.509** (0.242)

-0.023 (0.046) 0.426*** (0.143) 0.361** (0.171)

0.113** (0.046) 0.243** (0.094) 0.285 (0.203)

0.114* (0.059) 0.158 (0.108) -0.275 (0.622)

0.125** (0.057) 0.274** (0.109) 0.497** (0.227)

-0.030 0.115** (0.047) (0.046) 0.391** 0.261*** (0.147) (0.092) 0.402** 0.276 (0.177) (0.199)

0.074 (0.109) 0.453* (0.224)

0.324*** (0.081) 0.050 (0.164)

0.358*** (0.101) -0.194 (0.174)

0.240*** (0.065)

0.054 (0.110)

0.333*** (0.081)

0.303*** 0.252*** (0.090) (0.065)

0.091 (0.142)

0.191** 0.249*** (0.089) (0.066) 0.746*** 0.041 (0.157) (0.135)

Volatility of exports/GDP (3-yr sd) Volatility of partner growth (3-yr sd) Capital account Financial openness Log(broad money to GDP) Short term debt to GDP Mercantilist Exchange rate undervaluation

China dummy

0.769*** (0.157)

0.273 (0.166)

TE D

0.864*** 0.825*** (0.180) (0.163)

-2.453*** -2.441*** -2.430*** -2.449*** -2.554*** (0.133) (0.131) (0.130) (0.106) (0.100) 1,009 0.31 0.31

1,009 0.35 0.35

0.00

0.00 0.00

-2.422*** -2.553*** -2.608*** -2.361*** -2.096*** (0.128) (0.096) (0.120) (0.149) (0.170)

0.357** 0.700*** -0.112 0.433** (0.162) (0.223) (0.229) (0.200) -2.540*** -2.567*** -2.367*** -2.038*** -2.554*** (0.099) (0.127) (0.149) (0.171) (0.096)

1,009 0.38 0.37

1,009 0.50 0.49

1,009 0.55 0.54

1,009 0.39 0.38

1,009 0.57 0.56

449 0.45 0.43

296 0.53 0.50

264 0.53 0.51

1,009 0.57 0.56

449 0.44 0.43

296 0.53 0.50

264 0.53 0.51

1,009 0.57 0.56

0.00 0.00

0.00 0.02 0.00

0.00 0.23 0.00 0.00

0.00 0.00

0.00 0.09 0.00 0.00

0.01 0.13 0.00 0.05

0.36 0.00 0.02 0.00

0.79 0.44 0.01 0.00

0.00 0.10 0.00 0.00

0.02 0.11 0.00 0.12

0.32 0.00 0.00 0.00

0.84 0.33 0.02 0.00

0.00 0.09 0.00 0.00

EP

Observations R-squared R-squared adjusted Tests for groups' joined significance p-value Scale p-value Regime p-value CA p-value KA

0.438*** (0.139) 0.026 (0.088)

0.074 (0.133) -0.011** (0.005)

Current account Log(imports to GDP)

Constant

0.575*** 0.383*** (0.089) (0.081) -0.055 0.142** (0.063) (0.061)

-0.025 0.003 (0.144) (0.142) -0.021*** -0.016*** (0.005) (0.005)

Opportunity cost Interest rate differential w/ US

RIM countries dummy

+KA

SC

scale

M AN U

Sample Scale Log(per capita income)

scale scale regime scale regime opp. cost regime opp. cost +CA

RI PT

RIM

AC C

Source. Authors’ estimates. Notes. The table reports coefficients of OLS regressions of reserve demand with RIM or China intercepts included. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Robust standard errors clustered by country reported in parentheses (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

ACCEPTED MANUSCRIPT 31 Table 2. Principal Components Analysis: Current account, capital account and mercantilist determinants of reserve demand. (1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

RIM

Current account Capital account Undervaluation RIM dummy

0.927*** (0.175)

0.893*** (0.181)

0.842*** (0.175)

0.557*** (0.198)

China dummy Constant

Observations R-squared

-2.226*** -2.236*** -2.208*** -2.187*** -2.344*** (0.092) (0.094) (0.094) (0.089) (0.093) 1,009 0.250

1,009 0.277

1,009 0.301

1,009 0.353

1,009 0.413

-2.216*** -2.363*** -2.593*** -2.297*** -2.011*** (0.093) (0.092) (0.129) (0.105) (0.073) 1,009 0.315

1,009 0.444

449 0.320

296 0.457

264 0.394

full 80-97 98-04 05-10 -0.120 -0.267** 0.058 0.141 (0.089) (0.123) (0.090) (0.085) -0.097* 0.030 -0.036 -0.055 (0.058) (0.078) (0.067) (0.057) -0.036 -0.023 -1.090 0.229 (0.093) (0.094) (1.028) (1.007) 0.339*** 0.230* 0.263*** 0.349*** (0.079) (0.114) (0.090) (0.067) 0.258*** 0.335** 0.255*** 0.168*** (0.054) (0.142) (0.053) (0.033) 0.340*** 0.068 0.361*** 0.337*** (0.080) (0.100) (0.076) (0.111)

0.777*** 0.920*** 0.240 0.708*** (0.139) (0.203) (0.206) (0.204) -2.342*** -2.552*** -2.296*** -2.003*** (0.096) (0.134) (0.107) (0.073) 1,009 0.451

449 0.327

296 0.445

EP

TE D

Source. Authors’ estimates. Notes. The table reports coefficients of OLS regressions similar to those in Table 1 where the various proxies of scale, current account vulnerabilities, and capital account vulnerabilities are replaced by their respective first principal components. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Robust standard errors clustered by country reported in parentheses (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

AC C

(14)

RI PT

Opportunity cost

SC

Regime

scale scale regime -0.560*** -0.549*** (0.113) (0.111) -0.157** (0.070)

scale regime opp. cost +KA +mercantilist full 80-97 98-04 05-10 -0.181 -0.535*** -0.147 -0.321** -0.006 0.177* (0.114) (0.104) (0.106) (0.132) (0.101) (0.090) -0.093 -0.124* -0.088 0.026 -0.029 -0.019 (0.060) (0.068) (0.058) (0.078) (0.062) (0.057) -0.085 -0.377*** -0.079 -0.075 -1.169 -0.266 (0.089) (0.109) (0.093) (0.092) (0.975) (1.018) 0.252*** 0.287*** 0.162 0.205** 0.322*** (0.086) (0.083) (0.111) (0.099) (0.077) 0.237*** 0.251*** 0.235 0.242*** 0.179*** (0.050) (0.050) (0.156) (0.047) (0.032) 0.227** 0.342*** 0.119 0.312*** 0.337*** (0.092) (0.084) (0.106) (0.081) (0.124) 0.385** 0.812*** 0.293* 0.453* 0.290* 0.103 (0.173) (0.172) (0.159) (0.250) (0.155) (0.208)

M AN U

Sample Scale

scale scale regime regime opp. cost opp. cost +CA -0.534*** -0.319** (0.105) (0.122) -0.124* -0.088 (0.069) (0.067) -0.359*** -0.112 (0.095) (0.090) 0.304*** (0.083)

(12) (13) China

264 0.416

ACCEPTED MANUSCRIPT 32 Table 3. Principal Components Analysis and interactions: determinants of reserve demand.

Regime Regime x RIM Opportunity cost Opportunity cost x RIM

scale

scale regime

(3)

(4) scale scale regime regime opp. cost opp. cost +CA

-0.560*** -0.551*** -0.531*** (0.113) (0.110) (0.104) -0.160** -0.122* (0.070) (0.069) 0.020 0.002 (0.037) (0.030) -0.355*** (0.092) -2.087 (1.824)

Current account x RIM Capital account Capital account x RIM Undervaluation

Observations R-squared

0.927*** 0.902*** 0.907*** 0.785*** 0.906*** (0.175) (0.181) (0.180) (0.186) (0.192) -2.226*** -2.236*** -2.209*** -2.194*** -2.370*** (0.092) (0.094) (0.094) (0.090) (0.093) 1,009 0.250

1,009 0.278

1,009 0.303

EP

Constant

-0.161 (0.110) -0.073 (0.059) -0.001 (0.018) -0.110 (0.090) -0.386 (0.989) 0.226** (0.089) 0.103* (0.056) 0.253*** (0.050) -0.090*** (0.021)

TE D

Undervaluation x RIM RIM

+KA

(6) scale regime opp. cost +mercantilist -0.533*** (0.103) -0.116 (0.069) 0.028 (0.025) -0.371*** (0.103) -1.403 (1.624)

M AN U

Current account

-0.307** (0.120) -0.081 (0.067) 0.005 (0.021) -0.127 (0.090) -1.929 (1.970) 0.289*** (0.086) 0.090 (0.097)

(5)

1,009 0.357

1,009 0.446

(7)

(8)

(9)

(10)

full

80-97

98-04

05-10

RI PT

Scale

(2)

-0.135 -0.318** -0.010 0.171* (0.103) (0.132) (0.108) (0.085) -0.071 0.045 -0.036 -0.069 (0.059) (0.080) (0.064) (0.055) 0.018 0.015 0.039* -0.004 (0.013) (0.010) (0.023) (0.024) -0.102 -0.087 -1.153 0.829 (0.094) (0.089) (1.056) (1.007) -0.374 -0.797 1.553 -4.258** (0.822) (0.726) (1.413) (1.741) 0.257*** 0.147 0.191* 0.315*** (0.087) (0.112) (0.107) (0.072) 0.101* 0.116 0.060 -0.033 (0.054) (0.087) (0.044) (0.063) 0.262*** 0.235 0.247*** 0.190*** (0.051) (0.162) (0.048) (0.026) -0.073*** -0.081*** -0.015 -0.119*** (0.019) (0.027) (0.022) (0.015) 0.284*** 0.106 0.328*** 0.383*** (0.098) (0.114) (0.090) (0.138) 0.107 -0.169 -0.084 0.026 (0.132) (0.216) (0.118) (0.195) 0.726*** 1.207*** 0.473** 0.309 (0.183) (0.324) (0.203) (0.199) -2.380*** -2.615*** -2.307*** -2.036*** (0.092) (0.131) (0.108) (0.073)

SC

Sample

(1)

0.166 (0.103) 0.354* (0.199) 0.726*** (0.155) -2.207*** (0.093) 1,009 0.323

1,009 0.470

449 0.349

296 0.463

AC C

Source. Authors’ estimates. Notes. The table reports coefficients of OLS regressions similar to those in Table 2 allowing the slope coefficients for RIM and non-RIM countries to differ. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Robust standard errors clustered by country reported in parentheses (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

264 0.465

ACCEPTED MANUSCRIPT 33 Table 4. Reserve demand across quantiles.

Current account Log(imports to GDP) Volatility of exports/GDP (3-yr sd) Volatility of partner growth (3-yr sd) Capital account Financial openness Log(broad money to GDP) Short term debt to GDP Mercantilist Exchange rate undervaluation RIM countries dummy Constant

0.257*** (0.048) 0.092*** (0.032)

0.052 (0.047) -0.009 (0.032)

-0.026 (0.053) -0.032 (0.034)

-0.068 (0.063) -0.069 (0.044)

-0.079** (0.034) -0.023 (0.020)

-0.120** (0.047) -0.060* (0.033)

-0.041 (0.037) -0.036 (0.029)

0.0563 (0.118) -0.00788* (0.00454)

0.083 (0.069) -0.011* (0.006)

0.020 (0.045) -0.004 (0.003)

0.028 (0.039) -0.001 (0.002)

0.004 (0.061) -0.003 (0.004)

-0.063 (0.058) 0.007 (0.005)

-0.055 (0.068) 0.010* (0.005)

-0.079 (0.082) 0.008 (0.006)

0.008 (0.041) 0.002 (0.003)

-0.017 (0.062) 0.000 (0.004)

-0.024 (0.050) -0.002 (0.003)

-0.0248 (0.0805)

0.052 (0.061)

-0.131 (0.086)

-0.098 (0.060)

-0.152** (0.061)

-0.183** (0.080)

-0.151** (0.074)

-0.204** (0.081)

0.032 (0.078)

-0.021 (0.091)

-0.053 (0.058)

0.677*** (0.133) 0.212*** (0.0577) 0.233* (0.119)

0.804*** (0.087) 0.364 (1.033) 0.272* (0.155)

0.647*** (0.063) 0.190 (0.506) 0.195* (0.105)

0.634*** (0.050) 0.067 (0.582) 0.134 (0.123)

0.481*** (0.081) -0.091 (0.781) 0.120 (0.189)

-0.157** (0.078) -0.175 (0.864) -0.077 (0.149)

-0.170** -0.323*** (0.084) (0.108) -0.298 -0.455 (1.143) (1.256) -0.138 -0.152 (0.175) (0.227)

-0.013 (0.057) -0.123 (0.488) -0.061 (0.118)

-0.166* (0.087) -0.281 (0.811) -0.076 (0.200)

