Content uploaded by Eric Rougier
Author content
All content in this area was uploaded by Eric Rougier on Jan 08, 2019
Content may be subject to copyright.
The effect of macroeconomic instability on FDI flows:
Agravityestimationoftheimpactofregionalinte-
gration in the case of Euro-Mediterranean agreements
Dalila Chenaf-Nicet
a,
n
, Eric Rougier
b
a
University of Bordeaux, LAREFI Avenue Léon Duguit, 33008 Bordeaux, France
b
University of Bordeauux, GREThA Avenue Léon Duguit, 33608 Pessac, France
article info
Article history:
Received 25 September 2014
Received in revised form
5 October 2015
Accepted 7 October 2015
Available online 26 October 2015
JEL classification:
F21
F43
F44
Keywords:
FDI
Gravity model
European Union
Middle East and North Africa
Regional trade integration
abstract
In order to diversify their risks, firms facing uncertainty in their
domestic market may choose to increase their investment abroad by
transferring production to more stable host economies. By estimating a
gravity model of foreign direct investment (FDI) flows from Europe
and the Mediterranean region to the four main recipients of FDI in the
Middle East and North Africa (MENA) region from 1985 to 2009, this
article tests (1) the extent to which FDI inflows are affected by mac-
roeconomic volatility in the source country and (2) whether regional
trade and investment agreements could have increased this FDI sen-
sitivity to external macroeconomic volatility. We find that the inci-
dence of FDI between two countries increases with source GDP
instability and with host GDP stability. Moreover, FDI to MENA coun-
tries tends to be countercyclical with respect to the source country’s
business cycle. We also find that although FDI reactivity to host
country’suncertaintyisnotconditionedbyNorth–South trade and
investment agreements, it becomes negative for South–South regional
integration. Last, we show that although the source country’s
instability certainly matters when explaining bilateral FDI flows in our
sample, its impact may be less important when investments are driven
by cost differentials, that is, for vertical investment.
&2015 CEPII (Centre d’Etudes Prospectives et d’Informations
Internationales), a center for research and expertise on the world
economy. Published by ElsevierB.V.Allrightsreserved.
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/inteco
International Economics
http://dx.doi.org/10.1016/j.inteco.2015.10.002
2110-7017/&2015 CEPII (Centre d’Etudes Prospectives et d’Informations Internationales), a center for research and expertise on
the world economy. Published by Elsevier B.V. All rights reserved.
n
Corresponding author.
E-mail addresses: Dalila.chenaf-nicet@u-bordeaux.fr (D. Chenaf-Nicet), eric.rougier@u-bordeaux.fr (E. Rougier).
International Economics 145 (2016) 66–91
1. Introduction
Foreign investment is supposed to convey positive effects, such as technological upgrading and
trade expansion, to developing economies. Attracting FDI from multinational corporations (hereafter
MNCs) has therefore become a priority goal of most developing countries. Nonetheless, although
labor-abundant Middle East and North African (MENA) countries have made significant efforts, since
the mid-1990s, to increase their attractiveness through adjustment, stabilization, and liberalization
policies, they still receive few FDI flows when compared to other low- and middle-income
economies.
1
Weak institutional governance and limited market size have been pointed by many
studies as good candidate explanations for these disappointing outcomes (Malik and Awadallah,
2013; Chenaf-Nicet and Rougier, 2011). However, during the 1990s, MENA countries deeply reformed
their institutions and opened up their economies to foreign trade and investment notably via various
South–South (GAFTA, AMU) and North–South (Euro-Mediterranean) trade agreements (Alaya et al.,
2009; Mina, 2012). As a result, although FDI inflows have been significantly augmented for the four
main MENA recipient countries during the two last decades, FDI instability has simultaneously been
amplified (UNCTAD, 2009).
We argue in this article that source countries’macroeconomic conditions influencing the
decision of MNCs to invest abroad should be more closely investigated to understand the patterns
of FDI flows to the Arab region. Over the last three decades, MENA economies, especially the labor-
abundant ones, have become increasingly dependent on the European MNCs investment to
modernize their productive structures and provide jobs to their educated workers.
2
Like all MNCs,
European firms partially determine their investment decision by considering the demand con-
ditions on their domestic market, with horizontal investment being stimulated by a more unstable
demand home. Moreover, this dependence of FDI inflows to MENA on European demand
instability has probably gone stronger as trade integration between the two regions got deeper
over the last two decades. As a direct effect, the size and steadiness of European FDI flows to the
MENA region have become increasingly dependent on the source countries’macroeconomic
volatility.
Our aim in this article is to test this assumption by identifying the determinants of FDI flows
going to MENA economies, not from the point of view of their own factors of attraction but rather
by focusing our analysis on the way macroeconomic instability in source countries may condition
them. In other words, we seek to identify how FDI reacts to the source country’smacroeconomic
conditions, which may increase uncertainty for their MNCs, and to the synchronization of busi-
ness cycles in home and foreign economies. We also address the conditioning impact of regional
trade integration, between European and MENA economies and between MENA economies, on
this reaction.
In our article, macroeconomic uncertainty is assessed by the three-year GDP volatility
measuring short-term demand instability. Demand instability is supposed to have either a
positive or a negative impact on FDI flows.
3
On the one hand, firms facing increasing demand
uncertainty at home may be willing to invest abroad in order to diversify their portfolio of
consumption markets and to limit their exposure to the risk of instability of their revenue on
their domestic market. On the other hand, seeking lower production costs abroad through
1
Moreover, they still fail to experience the technological spillovers they initially expected. Sadik and Bolbol (2001)
explained this fact by the nature of FDI inflows, mostly resource-based, during the 1990s. Chenaf-Nicet and Rougier (2011) have
provided evidence based on more recent data that this failure could be because of the low absorption capacities of poorly
innovative MENA economies.
2
FDI sourced in Gulf countries has also become increasingly strategic for MENA countries. However, we do not introduce it
into our estimations since it is a more recent phenomenon on which we lack of a sufficiently long time perspective.
3
Productivity shock may also spur GDP trend instability over the longer run, but we do not measure and address this
dimension in our article. Moreover, we also control for the productivity shocks that may condition vertical investment by
introducing a proxy for the cost differential between the source and the host countries.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 67
vertical investment may help maintaining MNCs’margins, despite the instability of demand in
home markets. Conversely, during a period of higher revenue instability in their home market,
firms may nevertheless be unable to invest abroad because of financial restrictions.
4
Even
though the sign of the effect of GDP volatility on FDI is thus a priori uncertain, we could
anticipate that, if MNCs are not financially constrained, both horizontal and vertical FDI would
be increased by higher demand instability home.
Understandably, governments in developing host countries have no direct influence over source
countries’macroeconomic conditions. Trade and investment regional integration policies may,
nevertheless, condition the extent to which FDI inflows to the host country react to the uncertainty to
which the MNCs are exposed in their home country. Deeper trade integration between source and
host countries, notably via bilateral investment treaties (BITs) and free trade agreements (FTAs), may
well magnify the impact of source countries’volatility on FDI outflows by reducing the costs of
reallocating production abroad and re-exporting from abroad. By reducing taxation and transaction
costs, regional integration may lessen, for a given level of macroeconomic risks, the average risk
threshold below which MNCs would accept to invest abroad. For instance, higher trade and invest-
ment integration may ease production relocation abroad in the case of increased uncertainty in the
home market, therefore stimulating FDI outflows to the more stable host economies of the trading
zone. We thus test if trade agreements involving MENA countries –notably but not exclusively with
European countries –have worked as a magnifying force and increased FDI responsiveness to external
macroeconomic conditions or, on the contrary, if they have reduced it by promoting the substitution
of direct trade to horizontal FDI.
Although the determinants of FDI concerning the host country are now well known,
5
those
concerning the source country’s macroeconomic characteristics have seldom been studied.
6
In
particular, the sensitivity of FDI to uncertainty in the source country has hardly been inves-
tigated so far, even though this issue is certainly of considerable importance for those
developing countries whose external balance of trade and financing of growth rely heavily on
foreign capital inflows. Several studies have sought to explain aggregate FDI outflows or
inflows by aggregate measures of global instability (Albuquerque et al., 2005; Méon and
Sekkat, 2012). However, using such aggregate measures does not enable addressing the source
country’smacroeconomiccharacteristics,whichmayconditionFDI.Indeed,veryfewpapers
have tried to address the impact of the source country’smacroeconomicconditionsonbilat-
eral flows. By estimating a gravity model of bilateral FDI flows between OECD economies
covering the 1985–2007 period, Cavallari and D’Addona (2013) have found that FDI has tended
to increase when the source country had higher output volatility. Focusing on North–South
FDI, Levy-Yeyati et al. (2007) have estimated a gravity model and found that FDI sourced in
Europe and the United States tended to be countercyclical with respect to both output and
interest rate cycles in the source country. According to the authors, investor arbitrage among
different investment opportunities explains that FDI outflows and local investment tend to
move in opposite directions during cycles in the United States and Europe. We can see that FDI
sensibility to a source country’s output instability and business cycle are two important issues
for whoever wants to understand FDI instability. To our knowledge, the conditioning impact of
trade integration on the macroeconomic volatility–FDI relationship has never been tested
thus far.
By estimating a gravity model of foreign direct investment (FDI) flows from Europe and the
Mediterranean region to the four main recipients of FDI in the Middle East and North Africa (MENA)
region from 1985 to 2009, we fill this literature gap and find evidence that the incidence of FDI
4
On the revenue and substitution effects of output instability, see Chenaf-Nicet and Rougier (2014).
5
Good institutions, a low-cost and highly productive workforce, the availability of natural resources, and market size are,
among other things, key determinants of between-country differences in the attraction of FDI. See Bloningen (2005) for an
overview of the literature on the determinants of FDI.
6
For recent empirical analyses of the adverse effects of macroeconomic volatility on economic development, see Loayza
et al. (2007).Lensink and Morrissey (2006) have shown that economic growth is more reactive to FDI volatility than to FDI
levels.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9168
between two countries increases with source GDP instability and with host GDP stability. Moreover,
FDI to MENA countries tends to be countercyclical with respect to the source country’s business cycle.
We also find that although FDI reactivity to host country’s uncertainty is not conditioned by
North–South trade and investment agreements, it becomes negative for South–South regional inte-
gration. Last, we show that although the source country’s instability certainly matters when
explaining bilateral FDI flows in our sample, its impact may be less important when investments are
driven by cost differentials, that is, for vertical investment. In a nutshell, our estimations show that, in
our sample of countries, FDI tends to flow from the more volatile source countries to the less volatile
host countries and that the sensitivity of these FDI flows to source country volatility is, as expected,
affected by trade integration and by the type of investment.
The remainder of this article is organized in five sections. Section 2 draws our main assumptions
from the relevant literature and Section 3 discusses the indicators of macroeconomic instability
selected for the empirical study and the estimation strategy. In Section 4, we first present and then
discuss the results of our gravitational model panel data estimation, paying specific attention to
source countries and to several issues relating to the robustness of our results. Section 5 discusses
robustness checks and Section 6 concludes.
2. FDI, uncertainty, and trade integration: Hypotheses and existing empirical results
Does FDI increase or decrease with source country demand volatility? Does trade integration
amplify or mitigate FDI sensibility to source country’s instability? This section investigates theoretical
and empirical works that will help formulating the paper’s empirical approach.
Although the theoretical predictions concerning the impact of source country demand instability
on FDI are not convergent, they tend to predict a positive impact of source country’s demand
instability on FDI, notably when firms are not financially constrained.
In a model where the multinational making investment decisions faces demand shocks, Aizenman
and Marion (2004) have shown that higher volatility of demand reduces the expected profit asso-
ciated with both horizontal FDI and vertical FDI. They describe the specific mechanisms underlying
the adverse FDI impact of demand shocks and claim that they are applicable for both horizontal and
vertical FDI. However, their prediction of a systematically adverse impact relies on strong restrictions
7
and they assume a multiplicative demand instability combining instabilities on both home and for-
eign markets. Their prediction is therefore not relevant for our purpose since we aim at isolating the
FDI effect of source country’s demand volatility.
If we want to make the distinction between source and host country’s demand conditions, we
must consider that the decision to invest abroad in response to macroeconomic uncertainty, and the
ensuing level of FDI flows between two countries, is in fact the result of two simultaneous decisions
of investment under uncertainty in home and foreign markets. Firms actually choose whether they
will invest or not, and whether they will invest in home or in foreign economies. According to the
standard option-pricing analysis of investment under uncertainty, the return threshold that is
required for performing an irreversible investment, that is, an investment characterized by positive
sunk cost and low convertibility or liquidity, increases with uncertainty at home (Dixit, 1989; Dixit
and Pindyck, 1994; Marschak, 1949). In a context of increasing uncertainty concerning home
demand, delaying investment may therefore constitute an optimal strategy for a firm because
waiting for new information potentially raises the investment’s expected value (Bernanke, 1983;
McDonald and Siegel, 1986; Pindyck, 1988). Delaying investment therefore may turn out to be a
valuable option for the MNC reacting to the anticipated instability of its expected profits by holding
back on all its investment projects, including planned foreign investment (Aizenman, 2003; Wang
7
They assume that profits are concave with respect of the demand shocks. A positive demand shock, by inducing tensions
on the supply-side of the market, will increase price and may therefore reduce future sales and profits of firms, especially for
sectors such as manufacturing that feature high price elasticity. A negative demand shock will cut profits by adversely affecting
the current amount of sales, provided that the positive effect of decreasing prices on sales remains limited (Aizenman and
Marion, 2004: 131).
