Introduction
During recent years the world has experienced numerous political, social and cultural shocks: riots, revolutions, regime changes and severe crises. The latest events, regarding “Arab spring” and conflicts in Syria of 2011-2012 are widely discussed by journalists and economists as well. Here we can suggest a quotation from The Economist Intelligence Unit publication, “Spring Tide: Will the Arab risings yield democracy, dictatorship or disorder?”: “The movements inspired by the popular overthrow of autocratic regimes in Tunisia and
Egypt earlier this year have raised hopes of a widespread shift in the Arab world towards open and pluralistic regimes, ushering in economic change and ending generations of relative stagnation”. (The Economist), also we can mention an opinion by G. Osborne, a
British Conservative politician: “…the push for democracy in Arab spring countries could fail unless the international community speeded up financial support to the region” (the Financial Times, May 18, 2012). The democratic society‟s great concern about the political situation in African regions is supported by both socio-cultural aspects and also economic issues. Again, we can cite the Financial Times article: “The IMF, launching its twice yearly regional economic outlook, said countries such as Egypt and Tunisia were running out of policy options to battle a “slow and drawn-out economic recovery”, as the move from dictatorship to democracy takes longer than expected” (The Financial Times, 2 May, 2012). There is
evidence to believe that western democratic states fight for democracy in other countries, but why they consider democracy the best option for countries deciding on their political system and economic development?
There are numerous articles on the value of democracy: “Democracy as a Universal Value” by A. Sen (1999) in which the author refers to democracy as one of the most important phenomena of the 20th century, a more philosophical paper “Natural Capacities and Democracy as Good-in-Itself” by J.Ober (2008), “How People View Democracy: Findings from Public Opinion Surveys in Four Regions” reporting responses from post-communist Europe, also Latin American countries, Africa and some Asian states, by L. Diamond (2003). A question regarding this issue arises: under what conditions democracy may fail as a
political regime, outperformed by autocratic state? There is a paper by W. Easterly, not yet completed, to our knowledge, called “Benevolent Autocrats” (2011) in which he attempts to explain this issue. The author refers to the fact that under autocracies there may be higher economic growth, but often associated with downturns, as there is greater volatility, unlike in
democracies. Democracy implies smoothening of adverse shocks, due to more flexible policies, moreover, autocracies may suffer from spontaneous decisions of the leader. It seems that the topic is quite controversial and worth looking into.
Hence, we try to answer the following question in our research: if two countries have equal levels of urbanization, same percentage of FDI inflows and, moreover, similar development indicators, will they experience different rates of growth in their income if they have different political regimes? If we do this, we therefore build linkages between institutions,
urbanization, foreign direct investment and political regime for the sample of approximately
96 countries over a period 1980-2010 which can be considered as a valuable contribution to the existing scope of literature. In the existing literature authors mostly focus on dividing countries by income level or level of development. We in our study go further and consider two sub-samples within our sample: democratic and autocratic countries, so we examine the difference the regime makes for the model specification exhausted with multiple explanatory variables.
Literature Overview
Numerous papers were written in relation to Foreign Direct Investment efficiency, throwing light upon its great importance for economic growth process. For several decades many researches strived to tackle this issue. A very accurate statement made regarding this fact is “Historically, there have been as many regression specifications as there have been empirical papers on the determinants of growth” (Brulhart, Sbergami 2009:3). In our research we want to account for different factors affecting FDI, most important of them being urbanization level, „primacy‟ which stands for „urban concentration‟ (Henderson 2003) and also institutional factors. Therefore, first, we are going to refer to the papers written on the relationship between FDI and economic growth, then move to multifactor models, models with institutional variables and, finally, papers on urbanization. This way we can demonstrat that our contribution is valuable and will help to compare growth levels in countries with different institutional conditions but similar levels of investment and urbanization.
FDI and growth
First, we address the papers written on the relationship between FDI and growth which include not only the analysis of the relationship itself, but what influence other factors have on the efficiency of FDI flows.
In their paper on the endogenous relationship between FDI and growth, X. Li and X. Liu (2005) provide detailed tests and stress the importance of interaction of foreign direct investment with other factors, rather than of pure FDI effects. The main conclusion drawn from the research is that while FDI to developing countries interacting with human capital yields strictly positive effect on economic growth, the opposite happens with technology gap. In other words, the empirical results help to show that both promoted human capital and technology-absorptive capability lead to more FDI inflows. “This in turn will promote further economic growth and enhance competitiveness” (Li and Liu 2005: 404-405). So there we got the endogeneity issue.
One of the valuable aspects of this work is the sample size estimated by the authors. They do
a study of 84 countries over a period 1970-1999 using panel data approach, in order to capture country-specific features. Moreover, the study is focused on econometric techniques. Tests
are applied to the relationship between FDI and economic growth and it is concluded that the relationship becomes more endogenous in more recent years. It is well known that endogeneity arises from causal interrelationship between variables, as it is given in Blundell
and Powell “… that is, observable explanatory variables that are correlated with unobservable error terms”*. These authors also list possible reasons for this phenomenon: measurement errors, selection bias, correlation in random effects and heterogeneity and, most suitable in case of FDI and growth endogenously related – “simultaneity” (Blundell and Powell 2004). “Simultaneity” would describe the state when we have a system of equations where the variables of interest influence each other, so that there is a correlation in error terms. Simultaneous equations for the data in the Li and Liu study:
*
* Sourse: “FDI and Economic Growth: An increasingly endogenous relationship”. (X. Li and
X. Liu, 2005, p. 396-397)
X consists of variables included as controls by the authors: inflation rate, interest rate, changes of foreign exchange rate, black market premium; we take into account some of these variables in our work.
As we can state, there is a high probability of coefficients being biased in these equations. Li and Liu run Durbin-Wu-Hausman test and find that for the whole sample period 1970-1999 the results are insignificant, so there seems to be no evidence in favor of endogeneity. However, for 1985-1999, there is such evidence and hence, authors provide estimation tailored for simultaneous equations, particularly, 3SLS or three stage least squares. For the remaining sub-sample over 1970-1985 the appropriate single equations are used.
From the comparison of developed and developing countries Li and Liu draw the following conclusions: FDI affects growth in a positive way in both types of counties. Moreover, there is an interesting result related to interaction of FDI variable with technological gap. It turned out that for developed economies which possess initially high technology-absorptive capability end to gain a lot from large technology gap together with FDI term. In turn, developing countries having relatively low capability lose in growth from large technology gap.
Unfortunately, in our work we are not going to focus on technology gap, however we make use of other variables introduced. Li and Liu refer to political instability as possible factor and
measure it as the number of riots in a country for a given year, we, in turn, try to use our indicator for political stability taken from World Bank Governance Indicators database.
Regarding the results of the research by Li and Liu, we examine the estimation process in more detail. For instance, for single equations used for 1970-1985 the authors use interactions terms from the equation, FDI*Schooling, FDI*Technology gap and FDI* Telephone lines (measuring infrastructure) and for developing countries all three interaction terms yield significance on 5% level. These results are for the separate samples of developed and developing economies. We also want to do different country types separately, however, use a different classification. In simultaneous equations estimation, which we are not going to do in our work, initial logGDP becomes insignificant, while other variables maintain their value for the growth equation. Despite the fact that Telephone lines term as it is does not prove to be very significant (only for GDP growth, at 10% level), we will try to use it in our equation. In case we do not get anything significant from this measure, we will try to replace it with internet usage and also road density as compared to the land area estimate.
Li and Liu (2005), as it can be understood from the title of their work, also prove that growth itself affects FDI inflows, raising the question of endogeneity. The results they obtain checking for the reverse relationship yield statistically significant coefficients for GDP growth itself, also logarithm of GDP (initial) and Trade as a measure of openness for both types of countries considered, all on 1% of 5% levels of significance. Due to the fact that reverse causality phenomenon is proved in several papers, we concentrate on the one-way relationship, FDI affecting Growth and on the addition of specific terms to increase the explanatory power of the model.
There are many papers showing how important it is to pay attention to country classifications. One of the examples is the paper “Inappropriate pooling of wealthy and poor countries in empirical FDI studies” (2005) by B. Blonigen and M. Wang* with the title speaking for itself.