-0.153** (0.073) -0.158 (0.711) -0.015 (0.175)

0.113** (0.0487) 0.258*** (0.0930) 0.274 (0.203)

0.139*** (0.024) 0.306*** (0.065) 0.541*** (0.136)

0.082*** (0.016) 0.303*** (0.046) 0.359*** (0.083)

0.061*** (0.016) 0.349*** (0.038) 0.115* (0.067)

0.052** (0.022) 0.329*** (0.058) 0.034 (0.094)

-0.057*** -0.078*** -0.087*** -0.021 -0.030 (0.021) (0.025) (0.030) (0.015) (0.023) -0.003 0.043 0.023 0.046 0.026 (0.059) (0.063) (0.076) (0.039) (0.059) -0.181 -0.425*** -0.507*** -0.244*** -0.326*** (0.126) (0.140) (0.146) (0.076) (0.105)

-0.009 (0.019) -0.020 (0.048) -0.081 (0.081)

0.249*** 0.337*** 0.213*** 0.240*** 0.179*** (0.0664) (0.070) (0.052) (0.037) (0.053) 0.0409 0.077 -0.022 -0.273*** -0.277*** (0.135) (0.105) (0.064) (0.050) (0.086) -2.553*** -2.948*** -2.503*** -2.113*** -1.803*** (0.0963) (0.057) (0.033) (0.035) (0.042)

-0.125** -0.097 -0.158* 0.027 -0.033 (0.062) (0.066) (0.083) (0.042) (0.062) -0.099 -0.350*** -0.354*** -0.251*** -0.255*** (0.091) (0.103) (0.130) (0.058) (0.096) 0.444*** 0.834*** 1.144*** 0.390*** 0.700*** (0.049) (0.056) (0.063) (0.033) (0.044)

-0.061 (0.049) -0.005 (0.082) 0.310*** (0.036)

1,009

AC C

Observations Pseudo R2 Tests for groups' joined significance Scale variables p-value Regime variables p-value Current account variables p-value Capital account p-value

0.298*** (0.034) 0.128*** (0.019)

1,009 0.37

0.00 0.05 0.00 0.00

RI PT

Opportunity cost Interest rate differential w/ US

0.377*** (0.033) 0.151*** (0.023)

SC

Volatility of neer

0.324*** (0.052) 0.161*** (0.036)

M AN U

Regime Hard and Soft peg dummy

Inter-quantile regression results (5) (6) (7) (8) (9) (10) 25 vs. 50 25 vs. 75 25 vs. 90 50 vs. 75 50 vs. 90 75 vs. 90

0.383*** (0.0809) 0.142** (0.0614)

TE D

Log(population)

Quantile regression estimated coefficients (1) (2) (3) (4) 25 50 75 90

EP

Percentile Scale Log(per capita income)

OLS Table 1 (7)

1,009 0.37

1,009 0.37

1,009 0.38

1,009

1,009

1,009

1,009

1,009

1,009

0.00 0.33 0.00 0.00

0.00 0.53 0.00 0.00

0.00 0.62 0.00 0.00

0.47 0.13 0.16 0.02

0.61 0.11 0.09 0.00

0.21 0.22 0.01 0.00

0.05 0.71 0.90 0.00

0.02 0.96 0.18 0.00

0.30 0.78 0.18 0.42

Source. Authors’ estimates. Notes. The table reports coefficients of quantile regressions of reserve demand for the 25th, 50th, 75th, and 90th percentiles with RIM intercepts included (columns 1-4). Inter-quartile regressions in columns 5-10 test differences at different quantiles. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Quantile regression standard errors are obtained using bootstrapping with 1000 replications (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

ACCEPTED MANUSCRIPT 34 Table 5. Principal Components Analysis and interactions: reserve demand across quantiles. Quantile regression estimated coefficients (1) 25

Opportunity cost Opportunity cost x RIM Current account Current account x RIM Capital account Capital account x RIM Undervaluation Undervaluation x RIM RIM countries dummy Constant

Observations Pseudo R2

(6) 25 vs. 75

(7) 25 vs. 90

(8) 50 vs. 75

(9) 50 vs. 90

(10) 75 vs. 90

-0.177*** (0.057) -0.103** (0.044) 0.036** (0.018) -0.053 (0.073) -1.191 (1.513) 0.281*** (0.064) 0.065 (0.057) 0.259*** (0.032) -0.056*** (0.018) 0.277*** (0.090) 0.133 (0.133) 0.869*** (0.132) -2.776*** (0.044)

-0.093 (0.062) -0.090*** (0.031) 0.008 (0.011) -0.167** (0.078) -0.518 (0.838) 0.275*** (0.044) 0.098*** (0.037) 0.231*** (0.025) -0.060*** (0.010) 0.309*** (0.070) 0.026 (0.115) 0.632*** (0.109) -2.320*** (0.036)

-0.040 (0.057) -0.015 (0.027) 0.009 (0.011) -0.210** (0.101) 0.284 (0.614) 0.282*** (0.031) 0.107*** (0.030) 0.222*** (0.026) -0.073*** (0.011) 0.273*** (0.061) 0.119 (0.100) 0.389*** (0.093) -1.898*** (0.032)

0.069 (0.043) -0.008 (0.032) 0.014 (0.013) -0.123 (0.091) -0.165 (0.999) 0.291*** (0.041) 0.107*** (0.036) 0.182*** (0.041) -0.095*** (0.015) 0.268*** (0.075) 0.181 (0.120) 0.258*** (0.095) -1.612*** (0.047)

0.084 (0.057) 0.013 (0.037) -0.029* (0.016) -0.114 (0.077) 0.673 (1.268) -0.006 (0.055) 0.033 (0.052) -0.029 (0.026) -0.004 (0.016) 0.031 (0.079) -0.107 (0.120) -0.237* (0.124) 0.456*** (0.038)

0.137** (0.068) 0.088** (0.043) -0.027 (0.019) -0.157 (0.111) 1.475 (1.470) 0.001 (0.063) 0.042 (0.058) -0.038 (0.033) -0.017 (0.018) -0.004 (0.091) -0.014 (0.142) -0.480*** (0.138) 0.878*** (0.045)

0.247*** (0.065) 0.095* (0.050) -0.022 (0.021) -0.070 (0.106) 1.026 (1.671) 0.010 (0.070) 0.042 (0.065) -0.077 (0.048) -0.039* (0.022) -0.009 (0.106) 0.048 (0.161) -0.612*** (0.150) 1.164*** (0.058)

0.053 (0.055) 0.075*** (0.028) 0.002 (0.012) -0.043 (0.094) 0.802 (0.781) 0.007 (0.036) 0.009 (0.033) -0.009 (0.027) -0.013 (0.011) -0.035 (0.061) 0.093 (0.101) -0.242** (0.095) 0.422*** (0.033)

0.163** (0.066) 0.082** (0.038) 0.006 (0.015) 0.044 (0.094) 0.353 (1.173) 0.016 (0.051) 0.009 (0.043) -0.048 (0.044) -0.035** (0.015) -0.041 (0.087) 0.155 (0.136) -0.374*** (0.121) 0.708*** (0.051)

0.110** (0.050) 0.007 (0.030) 0.005 (0.012) 0.087 (0.081) -0.449 (0.879) 0.009 (0.036) -0.000 (0.034) -0.039 (0.036) -0.022* (0.013) -0.005 (0.071) 0.062 (0.112) -0.132 (0.091) 0.286*** (0.040)

1,009 0.310

1,009 0.300

1,009 0.299

1,009 0.314

1,009

1,009

1,009

1,009

1,009

1,009

RI PT

Regime x RIM

Inter-quantile regression results (5) 25 vs. 50

SC

Regime

(4) 90

M AN U

Scale

(3) 75

TE D

Percentile

(2) 50

AC C

EP

Source. Authors’ estimates. Notes. The table reports coefficients of quantile regressions similar to those in Table 4 allowing the slope coefficients for RIM and non-RIM countries to differ (columns 1-4). Inter-quartile regressions in columns 5-10 test differences at different quantiles. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Quantile regression standard errors are obtained using bootstrapping with 1000 replications (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

ACCEPTED MANUSCRIPT 35 Table 6. Extensions and robustness of OLS estimates.

Scale

Regime x RIM Opportunity cost Opportunity cost x RIM Current account Current account x RIM Capital account Capital account x RIM Undervaluation Undervaluation x RIM RIM

-0.157 (0.110) -0.085 (0.061) 0.009 (0.008) -0.096 (0.093) -0.166 (0.714) 0.256*** (0.087) 0.177*** (0.049) 0.254*** (0.052) -0.039** (0.016) 0.284*** (0.098) 0.044 (0.123) 0.692*** (0.186)

-0.160 (0.114) -0.083 (0.061) 0.006 (0.008) -0.098 (0.096) -0.100 (0.692) 0.253*** (0.092) 0.175*** (0.050) 0.255*** (0.052) -0.043** (0.018) 0.283*** (0.098) 0.072 (0.124) 0.703*** (0.210)

Excess exports Excess exports x RIM Asian crisis

Relative reserves 1 Relative reserves 1 x RIM Relative reserves 2 Relative reserves 2 x RIM

Asian crisis

-0.145 -0.120 (0.102) (0.096) -0.072 -0.003 (0.056) (0.054) 0.004 0.023* (0.015) (0.013) -0.081 -0.061 (0.087) (0.091) -0.410 -0.845 (1.156) (0.989) 0.205** 0.205** (0.088) (0.081) 0.077 0.106** (0.054) (0.043) 0.246*** 0.213*** (0.052) (0.042) -0.095*** -0.058*** (0.019) (0.014) 0.274*** (0.089) -0.054 (0.141) 0.868*** 0.888*** (0.171) (0.192) 0.260** (0.125) 0.080 (0.186) 0.516*** (0.116) -0.096 (0.189)

EP

Relative reserves 3 Relative reserves 3 x RIM

-2.380*** -2.438*** -2.374*** -2.373*** (0.092) (0.099) (0.094) (0.094)

AC C

Observations R-squared

(6)

TE D

Asian crisis x RIM

Constant

(5)

Different No China mercantilist End 2008 No China no Korea proxy

-0.135 -0.148 (0.103) (0.109) -0.071 -0.063 (0.059) (0.062) 0.018 0.011 (0.013) (0.009) -0.102 -0.085 (0.094) (0.097) -0.374 -0.542 (0.822) (0.814) 0.257*** 0.247*** (0.087) (0.091) 0.101* 0.117** (0.054) (0.057) 0.262*** 0.261*** (0.051) (0.055) -0.073*** -0.068*** (0.019) (0.020) 0.284*** 0.266*** (0.098) (0.098) 0.107 0.068 (0.132) (0.144) 0.726*** 0.809*** (0.183) (0.190)

Regime

(4)

1,009 0.470

877 0.461

982 0.466

969 0.460

(7) Percent of partners with increased reserves -0.120 (0.098) -0.078 (0.057) 0.019 (0.013) -0.144 (0.091) -0.291 (0.993) 0.238*** (0.083) 0.123** (0.053) 0.248*** (0.048) -0.073*** (0.017) 0.256** (0.097) 0.121 (0.130) 0.921*** (0.190)

(8) Average reserves of partners -0.235** (0.094) -0.011 (0.050) 0.030** (0.012) -0.115 (0.078) -0.690 (1.280) 0.104 (0.080) 0.126** (0.052) 0.183*** (0.041) -0.058*** (0.012) 0.277*** (0.085) -0.081 (0.145) 0.323 (0.461)

(9) Average change in reserves of partners

-2.376*** -2.456*** (0.099) (0.090) 971 0.463

1,009 0.525

(10)

(11)

(12)

(13)

Treatment effects undervaluation

Using Lags

IV1

IV2

-0.134 (0.099) -0.088 (0.056) 0.021* (0.012) -0.136 (0.092) -0.170 (0.938) 0.236*** (0.083) 0.116** (0.054) 0.254*** (0.049) -0.074*** (0.018) 0.274*** (0.101) 0.101 (0.133) 0.742*** (0.181)

-0.112*** -0.123 -0.116 -0.156** (0.040) (0.106) (0.076) (0.074) -0.060** -0.049 -0.090 -0.135* (0.024) (0.062) (0.074) (0.076) 0.027 0.005 0.083** 0.070* (0.019) (0.016) (0.038) (0.039) -0.113* -0.023 0.117 0.056 (0.060) (0.086) (0.180) (0.181) 0.154 0.343 2.518 2.245 (1.294) (0.449) (1.829) (1.894) 0.266*** 0.284*** 0.337*** 0.271*** (0.034) (0.087) (0.073) (0.073) 0.113*** 0.081 0.081 0.073 (0.043) (0.054) (0.056) (0.061) 0.266*** 0.259*** 0.254*** 0.250*** (0.023) (0.053) (0.039) (0.039) -0.074*** -0.080*** -0.058*** -0.061*** (0.015) (0.018) (0.019) (0.020) 0.448*** 0.298*** 0.484*** 0.501*** (0.074) (0.101) (0.162) (0.159) 0.115 0.089 0.326 0.211 (0.120) (0.128) (0.251) (0.260) 0.688*** 0.698*** 0.389** 0.476** (0.127) (0.170) (0.192) (0.206)

-2.486*** -2.353*** -2.352*** -2.343*** (0.038) (0.094) (0.060) (0.064)

RI PT

Sample

(3)

SC

Full model

(2)

M AN U

(1)

0.596*** (0.119) -0.367** (0.177) 0.623*** (0.105) -0.261 (0.176)

-2.367*** (0.092)

-2.321*** (0.090)

0.643*** (0.196) -0.312 (0.274) -2.351*** (0.091)

958 0.492

962 0.561

962 0.483

934

963 0.460

871 0.441

Source. Authors’ estimates. Notes. The table reports coefficients of various robustness checks and extensions for the OLS baseline specification in Table 3 column 7. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Robust standard errors clustered by country reported in parentheses (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

834 0.445

Table 7. Extensions and robustness of quantile regression estimates.

