Content uploaded by Laston Manja
Author content
All content in this area was uploaded by Laston Manja on Mar 17, 2022
Content may be subject to copyright.
1
Manja et al: Determinants of demand for international reserves in the SADC region
African Review of Economics and Finance | ISSN 2042-1478 |
Determinants of demand for international reserves in the
SADC region
Laston Petro Manja*, exLey BD siLuMBu anD regson DC Chaweza
Department of Economics, University of Malawi, Zomba, Malawi
*Corresponding author email: lmanja@unima.ac.mw
Abstract
What explains the unprecedented high levels of foreign reserves accumulated
in the Southern African Development Community (SADC)? This study seeks to
econometrically establish the key determinants of such high levels of reserves.
To do so, the study adopts two panel estimation techniques: the Blundell-Bond
System Generalized Method of Moments (GMM) and the Bias-Corrected Least
Squares Dummy Variable (LSDVC). The regression results show evidence of
precautionary (mainly reserves volatility) but not mercantilist motives in the
SADC region. Inertial and opportunity cost elasticities of reserves demand are
observed to be inelastic in both econometric models at about 0.85 and -0.70
respectively. The GMM identies membership to the Common Monetary Area
(CMA) as a signicant factor, as member countries of these sub-groupings
demand more reserves to meet subsequent reserve targets. In addition, while
GMM results reveal a U-shaped relationship between reserves hoarding and
national income, the LSDVC nds income to have no inuence on demand, an
observation attributed to strength of instruments within the GMM. From the
regression results, the study nds that empirical works on demand for reserves
need to clearly distinguish between the alternative measures of reserves - either
including or excluding gold; a measurement difference that has typically been
ignored in the literature. Various policy implications are drawn from the results.
Keywords: International Reserves; Buffer Stock Model; GMM; LSDVC; SADC.
JEL Classication: C23; E58
Article history:
Received: 28 July, 2020
Accepted: 12 March, 2022
African Review of Economics and Finance
2
1. Introduction
One notable characteristic of the global ofcial foreign reserve accumulation
is that although central banks of Asian economies hoard the biggest proportion
of global foreign reserves, African economies’ central banks particularly those
in the Southern African Development Community (SADC) have experienced
an increasing share of reserves overtime. Precisely, since the year 2000, the
increasing demand for reserves has been unprecedented with the regional average
sometimes reaching about 22 percent of GDP. The rationale for such reserve
hoardings, nonetheless, remains a debate for all countries (Ra, 2007). Though
these countries gradually amassed enormous reserves with time, all economies
including those in the SADC have become more market-oriented than centrally
regulated. This has deed key tenets of the central bank intervention model
which postulates that the wish to hold international reserves is justied only
when countries pursue xed exchange rate regimes in which case central banks
are expected to intervene in the market to counteract any speculative attacks
and defend parities (Bastourre, Carrera & Ibarlucia, 2009; Batten, 1982; Flood
& Marion, 2002). This means that as SADC countries become more liberalized,
other things held constant, stocks of reserves are expected to decline as shocks
will be self-absorbing and hence there will be little or no need for market
intervention. However, in spite of increased oating of currencies, reserve
hoarding has been more pronounced for instance with the national conicts
around 1995-2003 and the global nancial crisis of 2007-2009. This is seen in
Figure 1 which shows the trend of reserve hoarding in terms of both the average
ratio of reserves (excluding gold) to GDP and total reserves from the year 1995
to 2019.
Against this background, we investigate factors that inuence demand for
international reserves in the SADC region. International reserves are dened
as those external assets that are readily available to and controlled by monetary
authorities for direct nancing of payments imbalances through intervention
in exchange markets in order to inuence the currency exchange rate, among
others (IMF, 2009). The study nds that hoarding of international reserves in the
SADC region is mainly inuenced by the need to take precaution, rather than
for mercantilist motives, and to satisfy sub-regional reserve targets.
To esh out these ndings, the rest of the paper shows how the research was
carried out. The next section provides a brief background so as to motivate the
study within the context. While Section 3 reviews existing empirical literature
on the demand for reserves, Section 4 provides an overview of the SADC region
3
Manja et al: Determinants of demand for international reserves in the SADC region
and a synopsis of international reserves both on the global front and among
SADC countries. With the econometrics approach and results presented in
Section 5, Section 6 concludes the paper.
2. Theoretical background
Foreign reserves are hoarded around the world, inter alia, to manage monetary
and exchange rate policies; to abate the negative effects of external shocks
especially in the face of limited access to international nancial markets; to
nance imports and to meet foreign debt for borrowing economies; as well as
to reduce the cost of borrowing (Sinem & Nebiye, 2014; Bhattacharya, Mann
& Nkusu, 2019). However, these benets of reserve hoarding come at a cost.
Theoretically, based on the quantity theory of money, if monetary expansion
that corresponds to the hoarding of reserves is not fully sterilized and exceeds
growth of money demand, reserves hoarding tends to be inationary, both for
xed and exible exchange rate regimes (Steiner, 2009). Additionally, holding
excessively large quantities of reserves is costly because the yield from them,
usually invested in bonds, is much lower than the opportunity cost of holding
such reserves, although this cost is lower than the potential cost of another
crisis (Ra, 2007). Massive reserves accumulation also represents large foregone
consumption and investment in countries which would possibly have good
Figure 1: trenD in average (% oF gDP) anD totaL reserves For saDC (1995-2019)
- exCLuDing ziMBaBwe
Note: Reserves (% of GDP), LHS, and Total Reserves (Millions, US$), RHS
Source: IMF, IFS.
African Review of Economics and Finance
4
growth prospects. Further, the IMF (2009) observed existence of the Trifn
dilemma, whereby reserves hoarding poses a risk on stability of the international
monetary system (International Monetary Fund (IMF), 2009; Trifn, 1964).
The above points show that reserves hoarding markedly has its pros and
cons. Yet, regional motives for the tendency remain unclear. Historically, a
number of studies have been conducted, and the determinants mostly fall within
four categories: transactionary, precautionary (risk factors), mercantilist or
collateral asset motives. Analysis of the studies shows that an understanding
of the determinants is essential for reserve management, especially since there
are plans to launch a SADC central bank in the near future. Some previously
conducted studies in Africa observe that reserves are hoarded to satisfy
domestically set legal restrictions, to sustain domestic currency credibility in
the face of crises and to help with foreign borrowing (SADC, 2011), among
others. However, the continued reserve buildup is observed to have a weak return
effect besides posing a risk on stability of the international monetary system,
according to the IMF (2009). Such studies, however, have mainly considered
traditional variables, mostly transactionary, without concern for other areal-
specic determinants (Bhattacharya et al., 2019; Sanusi, Meyer & Hassan, 2019;
Elhiraika & Ndikumana, 2007). The studies have also ignored differences in the
alternative measures of reserves – either including or excluding gold. Why has
reserve accumulation in the SADC region been on the rise when investments in
some of the countries are below optimal? Is it self-insurance, protectionism or
other factors? Does the measure of reserves used in empirical analysis matter in
understanding the determinants? This study set out to answer these questions.