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 69
and Wong, 2007).
8
In the context of internationally integrated economies, however, if macro-
economic uncertainty is lower in the foreign economies with which the home economy trades,
investing abroad may constitute an additional option, besides delaying domestic investment,
available to the firms experiencing uncertainty in their home market (Brandão de Brito and de
Mello-Sampayo, 2005).
When uncertainty is lower in the foreign than in the home economy, however, setting up a subsidiary
abroad may constitute a possible alternative to delaying investment. In a nutshell, a firm will be particularly
responsive to demand uncertainty in the source country when it decides ex ante to invest abroad or not,
and simultaneously to uncertainty in the host country when it comes to choosing in which country to
invest.
9
Moreover, since risk-reducing FDI might increase when the home and foreign short run business
cycles are negatively correlated, FDI outflows accordingly may be higher from more volatile home countries,
while FDI inflows may be higher for those host economies where uncertainty is the lowest.
These theoretical predictions have hardly been empirically tested thus far. Cavallari and D’Addona
(2013) have found that FDI between OECD countries tended to increase during the period from 1985
to 2007 when the source country had higher output volatility. Although Cavallari and D’Addona
(2013) do not specify the type of investment for which their results hold, we may assume that it is
essentially North–North FDI, which is mostly horizontal. By estimating a gravity model of North–
South FDI, however, Levy-Yeyati et al. (2007) have found evidence that FDI flows tend to be coun-
tercyclical with respect to output cycles in the United States and Europe. These results would confirm
that investors choose between investment options at home and abroad on the basis of the volatility
differential between the source and host economies.
An additional issue raised by the latter study concerns the conditioning impact of trade integration.
Levy-Yeyati et al. (2007) finds evidence that recessions in industrial countries are likely to increase FDI flows
to developing countries with close ties to the United States and Europe. This latter result suggests that FDI
sensibility to uncertainty may well be magnified when the host and source economies are interlinked by
trade or investment treaties. It is now well documented that FDI tends to be triggered by global (Büthe and
Milner, 2008)aswellasbyregional(Busse et al., 2010; Daude et al., 2003; Medvedev, 2012)tradeinte-
gration. Daude and Stein (2007) and Jaumotte (2004) have provided convincing evidence of this positive
effect in the case of both North–South and South–South trade agreements. Similarly, bilateral investment
treaties have positive effects on FDI inflows to developing economies in general (Desbordes and Vicard,
2009)andtoMENAeconomiesinparticular(Mina, 2012). In addition, RTAs offering more liberal rules for
admission and provisions for foreign investment logically have a higher positive impact on FDI (Berger et al.,
2013). Lower tariffs and regulatory obstacles between two countries may be favorable to vertical FDI
because the cost of importing components and re-exporting is minimal and, on the contrary, detrimental to
horizontal FDI by enabling trade substitution for FDI (Caves, 1996; Markusen, 1984, 2002). As for horizontal
investment, it may be affected differently by trade integration (Markusen and Maskus, 2001). First, this type
of FDI is essentially relevant to countries with similar characteristics. Second, it may be adversely affected by
trade integration because trade costs reduction may prompt onshoring and exporting.
The net conditioning impact of free trade agreements on FDI responsiveness to demand instability may
therefore depend on the dominant form of FDI. Because it might be more vertical than horizontal (Blo-
ningen and Wang, 2005), European investment to MENA might react differently to source volatility than
South–South investment, which is more likely to be horizontal. We will control for these two possible
regimes in this article’sestimations.
3. Methodological issues
In order to assess FDI responsiveness to source country instability, we use a gravity model that links 32
countries that were, during the period 1985–2009, sources of investment to the four largest recipient
8
Of course, this prediction assumes that firms are not financially constrained as a consequence of growing uncertainty home.
9
When choosing the localization of its foreign investment, the MNC will therefore consider the host country’s char-
acteristics in terms of average growth and instability in the case of a horizontal investment and in terms of costs in the case of
vertical investment. See De Mello-Sampayo et al. (2010) and Brandão de Brito and de Mello-Sampayo (2005) for two recent
theoretical and empirical analyses of the option theory applied to FDI.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9170
countries in the MENA region (Egypt, Morocco, Tunisia, and Turkey).
10
MENA countries are particularly
concerned by the trends described in this article because they have had to overcome a significant increase
of inward FDI levels after the 1995 Barcelona Agreement (Chenaf-Nicet and Rougier, 2014), their economies
becoming increasingly dependent onFDIsourcedinEuropeaneconomiesaswellasinothercountriesof
the MENA region. We expect that European investment to MENA economies may have become more
reactive to source macroeconomic conditions with the extension of regional trade integration as well with
the bilateral trade agreements between these two regions. Insofar as the source countries of our sample
include the European Union (EU) economies, plus theMENAcountries(Mauritania,Morocco,Algeria,
Tunisia, Libya, Egypt, Jordan, Syria, and Turkey), our sample encompasses both North–South and South–
South trade agreements and resulting flows of FDI.
11
Because cross-sectional or time series studies of FDI determinants are constrained by their frame-
work to use a single average measurement of external conditions, thereby failing to address source-
related determinants of FDI,
12
we had to use a gravity model to properly assess source-related mac-
roeconomic determinants of FDI levels. The gravity model is increasingly used to explain bilateral flows
of FDI because it enables the effect of host country characteristics on FDI to be differentiated according
to a series of distance-related factors.
13
Gravity models consider that capital flows between a pair of
countries increase as a function of their national incomes (measured by GDP per capita) and decrease as
a function of the distance between them (measured by the kilometric distance between countries).
Control variables that relate to origin or destination individual countries can be incorporated into the
equations of gravity models. These control variables enable multilateral resistance factors to be taken
into account. Resistance factors explain why natural relationships can be blocked even when countries
are close (Anderson and van Wincoop, 2003). They include the existence of special transaction costs,
capital movement controls, information costs, trade or monetary agreements, or differences in com-
mercial practices and in languages. Empirical studies inspired by the works of Anderson and van
Wincoop (2003) and Anderson (2011) generally use an equation of the form:
G
ijt
¼g"M
s
i
it
"M
s
j
jt
d
n
ij
"#
"X
ijt
exp
θ
ij
"T
ijt
ð1Þ
where d
ij
is the kilometer distance between countries iand j,M
it
and M
jt
are attraction variables such as
economic size of markets, T
ijt
are resistance factors, and s
i
,s
j
, and θ
ij
are parameters to be estimated.
To be tested with standard estimators, Eq. (1) has to take a linear form. Log linearization is a robust
method (Deardorff, 1998) if the dependent variable does not take the value 0 and if there are no
heteroskedasticity problems (Arvis and Shepherd, 2013; Burger et al., 2009; Gómez-Herrera, 2013;
Santos Silva and Tenreyro, 2006).
14
When log-linearized, Eq. (1) becomes:
Log G
ijt
¼Log gðÞþS
i
Log M
it
ðÞþS
j
Log M
jt
!"
&nLog d
ij
!"
þX
ijt
θ
ij
"T
ijt
þϵ
ijt
ð2Þ
where Log(g) is a gravitational constant, M
it
and M
jt
represent the economic size of countries iand j,s
i
and s
j
are positive coefficients (attractive strength), and na negative coefficient (repulsive strength).
The sign of θ
ij
depends on hypotheses about variable T
ijt,
which may be resistance factors such as
differences in language, practices, the existence of capital flow controls, taxes on capital flows, the
10
Even though Algeria is also a big FDI recipient, this country was not included in the host country sample because inward
FDI is highly concentrated on oil and is likely to adopt a very different pattern. MENA Gulf countries are therefore considered
neither as host nor as source countries in our analysis because they are not as closely associated with European trade and
investment as Mediterranean ones.
11
The list of countries included in the panel is presented in Appendix 2,Table A2.
12
Méon and Sekkat (2012) is a recent illustration: they proxy external macroeconomic volatility using an aggregate ratio of
world FDI to world GDP.
13
Gravity models are inspired by equations of gravity in physics that relate the force with which two bodies attract each
other proportionally to the product of their masses and inversely to the square of the distance between them (Frankel, 1997).
For a theoretical analysis of gravity models in economics, see Anderson (1979). For recent studies using gravity models to
analyze FDI flows, see Bevan and Estrin (2004),Busse et al. (2010),Desbordes and Vicard (2009) and Frenkel et al. (2004).
14
In our calculations we tested for the presence of any heteroskedasticity problem with the Breuch–Pagan test and when
we detected heteroskedasticity we estimated a robust OLS equation using White’s correction.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 71
presence or absence of bilateral or multilateral agreement, exchange rate risk, and so on. The
expression ε
ijt
is a white noise. When masses and observable elements of multilateral resistance are
formally expressed, Eq. (2) becomes:
LnðFDI
ijt
Þ¼aþβ
1
LnðGDP
it
Þþβ
2
LnðGDP
jt
Þþβ
3
LnðD
ij
Þþβ
4
Source instability
t
þβ
5
Host instability
t
þβ
6
RTAs
ijt
þβ
7
BITs
ijt
þβ
8
Institut ional prof ile
jt
þu
i
þu
j
þv
t
þε
ijt
ð3Þ
where FDI
ijt
represents the value in dollars of the inflows of FDI from a country i(source country)
entering country j(host country) at time t.
15
If we now consider the right-hand side of Eq. (1b), Ln
(GDP
it
) and Ln(GDP
jt
) stand for the natural logarithm of GDP levels of the source and host countries,
respectively, and β
1
and β
2
take a positive sign if there is a “mass”effect operating in determining
bilateral direct investment flows. By extension, higher host country GDP is generally considered to
increase horizontal FDI because the size of the local market is worth being served by a multinational
firm’s production subsidiary.
D
ij
is the vector of the various concepts of distance controlling for the most typical sources of
transaction and transport costs involved in an investment moving from one country to another. The
physical bilateral distance (distance) corresponds to the distance between the countries’capitals; FDI
is generally taken as being inversely proportional to the distance between the two countries involved.
However, when the host country shares a common border, language, or a former colonial link with
the source country, it is generally considered that FDI will be higher. We use also two variables noted
adjacency and common language, which take the value 1 if the source and host countries respectively
share a common border or have a common language; otherwise, they take the value 0.
16
The variable
past colonial links takes value 1 if the source country had colonized the host country, and
0 otherwise.
17
In the literature, demand volatility is generally measured as the standard deviation of the annual
growth rate of GDP within a rolling five-year window (Aghion and Banerjee, 2005; Aizenman and Ito,
2012; Blanchard and Simon, 2001; Ramey and Ramey, 1995). Other methodologies exist although
they are less common and straightforward. A measure of GDP growth volatility based on the standard
deviation of the output gap has also been applied, but it is reported to overestimate short-term
volatility (Kent et al., 2005). Aizenman and Marion (2004) use the standard deviation of the inno-
vation from a first-order autoregressive process based on 20 years of annual data. This approach
requires a sufficiently long series of past data in order to be able to estimate autoregressive processes
for the first sample years, which is not our case. In our study, for each time period, host and source
instability were calculated as the standard deviation of GDP growth. Mean and standard deviation
values at time thave been computed as a five-year moving average over t-4, t-3, t-2, t-1,and t.We
have supposed that investors observe short-term past volatility and compare it for different potential
destinations. In order to avoid a null average value, we have chosen to compute absolute values of
standard deviations and then to express them in logarithmic form.
18
As argued, FDI inflows also
depend on the characteristics of the source country and source region in terms of GDP growth
instability. The expected sign of the source instability coefficient was discussed in the previous section.
It may be positive if FDI is countercyclical or if MNCs choose to invest abroad instead of simply
delaying domestic investment when source country’s demand instability increases. We can anticipate
that the coefficient for host instability could be either negative or positive, but the opportunity-driven
15
Data sources and definitions can be found in Appendix A1.
16
Bénassy-Quéré et al. (2007) and Abderrezak (2008) have provided evidence supporting the view that former colonial
links, through the institutional, linguistic, and cultural proximities that they produce between source and host countries, may
have a positive influence on the creation of international trade or FDI networks.
17
It should be noted that past colonial links is a good proxy for legal origin, which appears to be significant in explaining
bilateral portfolio investment flows (Lane and Milesi-Ferretti, 2008) as well as bilateral FDI flows (Daude and Stein, 2007).
