Vous êtes sur la page 1sur 35

Accepted Manuscript

Title: International capital flows to emerging markets: national and global


determinants
Author: Joseph P. Byrne, Norbert Fiess
PII:
DOI:
Reference:

S0261-5606(15)00194-1
http://dx.doi.org/doi:10.1016/j.jimonfin.2015.11.005
JIMF 1617

To appear in:

Journal of International Money and Finance

Please cite this article as: Joseph P. Byrne, Norbert Fiess, International capital flows to
emerging markets: national and global determinants, Journal of International Money and
Finance (2015), http://dx.doi.org/doi:10.1016/j.jimonfin.2015.11.005.
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.

International Capital Flows to Emerging Markets:


National and Global Determinants*

Joseph P. Byrne and Norbert Fiess

Department of Accountancy, Economics and Finance,


Heriot-Watt University, Edinburgh, UK

World Bank, Washington DC, USA


21st August 2015

Highlights
1. We examine the nature and determinants of aggregate and disaggregate portfolio flows to
emerging markets.
2. We identify substantial comovement in gross capital inflows, evidenced by a common
factor originating in the global environment.
3. Capital inflows are driven by commodity prices, US rates of return, uncertainty and growth
in advanced economies.
4. Financial openness and the quality of institutions are important country specific
characteristics driving capital inflows.
5. There is a common factor in the volatility of capital inflows, related to commodity prices
and US interest rates.

For their helpful comments the authors would like to thank Cline Azmar, Julia Darby, Rodolphe Desbordes,
Giorgio Fazio and Gregg Huff. We would also like to thank Serena Ng for the use of Matlab code. Finally, we
would like to thank the Editor and Reviewer for helpful and very detailed comments. Correspondence Address:
Department of Accountancy, Economics and Finance, Heriot-Watt University, Edinburgh, UK. Email:
<j.p.byrne@hw.ac.uk>.

Page 1 of 34

Abstract
Using a novel dataset for emerging markets, we empirically investigate the nature and
determinants of aggregate and disaggregate capital inflows. We present formal statistical
evidence of commonalities in capital inflows, with the strongest evidence for the level of
equity and bank flows. Advanced economy long-run bond yields and commodity prices are
identified as determinants of global capital flows. We also consider the national determinants
of capital flows, finding that financial openness and institutions matter for country flows.
Finally, we identify important commonalities in the volatility of bank inflows.
Keywords: Capital Flows; Emerging Markets; Global Factors; Idiosyncratic Flows.
JEL Classification Numbers: F32; F34.

Abstract
Using a novel dataset for emerging markets, we empirically investigate the nature and
determinants of aggregate and disaggregate capital inflows. We present formal statistical
evidence of commonalities in capital inflows, with the strongest evidence for the level of
equity and bank flows. Advanced economy long-run bond yields and commodity prices are
identified as determinants of global capital flows. We also consider the national determinants
of capital flows, finding that financial openness and institutions matter for country flows.
Finally, we identify important commonalities in the volatility of bank inflows.
Keywords: Capital Flows; Emerging Markets; Global Factors; Idiosyncratic Flows.
JEL Classification Numbers: F32; F34.

Page 2 of 34

1. Introduction
Historically, capital flows to emerging markets have mainly comprised foreign direct
investment. Recently, however, portfolio equity and bank-related flows to emerging markets
have increased substantially. Policy makers and academics are increasingly interested in the
nature and causes of these flows. For example, are international or domestic factors important
for capital flows? An existing strand of the literature highlights global characteristics, see
Calvo, Leiderman and Reinhart (1993) and Reinhart and Reinhart (2009). Although different
types of portfolio flows, whether this be equity, bond and bank portfolio inflows, behave
differently, see Contessi, DePace and Francis (2009). As well as focusing on global and
disaggregate behaviour this paper also considers the nature and relevance of country-specific
factors. Domestic structural characteristics may also be important for emerging market capital
inflows, such as financial openness, human capital or institutions, see Lucas (1988), North
(1994) and Alfaro, Kalemli-Ozcan and Volosovych (2008). This study makes use of a novel
panel time series dataset and innovations in panel methodology to examine both global and
national determinants of gross capital inflows.
According to Rothenberg and Warnock (2011) net capital flow dynamics may be
driven by capital inflows or outflows, which in turn may be related to different factors. Hence
capital in- and outflows require to be studied separately. Forbes and Warnock (2012) suggest
few papers have studied gross capital inflow data, previously focusing upon the more readily
available net flow data. Given Reinhart and Reinharts (2009) ocular evidence on common
capital inflow bonanzas, we statistically test for commonalities in our Bondware capital inflow
data. The extent of commonalities in global capital flows and their nature is assessed by Bai
and Ng (2004)s Panel Analysis of Nonstationarity in Idiosyncratic and Common components
(PANIC) methodology. The PANIC approach deals with potential nonstationarity by first
differencing the data, identifying a principal component and then re-cumulating the principal
component as a common factor. This avoids the identification of spurious common factors
based upon nonstationary data. When used in conjunction with Ngs (2006) test for cross
sectional correlation and Bai and Ngs information criteria, PANIC is useful since it allows us
to examine the existence and nature of common factors in global capital flows. This is
important in the current context for two reasons: if shocks to capital inflows are temporary
they are quickly reversed and thus less worrisome from a policy makers perspective. If
shocks are permanent this is more problematic. For example, if there is a permanent increase
3

Page 3 of 34

in volatility of capital inflows following openness this would be highly disadvantageous,


making economic growth volatile. Moreover, from a statistical perspective whether shocks are
permanent or temporary is important since it shall decide whether our methodology should be
robust to nonstationarity when assessing the determinants of commonalities in capital flows.
We examine the drivers of the common component in capital flows, and whether
economic developments in the global environment are important for common trends in capital
flows. This is related to the work by Levchenko and Mauro (2007), Reinhart and Reinhart
(2009) and Forbes and Warnock (2012). Makowiak (2008), Uribe and Yue (2006) and
Neumeyer and Perri (2005) emphasize international factors in driving interest rates and output
in emerging markets. Forbes and Warnock (2012) also identify important global variables that
drive extreme movements in capital inflows and outflows. However, we go beyond the
existing work on capital flows since we focus on identifying the global component which, by
construction, is orthogonal to idiosyncratic characteristics, such as country-specifics or the
domestic policy in a particular recipient country. These idiosyncratic movements in flows are
not truly global capital flows and indeed may mask important global determinants. We think
the common component in global capital flows may be influenced by international economic
activity and we test this hypothesis in our paper.
In this paper we extend existing work on capital flows in several regards: we first
assess the degree of commonality in capital flows, which provides a gauge for the importance
of common factors in determining the global supply of capital. We then extract this common
factor and relate it to economic fundamentals. As the level of aggregation of capital flow data
may impact both on the time series and economic determinants, we provide evidence for both
aggregate capital flows as well as disaggregated data based on portfolio equity, bank and bond
flows. We next explain the national determinants of aggregate capital flows. This is important
as it allows us to consider different conjectures as to how individual countries are impacted by
financial openness (Chinn and Ito, 2008), human capital (Lucas, 1990) and institutional
characteristics (North, 1994). We also consider standard recipient country explicators like
economic growth and interest rates.
To preview our main results, we identify important commonalities in capital inflows,
but these commonalities depend upon whether we consider aggregate or disaggregate capital
flows. Shocks have long lasting consequences for the common element in capital inflows. For
bank flows we find US long-run real interest rates are an important determinant of this

Page 4 of 34

common element, in parallel with Bernankes et al. (2011) suggestion that financial
globalisation operated through assets of a longer maturity. Also, there is a role for commodity
prices and uncertainty in driving equity flows, consistent with the evidence from Reinhart and
Reinhart (2009) and Forbes and Warnock (2012). Using panel econometrics we also present
formal statistical evidence that de jure financial openness and institutions explains why some
countries receive capital inflows. Finally, we identify commonalities in capital flow volatility;
these common shocks are not long lasting and relate to global determinants. Overall this
implies that the volatility of a countrys capital markets is influenced by external factors that
may be transient in nature.
This paper is structured as follows. Section 2 sets out the formal statistical methods
used in this study, including Uniform Spacings, PANIC and our approach to identifying
national and global determinants of flows. Section 3 introduces the dataset and presents our
main results. We examine evidence of commonalities in capital flows for aggregate and
disaggregate data, their time series properties and consider what drives these global capital
flows. Finally, we discuss what time-varying country characteristics influence whether a
country can attract idiosyncratic capital inflows. Section 4 concludes and makes policy
recommendations.