Compared to prior studies, this paper offers not only improvements in the scope
of countries covered within the region so as to have a better view of the entire
SADC region, but it also uses improved econometric techniques and variable
measurements, alongside a more robust model specication. The buffer stock
model of money demand (Frenkel & Jovanovic, 1981; Ra, 2007; Sinem & Nebiye,
2014) is applied in this case, dening reserve demand as a Weiner process.
3. Previous empirical studies
While many studies have dedicated efforts to understand international reserves
since the eld gained ground in the mid-1900s, studies done for African
developing economies remain scanty, with focus mainly on those countries
outside of Africa. These studies differ in terms of methodologies, assumptions
and what they focus on.
5
Manja et al: Determinants of demand for international reserves in the SADC region
The empirics of demand for international reserves were initialized mainly
by Machlup (1966) and Heller (1968) who found variability of trade to be a
much better measure of reserve demand, and not necessarily its level. It was
around this period that studies on the determinants of demand for international
reserves sparked remarkable interest amongst researchers. One notable study
was by Heller (1966) who revealed that international reserves are hoarded to
reduce adjustment costs that would be incurred if no reserves were held, though
the cost is balanced against the opportunity cost of holding these reserves. This
conrmed key tenets of the buffer stock model of demand. Additionally, a higher
marginal propensity to import (MPI) was found to reduce reserves demand
since the marginal cost of adjustment is lowered. Factors that quickly surfaced
in the search for determinants of international reserves include variability of
international receipts and payments (Clower & Lipsey, 1968; Kenen & Yudin,
1965); propensity to import (Clark, 1970; Heller, 1966; Kelly, 1970); and the
size of international transactions (Frenkel, 1978). However, like in many other
previous studies, propensity to import – representing the degree of openness –
appeared with a positive coefcient, contrary to popular expectations.
Later, Frenkel (1981) in a stochastic framework to explore optimal international
reserves used the Ordinary Least Squares (OLS) estimation technique for 22
developing countries from 1971 to 1975 and found imports as well as opportunity
and adjustment costs to be signicant determinants, in line with both Hume’s
(1752) price-specie-ow and the buffer stock model. Building on the work by
Heller (1966), Frenkel and Jovanovic (1981) proposed reserve volatility as a
proxy for adjustment costs. Although Frenkel (1983) discovered that international
reserves were increasing even with the oating of currencies in 1973, it was
later observed that the likely cause of the paradox was the prevalent capital
account liberalization (Grimes, 1993). These studies only considered traditional
determinants and focus was mainly on developed economies.
Edwards (1985) later observed that most studies on demand for international
reserves fail to nd the expected negative statistically signicant coefcient
of the opportunity cost because of incorrect measurements of the variable that
were employed in different studies. It was observed that opportunity cost must
be dened as the spread between the interest rate at which countries can borrow
from abroad and the London Interbank Offered Rate (LIBOR). Later studies
include Flood and Marion (2002) who presented a comprehensive extension of
Frenkel and Jovanovic’s (1981) study by modifying not only how volatility is
measured, but also incorporating nancial crises. Volatility was found to have a
African Review of Economics and Finance
6
signicant positive impact on reserve hoardings from this study. Finding similar
results, and surprisingly that countries with exible exchange rate regimes have
higher ratios of reserves to GDP, Bastourre et al. (2009) found trade openness,
regional imitation, persistence, development level and nancial and exchange
rate deregulation to be signicant factors affecting hoarding of reserves.
These results are in line with what was discovered by Aizenman and Marion
(2003), Aizenman and Lee (2007), Cheung and Ito (2009), and Hur and Kondo
(2016) who conrmed precautionary motives and found that the exchange-rate
arrangement, political considerations as well as size of international transactions
and their volatility are biggest determinants of demand for international reserves.
The results slightly contrast ndings by Dooley, Folkerts-Landau and Garber
(2004) who found existence of the mercantilist motive in China.
Bernard (2011) discovered a new contextual factor, called ‘keeping up with
the Joneses’ effect, where Central American countries’ policies were found to
consider reserve accumulation of large emerging markets in Latin America.
However, this effect is not expected for SADC countries as most of them are
just target oriented. More recently, studies have moved towards including more
nancial variables in the models, owing to Obstfeld, Shambaugh and Taylor
(2010) who adopted a nancial stability view for reserves demand. In that spirit,
Jung and Pyun (2015) employed difference-GMM and system-GMM estimators
and found that nancial deepening is positively associated with reserves, whereas
Oktay, Öztunç and Serín (2016) conrmed signicance of scale, precautionary,
mercantilist motives and nancial variables for G-7 countries. Most recently,
Mahraddika (2019) observed that reserve accumulation is positively associated
with domestic private investment in the long-run. Interestingly, Bhattacharya
et al. (2019) and Sanusi et al. (2019) attested to the existence of precautionary
motives, with the former focusing on low-income countries in general and the
latter only focusing on 10 selected southern African countries. These studies
conrm plausibility of, among others, the buffer stock model, thereby directing
on possible determinants of demand. In fact, beyond these previous empirical
works (including the closely related Sanusi et al. (2019)), this study goes a step
ahead for the SADC region by adopting a more exhaustive model specication
with better-dened variables (including the denition of foreign reserves),
covering a longer time frame for almost all SADC countries and employing
better econometric techniques.
7
Manja et al: Determinants of demand for international reserves in the SADC region
4. International reserves in the SADC region
The SADC is an inter-governmental organization with 16 countries as of 2021
including Angola, Botswana, Democratic Republic of Congo (DRC), Eswatini
(formerly Swaziland), Lesotho, Madagascar, Malawi, Mauritius, Mozambique,
Namibia, Seychelles, South Africa, Tanzania, Zambia, Zimbabwe and most
recently (since 2017) the Comoros. For development, the SADC region is guided
by the Regional Indicative Strategic Development Plan (RISDP), rst adopted
in 2003. The rst RISDP was a 15-year regional integration development
framework aimed at deepening integration in the region, detailing a timeline
for the transition of SADC from a free trade area (FTA) (achieved in 2008) to
an Economic and Monetary Union [EMU] (SADC, 2017). Later, the RISDP
2020-2030 was formulated guided by lessons learnt from the Revised RISDP
2015–2020. In this case, economic transition was prearranged to move from the
FTA to Customs Union in 2010, then to common market in 2015 and monetary
union by 2016. Overall, the region was expected to have a single currency by
2018, although other targets were set for percentage growth rates, ination rates,
scal decits, percentages of public debt and current account decits. For the
milestones, it was observed that member states only managed to achieve the
2008 targets for indicators, and macroeconomic convergence (MEC) targets
were best reached for public debt. Interestingly, worst performance for MEC
targets was for international reserves, mainly because the targets were observed
to be highly ambitious (Burgess, 2009; SADC, 2017; Simwaka, 2016).
