Master Thesis
Does the entry mode of Foreign Direct Investments have an impact on the real exchange rate? An empirical analysis
Author: Linn Rönnlöf
Supervisor: Teodora Borota Milicevic Department of Economics
June 5, 2020
2 Abstract
External finance is an important source of a country’s economic development but can contribute to adverse side-effects. FDI, a type of external financing, can be divided into either M&A or greenfield investment. I argue that, greenfield investment would have a stronger appreciation effect of the real exchange rate than M&A as a result of the Dutch Disease effect, spending and resource movement effect. Using a dynamic panel data technique from a sample of 87 countries during the time period 2003-2016, the paper finds indications that greenfield investment has an appreciation effect of the real exchange rate. Surprisingly it was also found that M&A tends to have a depreciation effect of the real exchange rate. The main implication of these results is that greenfield investment and M&A does not act homogenous in relation to the real exchange rate.
Keywords: FDI, Greenfield investment, M&A, real exchange rate, Generalized Method of moments (GMM)
3
Table of Contents
I. Introduction ... 4
II. Related literature ... 6
III. Theory ... 8
a. M&A and Greenfield investment ... 10
IV. Econometric approach ... 11
V. Data ... 15
a. Stationarity ... 17
VI. Results ... 18
a. Mechanisms – spending and resource movement effect ... 19
b. Unexpected effect by M&A ... 22
VII. Robustness test ... 24
A. Other robust tests ... 26
VIII. Conclusion ... 28
VIIII. Acknowledgement ... 29
Appendix ... 30
Appendix A – Theory of spending and movement effect ... 30
Appendix B - Countries ... 32
Appendix C – Description of variables ... 33
Appendix D – The real exchange rate ... 34
Appendix E – Correlation matrix ... 35
References ... 36
4 I. Introduction
External finance is an important ingredient for a successful process of economic development.
Over the past decades we have observed integration of developing countries and emerging markets into the global economy, a phenomenon associated with the increase of foreign capital inflow. Foreign capital inflow can take many forms, one of them is foreign direct investments (FDI) which have dominated the private capital flows and increased from a level of $500 000 million to approximately $2 000 000 million during the last 16 years1. Foreign capital flows are occasionally called a “mixed blessing”, since it might result in adverse side-effects.
Theoretically, inflow of foreign capital give rise to an appreciation of the real exchange rate of the receiving country’s currency, with negative consequences for the country's export sector, by potentially hurting competitiveness. This phenomenon is usually considered
5
two types of investment is that the latter involves a shift of assets from domestic to foreign hands. However, FDI through greenfield investment represents an enlargement of the host country's capital stock and an increase in employment (UNCTAD, 2000). This makes it evident that the two investment strategies are of different nature and therefore not substitutes for each other. Consequently, there is no reason to believe that these inflows will have the same effect on a country's exchange rate, which is a restriction that earlier empirical research has imposed.
A key feature of an appreciation of the real exchange rate is an increase of the relative price of non-tradable goods (relative to tradable goods). This could be referred to as the Dutch Disease effect, which channels either through the spending effect or the movement effect. The spending effect increases spending as a result of higher income of the exogenous shock generated by a boom of capital inflow. This will cause a rise in aggregate demand resulting in an increase in
the relative price of non-tradable goods, assuming the prices of 43.6(no567.2 ()0rng ) -43.6 ( (nc) 0.2 (re) 0.2 (a) 0.2 (s) -0.2 (e) 0.2( ) -25 18TT680.2 (-) Tj ET Q q 0.24 5 (ri) 06 T.2 (nc) 0.2 (re) 0.2 (a) 0.2 (s) -0.2 (e) 0.2( ) -25 1 9.T6 Tf (-) Tj ET Q q 0.24 5 (ri) 0 Tf [ (nc) 0.2 (re) 0.2 (a) 0.2 (s) -0.2 (e) 0.2( ) -25 (e) 0.2 (-) Tj ET Q q 0.4 0 0 0 0.24 18 621.92cm BT 50 0 0 50 (e) 0.2( ) -25 (i) 05 ) ] TJ ET Q q 0.24T (nt) 0..2 ()0rng idanc550.0519 -266 Tm /TT6 1 Tf [ (s) -0.2 39 -266 Tm /TT61 Tf 4/TT16 1 -5
6
specification, which controls for potential endogeneity using both internal and external instruments.
The remainder of this paper is organized as follow: Section II describes related literature within the field of FDI’s impact on the real exchange rate as well as earlier empirical literature on the mechanisms of the Dutch Disease. Section III describes the theoretical concept of the Dutch Disease and how it is affected by M&A and greenfield investment. Section IV concentrates on the empirical strategy and how to apply it to this research. Sector V will describe the data, addressing potential issues and how to solve them. Section VI reveals the results along with a discussion. Section VII presents robustness checks. Section VIII finish the paper by concluding the findings.
II. Related literature
When considering the impact of capital inflow on the real exchange rate, the earlier empirical literature is important for trying to understand the characteristics of FDI. Recent empirical studies have added focus to the composition of capital inflow and their impact on the real exchange rate. Several of these studies have conclusions in line with the suggested theoretical framework - a surge in capital inflow will appreciate the host country’s currency. Saborowski (2011) studied a panel of 84 developing and developed countries during the years 1995-2006.
He found that all types of private capital inflows appreciated the real exchange rate, though the effect of FDI was weaker compared to other types of flows. Further, Saborowski (2011) states that the appreciation effect becomes smaller as the level of financial development increases.
Lartey (2007) studies a data set of sub-Saharan African countries using a dynamic panel technique during 1980-2000 and finds that FDI inflows, as well as official aid, causes the real exchange rate to appreciate. Compared to earlier cross-country studies, Ibarra (2011) examined a country-specific study on Mexico. The study was conducted between the years 1988-2008 and included all types of private inflows giving rise to an appreciation of the real exchange rate in the receiving country.