2. Empirical Methods
This study considers both the nature and determinants of the common and
idiosyncratic element of emerging markets capital inflows. We posit that the common
element is global flows. The idiosyncratic component is country, or nation, specific. Ngs
(2006) Uniforms Spacings approach is our first evidence on capital flow commonalities. Ng
(2006) constructs a test statistic, from a standardized spacings variance ratio (svr) test, which
examines the null hypothesis of no correlation in a panel time series. The svr test examines the
probability integral transformation of the ordered correlations, rather than the sample
correlations themselves. Once these correlations are ordered it is more straightforward to
partition them into a sample of small and large correlations. Ng (2006) framework allows
us to ascertain the proportion of small ( ) and large (1 ) bivariate correlations in a panel
dataset, where 0 ,1 . The test utilizes small and large correlation subsets of size

0 , n ,

Page 5 of 34

where n = N(N1)/2 is the maximum number of correlations for N time series. The
standardized test statistic is:
svr ( )

SVR

, where

SVR N 0 , q

(1)

SVR is based upon the second moment of the actual correlations. Therefore, we have two svr
statistics for each of the small and large groups of correlations to initially test for comovement.
We can shed further light on the nature and determinants of capital co-movement in
our aggregate and disaggregate capital inflow data by using the Bai and Ng (2004) Panel
Analysis of Nonstationarity in Idiosyncratic and Common components (PANIC)
methodology. This has a number of advantages. We can identify pervasive or country specific
nonstationarity in the data, since we do not merely assume the latter and hence potential
idiosyncratic nonstationarity. Nonstationarity is relevant to the recent period of financial
globalisation, which involved increasing capital inflows to emerging markets. Also, this factor
model has the advantage that we are not required to know a priori if there is nonstationarity in
the data, since we first differences the data to identify the common component and then recumulating. This avoids spurious factors based on nonstationary data. Moreover, by extracting
a common factor and identifying co-movement, PANIC allows us to model global capital
flows. Kose et al. (2003) and Ciccarelli and Mojon (2010) also use factor models to examine
co-movement of real and nominal international data.
We focus on two key issues: the global and national determinants of capital inflows.
We consider these issues in several estimation steps. Our first estimation step is to examine
the global determinants of capital flows in a bivariate time series approach as follows:
Ft = f(Xt )

t=1,...,T

(2)

We proxy the level of the common factor in capital flows (Ft) at time t using the principal
component extracted by the PANIC methodology. Capital flows are a linear function f(.) of a
vector of potential explanatory variables Xt: these includes the level of real non-oil commodity
prices (RCPt), the real short term (RSRUSt) and long term (RLRUSt) US interest rates, VIX
uncertainty index (VIXt) and real GDP growth in the G7 (YtG7). We are specifically interested
in examining the correlation and evidence of Johansen (1988) cointegration between these
potential explanatory variables and the global component in capital inflows. Given that the

Page 6 of 34

common factor is central to our approach, we now go on to explain the PANIC methodology
in some detail.
The PANIC approach separates a panel time series of capital inflows (CAPit) into
country specific fixed effects (ci) for each country i, a common factor (Ft) which varies over
time t and is associated with corresponding factor loadings (i) and idiosyncratic components
(uit). It is unlikely to be the case that all countries capital flows are equally correlated and the
common factor may matter more for some countries rather than others. Hence factor loadings
i shall vary across country i. The PANIC specification is as follows:
CAPit = ci + iFt + uit

i=1,...,N; t=1,...,T

(3)

This approach identifies a common factor Ft taking account of nonstationarity by first


differencing the data, identifying a principal component and then re-cumulating this
component and testing its statistical properties. The PANIC method of differencing and recumulating is advantageous since it can be used to consistently identify commonalities and
nonstationarity in the panel dataset. The factor loadings i in equation (3) are obtained from
the loadings of the principal components analysis. That is, they are the eigenvalue associated
with the corresponding eigenvector using a principal components approach. The constant term
ci is latent and is removed by first-differencing the data. The country-specific (or
idiosyncratic) component in the factor model is the error term (uit) in equation (3).
We test whether the panel time series CAPit is nonstationary by examining the
statistical properties of the common factor and the errors term in equation (3). Using panel
unit root tests we can examine the null hypotheses of a nonstationary common factor Ft and/or
idiosyncratic component uit. We examine nonstationarity in the factor component using a
univariate Augmented Dickey Fuller (ADF) test, as follows:
Ft = Ft-1 + t

(4)

Based upon equation (4), the common factor ADF test has a null hypothesis H0: = 1 against
an alternative of HA: < 1. For this test we would reject the null hypothesis of factor unit root
for test statistics with large negative values, i.e. less than -2.89 at the 5% significance level,
but fail to reject it otherwise.
The idiosyncratic test statistic ( P uc ) is a Fisher-type pooled ADF test on the individual
errors uit in equation (3). This is distributed as standard normal as follow:

Page 7 of 34

P u 2
c

N
i 1

log p ( i ) 2 N /

4N

. Here p(i) is a probability value from an ADF test on the

following equation for each cross section i for N cross sections:


uit = i uit-1+ eit

(5)

The idiosyncratic statistic examines H0: i = 1 for all i in equation (5) against HA: i < 1, for
some i. The test statistic on the idiosyncratic component is based upon the adjusted sum of
probability values from i idiosyncratic ADF tests, and if this test statistics is greater than 1.65
we would reject the null of idiosyncratic nonstationarity at the 5% significance level, and fail
to reject the null otherwise. Bai and Ng (2002) also set out three information criteria to
identify whether there are common factors in the data.
What motivates our interest in particular global determinants in equation (2)? Reinhart
and Reinhart (2009) examine the relationship between capital flows and two measures of
global economic activity. The first measure is real per capital GDP growth in advanced
economies. A slowdown in growth in advanced economies leads to an expansion of capital
flows to emerging market economies, to take advantage of relatively stronger economic
activity and higher returns. The second global determinant is an index of real non-oil
commodity prices, since emerging markets are often exporters of primary commodities and an
increase in their price shall elicit higher investment. Moreover, Frankel (2008) illustrates a
potential link between commodity prices and real interest rates, with lower rates encouraging
speculation in commodities. A fall in interest rates will lead to a lower discounting of future
commodities, leading to an increase in the price of commodities today. Also, a decline in rates
is associated with an increase in investment and commodity prices. Reinhart and Reinhart
(2009) present evidence of a statistically significant and positive (negative) relationship
between commodity prices (economic growth) and capital inflows between 1967 and 2006.
See also Ahmed and Zlate (2014). Finally, Reinhart and Reinhart (2009) consider the direct
impact of short term real interest rates on capital flows. A fall in real rates of return in
advanced countries leads to an increase in capital flows to emerging market economies, as
investors search for yield. They measure capital inflows using current account data and we go
beyond this in our analysis by using actual capital inflow data.
In addition to advanced economies growth, commodity prices and short term interest
rates, there are other potential determinants of capital inflows to emerging markets and we
examine two not widely considered by the literature: long-term interest rates and global

Page 8 of 34

uncertainty. Long-run interest rates may be at least as important for capital flows as short-run

rates, if investors prefer to diversify assets with short maturities and assets of different
maturity are imperfect substitutes. This is related to Bernanke et al. (2011) discussion of
global saving being closely associated with the assets of longer maturities. Moreover,
investment opportunities may be driven by uncertainty in advanced economies. This is
connected to the literature on investment and uncertainty, exemplified by Dixit and Pindyck
(1994). A recent and popular measure of financial uncertainty is the VIX index of the Chicago
Board which measures the implied volatility of options on the S&P 500 equity index. It
reflects risk aversion in global capital markets to the extent that a rise in implied volatility
reflects a decline in investors risk appetite. Forbes and Warnock (2012) identify an important
role for global risk in influencing extreme aggregate capital flows. See also Bekaert et al.
(2012) for volatility commonalities in financial markets.
Beyond a consideration of purely global determinants of capital flows, we also seek to
investigate the country-specific determinants using a combined panel fixed effects
methodology. In particular, we consider whether financial openness, quality of institutions and
human capital in the recipient country are important in attracting aggregate capital flows.
Consequently, our empirical model of the global and country-specific determinants of
aggregate global capital flows to emerging markets is:
CAPit = 0i+ 0 + 1FOit+ 2Iit + 3HCit + 4Yit + 5Rit
+6RCP.t+7RSRUS.t+8RLRUS.t+9VIX.t+10Y.tG7+it

i=1,...,N; t=1,...,T

(6)

Equation (6) implies capital flows (CAPit) for country i at time t are a function of countryspecific characteristics financial openness (FOit), institutions (Iit) and human capital (HCit).
Financial openness is a necessary condition for capital inflows and is measured by the Chinn
and Ito (2008) index. Human capital as suggested by Lucas (1990) and institutions as
emphasized by North (1994). The remaining country-specific, or pull, determinants, or
push factors, of capital flows come in the form of domestic economic growth (Yit) and local
interest rates (Rit). Secondly, capital inflows are also impacted by global determinants
including non-oil commodity prices (RCP.t), the real US short term (RSRUS.t) and US long
term (RLRUS.t), VIX uncertainty index (VIX.t) and real GDP growth in the G7 (Y.tG7). The
subscript (.t) denotes that these explicators are global and do not vary across cross section,
hence the country i subscript is suppressed. In equation (6) parameters 0 to 10 are coefficients
estimated by panel fixed effects and it is the random error term.