In terms of economic performance, the SADC region represents the largest
regional structure in sub-Saharan Africa (SSA) in economic terms, being home
to one of Africa’s largest economies, South Africa. The region is the richest,
with a real per capita income higher than the continental average. In spite of
this, the SADC harbors some of Africa’s, and indeed world’s, poorer economies
indicating presence of disparities across the countries. For example, while
Seychelles ranked as a high-income economy using the 2015 World Bank
classication, countries such as Malawi and Tanzania remained as low-income
economies. Apart from the income differences highlighted above, SADC
countries also vary in their economic structures. While on a regional level
economic growth in the SADC is mainly driven by the services sector which
contributes more than half of GDP, some countries are highly dependent on
agriculture. Regionally in 2013, for example, the services, manufacturing and
agriculture sectors contributed respectively about 55, 31 and 14 percentages to
GDP. Nationally, Madagascar, Malawi, Mozambique and Tanzania reported the
African Review of Economics and Finance
8
highest dependence on agriculture (over a quarter), while Seychelles had only 2
percent from agriculture and 82 percent from services due to its vibrant tourism
sector. Angola had the greatest share of industry at about 57 percent. In 2015,
Madagascar and Mozambique had about 23 percent of GDP from agriculture,
with Seychelles still at 2 percent. Regional GDP decompositions by sectors in
2015 are illustrated by Figure 2.
Figure 2: saDC Countries' gDP DeCoMPosition By seCtors For 2015
(in MiLLion us$, Current PriCes)1
Note: 2012 gures used for Malawi data due to date availability.
Source: SADC Statistical Yearbook 2015 and WDI.
The SADC region managed to accumulate international reserves of about US$88
billion by 2016, from about US$51 billion in 2006. In terms of months of import
cover, the statistics indicate that only Angola, Botswana and Mauritius managed
to achieve the 6 months RISDP target in 2016, though Zimbabwe, Malawi, DRC
and Namibia consistently struggle to attain the internationally recommended 3
months of import cover. Nonetheless, the SADC average import cover has been
improving with time since 2011, as captured by Figure 3.
1 Other industry includes construction, mining and quarrying, electricity, gas and water. Services
include transport and communication, nance, insurance, real estate and business activities,
general government services, wholesale and retail trade, restaurant and hotels and other services.
9
Manja et al: Determinants of demand for international reserves in the SADC region
Figure 3: MaCroeConoMiC ConvergenCe: saDC average iMPort Covers
Source: SADC Database.
Figure 3 shows that the regional average import cover has consistently been
below the RISDP target, though the region could soon meet the convergence
criterion in terms of import cover. This is the case though some of the countries
(such as Zimbabwe) are far from meeting the national target level.
5. Economic approach
Having seen the qualitative relationship between various factors and reserves
hoarding, this section quantitatively explores the determinants of demand for
international reserves in the region. These determinants are mainly classied as
scale, precautionary or mercantilist factors. Scale variables included are real GDP
per capita (GDPc) and its square (GDPc2). Precautionary variables, capturing the
desire to self-insure again risks, were represented mainly by degree of openness
(API), capital inows (CapInf) and volatility measures in terms of exchange
rate volatility (ERVOL) and reserves volatility (ResVOL). Mercantilism is
about accumulating reserves in order to prevent or mitigate appreciation of
the local currency with the ultimate goal of increasing export growth. This
means reserves hoarding aims at maintaining export competitiveness and so
is a deliberate policy for an economy to exert negative externalities on trading
partners. This study measured the mercantilist motive using lagged export
growth (ExpGrowth). Other variables also included in the study include ination
(InfRat), the opportunity cost of hoarding reserves (OPP) as well as dummies
African Review of Economics and Finance
10
for membership in the Malawi-Mozambique-Zambia-Tanzania (MMZT) and
the common monetary area (CMADum) regions. These regressors are selected
based on literature and data availability.
5.1. Methodology
The study sought to explore factors that determine demand for international
reserves in the SADC region. Demand for reserves was measured by external
assets that are readily available to and controlled by monetary authorities, as
per the IMF’s (2007) denition of international reserves. Given that this is the
level of ofcial international reserves that monetary authorities (central banks)
wish to hoard, whereby the authority’s wish is a demand function, these assets
have been used in this study and in the literature alike to represent demand
for reserves (Aizenman & Marion, 2003; Aizenman & Lee, 2007). This
study made use of dynamic panel data techniques. Apart from offering more
informative data, increasing degrees of freedom and improving efciency, panel
models allow for the capturing of dynamics of adjustment so as to estimate
intertemporal relations (Hsiao, 2003). These form the benets of panel models,
and the rationale for using the methodology. Capturing dynamics of adjustment
was of essence in this study because reserves are dynamic in nature such that
the level of reserves at any point in time is a function of the immediate previous
level. Additionally, accumulation of reserves is heterogeneous both in time
and across countries. This calls for adoption of panel models so as to control
for individual heterogeneity to avoid obtaining biased results if the data was
simply pooled, ignoring the panel data structure. This would produce consistent
yet inefcient OLS estimates compared to Feasible Generalized Least Squares
(FGLS) estimates (Baltagi, 2005).
5.1.1. Blundell/Bond System GMM Estimator
Since reserves are dynamic in nature, a model of demand for reserves (y) must
have as one independent variable the level of reserves in the previous period, to
be given as in equation 1:
where i = 1, ... , N for N cross-sectional units and t = 1, ... , T for T time-periods;
δ is a scalar and x'it a vector of 1 x K independent variables. In this case, the error
component is assumed to be one-way, such that:
where the error components are independent even with each other; uit denotes the
(1)
(2)
11
Manja et al: Determinants of demand for international reserves in the SADC region
unobservable individual-specic effect and denotes the remainder disturbance.
In the dynamic model, the dynamic component (yi,t-1) is correlated with μi.
This correlation makes the use of OLS or linear static panel (Fixed and Random
effects) models yield biased and inconsistent estimates. As a solution, Arellano
and Bond (AB) (1991) derived a consistent GMM estimator for the model. This
estimator uses orthogonality conditions of the lag component and the remainder
error term to obtain additional instruments for estimation of a dynamic panel
data model. The resulting AB one-step GMM consistent estimator of δ with B,
a matrix of instruments, and G, an MA(1) Balestra matrix, is given by Baltagi
(2005), as:
However, the AB estimator can perform poorly if the autoregressive parameters
or the ratio of variance of the panel-level effect to variance of idiosyncratic error
are too large. This prompted Blundell and Bond (1998), building on the work of
Arellano and Bover (1995), to develop a system estimator that uses additional
moment conditions. This, known as the Arellano-Bover/Blundell-Bond linear
dynamic panel-data estimator, was employed in this study for estimation of
demand for international reserves, because it can capture dynamics of adjustment
and outperforms the AB estimator.
5.1.2. Bias-Corrected Least Squares Dummy Variable (LSDVC) Estimator
It was further observed that the traditional dynamic panel models, including
those by Arellano and Bond (1991) and Blundell and Bond (1998), may produce
biased and imprecise estimates when the number of cross-sectional units is small
or moderately large. An alternative approach to IV-GMM estimation in this
case is the Bias-Corrected Least Squares Dummy Variable (LSDVC) estimator
which was found to outperform the IV-GMM estimators in terms of bias and
root mean squared error (RMSE) (Judson & Owen, 1999). As such, the study
also employed Bruno’s (2005) LSDVC which extends the bias approximations
in Bun and Kiviet (2003) by allowing for estimation of unbalanced panels. In
the SADC context, it is important to accommodate the use of an unbalanced
panel knowing that some countries have missing observations in some periods.