The central idea of the paper is to check whether there is systematic difference in effects of FDI for developed (DC) and less developed countries (LDC). Hence, in their critical literature overview the authors concentrate specifically on the types of studies which fail to classify countries in their samples.
It is not that none of the literature preceding this work in 2004 avoids checking for regional and country-specific features in estimated samples, but it is rather the assumption proposed by the authors. According to the results of the research, countries indeed need to be separated
into categories in order for us to avoid biased estimates. Also stating that pooling countries
together in one sample is no good is Kemeny (2010): “Among poor countries the effect of FDI on upgrading is bolstered for those endowed with higher levels of social capability. The effect of FDI on upgrading in rich countries remains positive but is weaker, and social capability exerts little disparate influence among these similarly socially capable economies.”* Hence, it is crucial not to mix countries all in one big sample, as the significance of important interactions may simply disappear, become hidden.
Institutional variables
We can not avoid the topic of the effects of institutional factors on economic growth. Many papers attempted to examine the possibility of including multiple factors, not only institutional measures, into their growth equation, so that not only FDI term is the variable of interest. The paper of J. A. Batten and X.V. Vo simply called “An analysis of the relationship
between foreign direct investment and economic growth”* contains a research quite similar to the one we are going to do in our study. The main objective of their research is to evaluate the effects of different institutional conditions and hence, corresponding social policies, on the effectiveness of FDI regarding economic growth.
Batten and Vo examine a broad sample of 79 countries over a period from 1980 to 2003. In this sense, our research will be not of a very great contribution to the literature in terms of period length. Moreover, the paper by Batten and Vo concentrates on the interaction issue specifically, in order to differ from the scope of preceding articles. Differently from other papers, they propose four different measures for FDI, two stock and two flow variables, to capture all possible effects. The equation used as a basic specification is similar to the one in Li and Liu (2005) and Barro (2004)*:
*
* Sourse: “An analysis of the relationship between foreign direct investment and economic growth”, J. A. Batten and X.V. Vo, 2009:1625
Variables considered “frequently used”, or, most likely to affect growth according to previous studies are: ratio of government expenditure to GDP (size of government), percent of trade in GDP (openness), annual inflation, ratio of domestic credit to GDP (financial system stability), size of stock market, index for international country risk. These variables are pooled together as X with beta coefficient. I comprises of the variables, typically used in growth
equations: lagged initial GDP, percentage of domestic investment of GDP, secondary school enrollment rate, population growth rate etc.
The main implications, also useful for our research consist of general statement that there is a great effect of FDI on growth, also arising from complex interactions with other factors. Higher levels of education, greater openness to trade, well-developed stock market and lower population growth and risks tend to add to the benefits from FDI inflow, according to the study. Therefore, in our work will refer to the variables from Batten and Vo‟s research, as there is sufficient evidence from their empirical estimations. However, this paper fails to account for differences between developed and developing countries or any differences at all and pools all countries in the sample together.
There is no doubt that institutions affect growth as well as investment flows. As we see in papers previously mentioned, often FDI turns out much more effective if there is an interaction with another important term. Demekas (2007) argues that “strong legal and political system is required due to huge sunk costs of FDI”. Huge sunk costs associated with FDI arise as local companies and investors should be able to invest in absorbing foreign techniques and knowledge, as it is given in Alfaro et al (2010), knowledge spillovers can occur successfully only if firms are able to absorb it with the help of strong local financial markets. In fact, Alfaro et al (2010) and also Hermes and Lensink (2003) and Alfaro (2004) have papers written specifically on the importance of strong financial markets in the process of FDI – growth.
Crucial finding in the mentioned works on institutions is that well functioning financial market in a host economy positively influences the impact of FDI inflows. Alfaro in her work of 2004 shows that interaction term of FDI and financial market development is positive and significant on a large sample of countries. Also the author adds regional dummies, which helps the model gain explanatory power. Moreover, in order to find approximation for measuring financial market development, because it is not that straightforward, the author refers to articles in which the instruments are legal protection of investors and similar indicators. Then she uses these proxies in the specification and comes to an important conclusion: badly developed financial markets put actual limitations on the positive effectiveness of FDI.
In the article by Hermes and Lensink (2003), there is very similar research, tackling the issue of the role of financial markets. The authors give a fine explanation to why markets are so important – financial system itself helps to allocate resources more efficiently and thus, the
ability to receive positive effects from spillovers increases which makes FDI efficient, according to the reasoning. As a conclusion, authors make deeper policy propositions than those implied by Alfaro (2004) – first, the government should develop the financial system and only after it is done, open the economy to capital inflows.
Urbanization and Agglomeration
What we want to consider next, are the concepts of agglomeration and urbanization, important to our study. The research in the field of urbanization has a long and rich history. One of the examples is the work by DeLong and Shleifer (1993). They look far back in the past, times prior to industrial revolution. Despite the fact that the article is rather old, the
paper attempts to answer the question we deal with in our work: how urbanization and current regime are related to economic growth. In their research DeLong and Shleifer refer to the
sizes of Europe‟s largest cities as economic well-being of the regions. This means that in fact they link the concepts of urbanization and economic development really closely. Moreover, back in those days European largest cities were actual agglomerations, as they were centers of commerce and political power and, what is most important, they were very visible (thus, easy to identify) and growth in population in such countries strongly influenced overall economic growth.
The authors also introduce a classification of regimes, or, types of government for the large time period they considered (19 – 20th centuries). The classification is as follows: absolutist or “other”, for which it was true that “…the prince is bound by law”, “A second example … was city-state based rule by merchant oligarchies” (p. 680). Finally, “feudal” government is the third type of rule (DeLong, Shleifer 1993).
Urbanization measures used in the study are: number of cities larger than some N number(different for different data sets) and population growth in these cities. The principal independent variable included in the equation stands for absolutism, or the opposite. Controls, added to the main equation: for region and era; and with regional controls the authors
achieved far better results. The results yield increase in R-squared from 0.48 to 0.7 and from
0.36 to 0.54 for different periods considered. Such outcome is not surprising at all, as regions are indeed different, geographically, by land area, or due to some random occurrence. To overcome such effects of differences, the authors checks proportional growth instead of simple growth.
We would like to mention a paper that builds a framework in which growth and agglomeration are considered as „mutually self-reinforcing processes‟: P. Martin and G. I.P. Ottaviano, “Growth and Agglomeration” (2001). The authors provide evidence in favor of their argument: economic growth and agglomeration influence one another and the effects are measurable. To start with, we need a definition for agglomeration process. Basically, it is characterized as clustering, in urban terms, a cluster built up around some central point, usually main city. Referring to the paper by G. I. P. Ottaviano and D. Puga (1997) it can be described first at a smallest scale – when firms tend to come together in clusters for economic reasons, their sector being apparent. For larger scale agglomerations, often spread across boarders and being not simply firms, but cities, the explanation for clustering lies not only in technological spillovers, but in the market features and the aspects of infrastructure.
The authors‟ argument on the causal interrelationship is based on simple economic reasons. First, agglomeration affects growth in a positive way due to lower costs of research, development and innovation, also lower transaction costs. On the other hand, increasing growth in one of the regions affects the decisions of firms to relocate their production to this, more developed area. What is more, the relationship is not that simple, because as agglomeration increases in its level, with new incoming firms, productive incumbents tend to move their production „outwards‟, to the periphery. Hence, we need to remember that adding a measure for agglomeration or a proxy for it in our equation can lead us to confusing results. We need to pay attention to that with continuing economic growth agglomeration intensity is likely first to increase and then gradually fall to a certain level.
In his paper “Magnitude and causes of agglomeration” (2009) Diego Puga mentions several causes of agglomeration, the main one being the presence of „larger markets‟.* Larger markets, according to D. Puga imply better learning, infrastructure, better interaction, matching (meaning employers with employees), thicker labor market, improved individual specialization. From this we can infer that indeed, economically developed regions tend to attract clustering. Also the author proposes ways to quantify the agglomeration process: by comparing wages in different regions, one of them being an agglomeration. As well as looking at wages, one can also compare rents paid by companies, the higher are rents – the
more likely is that the company compensates this with higher gains from efficient production.