Page 9 of 34

What is more, our methodology allows us to focus upon the country-specific


determinants of capital inflows, by considering the drivers of idiosyncratic capital flows (uit)
in the following estimated panel equation:
uit = 0i + 0 + 1FOit + 2Iit + 3HCit + 4Yit + 5Rit + it

i=1,...,N; t=1,...,T

(7)

In equation (7), idiosyncratic capital flows uit are extracted from aggregate flows using
equation (3) and the Bai and Ngs PANIC approach. Estimated coefficients in equation (7) are
denoted by 0 to 5 and it is a random error term. Since we seek to explain idiosyncratic capital
flows which are country-specific, we primarily focus upon country-specific determinants in
equation (7). Hence we examine the importance of financial openness, institutions, human
capital, national economic growth and national interest rates.1 Having set out our empirical
methodology and research hypotheses we now proceed to discuss our data and present our
results.
<TABLE 1 HERE>
3. Dataset and Empirical Results
3.1 Data
In this study we use quarterly data on capital inflows for up to 64 emerging markets. A
list of countries is provided in the Data Appendix Table A1. The quarterly inflow data is from
Euromoney Bondware and Loanware, with the sample period 1993Q1 to 2009Q1. We scale
our capital flows using a period-by-period measure of economic activity in each country. We
have three types of disaggregate capital inflow data: Equity Issuance, Bond Issuance and
Syndicated Bank Lending. We avoid a difficulty flagged by Rothenberg and Warnock (2011),
since we use capital inflows and we do not conflate foreign and domestic investors which
occurs when net capital flows are examined, for example when using current account data. We
combined our three flow measures to represent our aggregate capital inflow data. Our gross
capital inflow dataset is preferable since it goes beyond the net data often used, has greater
frequency than other Balance of Payments data and greater granularity in allowing us to
consider disaggregate equity, bond and bank lending. The data may be considered to have
drawbacks however, as it uses only inflows, not offsetting outflows, and focuses upon primary

As robustness we also consider whether global determinants are important for idiosyncratic flows in equation
(7). However since we have extracted the orthogonal global component from capital flows to produce uit, we
have premia facia reasons to believe these global factors are unlikely to be important in equation (7).

10

Page 10 of 34

issuance, excluding the secondary market. However there is a good concordance between
Lane and Milesi-Ferreti (2007) External Wealth of Nations dataset and ours.2,3
<FIGURE 1 HERE>
During our sample period there have been waves of aggregate capital inflows across
emerging market economies. In the 1990s capital inflows increased substantially prior to the
Asian Crisis in 1997. Then, more recently a relatively more substantial wave preceded the
Global Financial Crisis. This is illustrated by Figure 1 which contains the first principal
components of aggregate and disaggregate inflows. Kose et al. (2007) argue that the ability of
emerging economies to share consumption risk is hindered by limited access to external debt.
However, the most recent financial wave has been associated with a deepening of financial
markets in emerging economies, see Lane and Milesi-Ferretti (2008). Early in the sample
period bond flows increased relative to bank and equity flows. However, bank and equity
flows have recently become important. Figure 1 also indicates bank flows have been volatile
during the crisis, consistent with Bankings significant role in the crisis.
<TABLE 1 HERE>
3.2 Uniform Spacings
We now formally test the extent of co-movement of capital inflows to emerging
markets by applying Ngs (2006) Uniform Spacings. The results are presented in Table 1.
They provide evidence of commonalities across disaggregate capital flows, although there are
some quantitative differences for each category of flow. The test statistic (svr) is based upon a
partition of the ordered correlations into small and large groups. Hence we have two svr test
statistics: one for large correlations and another for small. For aggregate flows we marginally
fail to reject the null hypothesis of no correlation for a subgroup of over 20% of large bivariate
correlations (large svr = 1.464). This is indicative of some aggregate co-movement. For
disaggregate flows we find greater evidence of a correlation between bank, bond and equity
flows since we reject the null hypothesis of no correlation for all three large svr statistics. This
is illustrative of a relatively greater degree of co-movement of disaggregate than aggregate
2

Altman et al. (2010) discuss the secondary market for bank and bond debt. Cerutti et al. (2014) differentiate
between syndicated and non-syndicated loans when examining cross border bank lending: syndicated loans are
typically held to maturity, but can be traded in secondary markets. Eichengreen and Mody (2000) provide a
discussion of the difference between primary and secondary data for interest rate spreads. For a firm level study
of Asian corporate bond issuance see Mizen and Tsoukas (2014).
3
Our gross capital inflow dataset does not allow direct comparability with net Balance of Payments data. We do
find that our dataset compares well on an annual basis with Lane and Milesi-Ferretti (2007) external wealth
dataset, with a correlation of our aggregate data of 0.95 in means and 0.75 in standard deviations.

11

Page 11 of 34

capital flows. Hence the disaggregate data provides a sharper indication of actual country
correlations of particular financial flows. This is interesting especially for equities given the
recent development of equity markets in emerging markets, see Lane and Milesi-Ferretti
(2008).4 Having statistically identified the global nature of flows we now turn to an analysis of
common factors in the aggregate and disaggregate data.
<TABLE 2 HERE>
3.3 Common Factors in Capital Flows
Our main approach to identify commonalities in based upon principal components.
Principal components are a means to reduce the dimensions of a large dataset such as ours.
The core PANIC results using this approach are set out in Table 2 Panel A. These identify
whether there is a principal component (Ft) in the inflow data and the nature of this
component. Information Criteria (IC) from Bai and Ng (2002) inform us whether there exists a
common component. Time series and panel unit root tests examine whether the common and
idiosyncratic components are nonstationary, respectively. For aggregate, equity and bank
flows there is evidence of a common or global component based on all three information
criteria (i.e. IC>0). This supports Reinhart and Reinharts (2009) informal evidence of capital
inflow bonanzas across countries and our uniform spacings evidence. For international bond
flows to emerging markets there is slightly less evidence of co-movement. That is, according
to Table 2, bond flows display evidence of a common factor for some but not all information
criteria (i.e. IC3=0). So whilst there are substantial capital inflow commonalities we can
differentiate financial flows across country and across asset. We proceed by imposing one
common factor on the data for the aggregate and disaggregate data.
We next test whether the capital inflow common factor is nonstationary in Table 2. We
do so using autoregressive factor equation (4) and examining the null hypothesis H0: = 1. In
terms of time series properties, the aggregate and disaggregate global flow factors are always
nonstationary. In other words, since the univariate factor ADF test statistics are greater than
the 5% critical value we are unable to reject the null hypothesis of a unit root in the common
component of aggregate and disaggregate bank, bond and equity flows. There appears to be
pervasive permanence in capital flows in response to global economic shocks. As can been
seen from Figure 1 which plots the first principal component of our four panel time series, the
4

We should note that there is not a substantial proportion of statistically significant correlations for bank and
equity flows in the uniform spacings test. In the next section therefore, we use Bai and Ngs (2002) information
criteria on the existence of a common component to buttress this evidence.

12

Page 12 of 34

aggregate flow factor is characterized by a sharp rise and fall towards the end of the sample
period, associated with the crisis. Equity and bank flows also have experienced a more
pronounced wave towards the end of the sample period, consistent with Lane and MilesiFerretti (2008) and IMF (2012). Also this upward trend took a significant reversal with the
crisis. Figure 1 also suggests that bonds flows have experienced earlier increases compared to
the rapid rise of equity and bank flows in the recent wave of financial globalisation. Hence,
the reduced evidence of a common component in bonds flows may be due to the lack of an
unambiguous upward global stochastic trend during our sample period.
Our PANIC results in Table 2 also allow us to characterise the nature of the
idiosyncratic (nation specific) capital inflows. In contrast to the common factor, the
idiosyncratic components all appear to be stationary. These test equation (5) using the null
hypothesis H0: i = 1 for all i in equation (5). That is we are able to reject the null hypothesis
of pooled panel unit root test since the pooled probabilities test statistic is greater than the 5%
critical value of 1.65, denoted by an asterisk. Domestic capital inflows, abstracting from
global components, have not experienced a permanent shock during our sample period. This
reinforces our interest in the global component, suggesting that individual country shocks have
not had permanent effects on their capital flows and we should look at the common
components for information on permanent changes during our sample period.