Therefore, the LSDVC was more appropriate than alternative techniques as
used in other studies, such as panel ARDL models (see Sanusi et al. (2019)).
For the LSDVC, from the dynamic model expressed in Equation 1, a selection
(3)
African Review of Economics and Finance
12
indicator rit equals 1 if (yit, xit) is observed and 0 otherwise. That is;
The (possibly) unbalanced dynamic panel data model is expressed as
Consequently, the LSDV estimator with bootstrapped standard errors is dened
by:
Where, is symmetric and idempotent.
Knowing that the LSDVC does not estimate coefcients of time-invariant
regressors, this study made use of both the system GMM (to estimate coefcients
of dummy variables) and the LSDVC. In this case, the LSDVC was used to
check if the system GMM estimates are biased or imprecise.
5.2. Variable measurement
The independent variables employed in the study were dened as follows:
Scale Variables: Real GDP per capita and its square were included to capture a
country i’s level of development at time t and the respective quadratic effects.
Precautionary variables: Volatility measures (reserves and exchange rate
volatility) were measured by the standard deviation of 12 monthly values within
each year, and their inclusion was in consonance with the Prebisch literature
on the consequences of unpredictable uctuations in earnings of raw material
exports. This measurement was adopted because data gaps in some countries
under observation could not permit the estimation of GARCH variances, as was
proposed by Mishra and Sharma (2011). Though, this approach is better than
that taken by Sanusi et al. (2019) who proxied the measure by the nominal
exchange rate. The degree of openness was measured as a ratio of imports to
GDP (average propensity to import – API) which is used in place of marginal
propensity to import. Beyond these, to check for robustness of the results,
capital inows – measured as the percentage of foreign direct investment [FDI,
net inows] to GDP – were also included following after Sanusi et al. (2019).
Opportunity Cost of Holding Reserves: The cost of foregone imports, expressed
in terms of the foregone investment, is measured by the difference between the
real return on reserves and the real return to domestic investments (that is, the
spread between domestic lending interest rate and real US Treasury bill rate)
(4)
(5)
(6)
13
Manja et al: Determinants of demand for international reserves in the SADC region
following Edwards (1985).
Mercantilist Factors are measured using export growth (ExpGrowth) (lagged
twice) as was done by Aizenman and Lee (2005).
Common Monetary Area (CMA) Dummy: This is a dummy taking a value of
1 for country in the Common Monetary Area and 0 for country not in the
Common Monetary Area at any time. Members of the CMA in the sample are
Lesotho, Namibia, South Africa and Swaziland.
MMZT Dummy: This dummy for Malawi, Mozambique, Zambia and Tanzania
was included in the study because of the countries’ preferential treatment in
terms of trade waivers, which is expected to reduce their exposure to risks and
eventually reserves demand. These countries were given a value of 1 for the entire
period, except for Zambia which also took a value of 0 after its reclassication
from low to lower middle-income countries in 2011.
Beyond these measures, the study also controlled for ination, following after
Sanusi et al. (2019), measured as the annual percentage change in consumer prices.
5.3. Empirical specication
The study adopted a specication and functional form by Jung and Pyun (2015)
who, in extension of the buffer stock model of demand, collected widely
regarded important determinants of international reserves from various studies
(Aizenman & Lee, 2007; Cheung & Ito, 2009; Obstfeld et al., 2010; Steiner,
2009). To allow for comparability of results from GMM estimation and the
LSDVC (which does not estimate the intercept coefcient), an intercept-less
model was estimated in both cases. This should not be problematic in this study
given that it is meaningless to talk about all regressors being equal to zero
anyways. The regression equation was specied as follows:
where R is Total Reserves (With or Without Gold) and εit is the disturbance term.
5.4. Econometric results
The study proxies demand for reserves by the level of ofcial international
reserves that monetary authorities (central banks) need to hoard, whereby the
authority’s wish is a demand function. Throughout history, various measurements
have been used, such as the level of international reserves or the import cover
(7)
African Review of Economics and Finance
14
(Edwards, 1984; Flanders, 1971; Frenkel, 1978; Frenkel, 1981). However, while
the level-of-reserves measure does not allow meaningful comparisons across
countries, months of import cover are not only less variable across countries but
also, they may not permit the inclusion of propensity to import as a factor in the
model (Bastourre et al., 2009; Bhattacharya et al., 2019; Cheung & Ito, 2009).
As such, the reserves to GDP ratio was used in this study. Given that within the
SADC some countries may hold signicant amounts of gold reserves (such as
South Africa, DRC and Mauritius), separate models with and without gold were
estimated in this study, to make recommendations for future studies seeking to
model the demand for reserves. This goes beyond the standard tradition of just
focusing on reserves without any regard for gold (Bhattacharya et al., 2019;
Sanusi et al., 2019; Bastourre et al., 2009; Bernard, 2011).
5.5 Data properties
Analyzed using Stata software version 15, the study employed an unbalanced
dynamic panel model for 15 SADC countries (N = 15) from 1980 to 2019 (T
= 40). Choice of years in the sample not only rested on data availability, but
also given that during the Bretton Woods system of xed exchange rates and
immediately thereafter, reserve accumulation was inspired by other motives than
what is of interest in modern studies (Bhattacharya et al., 2019). Out of the 16
SADC countries as of 2019, Zimbabwe was excluded from the analysis mainly
due to issues of data availability. Although the panel is unbalanced, randomness
of such missing observations means this is not worrisome. The data was mainly
sourced from the IMF’s International Financial Statistics (IFS) and the World
Bank’s World Development Indicators (WDI). Data on US Treasury Securities
was sourced from the Federal Reserve2. Appendix A shows a list of series, their
descriptions and sources. All main regressions in the study were based on the
full sample period for the 15 countries.
In standard microeconometric panel data models with N → ∞ and xed T, the
assumption of stationarity of the variables is justiable. However, stationarity
becomes more evident as the time dimension increases. Since the data used in
this study (with T = 40) starts to mimic the characteristics of a macro panel, it
was crucial that unit root properties of the variables employed in the empirical
model be tested in order to avoid the problem of spurious regressions among
non-stationary variables that are not cointegrated. To the best of our knowledge,
it is only Sanusi et al. (2019) who check for stationarity of the variables before
2 https://www.federalreserve.gov/releases/h15/
15
Manja et al: Determinants of demand for international reserves in the SADC region
adopting panel models. Of course, Sanusi et al. (2019) adopt a panel ARDL
technique on their sample of 10 southern African countries from 1990 to 2015.
Aside paying particular attention to the whole SADC region over almost the
entire post-Bretton woods institutions’ regime ahead of the planned regional
central bank, this study goes a step further to employ better denitions of
the different variables. In addition, on top of the traditionally hypothesized
determinants, the study also includes more regionally contextual variables, in
terms of the CMA membership dummy and the MMZT dummy. By doing this,
this study is better able at uncovering the dynamics of reserve management over
the time period so as to make policy recommendations towards the setting up
of the SADC central bank. Given that the number of cross-sectional units in the
study is small and less than the time dimension, cross-sectional dependence was
no concern.