Other studies have reached a less clear-cut conclusion, Bakardzhieva, Naceur and Kamar (2010) studied 57 developing countries between the years 1980-2007 and reached the conclusion that the private capital flows appreciate the host country's real exchange rate, except for the inflow of FDI which was statistically insignificant. Aizenman and Riera-Crichton
7
(2008) studies 80 developing and developed countries and found that, for some country groups depending on the level of trade openness, a depreciation effect comes to light.
To my best knowledge, there is no earlier literature estimating the effect of greenfield investment and M&A inflows on the real exchange rate. Jongwanich and Kohpaiboon (2013) briefly mentioned the topic in their study, which examined nine emerging Asian countries during the time period 2000-2009 and estimated how the real exchange rate was affected by capital inflows. They found an appreciation effect of the host country's currency affected by FDI inflow. This differs from the findings of a similar study using the same countries (Athukorala and Rajapatirana, 2003), although during an earlier time period. Jongwanich and Kohpaiboon (2013) suggests the explanation that M&A (not mentioning greenfield investment), as a part of FDI, had a bigger ratio invested into the non-tradable goods sector when their study took place than when Athukorala and Rajapatirana (2003) performed their study. To be specific, they stress that investments into the sector of non-tradable goods probably appreciate the real exchange rate more than investment into the tradable goods sector, because it targets the source of appreciation directly.
The empirical literature on Dutch Disease is extensive. Most of the literature focuses on the impact of high foreign exchange inflow (e.g. Capital inflow, aid and remittance) on the real exchange rate. Laverty et al (2012) were studying the existence of Dutch Disease effects from remittance, trying to capture both the spending and resource movement effect. They used a GMM estimator and found evidence which showed that an increase in remittance leads to an appreciation of the real exchange rate and a reduction in the ratio of tradable to non-tradable output, meaning that both spending effect and recourse movement effect were operative mechanisms. Moreover, Rajan and Subramanian (2005) analyses the effect of aid on economic growth, using an empirical strategy that exploits both cross-country as well as within-country variation. They find that aid inflows affects the competitiveness negatively using the real exchange rate as the channel.
Magud and Sosa (2010) conducted a study collecting all earlier literature concerning the Dutch Disease, both empirically and theoretically, and examined whether the studies provide a support for the potential negative effects on the economic growth. They conclude quite suggestive results, in over 80% of the cases the real exchange rate appreciates due to a Dutch Disease shock and by that causing a decline in the ratio of tradable to non-tradable output. In
8
more than 90% of the cases a Dutch Disease boom causes factor reallocation as well as an effect on the tradable sector reducing the relative productivity. Moreover, in 75% of all the cases there is a decline in the export sector. In the end they ask themselves if policy makers should be worried about the Dutch Disease effect, and the answer is yes when it comes to the effect on the real exchange rate. They argue that fiscal policy may reduce the impact on the mechanisms of the Dutch Disease. The spending effect associated with Dutch Disease can be mitigated by fiscal policy, which also plays a role in smoothing expenditures to lessen output fluctuations, which would in turn help to restrict the excess demand and therefore the spending effect.
Several studies in the early 1990’s concentrated on how macroeconomic policy could reduce the real exchange rate appreciation effect of capital inflow. The literature mainly focused on restricting the use of capital controls and various foreign exchange interventions (e.g., Calvo, Leiderman, & Reinhart, 1994; Corbo & Hernández, 1996).
To sum up, there is a split perception of how the real exchange rate is affected by FDI. Some studies illustrate that FDI appreciates the real exchange rate and some claims that the real exchange rate depreciates. Although, Magud and Sosa (2010) summary study shows that in over 80 % of the Dutch Disease studies the real exchange rate appreciates, which gives support to the suggested hypothesis. Further, several articles focus on policy action in relation to appreciation of the real exchange rate. Even though policy action can affect the results, it is not the focus of this study.
III. Theory
In this section I will explain the theoretical foundation of the underlying mechanisms which is suggested to drive M&A and greenfield investment to affect the real exchange rate. An appreciation of the real exchange rate will harm the competitiveness of export which leads to a reduction of a country’s current account. These negative economic side-effects are called the Dutch Disease effect and can be explained by the evident causal relationship between the economic improvements in one sector and the decline in other sectors. When the non-tradable sector increases its economic outcome, the given country’s currency becomes stronger in relation to other currencies, i.e. it appreciates. Originally the Dutch Disease corresponds to the massive natural gas finding in the Netherlands in the end of 1950’s, since then the Dutch
9
Disease have been used to describe natural resource booms or large foreign capital inflows, which cause an appreciation of the real exchange rate. Further, threaten the competitiveness of the tradable sector (Corden and Neary, 1982).
The strand of theoretical literature on the Dutch Disease begins with Corden (1981, 1984), Corden and Neary (1982), Van Wijenvergen (1984) and Edwards and Aoki (1983). As an early contributor Corden (1981) explains the Dutch Disease by a two-sector economy. By using a straightforward model, he illustrates that a finding of natural resources caused large capital inflow results in an appreciation of the real exchange rate. More precisely, the real appreciation results from large capital inflow and an excess demand for non-tradable goods. When empirically analysing the Dutch Disease effect Corden and Neary (1982) analyse a general equilibrium model and lean out the fundamental mechanism of the analysis. This is explained as the difference between two effects of the boom (surge in capital inflow), the spending effect and the resource movement effect (also called the movement effect).
Starting with the resource movement effect, which occur when a capital inflow boom’s an industry and affect the marginal productivity of the factor production, hence pulling resources out of the non-booming sector due to a rise in wages. The movement of labour out of the non- tradable goods sector results in a decrease in the output of non-tradable goods. Hence the resource movement effect leads to excess demand of non-tradable goods, therefore the price of non-tradable goods must increase to remove the excess demand. This results in an appreciation of the real exchange rate to restore the equilibrium.
Consider the spending effect, let’s assume that the sector where the capital inflow occur does not use any labour, in order to isolate the spending effect from the resource movement effect.
