The quantitative research does not give rise to particular ethical issues. The data on FDI flows is available free of charge from UNCTAD. The risk rating data from ICRG has been purchased and
100
the risk data from Global Insight has been obtained from Nyenrode University. In both cases, the use of the data is in conformance with the appropriate licensing agreement.
With respect to the case study and in-depth interviews, care has been taken not to divulge
information considered as proprietary, confidential or sensitive by the interviewees. When required, the identity of specific interviewees has been hidden in order to be able to report results of
significance without compromising confidentiality.
4.6 Conclusions
By combining quantitative and qualitative research methods, the topic of the research is approached from different perspectives and a comprehensive model of the determinants of FDI in the MENA region can be built and tested. Time series cross sectional analysis is chosen as an appropriate methodology for the study of Location factors, in a way that is consistent with other academic studies on FDI Location factors. Two different and detailed sets of risk ratings from reputable information publishers are used to analyse the role played by different risk factors.
A multiple case study based primarily on in-depth interviews is preferred to survey based
questionnaires. The case study is used to develop insights that can supplement, support or put into question the results of the statistical analysis and can lead to the development of new hypotheses and propositions that may be tested empirically in future research.
Based on the development of the hypotheses and case study objectives in this chapter, the initial conceptual model can be expanded as per Figure 6.
In this picture, the location factors are consistent with the hypotheses developed in Section 4.5. The available location choices are as per the country scope of the study, whereas the available mode choices are consistent with the academic literature on the subject. Only the potential transaction cost elements are not specified in advance. These will be identified and tested during the case study.
101
102
Chapter 5
Overall determinants of FDI in the MENA region 5.1 Introduction
This chapter deals with the testing of the model of the overall determinants of FDI flows into the MENA region and the testing of the hypotheses 1 to 5 described in Chapter 4. The model and the relevant sources of data are explained in Section 5.2. Section 5.3 contains the results of the
regression analyses, Section 5.4 provides a discussion of the results, followed by the conclusions in Section 5.5.
5.2 The model
In order to test the different hypotheses described in Chapter 4, a multiple regression model is built that includes the main parameters that are expected to be associated with FDI inflows according to the existing literature, as well as two new parameters related to natural resource endowments and oil prices.
The link between overall FDI inflows and the size of a country‘s economy as measured by total GDP is well established, as evidenced by the research quoted in Appendix E and the correlation coefficient between FDI and GDP of 0.625 for countries in the MENA region for the period 1987 - 2008. Therefore, as the dependent variable FDI inflows as a share of GDP is considered to be preferable to FDI inflows on their own. This definition of the dependent variable is in line with other studies on FDI determinants such as Jun & Singh (1995), Chan & Gemayel (2004) and Mina (2007). Given the near equivalence of the FDI/GDP measure to the FDI Performance Index published by UNCTAD‘s World Investment Report (see Chapter 3), the independent variable of FDI/GDP can also be referred to as a measure of a country‘s FDI performance. The alternative model specification would have used FDI as the dependent variable and include GDP as one of the independent variables. Such a model specification would increase the overall fit of the model, since GDP clearly explains a large part of the variation in FDI. However, in this case the risks of
multicollinearity and autocorrelation would be higher. For the purpose of this research, defining the dependent variable as FDI/GDP provides a clearer focus on what drives a country‘s FDI inflows in relation to the size of its economy.
103
The data on FDI flows is obtained from the relevant UNCTAD World Investment Reports. In order to correspond to the available data on the independent variables, annual data for the 22 year period between 1987 and 2008 are used for the 16 MENA countries that are part of the country sample. For the overall sample, this leads to a sample size that consists of 352 observations.
Whenever data is missing in terms of either the FDI score or one of the independent variables then the entire observation is not taken into account in the regression models.
The independent variables included in the model are designed to test the individual hypotheses and to give a comprehensive understanding of the drivers of FDI flows into the MENA region while keeping the overall number of explanatory variables small enough to be able to manage issues of multicollinearity.