While a lot of unit root tests have been proposed for panel models, only a
few of them can be applied to unbalanced data without inducing bias to the test
results. As proposed by Maddala and Wu (1999), this study adopted the Fisher-
type testing approach using the Phillips-Perron, rather than the Augmented
Dickey-Fuller (ADF) test, specifying a null hypothesis of non-stationarity.
Within this literature, the current study is the rst to adopt such unbalanced-
panel-friendly unit root testing approaches. Results of the Fisher-type test are
presented in Table 1.
The Fisher-type test results indicate stationarity in levels at 5 percent for all
variables except for reserves with gold as well as opportunity costs. For the
non-stationary variables, the cross-sectionally ADF (CADF) test from Pesaran
(2007) was used for validation. Here, with the null hypothesis specied that the
series is stationary, it is veried that all the variables are stationary in levels at
5 percent.
African Review of Economics and Finance
16
taBLe 1: Fisher-tyPe anD Pesaran PaneL unit root tests
Fisher-type Test
χ2-Statistic (p-value) [H0: Series is non-stationary]
Series Constant without trend Constant and time trend
Total Reserves (Incl. Gold) 42.1097 * (0.0700) 72.7994*** (0.0000)
Total Reserves (Excl. Gold) 45.6866 ** (0.0333) 62.0783*** (0.0005)
GDP per capita 46.1263 ** (0.0302) 87.4191*** (0.0000)
Squared GDP per capita 74.8798 *** (0.0000) 101.4461*** (0.0000)
Import-GDP 60.6074 *** (0.0008) 64.1435*** (0.0003)
Lagged Exports Growth 330.9098 *** (0.0000) 297.6141*** (0.0000)
Exchange Rate Volatility 204.1975 *** (0.0000) 270.4732*** (0.0000)
Reserves Volatility 105.7816 *** (0.0000) 241.6390*** (0.0000)
Opportunity Cost 66.9768 *** (0.0001) 41.4144* (0.2946)
Ination Rate 220.7344 *** (0.0000) 297.9905*** (0.0000)
Capital Inows 160.9468 *** (0.0000) 184.8706*** (0.2946)
Pesaran CADF Test
Z[t-bar] (p-value) [H0: Series is stationary]
Total Reserves (Incl. Gold) 0.718 (0.764) -
Opportunity Cost - -1.371* (0.085)
Notes: * p < 0.10, ** p< 0.05, *** p < 0.01
Source: Authors’ estimates.
5.6. Results and interpretations
Noting that the Blundell/Bond dynamic panel data model makes non-stationary
variables stationary through the differencing, a few tests were employed to
ensure that the results are reliable. First, the Sargan test was conducted to check
if over-identifying restrictions, and consequently moment conditions, are valid
so that GMM estimators are consistent. In this case, with the null hypothesis
specifying that over-identifying restrictions are valid for models with and
without gold reserves, chi-square statistics (345.266 and 321.807 respectively)
were observed to be statistically insignicant signifying that moment conditions
are valid and that there is no heteroskedasticity since the null hypothesis was
not over-rejected.
The Arellano-Bond test of serial correlation was conducted to ensure
that there is no serial correlation in the idiosyncratic errors so that moment
conditions are valid. With the null hypothesis, in this case, specied that there is
no serial correlation in rst-differenced errors, results from the AB test of serial
17
Manja et al: Determinants of demand for international reserves in the SADC region
autocorrelation for the system GMM with robust standard errors indicated that
although there is serial correlation in rst-differenced errors at the rst order for
both models at 5 percent signicance level, there is no serial correlation at the
second order, as desired.
Before conducting the multivariate analysis, a test of multicollinearity was also
conducted to check independent variables against perfect collinearity (>0.8) to
avoid obtaining, among other things, indeterminate regression coefcients and
innite standard errors. The variance-covariance matrix showed that there is no
signicantly high linear relationship between any regressors and the estimated
model is satisfactory.
While separately estimating the system GMM estimator, the LSDVC
estimator was also specied with the Blundell/Bond to initialize bias correction
with maximum accuracy of approximation. Bootstrapped standard errors were
also specied for the LSDVC since estimated asymptotic standard errors may
provide poor approximations in small samples, such that obtained t-statistics
and condence intervals are often not reliable (Bruno, 2005). If GMM and
LSDVC estimators gave similar results, then the system GMM estimates would
be deemed unbiased and precise, and so would be interpreted, otherwise LSDVC
estimates would be chosen. Table 2 presents the regression results.
Having estimated total reserves demand models with and without gold
reserves, the numbers of instruments were less than the groups in system
GMM, and chi-square statistics for the system GMM estimator were 3849.19
and 4068.16 respectively (both with P = 0.000). These statistics are strongly
signicant, showing that the independent variables included in the models
are jointly signicant. Comparing the models with and without gold, not
much difference was observed in statistical signicance, coefcient signs and
magnitudes. However, some difference is observed for ination in the system
GMM model whereby the coefcient for the ination rate is statistically
signicant with inclusion of gold reserves, yet insignicant without gold. This
shows that although most SADC countries hold low levels of gold reserves,
gold signicantly changes the dynamics in as far as the demand for reserves is
concerned. This empirically fails to vindicate studies that only focus on reserves
without gold. A comparison of system GMM and LSDVC estimators shows
similarity of the estimates, hence the system GMM estimates were deemed
unbiased and precise, and therefore interpreted. For the logged regressand,
individual coefcient interpretations employ the techniques of a log-lin model
for most regressors.
African Review of Economics and Finance
18
taBLe 2: MoDeL estiMation resuLts For saDC Countries
System GMM LSDVC
With Gold Without Gold With Gold Without Gold
Inertia 0.7986*** 0.8436*** 0.8971*** 0.9119***
GDP per capita -0.0001*** -0.0001*** -0.0000 -0.0000
Sq. GDP per capita 0.0000*** 0.0000*** 0.0000 0.0000
Ination Rate -0.0052** -0.0028 -0.0004 0.0006
Capital Inows -0.0026 0.0003 -0.0023 -0.0001
Opportunity Cost -0.0078*** -0.0057*** -0.0078*** -0.0062***
Degree of Openness 0.0018 0.0490 0.1641 0.1624
Lagged Export Growth 0.0011 0.0011 0.0007 0.0008
CMA Dummy 0.3708*** 0.3364*** - -
MMZT Dummy -0.0320 -0.0031 - -
ER Volatility 0.0009* 0.0010* 0.0008 0.0009
Reserves Volatility 0.0000*** 0.0000*** 0.0000*** 0.0000***
Wald x23849.193 4068.158 - -
P 0.000 0.000 - -
N316 316 316 316
Notes: * p < 0.10, ** p < 0.05, *** p < 0.01
Source: Authors’ estimates.