Given that the demand for non-tradable goods increase with income, the income will rise the aggregate demand. Once again there is an excess demand for non-tradable goods, thus an appreciation of the real exchange rate. Summarizing, either a surge in capital inflow causing an increase of marginal product of labour, drawing recourses out of the non-booming sector (movement effect) or causing a rise in income resulting in expansion of aggregate demand (spending effect). Both of these effect rises the relative price of non-tradable goods leading to an appreciation of the real exchange rate (Corden and Neary, 1982). See Appendix A for a more visual explanation.
10
a. M&A and Greenfield investment
The spending effect and the resource movement effect are the main mechanisms underlying this analysis. A surge in capital inflow, depending on composition of M&A and greenfield investment, could appreciate the real exchange rate trough the channels of spending and movement effect.
To the extent that cross-border M&A and greenfield investment both brings funds to the domestic country, there are no reason to distinguish them. However, to the extent on how they affect the host country there is a reason to distinguish them. The definition of a greenfield investment is described as a country from abroad establishing a new business in the host country, while M&A is described as acquiring or merging with an already existing local firm2. Arguably, the two entry modes are hypothetically different in nature and thus not perfect substitute for each other.
Nocke and Yeaple (2004) emphasise that M&A contributes less to a country’s economy than greenfield since cross-border M&A involves the acquisition of a local firm by a foreign multinational enterprise. Built on that, in a merger or an acquisition the responsibility of the existing employees transfers to the acquiring firm, meaning that that M&A does not increase the employment quantity. On the other hand, Greenfield investment potentially generates more new employees, due to the fact that new operational facilities are established, which indirectly generates new jobs and expands the capital stock (Canton and Solera, 2016). United Nations Conference on Trade and Development (henceforth UNTACD) (2000) stresses that both M&As and greenfield investments contributes with foreign financial funds to the host country, however the financial funds brought to the host country by M&A do not necessarily adds up to the capital stock for production, because of the fact that M&A shift the ownership from domestic to foreign hand. Therefore, the same amount of funds through M&A may equal the less productive investment than the same amount of funds through greenfield investment.
In times of entry, UNTACD (2000) emphasises that Greenfield investment is more likely to transfer new or better technologies and skills than M&A. There is a risk that merging and acquiring firms transfer the most skilled workers abroad, which is probably not the case for greenfield investment since they create new facilities and brings knowledge to the country. The
2 Nearly exclusively all of the M&As are acquisitions, only approximately 3% of the M&As are mergers.
11
result of extensive M&A activity in host countries is that competition may be reduced.
Greenfield investment on the other hand influence countries market structure and its competition by establishing new businesses and can therefore not diminish the market concentration.
The effect on the real exchange rate will likely depend on the composition of the capital inflow - M&A or greenfield. At the entry the characteristics of greenfield investment suggests enhancement of productive capacity, generating employment and contributing to advancement within technology and up-skilling employees as well as raising the capital stock. These outcomes can arguably trigger the mechanisms of the Dutch Disease effect, affecting income distribution and resource allocation causing the real exchange rate to appreciate. Therefore, the characteristics of greenfield investment suggests that greenfield investment work through the channels of spending and movement effect affecting the real exchange rate. On the other hand, inflow of M&As will most likely not work through neither spending effect nor movement effect. To be more precise, in the short run M&A does most likely not add up to the capital stock of production neither creating any new jobs. Therefore, there is reason to believe that M&A won’t cause a Dutch Disease effect or at least have a more attenuated appreciation effect on the real exchange rate.
IV. Econometric approach
In the empirical method I estimate the effect of M&A and greenfield investment on the real exchange rate. In the real exchange rate literature, it is common to use a dynamic panel data model, i.e. where a lagged dependent variable is included in the specification. This is also done in this study since the real exchange rate is thought to depend on its past values, meaning it is autocorrelated. Autocorrelation contributes to bias standard error, but by using a dynamic panel data model this can be moderated.
Given that M&A and greenfield investment is exogeneous to the real exchange rate, the estimated relationship will be causal. However, a
12
account for endogeneity, reversed causality as well as the fact that both past and current values of the real exchange rate are essential determinants of FDI, I will use a linear dynamic panel data model and estimate it with Generalised Method of Moments (GMM), a method often seen in similar studies (e.g. see Saborowski, 2010)
GMM is built on Method of Moments (MM), I use a trivial example for explanation. Suppose a model specified as; 𝑌 = 𝛽𝑥 + 𝑒, where 𝛽 is a parameter, 𝑥 is an endogenous explanatory variable, i.e. a variable correlated with the error term, and e is the error term. Rearranging the regression model gives 𝑒 = 𝑌 − 𝛽𝑥. Now assume 𝑧 be an instrument that is exogeneous to replace for the endogenous variable 𝑥. The properties of the population that MM condition on is 𝐸(𝑥𝑒) = 0. Insert the expression for the error term into the population moment, getting (𝐸[𝑧(𝑌 − 𝛽𝑥)] = 0 which equals 𝐸(𝑧𝑌) − 𝛽𝐸(𝑧𝑥) = 0. Suppose a sample of size 𝑛 is drawn, results in 1∑ (#$)!"& 2 − 𝛽 3∑ (#')!"& 4 = 0. By rearranging one can estimate the parameter 𝛽.
However, if there are no exact solutions, 𝛽 cannot be estimated. For the vast majority of cases, as well as in this study, it is unlikely that there is going to be exact solutions. This defines the benefit of GMM, where exact solutions are not needed. Instead, GMM minimizes a function of the sample moments (Roodman, 2009).
In this study I will include internal instruments, i.e. lagged variables of explanatory variables.
