Empir Econ (2010) 39:183–201 DOI 10.1007/s00181-009-0300-z ORIGINAL PAPER
Growth volatility and the interaction between economic and political development Jeffrey A. Edwards · Frank C. Thames
Received: 15 October 2007 / Accepted: 4 April 2009 / Published online: 2 May 2009 © Springer-Verlag 2009
Abstract Research on the effect of democracy on economic growth has not reached a definitive conclusion. Yet, research on the effect of democracy on economic growth volatility has consistently found that higher levels of democracy reduce volatility. Similarly, research has found that higher levels of economic development retard volatility. Using a novel empirical approach, this article presents evidence of an interactive effect between higher levels of democratization and economic development on growth volatility. Specifically, the marginal effect of political development on volatility is negative until countries reach per capita income levels of about $2,700, depending on the conditioning set. The marginal effect is insignificant for countries with higher levels of income. This implies that at a minimum, nearly 50% of the countries in our sample could enjoy less volatile economies with greater political development. Keywords
Growth · Volatility · Development · GMM
JEL Classification
O11 · O43 · C13 · C23
Authors would like to sincerely thank the reviewers of our manuscript in providing us with valuable suggestions that made this a better paper. J. A. Edwards (B) Department of Economics, North Carolina A&T State University, 1601 E. Market St., Greensboro, NC 27411, USA e-mail:
[email protected] F. C. Thames Department of Political Science, Texas Tech University, Box 41015, Lubbock, TX 79409, USA e-mail:
[email protected]
123
184
J. A. Edwards, F. C. Thames
1 Introduction In the political economy of institutions scholarship, few questions have elicited as much research as the question: does democracy affect economic growth? While some argue that democracy promotes growth, others argue that democracy undermines growth. In the end, there is still little consensus on how democracy affects growth (e.g., Baum and Lake 2003; Dreze and Sen 1989; Feng 1997; Helliwell 1994; Wittman 1995; Wittman 1989; Baba 1997; Knack and Keefer 1995; Guiso et al. 2004; Przeworski and Limongi 1993; Sirowy and Inkeles 1990). A related topic focuses on the effect of greater democratic institutionalization and economic development on growth volatility. Research on this question has two major findings. First, there is good evidence that strong democratic institutions reduce volatility in growth (e.g., Rodrik 1997; Rodrik 2000; Almeida and Ferreira 2002; Mobarak 2005). These studies suggest that democratic institutions reduce volatility by decreasing policy variance, strengthening institutional structures, and increasing the importance of risk averse voters in policymaking. Second, there is also strong evidence that economic development tends to reduce volatility (e.g., Acemoglu and Zilibotti 1997; Bekaert et al. 2006; Koren and Tenreyro 2007; Calderon et al. 2005). It does so by increasing financial depth and liberalization, international integration, and low sectoral concentration. It also does this by creating a class of economic agents who benefit from low volatility. We argue that there is an interactive effect between increasing political and economic development on growth volatility. Stronger democratic political institutions will increase the openness of the political system that, in turn, will make it easier for pro-stability forces to impact economic policy making. Using a dataset of 73 countries over 30 years, we test the marginal effect of the level of democratic institutionalization on growth volatility as a function of per capita real GDP. We use a two-stage-least-squares (2SLS) system of equations whereby the endogeneity of the control variables is corrected in a first-stage system generalized method of moments (GMM) estimation. With this approach, we find that the negative effect of democratization on volatility becomes insignificant once countries achieve a level of economic development around $2,700–$4,900 per capita income, depending on the conditioning set. We also find that this result is quite robust to outliers and an alternative specification of the democratization variable. (In fact, in our final robustness test, all except for two globally determined variables are treated as endogenous.) This implies that anywhere from about 50 to 60% of the countries in our sample will benefit from democratization with lower levels of volatility. Although the benefits of democratization on growth volatility may be limited to lower levels of economic development, there are a significant number of countries that would benefit from this improvement. The rest of the article will proceed as follows. First, we will analyze the existing literature on democracy, economic growth, and economic volatility. Second, we will discuss the methodological approaches used to assess the relationships between economic development, democratization, and growth volatility. Third, we will present our empirical findings. Finally, we draw a conclusion.