On the one hand, signicant determinants of demand for reserves in the
SADC region include previous reserve levels, income, opportunity cost of
holding reserves, ination, reserves volatility and the CMA membership
dummy. On the other hand, degree of openness, capital inows, membership to
the MMZT dummy, exchange rate volatility and lagged export growth are found
to be insignicant, implying a higher prevalence of precautionary rather than
mercantilist motives to hoarding reserves. This is in line with some ndings by
Bastourre et al. (2009), Sanusi et al. (2019) and Bhattacharya et al. (2019) for
low-income countries.
The results in Table 2 show that for both models there is a positive impact of
inertial forces on reserves demand at 1 percent level of signicance. This is an
elasticity whereby, ceteris paribus, a 1 percent increase in the previous ratio of
reserves to GDP leads to approximately 0.80 and 0.84 percentage increases in
the current demand for reserves with and without gold respectively for the GMM
model, and respective 0.90 and 0.91 percentage increases for the LSDVC. This
19
Manja et al: Determinants of demand for international reserves in the SADC region
shows that the inertial elasticity of reserves demand is inelastic. This result is
similar to what was found by, among others, Edwards (1984), Bastourre et al.
(2009) and Jung and Pyun (2015) in developing countries.
Income variables are statistically signicant at all levels. For the system
GMM estimator, a US dollar increase in GDP per capita results in decreased
reserve accumulation as measured by the reserves-GDP ratio. The variable
satises the a priori expected sign in line with Baumol’s (1952) square-root
rule for transaction demand as well as empirical ndings by Cheung and Ito
(2009) for both developed and developing countries. In this case, countries with
higher incomes are often more stable and therefore less critical of potential
external shocks whereas poorer economies are more prone to external shocks
and hence likely to demand more reserves. The square of income is also found to
be highly statistically signicant, though not economically signicant, exactly
as found by Aizenman and Marion (2003) and Bastourre et al. (2009). With
signicance of income only observed in GMM and not LSDVC estimator, it is
evident that the GMM adopts strong instruments capable of detecting original
relationships, hence adoption of system GMM estimates is justied (Bastourre
et al., 2009). The result depicts a U-shaped relationship between income and
reserves accumulation.
Of the three risk factors in the model, it is reserves volatility that is found to
be statistically signicant at the 5 percent level. The positive coefcient in all
models is in line with the prediction of the buffer stock model of international
reserves (Frenkel & Jovanovic, 1981). Reserves volatility was included as
a proxy for adjustment costs incurred by an economy that was not hoarding
enough reserves in the wake of a crisis. From all models it can be observed that
higher adjustment costs for an economy increase demand for reserve holdings.
This rightly aligns with ndings by Ben-Ltaifa, Kaendera and Dixit (2009) and
Bhattacharya et al. (2019) though it contrasts what was found by Bernard (2011)
and Bastourre et al. (2009). For example, within the SADC in light of the high
volatility, Lesotho and Swaziland as members of the Southern African Customs
Union (SACU) have been encouraged overtime to deal with revenue volatility
by building adequate reserve buffers to augment their resilience to risks from
volatility (AfDB, 2018). In addition, violent conicts such as the Mozambican
war of 1989-1992, DRC hostilities of 1996-1997 and Lesotho post-election
conicts of 1998 were always followed by depleted reserve buffers amidst an
increasing demand for economic restoration. Countries consequently demand
more reserves to survive such risks.
African Review of Economics and Finance
20
Contrary to expectations, the GMM results reveal that membership to the
CMA increases demand for reserves by about 35 percent. This is likely the case
because members of the CMA (Lesotho, Namibia, South Africa and Swaziland)
constantly set themselves targets to be met in terms of reserves accumulation, led
mainly by the regional de facto Central bank, the South African. For example,
smaller economies in the CMA are required to maintain foreign reserves of
at least equivalent to the total amount of local currencies they issue (Wang,
Masha, Shirono, & Harris, 2007). Similar to this is the requirement placed by
France on French colonies in Africa to hold 50 percent of foreign reserves in the
Bank of France. Such targets increase reserve demand. The special treatment
that Malawi, Mozambique, Zambia and Tanzania (MMZT dummy) get in trade
agreements due to their income ranking was found to have no effect on their
reserve management practices though the expectation was that this makes it less
risky for them, inclining them to demand less reserves for precaution.
For opportunity cost, the results conrm the a priori expected negative
coefcient. Precisely, the opportunity cost elasticity of reserves demand is
about -0.70. As an example, due to low interest rates in 2016, the Central Bank
of Seychelles reported to have increased investment and demand in foreign
currency (CBS, 2016). This shows that improvements in domestic investment
earnings reduce countries’ demand for reserves to be used in foreign earnings,
especially knowing that reserves are usually held in the form of short-term
interest-bearing assets.
5.7. Robustness checks
Various checks for robustness of the results were performed by estimating the
models while, among others, excluding countries with unique characteristics
that might affect the results. For example, Angola, being a net exporter of
petroleum, may have different experiences in balance of payments compared
to the other economies. In addition, while South Africa and Mauritius typically
hold higher proportions of reserves, the DRC has outlying indicators especially
for ination. As such, the models were estimated again in the absence of these 4
economies, as shown in Table 3.
For all models, no signicant differences in the estimates were observed,
alluding to the robustness of the results.
21
Manja et al: Determinants of demand for international reserves in the SADC region
taBLe 3: MoDeL estiMation resuLts exCLuDing angoLa, Mauritius, south aFriCa
anD DrC
System GMM LSDVC
With Gold Without Gold With Gold Without Gold
Inertia 0.7619*** 0.7590*** 0.9160*** 0.9241***
GDP per capita -0.0001*** -0.0001*** -0.0001 -0.0000
Sq. GDP per capita 0.0000*** 0.0000*** 0.0000 0.0000
Ination Rate 0.0028 0.0028 0.0043 0.0046
Capital Inows 0.0048 0.0052 0.0008 0.0007
Opportunity Cost -0.0036* -0.0037* -0.0063* -0.0054
Degree of Openness -0.0553 -0.0603 0.1146 0.1415
Lagged Export Growth -0.0006 -0.0005 -0.0001 -0.0001
CMA Dummy 0.5916*** 0.6055*** - -
MMZT Dummy -0.0601 -0.0700 - -
ER Volatility 0.0005 0.0007 0.0006 0.0007
Reserves Volatility 0.0000*** 0.0000*** 0.0000*** 0.0000***
Wald x23824.595 3703.514 - -
P 0.000 0.000 - -
N228 228 228 228
Notes: * p < 0.10, ** p < 0.05, *** p < 0.01
Source: Authors’ estimates.
6. Conclusion and policy recommendations
This study set out to empirically investigate factors that inuence demand for
international reserves in the SADC region from 1980 to 2019. Demand for
reserves is measured in terms of external assets that are readily available to and
controlled by central banks. During the study period, it is observed that reserve
demand generally increased in the SADC region as countries sought to actively
take precaution against risks, including the various national conicts between
1995 and 2003 as well as the global nancial crisis of 2007-2009. This is the
case in spite of the fact that these countries have grown more liberal over time.