Having established the strong correlation between FDI and GDP, market attractiveness is additionally measured by GDP per capita, based on the premise that markets with affluent
consumers are more attractive for market seeking FDI. There may be a reverse impact of GDP per capita on efficiency seeking FDI since high GDP per capita implies high wage rates, although the direction of this relationship does of course also depend on the distribution of income within a country and the mix between market seeking and factor seeking FDI. As stated in Chapter 3, the available evidence strongly suggests that there is little factor seeking FDI into the MENA region that is looking to benefit from low labour costs. It can therefore be expected that the market
attractiveness aspect of the GDP per capita measure on FDI is stronger than the wage impact and that GDP per capita is a determinant of FDI inflows.
Openness to trade is measured by a country‘s manufacturing exports as a percentage of GDP. Countries that are successful in exporting may also be successful in attracting foreign manufacturers that aim to export their products. The direction of causality between exports and FDI is not self evident; do open economies attract FDI or do countries that attract FDI increase their
manufacturing exports as a result? Jun & Singh (1995) came to inconclusive results regarding the direction of this relationship, but stated that the influence is most likely taking place in both directions, with the dominant impact being from openness to trade on FDI. In this study, only the strength of the relationship between openness to trade and FDI will be tested, assuming that if the two variables of FDI and openness to trade are positively associated with each other, there will be a mutually reinforcing relationship that operates in both directions.
104
As a measure of environmental risk, the Composite Risk rating score is used from the International Country Risk Guide published by the Political Risk Group (PRS). The choice for ICRG risk ratings has been explained in Chapter 4. Among the different risk scores available from ICRG, the
Composite Risk rating best reflects the overall concept of environmental risk as used in the International Business research. The role of a number of specific environmental risks is tested in Chapter 6.
A country‘s natural resource endowment is measured by its total oil and gas reserves. Oil reserves are measured in billions of barrels of proven reserves and gas reserves are measured in cubic meters (BP Statistical Review of World Energy, 2009). In order to arrive at one aggregate measure of a country‘s energy endowments, gas reserves are then transformed into equivalent oil reserves using the industry standard conversion ratio of 6.6 barrels of oil per 1000 cubic meters of gas
(www.rigzone.com).
Oil prices are measured by the world oil price at the start of a year (Source: BP Statistical Review of World Energy, 2009). Since changes in oil prices require some time to feed through to higher government revenues and potential changes to FDI, the impact that is tested is of a 1-year lagged effect of oil prices on FDI.
These parameters lead to the following model specification:
FDI/GDP = ƒ (GDP PC, OPENNESS, RISK, ENERGY, OIL PRICE)
With this model specification, it is expected that the most relevant factors have been included in the model while managing the risk of multicollinearity. Among other factors that appear in the
published literature, education (sometimes also measured as availability of skilled labor) and infrastructure are not included in the model. For both factors it is difficult to obtain relevant time series data in a consistent way for countries across the region. Education levels and the availability of skilled labor are difficult to quantify given the very large role played by foreign workers in the region, particularly in the GCC countries where qualified labor is brought into the country depending on the requirements at any particular time. In this sense, the supply of both cheap unskilled labor and highly educated staff are not constrained by what is available in the local labor market. Quality of infrastructure is difficult to quantify in a relevant way since most of the traditional measures, such as
105
telephone lines, are not likely to be adequate proxies for infrastructure development. The case study will test which aspects of infrastructure may be most relevant for multinational companies.
A total of three robustness checks are carried out in addition to the main regression equation. Given the different economic profiles of OPEC and non-OPEC countries, the regressions are run for all countries in the sample as well as separately for the groups of OPEC and non-OPEC countries. The split is made to determine if there are differences in the determinants of FDI inflows depending on a country‘s endowment of oil and gas resources. Such analysis is in line with Jun & Singh (1995) who split their country sample between recipients of high and low FDI flows to determine the underlying drivers for countries with very different profiles. A second robustness check is made by splitting the sample into two equal time periods (1987-1997 and 1998-2008) to ascertain whether the
determinants of FDI inflows have changed over time.
A third robustness check is made by replacing the ICRG risk rating with the risk rating of Global Insight. Since the Global Insight risk ratings are available for a shorter time period than ICRG ratings, the sample size is somewhat reduced and the regression is run only for the entire period for which Global Insight data is available. When considering the regression results, it needs to be kept in mind that the rating scales for ICRG and Global Insight work in opposite directions. For ICRG a high score is associated with relatively low risk, whereas for Global Insight ratings a high score is associated with high risk.