Regime Regime x RIM Opportunity cost Opportunity cost x RIM Current account Current account x RIM Capital account Capital account x RIM Undervaluation Undervaluation x RIM RIM countries dummy

(4) 90

(5) 25

(6) 50

(7) 75

(8) 90

-0.177*** (0.057) -0.103** (0.044) 0.036** (0.018) -0.053 (0.073) -1.191 (1.513) 0.281*** (0.064) 0.065 (0.057) 0.259*** (0.032) -0.056*** (0.018) 0.277*** (0.090) 0.133 (0.133) 0.869*** (0.132)

-0.093 (0.062) -0.090*** (0.031) 0.008 (0.011) -0.167** (0.078) -0.518 (0.838) 0.275*** (0.044) 0.098*** (0.037) 0.231*** (0.025) -0.060*** (0.010) 0.309*** (0.070) 0.026 (0.115) 0.632*** (0.109)

-0.040 (0.057) -0.015 (0.027) 0.009 (0.011) -0.210** (0.101) 0.284 (0.614) 0.282*** (0.031) 0.107*** (0.030) 0.222*** (0.026) -0.073*** (0.011) 0.273*** (0.061) 0.119 (0.100) 0.389*** (0.093)

0.069 (0.043) -0.008 (0.032) 0.014 (0.013) -0.123 (0.091) -0.165 (0.999) 0.291*** (0.041) 0.107*** (0.036) 0.182*** (0.041) -0.095*** (0.015) 0.268*** (0.075) 0.181 (0.120) 0.258*** (0.095)

-0.171*** (0.057) -0.086* (0.045) 0.026 (0.020) -0.034 (0.071) -1.036 (1.616) 0.301*** (0.066) 0.093 (0.063) 0.246*** (0.029) -0.053*** (0.019) 0.222** (0.092) 0.154 (0.135) 0.941*** (0.136)

-0.121** (0.061) -0.077** (0.035) 0.009 (0.011) -0.159** (0.075) -0.370 (0.717) 0.264*** (0.042) 0.101*** (0.035) 0.214*** (0.025) -0.061*** (0.009) 0.306*** (0.076) -0.009 (0.111) 0.731*** (0.101)

-0.053 (0.055) -0.008 (0.026) -0.001 (0.010) -0.208** (0.105) 0.077 (0.586) 0.276*** (0.033) 0.110*** (0.029) 0.205*** (0.031) -0.073*** (0.010) 0.282*** (0.057) 0.054 (0.084) 0.456*** (0.085)

0.058 (0.052) 0.006 (0.031) -0.007 (0.010) -0.101 (0.087) -0.603 (0.904) 0.294*** (0.048) 0.084** (0.037) 0.215*** (0.056) -0.082*** (0.014) 0.230*** (0.080) 0.112 (0.111) 0.209** (0.094)

(9) 25

Excess exports x RIM Asian crisis

Observations Pseudo R2

-2.776*** -2.320*** -1.898*** -1.612*** (0.044) (0.036) (0.032) (0.047) 1,009 0.310

1,009 0.300

1,009 0.299

1,009 0.314

-2.811*** -2.366*** -1.925*** -1.672*** (0.042) (0.039) (0.033) (0.049) 921 0.306

921 0.301

-2.772*** -2.311*** -1.886*** -1.597*** (0.045) (0.037) (0.032) (0.049)

AC C

Constant

921 0.297

921 0.308

Alternative mercantilist proxy (12) 90

EP

Asian crisis x RIM

Relative reserves 1 x RIM

(11) 75

-0.227*** -0.169** -0.073 0.050 (0.064) (0.071) (0.065) (0.050) -0.111** -0.095*** -0.012 -0.024 (0.046) (0.031) (0.026) (0.034) 0.018 -0.005 -0.003 -0.006 (0.022) (0.011) (0.012) (0.014) -0.051 -0.169** -0.214** -0.196** (0.072) (0.070) (0.100) (0.090) -0.073 0.272 0.068 -0.414 (1.324) (0.693) (0.592) (0.923) 0.269*** 0.256*** 0.280*** 0.289*** (0.071) (0.050) (0.033) (0.044) 0.170*** 0.154*** 0.111*** 0.097*** (0.059) (0.038) (0.030) (0.036) 0.255*** 0.210*** 0.199*** 0.169*** (0.031) (0.025) (0.026) (0.043) -0.019 -0.033** -0.065*** -0.089*** (0.021) (0.014) (0.016) (0.021) 0.254*** 0.319*** 0.293*** 0.250*** (0.086) (0.071) (0.061) (0.076) 0.105 0.012 0.077 0.114 (0.145) (0.119) (0.095) (0.120) 0.891*** 0.657*** 0.379*** 0.248** (0.167) (0.132) (0.104) (0.116)

Excess exports

Relative reserves 1

(10) 50

(13) 25

(14) 50

(15) 75

Asian crisis

(16) 90

-0.227*** -0.153*** 0.003 0.062* (0.073) (0.056) (0.064) (0.037) -0.089** -0.110*** -0.038 0.038 (0.044) (0.033) (0.033) (0.027) 0.001 0.011 -0.019 -0.018 (0.015) (0.011) (0.017) (0.016) -0.041 -0.151* -0.143* -0.136** (0.062) (0.079) (0.084) (0.068) -2.142* -0.635 -0.124 0.501 (1.261) (0.879) (1.037) (1.091) 0.141*** 0.154*** 0.261*** 0.333*** (0.052) (0.049) (0.047) (0.045) 0.054 0.086* 0.085* 0.056 (0.063) (0.051) (0.043) (0.037) 0.244*** 0.236*** 0.229*** 0.163*** (0.025) (0.025) (0.037) (0.042) -0.087*** -0.087*** -0.097*** -0.109*** (0.017) (0.012) (0.011) (0.013)

969 0.304

969 0.292

969 0.291

969 0.307

(17) 25

(18) 50

(19) 75

Relative reserves (20) 90

-0.105** -0.099*** 0.012 0.060 (0.053) (0.037) (0.055) (0.048) -0.008 -0.018 0.021 0.017 (0.039) (0.027) (0.027) (0.032) 0.024 0.024** 0.019 0.021 (0.019) (0.011) (0.013) (0.014) 0.033 -0.132 -0.017 -0.107 (0.070) (0.081) (0.114) (0.095) -1.159 -0.414 -0.454 -0.199 (1.681) (0.791) (0.677) (0.796) 0.231*** 0.206*** 0.274*** 0.284*** (0.047) (0.040) (0.033) (0.042) 0.106** 0.137*** 0.114*** 0.102*** (0.043) (0.028) (0.030) (0.035) 0.211*** 0.175*** 0.192*** 0.182*** (0.018) (0.019) (0.023) (0.044) -0.042*** -0.051*** -0.064*** -0.088*** (0.016) (0.010) (0.012) (0.013) 0.382*** 0.261*** 0.201*** 0.242*** (0.076) (0.055) (0.056) (0.072) -0.116 -0.010 0.079 -0.006 (0.143) (0.096) (0.098) (0.134) 1.307*** 1.002*** 0.474*** 0.199 (0.219) (0.128) (0.134) (0.147)

SC

Scale

No China no Korea

(3) 75

M AN U

Percentile

Period ending in 2008

(2) 50

TE D

Baseline (1) 25

RI PT

ACCEPTED MANUSCRIPT

1.138*** (0.201) 0.314*** (0.115) 0.143 (0.203)

0.802*** (0.141) 0.123 (0.079) 0.146 (0.116)

0.459*** (0.119) 0.093 (0.057) 0.110 (0.119)

0.336*** (0.122) 0.186*** (0.055) 0.045 (0.109)

0.806*** 0.592*** (0.107) (0.060) -0.447** -0.287*** (0.219) (0.100)

-2.800*** -2.344*** -1.902*** -1.559*** (0.051) (0.038) (0.041) (0.035) 977 0.314

977 0.285

977 0.277

977 0.306

0.360*** (0.081) -0.027 (0.126)

1,009 0.357

1,009 0.350

(22) 50

(23) 75

Using lagged values (24) 90

(25) 25

(26) 50

(27) 75

(28) 90

-0.135** -0.060 -0.020 0.091* (0.056) (0.058) (0.054) (0.049) -0.116** -0.092** -0.012 -0.013 (0.048) (0.037) (0.023) (0.027) 0.031* 0.008 0.009 0.012 (0.016) (0.011) (0.009) (0.012) -0.131 -0.170** -0.253*** -0.147 (0.089) (0.072) (0.098) (0.089) -1.697 -0.384 -0.021 0.769 (1.622) (0.782) (0.595) (0.897) 0.237*** 0.237*** 0.272*** 0.281*** (0.060) (0.042) (0.031) (0.043) 0.091* 0.127*** 0.131*** 0.097** (0.052) (0.034) (0.031) (0.039) 0.239*** 0.223*** 0.211*** 0.190*** (0.028) (0.024) (0.024) (0.036) -0.060*** -0.059*** -0.072*** -0.108*** (0.016) (0.009) (0.010) (0.014) 0.223*** 0.210*** 0.241*** 0.238*** (0.085) (0.069) (0.057) (0.069) 0.120 0.151 0.164* 0.092 (0.128) (0.104) (0.092) (0.118) 1.142*** 0.728*** 0.453*** 0.359* (0.196) (0.157) (0.137) (0.188)

-0.175*** -0.129** 0.009 0.062 (0.058) (0.063) (0.051) (0.045) -0.117** -0.095*** -0.012 0.023 (0.046) (0.033) (0.025) (0.023) 0.023 0.005 -0.007 -0.022 (0.017) (0.013) (0.012) (0.018) 0.059 -0.061 -0.028 -0.137** (0.088) (0.088) (0.069) (0.056) 0.105 0.497 -0.142 -0.651 (1.162) (0.815) (0.626) (1.131) 0.326*** 0.283*** 0.303*** 0.294*** (0.061) (0.048) (0.034) (0.034) 0.035 0.065* 0.088*** 0.068 (0.052) (0.035) (0.030) (0.044) 0.262*** 0.207*** 0.199*** 0.193*** (0.031) (0.028) (0.025) (0.039) -0.061*** -0.071*** -0.082*** -0.107*** (0.015) (0.012) (0.011) (0.018) 0.259*** 0.276*** 0.281*** 0.200*** (0.089) (0.085) (0.059) (0.069) 0.094 0.079 0.062 0.112 (0.125) (0.117) (0.091) (0.124) 0.788*** 0.653*** 0.396*** 0.265** (0.124) (0.096) (0.083) (0.114)

0.763*** 0.619*** 0.384*** 0.270* (0.200) (0.154) (0.093) (0.152) -0.460* -0.248 -0.163 -0.187 (0.251) (0.204) (0.175) (0.215) -2.734*** -2.298*** -1.887*** -1.628*** (0.052) (0.043) (0.032) (0.051)

-2.735*** -2.283*** -1.870*** -1.583*** (0.045) (0.038) (0.032) (0.037)

0.099 (0.099) 0.227 (0.170)

-2.826*** -2.413*** -2.044*** -1.787*** (0.050) (0.028) (0.031) (0.038) 1,009 0.348

(21) 25

1,009 0.350

958 0.328

958 0.315

958 0.313

958 0.324

965 0.307

Source. Authors’ estimates. Notes. The table reports coefficients of various robustness checks and extensions for the quantile regression baseline specifications in Table 5 columns 1-4. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Quantile regression standard errors are obtained using bootstrapping with 1000 replications (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

965 0.290

965 0.290

965 0.311

ACCEPTED MANUSCRIPT

Appendix A. Data and summary statistics Table A1. Countries in the sample, and variable definitions and sources.