“Do economies grow faster if they are concentrated in space?”. This is the cetral question M. Brulhart and F. Sbergami attempt to answer in “Agglomeration and Growth: Cross-country evidence” (2008). What is more, the authors also take into account that level of development,
as well as degree of openness, affect the FDI-growth nexus. In addition to the estimation of aggregate economic growth relationship with investment, there is also sectoral division of the sample. Brulhart and Sbergami argue that there is little probability that across all possible sectors the effect of agglomeration is the same.
The study follows the pattern of some of the previous papers (Barro and Sala-i-Martin (2004), Henderson (2003)), adding variables to the growth equation, indeed, authors use the estimators used in both Henderson (2003) and Barro et al (2004) specifications. The GMM approach and cross-country OLS estimation are applied to the analysis and as a result, authors’ general conclusion is that agglomeration positively affects growth, but up to a particular level only. Moreover, authors give an estimate of approximately 10 000 USD GDP/capita as a limit for GDP and its growth rate. What is most striking in the conclusion is that at a large, country scale, agglomeration can turn out detrimental, instead of efficient in terms of growth, and the suggestion hence is, to try to account for smaller scale, firm level, estimates. “Local clustering economies—which exist below the radar of this study—may be
as strong as ever in developed as well as in developing economies” (Brulhart, Sbergami 2008:
60).
In our work we are not going to look on wages or rents to quantify the effects of agglomeration. Also we are not going to consider small scale clusters of companies, although these measures can be more appropriate than simply taking urbanization and „primacy‟ as a proxy and adding them to the equation. Our choice of „primacy‟ variable can be supported by D. Puga (2009) who refers to previous works, stating that workers and firms themselves can increase their productivity functioning closer to the urban clusters hence, agglomeration levels can in fact be somehow accounted for with the use of „primacy‟ term, concentration of urban population. Urbanization level, in turn, seems less appropriate for adding it as agglomeration estimate, but we add it despite of this fact (see section on Methodology (Urbanization) for more details).
Data Description
Like in most papers related to the topic of economic growth affected by multiple factors, we first refer to the famous Barro and Sala-i-Martin framework*. (R.J. Barro and X. Sala-i- Martin: “Economic Growth”, 2-d edition, 2004) To obtain the variables that form the equation, similar to the one proposed by Barro and Sala-i-Martin, we first choose World Bank World Development Indicators database. For a time period covering 1980-2010 for our sample of 153 countries we obtain the following variables to form a basic regression equation*:
gdp_growth – Annual percentage growth rate of GDP per capita (4492 obs.)
gdp_const_lcu – GDP in terms of constant local currency (4531 obs.)
trade_percent – Trade, as percentage of GDP = sum of Exports and Imports divided by GDP
value (in current US$) (4422 obs.)
In his paper on determinants of FDI*, M. Al Nasser (2007) introduces Current Account/GDP term taken with a lag instead of trade in order to account for openness of the economy, so we keep this in mind in case we find that trade to GDP ratio is not sufficient:
ca_percent – Current account balance (% of GDP)
Other important variables can be found in the Appendix (List of Variables).
For measuring domestic investment level as compared to GDP we refer to IMF database
(www.imf.org), World Economic Outlook Report and Database for our time period 1980-
2010 and get the variable
invest – Investment as percentage of GDP (4262 obs.)
Even at this stage of analyzing available data, we observe that numbers of observations for basic variables we are going to include are not the same. This means that we are most likely to face missing values in our dataset and hence, provide appropriate estimation procedures.
Barro et al (2004) also make use of the variable for average years of schooling for each country:
bl_asyt25 – Average years of schooling for population aged over 25 years; we specifically consider this variable and transform it as the measurements were made every five years from
1945 and so we could not merge the Barro database for this variable with our dataset of indicators.
In addition, there are variables used by Barro and Sala-i-Martin such as Rule of Law indicator, measuring the strength of a country‟s legal system together with country‟s political risks. Moreover, the authors attempt to account for the role of the regime, introducing such
variables as Democracy and Democracy-squared. We add similar indicators to our model, but a little later in our work. In order to do this properly, we choose another data base – Polity IV introduced by Marshall and Jaggers in 2002 and currently available for years up to 2010. The indicators we are going to use are the following:
p_democ – or, Institutionalized Democracy, available as both time-series and cross-section; varies on the scale 0-10, “derived from coding of the competitiveness of political participation (variable p_parcomp), the openness and competitiveness of executive recruitment (variables p_xropen and p_xrcomp), and constraints on the chief executive (variable p_xconst)”. (QoG Codebook 2011, p.56)
p_autoc – or, Institutionalized Autocracy, available as both time-series and cross-section;
similar to p_democ in its nature being a complete opposite. Indicators are on the scale of 0 to
10, where 10 is most autocratic; “In mature form, autocracies sharply restrict or suppress competitive political participation”. (QoG Codebook 2011, p.57).
p_polity2 – or, Revised Combined Polity Score, available as both time-series and cross- section; equals the difference (p_democ – p_autoc) and varies from -10 to +10, where +10 is strongly autocratic.
Moreover, we also attempt to introduce a measure of political constraints to our equation. Equivalent estimate was added to the growth equation introduced by E. L. Glaeser, R. La Porta, F. Lopez-de-Silanes and A. Shleifer in their work “Do Institutions Cause Growth?”*. This indicator in presented as Political Constraints Index, or POLCON, created by Henisz in
2000. In simple words, the index shows whether any policy change is possible under existing government, or, in other words, measures the feasibility of policy alterations. We get
h_polcon5 – Political Constraints index, varying from 0 to 1, where 1 is most constrained.
As we are mostly interested in Foreign Direct Investment and its effects on growth in host countries, we definitely need to add
fdi_percent – Foreign direct investment inflow as percentage of GDP (4274 obs.)
This variable is not mentioned within Barro and Sala-i-Martin (2004) framework, but it is present in the research done by X. Li and X. Liu (2005) which we are now familiar with. The authors stressed the fact that FDI affects growth in both types of counties, developing and developed.
Variables chosen as measurements for infrastructure are again taken from World
Development Indicators data base and are presented in the Appendix.
The variables important to us in this study are those that measure urbanization levels and
“primacy”.
urban_tot – Total number for urban population
urban_percent – Percentage of urban population of Total population
urban_growth – Annual percentage growth rate of urban population
pop_in_large – Population in largest city, or, an alternative, as well as measure of “primacy”: pop_in_mil_percent – Percentage living in urban agglomerations of more than 1 mil population; also can act as a measure for „primacy‟;
Finally, we should not forget about the institutional aspect. Apart from indicators for the regime and political constraints, we as well include variables available from World Bank Governance Indicators by Kaufmann et al (2009). These variables are chosen to indicate the perception of the government. The whole dataset we use is quite complex in its origins (31 separate datasets from 25 different organizational bodies).
Note: we use “estimates” as variables to choose, so the variable names end in “e”; All values
vary from -2.5 to 2.5, estimates are standardized. For indicators listed, again, see Appendix. After deciding on what data we want to consider in our study, we understand that moving to
the composition of our equation may be followed by certain problems. First of all, as our approach implies trying different variables added as explanatory, in panel data analysis there is often an issue of omitted variables. This problem arises from the possibility that a variable correlated with the one or ones included in the estimated equation is excluded (M. Verbeek,
2004)*.
Secondly, the large number of variables implies possible correlation between them and hence, due to multicollinearity issue, possibly biased estimates resulting from our regression.
Methodology
We will describe the practical part of our work in accordance with step-by-step changes done to model specifications. An important thing to keep in mind, while changing equations, is that as we base our choices on previous works, we may easily face a problem of “publication bias” (S. Kolenikov 2000). In simple words, most authors choose to publish their findings only in cases when they find an interesting, or at least, significant relationship between variables.
Even if they have other specifications, we usually can not know the details, as they are not presented in their research papers. Hence, we definitely include those variables which are proved to affect growth, but we also try to look at the variables mentioned by the authors, but which on practice prove insignificant.