123
Growth volatility and the interaction between economic and political development
185
2 Democratic institutions, economic development, and economic growth volatility Democracies differ from authoritarian regimes primarily due to the extent governing officials are selected in elections that are not only free and fair, but feature high levels of contestation (Dahl 1972). Thus, political development can be thought of as the strengthening of democratic political institutions. In fact, much of the literature on the relationship between democracy and economic growth is based on an understanding of how democratic political institutions create sets of incentives that promote or undermine growth. Because democratic institutions are more transparent, they encourage a flow of information that allows citizens to better monitor excessive rent-seeking (Wittman 1995; Wittman 1989; Baba 1997). This reduces the opportunities for predation by state leaders that may undermine economic growth. In addition, since democratic institutions require leaders to obtain popular support, they tend to provide stronger property-rights regimes (Durham 1999; Kormendi and Meguire 1985; Knack and Keefer 1995; Pastor and Sung 1995; Leblang 1996). As the strength of property rights increases, so does the incentives for actors to invest. This investment, in turn, can spur growth. Finally, Lohman (1999) argues that democracies tend to select more qualified leaders than do authoritarian regimes. With more qualified leaders, democracies institute better policies to enhance economic growth.1 Those who argue that democracy undermines growth also focus on the effects of democratic political institutions. Olson (1982) argues that democratic governments are less able to resist interest group rent seeking, making their economies more stagnant. By opening the political process to more groups, democracies allow rent seeking to proliferate and undermine effective economic management, which leads to slower growth. Nordhaus (1975) writes that elected officials will have an incentive to manipulate the economy due to short-term electoral concerns. In this version of the political-business cycle, the need to win elections will create incentives for politicians to embrace policies that will aid their reelection in the short term, but will undermine growth in the long term.2 What about the effect of democratic institutions on growth volatility? The existing research suggests that democratic institutions retard growth volatility in two specific ways. First, democratic political institutions create incentives for political leaders to be risk averse. (Quinn and Woolley, 2001, p. 636) cogently argue that the studies examining the effect of democracy on economic growth are based on the “. . . implicit assumption that citizens have an unqualified preference for more growth rather than less growth.” Yet, using a logic based on Przeworski (1991) and Quinn and Woolley (2001) argue that democratic institutions create incentives for elites to be more risk averse, and thus avoiding riskier high growth economic strategies, for two reasons. 1 Democracy may have benefits for economic growth beyond simply institutions by stimulating the growth of human capital (Baum and Lake 2003; Dreze and Sen 1989; Feng 1997; Helliwell 1994), and increasing the level of social capital (e.g, Knack and Keefer 1995; Guiso et al. 2004). 2 In the end, there is still significant disagreement on the effects of democracy on growth (see, Przeworski and Limongi 1993; Sirowy and Inkeles 1990; Leblang 1997; Heo and Tan 2001; Gerring et al. 2005; Baum and Lake 2003; Helliwell 1994; Feng 1997; Przeworski et al. 2000; Tavaresa and Wacziarg 2001).
123
186
J. A. Edwards, F. C. Thames
First, since democratic institutions offer elites the possibility of future replacement, the costs of defeat are lower, decreasing the need for them to take bigger risks. In democratic systems, the cost of losing an election tend to be much less than losing the struggle for power in an authoritarian regime. Second, the behavior of elected elites in a democratic system will be affected by the median voter. In order to win an election, political elites have strong incentives to satisfy the demands of the median voter (Black 1948; Downs 1957), who would, as argued above, prefer a more stable economy. Thus, the incentive for risk-taking among elites is potentially mitigated by the more risk-averse median voter who would prefer certainty over uncertainty. This logic leads to the expectation that democratic economies will be less volatile due to the effect of democratic institutions. Second, democratic political institutions decentralize political power, which makes it difficult to change policy. Democratic institutions create more “veto players” than do authoritarian institutions (Henisz 2000). With more veto players the ability to make radical changes in economic policy decreases, which should increase stability. Nooruddin (2003) makes similar claims, showing that as the number of limitations on politicians increases, growth volatility decreases. Mobarak (2001) demonstrates that decentralization of political power tends to reduce instability in policy making thereby reducing growth volatility. Chandra (1999) finds that stable outcomes in countries tend to be positively correlated with diversity in terms of institutional portfolios. Thus, the decentralization of political power inherent to democratic political systems makes policy change difficult, creating a more stable economic environment. Consequently, growth volatility should be lower as political institutions become more democratic. While the strength of democratic institutions may be associated with lower volatility, so may higher levels of economic development. For