RI PT

Argentina, Armenia, Bosnia and Herzegovina, Brazil, Bulgaria, Chile, China, Colombia, Costa Rica, Croatia, Dominican Republic, Ecuador, Egypt, El Salvador, Georgia, Guatemala, Hungary, India, Indonesia, Jamaica, Jordan, Kazakhstan, Korea, Latvia, Lebanon, Lithuania, Malaysia, Mexico, Morocco, Pakistan, Panama, Peru, Philippines, Poland, Romania, Russia, South Africa, Thailand, Tunisia, Turkey, Ukraine, Uruguay, Venezuela, Vietnam. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Description

Log(reserves to GDP) Log(per capita income) Log(population) Log(imports to GDP) Volatility of neer Interest rate differential w/ US

Natural log of ratio of foreign exchange reserves to USD GDP Natural log of per capita income (at PPP) Natural log of population Natural log of ratio of imports to GDP Twelve month standard deviation of end of period nominal effective exchange rate ln[(1 + i)/(1 + iUS)], where iUS is the US interest rate corresponding to the definition used for the national interest rate (deposit, money market, t-bill rate, lending) 3 year standard deviation of export to GDP (goods) Chinn-Ito index measuring a country's degree of capital account openness Natural log of ratio of M2 to GDP Dummy variable equal to 1 if the currency pegged and zero otherwise Ratio of total short-term debt outstanding to GDP 3 year standard deviation of growth of trading partners' real GDP Dummy variable equal to 1 if the currency is undervalued and zero otherwise

M AN U

Volatility of exports/GDP (3-year sd) Financial openness Log(broad money to GDP) Peg dummy Short term debt to GDP Volatility of partner growth (3-year sd) Undervaluation dummy

Source IMF, IMF, IMF, IMF, IMF, IMF,

World Economic Outlook World Economic Outlook World Economic Outlook World Economic Outlook International Financial Statistics and authors' calculations International Financial Statistics and authors' calculations

SC

Variable

AC C

EP

TE D

Source. Authors’ estimates.

IMF, World Economic Outlook and authors' calculations Chinn and Ito (2009) IMF, World Economic Outlook Ghosh, Qureshi, and Tsangarides (2011) IMF, World Economic Outlook IMF, World Economic Outlook and authors' calculations Authors' calculations

ACCEPTED MANUSCRIPT 38 Table A2. Summary statistics RIM sample.

-1.96 14% 8.07 18.41 -0.97 12.41 0.03 0.03 0.17 -0.35 0.54 0.13 0.11 0.42

Source. Authors’ estimates.

Table A3. Summary statistics non-RIM sample. Variable

Mean

-2.39 9% 8.59 16.72 -1.23 13.18 0.15 0.03 0.14 -0.92 0.60 0.14 0.14 0.29

Std. Dev.

0.88 240% 0.68 1.46 0.59 8.27 0.43 0.10 1.58 0.57 0.49 0.19 0.18 0.46

AC C

EP

TE D

Log(reserves to GDP) Reserves to GDP Log(per capita income) Log(population) Log(imports to GDP) Volatility of neer Interest rate differential w/ US Volatility of exports/GDP (3-year sd) Financial openness Log(broad money to GDP) Peg dummy Short term debt to GDP Volatility of partner growth (3-year sd) Undervaluation dummy

0.79 220% 0.90 1.28 0.56 7.14 0.05 0.03 1.24 0.67 0.50 0.08 0.10 0.50

Source. Authors’ estimates.

Min -3.91 2% 6.08 16.44 -2.45 2.41 -0.09 0.00 -1.84 -2.30 0.00 0.02 0.01 0.00

Max -0.62 54% 10.30 21.02 0.01 45.45 0.28 0.17 2.48 0.66 1.00 0.45 0.66 1.00

RI PT

Log(reserves to GDP) Reserves to GDP Log(per capita income) Log(population) Log(imports to GDP) Volatility of neer Interest rate differential w/ US Volatility of exports/GDP (3-year sd) Financial openness Log(broad money to GDP) Peg dummy Short term debt to GDP Volatility of partner growth (3-year sd) Undervaluation dummy

Std. Dev.

SC

Mean

M AN U

Variable

Min

-5.42 0% 6.03 14.56 -3.07 0.81 -0.12 0.00 -1.84 -2.62 0.00 0.00 0.00 0.00

Max 0.07 107% 9.88 20.92 -0.06 41.22 5.08 2.83 2.48 1.05 1.00 1.73 2.37 1.00

ACCEPTED MANUSCRIPT Appendix A Table A1. Countries in the sample, and variable definitions and sources. Argentina, Armenia, Bosnia and Herzegovina, Brazil, Bulgaria, Chile, China, Colombia, Costa Rica, Croatia, Dominican Republic, Ecuador, Egypt, El Salvador, Georgia, Guatemala, Hungary, India, Indonesia, Jamaica, Jordan, Kazakhstan, Korea, Latvia, Lebanon, Lithuania, Malaysia, Mexico, Morocco, Pakistan, Panama, Peru, Philippines, Poland, Romania, Russia, South Africa, Thailand, Tunisia, Turkey, Ukraine, Uruguay, Venezuela, Vietnam. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Description

Source

Log(reserves to GDP) Log(per capita income) Log(population) Log(imports to GDP) Volatility of neer Interest rate differential w/ US

Natural log of ratio of foreign exchange reserves to USD GDP Natural log of per capita income (at PPP) Natural log of population Natural log of ratio of imports to GDP Twelve month standard deviation of end of period nominal effective exchange rate ln[(1 + i)/(1 + iUS)], where iUS is the US interest rate corresponding to the definition used for the national interest rate (deposit, money market, t-bill rate, lending) 3 year standard deviation of export to GDP (goods) Chinn-Ito index measuring a country's degree of capital account openness Natural log of ratio of M2 to GDP Dummy variable equal to 1 if the currency pegged and zero otherwise Ratio of total short-term debt outstanding to GDP 3 year standard deviation of growth of trading partners' real GDP Dummy variable equal to 1 if the currency is undervalued and zero otherwise

IMF, World Economic Outlook IMF, World Economic Outlook IMF, World Economic Outlook IMF, World Economic Outlook IMF, International Financial Statistics and authors' calculations IMF, International Financial Statistics and authors' calculations

Volatility of exports/GDP (3-year sd) Financial openness Log(broad money to GDP) Peg dummy Short term debt to GDP Volatility of partner growth (3-year sd) Undervaluation dummy

IMF, World Economic Outlook and authors' calculations Chinn and Ito (2009) IMF, World Economic Outlook Ghosh, Qureshi, and Tsangarides (2011) IMF, World Economic Outlook IMF, World Economic Outlook and authors' calculations Authors' calculations

AC C

EP

TE D

M AN U

SC

Source. Authors’ estimates.

RI PT

Variable

ACCEPTED MANUSCRIPT Table A2. Summary statistics RIM sample.

-1.96 14% 8.07 18.41 -0.97 12.41 0.03 0.03 0.17 -0.35 0.54 0.13 0.11 0.42

Min

0.79 220% 0.90 1.28 0.56 7.14 0.05 0.03 1.24 0.67 0.50 0.08 0.10 0.50

AC C

EP

TE D

M AN U

Log(reserves to GDP) Reserves to GDP Log(per capita income) Log(population) Log(imports to GDP) Volatility of neer Interest rate differential w/ US Volatility of exports/GDP (3-year sd) Financial openness Log(broad money to GDP) Peg dummy Short term debt to GDP Volatility of partner growth (3-year sd) Undervaluation dummy

Std. Dev.

-3.91 2% 6.08 16.44 -2.45 2.41 -0.09 0.00 -1.84 -2.30 0.00 0.02 0.01 0.00

Max -0.62 54% 10.30 21.02 0.01 45.45 0.28 0.17 2.48 0.66 1.00 0.45 0.66 1.00

RI PT

Mean

SC

Variable

(1)

(2)

(3)

ACCEPTED MANUSCRIPT (4) (5) (6) (7) (8) (9)

(10)

(11)

(12)

(13) China

(14)

(15)

05-10

full

80-97

98-04

05-10

full baseline

RIM

scale regime

0.579*** (0.088) -0.073 (0.071)

0.579*** (0.090) -0.061 (0.067)

Regime Hard and Soft peg dummy Volatility of neer

+KA

0.591*** (0.091) -0.048 (0.064)

0.515*** (0.087) 0.135** (0.062)

0.409*** (0.087) 0.138** (0.063)

0.575*** (0.089) -0.055 (0.063)

0.383*** (0.081) 0.142** (0.061)

-0.025 0.003 (0.144) (0.142) -0.021*** -0.016*** (0.005) (0.005)

0.074 (0.133) -0.011** (0.005)

0.033 (0.121) -0.007 (0.005)

0.016 (0.143) -0.017*** (0.005)

-0.390*** (0.061)

-0.065 (0.071)

-0.039 (0.082)

-0.393*** (0.069)

0.770*** (0.150) 0.170** (0.078) 0.311** (0.141)

Opportunity cost Interest rate differential w/ US Current account Log(imports to GDP) Volatility of exports/GDP (3-yr sd) Volatility of partner growth (3-yr sd) Capital account Financial openness Log(broad money to GDP) Short term debt to GDP

0.054 (0.092) 0.045 (0.070)

0.385*** (0.077) 0.135** (0.055)

0.056 (0.118) -0.008* (0.005)

0.308 0.057 (0.184) (0.120) -0.001 -0.019*** (0.006) (0.005)

-0.110 (0.100) -0.005 (0.006)

-0.025 (0.080)

-0.017 (0.095)

-0.068 (0.843)

0.648*** (0.134) 0.239** (0.093) 0.319** (0.128)

0.677*** (0.133) 0.212*** (0.058) 0.233* (0.119)

0.532*** (0.153) 0.197** (0.079) 0.395** (0.192)

0.100** (0.048) 0.278*** (0.100) 0.227 (0.212)

0.113** (0.049) 0.258*** (0.093) 0.274 (0.203) 0.249*** (0.066) 0.041 (0.135)

0.825*** (0.163)

0.769*** (0.157)

0.273 (0.166)

China dummy Constant

0.091 (0.142)

0.191** (0.089) 0.746*** (0.157)

0.438*** (0.139) 0.026 (0.088)

98-04

0.113 (0.093) 0.104 (0.076)

0.047 (0.095) -0.001 (0.070)

0.378*** (0.081) 0.148*** (0.052)

0.041 (0.119) -0.008* (0.005)

0.306* 0.065 (0.175) (0.137) -0.002 -0.019*** (0.006) (0.005)

-0.124 (0.095) -0.005 (0.006)

0.057 (0.119) -0.008* (0.005)

-0.342 (0.933)

-0.013 (0.079)

0.039 (0.102)

-0.079 (0.860)

-0.110 (0.948)

-0.021 (0.078)

0.457** (0.169) 1.385 (1.070) 0.206 (0.255)

0.509*** (0.149) 0.459 (1.989) 0.269 (0.579)

0.691*** (0.121) 0.215*** (0.058) 0.215* (0.119)

0.690*** (0.149) 0.182** (0.072) 0.292* (0.156)

0.476*** (0.134) 1.371 (1.093) 0.231 (0.261)

0.426*** (0.138) 0.386 (1.898) 0.163 (0.580)

0.690*** (0.121) 0.215*** (0.058) 0.231* (0.118)

0.060 (0.071) 0.195* (0.104) -1.068 (0.693)

0.123** (0.056) 0.263** (0.118) 0.509** (0.242)

-0.023 (0.046) 0.426*** (0.143) 0.361** (0.171)

0.113** (0.046) 0.243** (0.094) 0.285 (0.203)

0.114* (0.059) 0.158 (0.108) -0.275 (0.622)

0.125** (0.057) 0.274** (0.109) 0.497** (0.227)

-0.030 (0.047) 0.391** (0.147) 0.402** (0.177)

0.115** (0.046) 0.261*** (0.092) 0.276 (0.199)

0.074 (0.109) 0.453* (0.224)

0.324*** (0.081) 0.050 (0.164)

0.358*** (0.101) -0.194 (0.174)

0.240*** (0.065)

0.054 (0.110)

0.333*** (0.081)

0.303*** (0.090)

0.252*** (0.065)

M AN U

0.864*** (0.180)

80-97

0.117 (0.093) 0.093 (0.076)

Mercantilist Exchange rate undervaluation RIM countries dummy

full

-2.453*** -2.441*** -2.430*** -2.449*** -2.554*** (0.133) (0.131) (0.130) (0.106) (0.100)

0.417*** (0.146) 0.086 (0.076)

RI PT

Log(population)

scale

scale regime opp. cost +mercantilist

SC

Sample Scale Log(per capita income)

scale scale regime regime opp. cost opp. cost +CA

-2.422*** -2.553*** -2.608*** -2.361*** -2.096*** (0.128) (0.096) (0.120) (0.149) (0.170)

0.357** 0.700*** -0.112 0.433** (0.162) (0.223) (0.229) (0.200) -2.540*** -2.567*** -2.367*** -2.038*** -2.554*** (0.099) (0.127) (0.149) (0.171) (0.096)