3.4 Correlations of Global Determinants and Common Factors


Having identified commonalities and delineated their time series behavior, in this
section we investigate the relationship between the common elements of capital flows across
countries and their relationship to other macro variables. It is important to look at the
determinants of the common component in capital flows, since this global element may be less
related to individual country characteristics. In this sense we go beyond the existing literature
on capital flows and identify the global determinants of capital flows to particular countries. A
similar approach is set out in the methodological contributions by Bai (2004) and Gengenbach
et al. (2006). As explained above, after extracting the common factors from our PANIC
approach, we consider the relationship between the common factors in aggregate, bank, bond
and equity capital inflows to emerging markets and the following explanatory variables: the
real non-oil commodity prices (RCPt), the real short term (RSRUSt) and real long term

13

Page 13 of 34

(RLRUSt) US interest rate, VIX uncertainty index (VIXt) and real GDP growth in the G7
(YtG7).
<TABLE 3 HERE>
Table 3 presents evidence on the global determinants of capital inflows for both
aggregate and disaggregate data. Some heterogeneity is again manifest since the importance of
these explicators varies for aggregate and disaggregates flows: this supports our
methodological approach of considering aggregate and disaggregate data. There is a sizable
correlation between aggregate capital flows and real commodity prices, i.e. the correlation
coefficient is 0.39 in Table 3. Emerging markets are often commodity exporters and hence
capital inflows may be associated with increases in commodity prices. Moreover, there is
evidence of a negative correlation between aggregate flows and real long-run interest rates in
the US (i.e. -0.12). Low returns in advanced economies are pushing investment to emerging
markets. Uncertainty is also important in reducing flows, with a correlation coefficient of
-0.22. But, for aggregate flows, we find smaller correlations and counter-intuitive signs on
short rates and real economic growth rates, since they have positive and/or lower correlations
with the common factor. These warrant further disaggregate analysis.
Disaggregate data provides for a more granular analysis and Table 3 shows that the
determinants of capital flows vary for bank, bond and equity flows: for bank flows the most
important determinant is the long-run real US interest rate. Banks will actively lend to
emerging markets if there is a lower rate of return to long term US bonds. Relative to the size
of the correlation for long-run yields and given the lack of cointegration evidence below, there
is a less important role for short-term rates. This implies that longer maturity assets are more
important for global capital flows and the flows themselves are less directly attributable to US
monetary policy, contrasting with Reinhart and Reinhart (2009). They propose that capital
flows are related to economic growth, short-run interest rates and commodity prices. Forbes
and Warnock (2012) also highlight the importance of global factors, especially global risk
factors in driving gross capital flows. The VIX index is indicative of global uncertainty:
heightened global uncertainty can suppress investments due to potential irreversibility. This
uncertainty measure has a small correlation with bank or bond flows, the sign is counterintuitive or there is no evidence of cointegration, see below. Long rates appear more
connected to global financial developments than short rates. Bond flows are also influenced by
long-run interest rates. Table 4 indicates that economic growth matters for bank and bond

14

Page 14 of 34

flows (i.e. correlations with the common factor are -0.24 and -0.32 respectively). Common
equity inflows are substantially associated with real commodity prices, providing one of the
largest correlation of all our results (i.e. 0.48), and with long term interest rates (i.e. -0.32). As
mentioned earlier, Frankel (2008) highlights a negative link between real commodity prices
and interests rates. Since the signs of our bivariate relationships are consistent with this
hypothesis, we cannot rule out that this channel explains the link between capital flows,
interest rates and commodity prices. Although the correlation between VIX and equity flows
is relatively small, the sign is intuitive and below we confirm there is evidence of a long-run
relation. This replicates Forbes and Warnock (2012) evidence of uncertaintys importance for
aggregate gross inflows, but extended to disaggregate equity flows. Finally, smaller
correlation statistics suggest there is a less important role for short term interest rates and
economic growth in driving equity flows.
It was noted above that there is evidence of nonstationarity in the common factors, and
there is nonstationarity in the explanatory variables, available upon request. Consequently, we
should exercise caution when interpreting evidence of correlations in the data unless there is
complementary evidence of cointegration. Table 3 presents evidence of a cointegrating vector
between the common factors in capital flows and also our explicators; this is denoted by a and
b at the 5% and 10% significance level respectively. The results strongly support our
correlation analysis. We find evidence that the aggregate behaviour reflects components of the
disaggregate results. For bank flows we find evidence that long-run real interest rates RLRUSt
are an important explanatory variable, since we have evidence of cointegration using
Johansens (1988) Trace Test statistic. And in addition to -0.57 being the largest correlation in
Table 3, a simple regression of f_bankt (the common factor in bank flows) on a constant and
RLRUSt produced a negatively signed estimate on the long-run interest rate with a t-statistic of
5.36, see the scatter plot in Figure 2. Economic growth also cointegrated with bank flows but
the correlation coefficient was much smaller. Like bank flows, bonds flows are also influenced
by global interest rates and growth. However, for bond flows we caveat these results since
they may not have a common factor according to one of Bai and Ngs (2002) information
criteria. The contrasting results between different types of debt highlights the usefulness of our
dataset since it allows us the granularity to distinguish between bank and bond flows.
Equity flow results appear to drive the path of aggregate capital flows with respect to
commodity prices and uncertainty. Equities are especially related to real commodity prices:

15

Page 15 of 34

the correlation coefficient is 0.48; there is evidence of bivariate cointegration; and the
bivariate cross plot has a positive coefficient and t-statistics = 4.16, see Figure 3. Indeed Table
2 Panel B identifies a high correlations between aggregate, equity and bank factors, which
again reinforces Lane and Milesi-Ferrettis (2008) point on the development of equity markets
in emerging markets. Uncertainty is also important since there is evidence of correlation and
cointegration. This disaggregate equity evidence chimes with Forbes and Warnocks (2012)
result that risk is important in driving aggregate and extreme capital flows. Real interest rates
and growth are less important for equity flows since there is either no cointegration or a very
small correlation coefficient.

3.5 Panel Estimation: Global and Individual Country Determinants


Having identified commonalities in global capital flows, we now turn to address why
some individual countries receive more than others. We consider different hypotheses on why
some countries are affected more than others from the trends in financial globalisation. In
other words, why did some countries received substantial capital inflows as a consequence of
the deeper financial integration? At a fundamental level, financial openness would appear to
be an obvious reason why some countries receive a greater share of capital inflows. A country
decides to open markets and foreign capital should immediately flow in. This may be to ignore
the other potential obstacles to capital inflows. In this study we are interested in de jure
measures of financial openness since these give an indication of capital control liberalisation.
Other potential national determinants of capital inflows are the level of human capital
in a country and the general quality of institutions, based on a suggestion from Lucas (1990)
and North (1994) respectively. The main question in the literature on the Lucas (1990)
Paradox is why capital does not flow from rich to poor countries, despite a high relative
marginal product of capital for poor countries. Lucas (1990) suggests that accounting for
human capital can reduce or indeed completely eliminate the differential in marginal rates of
return to capital across countries, assuming that human capital spillovers are internalized
within a country. Hence, low human capital may be a bar to capital inflows. In contrast North
(1994) emphasizes institutions may be important for capital flows since economic returns
from investing in emerging markets may be dependent upon the quality of institutional
arrangements. The importance of institutions for capital flows is considered in a systematic
empirical framework by Alfaro et al. (2008) between 1970 and 2000. They suggest low

16

Page 16 of 34

institutional quality as the leading explanation for the Lucas Paradox. In summary, we seek to
discriminate between financial openness, the quality of institutions and also human capital in
our subsequent analysis in explaining why some countries receive substantial capital inflows
resulting from the recent period of financial globalisation.
We have a range of means of measuring potential country determinants of capital
inflows. Firstly, we have Chinn and Itos (2008) measure of financial openness (FOit) for each
country i. This is based on capital account transactions and the extent of capital controls and
their data is based upon the IMFs Annual Report on Exchange Arrangements and Exchange
Restrictions.5 We prefer Chinn and Itos de jure measure of capital controls rather than de
facto measures, as the latter are based on actual capital flows data which would make the
analysis somewhat circular. Secondly, for human capital (HCit) we use the Institute for Health
Metrics and Evaluation (IHME) data on the educational attainment of total population of 25
year olds and over.6 This is a proxy for human capital and should raises capital inflows based
upon the argument in Lucas (1990). Furthermore, we have a measure of the quality of
institutions (Iit), from the International Country Risk Guide; an increase in the index means an
improvement of institutions in that particular country. An improvement in a country's
institution should increase capital inflows. Finally, we consider standard macroeconomic
determinants of capital inflows, including recipient country i economic growth (Yit) and
interest rates (Rit).
<TABLE 4 HERE>
In Table 4 we examine the determinants of capital inflows across time and country
using panel fixed effects estimation of equation (6). In column [1] and [2] of Table 4 we
consider whether country specific explicators and global determinants impact upon aggregate
capital inflows. Column [1] includes all potential determinants and column [2] deletes
insignificant explanatory variables in a general-to-specific approach that we prefer. The
country specific explicators are financial openness, institutions, human capital, economic
growth and real interest rates. The estimated coefficients in column [2] on institutions and
financial openness are both positive and important, in that they are both statistically significant
at the 5% level. This is consistent with the suggestion of North (1994) and evidence in Alfaro
et al. (2008) that institutions matter. In contrast our measure of human capital is not a
5

This section uses annual observations, since our key determinants are provided on an annual basis.
As robustness we also considered the Barro and Lee (2000) dataset and results were not quantitatively different.
IHME data was preferred since it is available at an annual frequency.