18
Although source instability is not likely to be endogenous to FDI levels, host volatility theoretically may be affected by the
contemporary level of incoming FDI. To limit this risk, source instability in period tis computed as a three-year moving average
including periods t-3, t-2, and t-1. The same lags have been used to compute all our average variables: MENA instability, MENA
growth, and European instability.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9172
positive effect seems to be plausible for MENA economies, given the nature of the foreign investments
they tend to attract.
As for the factors associated with trade and investment integration, two variables have been
introduced. RTAs is a vector of dummy variables measuring each pair of countries’participation in a
regional trade agreement or investment treaty. This means that prior to the agreement being effec-
tive, the dummies take the value 0. For each consecutive year, the value 1 is given to the FDI flow
whose source and host countries are bound by an active RTA. Because our study uses a sample of both
MENA and European countries, we explicitly introduce controls for membership of three regional
trade agreements (GAFTA,AMU, and Euro-Mediterranean Free Trade Area,noted as MED). The peri-
meter and content of these three RTAs are fairly different. AMU (for Arab Maghreb Union) is the oldest
trade agreement among MENA countries. It was originally designed in 1989 to prepare for an eco-
nomic and future political unity among the Arab countries of North Africa (Algeria, Libya, Mauritania,
Morocco, and Tunisia) but has remained fairly ineffective because of political tensions and rivalries.
GAFTA (for Greater Arab Free Trade Area) was introduced in 1997 through an initiative made by the
Arab League. The agreement involved progressive reductions in customs duties and was extended to
the gradual elimination of trade barriers among 17 Arab countries (Algeria, Bahrain, Egypt, Iraq,
Kuwait, Lebanon, Libya, Morocco, Oman, Palestine, Qatar, Saudi Arabia, Sudan, Syria, Tunisia, United
Arab Emirates, and Yemen). The Euro-Mediterranean Free Trade Area (EMFTA) is a free trade zone,
introduced by the Barcelona Agreement adopted in 1995, which is built through (1) a series of
bilateral free trade agreements between the European Union and each state bordering the Medi-
terranean and (2) horizontal free trade agreements between the non-EU Mediterranean countries
themselves, such as the Agadir Agreement, which came into force in March 2007. The MENA countries
involved are Algeria, Egypt, Israel, Jordan, Lebanon, Morocco, Palestinian Territories, Syria, Tunisia,
and Turkey. Here, we focus exclusively on bilateral trade agreements between the European Union
and the MENA individual host countries of our sample because they are often associated with
increased export-processing FDI. Likewise, BITs is a vector of dummy variables measuring each pair of
countries’participation in an FDI agreement. These agreements cover both bilateral and multilateral
(regional) agreements such as those associating the European Union with each MENA host country.
Data on both RTAs and BITs are taken from UNCTAD.
Last, because investment decisions made by MNCs generally use a global evaluation of host
country business regulations (Ali et al., 2010), any empirical assessment of FDI flows requires the
introduction of a variable to control for institutional quality. Moreover, omitting indexes of institu-
tional quality biases typical gravity model estimates of trade, as was shown by Anderson and Mar-
couiller (2002). Accordingly, the ICRG investment profile comprehensive indicator
19
(denoted
investment profile) has been introduced into the estimations to control for these institutional elements
of transaction costs.
As is now standard in the gravity literature (Mátyás, 1997; Feenstra, 2004; Redding and Venables,
2004), time and source country and host country fixed effects (u
t
,u
i
,and u
j
) have been introduced in
order to control for the multilateral resistance terms identified by Anderson (1979) and popularized
by Anderson and van Wincoop (2003). Because our model does not include time-invariant expla-
natory factors, the inclusion of country fixed effects would be theoretically possible without causing
multicolinearity. Similar to various recent papers (Andrés et al., 2013; Cezar and Escoba, 2015;
Kleinert and Toubal, 2010), we have estimated a gravity FDI model including time-invariant country
fixed effects.
20
The Hausman tests that were conducted have confirmed that the fixed effect model
should be preferred to the random effect model because it is more consistent and more efficient.
19
This index captures the quality of the enforcement of business regulations and property rights by combining ratings of
contract viability, risks of expropriation, repatriation of profits, and delays in payments.
20
For panels with a sufficiently large number of years for the underlying factors of multilateral resistance to be able to
change, however, source and host fixed effects can be time varying (Head and Mayer, 2013). Because this would lead time-
varying country fixed effects to be collinear to our variables of interest, for example, source and host country volatility, we
could not opt for this strategy. One possible solution to deal with this problem is to estimate the gravity model with time-
varying fixed effects and to use the multiplicative term of source and host country volatility. However, the estimated coefficient
for the latter variable is difficult to interpret, which is inappropriate given our purpose of understanding how source and host
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 73
All data and variables used in our estimations are presented in Appendix 1:Table A1 (data and
definitions) and the list of countries in the sample is shown in Appendix 2:Table A2.
FDI datasets generally contain a large number of zeros. Several methods are used in the panel
gravity model literature to deal with the issue of zero-value FDI flows in a logarithmic model.
21
Santos Silva and Tenreyro (2006), however, showed that when the model does suffer from het-
eroskedasticity, the gravity model Poisson estimation is preferred. The Breush–Pagan tests reported
in the lower panel of columns A3.1 and A3.2 (Table A3) show that we cannot reject the hypothesis of
heteroskedasticity in our case. As a result, our preferred estimator is the Poisson Pseudo-Maximum
Likelihood. Results of our Poisson regressions are reported in Sections 4 and 5. Because the coef-
ficients of interaction terms in nonlinear models such as Poisson Pseudo-Maximum Likelihood
cannot be directly interpreted, as discussed in Gill (2001), we follow the literature (Andrés et al.,
2013) by estimating the model in incidence rate ratios (IRRs). IRRs can be interpreted directly as the
odds of a source country choosing to invest in a host country versus the odds of not choosing that
host country, instead of in mere probabilities.
22
Note that IRRs less than 1 reveal a negative impact
of the corresponding determinant on bilateral FDI flows, whereas ratios greater than 1 reveal a
positive impact.
4. FDI, macroeconomic volatility, and trade integration
4.1. The effect of source GDP volatility on FDI
In Table 1, the gravity model was estimated first by the now standard Pseudo-Poisson Maximum
Likelihood method (column 1.1) and then, with the coefficients expressed in IRR (column 1.2). Source
instability, host instability, and the ICRG investment profile are being introduced step by step in
columns (1.3) and (1.4).
23
First of all, the estimation of the standard gravity model (column 1.2) is in
accordance with the typical results reported in the literature, although both source GDP and host GDP
significantly increase FDI flows to the four MENA countries. As far as physical distance is concerned,
our results are contrasting. Although FDI flows between two countries seem to be unaffected by the
existence of common borders, they nevertheless tend to decrease with geographical distance. A
common language shared by the source and host countries increases FDI flows whereas the existence
of past colonial links between two countries has no effect.
If we now turn to our variables of interest, that is, source and host country instability, column
1.3 confirms that they leave column 1.2’s results unchanged. Nonetheless, column 1.3 indicates that
source instability and host instability have opposite effects on FDI. Host instability has a significant and
negative impact on FDI flows whereas source instability has a significant and positive impact. In
column 1.4, the positive coefficient for host investment profile indicates that good institutional gov-
ernance has a strong influence on FDI to MENA countries. Moreover, the introduction of host
investment profile does not modify the results for source and host instabilities, thereby indicating that
(footnote continued)
volatility independently affect FDI flows between two countries. Moreover, for panels with limited time variation, such as ours,
it is reasonable to assume that sources of multilateral resistance move only slowly (Bergstrand and Egger, 2007).
21
The most common is the Eichengreen correction, coupled with random effect estimation, which consists of using a
transformation of the form ln(1 þFDI). This method is widely used because it is simple and it enables the coefficient to be
interpreted as elasticity when the value ln(1þFDI) is approximately equal to ln(FDI), which is accepted as a reasonable
assumption (Eichengreen and Irwin, 1998). The model can also be estimated by using the Tobit method, which explicitly
accounts for zero FDI flows, without excluding them. This increases the variation of the dependent variable, thereby producing
higher values and significance for the estimated coefficients of the various determinants of FDI (Eaton and Tamura, 1994; Head
and Ries, 2008; Wei, 2000). Our baseline model random effect estimations, with the Eichengreen correction and with the
random effect Tobit estimator, are reported in Table A4.
22
According to Gill (2001) and Andrés et al. (2013), this transformation renders the specification of interaction terms
straightforward as in a linear model so that they can be estimated with standard numerical procedures for maximum
likelihood.
23
Time series’stationarity tests for the variables of interest have been reported in Table A5 of the Appendix. The results of
the Levin–Lin–Chu unit-root test have been confirmed for the Im–Pesaran–Shin unit-root test.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9174
the risks raised by macroeconomic characteristics must not be confused with the risks imposed by
bad institutional governance and high transaction costs. Last, a multiplicative variable has been
introduced to test the assumption that source and host uncertainties may have cumulated effects on
FDI bilateral flows. The results reported in column 1.5 indicate that uncertainties in source and host
countries do not cumulate their individual impact on FDI.
A related issue concerns pro-cyclical or countercyclical behavior of bilateral FDI in our sample of
countries. If firms seek to diversify their risks by investing more in a dynamic foreign market when
their home economy is depressed, then FDI will increase when home and foreign country business
cycles are negatively correlated. The fact that bilateral FDI flows tend to be higher when the source
and destination business cycles are not synchronized therefore signals that MNCs tend to substitute
foreign production to domestic production so as to reduce the microeconomic risks linked with
macroeconomic cyclical volatility. In order to test the existence of such a countercyclical pattern of
bilateral FDI, we have created a dummy de-synchro in three steps. As a first step, we have used a
Hodrick–Prescott filter in order to single out the source and host countries’GDP growth trends over
the whole period and kept the yearly cyclical component. Then, we have computed for each source
(host) country a dummy variable called source cycle (resp. host cycle) taking the value 1 for the years
when GDP growth was lower than the trend and taking 0 when GDP growth was above the trend. We
Table 1
Poisson Pseudo-Maximum Likelihood model with country and time fixed effects: Baseline model and baseline with controls.
Dependent variable FDI
levels
1.1 1.2 1.3 1.4 1.5
PPML IRR IRR IRR IRR
Source GDP 0.561
(16.62)
nnn
1.754 (16.62)
nnn
1.566 (11.04)
nnn
1.637 (13.28)
nnn
1.639
(13.27)
nnn
Host GDP 0.330 (11.17)
nnn
1.391 (11.17)
nnn
1.599 (13.21)
nnn
1.2172 (4.46)
nnn
1.216 (4.46)
nnn
Distance &0.608
(&2.63)
nnn
0.544
(&2.63)
nnn
0.597 (&2.18)
nnn
0.645 (&1.93)
nn
0.645
(&1.93)
nn
Adjacency &0.335
(&1.20)
0.715 (&1.20) 0.732 (&1.12) 0.730(&1.17) 0.729 (&1.17)
Past colonial links &0.191
(&0.33)
0.825 (&0.33) 1.056 (&0.09) 0.909 (&0.17) 0.907 (0.17)
Common language 0.831 (2.95)
nnn
2.296 (2.95)
nnn
1.957 (2.38)
nn
1.805 (2.16)
nn
1.808 (2.16)
nn
Source instability –– 1.030 (2.29)
nn
1.037 (2.87)
nnn
1.027 (0.85)
Host instability –– 0.214 ( &10.66)
nnn
0.470
(&4.63)
nnn
0.466
(&4.63)
nnn
Host investment profile –– – 1.089 (10.16)
nnn
1.089 (10.13)
nnn
Source instability
n
host
instability
–– – – 1.0407 (0.36)
Constant &10.81
(&6.00)
nnn
0.00002
(&6.00)
nnn
9.47E
&06
(&6.20)
nnn
0.0001
(&4.42)
nnn
0.0001
(4.92)
nnn
Log-Likelihood &7024.5758 &6011.887 &5960.4606 &5960.3967
Log-Likelihood ratio test χ
2
¼5337.16
nnn
χ
2
¼4945.80
nnn
χ
2
¼4699.39
nnn
χ
2
¼4684.4
nnn
Wald test χ
2
¼2182.49
nnn
χ
2
¼1246.37
nnn
χ
2
¼1311.78
nnn
χ
2
¼1311.9
nnn
Country FE vs. pooled Yes
nnn
Yes
nnn
Yes
nnn
Yes
nnn
Time effects vs. pooled Yes
nnn
Yes
nn
Yes Yes
Hausman test χ
2
(6)¼44.45 χ
2
(8)¼44.22 χ2 (9)¼85.34 χ2 (9)¼55.93
Prob4χ
2
¼0.00 Prob4χ
2
¼0.00 Prob4χ
2
¼0.00 Prob4χ
2
¼0.00
Note: IRRs of less than 1 reveal a negative impact of the corresponding determinant on bilateral FDI flows, whereas ratios
greater than 1 reveal a positive impact.
n
Significant at 10% risk.
nn
Significant at 5% risk.
nnn
Significant at 1% risk.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 75
finally have ascribed to each country pair a dummy labeled de-synchro taking the value 1 in the years
when the two cycles were not synchronized and 0 otherwise.