instance, Acemoglu and Zilibotti (1997) argue that in early stages of development there is less economic diversification. This lack of “risk spreading” ultimately leads to economic uncertainties. Therefore, patterns of development will first consist of highly variable output, followed by financial deepening and more steady growth. Bekaert et al. (2006), find that financial liberalization is associated with lower consumption growth volatility. Financial liberalization increases international risk sharing thereby reducing volatility in consumption, and volatility in the economy. Koren and Tenreyro (2007), find that production tends to shift towards less risky sectors with the level of development. They also find that high sectoral concentration in early stages of development tends to occur in high risk sectors, therefore increasing exposure to risk during these stages. While the evidence points toward an inverse relationship between economic development and volatility, further development will propagate the need for even greater stability. Since agents want “smooth” consumption and income paths, high overall volatility will make this feat difficult. Through economic development, these same agents become more savvy investors which then increase the desire to smooth their consumption and/or income even more. This idea can be reasoned with the following argument outlined in Aghion et al. (1999). The argument is simple. At low levels of development, there tends to be larger degrees of separation between savers and investors. This separation may be a physical one whereby savers are simply in no position to invest directly in physical capital. Investing requires education, imagination, and networking. Agents in lesser-developed
123
Growth volatility and the interaction between economic and political development
187
economies are usually lacking in at least one of these characteristics. For instance, the idea that one can use their saving to borrow funds and invest in capital for future gain is not an idea that crosses most of these individual’s minds. It is also the case that investment can require relatively large initial outlays of liquidity that simply may not be available to those in poorer economies. Furthermore, investment usually requires some form of cooperation with individuals in similar investment circles. This would thereby limit the accessibility of funds to those individuals not directly connected with existing financial markets. In lesser-developed economies, sophisticated social and financial connections between social groups at high and low income levels may not exist in a way that would provide the saver the means to network their way toward increasing investment. Increased economic development decreases the separation between investors and savers by creating a more financially savvy, connected, and wealthier demographic. Aghion et al. (1999), show that when this separation is low, volatility is also low. In this society, when growth and interest rates are low, savvy investors can manipulate their assets to retain a relatively high proportion of their profits. This lets them rebuild their reserves and expand their investment. Hence, increased economic development creates a society whereby individuals can and will spend much of their efforts neutralizing periods of volatile growth. If this effort is costly, it is therefore reasonable to assume that these agents would want to reduce this cost. Therefore, economic development leads to agents who want low volatility so these efforts can be minimized. We argue that the extent to which the economic agents who favor low volatility can impact policy will be increased under more open political regimes. Higher levels of economic development are associated with low volatility due, in part, by creating a coalition of political actors who, like the median voter, have preferences for stable economic growth. As the gap between investors and savers decreases as development increases, the number of individuals who prefer low volatility increases as well. The ability of these economic agents to affect policy will increase to the extent that political institutions allow more actors to impact policy formation. Thus, as democratic political institutions strengthen so should the ability of these actors to push policymakers towards stable economic growth. Put another way, economic development creates societal demands for greater growth stability; however, the ability of a pro-stability coalition to achieve its goal will depend on the nature of the country’s political institutions. The more open the polity, the more likely pro-stability forces will be able to influence economic policy. Consequently, we expect an interactive effect between the strength of democratic political institutions and economic development on growth volatility.3 3 It is possible that both economic and political development are highly correlated with each other. There is a long-standing belief that democratization is linked in linear fashion with economic development (Lipset 1959, 1981). Yet, others disagree with a linear interpretation of this relationship, suggesting that wealth creates stability irrespective of whether the system is democratic or authoritarian (Przeworski and Limongi 1993; Przeworski et al. 2000). Thus, the levels of economic and political development may be correlated; but, it is highly possible they are measuring different things. In addition, as discussed above, previous research suggests different mechanisms by which political and economic development will affect growth volatility. By using proxies for both economic and political development not only can we determine whether they independently affect volatility, but we can ascertain how they work together to affect volatility.