Before we go into any detail on model specifications, we first check cross-correlations between the key variables. For the variable for GDP growth we can observe that variables most correlated with growth are: lending (0.192), telephone (0.343), last year‟s GDP term (0.23), real interest rate (-0.484), percentage of urban population (0.2), h_polcon5 for political constraints (0.189), schooling (0.253), and the set of institutional variables: Corruption index (wbgi_cce) (0.29), Rule of Law (wbgi_rle) (0.33), Political Stability (wbgi_pse) (0.26),
Regulatory Quality (wbgi_rqe) (0.35). Hence, from this we can think of a concept of our final, most exhausted, specification before we use the basic options studied by other authors.
Basic specification
In the section devoted to data sources we stated that the base equation we first take into consideration is the basic ones proposed by Barro and Sala-i-Martin in 2004 and by Li and Liu in 2005 already mentioned in our overview. For obtaining all the necessary data we first
look at the variables we get from WDI data set. The principal variables for our equation, GDP growth rate, population growth, average years of schooling etc. are all present in the data set, all we needed to do is to deflate the variables which are not given as percentages, so that the variables are appropriate for modeling. From now on we will assume that the necessary deflation procedures are made for certain variables. Our initial specification, therefore is:
(1). GROWTH = C + β1ln(GDP(-1)) + β2INVEST(-1) + β3POP + β4SCHOOL(-1) + β5OPEN(-1) + ε,
where ln(GDP(-1)) is lagged GDP in absolute terms taken with logarithm, INVEST(-1) is
lagged investment as percentage of GDP, POP is the rate of population growth which can also be taken with a lag, if there are problems with significance. SCHOOL is the average years of schooling for the population older than 25 years (lagged) and OPEN is the measure of openness for which we take Trade as percentage of GDP (lagged). Variables lnGDP, INVEST are taken with a lag for a simple reason of possible endogeneity, as investment is itself a part of aggregate demand and GDP is obviously related to GDP growth. Other variables we lag simply to be cautious, although there may be no obvious reverse causation.
Our regression equations are estimated using simple OLS (Ordinary Least Squares) approach. Before commenting on the results for this specification we run the regressions, with both
fixed effects (FE) and random effects (RE) options. The choice of these characteristics originates from how we treat individual effects for the items in our sample. Deciding on whether to use FE or RE for the regression depends on the results of the Hausman test. The zero hypothesis for the Hausman test implies the indifference between random or fixed effects estimation. We apply the test after running the regressions and get the p-value of 0.000 and very high χ2 of 71.1, therefore, there is significant evidence that we should reject H0 and
decide on choosing FE over RE. Fixed effects implies that we are going to have “within
effects” or “fixed effects” estimators (M. Verbeek, 2004, p. 346).
In addition to choosing FE, we apply „robust‟ option to the regression estimation so that we attempt to overcome the problem of heteroscedasticity in error variances. The results of the regression estimation prove that for the basic model with no FDI added, the variables for lagged GDP, openness and schooling appear to be significant on 1% level (see Table 1). Also, the results are consistent with economic intuition. Similar to the results from Batten and Vo (2010), there is a negative and significant relationship between lagged GDP per capita (estimate for β1 = -2.072 with s.e. (0.352)) and GDP growth, positive relationship with Trade – lagged measure for openness (trade/GDP) increasing by 1% causes an increase in growth by
0.035% on average, so openness indeed positively affects growth. Also schooling corresponds to expectations, as we have estimate for β4 = 1.134 with s.e. (0.293), hence years of schooling definitely add to the benefit of the economy, creating more sophisticated labor market. The R- squared is very low (6.4%) in this specification.
After considering the basic model, we add FDI measure to the equation, first, in a simple form: FDI as percentage of GDP taken with a lag
(2). GROWTH = C + β1ln(GDP(-1)) + β2INVEST(-1) + β3POP + β4SCHOOL(-1) + β5OPEN(-1) + β6FDI(-1) +
ε
We add the measure for FDI taken with a lag, as the relationship between GDP and FDI is proved to be endogenous, especially for the years after 1985 (Li, Liu 2005), however, simple lag is probably not the best instrument variable we can use. In fact, there are multiple papers that attempt to introduce instruments for FDI estimation: simple lagged FDI as in Alguacil, Cuadro and Orts (2011), spatially weighted measures of FDI determinants, as in Poelhekke and Van der Ploeg (2008).
The estimation procedure of the model with added FDI is similar to the previous one (see
Table 1). On 5% level the coefficient for FDI(-1) term is positive and significant (estimated β6
= 0.0938 with s.e. (0.0421)), also the variables from the basic specification remain significant, all on 1% level. R-squared is still very low, but higher than in initial specification (6.9%). Hence, on this stage we can conclude that for our sample of 96 countries FDI positively
affects economic growth and the effect is significant.
For further specifications we will assume specification (2) as our base model and from now on regard the set of explanatory variables from GROWTH = C + β1ln(GDP(-1)) +
β2INVEST(-1) + β3POP + β4SCHOOL + β5OPEN + β6FDI(-1) + ε as a set of control variables
we are going to add to every specification we try.
Infrastructure
Here we proceed to exhausting or model with variables that appear to have some correlation with economic growth. First, we try to add lagged INFLATION as an explanatory variable. As a result we get FDI coefficient still significant on 5% level, positively affecting growth. No changes in significance are present for our controls. However, the inflation term appears insignificant, with negative coefficient. We try using real interest rate, which is of very low correlation with inflation, but we again get no significance, which is unfortunate, as real rate is well correlated with growth.
As in Li and Liu (2005) they found that infrastructure may contribute to the growth equation, we want to check for this fact. We have three possible variables which can act as proxies for infrastructure: TELEPHONE(-1), as number of telephone lines per 100 people (lagged) and INTERNET(-1)- Internet users per 100 people (lagged). Also we can add road density measure and for this we specially create a variable ROAD being a ratio of total road length in km divided by the land area, plus, we create a lagged variable ROAD(-1). After running three
specifications with each of these infrastructure proxies (see Table 2), we come to a conclusion that only TELEPHONE as infrastructural measure can actually contribute to our model, as R- squared increases to 7.1%. Good infrastructure positively affects growth, as the coefficient
estimate for TELEPHONE is 0.0462 with s.e. (0.0249), and hence, positive significant on
10% level. To mention here, in Li and Liu research the same infrastructure proxy proved insignificant on all levels. As proportion of telephone lines per 100 people rises by 1%, growth increases by 0.046% on average. Moreover, this variable does not harm to the significance of control variable coefficients. FDI coefficient is significant on 5% level, which is good for us and the estimate is 0.0936 with s.e. (0.0418), hence, if FDI increases by 1%, GDP growth will on average increase by 0.094%. As a result, we can add a new variable to our model and get specification:
(3). GROWTH = C + β1ln(GDP(-1)) + β2INVEST(-1) + β3POP + β4SCHOOL(-1) + β5OPEN(-1) + β6FDI(-1) +
+ β7 TEL(-1) + ε
Urbanization
After coming up with a specification (3) which replaces (2) as our main specification we turn to the issue of urbanization. In the Literature Overview we defined urbanization and agglomeration, the latter being a big issue nowadays. In our data set, we have measures for urbanization, both in growth and in percentage of total population. Moreover, we have measures for „primacy‟, or urban concentration level. Primacy, according to Henderson (2003), can be defined as “the share of the largest city in national urban population” (Henderson 2003: 51). So hence, simple urbanization is more general than primacy, as it
accounts for any city in the country, while primacy refers to the „largest‟ cities. Hence, we can use our variable called pop_in_mil_percent (population in urban agglomeration of > 1 mil people, percent of total urban population) as a possible proxy for agglomeration. Urbanization is not that good a proxy for agglomeration, according to this logic, but nevertheless, if we recall the assumptions in the paper by Shleifer et al(1993), it is possible to look at its effects and what is more, introduce interaction terms.
First of all, the following specification is done:
(4). GROWTH = C + β1ln(GDP(-1)) + β2INVEST(-1) + β3POP + β4SCHOOL(-1) + β5OPEN(-1) + β6FDI(-1) +
+ β7 INFL(-1) + β8TEL(-1) + β9URBAN(-1) + ε,
where URBAN(-1) stands for lagged percentage of urban population of total population for a given year. For this model specification the coefficient estimates for regressors we used in (3) partly lose their significance (see Table 3), average schooling (SCHOOL) and TELEPHONE
terms become insignificant, while trade (OPEN) measure loses significance, but still is significant on 5% level. Only the lagged GDP term is significant of 1% level, as before (estimate for β1 = -2.479 with s.e. (0.0357)). We observe loss of significance for FDI term (10%), for trade (openness) (now 5%). Urbanization itself, being included into the equation as it is proves significant on 10% level.