In summary, a total of six regression models are produced as follows:
Model 1: All MENA countries, all years (1987 – 2008), ICRG risk data Model 2: OPEC countries, all years (1987 – 2008), ICRG risk data Model 3: Non-OPEC countries, all years (1987 – 2008), ICRG risk data Model 4: All MENA countries, early years (1987 – 1997), ICRG risk data Model 5: All MENA countries, recent years (1998 – 2008), ICRG risk data Model 6: All MENA countries, all years (1987 -2008), Global Insight risk data
106
5.3 Results
Table 11 shows a correlation matrix between the dependent and independent variables for all countries for the period 1987-2008, as well as for the variables FDI and GDP separately. From Table 11, it can be observed that the dependent variable (FDI/GDP) is significantly correlated with all the independent variables in the model (at the 1% level of significance). The relationship is positive in all cases, except for Oil & Gas Reserves, which are negatively correlated with FDI/GDP. Among the independent variables in the model several significant correlations exist, with the highest correlation being 0.435 between GDP per capita and Composite Risk. This points to a potential concern of multicollinearity between the dependent variables, for which the relevant tests will be carried out.
Table 11: Bivariate Correlation Coefficients of model variables
FDI GDP FDI / GDP GDP Per Capita Manuf. Exports/ GDP Composite Risk
Oil & Gas Reserves Oil Price FDI 1 GDP .625 ** 1 FDI / GDP .363 ** -.027 1 GDP Per Capita .237 ** .179 ** .123 * 1 Manufacturing Exports/GDP .210 ** -.055 .281 ** .154 ** 1 Composite Risk .261 ** .237 ** .160 ** .435 ** .277 ** 1 Oil & Gas
Reserves .211 ** .738 ** -.193 ** .340 ** -.124 * .188 ** 1
Oil Price .509 ** .373 ** .380 ** .282 ** .177 ** .374 ** .069 1
Note:
** Correlation is significant at the 1% level (2-tailed). * Correlation is significant at the 5% level (2-tailed).
107
The results of the regression models are shown in Table 12 for each of the models specified. Model 1 represents the basic model, while models 2 – 6 represent robustness checks based on subsamples of the overall dataset.
Definition of models 1- 6:
Model 1: All MENA countries, all years (1987 – 2008), ICRG risk data Model 2: OPEC countries, all years (1987 – 2008), ICRG risk data Model 3: Non-OPEC countries, all years (1987 – 2008), ICRG risk data Model 4: All MENA countries, early years (1987 – 1997), ICRG risk data Model 5: All MENA countries, recent years (1998 – 2008), ICRG risk data Model 6: All MENA countries, all years (1987 -2008), Global Insight risk data
Table 12: Regression results
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
Constant -0.526 (0.660) -0.873 (0.214) -0.69 (0.971) -1.552 (0.613) 10.660 (0.000) ** -4.076 (0.040) * GDP Per Capita 0.000 (0.231) 0.000 (0.005) ** 0.000 (0.001) ** 0.000 (0.522) 0.000 (0.008) ** 0.000 (0.010) * Manufacturing Exports/GDP 0.088 (0.002) ** 0.098 (0.000) ** 0.019 (0.701) 0.022 (0.691) 0.139 (0.000) ** 0.0134 (0.000) ** Composite Risk 0.000 (0.995) -0.007 (0.536) -0.004 (0.894) 0.039 (0.152) -0.171 (0.000) ** 1.398 ('0.021) * Oil & Gas
Reserves -0.010 (0.000) ** -0.001 (0.362) -0.231 (0.012) * -0.010 (0.013) * -0.009 (0.001) ** -0.011 (0.000) ** Oil Price 0.095 (0.000) ** 0.057 (0.000) ** 0.115 (0.000) ** 0.037 (0.778) 0.108 (0.000) ** 0.054 (0.011) * Adjusted R- squared 0.216 0.546 0.236 0.029 0.401 0.227 Observations 323 138 184 163 159 143 Durbin Watson 1.204 0.827 1.463 2.094 1.694 2.197 Note: t-statistic in brackets ** Significant at 1% level * Significant at 5% level
108
Considering first the model for all MENA countries for all years (1987 – 2008), three of the parameters in the model are significant and two parameters are not significant. In all cases, the results refer to significance after controlling for the other variables in the model. Manufacturing exports as a share of GDP and oil prices are positively associated with FDI, whereas a country‘s oil and gas reserves are significant and negatively associated with FDI. The variables GDP per capita and Composite Risk are both found to be not significantly associated with FDI performance. The overall adjusted R-squared values range from a low of 21.6% (Model 1) to a high of 54.6% (Model 2). This apparently