The internal instruments are M&A, greenfield investment, terms of trade, investment by residents and trade openness, since I believe them to be endogenous or predetermined. Most likely will the moment conditions (instruments) not be exactly identified, therefore the choice of GMM in this study. I will use the following specification:
𝑟𝑒𝑒𝑟() = 𝛽*𝑟𝑒𝑒𝑟()+*+ 𝛽,𝐹𝐷𝐼()+ 𝛽-𝑍()+ 𝜇( + 𝜀()
where 𝑟𝑒𝑒𝑟 is the growth rate of the real effective exchange rate estimated as the dependent variable, FDI is a variable either taking the values of greenfield investment or M&As, Z is a vector of control variables (terms of trade, investment by residents abroad and trade openness3),
3Depending on the country’s adjustments of trade restrictions (e.g., tariffs or quotas), prices on tradable and non- tradable goods will be affected causing terms of trade to either increase or decrease. Thus, causing an appreciation of the real exchange rate (Lartey, 2007). Further, terms of trade are expected to appreciate the real exchange rate trough the wealth effect (Saborowski 2010). M&A and greenfield measures net inward investment by non-
(1)
13
𝜇 is the country-specific effect and 𝜀 is the error term. Moreover, 𝑖 indicates the country and 𝑡 indicates the time period (year). Logarithms are taken on all variables, except the variables that can present negative values. The error term is explained by 𝜀() = 𝜇( + 𝜂(), where 𝜇( is the country-specific effect and 𝜂() is the disturbance term. There is a correlation between the lagged dependent regressor, 𝑟𝑒𝑒𝑟()+*, and the error term, 𝜀(), since if an unmodelled shock of the real exchange rate occurs it will be captured by the error term. Because of this OLS estimates would appear bias, to solve this GMM transforms the equation through differencing:
∆𝑟𝑒𝑒𝑟() = ∆𝛽*𝑟𝑒𝑒𝑟()+*+ ∆𝛽,𝐹𝐷𝐼()+ ∆𝛽-𝑍()+ ∆𝜀()
Because the country-fixed effect does not vary over time, it will be eliminated when differentiating (𝜇(− 𝜇( = 0). By equation (2) we can see that the dependent variable lagged two or more years will be qualified as instruments for ∆𝑟𝑒𝑒𝑟()+*, since the error term and the lagged dependent variable in equation (2) are uncorrelated. Turning to the endogenous variables (M&A, greenfield investment and the control variables) lagged levels of the variables themselves are valid instruments.
There are two types of GMM estimators, Difference-GMM and System-GMM. Difference GMM is the underlying method, introduced by Arellano and Bond (1991) which uses the recent explained procedure with population moments and instruments. Yet, if the explanatory regressors depends on past values of themselves, when the variables are lagged by levels, they are weak instruments for differences. This will result in bias estimates. In order to increase the efficiency, one can use system-GMM which is generated by Arellano and Bover (1995) and Blundell and Bond (1998). The system-GMM builds a system of two equations, an original equation and a transformed one, equation (1) and (2) respectively. System-GMM can adequately improve the efficiency because of the introduction of additional instruments, by assuming that first difference of instrument variables is uncorrelated with the country fixed effect. Because of the advantages of system-GMM it is the preferred approach in this study.
residents as a shar of GDP, by including capital investment by non-residents I control for the capital flows from non-foreign actors. In similar models it is common to use some measure of productivity as a control variable, I will not include such a control variable since I argue that it can potentially capture the effect laid out in the theory.
(2)
14
Meaning that the population moments are expressed as 𝐸A𝑟𝑒𝑒𝑟()+.∆𝜀()B = 0, 𝐸A𝐹𝐷𝐼()+.∆𝜀()B = 0 and 𝐸A𝑍()+.∆𝜀()B = 0.
One can either choose one-step system-GMM or two-step system-GMM. The two-step system- GMM provides more efficient asymptotic properties when heteroskedasticity is percent in the error term (Arellano and Bond, 1991; Blundell and Bond, 1998). Windmeijer (2005) show that the two-step GMM can potentially provide standard errors being biased downwards. In order to deal with this bias, I employ the Windmeijer (2005) corrected standard errors in all my regressions.
While the above specification is linear and constitutes my baseline specification, to get a deeper understanding of the mechanisms driving the effect of Dutch Disease I add interaction terms into the model. The interaction terms will give an indication whether the spending or movement effect is deepened by M&A and/or greenfield investment. Worth mentioning, the interaction term will not show if the different investment types matter through the mechanisms. To capture the spending effect, I include an interaction term of the indicative variable of the mechanism together with one of the two investment types, and perform the same procedure capturing the movement effect. A positive interaction term indicates that the investment type included deepens the Dutch Disease effect, spending or movement effect, causing an appreciation of the real exchange, and vice versa for a negative interaction term. For comparison purpose I will estimate these effects with OLS and get an indication whether the mechanism has an effect or not. I will include fixed effects since I believe the data set includes individual effect that are unique to each country. Theoretically, it is likely that both the magnitude of income and the efficiency of M&A and greenfield is affected by the real exchange rate, meaning that the estimates would turn out bias. I follow the OLS regression by valid estimates of GMM.
A potential concern with GMM is that it can be difficult to stipulate the correct specification, due to the fact that the many choices can lead to manipulation of the model. However, there are restrictions evaluating the instruments validity, for which there are two tests. The first one tests the autoregressive serial correlation, where the null hypothesis states; the error term does not contain serial correlation. There are test for both first-order and second-
15
Sargan test and the Hansen test can be applied. In this study I will entrust the Hansen test since it is more robust when the estimation contains heteroscedasticity and autocorrelation. Where heteroscedasticity origins from the differences in country size included in the sample and autocorrelation from variables dependent on its past values, most probably the real exchange rate.
V. Data
The panel data set I use in this study comprises of 87 countries, both developed and developing, see Appendix B for a complete list of the countries. I use annual data during the time period 2003-2016. The choice of sample was due to data availability4, however since the sample includes countries from the area of high-income, upper-middle-income, lower-middle-income and low-middle income5, it is intended to give a more or less representative view of the world.