Basing on the literature, for instance, the article by Poelhekke and Van der Ploeg (2008), we can transform urbanization term, just as the authors transform primacy measure. So we use urbanization level squared, urbanization growth, moreover, interaction terms with FDI, in order to check whether there is marginal positive effect of urbanization on FDI and hence, additional effect on growth. If we replace our usual control variables lnGDP(-1), investment, population growth, schooling, openness, inflation FDI itself and telephone with set of all these variables X and estimate the following, we need to choose the best of:
(5). GROWTH = C + β1X + β2URBAN(-1) + ε, GROWTH = C + β1X + β2 URBAN(-1)2 + ε, GROWTH = C + β1X + β2URBANGROWTH(-1) + ε,
GROWTH = C + β1X + β2URBAN(-1) + β3FDI(-1)*URBAN(-1) + ε,
GROWTH = C + β1X + β2URBAN(-1)2 + β3FDI(-1)*URBAN(-1) + ε
As a result (see Table 3.3.1 and Table 3.3.2 for detailed results), we see that the best specification due to its R-quared and, more importantly, its coefficient estimate for included variable and FDI term is number (11) (Table 3.3.2) with URBAN(-1) squared and interaction term of FDI and urbanization. R-squared in this specification is higher than before – 8.6%. We see that FDI coefficient estimate is significant on 1% level, as interaction term with urbanization. However, while for pure FDI, if it increases by 1%, growth will on average increase by 0.3%, with interaction term, the marginal effect of urbanization is negative, estimate is -0.0035. This can mean that while urbanization, specifically, squared percentage, positively affects growth, it creates negative effect for FDI. Here we can also recall the argument presented in Brulhart and Sbergami (2008) on the falling importance of agglomeration on a large scale. In other words, the more is the level of urbanization, the less efficient are FDI inflows for growth. This can be viewed as a marginal element of productivity, which, if we recall microeconomic framework, is first increasing and then decreasing after certain level. But still, overall FDI has positive effect on economic growth in our chosen specification.
Primacy
Similar approach as in case of urbanization can be used to analyze the effects of primacy on growth and its interaction with FDI. Without referring to the particular specifications, as they are equivalent to those for urbanization (see (5).), only with no term for growth in primacy, we obtain the results shown in Table 3.4. The best specification with primacy variables is number (13), with primacy itself and interaction of FDI with primacy term. For this model specification FDI coefficient is significant on 10% level, estimate being positive equal to
0.142 (0.083) and the interaction term with primacy yields beta estimated -0.0027 (0.018), unfortunately, insignificant even at 10% level. In fact, if the negative coefficient were significant, we would apply the same logic as in case of urban population percentage. The following items: trade (openness), lnGDP(-1), population growth, and primacy itself are significant at 1% level which can be considered an improvement compared to specification we chose for urbanization. All coefficients have expected signs. Also model can possibly be considered better than before due to R-squared now equal to 10.3%.
As an outcome of regression estimations of model specifications in sections related to urbanization and primacy, we got two specifications: for urbanization number (11) and for primacy number (13), both showing improvements over models with no such variables added, also reasonable significance of coefficients of our interest. As we state in our research, both variables can be viewed as proxies for agglomeration which we could not include due to data insufficiency and measurement issues. We use these two measures separately in different equations as they are strongly correlated (ro = 0.56, from cross-correlation matrix we calculated).
Institutions
In this sub-section we need to add the institutional variables to both specifications we obtained in previous sub-section on agglomeration. Our prediction is that with better institutions the rate of economic development (growth) will be enlarged. And from the cross- correlation matrix we examine at the beginning of the Methodology section, we get that we can take icrg_qog, wbgi_rle, wbgi_rqe and wbgi_cce variables separately due to their obviously high correlation with growth term. Also, despite the fact that all institutional variables are strongly correlated with each other, we are extremely cautions with taking some
of them together. We therefore try some of the indicators in pairs, if their correlation is around
50% or less.
The results for the specification with urbanization and interaction term (number (11)) with added institutional terms are presented in Table 4. We focus only on those of the 7 specifications that have estimated coefficients for FDI inflow percentage are significant at least at 10% level. We have equations with numbers (15), (19), (20) and (21) to compare. The level of significance for FDI term is the same for the latter three – 10% which is worse than we had before, however, the level is 5% for the (15) equation. R-squared measures, although
it is not what we have to look at in studying the relationships between variables, still, for equations (20) and (15), equal to 14.6% and 9.8% respectively, now similar to the values obtained in Li and Liu research estimations, for instance, as they on average got specifications with R-squared of about 12 – 14%. We are interested most in the effects of FDI, we better
look closer at (15) specification. The coefficient estimate is 0.245 (0.103) at 5% level, FDI interaction with urbanization is negative, as in specification with no institutional variable, estimated -0.0026 (0.0013) at 10% level of significance. Due to the fact that for pure FDI coefficient is almost ten times greater than for interaction variable, FDI still has positive effect on growth while the marginal effect of urbanization on FDI is negative. Of our control variables from the rest of the set X, we have estimates of coefficients for population growth, urban percentage squared and institutional variable, ICRG Quality of Government index all significant on 10% level and having appropriate signs. For the estimates for lhGDP(-1) and trade the levels are 1% and 5% respectively.
At this stage we have a specification, satisfying our needs:
(6). GROWTH = C + β1X + β2URBAN(-1)2 + β3FDI(-1)*URBAN(-1) + β4 ICRG + ε
For this specification to be more proper, we correct ourselves and also add the term URBAN
with no squared term, hence, for the results see Table 4.1.
(7). GROWTH = C + β1X + β2URBAN(-1)2 + β3URBAN(-1) + β4FDI(-1)*URBAN(-1) + β5 ICRG + ε
Unfortunately, the result is not promising, probably, due to the fact that squared term adds complexity to the estimated model. Overall, we do not get many significant coefficients, only lnGDP(-1), population growth and trade coefficients have 1% level of significance for the estimates, ICRG term estimate is significant on 10%, other variables carry insignificant coefficients. Hence, instead of using squared term in our estimations, we better choose a model, with probably less significant coefficient for FDI proportion of GDP term, and that is
the specification from Table 3.3.1, with added urbanization term and interaction FDI*Urban. term.
Now we move to the primacy variable and specifications for it. Running similar regressions, as we do for urbanization, we get the results available in Table 5. We decide on choosing model specification number (22), as it yields significant on 5% FDI coefficient estimate, however, not significant interaction term with primacy which is negative, as before. For every FDI increasing by 1%, economic growth rate will rise by 0.15% on average, other things equal, insignificantly affected by increasing primacy measure multiplied by FDI percentage. lnGDP(-1), population growth, trade percentage and schooling coefficients are all significant on 1% or 5% levels. Unfortunately, the coefficient for country risk and quality of government measure (ICRG_QOG) is significant only on 20% level (t-test for coefficient yields 1.34, p- value 0.184). All in all, for primacy we end up with the specification (22) which is:
(8). GROWTH = C + β1X + β2PRIMACY(-1) + β3FDI(-1)*PRIMACY(-1) + β4 ICRG + ε.
Separate samples estimation
We now need to test specifications we decided on in previous sub sections which include urbanization/ /primacy, and interaction terms and institutional variable ICRG_QOG. We estimate samples divided according to the level of democracy/autocracy, so we use p_polity2 indicator, and classify our countries with more democratic (p_polity2 > 0) and more autocratic (p_polity2 <=0) regimes using the scale suggested by Marshall and Jaggers 2002, on which the values vary from -10 to +10, where +10 is strongly democratic. As we run two separate regressions, we obtain the results presented in Table 6.