We suppose that (1) the dummy source cycle will have a positive coefficient if bilateral FDI is
higher when the source country’s GDP growth cycle is located below its trend and (2) the de-synchro
dummy will have a positive coefficient if bilateral FDI is higher when the source and host countries’
cycles are not synchronized. De-synchro alone, however, does not inform about which pattern of de-
synchronization leads to the highest FDI levels. We may expect that FDI will be higher when the
source country is in a low conjuncture and host countries are in high conjuncture. In this case, FDI will
be countercyclical in the source country because it tends to increase when the source economy is in a
contraction period and procyclical in the host economy because it tends to increase in a period of
expansion. Because the interaction of the source cycle with de-synchro dummies gives the value 1 to
the episodes of de-synchronization with low conjuncture in the source country and 0 otherwise, a
positive value of its estimated coefficient would enable identifying this procyclical pattern. In addi-
tion, a positive and significant coefficient for de-synchro*source volatility might inform about the
prospect that the positive impact of source volatility on FDI flows would be higher when cycles are
not synchronized.
Results of the estimations of the gravity equation including the described dummies are reported in
Table 2. In column 2.1, the positive coefficient for source cycle first indicates that, other things being
equal, FDI increases when the source country’s GDP growth is low. Next, column 2.2 shows that the
de-synchro’s coefficient is positive and significant indicating only that FDI flows tend to be higher
when the source and host countries have de-synchronized cycles. However, in column 2.3, the
interaction de-synchro*source cycle takes a positive and significant coefficient making it clear that FDI
Table 2
Poisson Pseudo-Maximum Likelihood model with country and time fixed effects: With business cycles.
Dependent variable FDI levels 2.1 2.2 2.3 2.4
IRR IRR IRR IRR
Source GDP 1.717 (13.89)
nnn
1.668 (13.72)
nnn
1.695 (13.84)
nnn
2.231 (14.97)
nnn
Host GDP 1.203 (4.22)
nnn
1.243 (4.94)
nnn
1.217 (4.48)
nnn
1.028 (0.57)
Distance 0.620 (&2.12)
nn
0.622 (&2.08)
nn
0.623 (&2.09)
nn
0.498 (&3.00)
nnn
Adjacency 0.730 (&1.16) 0.734 (&1.14) 0.733 ( &1.15) 0.686(&1.31)
Past colonial links 0.824 (&0.34) 0.881 (&0.22) 0.849 (&0.29) 0.480 (&1.26)
Common language 1.968 (2.46)
nnn
1.916 (2.36)
nn
1.939 (2.41)
nnn
3.232 (3.89)
nnn
Source instability 1.042 (3.24)
nn
1.042 (3.20)
nn
1.042 (3.24)
nnn
1.040 (3.05)
nnn
Host instability 0.545 (&3.62)
nnn
0.655 (&2.43)
nnn
0.592 (&3.07)
nnn
0.710 (&2.04)
nn
Host investment profile 1.076 (8.06)
nnn
1.069 (7.23)
nnn
1.073 (7.61)
nnn
1.056 (6.01)
nnn
Source cycle 1.093 (3.77)
nnn
––0.483 (&4.42)
nnn
De-synchro –1.084 (5.26)
nnn
––
De-synchro
n
source cycle ––1.062 (4.32)
nnn
–
Source cycle
n
source instability –––&0.883 ( &0.73)
Constant 0.000092
(&5.16)
nnn
0.000085
(&5.15)
nnn
0.00009
(&5.14)
nnn
0.000063 (&5.35)
Log-Likelihood &59,553.3733 &5946.688 &5951.1972 &5921.3883
Log-Likelihood ratio test χ
2
¼4713.46
nnn
χ
2
¼4713.82
nnn
χ
2
¼4717.60
nnn
χ
2
¼4763.21
nnn
Wald test χ
2
¼1323.49
nnn
χ
2
¼1330.17
nnn
χ
2
¼1325.20
nnn
χ
2
¼1351.02
nnn
Country FE vs. pooled Yes
nnn
Yes
nnn
Yes
nnn
Yes
nnn
Time effects vs. pooled Yes
nnn
Yes
nn
Yes Yes
Hausman test χ
2
(10)¼60.13 χ
2
(10)¼265.7 χ
2
(10)¼53.25 χ
2
(11) ¼33.43
Prob4χ
2
¼0.0000 Prob4χ
2
¼0.0000 Prob4χ
2
¼0.0004 Prob 4χ
2
¼0.0004
Note: IRRs of less than 1 reveal a negative impact of the corresponding determinant on bilateral FDI flows, whereas ratios
greater than 1 reveal a positive impact.
nnn
Significant at 1% risk.
nn
Significant at 5% risk.
n
Significant at 10% risk.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9176
is higher when the source country is in the downward phase of its cycle and the host country is in its
upward phase.
24
The assumption that FDI is procyclical for the source and host countries is thus
supported by our estimations: not only does bilateral FDI increase when business cycles are not
synchronized but also it tends to depress investment in the source country when it is in a low
conjuncture while stimulating investment in booming host countries. In column 2.4, however, the
interaction between source cycle and source instability is not significantly different from 0, therefore
indicating that the effect of source instability is neither magnified nor minored by economic down-
turns in the source country. These two mechanisms actually independently affect FDI flows in our
sample of countries.
To conclude, FDI tends to increase when a source country’s demand instability increases and when
source and host countries’GDP cycles are de-synchronized. Therefore, not only do MNCs delay
investment when demand volatility increases in domestic economy but also they substitute pro-
duction abroad to production home, most particularly by choosing destinations where GDP per capita
growth is more stable and follows a de-synchronized cycle. This result therefore confirms the
assumption that in our sample, GDP instability is driven by demand shocks.
4.2. Does trade integration matter? Regional economic integration, source volatility, and FDI
Now that our model with instabilities has been estimated and a positive impact on FDI of source
country instability has been highlighted for our sample of countries, several questions arise, all
connected with the intermediary role of trade integration. Whole sample results could hide the fact
that European and MENA firms may react differently to host country instability in terms of their
foreign investment decisions. Such a distinct behavior could be because of differences in the European
and MENA economies’exposure to foreign trade and to trade-led macroeconomic instability. More
open economies often suffer from a higher level of macroeconomic instability, and firms from more
open economies may also be more internationalized. As a result, the positive FDI effect of source
country GDP instability observed for the whole sample may well reflect the fact that trade openness
of the source country, which is correlated to GDP instability, has an influence on its FDI outflows.
Furthermore, trade integration via RTAs may well, under these conditions, intensify the positive effect
of source country instability on FDI by increasing trading opportunities and reducing transaction and
fiscal costs of investing abroad and trading from abroad. Likewise, bilateral investment treaties (BITs)
may increase FDI by reducing transaction and relocation costs.
Variables accounting for South–South (here GAFTA and AMU) and North–South (here MED)
regional trade agreements have been included in the gravity equation. Although FDI flows to MENA
countries have been increased by the GAFTA, they were influenced neither by Euro-Mediterranean
trade agreements nor by the AMU agreement. However, having concluded a BIT has a positive impact
on FDI flows between the signatory countries.
An additional test consists of checking if RTAs and BITs have amplified the positive relationship between
source volatility and FDI. It was argued that deeper trade integration between source and host countries, via
BITs and FTAs, may amplify the positive effect of source country volatility on FDI outflows by reducing the
costs of reallocating production abroad and re-exporting from abroad (Aizenman and Marion, 2004; De
Mello-Sampayo et al., 2010). It can therefore be expected that, by easing production reallocation abroad in
the case of higher home uncertainty, trade and investment integration will increase FDI to the more stable
host economies of the trading zone. In order to assess the extent to which North–South (here the Euro-
Mediterranean trade agreement MED)andSouth–South RTAs (here GAFTA and AMU) have increased or not
the sensitivity of FDI to host country macroeconomic conditions, we have successively estimated the gravity
model with source and host volatilities and each one of the following four multiplicative terms: source
instability*MED, source instability*AMU, source instability*GAFTA, and source instability*BITs.Resultshavebeen
reported in Table 3.
24
As we intended to filter out the source instability impact for the de-synchronization episodes, not to test the interactive
term, we did not introduce all the components of the interaction terms alone. This is why de-synchro is not included in
Estimation 2.3.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 77
When a continuous variable Xis interacted with a dummy Z, the coefficient of the interactive term
measures the extent to which the dependent variable Yincreases with Xwhen condition Zis met
(Brambor et al., 2006: 65). In our specific case, the interaction term’s coefficient must therefore be
interpreted as the impact of a variable (here source instability) on the dependent variable (FDI
bilateral flow) when the population is limited to the individual observations for which the condition Z
(here participation to a bilateral trade or investment agreement) is met. Column 3.1 in Table 3 first
shows that neither the significance nor the sign of the source instability’s individual FDI impact are
removed when the four interactive terms are simultaneously introduced in the regression. If we now
turn our attention to the interactions, we can check that the coefficient of the source instability*MED
interaction is not significant (column 3.2) whereas, on the contrary, source instability*UMA and source
instability*GAFTA both show a lower-than 1 estimated coefficient indicating a negative impact (col-
umns 3.3 and 3.4). Similar to the case of MED, the coefficient of the source instability*BITs interaction is
not significant (column 3.5). This suggests that whereas the positive FDI impact of source instability is
not conditioned by the existence of North–South trade integration (MED) and bilateral investment
treaties (BITs), it is reduced in the case of South–South trade agreements (notably GAFTA). Indeed,
Table 3
Poisson Pseudo-Maximum Likelihood model with country and time fixed effects: Regional trade agreements and bilateral
investment agreements.
Dependent variable FDI
levels
3.1 3.2 3.3 3.4 3.5
IRR IRR IRR IRR IRR
Source GDP 1.481 (9.82)
nnn
1.472 (9.39)
nnn
1.481 (9.76)
nnn
1.497 (14.10)
nnn
1.474 (9.66)
nnn
Host GDP 1.213 (3.78)
nnn
1.188 (3.34)
nnn
1.192 (3.42)
nnn
1.203 (3.60)
nnn
1.187 (3.33)
nnn
Distance 0.673 (&1.90)
nn
0.668
(&1.90)
nn
0.662 (&1.95)
nn
0.654 (&2.01)
nn
0.667 (&1.91)
nn
Adjacency 0.735 ( &1.37) 0.729 (&1.42) 0.727 (&1.43) 0.729 (&1.42) 0.729 (&1.42)
Past colonial links 1.278 (0.48) 1.245 (0.43) 1.243 (0.43) 1.212 (0.38) 1.240 (0.42)
Common language 1.058 (0.22) 1.055 (0.21) 1.054 (0.21) 1.083 (0.31) 1.057 (0.22)
Source instability 1.040 (3.02)
nnn
1.043 (3.26)
nnn
1.043 (3.25)
nnn
1.042 (3.22)
nnn
1.057 (3.41)
nnn
Host instability 0.589 (&2.69)
nnn
0.582
(&2.69)
nnn
0.580 (&2.77)
nnn
0.584 (&2.73)
nnn
0.5691 (&2.87)
nnn
Host investment profile 1.085 (9.38)
nnn
1.079 (8.59)
nnn
1.079 (8.61)
nnn
1.077 (8.40)
nnn
1.078 (8.60)
nnn
MED 1.034 (1.05) 1.043 (0.87) 1.028 (0.85) 1.017 (0.53) 1.035 (1.07)
AMU 1.879 (1.39) 1.893 (1.41) 2.122 (1.64) 1.944 (1.47) 1.896 (1.41)
GAFTA 1.239 (2.97)
nnn
1.202 (2.55)
nnn
1.238 (2.89)
nnn
1.577 (4.91)
nnn
1.202 (2.54)
nnn
BITs 1.153 (3.69)
nnn
1.153 (3.69)
nnn
1.151 (3.65)
nnn
1.153 (3.69)
nnn
1.160 (3.83)
nnn
Source instability
n
MED –0.940 (&0.22) –––
Source instability
n
AMU ––0.471 (&1.86)
n
––
Sourceinstability
n
GAFTA –––0.188 (&4.51)
nnn
–
Source instability
n
BITs ––––0.965 ( &1.32)
Constant 0.0008 (4.07)
nnn
0.001
(&3.72)
nnn
0.0013 (&3.72)
nnn
0.0009
(&3.97)
nnn
0.0014 (&3.73)
nnn
Log-Likelihood &5277.964 &5268.7747 &5267.0546 &5258.4314 &5267.921
Log-Likelihood ratio test χ
2
¼2572.83
nnn
χ
2
¼2561.87
nnn
χ
2
¼2567.23
nnn
χ
2
¼2566.66
nnn
χ
2
¼2565.36
nnn
Wald test χ
2
¼1076.08
nnn
χ
2
¼1084.39
nnn
χ
2
¼1085.76
nnn
χ
2
¼1103.70
nnn
χ
2
¼1086.22
nnn
Country FE vs. pooled Yes
nnn
Yes
nnn
Yes
nnn
Yes
nnn
Yes
nnn
Time effects vs. pooled Yes
n
Yes
nn
Yes
nn
Yes
nn
Yes
nn
Hausman test χ
2
(13)¼41.70 χ
2
(13)¼22.25 χ
2
(14)¼48.95 χ
2
(14)¼36.39 χ
2
(14)¼73.73
Prob4χ
2
¼0.0000 Prob4χ
2
¼0.07 Prob4χ
2
¼0.0000 Prob4χ
2
¼0.0000 Prob4χ
2
¼0.0000
Note: IRRs of less than 1 reveal a negative impact of the corresponding determinant on bilateral FDI flows, whereas ratios
greater than 1 reveal a positive impact.
n
Significant at 10% risk.
nn
Significant at 5% risk.
nnn
Significant at 1% risk.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9178
when model 3.1 is estimated only for the subsample of GAFTA countries, the value of the coefficient of
source instability is negative (IRR is lower than 1), indicating that, in the case of South–South trade
integration, macroeconomic instability in source countries even has a negative impact on FDI flows.