17

Page 17 of 34

statistically significant determinant of capital inflows: hence, we have no evidence in this table
to support the Lucas (1990) argument that human capital explains the level of capital flows to
emerging markets. In addition, country specific macroeconomic determinants output and
interest rates are insignificant pull factors. Table 4 column [2] also considers global
determinants and finds an important role for real commodity prices and advanced economies
economic growth, consistent with Table 3, also in addition to short and long US interest rates.
Whilst column [2] has a significant F-statistic, rejecting the joint null hypothesis that our
coefficients are equal to zero, the R2 statistic indicates that we explain around a fifth of the
total variation in the capital inflow data.7
In Table 4 we can also examine the determinants of idiosyncratic capital inflows using
equation (7). This is the aggregate data filtering out common components. Using panel fixed
effects we find that institutions and openness are again highly important for this idiosyncratic
country data in column [4]. Human capital is again unimportant for idiosyncratic capital
inflow data, since it is not significant at the 10% level in column [3] and we delete this
determinant. In column [3] of Table 4, we see that domestic factors like output and interest
rates are again unimportant and they are deleted from estimation in column [4]. Given the
idiosyncratic has extracted global components we anticipate that the global determinants shall
be unimportant for idiosyncratic capital flows. Column [3] indicates that our method for
accounting for global factors is coherent since none of the global determinants are in fact
significant. Indeed, de-factoring the data means only country specific factors now matter and
the global determinants have been filtered out. In this case the R2 statistic indicates that we
explain around 10% of the total variation in the capital inflow data. This implies that the
proportion of variation in the data that we explain is approximately equivalent between the
common factor and the idiosyncratic element.8
<TABLE 5 HERE>
We also examine the extent to which the country specific determinants are important
for disaggregate capital flows using panel fixed effects estimation. We use a general to
specific methodology and present only the final regression results in Table 5. Overall these
7

As recommend by a referee we experimented with including a time trend in Table 4 and also tested for panel
cointegration. These estimations, available upon request, indicate that key results are not sensitive to including a
deterministic trend nor indicative of a spurious regression.
8
We also examined whether emerging markets are exporters of commodities when assessing the impact of
commodity prices on capital flows. We used data from UNCTAD and WTO to construct an interaction dummy
for export commodity dependence. However, this interaction was not statistically significant.

18

Page 18 of 34

highlight the importance of financial openness and institutions for bank and bond flows to
emerging markets. We find a role for human capital, although this seems paradoxical. Human
capital is important for total disaggregate flows and for idiosyncratic flows, but with the
opposite sign. This negative sign was not robust to the inclusion of a time trend. Real
commodity prices are positive for total bank and total equity flows: this mirrors the results in
Table 4. There are some counter-intuitive signs for short-run yields on bank and bond flows,
but not for equity. The uncertainty measure is important for bond and equity flows, although
this is more believable for equity since the sign is consistent with results in Table 3 and this
global determinant operates through total equity flows.

3.6 Volatility of Capital Flows


For an emerging market it is not only the level of capital flows that matters, it is also
important to consider the nature and determinants of the volatility of inflows. This is
suggested in the related literature, in which it is assumed that rapid reversals of capital flows
(i.e. high volatility) have negative economic consequences. In this section we investigate the
degree of co-movement across countries in volatility of capital inflows focusing upon global
volatility flows. This provides formal statistical information on the extent to which individual
countries themselves are dependent upon global capital flows and hence are not entirely
responsible for the behaviour of capital markets that confronts them. We use a rolling window
of the standard deviation of 12 monthly observations to measure the volatility of capital
inflows. Table 6 presents our PANIC results for the co-movement of inflow volatility.
<TABLE 6 HERE>
Table 6 provides evidence of co-movement of volatility for the aggregate flows.
Following Bai and Ng (2002), the information criteria indicate at least one principal
component in the aggregate data. This suggests that capital flow volatility, in addition to
potential country-specific determinants, has many commonalities across countries. This result
also stands for a panel time series of disaggregate inflows, since there is evidence of comovement in the volatility of bank, bond and equity flow indicated by the information criteria.
The data is also stationary as suggested by the rejection of the null of nonstationarity for the
factor and idiosyncratic data.
<TABLE 7 HERE>

19

Page 19 of 34

What determines this global volatility in capital inflows to emerging markets? From
Table 7 the main result is that the volatility of aggregate flows is determined by real
commodity prices and long-run US interest rates. RCPt and RLRUSt appear to be highly
correlated with aggregate volatility, i.e. correlation coefficient of 0.50 and -0.53 respectively.
Using bivariate regressions we find that RCPt and RLRUSt are statistically significantly related
to aggregate flow volatility, denoted by the superscripts in Table 7. Short rates are also
important for some of the short term volatility of capital flows but with a less negative
correlation. It again may be the case that as interest rates fall and become more stable
investors look for alternative, higher and more risky rates of return elsewhere. This aggregate
volatility evidence is replicated most strongly for the disaggregate bank and equity flows,
consistent with the evidence for the level of capital inflows.

4. Conclusion
This paper considers the nature and determinants of capital inflows to emerging
markets. We examine both aggregate and disaggregate capital inflows since they do not
display the same time series behaviour, depend upon the same shocks and have the same
economic implications. We find important similarities and differences in the cross country
behaviour of financial flows of different asset types. For example, we find evidence of
considerable cross country correlation in bank and equity flows according to our PANIC
approach. These consequently influence aggregate flows, which may primarily be a reflection
of bank and equity capital inflows. In contrast there was slightly less PANIC evidence that the
level of bond flows is correlated across countries.
We went on to consider the potential determinants of the waves in financial
globalization. We set out an important channel for financial globalization operating through
long rates and impacting emerging markets. We find that real US long-run interest rates are an
important determinant of disaggregate bank and equity capital inflows. Bernanke et al. (2011)
and Byrne et al. (2012) set out how the rapid increase in global savings that preceded, and may
have caused the recent Global Financial Crisis (the Global Savings Glut), operated on longrun rather than short-run interest rates and would in turn have important consequence for
emerging markets. A fall in long-run returns in bonds causes investors to direct funds to
emerging markets. There is evidence of a less important role for short term interest rates in
driving capital inflows in our results. Hence, US monetary policy which operates through

20

Page 20 of 34

short term interest rates may be having a relatively less powerful effect in emerging markets.
Real commodity prices also appear to be important for equity and aggregate capital flow data,
but less so for bank flows, where US long rates clearly dominate. Frankels (2008) suggestion
of a symbiotic relationship between commodity prices and interest rates is therefore unlikely
to be the whole story explaining bank flows. Risk is important for capital flows, in particular
for disaggregate equity flows. We also were able to identify common elements in the volatility
of capital inflows. This suggests that some of the negative implications of capital flows (i.e. an
abrupt discontinuation of capital inflows or Sudden Stops) may be less the result of specific
policies associated with particular emerging market economies but may be largely generic to
this investment group.
Finally, our methodology allows us to consider the potential determinants of an
individual countrys experience with capital inflows and to discriminate between competing
hypothesis from North (1994) and Lucas (1990). We found evidence of an important role for
de jure financial openness, see Chinn and Ito (2008), and institutions, following North (1994)
and Alfaro et al. (2008). In contrast the idea that human capital is less important for the level
of aggregate or idiosyncratic capital inflow stands, which undermines the idea that human
capital explains the Lucas Paradox that financial capital does not flow to emerging markets
despite a high marginal product of capital. This result was robust to alternative measures of
human capital. These results matter for emerging market economies because they imply it
may not be sufficient to remove capital controls to benefit from global capital flows. To gain
from future waves of financial globalization, emerging markets economies should therefore
have increased financial openness and aim to strengthen their institutions.