We at first consider the Democracy Sample and simply looking at models (1) and two which deal with the effects of primacy on growth and FDI, we see that none of the coefficients we are interested in are close to significance which means we can immediately move to the urbanization specifications. For the model with added urbanization and interaction term, the coefficient estimate for FDI and interaction term are insignificant, moreover, with no interaction, urbanization level still does not significantly affect growth, nor does FDI inflow percentage. So, overall we can conclude that for Democracy Sample, FDI does not significantly affect growth as much as inflation, trade and population growth do. If we simply look on average value of GDP in nominal terms, for Democratic countries this value is almost three times larger than that of Autocratic states, same is true for minimum values (see Figure
1). Hence, we can roughly assume that there are more economically developed economies in
the Democratic sample and hence, such economies probably do not rely that much on foreign investment and consequent spillovers as less developed countries do. Moreover, we can refer to the argument presented by D. Puga (2009) in his article on agglomeration, that as in developed countries agglomerations build up great infrastructure, with greater infrastructure and less transportation costs we may infer that the idea of agglomeration becomes less relevant.
If we turn to Autocracy Sample, the picture becomes slightly different in the case of specification with urban population percentage. Looking at model in which we add single urbanization term without interaction with FDI, we only get positive and significant coefficient for urbanization itself, estimated coefficient is 0.247 (0.0786), this can be explained with the plausible tendency to move closer to cities in autocratic states, due to privileges, infrastructure reasons and for the sake of being closer to the bureaucratic bodies. Hence, urbanization adds positive and significant effect to economic growth, causing average
0.25% increase in economic growth if it increases by 1% other things equal. Most interesting for us are the results in specifications with primacy term for Autocracy Sample. In first specification, with no interaction, both estimates, for FDI term itself and primacy alone have significant on 1% level coefficients which can be interpreted as if primacy level increases by
1% affect growth leading to average growth rate increase by 0.279%. We can note that this effect is greater than one we see with urbanization. On the other hand, we have another specification, with FDI*primacy term and it yields positive and significant effects of both FDI and primacy, on 1% and 5% levels respectively, moreover, there is an interaction term coefficient which is small and negative, significant on 5% level. Hence, we can say that in Autocratic Sample of countries, states with large urban concentration (agglomeration) have weaker effects of FDI on economic growth, due to interaction with coefficient -0.00628 (0.0024). Nevertheless, the overall FDI effect is positive as the coefficient for pure FDI is almost 90 times greater than negative interaction coefficient in absolute terms. Regarding
other variables, both models yield significant estimates for logGDP(-1), population growth and inflation, almost all with expected signs except inflation term, having very little but positive effect on growth, unlike in previous models.
We observe that comparing the effects of FDI and primacy on economic growth in our two samples, we get that on average, similar increases in FDI cause much greater and, most importantly, more significant increases in GDP growth in Autocratic Sample, compared to Democratic. Moreover, rising primacy levels have double effects: on one hand, with no interaction taken into account, primacy also positively adds to growth, almost as much as FDI
inflows do; on the other hand, with large agglomeration levels, more primacy may harm FDI
effectiveness at a margin.
Results
After consistent and detailed description of our estimations we can proceed to the main implications which can be drawn from our study. First, we check basic specifications combining models from Barro et al (2004) and Liu and Lee (2005) in order to be sure that the model will work on our initial sample of 95-96 countries for the period from 1980 to 2010. After doing this we augment the specification with additional explanatory variables that are proved to cause significant changes in GDP growth. Finally we consider institutions to pick one or a few most significant institutional parameters.
In the final, and main, part of our analysis we separate countries, and not by their level of development, or income, like in preceding literature. We classify them by the index corresponding to the trend in the political regime. We have around 61-69 countries in our Democracy Sample and 33-36 in Autocracy Sample, due to data availability issue.
The estimates for coefficients for control variables we obtained are consistent with our expectations: negative coefficients for previous level of GDP, as we take GDP growth as percentage, and if the previous level is large, this percentage change can seem very small. Also, negative coefficient for population growth, as we take GDP per capita as growth measure, with greater population, the per capita transformation results in lower growth. Trade, as measure of openness of the economy is a positive driver for the economy, as it has been proved in previous papers as well. The effects of FDI and urbanization/primacy as proxies for agglomeration are not similar for different model specifications. What is more, they differ in our final estimation of two separate samples. We proved that for our large sample of
countries, for autocracies both urbanization and primacy play a very important role, taken in interaction with FDI, or separate. In democracies, in tern, neither FDI inflows, nor urbanization/primacy effects are insignificant for economic development.
Conclusion
All in all, is democracy the best option for the countries with unstable political situation? We try to find an answer to this question on economic level. We use econometric techniques to estimate the effects of foreign direct investment and other important factors on the sample of about 96 countries divided into two samples, autocratic and democratic. Despite the fact that on average, democratic state experience greater economic growth, autocracies show greater range between the minimum and the maximum values of GDP growth. This illustrates the argument by W. Easterly (2011) on greater volatility in income for dictatorship countries. We prove that FDI can be more efficient in autocracies in terms of growth, therefore, investing
into autocratic states, or ones close to this condition, can yield large benefit in terms of growth and overall productivity. However, we can suggest investing into autocracies very cautiously. There are strong political issues involved, as well as danger of over-concentration in urban agglomerations which we proved to yield small but negative effect on FDI efficiency other things equal. As suggested by Alfaro (2010) and Hermes and Lensink (2003), it is very important to take into account financial market development. Hence, it can be interesting to add commercial lending term, or net domestic credit measure to our specification to observe, whether these terms can affect our results. There is a possibility that, as democracies tend to rely more on the market and financial markets in particular, rather than on individual
decisions, and hence, remembering Hermes and Lensink‟s (2003) opinion on efficiency gained due to financial market development, adding it as a term in interaction with FDI can change our results in favor of democracies.
Another possible research to base on the model we evaluate in our research can be based on better measurement of agglomeration. There may be more economic approach to measuring agglomeration, than simply taking urban concentration as a proxy, based on wage and rent differentials, for instance (D.Puga, 2009).
To sum up, in the light of current political transformations experienced by some parts of the world, our model, probably transformed according to one‟s needs, may help to evaluate differences between countries and to advise on the decisions regarding openness to international capital flows, or, Foreign Direct Investment.