Given the nature of the dependent variable, as well as the (unbalanced) nature
of the panel itself, both Blundell/Bond system GMM and LSDVC estimators
were employed. Apart from including the key traditional factors adapted other
empirical studies, this study adds regional-specic dummies and uses improved
methodologies (including stationarity testing, among other diagnostic tests) and
African Review of Economics and Finance
22
variable measurements. The study weighs in on a decision which is typically
taken for granted by empirical studies; that is, whether to ignore the gold
component of reserves or not. Particularly, it is observed from this study that in
spite of making up only a small proportion of foreign reserves, gold holdings
actually inuence dynamics for the demand for foreign reserves, and hence
future studies should model reserves with and without gold separately. From
the diagnostic tests, regression models were deemed to be well specied. The
long-run results were then estimated using the system GMM estimator whose
estimates’ unbiasedness and precision was veried by the LSDVC.
In terms of the econometric results, it is observed from the system GMM
model (with gold) that income (as measured by GDP per capita), ination,
opportunity cost, CMA dummy, as well as reserves volatilities play a role in
inuencing reserve hoarding. Precisely, income, ination and opportunity cost
have negative effects on demand for reserves, while CMA membership and
volatility have positive effects. One variable that is unexpectedly found to be
statistically insignicant using both estimators is the degree of openness. This
is the case for the SADC region probably because the picture is too mixed,
with some countries being net importers, suffering BOP decits, yet others
enjoy positive trade balances (such as Angola, South Africa and Zambia). In
that regard, the traditional determinant has no effect on demand for reserves.
Generally, obtained results are supported by, inter alia, the buffer stock model.
Several policy recommendations can be made from the results. Given that
reserves are accumulated for precautionary motives, it is evident that improving
stability of the monetary system is essential, in order to maintain a sustainable
demand for reserves. One way to do this is to put in place sound reserve
management policies which in turn effect on macroeconomic performance in
terms of, among others, economic growth and price stability by reducing various
costs, including opportunity and adjustment costs. It is also observed from this
study that high opportunity costs reduce demand for reserves. This calls for
improvement of returns that central banks get in the case that they decide to
invest and not hoard reserves, through among other things reducing risk and
offering secure and rewarding nancial assets. Promotion of trade zones and
unions is also a key policy implication from the results. It is pivotal for countries
to increase stability and economic performance by opening up to regional unions
from which they gain through trade and the set agreements.
23
Manja et al: Determinants of demand for international reserves in the SADC region
Biographical notes
Laston Petro Manja is a Lecturer of Economics at the University of Malawi.
He is also a pre-doctoral research fellow and a UMAPS (University of Michigan
African Presidential Scholars Program) scholar at the University of Michigan,
Ann Arbor. He has vast work and consulting experience, including with the
World Bank. He is a Malawian citizen.
Exley BD Silumbu is currently a Senior Lecturer in Economics at the
Economics Department, University of Malawi. He holds a PhD in Economics
from the University of Notre Dame, Indiana, USA. His specializations include
International Economics, Macroeconomics and Development Economics. He
has worked as Principal Economist in the Malawi Civil Service; Chief Economist
at the Malawi Confederation of Chambers of Commerce and Industry (MCCCI);
and has also served as a member of the Monetary Policy Committee (MPC) of
the Reserve Bank of Malawi.
Regson DC Chaweza holds a PhD in Development Economics and an MA
in Economics and is currently a Lecturer in Economics at the Economics
Department, University of Malawi. He previously held senior national policy
advisory position as a member of the MPC of the Reserve Bank of Malawi.
He has conducted various consultancy projects funded by the Government of
Malawi, international donors and development partners including UNDP and
the World Bank.
Acknowledgements
We are thankful to the handling Editor, Associate Professor Franklin Obeng-
Odoom and for the anonymous reviews which made this article better.
Conicts of interest
The authors declare no conict of interest.
References
AfDB. (2018). Southern Africa Economic Outlook 2018. African Development
Bank.
Aizenman, J., & Lee, J. (2005). International Reserves: Precautionary versus
Mercantilist Views, Theory and Evidence. NBER Working Paper 11366.
African Review of Economics and Finance
24
Aizenman, J., & Lee, J. (2007). The nancial crisis and sizeable international
reserves depletion: From fear of oating to the fear of losing international
reserves? International Review of Economics and Finance, 24, 250-269.
Aizenman, J., & Marion, N. (2003). The high demand for international reserves
in the Far East: What is going on? Journal of the Japanese and International
Economics, 17, 370-400.
Arellano, M., & Bond, S. (1991). Some test of specication for panel data:
Monte Carlo evidence and an application to employment equations. Review of
Economic Studies, 277-297.
Arellano, M., & Bover, O. (1995). Another look at the instrumental variable
estimation of the error component model. Journal of Econometrics, 68, 29-51.
Baltagi, B. H. (2005). Econometric analysis of panel data (3rd ed.). West Sussex:
John Wiley & Sons.
Bastourre, D., Carrera, J., & Ibarlucia, J. (2009). What is driving reserve
accumulation? A Dynamic Panel Approach. Review of International
Economics, (17), 861-877.
Batten, D. S. (1982). Central banks’ demand for foreign reserves under xed
and oating exchange rates. Federal Reserve Bank of St. Louis.
Baumol, W. (1952). The transactions demand for cash: An inventory theoretic
approach. The Quarterly Journal of Economics, 66(4), 545-556.
Ben-Ltaifa, N., Kaendera, S., & Dixit, S. (2009). Impact of the Global Financial
Crisis on Exchange Rates and Policies in Sub-Saharan Africa. IMF African
Department 09/3.
Bernard, K. M. (2011). International reserve adequacy in Central America. IMF
Working Paper, WP/11/144.
Bhattacharya, R., Mann, K., & Nkusu, M. (2019). Estimating the demand for
reserve assets across diverse groups of countries. Rev Int Econ., 27, 822-853.
doi:10.1111/roie.12399
Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in
dynamic panel data models. Journal of Econometrics, 87, 115-143.
Bruno, G. (2005). Approximating the bias of the LSDV estimator for dynamic
unbalanced panel data models. Economics Letters, 87, 361-366.
Bun, M., & Kiviet, J. (2003). On the diminishing returns of higher order terms
in asymptotic expansions of bias. Economics Letters, 79, 145-152.
25
Manja et al: Determinants of demand for international reserves in the SADC region
Burgess, R. (2009). The Southern African Development Community's
Macroeconomic Convergence Program: Initial Performance. IMF Staff
Postition Note, SPN/09/14.
CBS. (2016). International Reserves Management Report 2016. Central Bank
of Seychelles.
Cheung, Y.-W., & Ito, H. (2009). A cross-country empirical analysis of
international reserves. International Economic Journal, 23, 447-481.
Clark, P. B. (1970). Demand for international reserves: A cross-country analysis.
Canadian Journal of Economics III, 577-594.
Clower, R., & Lipsey, R. (1968). The present state of international liquidity
theory. American Economic Review, 586-594.
Dooley, M., Folkerts-Landau, D., & Garber, P. (2004). The revived Bretton
Woods System: The effects of periphery intervention and reserve management
on interest rates and exchange rates in center countries. NBER Working Paper,
no. 10332.