Table 1 reports the summary statistics. M&A (% GDP) presents an average of approximately 1.3%, but vary greatly across the sample, from -18% for Switzerland to 81% for Luxembourg.
Greenfield investment (% GDP) has a mean of 4%, from 0.06% for Japan to 58% for Mongolia, also indicating a great variation across the sample. Finally, the real effective exchange rate also varies significantly across the sample with an average of 105.6, from 68.1 for Dominican Republic to 239.5 for Kenya. For a deeper understanding of the effects sector-level data would have been desirable, but there is no such data available for this study.
In the baseline model the dependent variable is the real effective exchange rate, a rise in the variable indicates a real appreciation of the host country's currency, see Appendix D for a further explanation of the real exchange rate. The key regressors of interest are the two measures of FDI, disaggregated into cross-border merger and acquisitions (M&A) and greenfield investments. Inflow of cross-border M&A is defined as the mergers with, or acquisitions of, a lasting investment of an already existing domestic firm by a foreign investor6. Greenfield investment is defined as an establishment of a new plant in a foreign country7
4 The sample of country data is limited due to lack of data on M&A. The time sample is limited due to data availability on greenfield investment.
5The classification follows the definition of the United Nations World economic situation prospect (2019). The classification is done through Economies by per capita GNI in June 2018.
6This variable does not include sales of foreign owned affiliates to another foreign MNE.
7 UNCTAD stresses that data may include investments that are not qualified as FDI. Joint ventures are also included, but only where they lead to a new physical operation.
16
(UNCTAD, 2000). See Appendix C for description, expected sign and source of the variables.
The data concerning FDI is sourced from UNCTAD and are based on information reported by Thomson Reuters. The other variables are mainly sourced from The world bank.
Table 1. Summary statistics
Obs. Mean Std. Dev. Min Max
Real effective exchange rate 1177 105.559 16,810 68,1316 239.4655
M&A (% GDP) 1142 0.01025 0.0369 -0.18309 0.8104
Greenfield (% GDP) 1213 0.040 0,0622 0,0006142 0,5808
Terms of trade 1218 111.862 30,699 50,19265 233, 5779
Trade openness 1169 0.9629 0,667 0,2070 4,4262
Inv. by residents 1159 0.0281 0,0417 0,000015 0,26431
Income 1215 23631.64 18895,95 1368,701 1081165,8
TNT 1189 0.392887 0,2311875 0,01289 1,6282
Inflation (%) 1155 4.5378 5,2571 -4,4781 59,2197
Employment in service (%
total emp.) 1204 59.7434 15,37086 20,602 87,8
Note: All 87 countries are included. Neither of the variables are transformed into logarithms.
To get an indication whether the suggested channel driving the real exchange rate I include two interaction terms. To capture the spending effect, I will use income as an indicator, since the income distribution is affected by the spending effect. The income will rise when the aggregate demand of non-tradable goods increase. Accompanied by that, for robustness, I will use inflation as the measure of spending effect, which is a measure used by Papyrakis and Raveh (2014). To capture the resource movement effect, I use the ratio of tradable to non-tradable goods output (related mechanism used by Acosta et al. (2012)). Due to lack of systematic and comparable data on tradable and non-tradable good output I use the sum of agriculture and manufacturing output to approximate tradable goods. Further I use service sector output to approximate non-tradable output. Agricultural, manufacturing and service output is measured as the share of GDP and are acquired from The world bank. For robustness I use a more traditional measure of non-tradable goods – employment in service. In 2003, the starting point of my sample, service is no longer exclusively non-tradable goods. The reasoning behind the mechanisms is to explain the movement in labour over the different sectors.
Along with the above-mentioned variables of interest, I will include control variables to limit the omitted variable bias in the model. These variables can be seen as determinants of the real
17
exchange and are used among other papers (e.g. Saborowski 2010; Combes, Kinda and Plane, 2011). Outstanding determinants of the real exchange rate are trade openness, terms of trade, and capital investments by residents abroad which are included as control variables in the baseline model.
a. Stationarity
A consistent GMM estimator is built on stationary variables, meaning that non-stationarity is a disadvantageous in the model, resulting in a spurious regression. When using an autoregressive model, like in my case, the main concern is the presence of unit root. If a variable contains a unit root, it is an indication of non-stationarity.
There are many different types of unit root tests. I will perform an Im–Pesaran–Shin (IPS) test of a unit root, since it is applicable to panel data and allows for heterogeneous panels. Also, IPS performs better when the time sample is small compared to other similar tests, meaning it is applicable to this study (Im et al. 2003).
Table 2. Im-Pesaran-Shin test for unit root
Variable p-value
No time trend included Time trend included
Real exchange rate (ln) 0.0096 0.0799
M&A (% GDP) 0.0000 0.0000
Greenfield inv. (% GDP) 0.0000 0.0000
Note: Not all countries are included since it is not possible to run the test when the data is to unbalance.
In the IPS test the null hypothesis is stated; there exists a unit root and I will evaluate the test at a 5% significant level. Table 2 implies that M&A and greenfield investment can’t be rejected, which implies that these variables does not contain any stationarity. Although, the real exchange rate is not rejected when a time trend is excluded but rejected when a time trend is included. To decide whether a time trend should be included or not depends on the data generating process. To get an indication whether the data contains a linear time trend or not, I visually review the data of the dependent variable and conclude that majority of the panels does not contain a linear time trend. This suggests us to focus on the unit root test who excluded the time trend, where the p-value is not rejected meaning that the real exchange rate is most probably stationary.
18
A repeated criticism of unit root tests concerns the poor power and size properties that many of such tests present. On the other hand, when using system GMM there is most likely not a problem with non-stationarity since the issue concerning autocorrelation is much less severe and can be corrected by accounting for time-specific effect, i.e. including a time-dummy. This I have done in all regressions. I conclude that there is no need to worry for spurious regressions, cause by non-stationarity in the variables.