AC C

EP

TE D

Observations 1,009 1,009 1,009 1,009 1,009 1,009 1,009 449 296 264 1,009 449 296 264 1,009 R-squared 0.31 0.35 0.38 0.50 0.55 0.39 0.57 0.45 0.53 0.53 0.57 0.44 0.53 0.53 0.57 R-squared adjusted 0.31 0.35 0.37 0.49 0.54 0.38 0.56 0.43 0.50 0.51 0.56 0.43 0.50 0.51 0.56 Tests for groups' joined significance p-value Scale 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.36 0.79 0.00 0.02 0.32 0.84 0.00 p-value Regime 0.00 0.00 0.02 0.23 0.00 0.09 0.13 0.00 0.44 0.10 0.11 0.00 0.33 0.09 p-value CA 0.00 0.00 0.00 0.00 0.02 0.01 0.00 0.00 0.00 0.02 0.00 p-value KA 0.00 0.00 0.05 0.00 0.00 0.00 0.12 0.00 0.00 0.00 Source. Authors' estimates. Notes. The table reports coefficients of OLS regressions of reserve demand with RIM or China intercepts included. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Robust standard errors clustered by country reported in parentheses (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, T1, 20-05-2014

Principal Components Analysis: Current account, capital account and mercantilist determinants of reserve demand. (1)

(2)

(3)

(4)

ACCEPTED (5) (6) MANUSCRIPT (7) (8) (9)

(10)

(11)

RIM

scale -0.560*** (0.113)

Regime

scale regime -0.549*** (0.111) -0.157** (0.070)

Opportunity cost Current account Capital account Undervaluation RIM dummy

0.927*** (0.175)

0.893*** (0.181)

0.842*** (0.175)

0.557*** (0.198)

scale regime opp. cost +KA +mercantilist -0.181 -0.535*** (0.114) (0.104) -0.093 -0.124* (0.060) (0.068) -0.085 -0.377*** (0.089) (0.109) 0.252*** (0.086) 0.237*** (0.050) 0.227** (0.092) 0.385** 0.812*** (0.173) (0.172)

China dummy Constant

-2.226*** (0.092)

-2.236*** (0.094)

-2.208*** (0.094)

-2.187*** (0.089)

-2.344*** (0.093)

-2.216*** (0.093)

full -0.147 (0.106) -0.088 (0.058) -0.079 (0.093) 0.287*** (0.083) 0.251*** (0.050) 0.342*** (0.084) 0.293* (0.159)

80-97 -0.321** (0.132) 0.026 (0.078) -0.075 (0.092) 0.162 (0.111) 0.235 (0.156) 0.119 (0.106) 0.453* (0.250)

98-04 -0.006 (0.101) -0.029 (0.062) -1.169 (0.975) 0.205** (0.099) 0.242*** (0.047) 0.312*** (0.081) 0.290* (0.155)

05-10 0.177* (0.090) -0.019 (0.057) -0.266 (1.018) 0.322*** (0.077) 0.179*** (0.032) 0.337*** (0.124) 0.103 (0.208)

full -0.120 (0.089) -0.097* (0.058) -0.036 (0.093) 0.339*** (0.079) 0.258*** (0.054) 0.340*** (0.080)

80-97 -0.267** (0.123) 0.030 (0.078) -0.023 (0.094) 0.230* (0.114) 0.335** (0.142) 0.068 (0.100)

RI PT

Sample Scale

scale scale regime regime opp. cost opp. cost +CA -0.534*** -0.319** (0.105) (0.122) -0.124* -0.088 (0.069) (0.067) -0.359*** -0.112 (0.095) (0.090) 0.304*** (0.083)

(12) (13) China

-2.363*** -2.593*** -2.297*** -2.011*** (0.092) (0.129) (0.105) (0.073)

98-04 0.058 (0.090) -0.036 (0.067) -1.090 (1.028) 0.263*** (0.090) 0.255*** (0.053) 0.361*** (0.076)

(14)

05-10 0.141 (0.085) -0.055 (0.057) 0.229 (1.007) 0.349*** (0.067) 0.168*** (0.033) 0.337*** (0.111)

0.777*** 0.920*** 0.240 0.708*** (0.139) (0.203) (0.206) (0.204) -2.342*** -2.552*** -2.296*** -2.003*** (0.096) (0.134) (0.107) (0.073)

AC C

EP

TE D

M AN U

SC

Observations 1,009 1,009 1,009 1,009 1,009 1,009 1,009 449 296 264 1,009 449 296 264 R-squared 0.250 0.277 0.301 0.353 0.413 0.315 0.444 0.320 0.457 0.394 0.451 0.327 0.445 0.416 Source. Authors' estimates. Notes. The table reports coefficients of OLS regressions similar to those in Table 1 where the various proxies of scale, current account vulnerabilities, and capital account vulnerabilities are replaced by their respective first principal components. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Robust standard errors clustered by country reported in parentheses (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, T2, 20-05-2014

Principal Components Analysis and interactions: determinants of reserve demand.

scale

scale regime

-0.560*** (0.113)

Regime Regime x RIM

(3)

-0.551*** (0.110) -0.160** (0.070) 0.020 (0.037)

Opportunity cost Opportunity cost x RIM

(4) scale scale regime regime opp. cost opp. cost +CA -0.531*** (0.104) -0.122* (0.069) 0.002 (0.030) -0.355*** (0.092) -2.087 (1.824)

Current account Current account x RIM

-0.307** (0.120) -0.081 (0.067) 0.005 (0.021) -0.127 (0.090) -1.929 (1.970) 0.289*** (0.086) 0.090 (0.097)

Capital account Capital account x RIM

(5)

+KA -0.161 (0.110) -0.073 (0.059) -0.001 (0.018) -0.110 (0.090) -0.386 (0.989) 0.226** (0.089) 0.103* (0.056) 0.253*** (0.050) -0.090*** (0.021)

Undervaluation x RIM RIM Constant

-0.533*** (0.103) -0.116 (0.069) 0.028 (0.025) -0.371*** (0.103) -1.403 (1.624)

0.166 (0.103) 0.354* (0.199) 0.726*** (0.155) -2.207*** (0.093)

M AN U

Undervaluation

(6) scale regime opp. cost +mercantilist

0.927*** 0.902*** 0.907*** 0.785*** 0.906*** (0.175) (0.181) (0.180) (0.186) (0.192) -2.226*** -2.236*** -2.209*** -2.194*** -2.370*** (0.092) (0.094) (0.094) (0.090) (0.093)

(7)

(8)

(9)

(10)

full

80-97

98-04

05-10

-0.135 -0.318** (0.103) (0.132) -0.071 0.045 (0.059) (0.080) 0.018 0.015 (0.013) (0.010) -0.102 -0.087 (0.094) (0.089) -0.374 -0.797 (0.822) (0.726) 0.257*** 0.147 (0.087) (0.112) 0.101* 0.116 (0.054) (0.087) 0.262*** 0.235 (0.051) (0.162) -0.073*** -0.081*** (0.019) (0.027) 0.284*** 0.106 (0.098) (0.114) 0.107 -0.169 (0.132) (0.216) 0.726*** 1.207*** (0.183) (0.324) -2.380*** -2.615*** (0.092) (0.131)

-0.010 (0.108) -0.036 (0.064) 0.039* (0.023) -1.153 (1.056) 1.553 (1.413) 0.191* (0.107) 0.060 (0.044) 0.247*** (0.048) -0.015 (0.022) 0.328*** (0.090) -0.084 (0.118) 0.473** (0.203) -2.307*** (0.108)

RI PT

Scale

(2)

SC

Sample

ACCEPTED MANUSCRIPT

(1)

0.171* (0.085) -0.069 (0.055) -0.004 (0.024) 0.829 (1.007) -4.258** (1.741) 0.315*** (0.072) -0.033 (0.063) 0.190*** (0.026) -0.119*** (0.015) 0.383*** (0.138) 0.026 (0.195) 0.309 (0.199) -2.036*** (0.073)

AC C

EP

TE D

Observations 1,009 1,009 1,009 1,009 1,009 1,009 1,009 449 296 264 R-squared 0.250 0.278 0.303 0.357 0.446 0.323 0.470 0.349 0.463 0.465 Source. Authors' estimates. Notes. The table reports coefficients of OLS regressions similar to those in Table 2 allowing the slope coefficients for RIM and non-RIM countries to differ. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Robust standard errors clustered by country reported in parentheses (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, T3, 20-05-2014

Reserve demand across quantiles.

Opportunity cost Interest rate differential w/ US Current account Log(imports to GDP) Volatility of exports/GDP (3-yr sd) Volatility of partner growth (3-yr sd) Capital account Financial openness Log(broad money to GDP) Short term debt to GDP Mercantilist Exchange rate undervaluation RIM countries dummy Constant

0.377*** (0.033) 0.151*** (0.023)

0.298*** (0.034) 0.128*** (0.019)

0.257*** (0.048) 0.092*** (0.032)

0.052 (0.047) -0.009 (0.032)

-0.026 (0.053) -0.032 (0.034)

-0.068 (0.063) -0.069 (0.044)

-0.079** (0.034) -0.023 (0.020)

-0.120** (0.047) -0.060* (0.033)

-0.041 (0.037) -0.036 (0.029)

0.0563 (0.118) -0.00788* (0.00454)

0.083 (0.069) -0.011* (0.006)

0.020 (0.045) -0.004 (0.003)

0.028 (0.039) -0.001 (0.002)

0.004 (0.061) -0.003 (0.004)

-0.063 (0.058) 0.007 (0.005)

-0.055 (0.068) 0.010* (0.005)

-0.079 (0.082) 0.008 (0.006)

0.008 (0.041) 0.002 (0.003)

-0.017 (0.062) 0.000 (0.004)

-0.024 (0.050) -0.002 (0.003)

-0.0248 (0.0805)

0.052 (0.061)

-0.131 (0.086)

-0.098 (0.060)

-0.152** (0.061)

-0.183** (0.080)

-0.151** (0.074)

-0.204** (0.081)

0.032 (0.078)

-0.021 (0.091)

-0.053 (0.058)

0.677*** (0.133) 0.212*** (0.0577) 0.233* (0.119)

0.804*** (0.087) 0.364 (1.033) 0.272* (0.155)

0.647*** (0.063) 0.190 (0.506) 0.195* (0.105)

0.634*** (0.050) 0.067 (0.582) 0.134 (0.123)

0.481*** (0.081) -0.091 (0.781) 0.120 (0.189)

-0.157** (0.078) -0.175 (0.864) -0.077 (0.149)

-0.170** (0.084) -0.298 (1.143) -0.138 (0.175)

-0.323*** (0.108) -0.455 (1.256) -0.152 (0.227)

-0.013 (0.057) -0.123 (0.488) -0.061 (0.118)

-0.166* (0.087) -0.281 (0.811) -0.076 (0.200)

-0.153** (0.073) -0.158 (0.711) -0.015 (0.175)

0.113** (0.0487) 0.258*** (0.0930) 0.274 (0.203)

0.139*** (0.024) 0.306*** (0.065) 0.541*** (0.136)

0.082*** (0.016) 0.303*** (0.046) 0.359*** (0.083)

0.061*** (0.016) 0.349*** (0.038) 0.115* (0.067)

0.052** (0.022) 0.329*** (0.058) 0.034 (0.094)

-0.057*** (0.021) -0.003 (0.059) -0.181 (0.126)

-0.078*** (0.025) 0.043 (0.063) -0.425*** (0.140)

-0.087*** (0.030) 0.023 (0.076) -0.507*** (0.146)

-0.021 (0.015) 0.046 (0.039) -0.244*** (0.076)

-0.030 (0.023) 0.026 (0.059) -0.326*** (0.105)

-0.009 (0.019) -0.020 (0.048) -0.081 (0.081)

0.249*** (0.0664) 0.0409 (0.135) -2.553*** (0.0963)

0.337*** (0.070) 0.077 (0.105) -2.948*** (0.057)

0.213*** (0.052) -0.022 (0.064) -2.503*** (0.033)

0.240*** (0.037) -0.273*** (0.050) -2.113*** (0.035)

0.179*** (0.053) -0.277*** (0.086) -1.803*** (0.042)

-0.125** (0.062) -0.099 (0.091) 0.444*** (0.049)

-0.097 (0.066) -0.350*** (0.103) 0.834*** (0.056)

-0.158* (0.083) -0.354*** (0.130) 1.144*** (0.063)

0.027 (0.042) -0.251*** (0.058) 0.390*** (0.033)

-0.033 (0.062) -0.255*** (0.096) 0.700*** (0.044)

-0.061 (0.049) -0.005 (0.082) 0.310*** (0.036)