25
One possible explanation is that MNCs from MENA countries are more likely to be financially con-
strained than those from European economies during periods of high demand instability in their
home market. The former are consequently more likely to reduce or delay their operations abroad
because they are financially constraint.
This section’s results thus suggest that only South–South intraregional trade agreements seem to
have had an impact on FDI responsiveness to source instability. For the countries linked by GAFTA
agreements, the FDI impact of the source country’s demand instability becomes negative. As
explained in Section 2, however, the theory predicts that vertical and horizontal investment might
react differently to increased uncertainty in source countries. Our main result that FDI flows from
highly volatile to weakly volatile economies must therefore be further investigated in a gravity model
that is fitted to explain vertical investment.
4.3. Regional integration, source volatility, and vertical–horizontal FDI
The model described by Eq. (2) is specifically fitted to analyze horizontal FDI because it includes
the size of the host economy as an explaining factor (Kleinert and Toubal, 2010). It can nevertheless be
adapted easily to account for both horizontal and vertical FDI by adding an indicator of the relative
factor endowment and the joint size of home and host country (the sum of their GDP), imposing no
restrictions on the individual country sizes (Kleinert and Toubal, 2010: 8).
26
According to the
knowledge-capital model, skill differences between the labor force in the source and the host
countries would be the ideal indicator to identify drivers of vertical investment (Carr et al., 2001).
Because relevant data are missing for developing host countries, the difference in GDP per capita
between the two countries is generally used as a proxy for the differences in factor endowments or in
the level of economic and technological development in each country (Busse et al., 2010). The dif-
ference in GDP per capita’s coefficient will take a positive sign if FDI is attracted by low labor costs.
As shown by Kleinert and Toubal (2010), an additional determinant of vertical FDI, the sum of
source and host countries’GDP (market size), can be derived from the theoretical knowledge-capital
model of FDI. It is therefore expected that in the case of vertical investment, the GDP per capita
difference as well as the size of the demand by the two countries will take a positive coefficient
(Kleinert and Toubal, 2010).
27
Moreover, the estimated coefficient on adjacency also informs about
the nature of FDI: whereas a positive sign might be reflecting the fact that FDI may be predominantly
of the vertical type since proximity eases investment for re-exporting, a negative sign would suggest
FDI to be predominantly of the horizontal type, since low transport costs should render exporting
more advantageous than FDI (De Mello-Sampayo, 2009).
Because vertical FDI is essentially conditioned by productivity and labor cost differences, it can be
expected that it will be affected only partially by the source country’s volatility outcomes. Vertical FDI
may therefore feature a lower sensibility to domestic demand volatility than horizontal FDI.
Accordingly, the coefficients of the vertical FDI’s two main drivers, that is, the difference in GDP per
capita and market size, may be only marginally affected by an increase of source country’s GDP
volatility. On the contrary, the coefficient of the horizontal FDI’s main driver (host GDP) may increase
with volatility in the source country.
25
Estimations made for UMA subsamples give similar results.
26
Various empirical specifications of the gravity model can be used to explain either horizontal or vertical FDI, with all of
them being based on sound theoretical foundations (Anderson, 2011). Kleinert and Toubal (2010), for example, have derived
two different empirical specifications of the FDI gravity model. The first one is derived from a proximity-concentration model
explaining horizontal investment whereas the second one is derived from a factor-proportions model explaining vertical
investment.
27
Busse et al. (2010) also take into account the fact that the host country’s openness to trade may induce vertical FDI,
arguing that closed economies are hardly attractive to vertical FDI, which involves fragmented production patterns and
international trade in intermediates. We have tested this variable, but because it is never significant and may be correlated to
RTAs, it is not included in our preferred specification.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 79
Table 4 shows the results of the estimations of the gravity model with volatilities, the determi-
nants of vertical FDI, and the interactions of source volatility with the determinants of vertical and
horizontal investment.
First, the two determinants of vertical FDI, the difference in GDP per capita and market size, have
been successively included in Eq. (2), with the results being reported in columns 4.1 and 4.2.
The positive signs of these two variables’coefficients indicate that, for the countries in our sample,
the technological distance and the aggregate market size between the two countries have a positive
impact on FDI between these countries.
28
Our sample’s bilateral FDI is thus partially vertical.
We have then introduced the interactions in the estimations, with the results being reported in
columns 4.3–4.5. When two continuous variables Xand Zinteract, the coefficient of the interactive
Table 4
Poisson Pseudo-Maximum Likelihood model with country and time fixed effects: Source and host instabilities and types of FDI.
Dependent variable FDI
levels
4.1 4.2 4.3 4.4 4.5
IRR IRR IRR IRR IRR
Source GDP 0.836 (3.21)
nnn
1.344 (6.51)
nnn
0.831 (3.23)
nnn
1.328 (6.15)
nnn
1.636 (13.21)
nnn
Host GDP 0.950 (3.66)
nnn
1.186 (3.27)
nnn
0.949 (3.79)
nnn
1.1913 (3.34)
nnn
1.218 (4.47)
nnn
Distance 0.722 (&1.37) 0.720 (&1.50) 0.723 (&1.37) 0.724 ( &1.47) 0.645 (&1.93)
nn
Adjacency 727 (&1.32) 735 ( &1.37) 0.727 (&1.32) 0.822 (&0.87) 0.730 (&1.17)
Past colonial links 1.152 (0.26) 0.813 (0.92) 1.153 (0.26) 1.378 (0.63) 0.911 (&0.17)
Common language 0.969 (0.11) 1.152 (0.54) 0.968 (0.12) 1.164 (0.57) 1.803 (2.15)
nn
Source instability 1.035 (2.67)
nnn
1.048 (3.62)
nnn
1.165 (0.27) 1.122 (3.03)
nnn
1.091 (&0.55)
Host instability 0.551 (&3.03)
nn
0.567 (&2.88)
nnn
0.705 (&1.74) 0.555 ( &2.99)
n
0.470 (&4.62)
nnn
Host investment profile 1.084 (9.04)
nnn
1.079 (8.63)
nnn
1.084
(&9.04)
nnn
1.079 (8.61)
nnn
1.090 (10.17)
nnn
MED 1.042 (1.28) 1.019 (0.60) 1.043 (1.28) 1.018 (0.50) –
AMU 2.104 (1.48) 1.947 (1.43) 2.106 (1.48) 1.953 (1.43) –
GAFTA 1.216 (2.69)
nnn
1.228 (2.83)
nnn
1.216 (2.69)
nnn
1.228 (2.83)
nnn
–
BITs 1.151 (3.63)
nnn
1.153 (3.70)
nnn
1.150 (3.69)
nn
1.153 (3.68)
nnn
–
Market size 2.213 (4.01)
nnn
–2.230 (3.98)
nnn
––
GDP per capita
difference
–1.159 (5.16)
nnn
–1.180 (5.40)
nnn
1.157 (5.14)
nnn
Market size
n
source
instability
–&0.994 (&0.21) ––
GDPpcDiff.
n
source
instability
–&&0.979 ( &1.87)
n
&
Host GDP
n
instability ––––0.997 ( &0.29)
Constant 0.0004 (&4.3)
nnn
0.003( &3.17)
nnn
0.0003
(&4.03)
nnn
0.0033
(&3.13)
nnn
0.0001
(&4.49)
nnn
Log-Likelihood &5260.51 &5254.06 &5262.27 &5252.43 &5960.41
Log-Likelihood ratio test χ
2
¼2467.52
nnn
χ
2
¼2507.47
nnn
χ
2
¼2451.75
nnn
χ
2
¼2469.95
nnn
χ
2
¼4687.25
nnn
Wald test χ
2
¼1093.59
nnn
χ
2
¼1102.55
nnn
χ
2
¼1093.62
nnn
χ
2
¼1103.29
nnn
χ
2
¼1311.85
nnn
Country FE vs. pooled Yes
nnn
Yes
nnn
Yes
nnn
Yes
nnn
Yes
nnn
Time effects vs. pooled Yes
nn
Yes
nn
Yes
nnn
Yes
nn
Yes
Hausman test χ
2
(14)¼25.48 χ
2
(14)¼33.38 χ
2
(15)¼30.48 χ
2
(15)¼32.75 χ
2
(10)¼70.32
Prob4χ
2
¼0.0000 Prob 4χ
2
¼0.0025 Prob4χ
2
¼0.01 Prob4χ
2
¼0.005 Prob4χ
2
¼0.0000
Note: IRRs of less than 1 reveal a negative impact of the corresponding determinant on bilateral FDI flows, whereas ratios
greater than 1 reveal a positive impact.
n
Significant at 10% risk.
nn
Significant at 5% risk.
nnn
Significant at 1% risk.
28
The non-significant estimated coefficient on adjacency does not confirm this result, however. This may signal contra-
dictory patterns of vertical and horizontal FDI over the whole sample, as in De Mello-Sampayo (2009).
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9180
term measures the extent to which the variable Yreacts with Xfixed at an average level but when
variable Zvaries. In column 4.4, the coefficient of the interaction of GDP per capita difference with
source instability takes a significant and negative sign, albeit at 10%, indicating that the positive FDI
impact of a given level of source instability is reduced when the cost differential increases. This
indicates that when the investment toward MENA economies is driven by strong cost differentials, as
is the case for vertical FDI, it is less sensible to source instability. As for the second determinant of
vertical investment, that is, the aggregate market size, it also has a positive impact on the FDI flows
between two countries (column 4.1), although its interaction with source instability is not statistically
significant (column 4.3). Likewise, column 4.5 shows that the positive impact of source volatility on
FDI flows is not significantly modified by an increase in host GDP, therefore confirming that FDI is
mostly vertical in the overall sample. This would mean that if a source country’s instability certainly
matters to explain bilateral FDI flows in our sample, its impact may be less important when invest-
ments are driven by cost differentials, that is, for vertical investment.
5. Robustness checks
In this section, we address additional issues relating to the robustness of the results discussed
in Section 4.
5.1. Robustness 1: Alternative sources of uncertainty
In order to test whether the positive effect of source countries’GDP instability on FDI is not an
artifact produced by the correlation between our sample countries outward FDI but a more global
trend of increased FDI, an indicator of global waves of FDI
29
and an indicator of European waves of FDI
also have been successively introduced into the estimation of the gravity model with uncertainties.
The results, reported in Table 5 columns 5.1 and 5.2, indicate that these two controls have a non-
significant impact on FDI flows and their addition to the estimated model leaves the coefficient for
source country instability unchanged.
Because there is a risk that source country macroeconomic instability may be correlated to a global
or at least regional trend, we have to check whether the effects estimated for our overall sample hold
when the perimeter of external instability is extended to the source country’s region or to the world
economy. Two alternative measurements of global and regional macroeconomic trends have been
introduced successively as additional controls in the complete gravity model with instabilities,
regional agreements, and vertical–horizontal FDI determinants: (1) the lagged three-year averaged
world GDP growth and (2) the lagged three-year averaged standard deviation of world GDP growth.
Results reported in columns 5.3 and 5.4 show that, as expected, the estimated coefficient of the first
two variables takes a positive and significant value, whereas that of the last two variables is negative
and significant. It means that although world GDP growth has a positive impact on FDI flows, growth
instability has a negative impact. Moreover, the inclusion of these four variables in the standard
gravity model does not change the values of the source and host instability estimated incidence ratios.
These results therefore suggest that the positive FDI effect is not driven by a global trend of
macroeconomic instability.