References
Alguacil M., Cuadros, A., Orts, V. (2011) Inward FDI and Growth: The Role of Macroeconomic and Institutional Environment, Journal of Policy Modeling, 2011, vol. 33, issue 3
Alfaro, L., Chanda, A., Kalemli-Ozcan, S. and Sayek, S. (2004) FDI and economic growth:
the role of local financial markets, Journal of International Economics
Alfaro, L., Chanda, A., Kalemli-Ozcan, S.,&Sayek, S. (2010). Does foreign direct investment promote growth? Exploring the role of financial markets on linkages, Journal of Development Economics, 91(2)
Al Nasser, O. (2007). The Determinants of the U.S. Foreign Direct Investment: Does the
Region Matter? Global Economic Review,1(36)
Bradford, J., DeLong and Shleifer, A. (1993), “Princes and Merchants: City Growth before the Industrial Revolution,” Journal of Law and Economics 36
Batten, J. and Vo, X. (2009) An analysis of the relationship between foreign direct investment and economic growth, Applied Economics, Taylor and Francis Journals, vol. 41(13)
Blomstrom, M., Lipsey, R., Zejan, M., (1992) What Explains Developing Country Growth,
NBER Working Paper No. 4132
Blonigen, B. A., &Wang, M. G. (2005) Inappropriate pooling of wealthy and poor countries
in empirical FDI studies. In T. Moran, E. Graham, & M. Blömstrom (Eds.) Does FDI promote development? (pp. 221–243), Washington, D.C, Institute for International Economics
Blundell, R. and Powell, J. (2004) Endogeneity in Semiparametric Binary Response Models,
Review of Economic Studies, Vol. 71, No. 3
Brülhart, M. and Sbergami, F. (2008) Agglomeration and Growth: Cross-Country
Evidence, CEPR Discussion Papers 6941, C.E.P.R. Discussion Papers
Demekas, D., Horváth, B., Ribakova, E., & Wu, Y. (2007) Foreign direct investment in
European transition economies: The role of policies. Journal of Comparative Economies,
35(2)
Diamond, L. (2001) How People View Democracy: Findings from Public Opinion Surveys in
Four Regions, presentation, 2001
Easterly,W. (2005) National policies and economic growth: A reappraisal. Handbook of economic growth
Easterly, W. (2011) Benevolent Autocrats, Working paper
Glaeser E., La Porta R., Shleifer A. (2004) Do Institutions Cause Growth? Journal of
Economic Growth
Henderson, V. (2003) The Urbanization Process and Economic Growth: The So-What
Question, Journal of Economic Growth, Springer, vol. 8(1)
Hermes, N. and Lensink, R. (2003) Foreign direct investment, financial development and economic growth, Journal of Development Studies
Kemeny, T. (2010) Does foreign direct investment drive technological upgrading? World
Development, 38(11)
Li, X. and Liu, X. (2005) Foreign direct investment and economic growth: an increasingly endogenous relationship, World Development
Mallick and Moore
Martin, P. & Ottaviano, G. I. P. (2001) Growth and Agglomeration, International Economic Review, Department of Economics, University of Pennsylvania and Osaka University Institute of Social and Economic Research Association, vol. 42(4)
Marshall, M. and Jaggers, K. (2002) Polity IV Project: Political Regime Characteristics and
Transitions, 1800-2010
Ober, J. (2008) Natural Capacities and Democracy as a Good-in-Itself, Philosophical Studies,
132
Puga, D. (2009) The magnitude and causes of agglomeration economies, Working
Papers 2009-09, Instituto Madrileño de Estudios Avanzados (IMDEA)
Poelhekke S. and Van der Ploeg F. 2008 Globalization and the Rise of Mega-Cities in the
Developing World, CESifo Working Paper Series, 2208
Sen, A. (1999) Democracy as a Universal Value, Journal of Democracy 10.3
Verbeek, M. (2004) Modern Econometrics, 2nd ed., John Wiley & Sons, Ltd
World Bank: World Bank Governance Indicators (by Kaufmann et al (2009)) World Bank: World Development Indicators, latest issue
Rothstein, B. et al (2011) The Quality of Government Dataset Codebook, The Financial Times, May 18, May 2, 2012
The Economist, Economist Intelligence Unit
Колеников С.О. Прикладной эконометрический анализ в статистическом пакете
STATA
Appendix
Table 3.1
BASE
Variables (1)
Full Sample
Full Sample (2)
fdi_percent_t1
Log_gdp_cur_us_t1
-2.072***
(0.352)
0.0938**
-2.199***
(0.0421) (0.354)
Pop_growth -0.393 -0.431
(0.321) (0.333)
invest_t1 -0.0203 -0.0235
trade_percent_t1 school25_av_t1
Constant
0.0345***
1.134***
10.07*** (0.0228) (0.0103) (0.293)
(1.899)
0.0299***
1.062***
11.75*** (0.0227) (0.0104) (0.295)
(2.112)
Observations
R-squared
2,247
0.064
2,202
0.069
Number of country_id 96 96
Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1
Table 3.2
INFRASTRUCTURE
Variables fdi_percent_t1 log_gdp_cur_us_t1 pop_growth invest_t1 trade_percent_t1 school25_av_t1 telephone_t1
Internet_t1 road_land_t1
Inflation_t1
Constant
Observations
R-squared
Number of country_id (3) (4) (5)
Full Sample
0.0936** 0.0372 0.0212 (0.0418) (0.0346) (0.039)
-2.507*** -1.493** -2.814***
-0.407 -0.568 -0.914
-0.434 -0.605* -0.253
-0.331 -0.345 -0.5
-0.0229 -0.0401 -0.0475
-0.0233 -0.032 -0.0362
0.0294*** 0.00722 0.0254
-0.0101 -0.0135 -0.0205
0.946*** 0.376 1.258*
(0.298) (0.498) (0.73)
0.0462* (0.0249)
0.00196 (0.00795)
0.0978 (0.851)
14.01*** 13.04*** 16.34*** (2.559) (4.841) (5.038)
2,200 1,081 858
0.071 0.042 0.048
96 96 95
n parentheses: *** p<0.01, ** p<0.05, * p<0.1 (6)
0.0854** (0.0404)
-2.516*** (0.403)
-0.432 (0.331)
-0.0191 (0.023)
0.0305*** (0.00996)
0.959*** (0.296)
0.0434* (0.0246)
-6.24E-05 (5.04E-05)
13.93*** (2.522)
2,196
0.071
96
Robust standard errors i
Table 3.3.1
URBANIZATION
variables fdi_percent_t1 log_gdp_cur_us_t1 pop_growth invest_t1 trade_percent_t1 school25_av_t1 telephone_t1
Inflation_t1 urban_percent_t1 urban_growth_t1 fdi_urban fdi_urban_gr Constant
Observations
R-squared
Number of country_id (7) (8) (9)
Full Sample
0.0741* 0.0846** 0.284*** (0.0375) (0.0402) (0.0966)
-2.479*** -2.485*** -2.426*** (0.364) (0.397) (0.361)
-0.652** -0.569 -0.621** (0.305) (0.384) (0.3)
-0.0288 -0.0185 -0.0303 (0.025) -0.0229 -0.0245
0.0207** 0.0293*** 0.0187* (0.0104) -0.0098 -0.0101
0.52 1.038*** 0.555*
(0.313) (0.304) (0.301)
0.0383 0.0409* 0.0391 (0.0247) (0.0242) (0.0248)
-6.94E-05 -6.27E-05 -6.68E-05 (5.01E-05) (5.02E-05) (4.51E-05)
0.0926* 0.0899* (0.0505) (0.0486)
0.187 (0.195)
-0.00321**
-0.00128
12.31*** 13.10*** 11.96*** (3.096) (2.597) (2.988)
1,954 2,196 1,954
0.081 0.073 0.085
95 96 95
in parentheses: *** p<0.01, ** p<0.05, * p<0.1 (10)
0.172*
(0.1)
-2.167*** (0.348)
-0.973** (0.385)
-0.0314 (0.0252)
0.0214** (0.00925)
0.755** (0.321)
0.0404* (0.0237)
-6.09E-05 (4.18E-05)
0.0697 (0.239)
-0.00265** (0.00126)
0.0386* (0.0208)
14.15*** (2.727)
1,954
0.085
95
Robust standard errors
Table 3.3.2
URBAN_SQ (11)
variables Full Sample
fdi_percent_t1 0.305*** (0.0955)
fdi_urban -0.00353*** (0.00127)
log_gdp_cur_us_t1 -2.436*** (0.37)
pop_growth -0.606** (0.299)
invest_t1 -0.0302 (0.0253)
trade_percent_t1 0.0191* (0.00972)
school25_av_t1 0.525* (0.297)
telephone_t1 0.033 (0.0252)
inflation_t1 -0.0000653 (0.0000451)
urban_percent2 0.000946** (0.000408)
Constant 13.85*** (2.544)
Observations 1,954
Number of country_id 95
R-squared 0.086
Robust standard errors in parentheses: *** p<0.01, ** p<0.05, *
p<0.1
Table 3.4
PRIMACY
variables fdi_percent_t1 log_gdp_cur_us_t1 pop_growth invest_t1 trade_percent_t1 school25_av_t1 telephone_t1 inflation_t1 primacy_t1 primacy_2 fdi_primacy_t1 fdi_primacy_2
Constant
Observations
R-squared
Number of country_id (11) (12) (13)
Full Sample
0.0765 0.0757 0.142*
(0.0542) (0.0555) (0.0828)
-2.234*** -2.090*** -2.258*** (0.392) (0.381) (0.386)
-1.402*** -1.378*** -1.340*** (0.237) (0.253) (0.262) (0.0197) (0.015) (0.0225) (0.0283) (0.0336) (0.028)
0.0234** 0.0262** 0.0232** (0.0112) (0.0112) (0.0111)
0.516* 0.543* 0.543*
(0.286) (0.276) (0.275)
0.0332 0.0278 0.0349 (0.0277) (0.0272) (0.0275)
-0.000138 -0.000142* -0.000134 (8.71E-05) (8.46E-05) (8.78E-05)
0.226*** 0.235*** (0.062) (0.0633)
0.00220*** (0.000392)
-0.00267 (0.00178)
11.32*** 13.41*** 11.09*** (2.749) (2.802) (2.775)
1,635 1,635 1,635
0.102 0.098 0.103
71 71 71
rentheses: *** p<0.01, ** p<0.05, * p<0.1 (14)
0.0945 (0.0634)
-2.18e-05* (1.20E-05)
-1.322*** (0.281) (0.0175) (0.0336)
0.0264** (0.0112)
0.570**
(0.27)
0.0284 (0.0272)
-0.00014 (8.49E-05)
0.00225*** (0.000415)
-2.18e-05* (1.20E-05)
13.31*** (2.817)
1,635
0.099
71
Robust standard errors in pa
Table 4
INSTITUTIONS
variables
fdi_percent_t1 fdi_urban
log_gdp_cur_us_t1
pop_growth invest_t1 trade_percent_t1
school25_av_t1
telephone_t1
inflation_t1 urban_percent2
ICRG for QoG
Rule of Law
Control of
Corrupt.