Edwards, S. (1984). The demand for international reserves and monetary
equilibrium: Some evidence from developing countries. NBER Working
Paper Series, WP 1307.
Edwards, S. (1985). On the interest rate elasticity of the demand for international
reserves: Some evidence from developing countries. NBER Working Paper
Series, WP 1532.
Elhiraika, A., & Ndikumana, L. (2007). Reserves accumulation in African
countries: Sources, motivations, and effects. UNECA and University of
Massachusetts Amherst, Working Paper 2007-12.
Flanders, J. M. (1971). The demand for international reserves. Princeton Studies
in International Finance, No. 27, 1-50.
Flood, R., & Marion, N. (2002). Holding reserves in an era of high capital
mobility. IMF Working Paper 02/62.
Frenkel, J. (1978). International reserves under alternative exchange rate regimes
and aspects of the economics of managed oat. NBER Working Paper Series,
WP 287.
Frenkel, J. (1983). International liquidity and monetary control. In J. Frenkel,
& G. Von Furstenberg (Ed.). Washington, DC: International Monetary Fund.
African Review of Economics and Finance
26
Frenkel, J., & Jovanovic, B. (1981). Optimal international reserves: A stochastic
framework. The Economic Journal, 91, 507-514.
Grimes, A. (1993). International reserve under oating exchange rates: Two
paradoxes explained. The Economic Record, 69, 411-415.
Heller, H. (1966). Optimal international reserves. Econ. J., 76, 296-311.
Heller, H. (1968). The transactions demand for international means of payment.
J. Polit. Econ., 76, 141-145.
Hsiao, C. (2003). Analysis of panel data. Cambridge: Cambridge University
Press.
Hume, D. (1752). Political Discourses. Edinburgh.
Hur, S., & Kondo, I. O. (2016). A theory of rollover risk, sudden stops, and
foreign reserves. Journal of International Economics, 103, 44-63.
IMF. (2009). Balance of payments and international investment position manual
(BPM6, 6th ed.). Washington, D.C.: IMF.
Independent Evaluation Ofce of the International Monetary Fund (IEO).
(2012). International reserves: IMF concerns and country perspectives.
Washington, D.C.: International Monetary Fund.
International Monetary Fund (IMF). (2009). Balance of payments and
international investment position manual (BPM6, 6th ed.). Washington, D.C.:
IMF.
Judson, R. A., & Owen, A. L. (1999). Estimating dynamic panel data models: A
guide for macroeconomists. Economics Letters, 65, 9-15.
Jung, K. M., & Pyun, J. H. (2015). Out-of-sample analysis of international
reserves for emerging economies with a Dynamic Panel model. Korea: Korea
University Business School.
Kelly, M. G. (1970). The demand for international reserves. American Economic
Review, 655-667.
Kenen, P. B., & Yudin, E. B. (1965). The demand for international reserves.
Review of Economics and Statistics, 242-250.
Machlup, F. (1966). The need for monetary reserves. Reprints in International
Finance, Princeton University.
Maddala, G., & Wu, S. (1999). A comparative study of unit root tests with panel
data and a new simple test. Oxford Bulletin of Economics and Statistics, 61,
631-652.
27
Manja et al: Determinants of demand for international reserves in the SADC region
Mahraddika, W. (2019). Does international reserve accumulation crowd out
domestic private investment? International Economics. doi:https://doi.
org/10.1016/j.inteco.2019.02.003
Mishra, R. K., & Sharma, C. (2011). The demand for international reserves and
monetary equilibrium: New evidence from India. Journal of Policy Modeling,
33(6), 901-919.
Obstfeld, M., Shambaugh, J. C., & Taylor, A. M. (2010). Financial stability,
the trilemma, and international reserves. American Economic Journal:
Macroeconomics, 2, 57–94.
Oktay, B., ÖZTUNÇ, H. Z., & SERİN, V. (2016). Determinants of gold reserves:
An empirical analysis for G-7 countries. Procedia Economics and Finance,
38, 8 – 16.
Pesaran, H. (2007). A simple panel unit root test in the presence of cross-section
dependence. Journal of Applied Econometrics, 22, 265-312.
Ra, H.-R. (2007). Demand for international reserves: A case study for Korea.
The Journal of the Korean Economy, 8(1), 147-175.
SADC. (2011). Explanatory Guide to the SADC Central Bank Model Law.
SADC. (2017). Summary of the SADC Revised Regional Indicative Strategic
Development Plan 2015-2020. Gaborone, Botswana: SADC.
Sanusi, K. A., Meyer, D. F., & Hassan, A. S. (2019). An investigation of the
determinants of foreign exchange reserves in Southern African countries.
Journal of International Studies, 12(2), 201-212. doi:10.14254/2071-
8330.2019/12-2/12
Simwaka, K. (2016). Macroeconomic convergence in Southern African
Development Community. African Economic Research Consortium Research
Paper 325.
Sinem, E., & Nebiye, Y. (2014). Demand for international reserves in Turkey.
The Romanian Economic Journal, 63-76.
Steiner, A. (2009). Does the accumulation of international reserves spur
ination? A Panel Data analysis. University of Mannheim: Department of
Economics.
Trifn, R. (1964). The evolution of the international monetary system: Historical
reappraisal and future perspectives. Princeton, NJ: International Finance
Section, Princeton University.
African Review of Economics and Finance
28
Wang, J.-Y., Masha, I., Shirono, K., & Harris, L. (2007). The Common Monetary
Area in Southern Africa: Shocks, adjustment, and policy challenges. IMF
Working Paper WP/07/158.
Appendix A: Data Sources
Variables Description Source
Total Reserves (Including Gold) Natural log of the ratio of reserves (with gold)
to the value of current GDP (in US$)
IFS
Total Reserves (Excluding Gold) Natural log of the ratio of reserves (without
gold) to the value of current GDP (in US$)
IFS
GDP per capita Taken on PPP basis (constant 2017, int’l $) WDI
Squared GDP per capita Square of the GDP per capita WDI
Imports Imports of goods and services as a ratio to
current GDP (in US$)
WDI
Lagged Exports Growth Lags the annual percentage growth of exports
of goods and services by 2 periods
WDI
Lending Rates Lending Rates IFS
10yr Market Yield on U.S.
Treasury Securities
10yr Market Yield on U.S. Treasury Securities Federal Reserve
Opportunity Cost Real US treasury bill rate minus the domestic
lending interest rate
-
CMA Dummy Takes 0 for country in the CMA at time t and
1 for country not in the CMA at the time
-
MMZT Dummy Takes 1 for Malawi, Mozambique, Zambia
and Tanzania and 0 otherwise
-
Exchange Rate Volatility Standard deviation of monthly (period
average) exchange rates in each year
IFS
Reserves Volatility Standard deviation of reserves (excl gold) IFS
Ination Rate Ination, consumer prices (annual %) WDI
Capital Inows Foreign direct investment, net inows (% of
GDP)
WDI
Notes: Federal Reserve data is obtained from https://www.federalreserve.gov/releases/h15/.
IFS - International Financial Statistics; WDI - World Development Indicators.
.