21

Page 21 of 34

Data Sources Appendix


We use quarterly Capital Inflow in US Dollars from Euromoney Bondware and Loanware.
This disaggregate data is for Equity, Bond and Bank flows. We sum the data to produce an
aggregate flow of portfolio capital. Capital Inflow has been divided by period-by-period
nominal GDP from IMF World Economic Outlook, to account for relative size of flows across
countries. Since international capital flows are typically measured in US Dollars and following
Reinhart and Reinhart (2009) all data is in US Dollars. Table A1 presents the countries that we
have data from Euromoney. We apply a four quarter moving average. We restrict attention to
countries that participate in International Capital Inflows. Hence we exclude those countries
for which we have less than four quarters of observations between 1993Q1 to 2009Q1. Outlier
countries were removed. In the section of the volatility of capital flows we measure volatility
as a rolling standard deviation with a window of 12 monthly observations.
Financial Openness (FOit) Chinn and Ito (2008) produce a de jure measure of financial
openness based on capital account transactions and the extent of capital controls in 2000.
From the IMFs Annual Report on Exchange Arrangements and Exchange Restrictions. The
index has a mean zero and an increase is the index indicates increasing openness.
G7 Real GDP Growth (YtG7) from OECD Main Economic Indicators and varies over time t.
Human Capital (HCit) We use Institute for Health Metrics and Evaluation data from
GapMinder on the average number of years of schooling for each country i at time t. We also
used for robustness Barro and Lee (2000) measure of the average number of years of
schooling in 2000 for each country i.
Institutional Quality (Iit) A composite index from International Country Risk Guide (ICRG).
The measure is from 40 to 87 and a rise in the index is associated with an improvement in
institutions. The twelve different institutional measures include: Government Stability,
Socioeconomic Conditions, Investment Profiles, Internal Conflict, External Conflict,
Corruption, Military Involvement in politics, Religious involvement in politics, Law and
Order, Ethnic Tensions, Democratic Accountability and Bureaucratic Accountability.
Real Commodity Prices (RCPt) are from IMF International Financial Statistic. Based upon
Non-oil commodity prices deflated by US wholesale price index following Reinhart and
Reinhart (2009).
Real GDP Growth (Yit) is from the IMF International Financial Statistics and World Bank
World Development Indicators and varies over country i and time t.
Real Interest Rates are from IMF International Financial Statistic. They are 3 Month US
Treasury Bill Rate (RSRUSt) and 10 year US government bond yield (RLRUSt) deflated ex
post by the annual US Consumer Price inflation. National real interest rates (Rit) are from
World Bank World Development Indicators and varies over country i and time t.
VIX Index (VIXt) is a measure of US stock market uncertainty from the Chicago Board
Option Exchange.

22

Page 22 of 34

Table A1. Sample of Countries


Aggregate (N = 29)
Algeria
Argentina
Bolivia
Bulgaria

Colombia

Bank (N = 46)
Algeria
Argentina
Bangladesh
Belarus
Bolivia
Burkina Faso
Cameroon
Colombia
Costa Rica
Cote D'Ivoire
Croatia
Dominican Republic
Ecuador
El Salvador

Georgia

Indonesia
Iran

Ghana
Guatemala
Guinea
Honduras
Indonesia
Iran

Jordan

Jordan

Kenya
Latvia

Kenya
Latvia
Lebanon
Lithuania
Macedonia

Lithuania

Mauritius
Mexico
Morocco
Nigeria

Mexico
Morocco
Mozambique
Namibia

Pakistan
Peru
Philippines
Poland

Romania

Romania
Senegal

South Africa
Sri Lanka

South Africa
Sri Lanka
Tanzania
Thailand

Tunisia
Turkey
Ukraine
Venezuela
Vietnam

Equity (N = 34)

Argentina

Argentina

Belarus
Bulgaria

Bulgaria

Chile
Colombia
Costa Rica

Chile
Colombia

Croatia
Dominican Republic

Croatia

Egypt
El Salvador
Estonia

Egypt

Guatemala
Guinea

Tunisia
Turkey
Ukraine
Uruguay
Venezuela
Vietnam
Zimbabwe

Estonia
Ghana

Indonesia

Indonesia

Jamaica
Jordan
Kazakhstan

Jordan
Kazakhstan

Latvia
Lithuania

Lebanon
Lithuania
Malawi
Malaysia

Morocco

Pakistan

Peru
Philippines
Poland

Thailand

Bond (N = 35)

Peru
Philippines
Poland
Qatar
Romania

Mexico
Morocco

Oman
Pakistan
Panama
Papua New Guinea
Peru
Philippines
Poland
Romania

Slovak Republic
South Africa
Sri Lanka

Slovak Republic
South Africa
Sri Lanka

Thailand
Trinidad and Tobago
Tunisia
Turkey
Ukraine
Uruguay
Venezuela

Thailand
Tunisia
Turkey
Ukraine
Vietnam
Zimbabwe

23

Page 23 of 34

References
Acemoglu, D. (2009) Introduction to Modern Economic Growth. MIT Press: Cambridge.
Ahmed, S. and A. Zlate (2014) Capital flows to emerging market economies: A brave new
world? Journal of International Money and Finance, vol. 48, pp. 221-248.
Alfaro, L., S. Kalemli-Ozcan and V. Volosovych (2008) Why doesn't capital flow from rich to
poor countries? An empirical investigation. Review of Economics and Statistics, vol. 90(2), pp
347-368.
Altman, E. I., A. Gande and A. Saunders (2010) Bank debt versus bond debt: Evidence from
secondary market prices. Journal of Money, Credit, and Banking, vol. 42(4), pp. 755767.
Bai, J. (2004) Estimating cross-section common stochastic trends in nonstationary panel data.
Journal of Econometrics, vol. 122(1), pp. 137-183.
Bai, J. and S. Ng (2002) Determining the number of factors in approximate factor models.
Econometrica, vol. 70(1), pp. 191-221.
Bai, J. and S. Ng (2004) A PANIC attack on unit roots and cointegration. Econometrica, vol.
72(4), pp. 1127-1177.
Barro, R.J. and J.-L. Lee (2000) International Data on Educational Attainment: Updates and
Implications, manuscript, Harvard University, February 2000.
Bekaert, G., R.J. Hodrick, and X. Zhang (2012) Aggregate idiosyncratic volatility. Journal of
Financial and Quantitative Analysis, vol. 47(6), pp. 11551185.
Bernanke, B.S., C. Bertaut, L.P. DeMarco, and S. Kamin (2011) International Capital Flows
and the Returns to Safe Assets in the United States, 20032007. International Finance
Discussion Paper 1014, Board of Governors.
Byrne, J.P, G. Fazio, and N. Fiess (2012) Interest rate co-movements, global factors and the
long end of the term spread. Journal of Banking and Finance, vol. 36(1), pp. 183-192.
Caballero, R.J., F. Emmanuel, and P.-O. Gourinchas (2008) An equilibrium model of Global
Imbalances and low interest rates. American Economic Review, vol. 98(1), pp. 358-393.
Calvo, G., L. Leiderman, and C.M. Reinhart (1993) Capital inflows and real exchange rate
appreciation in Latin America: The role of external factors. IMF Staff Papers, vol. 40(1), pp.
108151.
Calvo, G. and C.M. Reinhart (2000) When capital inflows come to a sudden stop:
consequences and policy options. In P.B. Kenen and A.K. Swoboda (eds.) Reforming the
International Monetary and Financial System, International Monetary Fund, pp 175201.

