Estimating the demand for reserve assets across diverse groups of countries

• DOI: http://doi.org/10.1111/roie.12399
• Published date: 01 August 2019
• Date: 01 August 2019
• Author: Rina Bhattacharya,Katja Mann,Mwanza Nkusu

822

|

wileyonlinelibrary.com/journal/roie Rev Int Econ. 2019;27:822–853.

© 2019 John Wiley & Sons Ltd

Received: 24 May 2018

|

Revised: 19 October 2018

|

Accepted: 10 February 2019

DOI: 10.1111/roie.12399

ORIGINAL ARTICLE

Estimating the demand for reserve assets across

diverse groups of countries

RinaBhattacharya1

|

KatjaMann2

|

MwanzaNkusu1

1International Monetary Fund,

Washington, D.C.

2Copenhagen Business School,

Frederiksberg, Denmark

Correspondence

Katja Mann, Copenhagen Business School,

Frederiksberg, Hovedstaden, Denmark.

Email: kma.eco@cbs.dk

Abstract

This paper takes a fresh look at the determinants of the holding

of reserves with the aim of highlighting similarities and differ-

ences among emerging markets (EMs), advanced economies

(AEs), and low‐income countries (LICs). We apply two panel

estimation techniques: fixed effects (FE) and common corre-

lated effects pooled mean group (CCEPMG). FE regression

results suggest that precautionary savings’ motives, both cur-

rent account‐ and capital account‐related, are generally the

most important determinants of reserves’ holding for all coun-

try groups. Nonetheless, there is considerable heterogeneity

across country groups and over time. The intertemporal mo-

tive, a novelty of this paper, has gained importance every-

where. The CCEPMG results confirm the importance of

precautionary motives and suggest that current account mo-

tives matter only for EMs and LICs and capital account mo-

tives matter for all groups while being more relevant for EMs.

The CCEPMG results also point to the importance of taking

into account the heterogeneous impact of unobserved com-

mon factors that affect coefficient estimates and the dynamic

process through which reserves adjust to changes.

JEL CLASSIFICATION

C23, E58, F31, G01

1

|

INTRODUCTION

Since the early 2000s, the amount of reserve assets held by countries’ monetary authorities has been

rising all over the world, and so has researchers’ interest in the topic. A large literature has tried to

|

823

BHATTACHARYA eT Al.

determine the optimal amount of reserves or to explain the actual amounts held by countries, but since

the GFC, empirical studies had been mostly focused on emerging markets (EMs).1

The importance

of holding adequate reserves for advanced economies (AEs) and, low‐income countries (LICs) with

limited access to international financial markets was not featured prominently in analyses. Research

papers that extend their analyses beyond EMs either consider EMs jointly with other countries or dis-

tinguish between country groups along a different dimension than we do.2

The literature puts forth a number of arguments to rationalize the greater focus on reserves’ hold-

ing in EMs relative to the remaining two country groups (see International Monetary Fund [IMF],

2011, and further references therein). It suggests that the EMs’ country group has both the need to

hold reserves, facing frequent balance of payments disruptions, and the means to acquire reserves in

international markets. For countries in the AEs’ group, it is argued that their external positions are

too large to be insurable and that they do not need to hold a large amount of reserves in the first place,

given that their trade and capital flows are more stable and their financial markets are more liquid.

Moreover, these countries can rely upon other forms of insurance, for example swap lines. Some of

the AEs are reserve issuers themselves, and hence the traditional motives of reserves’ holding do not

apply to them.3

For LICs, it is argued that they have limited access to financial markets and that the

balance of payments risks they face are therefore different in nature from those of EMs.

This paper distinguishes itself from most of the related literature that not only implicitly assumes

that existing models mostly applied to EMs are of only limited relevance to AEs and LICs, but also

abstracts from cross‐country heterogeneity and cross‐sectional dependence. Our stance that models of

the determinants of reserves used for EMs could be relevant for AEs and LICs stems from two main

reasons. First, as illustrated in Figure 1, AEs and LICs are not different from EMs when it comes to

both the level and the trend of reserves’ holding. In particular, relative to GDP, reserves in EMs, AEs,

and LICs have been following a similar trend, rising from about 1.5% to around 3% of GDP between