1,009

1,009 0.37

1,009 0.37

1,009 0.37

1,009 0.38

1,009

1,009

1,009

1,009

1,009

1,009

0.00 0.05 0.00 0.00

0.00 0.33 0.00 0.00

0.00 0.53 0.00 0.00

0.00 0.62 0.00 0.00

0.47 0.13 0.16 0.02

0.61 0.11 0.09 0.00

0.21 0.22 0.01 0.00

0.05 0.71 0.90 0.00

0.02 0.96 0.18 0.00

0.30 0.78 0.18 0.42

AC C

EP

Observations Pseudo R2 Tests for groups' joined significance Scale variables p-value Regime variables p-value Current account variables p-value Capital account p-value

0.324*** (0.052) 0.161*** (0.036)

RI PT

Volatility of neer

(10) 75 vs. 90

0.383*** (0.0809) 0.142** (0.0614)

SC

Regime Hard and Soft peg dummy

Inter-quantile regression results (6) (7) (8) (9) 25 vs. 75 25 vs. 90 50 vs. 75 50 vs. 90

(5) 25 vs. 50

M AN U

Log(population)

Quantile regression estimated coefficients (1) (2) (3) (4) 25 50 75 90

TE D

Percentile Scale Log(per capita income)

ACCEPTED MANUSCRIPT

OLS Table 1 (7)

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, T4, 20-05-2014

Principal Components Analysis and interactions: reserve demand across quantiles. Quantile regression estimated coefficientsMANUSCRIPT ACCEPTED

Regime x RIM Opportunity cost Opportunity cost x RIM Current account Current account x RIM Capital account Capital account x RIM Undervaluation Undervaluation x RIM RIM countries dummy Constant

-0.093 (0.062) -0.090*** (0.031) 0.008 (0.011) -0.167** (0.078) -0.518 (0.838) 0.275*** (0.044) 0.098*** (0.037) 0.231*** (0.025) -0.060*** (0.010) 0.309*** (0.070) 0.026 (0.115) 0.632*** (0.109) -2.320*** (0.036)

-0.040 (0.057) -0.015 (0.027) 0.009 (0.011) -0.210** (0.101) 0.284 (0.614) 0.282*** (0.031) 0.107*** (0.030) 0.222*** (0.026) -0.073*** (0.011) 0.273*** (0.061) 0.119 (0.100) 0.389*** (0.093) -1.898*** (0.032)

(4) 90 0.069 (0.043) -0.008 (0.032) 0.014 (0.013) -0.123 (0.091) -0.165 (0.999) 0.291*** (0.041) 0.107*** (0.036) 0.182*** (0.041) -0.095*** (0.015) 0.268*** (0.075) 0.181 (0.120) 0.258*** (0.095) -1.612*** (0.047)

(5) 25 vs. 50 0.084 (0.057) 0.013 (0.037) -0.029* (0.016) -0.114 (0.077) 0.673 (1.268) -0.006 (0.055) 0.033 (0.052) -0.029 (0.026) -0.004 (0.016) 0.031 (0.079) -0.107 (0.120) -0.237* (0.124) 0.456*** (0.038)

(6) 25 vs. 75 0.137** (0.068) 0.088** (0.043) -0.027 (0.019) -0.157 (0.111) 1.475 (1.470) 0.001 (0.063) 0.042 (0.058) -0.038 (0.033) -0.017 (0.018) -0.004 (0.091) -0.014 (0.142) -0.480*** (0.138) 0.878*** -0.045

(7) 25 vs. 90 0.247*** (0.065) 0.095* (0.050) -0.022 (0.021) -0.070 (0.106) 1.026 (1.671) 0.010 (0.070) 0.042 (0.065) -0.077 (0.048) -0.039* (0.022) -0.009 (0.106) 0.048 (0.161) -0.612*** (0.150) 1.164*** (0.058)

(8) 50 vs. 75 0.053 (0.055) 0.075*** (0.028) 0.002 (0.012) -0.043 (0.094) 0.802 (0.781) 0.007 (0.036) 0.009 (0.033) -0.009 (0.027) -0.013 (0.011) -0.035 (0.061) 0.093 (0.101) -0.242** (0.095) 0.422*** (0.033)

RI PT

Regime

-0.177*** (0.057) -0.103** (0.044) 0.036** (0.018) -0.053 (0.073) -1.191 (1.513) 0.281*** (0.064) 0.065 (0.057) 0.259*** (0.032) -0.056*** (0.018) 0.277*** (0.090) 0.133 (0.133) 0.869*** (0.132) -2.776*** (0.044)

(3) 75

SC

Scale

(2) 50

M AN U

(1) 25

Percentile

Inter-quantile regression results (9) 50 vs. 90 0.163** (0.066) 0.082** (0.038) 0.006 (0.015) 0.044 (0.094) 0.353 (1.173) 0.016 (0.051) 0.009 (0.043) -0.048 (0.044) -0.035** (0.015) -0.041 (0.087) 0.155 (0.136) -0.374*** (0.121) 0.708*** (0.051)

(10) 75 vs. 90 0.110** (0.050) 0.007 (0.030) 0.005 (0.012) 0.087 (0.081) -0.449 (0.879) 0.009 (0.036) -0.000 (0.034) -0.039 (0.036) -0.022* (0.013) -0.005 (0.071) 0.062 (0.112) -0.132 (0.091) 0.286*** (0.040)

AC C

EP

TE D

Observations 1,009 1,009 1,009 1,009 1,009 1,009 1,009 1,009 1,009 1,009 Pseudo R2 0.310 0.300 0.299 0.314 Source. Authors' estimates. Notes. The table reports coefficients of quantile regressions similar to those in Table 4 allowing the slope coefficients for RIM and non-RIM countries to differ (columns 1-4). Inter-quartile regressions in columns 5-10 test differences at different quantiles. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Quantile regression standard errors are obtained using bootstrapping with 1000 replications (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, T5, 20-05-2014

Extensions and robustness of OLS estimates

Scale Regime Regime x RIM Opportunity cost Opportunity cost x RIM Current account Current account x RIM Capital account Capital account x RIM Undervaluation Undervaluation x RIM RIM

Percent of ACCEPTED MANUSCRIPT

(3)

(4)

Different No China mercantilist End 2008 No China no Korea proxy

-0.135 -0.148 (0.103) (0.109) -0.071 -0.063 (0.059) (0.062) 0.018 0.011 (0.013) (0.009) -0.102 -0.085 (0.094) (0.097) -0.374 -0.542 (0.822) (0.814) 0.257*** 0.247*** (0.087) (0.091) 0.101* 0.117** (0.054) (0.057) 0.262*** 0.261*** (0.051) (0.055) -0.073*** -0.068*** (0.019) (0.020) 0.284*** 0.266*** (0.098) (0.098) 0.107 0.068 (0.132) (0.144) 0.726*** 0.809*** (0.183) (0.190)

-0.157 (0.110) -0.085 (0.061) 0.009 (0.008) -0.096 (0.093) -0.166 (0.714) 0.256*** (0.087) 0.177*** (0.049) 0.254*** (0.052) -0.039** (0.016) 0.284*** (0.098) 0.044 (0.123) 0.692*** (0.186)

-0.160 (0.114) -0.083 (0.061) 0.006 (0.008) -0.098 (0.096) -0.100 (0.692) 0.253*** (0.092) 0.175*** (0.050) 0.255*** (0.052) -0.043** (0.018) 0.283*** (0.098) 0.072 (0.124) 0.703*** (0.210)

Excess exports Excess exports x RIM

(6)

(7)

(8)

Asian crisis

partners with increased reserves

Average reserves of partners

-0.145 -0.120 (0.102) (0.096) -0.072 -0.003 (0.056) (0.054) 0.004 0.023* (0.015) (0.013) -0.081 -0.061 (0.087) (0.091) -0.410 -0.845 (1.156) (0.989) 0.205** 0.205** (0.088) (0.081) 0.077 0.106** (0.054) (0.043) 0.246*** 0.213*** (0.052) (0.042) -0.095*** -0.058*** (0.019) (0.014) 0.274*** (0.089) -0.054 (0.141) 0.868*** 0.888*** (0.171) (0.192) 0.260** (0.125) 0.080 (0.186) 0.516*** (0.116) -0.096 (0.189)

-0.120 (0.098) -0.078 (0.057) 0.019 (0.013) -0.144 (0.091) -0.291 (0.993) 0.238*** (0.083) 0.123** (0.053) 0.248*** (0.048) -0.073*** (0.017) 0.256** (0.097) 0.121 (0.130) 0.921*** (0.190)

(9) Average change in reserves of partners

-0.235** (0.094) -0.011 (0.050) 0.030** (0.012) -0.115 (0.078) -0.690 (1.280) 0.104 (0.080) 0.126** (0.052) 0.183*** (0.041) -0.058*** (0.012) 0.277*** (0.085) -0.081 (0.145) 0.323 (0.461)

M AN U

Asian crisis Asian crisis x RIM Relative reserves 1

(10)

(11)

(12)

(13)

Treatment effects undervaluation

Using Lags

IV1

IV2

-0.134 (0.099) -0.088 (0.056) 0.021* (0.012) -0.136 (0.092) -0.170 (0.938) 0.236*** (0.083) 0.116** (0.054) 0.254*** (0.049) -0.074*** (0.018) 0.274*** (0.101) 0.101 (0.133) 0.742*** (0.181)

-0.112*** -0.123 -0.116 -0.156** (0.040) (0.106) (0.076) (0.074) -0.060** -0.049 -0.090 -0.135* (0.024) (0.062) (0.074) (0.076) 0.027 0.005 0.083** 0.070* (0.019) (0.016) (0.038) (0.039) -0.113* -0.023 0.117 0.056 (0.060) (0.086) (0.180) (0.181) 0.154 0.343 2.518 2.245 (1.294) (0.449) (1.829) (1.894) 0.266*** 0.284*** 0.337*** 0.271*** (0.034) (0.087) (0.073) (0.073) 0.113*** 0.081 0.081 0.073 (0.043) (0.054) (0.056) (0.061) 0.266*** 0.259*** 0.254*** 0.250*** (0.023) (0.053) (0.039) (0.039) -0.074*** -0.080*** -0.058*** -0.061*** (0.015) (0.018) (0.019) (0.020) 0.448*** 0.298*** 0.484*** 0.501*** (0.074) (0.101) (0.162) (0.159) 0.115 0.089 0.326 0.211 (0.120) (0.128) (0.251) (0.260) 0.688*** 0.698*** 0.389** 0.476** (0.127) (0.170) (0.192) (0.206)

0.643*** (0.196) -0.312 (0.274) -2.351*** (0.091)

-2.486*** -2.353*** -2.352*** -2.343*** (0.038) (0.094) (0.060) (0.064)

0.596*** (0.119) -0.367** (0.177)

Relative reserves 1 x RIM Relative reserves 2

0.623*** (0.105) -0.261 (0.176)

Relative reserves 2 x RIM Relative reserves 3

TE D

Relative reserves 3 x RIM Constant

(5)

RI PT

Sample

Full model

(2)

SC

(1)

-2.380*** -2.438*** -2.374*** -2.373*** (0.092) (0.099) (0.094) (0.094)

-2.376*** -2.456*** (0.099) (0.090)

-2.367*** (0.092)

-2.321*** (0.090)

AC C

EP

Observations 1,009 877 982 969 971 1,009 958 962 962 934 963 871 834 R-squared 0.470 0.461 0.466 0.460 0.463 0.525 0.492 0.561 0.483 0.460 0.441 0.445 Source. Authors' estimates. Notes. The table reports coefficients of various robustness checks and extensions for the OLS baseline specification in Table 3 column 7. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Robust standard errors clustered by country reported in parentheses (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, T6, 20-05-2014

ACCEPTED MANUSCRIPT Extensions and robustness of quantile regression estimates.