24

Page 24 of 34

Cardarelli, R., S. Elekdag, and M.A. Kose (2010) Capital inflows: Macroeconomics
implications and policy responses. Economic Systems, vol. 34(4), pp. 333-356.
Cerutti, E., H. Galina and C. Minoiu (2014) Financial crises and the composition of crossborder lending. Forthcoming Journal of International Money and Finance. IMF Working
Paper 14/185.
Chinn, M.D. and H. Ito (2008) A new measure of financial openness. Journal of Comparative
Policy Analysis, vol. 10(3), pp. 309 - 322.
Ciccarelli, M. and B. Mojon (2010) Global Inflation, Review of Economics and Statistics, vol.
92(3), pp. 524-535.
Contessi, S., P. DePace and J. Francis (2009) The cyclical properties of disaggregated capital
flows. Working Papers 2008-041, Federal Reserve Bank of St. Louis.
Contessi, S. and P. DePace (2009) Do European capital flows comove? North American
Journal of Economics and Finance, vol. 20(2), pp. 145-161.
Dixit, A. and R.S. Pindyck (1994) Investment Under Uncertainty. Princeton University Press,
Princeton.
Eichengreen, B. and A. Mody (2000) What Explains Changing Spreads on Emerging Market
Debt? In Capital Flows and the Emerging Economies: Theory, Evidence, and Controversies.
Ed. S. Edwards, NBER, University of Chicago, Chicago.
Forbes, K.J. and F.E. Warnock (2012) Capital flow waves: Surges, stops, flight, and
retrenchment. Journal of International Economics, vol. 88(2), pages 235-251.
Frankel, J.A. (2008) The effect of monetary policy on real commodity prices. In Asset Prices
and Monetary Policy. Ed. J.Y. Campbell, NBER, University of Chicago, Chicago.
Gengenbach, C., F.C. Palm and J.-P. Urbain (2006) Cointegration testing in panels with
common factors. Oxford Bulletin of Economics and Statistics, vol. 68(s1), pp. 683-719.
Henry, P.B. (2007) Capital account liberalization: Theory, evidence, and speculation. Journal
of Economic Literature, vol. 45(4), pp. 887-935.
Herrera, A.M., I. Murtazashvili and E. Pesavento (2008) The comovement in inventories and
in sales: Higher and higher. Economics Letters, vol. 99(1), pp. 155-158.
IMF (2012) The liberalization and management of capital flows: an institutional view. IMF,
Washington DC, USA.
Johansen, S. (1988) Statistical analysis of cointegrating vectors. Journal of Economic
Dynamics and Control, vol. 12(2-3), pp. 231-254.

25

Page 25 of 34

Kose, M.A., E.S. Prasad and M.E. Terrones (2009) Does openness to international financial
flows raise productivity growth? Journal of International Money and Finance, vol. 28(4), pp.
554-580.
Kose, M.A., C. Otrok and C.H. Whiteman (2003) International business cycles: World, region
and country specific factors. American Economic Review, vol. 93(4), pp. 1216-1239.
Lane, P.R. and G.M. Milesi-Ferretti (2007) The external wealth of nations mark II. Journal of
International Economics, vol. 73, pp. 223-250.
Lane, P.R. and G.M. Milesi-Ferretti (2008) The drivers of financial globalization. American
Economic Review, vol. 98(2), pp. 327-32.
Levchenko, A. and P. Mauro (2007) Do some forms of financial flows protect from sudden
stops? World Bank Economic Review, vol. 21(3), pp. 389-411.
Lucas, R.E. (1990) Why doesnt capital flow from rich to poor countries? American Economic
Review, vol. 80(2), pp. 92-6.
Makowiak, B. (2007) External shocks, U.S. monetary policy and macroeconomic
fluctuations in emerging markets. Journal of Monetary Economics, vol. 54(8), pp. 2512-2520.
Mishkin, F.S. (2008) Is financial globalization beneficial? Journal of Money, Credit, and
Banking, vol. 39(2-3), pp. 259-294.
Mizen, P., and S. Tsoukas (2013) What promotes greater use of the corporate bond market? A
study of the issuance behaviour of firms in Asia. Oxford Economic Papers, vol. 66 (1). pp.
227-253.
Neumeyer, P.A. and F. Perri (2005) Business cycles in emerging economies: the role of
interest rates. Journal of Monetary Economics, vol. 52(2), pp. 345-380.
North, D.C. (1994) Economic Performance through time. American Economic Review, vol.
84(3), pp. 359-68.
Ng, S. (2006) Testing cross-section correlation in panel data using Spacings. Journal of
Business and Economics Statistics, vol. 24(1), pp. 12-23.
Obstfeld, M. (1994) Risk-taking, global diversification, and growth. American Economic
Review, vol. 85(5), pp. 1310-29.
Obstfeld, M. (2009) International finance and growth in developing countries: What have we
learned? IMF Staff Papers, vol. 56(1), pp. 63-111.
Obstfeld, M. and A.M. Taylor (2004) Global Capital Markets: Integration, Crisis, and
Growth. Cambridge University Press, Cambridge.
Political Risk Service Group (2001) International Country Risk Guide. Syracuse, New York.

26

Page 26 of 34

Prasad, E., K. Rogoff, S.-J. Wei and M.A. Kose (2003) Effect of financial globalization on
developing countries: Some Empirical Evidence. IMF Occasional Paper #220.
Reinhart, C.M. and V.R. Reinhart (2009) Capital flow bonanzas: An encompassing view of
the past and present. In NBER International Seminar on Macroeconomics, pp. 9-62. Eds. J.
Frankel and C. Pissarides, National Bureau of Economic Research.
Reinhart, C.M. and K. Rogoff (2004) Serial default and the paradox of rich-to-poor capital
flows. American Economic Review: Papers and Proceedings, vol. 94(2), pp. 5358.
Rodrik, D. (1998) Who needs capital-account convertibility? In Should the IMF Purse capitalaccount convertibility? Ed. S. Fischer et al. Princeton Essays in International Finance 207.
Rothenberg, A.D. and F.E. Warnock (2011) Sudden flight and true sudden stops. Review of
International Economics, vol. 19(3), pp. 509-524.
Stiglitz, J.E. (2002) Globalization and Its Discontents. New York: Norton.
Uribe, M. and V.Z. Yue (2006) Country spreads and emerging countries: Who drives whom?
Journal of International Economics, vol. 69(1), pp. 636.
Warnock, F.E. (2002) Home bias and high turnover reconsidered. Journal of International
Money and Finance, vol. 21(6), pp. 795-805.

27

Page 27 of 34

Figure 1. Principal Component of Capital Inflows to Emerging Markets

Notes: this figure contains the first principal component extracted from the panel dataset of Bank
(f_bank), Bond (f_bond) and Equity (f_equity) and Aggregate Capital Inflows (f_agg).

28

Page 28 of 34

Figure 2. Cross Plot of Real Long US Interest Rate and Bank Factor

f_bankt = 0.008 0.001RLRUSt


(t=10.10) (t=5.36)

R2 = 0.33

Notes: OLS estimation indicates there is a strong and statistically significant negative relationship
between Real Long US Interest Rates (RLRUSt) and the common factors of bank capital inflows
(f_bankt). Time period is 1993Q3 to 2008Q3 and T = 61.

Figure 3. Cross Plot of Real Commodity Prices and Equity Factor

f_equityt = 0.007 + 0.0001RCPt


(t=3.53) (t=4.16)

R2 = 0.23

Notes: There is a positive relationship between the common factor in capital flows for equity
(f_equityt) and real commodity prices (RCPt). This relationship is statistically significant,
although there is a slightly smaller R2 and t-statistic than for regression between the factor and
real long-run interest rates. Time period is 1993Q3 to 2008Q3 and T = 61.

29

Page 29 of 34

Table 1. Uniform Spacings Analysis of Capital Inflows

Aggregate

0.793

Number of small
correlation pairings
322 out of 406

Disaggregate
Bank

0.894

925 out of 1035

-0.467

2.021*

Bond

0.882

525 out of 595

1.659*

1.857*

Equity

0.856

480 out of 561

0.542

2.484*

Small svr

Large svr

-0.658

1.464

Notes: This table presents evidence on the degree of cross sectional correlation for our aggregate and
disaggregate capital inflow data. is the proportion of all possible correlations (n) that are small ( ). Ng
(2006) Spacings Variance Ratio test statistic (svr) provides evidence of whether correlation is significantly
different from zero, distributed as standard normal, therefore the 5% critical value is 1.65, and significance
at the 5% level is denoted by an asterisk (*). First order serial correlation is removed following Ng (2006),
assuming an AR(1) model. There are n = N(N-1)/2 correlations, for N = 29, 46, 35 and 34 respectively for
Aggregate, Bank, Bond and Equity flows. The time dimension is 1993Q1 to 2009Q1.