relatively low value of R-squared is a direct result of choosing FDI/GDP as the dependent variable. The model aims to explain the part of FDI inflows into a country that is not already explained by the size of a country‘s economy, as measured by GDP. If the model had used FDI as its dependent variable and included GDP among the independent variables, clearly the R-squared values would have been much higher, but less precision would have been obtained regarding the role played by the other determinants in the model and the risk of multicollinearity would have increased. Among the six models that are tested, the highest R-squared levels are for Model 2 (OPEC countries, all years) and Model 5 (all countries, recent years).
When the sample is split between OPEC and non-OPEC countries (Models 2 and 3), there are some similarities and differences compared to the overall MENA model. In line with the overall model, the role of oil prices in determining FDI flows is significant and positive in both OPEC and non- OPEC countries. Also in line with the overall model, the role of the Composite Risk measure is not significant in both sub-samples of OPEC and non-OPEC countries. Contrary to the overall model, GDP per capita is found to be significantly associated with FDI performance in both of the sub- samples of OPEC and non-OPEC countries. This finding indicates that GDP per capita is associated with FDI performance within the group of OPEC countries and within non-OPEC countries but not across OPEC and non-OPEC countries in the MENA region.
Again with respect to Models 2 and 3, the role of manufacturing exports is found to be positive in OPEC countries. The role of oil and gas reserves is not significant within the group of OPEC countries but is significant and negative among non-OPEC countries.
When the results of the first 11 years (1987 – 1997) are compared to the second 11 year (1998 – 2008) period (models 4 and 5), the model is more robust for the later period, as demonstrated by the significance of the individual parameters and the much higher R-squared values. In the earlier
109
period, only a country‘s oil and gas reserves are significant and negatively associated with FDI performance. In the later period, all variables are significant, with the parameters GDP per capita, Manufacturing Exports and Oil Price having a positive sign and Composite Risk and Oil & Gas Reserves with a negative sign.
Model 6 displays the results using Global Insight as the data source for the environmental risk score. Since Global Insight data is only available for the 1997 – 2005 period, the results are most
comparable to model 5 which is based on ICRG risk data for the 1997 – 2008 period. The results based on ICRG and Global Insight data are very similar, keeping in mind that for ICRG a high risk score means relatively little risk, whereas for Global Insight a high risk score indicates high risk. Surprisingly, for the later period, using either ICRG or Global Insight data, high levels of environmental risk are associated with high levels of FDI inflows.
Tests point out that the model does not suffer from multicollinearity issues. For each regression, multicollinearity statistics have been obtained and the Variance Inflation Factor (VIF) is below 4 for all regressions, whereas typically only VIF values over 5 or 10 give concern for multicollinearity in the model (Field, 2009). The full model results and multicollinearity statistics are displayed in Appendix C.
The Durbin Watson statistics reported in Table 12 do point towards a concern of autocorrelation, particularly for Model 2 which includes only the OPEC countries in the sample. The Durbin Watson test measures the level of correlation between residual terms in the regression equation. The test produces values ranging from 0 to 4, with a value of 2 meaning that the residuals are uncorrelated, a value higher than 2 indicating a negative correlation between adjacent residuals and a value less than 2 indicating a positive correlation between residuals. As a conservative rule of thumb, values of less than 1 and greater than 3 pose a cause for concern (Field, 2009). The presence of autocorrelation does not lead to biased estimates of the coefficients in the model but potentially leads to an