Regul. Quality(-1) Political Constraints Political Stability Regul. Quality Constant
Observations
R-squared Number of country_id (15) (16) (17) (18) (19) (20)
Full Sample
0.245** -0.0039 -0.0864 -5.46E-05 0.206* -0.280*
(0.103) (0.112) -0.117 -0.113 -0.11 -0.168
-0.00261* 0.000275 0.00154 0.000273 -0.00194 0.00411* (0.00132) (0.0015) (0.00151) (0.00151) (0.00152) (0.00231)
-1.693*** -2.962*** -3.258*** -2.847*** -1.556*** -2.034 (0.464) (0.826) (1.033) (0.819) (0.535) (1.52)
-0.687* -0.617*** -0.542*** -0.638*** -0.61 -0.883*** (0.35) (0.166) (0.171) (0.171) (0.414) (0.13)
-0.0381 -0.0872 -0.09 -0.0808 -0.0436 -0.0706 (0.0274) (0.0534) (0.0651) (0.0517) (0.0286) (0.0782)
0.0240** -0.0202 0.00137 -0.0228* 0.0288*** 0.0425** (0.0101) (0.0142) (0.0175) (0.0135) (0.0102) (0.0186)
0.47 0.5 (0.296) (0.327)
-0.0266 0.0830** 0.170*** 0.0800* -0.0363 0.00925 (0.0309) (0.0374) (0.0413) (0.0404) (0.035) (0.0508)
-8.66e- – -8.28e-
-0.000175 05*** 0.000114*** 05*** -0.000156 -0.00295 (0.00011) (0.0000108) (0.0000125) (0.0000117) (0.000107) (0.00542)
0.000955* 0.00417*** 0.00317 0.00384*** 0.000773 0.00846** (0.000505) (0.00123) (0.0022) (0.0012) (0.000526) (0.00352)
2.345* 2.184 (1.377) (1.369)
-0.408 (1.197)
2.753** (1.132)
-1.221 (1.15)
0.265 -0.75
0.747 (1.538)
0.281 (1.007)
8.133*** 11.67*** 13.60*** 12.57*** 7.428** -16.39** (2.972) (3.87) (4.406) (4.113) (3.425) (7.166)
1,450 751 680 751 1,358 500
0.098 0.117 0.146 0.12 0.091 0.146
83 93 92 93 82 90
rors in parentheses: *** p<0.01, ** p<0.05, * p<0.1 (21)
-0.279*
-0.167
0.00411* (0.00233)
-2.031 (1.534)
-0.891*** (0.129)
-0.0727 (0.0723)
0.0429** (0.0189)
0.0111 (0.0508)
-0.00277 (0.00575)
0.00848** (0.00349)
-0.743 (1.508)
0.242 (1.662)
-16.57** (6.89)
500
0.146
90
Robust standard er
Table 4.1
URBAN_SQ_INST
variables
Full Sample
FDI/GDP
0.0785
(0.239)
FDI*Urbanization -0.000606
lnGDP(-1) Pop. growth (0.00299)
-1.507*** (0.469)
-1.142*** (0.258)
Investment/GDP -0.0105
Trade/GDP (0.0298)
0.0340*** (0.0120)
Av. schooling 0.0428
(0.258)
Telephone -0.0171
(0.0307)
Inflation -0.000119
ICRG for QoG (0.000146)
2.825* (1.509)
Urbanization -0.0656
(0.130)
Urbanization squared 0.00146
Constant (0.00129)
10.16** (4.979)
Observations
1,305
Number of country_id 80
R-squared 0.093
INSTITUTIONS
variables fdi_percent_t1 fdi_primacy_t1 log_gdp_cur_us_t1 pop_growth invest_t1 trade_percent_t1 school25_av_t1 telephone_t1
inflation_t1
ICRG for QoG Rule of Law
Control of Corrupt. Political Constraints Political Stability Regul. Quality(-1) Constant
Observations
R-squared
Number of country_id (22) (23) (24) (25) (26)
Full Sample
0.155** -0.0436 -0.0902 0.150** -0.122** (0.0711) (0.0683) (0.0702) (0.0707) (0.0542)
-0.00290 0.00148 0.00409** -0.00281 0.00350 (0.00192) (0.00202) (0.00195) (0.00193) (0.00250)
-1.206*** -1.487** -2.379*** -1.266*** -0.171 (0.437) (0.570) (0.677) (0.439) (0.753)
-1.408*** -1.428** -1.259* -1.398*** -2.418*** (0.326) (0.645) (0.683) (0.320) (0.563)
-0.0178 -0.0718 -0.0652 -0.0192 -0.0324 (0.0333) (0.0672) (0.0796) (0.0331) (0.0806)
0.0282** -0.00524 0.0307 0.0302** 0.0625** (0.0115) (0.0183) (0.0265) (0.0115) (0.0288)
0.525** 0.425 (0.232) (0.261)
-0.0277 0.0818** 0.191*** -0.0243 0.0696** (0.0291) (0.0397) (0.0391) (0.0287) (0.0334)
–
-0.000154 -0.0220*** -0.00158 -0.000157 0.0161***
(0.000119) (0.00326) (0.00653) (0.000120) (0.00233)
1.811 1.819 (1.351) (1.326)
-0.997 (1.442)
2.699** (1.350)
0.561 -0.361 (0.877) (1.266)
-0.599 (0.587)
8.721** 17.34*** 17.71*** 9.313*** 2.116 (3.319) (4.719) (4.591) (3.318) (5.733)
1,326 616 551 1,318 481
0.103 0.087 0.131 0.105 0.145
71 71 71 71 71
n parentheses: *** p<0.01, ** p<0.05, * p<0.1 (27)
-0.0965* (0.0551)
0.00283 (0.00243)
-0.183 (0.710)
-2.254*** (0.592)
-0.00965 (0.0808)
0.0509* (0.0261)
0.0586* (0.0323)
-0.0201*** (0.00274)
0.422 (1.372)
-3.196** (1.454)
3.263 (5.628)
481
0.174
71
Robust standard errors i
(1) (2) (3) (4) (1) (2) (3) (4)
pop_growth – Annual percentage growth in national population (4739 obs.)
inflation – Annual (%), averaged (4478 obs.)
lending – Commercial banks and other lending, to approximate the level of financial sector activity;
telephone – Telephone lines per 100 people and (4691 obs.)
internet – Internet users per 100 people (2622) (here possible to cite wiki)
road_density – kilometers of road per 100 sq. km of land area
wbgi_pse – Political Stability indicator; represents the soundness of the regime
wbgi_vae – “Voice and Accountability” indicator represents a mix of factors: civil liberties, political rights, independence of the media, mainly indicating to what extent citizens are able to affect the selection of their government
wbgi_gee – Government Effectiveness indicator stands for a measure of bureaucratic system quality and the competence of its members;
wbgi_cce – Control of Corruption index, corruption defined as an “exercise of public power for public gain” (QoG Codebook).
wbgi_rle – the “Rule of Law” indicator measures the confidence in the nation‟s legal system,
contract system and, more general, in the social normative framework.
wbgi_rqe – the Regulatory Quality estimate plays a role in the movements of investments, as it stands for measurement of regulatory burden on such aspects as international trade and overall business environment;
Figure 1 – GDP comparison
Democracy Sample:
Variable Obs Mean St Dev Min Max
gdp_cur_us 2379 10325.95 13712.54 104.81 118218.8
Autocracy Sample:
Variable Obs Mean St Dev Min Max
gdp_cur_us 1003 2013.206 4190.095 86.76 36738.45
*Sourse: author’s calculations