VI. Results
The estimated effect of M&A and greenfield investment on the real exchange rate is shown in Table 3. In all regressions the baseline control variables are; trade openness (export + import as a share of GDP), terms of trade and capital investments by residents abroad (called inv. by residents). A positive estimate infers a real appreciation of the exchange rate and vice versa for a negative estimate. In Regression I both types of capital inflows are included and Regression II and III include one investment type at the time. In Table 3 all the regressions are obtained using the full amount of observations. In Regression I the majority of the estimates obtain the expected sign (see Appendix B for explanation of the variables) suggested by the theory. In Regression II several of the estimated coefficients have the expected sign, including greenfield investment. Greenfield investment will have an appreciation effect of the real exchange rate, in line with the theory laid out. In Regression III most of the estimates corresponds to the results of Regression I and II, including M&A, which indicates that if M&A increases the real exchange rate would depreciate, which is a somewhat surprising result. However, in all three regressions there is little evidence of significant effects from the regressors of interest. Table 3 suffers from inconsistency in the estimates, although the estimates indicate some support to the suggested theory.
Table 3. Core estimates Dependent variable Estimation
Period
Unit of observations
Real effective exchange rate
Two-step system GMM estimation with Windmeijer standard errors 2003-2016
Annual
Regression I Regression II Regression III
Lagged dependent
variable 0.938***
(0.148) 1.103***
(0.146) 0.917***
(0.0974)
M&A -0.836
(0.691) -1.157
(0.873)
19
Greenfield 0.116
(0.731)
0.117 (0.380)
Inv. by residents 0.882*
(0.471) 0.392
(0.893) 0.581
(0.435)
Trade openness (ln) 0.0585
(0.0712)
0.0427 (0.0427)
0.107* (0.0631)
Terms of trade (ln) 0.0726
(0.0492) 0.0829*
(0.0426) 0.0761*
(0.0410)
Number of obs. 982 1037 982
Number of lags 2 2 2
Number of groups 84 85 84
Number of instruments 14 12 12
Obs/group
min 3 1 3
avg 11,69 12,20 11,69
max 13 13 13
Specification tests
Hansen test (p-value) 0,860 0,204 0,988
AR(2) (p-value) 0,181 0,144 0,123
Note: All regressions include a year dummy. Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01.
The null hypothesis of the Hansen test is; instruments are not correlated with the error term. The null hypothesis for the AR(2) test state; there is no serial correlation. The table states that we cannot reject the null hypothesis on neither of the tests.
The absence of support to the suggested theory could be explained by policy actions preventing an effect on the real exchange rate. Magud and Sosa (2010) stresses that fiscal policy has an effect on the real exchange rate. However, the potential effect policy actions could have on the real exchange rate are beyond the scope of this analysis.
a. Mechanisms – spending and resource movement effect
In absence of convincing outcomes of the different investment types on the real exchange rate, the results from Table 3 suggest the need to further explore the channels of M&A and greenfield investment affecting the real exchange rate. Me arguing that the effect on the real exchange rate will be channelled through the spending and movement effect. To get a deeper understanding of the mechanisms I include an interaction term between M&A or greenfield investment and the mechanism itself.
The regressions in Table 4 present the mechanisms of spending and movement effect with income and tradable-to-non-tradable goods output (TNT) as indicative variables respectively, accompanied by an interaction term between each investment type and the mechanism at focus.
The interaction term will indicate if the two investment types deepens the effect of the spending
20
and/or movement effect. Yet, as mentioned, the interaction term will not capture if the suggested effects are channelled through the mechanisms. However, given that the direct links of FDI are questionable this will give an indication if the different investment types extends the effect of the mechanism or not. Table 4 also includes OLS estimates for comparison purpose. Spending effect and movement effect is presented in Table 4, panel (a) and panel (b) respectively.
Table 4 panel (a) focuses on the spending effect, the first two columns examine the relationship measured with OLS and the latter two with GMM, including the same variables. A first glance at the results, Regression I and II shows that income matters for the real exchange rate since it is strongly significant. This implies that a higher level of income would appreciate the real exchange rate, as the theory suggests. However, the interaction terms are insignificant. In Regression IV the interaction term including M&A suggests that a higher level of income will attenuate the appreciation of the real exchange rate of M&A inflow. Regression III illustrate by the sign of the estimates that the interaction terms are in line with the suggested theory, the interaction term including greenfield investment suggests that a higher level of income will increase the appreciation of the real exchange rate of greenfield investment. Although, the income estimate suggests that an increase in income will depreciate the real exchange rate, although the estimates are insignificant.
Table 4 panel (b) focus on the resource movement effect and uses the same structure as in Table 4 panel (a). Starting off with the OLS results in Regression I and II, which illustrate that TNT does not have any significant effect on the real exchange rate, this corresponds to the results in Regression III and IV presenting the GMM results. However, in Regression III and IV the estimates of TNT are significant for the interaction terms. The interaction term between greenfield investment and TNT is positive and significant at a 5% level, which indicates that a real appreciation effects of greenfield are enhanced through the resources moving to the non- tradable sector. This is in line with what was earlier argued, that greenfield investment will appreciate the real exchange rate through the resource movement effect. The estimate of M&A shows what has been acknowledge before, the interaction term between M&A and TNT is negative, which infer that an appreciation effect of real exchange rate attenuates through the resources moving to the non-tradable sector. In Table 4 (panel a and panel b) neither of the channels (income nor TNT) itself suggest any clear-cut results. However, I will not focus too
21
much on the mechanism variable itself, partly because of the possibility that other effect might be included, and partly because the OLS estimates are not valid.