5.2. Robustness 2: Alternative sources of uncertainty in the host country
It could be objected that our main results are driven by our measure of GDP instability. The more
volatile domestic markets (as measured by the standard deviation) may also be the most dynamic
ones (as measured by average GDP growth). The MNCs operating in these more open markets may
therefore invest more abroad than those operating in less open economies because their revenues are
29
Similarly to Méon and Sekkat (2012), the world FDI wave indicator consists of the annual value of world FDI outflows;
similarly, the European FDI wave is computed as the annual value of European FDI outflows.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 81
higher and not because they want to trade home instability against foreign stability. In order to rule
out this possible source of misinterpretation, a GDP growth coefficient of variation has been used as a
replacement for the standard deviation. Column 6.1 of Table 6 shows that the coefficient of source
instability remains positive even when the size effect is controlled for by measuring GDP instability by
a coefficient of variation instead of a standard deviation. We can therefore rule out the argument that
the positive impact of the standard deviation of GDP growth is driven by the fact that the most
volatile countries are also those where the growth rates of aggregate income and corporate revenues
are the highest.
Table 5
Poisson Pseudo-Maximum Likelihood model with country and time fixed effects: Instabilities, RTAs, and BITs and global
instability.
Dependent variable
FDI levels
7.1 7.2 7.3 7.4
IRR IRR IRR IRR
Global FDI waves European FDI
waves
Three-year world GDP
growth
Three-year world GDP standard
deviation
Source GDP 1.333 (6.26)
nnn
1.339 (6.38)
nnn
1.337 (6.38)
nnn
1.386 (7.05)
nnn
Host GDP 1.140 (2.13)
nn
1.169 (2.76)
nnn
1.225 (3.78)
nnn
1.204 (3.59)
nnn
Distance 0.733 (&1.41) 0.725 ( &1.47) 0.709 (&1.59) 0.708 ( &1.60)
Adjacency 0.812 (&0.93) 0.813 (&0.98) 0.813 (&0.91) 0.819 (&0.89)
Past colonial links 1.371 (0.66) 1.366 (0.61) 1.381 (0.64) 1.283 (0.49)
Common language 1.111 (0.40) 1.136 (0.48) 1.157 (0.55) 1.264 (0.88)
Source instability 1.047 (3.55)
nnn
1.048 (3.58)
nnn
1.047 (3.55)
nnn
1.051 (3.80)
nnn
Host instability 1.879 (3.05)
nnn
0.583 (&2.68)
nnn
0.522 (&3.27)
nnn
0.662 (&2.00)
nnn
Host investment
profile
1.073 (7.78)
nnn
1.076 (7.48)
nnn
1.075 (8.20)
nnn
1.074 (8.00)
nnn
MED 1.007 (1.22) 1.013 (0.39) 1.006 (1.20) 1.037 (1.10)
AMU 1.940 (1.43) 1.942 (1.43) 1.912 (1.39) 1.998 (1.50)
GAFTA 1.212 (2.62)
nnn
1.220 (2.71)
nnn
1.213 (2.66)
nnn
1.230 (2.85)
nnn
BITs 1.155 (3.72)
nnn
1.154 (3.71)
nnn
1.148 (3.55)
nnn
1.154 (3.70)
nnn
Market size 2.248 (4.49)
nnn
2.471 (4.67)
nnn
2.272 (4.02)
nnn
2.625 (4.76)
nnn
GDP per capita
difference
1.1560 (5.01)
nnn
1.157 (5.10)
nnn
1.150 (4.86)
nnn
1.170 (5.44)
nnn
World FDI waves 1.035(1.19) –– –
Europe FDI waves –1.015 (0.62) ––
Three-year world GDP
growth
––1.020(2.91)
nnn
–
Three -year world GDP
SD
––– 0.999 (&2.49)
nnn
Constant 0.0026
(&3.26)
nnn
0.002 (&3.21)
nnn
0.0022 (&3.34)
nnn
.001 (&3.58)
nnn
Log-Likelihood &5253.3654 &5253.8557 &5249.7904 &5250.9783
Log-Likelihood ratio
test
χ
2
¼2503.74
nnn
χ
2
¼2502.15
nnn
χ
2
¼2510.05
nnn
χ
2
¼2488.99
nnn
Wald test χ
2
¼1104.54
nnn
χ
2
¼1103.54
nnn
χ
2
¼1107.69
nnn
χ
2
¼1106.67
nnn
Country FE vs. pooled Yes
nnn
Yes
nnn
Yes
nnn
Yes
nnn
Time effects vs. pooled Yes
nn
Yes
nn
Yes
nn
Yes
nn
Hausman test χ2 (16)¼29.86 χ
2
(16)¼28.09 χ
2
(16)¼26.00 χ
2
(16)¼17.17
Prob4χ
2
¼0.0142 Prob4χ
2
¼0.0138 Prob4χ
2
¼0.025 Prob4χ
2
¼0.24
Note: IRRs of less than 1 reveal a negative impact of the corresponding determinant on bilateral FDI flows, whereas ratios
greater than 1 reveal a positive impact.
n
Significant at 10% risk.
nn
Significant at 5% risk.
nnn
Significant at 1% risk.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9182
Nominal instability in the host country may also affect the level of FDI inflows. Because inflation
increases uncertainty about the future value of liabilities and assets acquired by the MNC, it should
adversely influence FDI inflows because incoming investments that could have long-term higher
returns generally are not implemented. The existing empirical evidence for this adverse effect is
mixed: whereas more inflation does not appear to be a significant determinant of FDI inflows for
Frenkel et al. (2004), it does significantly reduce incoming investment for Garibaldi et al. (2001) and
Tapsoba (2012). Column 6.2 shows that although a higher inflation rate in the host countries reduces
FDI inflows, its inclusion leaves the coefficients unchanged for source and host instabilities.
As for exchange rate instability, its FDI impact is also ambiguous, theoretically and empirically.
30
Because we are interested mainly in the effect of nominal instabilities on FDI levels, we have chosen
to test the impact of extreme forms of exchange rate volatility such as exchange rate crises. According
to Kaminsky et al. (1998), an exchange rate crisis is typically a situation in which a speculative attack
Table 6
Poisson Pseudo-Maximum Likelihood model with country and time fixed effects: Instabilities, RTAs, and BITs and other sources
of macroeconomic instabilities.
Dependent variable FDI levels 8.1 8.2 8.3
IRR IRR IRR
Coefficient variation Inflation Exchange rate crisis
Source GDP 1.740 (2.72)
nnn
1.788 (3.51)
nnn
1.742 (2.92)
nnn
Host GDP 1.944 (2.22)
nn
1.938 (2.82)
nnn
1.944 (2.57)
nnn
Distance 0.776 (&1.08) 0.783 (&1.07) 0.773 (&1.05)
Adjacency 0.819 (&0.82) 0.817 (&0.84) 0.813 ( &0.87)
Past colonial links 1.287 (0.46) 1.244 (0.40) 1.279 (0.44)
Common language 1.052 (1.18) 1.050 (0.18) 1.047 (0.16)
Source instability (SE) –1.040 (2.98)
nnn
1.042 (3.11)
nnn
Source instability (CV) 1.879 (3.05)
nnn
––
Host instability 0.542 (&3.11)
nnn
.530 (&3.22)
nnn
0.498 (&3.51)
nnn
Host investment profile 1.083 (9.00)
nnn
1.075 (7.66)
nnn
1.080 (8.58)
nnn
MED 2.218 (1.54) 1.029 (0.97) 1.034 (1.02)
AMU 1.927 (1.22) 2.225 (1.57) 2.208 (1.52)
GAFTA 1.240 (2.96)
nnn
1.238 (2.93)
nnn
1.246 (3.02)
nnn
BITs 1.152 (3.66)
nnn
1.141 (3.40)
nnn
1.150 (3.60)
nnn
Market size 2.288 (4.09)
nnn
2.151 (3.76)
nnn
2.289 (4.08)
nnn
GDPpc. differential 1.161 (5.21)
nnn
1.163 (5.29)
nnn
1.151 (4.91)
nnn
Crude oil price –––
Commodity price index –––
Host inflation –0.964 (&2.34)
nnn
–
Host exchange rate instability 0.0008 ( &3.48)
nnn
–0.938 (&2.89)
nnn
Constant 0.001 (&3.46)
nnn
0.0009 ( &3.46)
nnn
Log-Likelihood &5245.3218 &5242.739 &5241.2839
Log-Likelihood ratio test χ
2
¼2425.12
nnn
χ
2
¼2429.58
nnn
χ
2
¼2430.57
nnn
Wald test χ
2
¼1116.19
nnn
χ
2
¼1119.59
nnn
χ
2
¼1119.28
nnn
Country FE vs. pooled Yes
nnn
Yes
nnn
Yes
nnn
Time effects vs. pooled Yes
nn
Yes
nn
Yes
nn
Hausman test χ
2
(16)¼29.48 χ
2
(16)¼52.99 χ
2
(16)¼27.39
Prob4χ
2
¼0.0142 Prob4χ
2
¼0.0000 Prob4χ
2
¼0.0374
Note: IRRs of less than 1 reveal a negative impact of the corresponding determinant on bilateral FDI flows, whereas ratios
greater than 1 reveal a positive impact.
n
Significant at 10% risk.
nn
Significant at 5% risk.
nnn
Significant at 1% risk.
30
Because exchange rate volatility increases the risks related to export to and from developing countries, it tends to
depress vertical FDI and stimulate horizontal investments (Aizenman and Marion, 2004). Takagi and Shi (2011) have shown
that Japanese FDI to the region, mostly vertical, was positively affected by exchange rate volatility from 1987 to 2008.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 83
leads to a sharp depreciation of the local currency and to considerable losses in exchange reserves,
especially in the case of fixed or pegged regimes. In line with Kaminsky et al. (1998) and Ahluwalia
(2000), we have implemented for each host country an ex post identification of the periods during
which they were affected by such a crisis between 1985 and 2009.
31
To this end, an index combining
variations of the nominal exchange rates and variations in the foreign exchange reserves has been
computed (see Appendix 5 for details). Results of estimations are reported in columns 6.3 and show
that more exchange rate instability reduces FDI inflows to MENA, without modifying the results for
host and source real instabilities.
Our results concerning nominal instability therefore support the assumption that Europe-MENA
foreign investment is mainly vertical because it tends to decrease with inflation and exchange rate
instability. Although they reduce the incidence of FDI flows, nonetheless, nominal sources of host
country instability (inflation and exchange rate instability) modify neither the magnitude of IRR nor
the significance of host country instability.
6. Conclusion
Using a gravity model, this article tests the assumption that FDI is reactive to macroeconomic
instability in source and host countries for a sample of European and MENA countries for the period
from 1985 to 2009. The gravity model enables identifying the FDI impact of macroeconomic risks in
both source and host countries while controlling for the other sources of risks and costs associated
with distance (geographical, linguistic, and legal). Finally we could show that the incidence of FDI
between two countries increases with source GDP instability and with host GDP stability. Both source
country’s instability and source host countries’cycle de-synchronization tend to increase FDI to
MENA, with the less instable MENA countries receiving more FDI. Therefore, not only do MNCs delay
investment when volatility increases in domestic economy but also they substitute production abroad
to production home, most particularly by choosing the destinations where GDP per capita growth is
more stable and follows a de-synchronized cycle. Our findings therefore indicate that FDI may con-
stitute a genuine and valuable option for firms undergoing strong instability or downward con-
juncture in their home market. Delaying their domestic investment is not the unique option open to
them. The fact that FDI tends to flow from the most to the least instable macro-economies can
indicate that both northern and southern firms trade between investing home and abroad through a
function of macroeconomic features and not only for costs or market size considerations.
Moreover, we have also found that this reactivity is conditioned by (1) trade and investment agree-
ments and (2) the type of FDI (vertical or horizontal). First, regional trade and investment agreements had
an impact on these patterns during the period under study, but this effect is nonetheless confined to
South–South RTAs. South–South trade integration (GAFTA), as well as bilateral investment treaties, directly
increased FDI flows to MENA. For our four MENA host economies, however, the GAFTA agreement has
significantly reduced the FDI responsiveness to the source country’sinstabilities.MNCsfromMENA
countries are more likely to be financially constrained than those from European economies during
periods of high demand instability in their home market and consequently reduce or delay their opera-
tions abroad. By contrast, because of their implementation in the 1990s, the North–South Euro-
Mediterranean agreements have neither spurred FDI flows to the MENA economies nor altered the FDI
responsiveness to European countries’macroeconomic conditions.
Additionally, we find some evidence, albeit weak, that the sensibility of FDI to uncertainty in the source
country decreases with the technological distance between the source and host countries. Put differently,
if considering source country’sinstabilitycertainlyhelpsexplainoverallbilateralFDIflows in our sample,
its impact may be less important when investments are driven by cost differentials, that is, for vertical
investment. These results could indicate that by reducing the costs of investment in the least costly MENA
31
We do not use the evolution of the exchange rate as an indicator for external stability: that would be somewhat
meaningless when only annual data are used for estimations. Nor do we introduce exchange rate regimes, as in Frenkel et al.