30

Page 30 of 34

Table 2. PANIC Analysis of Capital Inflows


FACTOR (Ft)
Panel A
Aggregate
Bank
Bond
Equity
Panel B

IDIOSYNCRATIC(uit)

-1.208

3.158*
Disaggregate
5.389*
5.197*
8.436*

-1.334
-2.094
0.056

IC1

IC2

IC3

5
5
5

3
5
5

1
0
1

Factor Correlations
Bank Factor
Bond Factor
Equity Factor

0.46
0.36
0.77
Aggregate Factor

0.40
0.71
Bank Factor

0.31
Bond Factor

Notes: This Table examines the statistical properties of our capital inflow data. Panel A presents evidence on whether the
common factor and idiosyncratic components are nonstationary, using Bai and Ngs (2004) PANIC approach, and the
number of common factors. We use equation (3) to decompose the dataset into common factor (Ft ) and idiosyncratic
component (uit). For the factor Ft using equation (4) we reject the null hypothesis of a unit root in the common
component for large negative values for the test statistic (less than -2.89). For the idiosyncratic component using
equation (5), we reject the null hypothesis of a unit root for large positive values of the test statistic (greater than 1.65).
Rejected null hypotheses of nonstationarity are denoted by an asterisk (*) and in bold. We identify the factor structure
using information criteria from Bai and Ng (2002), i.e. IC1 to IC3. Panel B contains correlations of the first principal
component of the Aggregate and Disaggregate data. We use Aggregate data and Disaggregate data for Bank, Bond and
Equity flows. The number of cross sections are N = 29, 46, 35 and 34 respectively for Aggregate, Bank, Bond and Equity
flows. The time span of the capital inflow dataset is 1993Q1 to 2009Q1 (T=65).

Table 3. Correlation and Cointegration of Capital Flow Factors and Explicators


RCPt

RSRUSt

RLRUSt

VIXt

YtG7

Aggregate

0.39b

0.27

-0.12

-0.22b

0.11a

Bank

0.08

-0.43

-0.57a

0.20

-0.24a

Bond

-0.13

-0.16b

-0.46a

-0.13

-0.32a

Equity

0.48b

0.00

-0.32

-0.09b

-0.02a

Notes: This table includes bivariate numerical correlation of common factors (Ft) in capital inflows with
potential explicators. Also, this table presents evidence of the existence of one cointegrating vectors
between the type of capital flow and explanatory variable, in bold and denoted by a at 5% and b at 10%
level of statistical significance. This is based on the Johansen (1988) Trace Test Statistic, where the null
hypothesis is no cointegration. The time period is 1993Q2 to 2008Q3. Lag length determined by AIK.
RCPt is real commodity prices excluding oil, RSRUSt is the real short-run US interest rate, RLRUSt is the
real long-run US interest rate, VIXt is a measure of market uncertainty and YtG7 is real GDP growth in
the G7.

31

Page 31 of 34

Table 4. Determinants of Aggregate and Idiosyncratic Country Portfolio Flows


Aggregate
FOit

[1]
1.382***

[2]
1.450***

Idiosyncratic
[3]
[4]
1.435***
1.256***

Iit

0.103**

0.102**

0.122***

HCit

1.372

-1.053

Yit

0.086

0.076

Rit

0.005

0.005

RCP.t

0.040**

0.034**

0.003

RSRUS.t

0.605***

0.679***

-0.129

RLRUS.t

-0.112

-0.829***

-0.046

VIX.t

0.028

Y.t

0.476

0.511**

0.335

Constant

-18.596***

-5.283*

0.329

-5.054*

NxT

365

385

365

385

25

25

25

25

R2

0.21

0.19

0.12

0.09

F-statistic

8.983***

13.861***

4.486***

17.157***

Explicators

G7

0.125***

0.083

Notes: This table presents evidence on the determinants of aggregate and idiosyncratic capital flows to
emerging markets. Table 4 estimates equation (6) and (7) in the main text by panel fixed effects.
Column [1] seeks to examine whether the following determinants are important for aggregate capital
inflows: Financial Openness (FOit), Institutions (Iit), Human Capital (HCit), country specific economic
growth (Yit) and interest rates (Rit) from recipient countries. Also column [1] contains global
determinants of capital inflows: RCP.t is real commodity prices excluding oil, RSRUS.t is the real shortrun US interest rate, RLRUS.t is the real long-run US interest rate, VIX.t is a measure of market
uncertainty and Y.tG7 is real GDP growth in the G7. Column [2] is a general to specific approach.
Column [3] repeats the analysis on the idiosyncratic or nation specific capital inflows. Idiosyncratic data
is obtained by removing the global factor using Bai and Ng (2004) from the aggregate data, see
equation (3). The time dimension is 1993 to 2009, and the data is annual here due to data availability.
Explicators that are statistically significant are denoted by asterisk: *** at 1%, ** at 5% and * at 10%
level of statistical significance. The F-statistic tests the joint null hypothesis that all estimated
coefficients are equal to zero.

32

Page 32 of 34

Table 5. Determinants of Disaggregate Portfolio Flows


Bank

Bank
Idiosync.

Bond

Bond
Idiosync.

Equity

Equity
Idiosync.

[1]
0.477***
0.078***
0.939***

[2]
0.564***
0.070***
-0.745***

[3]
0.480**
0.116***
1.106***

[4]
0.467**
0.099***
-0.665**

[5]

[6]

0.448***

-0.375***

0.283***

-0.037***
0.411***

Explicators
FOit
Iit
HCit
Yit
Rit
RCP.t
RSRUS.t
RLRUS.t
VIX.t
Y.t G7
Constant
NxT
N
R2
F-statistic

0.019***
0.269***

-11.177***
694
45
0.120
17.629***

0.440
694
45
0.135
25.072***

-9.804***
522
35
0.096
10.286***

0.019***
-0.134***
0.061**
0.447***
-2.254
522
35
0.069
7.101***

-0.027***
-4.365***
544
34
0.112
21.357***

3.200***
544
34
0.032
8.274***

Notes: This table presents evidence on the determinants of disaggregate total and idiosyncratic capital flows to
emerging markets. Table 5 estimates equation (6) and (7) in the main text for disaggregate data. Column [1] seeks to
examine whether the following determinants are important for disaggregate Bank inflows: Financial Openness (FOit),
Institutions (Iit), Human Capital (HCit), country specific economic growth (Yit) and interest rates (Rit) from recipient
countries. Also column [1] contains global determinants of capital inflows: RCP.t is real commodity prices excluding
oil, RSRUS.t is the real short-run US interest rate, RLRUS.t is the real long-run US interest rate, VIX.t is a measure of
market uncertainty and Y.tG7 is real GDP growth in the G7. Column [2] is the estimation results for idiosyncratic bank
flows. Column [3] and [4] are for bond flows. [5] and [6] repeats the analysis equity inflows. Idiosyncratic data is
obtained by removing the global factor using Bai and Ng (2004) from the aggregate data, see equation (3). The time
dimension is 1993 to 2009, and the data is annual here due to data availability. Explicators that are statistically
significant are denoted by asterisk: *** at 1%, ** at 5% and * at 10% level of statistical significance. The F-statistic
tests the joint null hypothesis that all estimated coefficients are equal to zero.

33

Page 33 of 34

Table 6. PANIC Analysis of Capital Flow Volatility


FACTOR
Panel A
Aggregate
Bank
Bond
Equity
Panel B

-3.767*
-4.110*
-3.768*
-3.742*

IDIOSYNCRATIC
9.362*
Disaggregate
18.335*
15.508*
21.758*

IC1

IC2

IC3

5
5
5

5
5
5

0
1
5

Factor Correlations
Bank Factor
Bond Factor
Equity Factor

0.77
-0.05
0.89
Aggregate Factor

-0.03
0.88
Bank Factor

-0.11
Bond Factor

Notes: This Table examines the statistical properties of the volatility of our capital inflow data. Panel A presents
evidence on the number of common factors and whether the common factor and idiosyncratic components of volatility
are nonstationary, using Bai and Ngs (2004) PANIC approach. Time period is 1994Q1 to 2008Q4. Number of cross
sections is N = 29, 46, 35 and 34 respectively for Aggregate, Bank, Bond and Equity flows. Volatility is measured as a
rolling standard deviation with a window of 12 monthly observations. Rejected null hypotheses of nonstationarity are
denoted by an asterisk (*) and in bold. Panel B contains correlations of the first principal component of the Aggregate
and Disaggregate volatility data. See Notes to Table 2 for more details.

Table 7. Explicators of Common Factors in Flow Volatility


RCPt

RSRUSt

RLRUSt

VIXt

YtG7

Aggregate

0.50a

-0.29a

-0.53a

-0.04

-0.22

Bank

0.36a

-0.57a

-0.55a

0.23b

-0.48a

Bond

0.00

0.25b

-0.05

-0.09

0.18

0.17

-0.35a

Equity

0.44

-0.43

-0.56

Notes: The values in this Table represent bivariate correlations of volatility of the various capital flows
factors with explanatory variables. The time period is 1994Q1 to 2008Q4. Lag length determined by
AIK. RCPt is real commodity prices excluding oil, RSRUSt is the real short-run US interest rate, RLRUSt
is the real long-run US interest rate, VIXt is a measure of market uncertainty and YtG7 is real GDP
growth in the G7. Statistical significance of the correlation in bold denoted by superscript a at 5% and b
at 10% level of statistical significance.

34

Page 34 of 34