Table 4. (a) spending effect and (b) resource movement effect (a)
Dependent variable Estimation
Period
Unit of observations
Real effective exchange rate
OLS/Two-step system GMM estimation with Windmeijer standard errors 2003-2016
Annual
Regression I OLS
Regression II OLS
Regression III GMM
Regression IV GMM Lagged dependent
variable 0.782***
(0.124)
0.786***
(0.108)
Income (ln) 0,255***
(0.0578)
0,249***
(0.0601)
-0.0229 (0.0144)
-0.00405 (0.0407) Income * M&A -0,00259
(0.00437)
-0.102 (0.0652)
Income * Greenfield -0,00834
(0.00875)
0.0292 (0.0670) Inv. by residents -0.401
(1.015)
-0.320 (1. 003)
0.0831 (0.830)
0.706 (1.245) Terms of Trade (ln) 0,156**
(0.0631)
0,155**
(0.0647)
0.0499 (0.0492)
0.101**
(0.0394) Trade openness (ln) -0,310***
(0.0527)
-0.301***
(0.0529)
0.00877 (0.0588)
-0.00814 (0.0507)
Fixed effect yes yes
Number of obs. 1052 1117 982 1037
Number of lags 2 2
Number of groups 84 85 84 85
Number of
instruments 12 12
Obs/group
min 5 2 3 1
avg 12.5 13.1 11.69 12.20
max 14 14 13 13
Specification tests
Hansen test (p-value) 0.869 0.058
AR(2) (p-value) 0.148 0.174
(b)
Regression I OLS
Regression II OLS
Regression III GMM
Regression IV GMM Lagged dependent
variable 0.896***
(0.134) 0.778***
(0.131)
TNT -0.0525
(0.127)
0.00379 (0.118)
0.183 (0.167)
-0.163 (0.115)
TNT * M&A 0.580
(0.509) -7.606**
(3.788)
TNT * Greenfield -0.0274
(0.147) 1.651**
(0.638)
22
Inv. by residents -0.418
(1.017) -0.337
(0.990) 0.756
(1.752) 1.290
(1.365) Trade openness (ln) -0.298***
(0.0558) -0.296***
(0.0536) 0.0759
(0.0654) -0.0485
(0.0652) Terms of trade (ln) 0.219***
(0.0630)
0.222***
(0.0642)
0.0803 (0.0490)
0.0511 (0.0537)
Fixed effect yes yes
Number of obs. 1043 1105 974 1028
Number of lags 2 2
Number of groups 83 83 83 83
Number of instruments 12 12
Obs/group
min 5 7 3 6
avg 12.6 13.3 11.73 12.39
max 14 14 13 13
Specification tests
Hansen test (p-value) 0.899 0.247
AR(2) (p-value) 0.392 0.297
Note: All regressions include year dummy. Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01. The null hypothesis of the Hansen test is; instruments are not correlated with the error term. The null hypothesis for the AR(2) test state; there is no serial correlation. The table states that we cannot reject the null hypothesis on neither of the tests.
b. Unexpected effect by M&A
I expected to find an effect of greenfield investment appreciating the real exchange rate and a weaker or non-appreciating effect of M&A. Even though, the estimates of greenfield investment didn’t present a convincing result that supports the suggested theory, I captured something I didn’t expect. The results of Table 3 and 4 illustrates that M&A appears to depreciate the real exchange rate. To investigate what drives this effect I have divided the sample into low- and high-income countries, 42 and 44 countries respectively8. Table 5 reports the results from regressing the different investment types on the real exchange rate with a divided sample. The table illustrates that the relationship between investment and the real exchange rate is driven by the low-income countries. Regressions concerning the high-income countries does not provide any significant results, neither provide the AR(2) test nor the Hansen test comfortable results. Meaning that the results in Table 5 concerning the high-income countries are questionable because of the poor test results.
8 The classification follows the definition of the United Nations World economic situation prospect (2019). The classification is done through Economies by per capita GNI in June 2018.
23
Focusing on the low-income countries (Regression I, II and III). Regression I and II illustrates that the majority of the coefficient estimates obtain the anticipated sign, and several of the estimates are significant. M&A are significant in both of the regressions9. Focusing on Regression II, if M&A increases it will result in a negative effect on the real exchange rate, meaning a depreciation. Athukorala and Rajapatirana (2003) found a depreciation effect when measuring FDI on the real exchange rate. They explain this effect by arguing that FDI is bias towards the tradable goods sector compared to other capital flows. To investigate if that is the case another type of data would be needed, which is outside the scope of this analysis.
Further, Regression III obtains estimates in line with the theory. Greenfield investment contains the expected sign, which indicates that even in low income countries the laid-out theory seems to be correct. Although, one can conclude that there are no obvious signs that the effect of greenfield investment on real exchange rate is driven neither by low-income countries nor high- income countries.
Another effect to keep in mind, UNCTAD highlights that an appreciation effect can occur from a pressure on the domestic currency due to a large transaction. M&A transactions often occur immediately while greenfield investments usually are spread out over a period of time. That is, if the volume of the transaction is the same for both M&As and greenfield investment, M&As tend to appreciate the currency more because of one single transaction rather than several transactions that are spread out (UNCTAD, 2000). This could potentially explain why M&A is significant, but not why M&A presents a negative estimate which could possibly be explain by the sector of investment.
Table 5. Sample divided into low- and high-income countries Dependent variable
Estimation Period
Unit of observations
Real effective exchange rate
Two-step system GMM estimation with Windmeijer standard errors 2000-2016
Annual Reg. I
Low income
Reg. II
Low income
Reg. III
Low income
Reg. IV
High income
Reg. V
High income
Reg. VI
High income
Lagged dependent variable
0,684***
(0,139)
0,665***
(0101)
0,708***
(0,183)
0,842***
(0,0867)
0,0896***
(0,0821)
0,812***
(0,0914)
9 In Regression I both terms of trade, Inv. by residents and M&A are significant at a 10% level and becomes significant at a 1% level in Regression II when greenfield is excluded from the model. This could indicate that that in Regression II M&A “absorbs” the effect of greenfield investment. That is, greenfield investment is significantly associated with M&A and the real effective exchange rate by controlling for greenfield investment biased the parameter estimates of M&A, making seem more significant than it really it.