(2004), because that would not enable the effect of external shocks on FDI inflows to be grasped.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9184
economies, regional integration has reduced FDI reactiveness to source macroeconomic conditions and
strengthened its dependence on the standard drivers of vertical investment. A more fine-grained analysis
of the complex articulation among instabilities, trade integration, and the types of FDI would nevertheless
be required at a next stage. Addressing this articulation would certainly necessitate using more dis-
aggregated data on vertical and horizontal FDI, which are, unfortunately, not available.
Our results raise several policy issues. Two decadesofsuddenstopsundergonebyemergingeconomies
have illustrated that attraction policies and a high GDP share of FDI are not sufficient conditions to stabilize
capital inflows (Calvo, 1998). Obviously, developing countries’policies cannot influence source countries’
macroeconomic conditions. Nonetheless, we show in this article that trade and investment policies aimed
at more deeply integrating host and source countries certainly condition the extent to which FDI inflows to
the former react to the uncertainty to which the latter’sMNCsareexposed.Furthermore,wehavefound
evidence of higher levels of North–South investment when the source and host country’scyclesarenot
synchronized, with FDI flows tending to depress private investment in the source country in a bust cycle
while increasing private investment in booming host countries. It follows that, in a world of growing trade
integration, it is even more essential than before for DCstoprudentlyopentheircapitalaccount.Moreover,
even though FDI tends to be less unstable than portfolio investment (Hausmann and Fernández-Arias,
2000; Lipsey, 2001), FDI reactivity to external macroeconomic instability requires the consideration of the
latter as a potential source of shock diffusion from northern to southern economies.
In addition, our results show that economic integration via regional trade or bilateral and
investment agreements does not necessarily improve developing countries’capacity to attract vertical
foreign investment. On the one hand, the reduction of microeconomic transaction costs and the
increase of regional market size prompted by RTAs and BITs have tended to increase levels of FDI to
MENA economies, but only in the case South–South trade integration. North–South trade agreements,
the most likely to bring out vertical investment to MENA countries, have not triggered FDI. Moreover,
our findings show that RTAs do not always smooth the constraints imposed on MNCs by the mac-
roeconomic volatility in their home market and reduce the option price of delaying investment. In our
sample, this is especially true of the FDI flows coming from European sources. Consequently, a major
issue for MENA countries is certainly that they could end up individually competing one with another,
spending high amounts of fiscal resources to attract European firms’vertical investment, as found by
Chenaf-Nicet and Rougier (2009) in the case of Morocco and Tunisia. Cherif and Dreger (2015) have
recently found that agglomeration effects are weaker for the MENA region than for Latin America and
Southeast Asia, therefore confirming that vertical and platform investments, which are the most
likely to agglomerate and generate technological spillovers, are underrepresented in the MENA
region. The social cost of the policies aimed at vertical FDI attraction can therefore turn out to be
considerably high, and their economic efficiency is limited to enclaves, as shown by Piveteau and
Rougier (2011) in the case of Morocco. This is all the more the case as FDI tends to be highly sensitive
to source countries’macroeconomic fluctuations. Further investigations would therefore be required
to understand how the industrial and trade policies of MENA labor-abundant countries have evolved
in reaction to deepening regional integration, notably in order to stabilize vertical FDI inflows and
organize the regional supply chain integration that seems necessary to make these countries less
dependent on short-term demand instabilities in source countries.
Appendix 1: Data sources and definitions
See Table A1.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 85
Appendix 2: List of countries in the sample
See Table A2.
Table A1
Variables Description Data source
FDI The value in thousands of US dollars of flows of foreign direct
investment (FDI) from one country (source country) toward
another country (host country)
OECD, UNCTAD; Balance of pay-
ments of Morocco, Central Bank of
Tunisia
Host GDP GDP in thousands of US dollars CEPII, CHELEM database
Source GDP GDP in thousands of US dollars
GDP per capita
difference
Difference in GDP per capita (thousands of US dollars) between
source and host country
CEPII and IMF International
Financial statistics for population
data
Distance Distance in kilometers between source and host countries’capitals CEPII, Geo dataset
Adjacency Common border between source and host countries (takes the
value 1 if the two countries share a common border and
0 otherwise)
CEPII, Geo dataset
Common
language
Common official language for source and host countries (takes the
value 1 if the two countries share a common language and
0 otherwise)
CEPII, Geo dataset
Common colo-
nial power
Common colonizer for source and host countries (takes the value
1 if the two countries had a common colonizer and 0 otherwise)
CEPII, Geo dataset
Past colonial
links
Dummy variable taking the value 1 if host country was colonized
by source country and 0 otherwise
CEPII, Geo dataset
Investment
profile
Host country’s score for institutional risk to FDI including ratings of
contract viability, risks of expropriation, profit repatriation, and
payment delays. Highest score equates to very low risk.
ICRG database
Source instabil-
ity Host
instability
Three-year standard deviations of GDP growth for host and source
countries
Authors’calculations CEPII, CHE-
LEM database
De-synchro Dummy taking the value 1 (the sum of the source and host dum-
mies was equal to 1) and taking 0 when the cycles were synchro-
nized (the sum of the source and host dummies was equal to 0 or 2)
Authors’calculations
MED, GAFTA,
AMU
Dummy variables taking the value 1 for the country years covered
by these bilateral or multilateral trade agreements and 0 otherwise
Authors’calculations based on
UNCTAD
BITs Dummy variable taking the value 1 for the country years covered
by a bilateral international investment agreement and 0 otherwise
Authors’calculations based on
UNCTAD
Market size Sum of source and host GDPs Authors’calculations CEPII, CHE-
LEM database
GDP per capita
difference
Source GDP per capita minus host GDP per capita Authors’calculations CEPII, CHE-
LEM database
World FDI wave World levels of FDI flows in value UNCTAD
Europe FDI wave European Union (UE25) levels of FDI flows in US dollars UNCTAD
Three-year
world GDP
growth
Three-year moving average of world GDP growth Authors’calculations on the basis
of IMF data
Three-
year world
GDP SD
Three-year moving standard deviation of world GDP growth Authors’calculations on the basis
of IMF data
Commodity
price index
Commodity industrial inputs (including agricultural raw materials
and metals) price index
IMF WEO
Crude oil price Crude oil price IMF WEO
Inflation host Annual rate of inflation in the host country Authors’calculations on the basis
of IMF data
Exchange rate
instability
Index taking the value 1 if the country has experienced a large
variation in the value of the real exchange rate or of the foreign
currencies reserves and 0 otherwise
Authors’calculations on the basis
of IMF data
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9186
Appendix 3: FDI levels: Eichengreen correction and Tobit RE estimations
See Table A3.
Table A2
Algeria Germany Mauritania Sweden
Austria Greece Morocco Switzerland
Belgium–Luxembourg Hungary Netherlands Syria
Czech Republic Ireland Norway Tunisia
Denmark Italy Poland Turkey
Egypt Jordan Portugal United Kingdom
Finland Libya Romania
France Malta Spain
Note: The four MENA host countries are shown in bold.
Table A3
Estimator Eichengreen’s correction RE estimator Tobit RE estimator
A3.1 A3.2 A3.3 A3.4
GDP source 1.608 (16.29)
nnn
1.532 (14.34)
nnn
1.608 (16.45)
nnn
1.522 (13.78)
nnn
GDP host 1.081 (11.09)
nnn
1.072 (9.70)
nnn
1.079 (11.13)
nnn
1.081 (9.58)
nnn
Distance &1.016 (&2.85)
nnn
&0.996 (&2.70)
nnn
&1.016 (&2.89)
nnn
&0.994 (&2.59)
nnn
Adjacency &0.570 (&1.20) &0.679 (&1.39) &0.569 (&1.22) &.676 (&1.34)
Past colonial
links
0.130 (0.13) 0.649 (0.62) 0.130 (0.13) 0.676 (0.62)
Common
language
1.862 (3.93)
nnn
1.720 (3.50)
nnn
1.862 (3.99)
nnn
1.711 (3.34)
nnn
Instability
source
&.237 (2.68)
nnn
–.237 (2.69)
nnn
Instability host –&.634 (&1.04) –&.617 (&1.09)
Constant &37.651 (&11.94)
nnn
&36.047 (&10.57)
nnn
&37.630 ( &12.07)
nnn
&36.041 (&10.24)
nnn
R
2
within 0.20 0.15
R
2
between 0.53 0.53
R
2
total 0.39 0.39
Tests Fisher test: MCO vs. indi-
vidual FE
Fisher test: MCO vs. indi-
vidual FE
Log–
Likelihood¼&8472.271
Log-Likelihood
¼&7087.8369
F(134, 3238)¼21.32 F(134, 2696)¼21.34 Wald χ
2
¼992.20
nnn
Wald χ
2
¼642.25
nnn
Pr4F¼0.000 Pr 4F¼0.000 Prob4χ
2
¼0.0000 Prob 4χ
2
¼0.0000
Fisher test: MCO vs. time:
FE
Fisher test: MCO vs. time:
FE
F(24, 3244)¼3.61 F(20, 2806)¼2.27
Pr4F¼0.000 Pr 4F¼0.000
Hausman test χ
2
¼3.61 Hausman test χ
2
¼22.52
Pr4χ
2
¼0.1604 Pr4χ
2
¼0.040
Wald χ
2
¼990.96
nnn
Wald χ
2
¼651.60
nnn
Breush–Pagan
χ
2
(1)
test
χ
2
¼6654.43 χ
2
¼5727.60
Pr4χ
2
¼0.000 Pr 4χ
2
¼0.000
Note: Number of observations: 3375; number of years: 23; number of country pairs: 27
n
5¼135.
n
Significant at 10% risk.
nn
Significant at 5% risk.
nnn
Significant at 1% risk.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 87
Appendix 4: Stationarity test
See Table A4.
Table A4
Panel A: Levin–Lin–Chu unit-root test.
Variables Adjusted t
n
p-value
Source instability
&9.6729
0.0000
Host instability
&9.9008
0.0000
Host investment profile
&8.6225
0.0000
GDP source
&1.8616
0.0313
GDP host
a
&5.5781
0.0000
GDP per capita difference
&8.0229
0.0000
FDI level
&3.1045
0.0010
Ho: Panels contain unit
roots
Number of panels¼135
Ha: Panels are stationary Number of periods¼25
AR parameter: common
Panel means: included
Time trend: not included
Panel B: Im–Pesaran–Shin unit-root test
Variables Z-t-tilde-bar p-value
Source instability
&3.6954
0.0001
Host instability
&6.1957
0.0000
Host investment profile
a
&5.1270
0.0000
GDP source
&10.7791
0.0000
GDP host
a
&7.5199
0.0000
GDP per capita difference
&10.3557
0.0000
FDI level
a
&10.7478
0.0000
Ho: All panels contain unit
roots
Number of panel¼135
Ha: Some panels are
stationary
Number of periods¼25
AR parameter: panel-specific
Panel means: included
Time trend: not included
a
Time trend included.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9188
Appendix 5: The exchange rate crisis indicator
In line with Kaminsky et al. (1998) and Ahluwalia (2000), we have computed an index of
exchange rate instability. Eq. (A1) shows that the indicator consists of a weighted average of the
variations of the nominal exchange rate and in the exchange reserves. These two variables,
computed as quarterly variations on the basis of monthly average levels, are respectively named
ΔTCN and ΔRES. The weights respectively measure the shares of the variance of the exchange
rate and the foreign exchange reserves in the sum of these variances.
IND ¼
1
σ
2TCN
1
σ
2TCN
þ
1
σ
2RES
#$
0
@1
A"ΔTCNþ
1
σ
2
RES
1
σ
2TCN
þ
1
σ
2RES
#$
0
@1
A"&1ðÞ"ΔRES ðA1Þ
This index, designed to reflect the intensity of the pressures that a national currency undergoes during
an episode of balance of payments crisis, enables the severity of those periods of external instability to be
assessed. It should be noted that a negative sign for the average monthly variation of the foreign exchange
reserves enables obtaining the highest leveloftheindexwhenthecrisisisimminent(Ahluwalia, 2000).
The instability threshold above which whether a country jis affected by a crisis at date tis defined on a
country-by-country basis and not for the whole sample. It is obtained by considering both the average
level (mean
IND
)andthestandarddeviation(σ
IND
)ofIND.AcrisisisdetectedwhenINDissuperiororequal
to mean
IND
þσ
IND
.Thatthresholddefinition corresponds to the minimal bound found in the literature. A
higher threshold (1.5*β
IND
þσ
IND
)wouldhaveresultedinaninsufficient variation of the variable. Subse-
quently, a binary variable exchange rate instability has been created, with this variable taking the value 1 if
INDZmean
IND
þσ
IND
and 0 otherwise. In order to transpose these quarterly crises data into annual
impacts, it is commonly admitted that any episode of crisis duration exceeding a period of 3 months will
have effects on the current year, which means that the crisis could be regarded as annual. In order to avoid
reverse causality with FDI inflows, a crisis in year thas been imputed in estimation as a determinant of FDI
inflows in tþ1(Ahluwalia,1998;Kaminsky et al., 1998).