1980 and 2014. Second, both AEs and LICs can benefit from holding adequate reserves to insure

against adverse shocks. For AEs, while episodes of market dysfunction are less frequent and of shorter

duration than in EMs, they can have very disruptive effects on macroeconomic performance as seen

during the global financial crisis (GFC).4

For LICs, reserves can help smooth disruptions to domestic

demand associated with the volatility of export proceeds and other external flows. According to IMF

(2013a), LICs that had higher levels of international reserves in the year preceding an adverse external

shock event were better able to cushion economic activity. In our dataset, LICs are the country group

with the highest volatility of terms of trade and volatility of capital inflows (see Appendix S1, Table

A3, in the Supporting Information—for access see note at the end of the paper). Against this back-

ground, we argue that it does make sense to apply models developed for EMs also to AEs and LICs to

investigate what drives their demand for reserve assets. As detailed below, we also take into account

the fact that countries included in panels are very heterogeneous, including across groups, and that

their demands for reserves could be differently affected by unobserved common factors that, in turn,

may be correlated with the regressors. If ignored, these panel properties could lead to biased estimates

and spurious inference.

Our aim is to investigate the determinants of the demand for reserves and highlight possible simi-

larities and differences in the motives for reserves’ holding across the three country groups, and to see

how these motives have evolved over time, using various econometric specifications. Our economet-

ric strategy is twofold. First, we carry out static panel data estimations, which include country fixed

effects (FE) to capture time‐invariant country‐specific factors. We compare estimation results both

for the complete sample period of 1980 to 2014, and for three subperiods to find out how the motives

for reserves’ holding may have been affected by the experiences of two severe crises, the Asian crisis

and the GFC. Second, we apply a relatively new estimation technique, common correlated effects

824

|

BHATTACHARYA eT Al.

pooled mean group (CCEPMG) estimation that allows for addressing the heterogeneity of the coun-

tries included in the sample and other issues that have a bearing on the robustness of the estimates.

In particular, CCEPMG estimation takes into account (a) possible existence of common unobserved

factors that affect the accumulation of reserves while not being captured by explanatory variables and

give rise to cross‐sectional dependence in the regressions’ residuals, and (b) the fact that reserves are

likely to adjust to changes in the explanatory variables through a dynamic process. In capturing these

dynamics, the CCEPMG methodology, unlike FE, makes a clear distinction between the short‐run and

the long‐run impacts of the explanatory variables. It allows for the slope parameters to differ in the

short run, while assuming convergence in the long run.

In applying these different estimation techniques to a sample of all countries and separately for

EMs, AEs, and LICs, we obtain a detailed picture of the differences and commonalities in the motives

for reserves’ holding of the three country groups. The results are useful for policymakers and can in-

form the general debate about how models of reserves’ holding, originally developed for EMs, could

be applicable to AEs and LICs. We make at least three further contributions to the literature. First, to

the best of our knowledge, we are the first to apply the CCEPMG estimation technique to the analysis

of reserves’ holding, thereby separating short‐run from long‐run determinants of the accumulation

of reserves. Second, we consider new explanatory variables for reserves’ holding. In particular, in

addition to the standard variables analyzed in previous studies, we investigate intertemporal savings

motives by including a measure of the change in the age dependency ratio, and also analyze whether

an economic crisis situation (as proxied in this study by participation in an IMF program, especially

for AEs and EMs) makes a difference to reserves’ holding. Third, in the fixed‐effect regressions, we

test for and address cross‐sectional dependence and serial correlation, which the previous literature on

the determinants of reserves has been silent upon.

FIGURE 1 Measures of reserves for AEs, EMs and LICs [Colour figure can be viewed at wileyonlinelibrary.

com]

20 40 60

percent

1980 1985 1990 1995 2000 2005 2010 2015

Year

EMs AEsLICs

Reserves relative to imports

020 40 60

percent

1980 1985 1990 1995 2000 2005 2010 2015

Year

EMsAEs LICs

Reserves relative to broad money

123

percent

1980 1985 1990 1995 2000 2005 2010 2015

Year

EMs AEsLICs

Reserves relative to GDP

05 10

US$ bn (LICs)

050 100 150

US$ bn (EMs and AEs)

1980 1985 1990 1995 2000 2005 2010 2015

Year

EMsAEs LICs

Real reserves