24
M&A -0,505*
(0,276)
-0,521***
(0,180)
-0,312 (0,325)
-0,243 (0,215)
Greenfield -0,0522
(0,541)
1,207 (0,755)
0,282 (0,466)
0,667 (0,570)
Inv. by residents 1,383*
(0,0807)
1,248**
(0,614)
1,582 (1,534)
0,625 (1,169)
0,420 (0,861)
0,882 (1,261) Trade openness (ln) -0,0548
(0,0853)
-0,0592 (0,0707)
-0,196 (0,141)
-0,0275 (0,0380)
-0,00513 (0,0314
-0,0562 (0,0451) Terms of trade (ln) 0,163*
(0,0772)
0,163**
(0,0704)
0,0806 (0,113)
0,0398 (0,0446)
0,0541 (0,0414)
0,0398 (0,0451)
Number of obs. 466 466 511 516 516 526
Number of lags 2 2 2 2 2 2
Number of groups 42 42 43 42 42 42
Number of instruments 14 12 12 14 12 12
Obs/group
min 3 3 1 7 7 7
avg 11,10 11,10 11,88 12,29 12,29 12,52
max 13 13 13 13 13 13
Specification tests
Hansen test (p-value) 0,729 0,934 0,171 0,062 0,146 0,076
AR(2) (p-value) 0,435 0,352 0,680 0,010 0,014 0,002
Note: All regressions include year dummy. Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01.
There are 43 low-income countries and 44 high-income countries. The null hypothesis of the Hansen test is;
instruments are not correlated with the error term. The null hypothesis for the AR(2) test state; there is no serial correlation. The table states that we cannot reject the null hypothesis on neither of the tests.
VII. Robustness test
25
Inflation -0.00409***
(0.000944)
-0.00440***
(0.000876)
0.0140 (0.0126)
-0.00505 (0.00706) Inflation * M&A 0.0314
(0.0269)
-0.660 (0.518)
Inflation * Greenfield 0.00574
(0.00520)
0.0742 (0.0643) Inv. by residents -0.241
(1.097)
-0.141 (1.040)
1.362 (1.107)
1.399 (1.272) Trade openness (ln) -0.312***
(0.0534)
-0.297***
(0.0538)
0.0400 (0.0779)
-0.0651 (0.0452) Terms of trade (ln) 0.209***
(0.0680)
0.213***
(0.0700)
0.0474 (0.0448)
0.00654 (0.0616)
Fixed effect yes yes
Number of obs. 1016 1081 951 1006
Number of lags 2 2
Number of groups 81 82 81 82
Number of instruments 12 12
Obs/group
min 4 2 3 1
avg 12.5 13.2 11.74 12.27
max 14 14 13 13
Specification tests
Hansen test (p-value) 0.973 0.163
AR(2) (p-value) 0.241 0.765
(b)
Dependent variable Estimation
Period
Unit of observations
Real effective exchange rate
OLS/Two-step system GMM estimation with Windmeijer standard errors 2003-2016
Annual Regression I
OLS
Regression II OLS
Regression III GMM
Regression IV GMM Lagged dependent
variable 0.752***
(0.134)
0.730***
(0.0956) Employment in service 0.0545
(0.157)
0.0171 (0.179)
-0.0370 (0.0628)
-0.0120 (0.0948) Employment in service
* M&A 0.0131
(0.0157)
-0.402* (0.241) Employment in service
*Greenfield -0.0241
(0.0222)
0.0382 (0.167) Inv. by residents -0.379
(0.996)
-0.232 (0.968)
1.208 (1.321)
1.416 (1.040) Trade openness (ln) -0.315***
(0.0526)
-0.306***
(0.0524)
0.0131 (0.0437)
-0.0127 (0.0419) Terms of trade (ln) 0.222***
(0.0636)
0.218***
(0.0649)
0.0587 (0.0522)
0.0938**
(0.0404)
Fixed effect yes yes
Number of obs. 1047 1112 977 1032
Number of lags 2 2
Number of groups 83 84 83 84
Number of instruments 12 12
Obs/group
26
min 5 2 3 1
avg 12.6 13.2 11.77 12.29
max 14 14 13 13
Specification tests
Hansen test (p-value) 0.443 0.123
AR(2) (p-value) 0.261 0.129
Note: All regressions include year dummy. Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01. The null hypothesis of the Hansen test is; instruments are not correlated with the error term. The null hypothesis for the AR(2) test state; there is no serial correlation. The table states that we cannot reject the null hypothesis on neither of the tests.
In Table 6 panel (a) we see similar results as when the spending effect was measured by income. Regression I and II presents the OLS results which reveals that inflation has a negative impact on the real exchange rate. In the corresponding table where the spending effect is proxied by income, Table 4 panel (a), the income measure is positive, instead of negative. This can be explained by the fact that the two variables are systematic diverse. The interaction terms in Regression III and IV indicate that M&A attenuates the spending effect and that greenfield investment deeper the spending effect. The estimate of the interaction term including greenfield investment is in line with the suggested theory. It is clear that the two interaction terms measured with either, income or inflation, corresponds to each other. These estimates are not subject to any substantive changes, indicating that the results are somewhat robust.
Table 6 panel (b) illustrates the resource movement effect. Overall the estimates are similar to the corresponding table, Table 4 panel (b). Also, Regression III illustrates a significant negative estimate of the interaction term between employment in service and M&A. However, the interaction term in Regression IV indicates by the sign that the theory laid out is correct. The lack of significant estimates could be explained by the more traditional measure of employment in non-tradable sector than the measure of TNT, which includes several sectors and does a comparison.
A. Other robust tests
The dynamic panel data model I perform in this study allows for weak endogeneity of the main regressors. Endogeneity is less likely when it comes to interaction terms, but the variables of interest in the core regressions could potentially give rise to endogeneity, I check the validity of the instruments with the Hansen test of over-identification. The Hansen test estimates the joint validity of all the instruments. In almost neither of the cases the null hypothesis is rejected,