Eq. (1) shows that the indicator consists of a weighted average of the variations of the nominal exchange
rate and the exchange reserves. These two variables, computed as quarterly variations on the basis of
monthly average levels, are respectively named ΔTCN and ΔRES. The weights respectively measure the
shares of the variance of the exchange rate and the foreign exchange reserves in sum of these variances.
References
Abderrezak A. Foreign direct investment in the MENA region: Is there a colonisation memory effect? J North Afr Stud 2008;13
(2):219–26.
Aghion P, Banerjee A. Volatility and growth. Clarendon lectures in economics. Oxford: Oxford University Press; 2005.
Ahluwalia P. Discriminating contagion: An alternative explanation of contagious currency crises in emerging markets. IMF
working paper WP/00/14. Washington, DC: The International Monetary Fund; 2000.
Aizenman J, Ito H. Trilemna policy convergence patterns and output volatility. North Am J Econ Fin 2012;23(3):269–85.
Aizenman J, Marion N. The merits of horizontal versus vertical FDI in the presence of uncertainty. J Int Econ 2004;62(1):125–48.
Aizenman J. Volatility, employment and the patterns of FDI in emerging markets. J Dev Econ 2003;72(2):585–601.
Alaya M, Chenaf-Nicet D, Rougier E. Une analyse macroéconomique de l’attractivité des PSEM. In: Mezouhagui M, editor.
L’attractivité des pays du Maghreb. Paris: IRMC et Khartala; 2009.
Albuquerque R, Loayza N, Servén L. World market integration through the lens of foreign direct investors. J Int Econ 2005;66
(2):267–95.
Ali FA, Fiess N, MacDonald R. Do institutions matter for foreign direct investment? Open Econ Rev 2010;21(1):201–19.
Anderson JE, Marcouiller D. Insecurity and the pattern of trade: an empirical investigation. Rev Econ Stat 2002;84(2):342–52.
Anderson JE, van Wincoop E. Gravity with gravitas: a solution to the border puzzle. Am Econ Rev 2003;93(1):170–92.
Anderson J. A theoretical foundation for the gravity equation. Am Econ Rev 1979;69(1):106–16.
Anderson JE. The gravity model. Annu Rev Econ 2011;3:133–60.
Andrés MS, Nunnenkamp P, Busse M. What drives FDI from non-traditional sources? A comparative analysis of the determi-
nants of bilateral FDI flows Econ: Open-Access Open-Assess E-J 2013;7(1):1–53.
Arvis J, Shepherd B. The Poisson quasi-maximum likelihood estimator: A solution to the “adding up”problem in gravity models.
Appl Econ Lett 2013;1:37–41.
Berger A, Busse M, Nunnenkamp P, Roy M. Do trade and investment agreements lead to more FDI? Accounting for key pro-
visions inside the black box Int Econ Econ Policy 2013;10(2):247–75.
Bergstrand JH, Egger P. A knowledge-and-physical-capital model of international trade flows, foreign direct investment, and
multinational enterprises. J Int Econ 2007;73(2):278–308.
Bernanke BS. Irreversibility, uncertainty, and cyclical investment. Q J Econ 1983;98(1):85–106.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 89
Bevan AA, Estrin S. The determinants of foreign direct investment into European transition economies. J Comp Econ 2004;32
(4):775–87.
Blanchard O, Simon J. The long and large decline in U.S. output volatility. Brook Pap Econ Act 2001;1:135–74.
Bloningen BA, Wang M. Inappropriate pooling of wealthy and poor countries in empirical FDI studies In: Moran Theodore H,
Graham Edward M, Blomström Magnus, editors. Does foreign direct investment promote development? Washington, DC:
Institute for International Economics; 2005. p. 221–44.
Bloningen BA. A review of the empirical literature on FDI determinants. NBER Working Paper No. 11299. Cambridge, MA:
National Bureau of Economic Research; 2005.
Brambor T, Clark WR, Golder M. Understanding interaction models: improving empirical analyses. Political Anal 2006;14
(1):63–82.
Brandão da Brito JM, de Mello-Sampayo F. The timing and probability of FDI: an application to the United States multinational
enterprises. Appl Econ 2005;37(4):417–37.
Burger M, van Oort FG, Linders GM. On the specification of the gravity model of trade: zeros, excess zeros and zero-inflated
estimation. Spat Econ Anal 2009;4(2):167–90.
Busse M, Königer J, Nunnenkamp P. FDI promotion through bilateral investment treaties: more than a bit? Rev World Econ
2010;146(1):147–77.
Bénassy-Quéré A, Coupet M, Mayer T. Institutional determinants of foreign direct investment. World Econ 2007;30(5):764–82.
Büthe T, Milner HV. The politics of foreign direct investment into developing countries: increasing FDI through international
trade agreements? Am J Polit Sci 2008;52(4):741–62.
Calvo GA. Capital flows and capital market crises: the simple economics of sudden stops. J Appl Econ 1998;1(1):35–54.
Carr DL, Markusen JR, Maskus KE. Estimating the knowledge capital model of the multinational enterprise. Am Econ Rev
2001;91(3):693–708.
Cavallari L, D’Addona S. Nominal and real volatility as determinants of FDI. Appl Econ 2013;45:2603–10.
Caves RE. Multinational Enterprise and Economic Analysis. 2nd ed. New York: Cambridge University Press; 1996.
Cezar R, Escoba OR. Institutional distance and foreign direct investment. Rev World Econ 2015;151(4):713–33.
Chenaf -Nicet D, Rougier E. Output volatility and FDI to Middle East and North African countries: a close-up on the source
countries. Rég Dév 2014;40:139–65.
Chenaf-Nicet D, Rougier E. Concurrence des territoires ou effets de déversement d’attractivité? Nature des relations entre les
IDE marocains et les IDE tunisiens. Rég Dév 2009;29:105–40.
Chenaf-Nicet D, Rougier E. New exports matter: FDI, export discoveries and growth in MENA countries. J Int Trade Econ Dev
2011;20(4):507–33.
Cherif M, Dreger C. The effect of south–south trade integration on FDI. DIW Discussion Papers 1461. Berlin: Deutsches Institut
für Wirtschaftsforschung; 2015.
Daude C, Stein E. The quality of institutions and foreign direct investment. Econ Polit 2007;19(3):317–44.
Daude C, Levy-Yeyati E, Stein E. Regional integration and the location of FDI. Working Paper 492. University of Maryland. Inter-
American Development Bank; 2003.
De Mello-Sampayo F, de Sousa-Vale S, Camões F. Delaying the timing of offshoring low-skilled tasks. Econ Model 2010;27
(5):951–8.
De Mello-Sampayo F. Competing destinations formulation of the gravity model: an application to the geographic distribution of
USA multinational enterprises' activities. Appl Econ 2009;41(17):2237–53.
Deardorff AV. Determinants of bilateral trade: does gravity work in a neoclassical world? In: Frankel Jeffrey A, editor. The
regionalization of the world economy. University of Chicago Press NBER; 1998. p. 7–22.
Desbordes R, Vicard V. Foreign direct investment and bilateral investment treaties: an international political perspective. J
Comp Econ 2009;37(3):372–86.
Dixit A, Pindyck R. Investment under uncertainty. Princeton, NJ: Princeton University Press; 1994.
Dixit A. Entry and exit decisions under uncertainty. J Polit Econ 1989;97:620–38.
Eaton J, Tamura A. Bilateralism and regionalism in Japanese and U.S. trade and direct foreign investment patterns. J Jpn Int Econ
1994;8(4):478–510.
Eichengreen BJ, Irwin DA. The role of history in bilateral trade flows. In: Frankel JA, editor. The regionalization of the world
economy. Cambridge, MA: National Bureau of Economic Research; 1998.
Feenstra R. Advanced international trade: theory and evidence. Princeton, NJ: Princeton University Press; 2004.
Frankel J. Regional trading blocs in the world economic system. Washington, DC: Institute for International Economics; 1997.
Frenkel M, Funke K, Stadtmann G. A panel analysis of bilateral FDI flows to emerging economies. Econ Syst 2004;28(3):281–
300.
Garibaldi P, Mora N, Sahay R, Zettelmeyer J. What moves capital in transition economies? IMF Staff Pap 2001;48(1):109–45.
Gill J. Interpreting interactions and interaction hierarchies in generalized linear models: issues and applications. Paper presented at
the annual meeting of the American Political Science Association, September. San Francisco: University of Florida; 2001.
Gómez-Herrera E. Comparing alternative methods to estimate gravity models of bilateral trade. Empir Econ 2013;44(3):1087–111.
Hausmann R, Fernández-Arias E. Foreign direct investment: good cholesterol? Inter-American Development Bank Working
Paper No. 417 Washington, DC: Inter-American Development Bank; 2000.
Head K, Mayer T. Gravity equations: workhorse, toolkit, and cookbook. CEPII Working Paper no. 2013-27; September 2013.
Head K, Ries J. FDI as an outcome of the market for corporate control: theory and evidence. J Int Econ 2008;74(1):2–20.
Jaumotte F. Foreign trade investment and regional trade agreements: the market size effect revisited. IMF Working Paper 04/
206. Washington, DC: The International Monetary Fund; 2004.
Kaminsky G, Lizondo S, Reinhart C. Leading indicators of currency crises. IMF Staff Pap 1998;5(1):1–48.
Kent C, Smith K, Holloway J. Declining output volatility: what role for structural change? Research Discussion Paper 2005–
2008. Economic Group Reserve Bank of Australia; 2005.
Kleinert J, Toubal F. Gravity for FDI. Rev Int Econ 2010;18(1):1–13.
Lane P, Milesi-Ferretti GM. International investment patterns. Rev Econ Stat 2008;90(3):538–49.
Lensink R, Morrissey O. Foreign direct investment: flows, volatility and the impact on growth. Rev Int Econ 2006;14(3):478–93.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–9190
Levy-Yeyati E, Panizza U, Stein E. The cyclical nature of north–south FDI flows. J Int Money Finance 2007;26(1):104–30.
Lipsey RE. Foreign direct investors in three financial crises. NBER Working Paper No. 8084. Cambridge, MA: National Bureau of
Economic Research; 2001.
Loayza NV, Rancière R, Servén L, Ventura J. Macroeconomic volatility and welfare in developing countries: an introduction.
World Bank Econ Rev 2007;21(3):343–57.
Malik A, Awadallah. B. The economics of the Arab spring. World Dev 2013;45(1):296–313.
Markusen JR, Maskus KE. General-equilibrium approaches to the multinational firm: a review of theory and evidence. NBER
Working paper 8334. Cambridge, MA: The National Bureau of Economic Research; 2001.
Markusen JR. Multinationals, multi-plant economies, and the gains from trade. J Int Econ 1984;16(3–4):205–26.
Markusen JR. Multinational firms and the theory of international trade. Cambridge, MA: MIT Press; 2002.
Marschak J. Role of liquidity under complete and incomplete information. Am Econ Rev 1949;39(3):182–95.
McDonald RL, Siegel D. The value of waiting to invest. Q J Econ 1986;101(4):707–28.
Medvedev D. Beyond trade: The impact of preferential trade agreements on FDI inflows. World Dev 2012;40(1):49–61.
Mina WM. The institutional reforms debate and FDI inflows to the MENA region: the “best”ensemble. World Dev 2012;40
(9):1798–809.
Mátyás L. Proper econometric specification of the gravity model. World Econ 1997;20(3):363–8.
Méon P-G, Sekkat K. FDI waves, waves of neglect of political risk. World Dev 2012;40(11):2194–205.
Pindyck RS. Irreversible investment, capacity choice, and the value of the firm. Am Econ Rev 1988;78(5):969–85.
Piveteau A, Rougier E. Le retour en trompe l’œil de la politique industrielle. L’expert, l’Etat et l’économie politique locale. Rev
Tiers-Monde 2011(208):177–92.
Ramey G, Ramey V. Cross-country evidence on the link between volatility and growth. Am Econ Rev 1995;85:1138–51.
Redding S, Venables T. Economic geography and international inequality. J Int Econ 2004;62(1):53–82.
Sadik A, Bolbol A. Capital flows, FDI and technology spillovers: evidence from Arab countries. World Dev 2001;29(12):2111–25.
Santos Silva J, Tenreyro S. The log of gravity. Rev Econ Stat 2006;88(4):641–58.
Takagi S, Shi Z. Exchange rate movements and foreign direct investment: Japanese investment in Asia, 1987–2008. Jpn World
Econ 2011;23(4):265–72.
Tapsoba R. Does inflation targeting matter for attracting foreign direct investment in developing countries. Working Paper
201203. University of Clermont-Ferrand. CERDI; 2012.
UNCTAD. World investment report. Geneva: United Nations Conference on Trade and Development; 2009.
Wang M, Wong S. Foreign direct investment outflows and business-cycle fluctuations. Rev Int Econ 2007;15(1):146–63.
Wei S-J. How taxing is corruption on international investors? Rev Econ Stat 2000;82(1):1–11.
D. Chenaf-Nicet, E. Rougier / International Economics 145 (2016) 66–91 91