Regime Regime x RIM Opportunity cost Opportunity cost x RIM Current account Current account x RIM Capital account Capital account x RIM Undervaluation Undervaluation x RIM RIM countries dummy

(4) 90

(5) 25

(6) 50

(7) 75

(8) 90

-0.177*** (0.057) -0.103** (0.044) 0.036** (0.018) -0.053 (0.073) -1.191 (1.513) 0.281*** (0.064) 0.065 (0.057) 0.259*** (0.032) -0.056*** (0.018) 0.277*** (0.090) 0.133 (0.133) 0.869*** (0.132)

-0.093 (0.062) -0.090*** (0.031) 0.008 (0.011) -0.167** (0.078) -0.518 (0.838) 0.275*** (0.044) 0.098*** (0.037) 0.231*** (0.025) -0.060*** (0.010) 0.309*** (0.070) 0.026 (0.115) 0.632*** (0.109)

-0.040 (0.057) -0.015 (0.027) 0.009 (0.011) -0.210** (0.101) 0.284 (0.614) 0.282*** (0.031) 0.107*** (0.030) 0.222*** (0.026) -0.073*** (0.011) 0.273*** (0.061) 0.119 (0.100) 0.389*** (0.093)

0.069 (0.043) -0.008 (0.032) 0.014 (0.013) -0.123 (0.091) -0.165 (0.999) 0.291*** (0.041) 0.107*** (0.036) 0.182*** (0.041) -0.095*** (0.015) 0.268*** (0.075) 0.181 (0.120) 0.258*** (0.095)

-0.171*** (0.057) -0.086* (0.045) 0.026 (0.020) -0.034 (0.071) -1.036 (1.616) 0.301*** (0.066) 0.093 (0.063) 0.246*** (0.029) -0.053*** (0.019) 0.222** (0.092) 0.154 (0.135) 0.941*** (0.136)

-0.121** (0.061) -0.077** (0.035) 0.009 (0.011) -0.159** (0.075) -0.370 (0.717) 0.264*** (0.042) 0.101*** (0.035) 0.214*** (0.025) -0.061*** (0.009) 0.306*** (0.076) -0.009 (0.111) 0.731*** (0.101)

-0.053 (0.055) -0.008 (0.026) -0.001 (0.010) -0.208** (0.105) 0.077 (0.586) 0.276*** (0.033) 0.110*** (0.029) 0.205*** (0.031) -0.073*** (0.010) 0.282*** (0.057) 0.054 (0.084) 0.456*** (0.085)

0.058 (0.052) 0.006 (0.031) -0.007 (0.010) -0.101 (0.087) -0.603 (0.904) 0.294*** (0.048) 0.084** (0.037) 0.215*** (0.056) -0.082*** (0.014) 0.230*** (0.080) 0.112 (0.111) 0.209** (0.094)

(9) 25

(10) 50

(11) 75

Alternative mercantilist proxy (12) 90

(13) 25

-0.227*** -0.169** -0.073 0.050 (0.064) (0.071) (0.065) (0.050) -0.111** -0.095*** -0.012 -0.024 (0.046) (0.031) (0.026) (0.034) 0.018 -0.005 -0.003 -0.006 (0.022) (0.011) (0.012) (0.014) -0.051 -0.169** -0.214** -0.196** (0.072) (0.070) (0.100) (0.090) -0.073 0.272 0.068 -0.414 (1.324) (0.693) (0.592) (0.923) 0.269*** 0.256*** 0.280*** 0.289*** (0.071) (0.050) (0.033) (0.044) 0.170*** 0.154*** 0.111*** 0.097*** (0.059) (0.038) (0.030) (0.036) 0.255*** 0.210*** 0.199*** 0.169*** (0.031) (0.025) (0.026) (0.043) -0.019 -0.033** -0.065*** -0.089*** (0.021) (0.014) (0.016) (0.021) 0.254*** 0.319*** 0.293*** 0.250*** (0.086) (0.071) (0.061) (0.076) 0.105 0.012 0.077 0.114 (0.145) (0.119) (0.095) (0.120) 0.891*** 0.657*** 0.379*** 0.248** (0.167) (0.132) (0.104) (0.116)

(14) 50

(15) 75

Asian crisis

(16) 90

-0.227*** -0.153*** 0.003 0.062* (0.073) (0.056) (0.064) (0.037) -0.089** -0.110*** -0.038 0.038 (0.044) (0.033) (0.033) (0.027) 0.001 0.011 -0.019 -0.018 (0.015) (0.011) (0.017) (0.016) -0.041 -0.151* -0.143* -0.136** (0.062) (0.079) (0.084) (0.068) -2.142* -0.635 -0.124 0.501 (1.261) (0.879) (1.037) (1.091) 0.141*** 0.154*** 0.261*** 0.333*** (0.052) (0.049) (0.047) (0.045) 0.054 0.086* 0.085* 0.056 (0.063) (0.051) (0.043) (0.037) 0.244*** 0.236*** 0.229*** 0.163*** (0.025) (0.025) (0.037) (0.042) -0.087*** -0.087*** -0.097*** -0.109*** (0.017) (0.012) (0.011) (0.013)

Excess exports Excess exports x RIM Asian crisis Asian crisis x RIM Relative reserves 1

(17) 25

(18) 50

(19) 75

Relative reserves (20) 90

-0.105** -0.099*** 0.012 0.060 (0.053) (0.037) (0.055) (0.048) -0.008 -0.018 0.021 0.017 (0.039) (0.027) (0.027) (0.032) 0.024 0.024** 0.019 0.021 (0.019) (0.011) (0.013) (0.014) 0.033 -0.132 -0.017 -0.107 (0.070) (0.081) (0.114) (0.095) -1.159 -0.414 -0.454 -0.199 (1.681) (0.791) (0.677) (0.796) 0.231*** 0.206*** 0.274*** 0.284*** (0.047) (0.040) (0.033) (0.042) 0.106** 0.137*** 0.114*** 0.102*** (0.043) (0.028) (0.030) (0.035) 0.211*** 0.175*** 0.192*** 0.182*** (0.018) (0.019) (0.023) (0.044) -0.042*** -0.051*** -0.064*** -0.088*** (0.016) (0.010) (0.012) (0.013) 0.382*** 0.261*** 0.201*** 0.242*** (0.076) (0.055) (0.056) (0.072) -0.116 -0.010 0.079 -0.006 (0.143) (0.096) (0.098) (0.134) 1.307*** 1.002*** 0.474*** 0.199 (0.219) (0.128) (0.134) (0.147)

RI PT

Scale

No China no Korea

(3) 75

SC

Percentile

Period ending in 2008

(2) 50

M AN U

Baseline (1) 25

1.138*** (0.201) 0.314*** (0.115) 0.143 (0.203)

0.802*** (0.141) 0.123 (0.079) 0.146 (0.116)

0.459*** (0.119) 0.093 (0.057) 0.110 (0.119)

0.336*** (0.122) 0.186*** (0.055) 0.045 (0.109)

0.806*** 0.592*** (0.107) (0.060) -0.447** -0.287*** (0.219) (0.100)

0.360*** (0.081) -0.027 (0.126)

-2.811*** -2.366*** -1.925*** -1.672*** (0.042) (0.039) (0.033) (0.049)

TE D

-2.776*** -2.320*** -1.898*** -1.612*** (0.044) (0.036) (0.032) (0.047)

-2.772*** -2.311*** -1.886*** -1.597*** (0.045) (0.037) (0.032) (0.049)

-2.800*** -2.344*** -1.902*** -1.559*** (0.051) (0.038) (0.041) (0.035)

(22) 50

(23) 75

Using lagged values (24) 90

(25) 25

(26) 50

(27) 75

(28) 90

-0.135** -0.060 -0.020 0.091* (0.056) (0.058) (0.054) (0.049) -0.116** -0.092** -0.012 -0.013 (0.048) (0.037) (0.023) (0.027) 0.031* 0.008 0.009 0.012 (0.016) (0.011) (0.009) (0.012) -0.131 -0.170** -0.253*** -0.147 (0.089) (0.072) (0.098) (0.089) -1.697 -0.384 -0.021 0.769 (1.622) (0.782) (0.595) (0.897) 0.237*** 0.237*** 0.272*** 0.281*** (0.060) (0.042) (0.031) (0.043) 0.091* 0.127*** 0.131*** 0.097** (0.052) (0.034) (0.031) (0.039) 0.239*** 0.223*** 0.211*** 0.190*** (0.028) (0.024) (0.024) (0.036) -0.060*** -0.059*** -0.072*** -0.108*** (0.016) (0.009) (0.010) (0.014) 0.223*** 0.210*** 0.241*** 0.238*** (0.085) (0.069) (0.057) (0.069) 0.120 0.151 0.164* 0.092 (0.128) (0.104) (0.092) (0.118) 1.142*** 0.728*** 0.453*** 0.359* (0.196) (0.157) (0.137) (0.188)

-0.175*** -0.129** 0.009 0.062 (0.058) (0.063) (0.051) (0.045) -0.117** -0.095*** -0.012 0.023 (0.046) (0.033) (0.025) (0.023) 0.023 0.005 -0.007 -0.022 (0.017) (0.013) (0.012) (0.018) 0.059 -0.061 -0.028 -0.137** (0.088) (0.088) (0.069) (0.056) 0.105 0.497 -0.142 -0.651 (1.162) (0.815) (0.626) (1.131) 0.326*** 0.283*** 0.303*** 0.294*** (0.061) (0.048) (0.034) (0.034) 0.035 0.065* 0.088*** 0.068 (0.052) (0.035) (0.030) (0.044) 0.262*** 0.207*** 0.199*** 0.193*** (0.031) (0.028) (0.025) (0.039) -0.061*** -0.071*** -0.082*** -0.107*** (0.015) (0.012) (0.011) (0.018) 0.259*** 0.276*** 0.281*** 0.200*** (0.089) (0.085) (0.059) (0.069) 0.094 0.079 0.062 0.112 (0.125) (0.117) (0.091) (0.124) 0.788*** 0.653*** 0.396*** 0.265** (0.124) (0.096) (0.083) (0.114)

0.763*** 0.619*** 0.384*** 0.270* (0.200) (0.154) (0.093) (0.152) -0.460* -0.248 -0.163 -0.187 (0.251) (0.204) (0.175) (0.215) -2.734*** -2.298*** -1.887*** -1.628*** (0.052) (0.043) (0.032) (0.051)

-2.735*** -2.283*** -1.870*** -1.583*** (0.045) (0.038) (0.032) (0.037)

0.099 (0.099) 0.227 (0.170)

Relative reserves 1 x RIM Constant

(21) 25

-2.826*** -2.413*** -2.044*** -1.787*** (0.050) (0.028) (0.031) (0.038)

AC C

EP

Observations 1,009 1,009 1,009 1,009 921 921 921 921 969 969 969 969 977 977 977 977 1,009 1,009 1,009 1,009 958 958 958 958 965 965 965 965 Pseudo R2 0.310 0.300 0.299 0.314 0.306 0.301 0.297 0.308 0.304 0.292 0.291 0.307 0.314 0.285 0.277 0.306 0.348 0.357 0.350 0.350 0.328 0.315 0.313 0.324 0.307 0.290 0.290 0.311 Source. Authors' estimates. Notes. The table reports coefficients of various robustness checks and extensions for the quantile regression baseline specifications in Table 5 columns 1-4. RIM countries are China, Indonesia, Korea, Malaysia, Philippines, Thailand, and Vietnam. Quantile regression standard errors are obtained using bootstrapping with 1000 replications (*, **, *** indicate significance at the 10, 5, and 1 percent levels, respectively).

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, T7, 20-05-2014

35

1

1980-2010

2 3 4 5 6 7

1980-1997 1998-2004 2005-2010 2000-2007 2007 2010 Reserves

Full sample 13.9

RIM 18.5

non-RIM 12.9

8.5 15.1 21.6 17.3 21.2 24.6

10.8 22.2 30.4 24.9 32.1 32.7

7.9 13.8 20.0 15.9 19.1 23.0

ACCEPTED MANUSCRIPT

30

25

20

15

10

5

0

1980-1997

Full sample

RIM

1998-2004

2005-2010

non-RIM

Source: World Economic Outlook and authors' calculations. Figure 2. Proxies for precautionary and mercantilist motives for RIM and non-RIM countries.

RI PT

1980-2010

RIM

1980-2010 110

90

90

70

70

50

50

30

30

10

10

-10

-10

-30

-30

M2/GDP

ST Debt/GDP

Misalignment

Imports/GDP

M2/GDP

ST Debt/GDP

Misalignment

2005-2010

EP

TE D

1998-2004

Imports/GDP M2/GDP ST Debt/GDP 1980-2010 37.9 70.3 12.8

AC C

Imports/GDP

M AN U

110

SC

1980-1997

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, F1&F2

1980-1997 1998-2004 2005-2010 2000-2007 2007 2010

30.4 45.4 48.7 48.4 49.9 45.9

52.2 89.9 98.5 90.6 97.3 107.1

13.1 13.9 10.9 12.0 10.9 9.8

Non-RIM Imports/GDP M2/GDP ST Debt/GDP 1980-2010 29.2 39.8 14.1 1980-1997 1998-2004 2005-2010 2000-2007 2007 2010

23.0 33.0 37.5 35.5 39.7 35.1

Imports/GDP M2/GDP ST Debt/GDP Misalignment

36.1 37.9 49.3 42.3 49.6 52.9

10.7 15.9 17.9 17.7 19.5 17.1

1980-2010 RIM non-RIM 37.9 29.2 70.3 39.8 12.8 14.1 -7.6 -0.7

Figure 3. Reserves Distribution, various samples and periods.

0

M AN U

20

SC

40

RI PT

60

80

100

120

ACCEPTED MANUSCRIPT

100

AC C

EP

80 60 40 0

20

RIM sample

TE D

120

Full sample non-RIM sample

1980-1997

1998-2004 Full sample non-RIM sample

2005-2010 RIM sample

Source: World Economic Outlook and authors' calculations. C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, F3

ACCEPTED MANUSCRIPT

95% lowerpredavg 0.0890016 0.0584162 0.1457506 0.0657728 0.3044271 0.1605755 0.1840171 0.0498044 0.1369991 0.1777361 0.0959499 0.1125288 0.180826 0.1290074 0.0887775 0.1143487 0.2887144 0.0753902 0.1024185 0.1261185 0.214312 0.0882571 0.2343176 0.2207466 0.3376947 0.2407883 0.3557285 0.1401188 0.1322077 0.0613057 0.1251988 0.0994235 0.1240478 0.1544337 0.1431847 0.1384261 0.1006815 0.234693 0.1435912 0.0757054 0.1302408 0.0893418 0.0954819 0.1606919

TE D

95% upperpredavg 0.1588474 0.1087722 0.2149184 0.1467765 0.4790427 0.2998862 0.3959261 0.094916 0.2018663 0.2777755 0.1598504 0.2132898 0.3589288 0.2033214 0.1387815 0.1796463 0.5127646 0.1333412 0.1756772 0.2030621 0.3685752 0.1515247 0.5169607 0.3688084 0.9163291 0.3829617 0.6030923 0.2865647 0.2094764 0.0856976 0.2296926 0.1941801 0.1826876 0.2571673 0.2469251 0.2830342 0.181643 0.3820334 0.2299561 0.1412732 0.1998084 0.1526867 0.1602181 0.2818893

RI PT

actual 0.1708871 0.1802125 0.2971453 0.131379 0.4184182 0.1025381 0.43825 0.0987073 0.1562746 0.2333393 0.0621537 0.0629204 0.2326288 0.1041809 0.1331378 0.1214291 0.1732704 0.2323837 0.1274926 0.1456717 0.4259632 0.1541601 0.2498721 0.1933319 0.5355707 0.1937784 0.540543 0.0849427 0.3212096 0.0988835 0.0977612 0.2512418 0.2113184 0.1484754 0.2190788 0.359737 0.1042527 0.3454744 0.2020096 0.1133685 0.2230417 0.1717874 0.1097619 0.3339515

SC

fitted 0.118902 0.0797124 0.1769873 0.0982543 0.3818816 0.2194411 0.2699207 0.0687549 0.1662994 0.2221953 0.1238452 0.1549233 0.254762 0.1619567 0.1109985 0.1433259 0.3847629 0.1002627 0.1341365 0.1600309 0.2810517 0.1156423 0.3480417 0.28533 0.5562729 0.3036654 0.4631815 0.2003824 0.1664163 0.0724828 0.1695796 0.1389463 0.150539 0.199287 0.1880317 0.1979377 0.1352335 0.2994338 0.1817132 0.1034173 0.1613171 0.116796 0.1236848 0.2128317

M AN U

country abbr Argentina ARG Armenia ARM Bosnia and Herzegovina BIH Brazil BRA Bulgaria BGR Chile CHL China CHN Colombia COL Costa Rica CRI Croatia HRV Dominican Republic DOM Ecuador ECU Egypt EGY El Salvador SLV Georgia GEO Guatemala GTM Hungary HUN India IND Indonesia IDN Jamaica JAM Jordan JOR Kazakhstan KAZ Korea KOR Latvia LVA Lebanon LBN Lithuania LTU Malaysia MYS Mexico MEX Morocco MAR Pakistan PAK Panama PAN Peru PER Philippines PHL Poland POL Romania ROM Russia RUS South Africa ZAF Thailand THA Tunisia TUN Turkey TUR Ukraine UKR Uruguay URY Venezuela VEN Vietnam VNM

EP

year 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007

AC C

2007 ifs 612 213 963 223 918 228 924 233 238 960 243 248 469 915 258 944 534 536 436 343 439 916 542 941 446 946 962 548 273 686 564 283 293 566 964 968 922 199 524 578 744 186 926 298

ACCEPTED MANUSCRIPT

Figure 4. Actual vs. predicted reserves in 2007 (no differentiation between

RI PT

60%

M AN U

SC

50%

AC C

CHN JOR

Actual 2007

40%

RUS THA VNM MAR

30%

TE D

5% 10% 12% 3% 4% -12% 17% 3% -1% 1% -6% -9% -2% -6% 2% -2% -21% 13% -1% -1% 14% 4% -10% -9% -2% -11% 8% -12% 15% 3% -7% 11% 6% -5% 3% 16% -3% 5% 2% 1% 6% 5% -1% 12%

EP

2 deviations above 0 above 0 above 0 0 0 0 0 0 below above 0 above 0 0 0 0 0 0 below 0 below 0 0 0 below 0 0 0 0 0 below above 0 0 0 0 0 above 0 above 0 0 0 0 below 0 0 0 below 0 0 0 below above 0 above 0 0 below above 0 above 0 0 below 0 0 above 0 0 0 0 0 0 0 0 0 above 0 above 0 0 0 above 0

BIH

PER IND

UKRROM PHL TUN

20%

ARM

URY ARG KAZ

LVALTU

CRI JAM POL

GEOIDN BRA GTM TURVEN ZAF SLV PAK COL PAN

10%

HRV EGY

CHL MEX

DOM ECU

0% 0%

10%

20%

30%

Fitted Notes. Slope coefficients on the regressors are 2007 constrained to be the s are marked in squares. RIM countries whose actual reserves are abov are marked in triangles. Source. World Economic Outlook and authors' calculations.

ACCEPTED MANUSCRIPT

MYS

LBN

SC

JOR

RI PT

(no differentiation between RIM and non-RIM countries).

M AN U

BGR

KOR

LVALTU

AC C

EP

HUN

TE D

THA

40%

50%

60%

Fitted ors are 2007 constrained to be the same for RIM and non-RIM countries. RIM countries hose actual reserves are above 2 standard deviations of the predicted reserves thors' calculations.

ACCEPTED MANUSCRIPT

95% upperpredavg 0.1683724 0.1178431 0.2245893 0.1581646 0.5085692 0.3264442 0.3979149 0.0980851 0.2122762 0.3029514 0.1657261 0.233057 0.4070862 0.2137192 0.145798 0.1915981 0.5499991 0.1424282 0.1358717 0.2145656 0.3947494 0.1592941 0.4196204 0.3943273 0.9242514 0.4027994 0.5026607 0.3042008 0.2235557 0.0853485 0.2501926 0.2121346 0.170585 0.2676417 0.2634527 0.3046997 0.1927934 0.3420666 0.2451642 0.144572 0.2006819 0.1660481 0.1699353 0.336609

RI PT

95% lowerpredavg 0.0854813 0.0584395 0.1375752 0.0612064 0.2798164 0.1550023 0.281543 0.0498127 0.1334131 0.170421 0.0958388 0.1048782 0.144639 0.1247526 0.0846636 0.1047205 0.2648151 0.0575095 0.1114667 0.1234094 0.1868622 0.0846753 0.1921405 0.2257219 0.3374429 0.2412751 0.3914654 0.1247551 0.1103673 0.0538796 0.1214057 0.0929141 0.1419894 0.1462808 0.1368224 0.1230874 0.0909323 0.2700559 0.129503 0.0734708 0.1173334 0.0915688 0.0907023 0.2298786

SC

actual 0.1708871 0.1802125 0.2971453 0.131379 0.4184182 0.1025381 0.43825 0.0987073 0.1562746 0.2333393 0.0621537 0.0629204 0.2326288 0.1041809 0.1331378 0.1214291 0.1732704 0.2323837 0.1274926 0.1456717 0.4259632 0.1541601 0.2498721 0.1933319 0.5355707 0.1937784 0.540543 0.0849427 0.3212096 0.0988835 0.0977612 0.2512418 0.2113184 0.1484754 0.2190788 0.359737 0.1042527 0.3454744 0.2020096 0.1133685 0.2230417 0.1717874 0.1097619 0.3339515

M AN U

fitted 0.1199696 0.0829861 0.1757781 0.0983905 0.3772347 0.2249435 0.334709 0.0698991 0.1682868 0.2272208 0.1260278 0.1563413 0.2426531 0.1632851 0.1111026 0.1416483 0.3816387 0.090504 0.1230657 0.162725 0.2715948 0.1161391 0.2839473 0.2983426 0.5584641 0.3117458 0.4435925 0.1948091 0.1570772 0.0678125 0.1742837 0.1403933 0.1556318 0.1978657 0.1898584 0.1936613 0.1324053 0.303936 0.1781839 0.1030622 0.1534493 0.1233078 0.1241512 0.2781712

TE D

country abbr Argentina ARG Armenia ARM Bosnia and Herzegovina BIH Brazil BRA Bulgaria BGR Chile CHL China CHN Colombia COL Costa Rica CRI Croatia HRV Dominican Republic DOM Ecuador ECU Egypt EGY El Salvador SLV Georgia GEO Guatemala GTM Hungary HUN India IND Indonesia IDN Jamaica JAM Jordan JOR Kazakhstan KAZ Korea KOR Latvia LVA Lebanon LBN Lithuania LTU Malaysia MYS Mexico MEX Morocco MAR Pakistan PAK Panama PAN Peru PER Philippines PHL Poland POL Romania ROM Russia RUS South Africa ZAF Thailand THA Tunisia TUN Turkey TUR Ukraine UKR Uruguay URY Venezuela VEN Vietnam VNM

EP

year 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007 2007

AC C

2007 ifs 612 213 963 223 918 228 924 233 238 960 243 248 469 915 258 944 534 536 436 343 439 916 542 941 446 946 962 548 273 686 564 283 293 566 964 968 922 199 524 578 744 186 926 298

ACCEPTED MANUSCRIPT

Figure 5. Actual vs. predicted reserves 2007 (differentiating between RIM a

RI PT

60%

SC

50%

M AN U

Actual 2007

40%

AC C

JOR

RUS THA VNM MAR

30%

TE D

5% 10% 12% 3% 4% -12% 10% 3% -1% 1% -6% -9% -1% -6% 2% -2% -21% 14% 0% -2% 15% 4% -3% -11% -2% -12% 10% -11% 16% 3% -8% 11% 6% -5% 3% 17% -3% 4% 2% 1% 7% 5% -1% 6%

EP

2 deviations above 0 above 0 above 0 0 0 0 0 0 below above 0 above 0 0 0 0 0 0 below 0 below 0 0 0 below 0 0 0 0 0 below above 0 0 0 0 0 above 0 0 0 0 0 0 below 0 0 0 below above 0 0 below above 0 above 0 0 below above 0 above 0 0 0 0 0 above 0 0 0 above 0 0 0 0 0 above 0 above 0 0 0 0 0

BIH

PER IND

UKR ROM PHL TUN

20%

ARM

KOR

URY ARG KAZ

HRV EGY LVA

CRI JAM POL

GEO BRA IDN GTM TURVEN ZAF SLV CHL PAK COL PAN MEX

10%

DOM ECU

0% 0%

10%

20%

30%

Fitted 2007

Notes. Slope coefficients are allowed to vary between the RIM and non squares. RIM countries whose actual reserves are above 2 standard d triangles. Source. World Economic Outlook and authors' calculations.

ACCEPTED MANUSCRIPT

MYS

RI PT

ifferentiating between RIM and non-RIM countries).

SC

LBN

CHN

M AN U

BGR

THA VNM

TE D

KOR

LTU LVA

AC C

EP

HUN

40%

50%

60%

Fitted 2007

vary between the RIM and non-RIM samples. RIM countries are marked in serves are above 2 standard deviations of the predicted reserves are marked in thors' calculations.

AC C

EP

TE D

M AN U

SC

RI PT

ACCEPTED MANUSCRIPT

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, F6

ACCEPTED MANUSCRIPT

Figure 7. Non-RIM countries cumulative differences decomposition 1990-2010 (specification with both dummy and interactio Source: Authors' calculations.

RI PT

1.4 1.3

SC

1.2 1.1

M AN U

1.0 0.9 0.8

TE D

0.7 0.6

0.4

AC C

0.3

EP

0.5

0.2 0.1 0.0

-0.1 -0.2 1990

1992

1994

1996

1998

2000

Capital account Current account MercantilistF7 C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX,

ACCEPTED MANUSCRIPT

Figure 8. China cumulative differences decomposition 1990-2010 Source: Authors' calculations.

RI PT

1.8 1.6

SC

1.4 1.2

M AN U

1.0 0.8 0.6

TE D

0.4 0.2

AC C

-0.2

EP

0.0

-0.4 -0.6 -0.8 -1.0 -1.2

1990

1992

1994

1996

1998

2000

C:\pdfconversion\WORK\authordoc\Ghosh_Ostry_Tsangarides_TablesAndFigures_Apr2014_v1_JIMF.XLSX, F8

ACCEPTED MANUSCRIPT

Highlights

AC C

EP

TE D

M AN U

SC

RI PT

-We compare demand for reserves determinants of Pacific Rim countries and their peers. - RIMs hold more reserves overall, but fewer reserves for capital account shocks. - RIMs hold more reserves for current account shocks. - Mercantilism more important for RIMs but explains small share of recent reserve rise. - Shifting motives for holding reserves across time and the reserves’ distribution.