123
188
J. A. Edwards, F. C. Thames
There is some evidence of this in the literature. For example, Mobarak (2005) found that the negative effect of higher levels of democracy was found in a model that included only low-income countries, but not in high-income countries. While this study was empirically novel, and effectively answered the two questions posed above, it was incomplete. Mobarak shows that the marginal effect of political development on volatility is only negative for countries with incomes below $4,000 per capita GDP. But the “cutoff” level of income he chooses seems somewhat arbitrary and the delineation is binary; i.e., either a country is higher than $4,000, or less than $4,000. He also uses decadal data thereby holding the determinants constant across widely popular, nonetheless arbitrary, blocks of time. We add to this outcome in three important ways. First, we specifically analyze the marginal effect of political development on volatility across all levels of income—i.e., there is no binary delineation of the data. In doing this, we hope to pinpoint where the upper bound of the marginal effect’s 90% confidence interval crosses zero. Analyzing the upper bound of the confidence interval will give us the precise income location whereby the mean marginal effect becomes statistically insignificant. Second, we use annualized data, giving us the full scope and influence generated by annual changes in the conditioning set. Thirdly, we investigate whether this effect is robust to different conditioning sets across more precisely defined levels of income than just the broadly defined dichotomy described above; in other words, we can tell how the influence income has on the marginal effect of political development changes as we control for different variables. 3 Model Given our desire to present robust results, we must deal with several specification issues. First, we must decide how to measure growth volatility. What first comes to mind is the standard deviation of the growth rate in real per capita GDP; however, this variable is limited when investigating annual levels because it must be held constant for an extended period of time. Since this study is one of examining the effect democracy has on levels of volatility after controlling for economic characteristics that are temporal in nature, using as small a unit of time as possible is optimal. Therefore, we intend to follow the example of Edwards et al. (2006b), and Edwards (2007) by using the absolute value of the annual deviation from the average of growth in real per capita GDP as a proxy for levels of volatility. These are essentially annual deviations from what can be considered long-run growth. Debating the functional form of volatility leads us directly into our second concern: what should we use as a proxy for democratic political institutions? There are several measures of democracy available, all of which have their strengths and weaknesses.4 Since our primary goal is to examine the development of democratic political institutions, we use the Polity2 variable found in the Polity IV Project (2004) dataset. Of the available democracy measures, Polity2 most directly assesses the strength of democratic political institutions, since it is created by coding countries based on the competitiveness of elections and the competiveness and openness of executive recruitment 4 See Munk and Verkuilen (2002) for an excellent review of the available measures of democracy.
123
Growth volatility and the interaction between economic and political development
189
(Polity IV Project 2004). Each country is scored on a scale from −10 to 10 with 10 being a consolidate democracy. The other alternative, the Freedom House/Gastil measure, suffers from problems of both inter-coder reliability and the lack of clear rules governing coding decisions that the Polity IV dataset does not suffer from (Munk and Verkuilen 2002).5 Third, we must decide what to include in the conditioning set if the objective is to control for most things economic that could affect levels of volatility? Edwards et al. (2006b) showed that current levels of volatility can be significantly influenced by past volatility; i.e., volatility has a business-cycle dynamic just like growth. Consequently, we include lagged volatility in our conditioning set. Classical growth theory dictates that over a particular time period different countries will have started on different points of their production functions, implying inherently different levels of volatility; country-specific fixed effects should be a sufficient proxy for this heterogeneity (e.g., Islam 1995). Following this methodological standard for growth equations, we include the ratio of investment-to-GDP (Investment), the population growth rate (Population Growth), the percentage of the population over 25 with a secondary education (Schooling), the level of trade openness (Openness), and the level of inflation (Inflation).6 To complicate matters further, we eventually include volatilities in world growth rates and oil prices (World Volatility and Oil Volatility respectively). They have been shown to be important for capturing inelastic commodity price adjustments (via volatile oil prices) and for the most part, completely exogenous inertia in world markets (via volatility in world growth) (Edwards et al. 2006b).7 As stated earlier, the data set is perfectly balanced across panels—i.e., each year covers exactly the same countries. Doing this greatly reduces any possibility for bias that may result from different years covering different countries. Having said this, we also eventually control for gross capital flows (GCF), and foreign direct investment (FDI). These have been shown to be important when determining how volatility affects growth rates especially during this era of globalization (Kose et al. 2006; Edwards 2007). The data for these variables is limited in availability and their inclusion results in highly unbalanced panels. On the bright side, evaluating results from this unbalanced regression will add to our test of robustness. The schooling data comes from Barro and Lee (2000), while the remaining variables come from the World Bank’s, World Development Indicators database (2006). The dataset covers 73 countries over the years 1973 through 2003; loosing two years to lags, the final data covers 73 countries from 1975 through 2003. The unbalanced
5 In political science, Polity IV is increasingly the “industry standard” measure of democracy, in particular in the International Relations literature. One reason for this, which is beyond our concern, is that Polity IV maintains data for countries as far back as 1800. Freedom House/Gastil measures only go back as far as 1972. 6 See countless empirical growth papers for examples of these standard variables starting with Mankiw et al. (1992), Levine and Renelt (1992), and Islam (1995). Many arguments remain as to what variables are necessarily “robust” determinants of growth, e.g., Hoover and Perez (2004), Sala-I-Martin (1997), and Edwards et al. (2006a), but this set of variables seem to show up much more frequently than others. 7 To draw valid inference from these regressions, the linear form of GDP is also included (Brambor et al.
2005).
123
190
J. A. Edwards, F. C. Thames
data covers the same 73 countries, but the number of years per country varies from 13 to 29 years. To this point, we have four models to estimate. The basic form is Volatilityit = ai0 + a1 Volatilityit−1 + a2 Polity2it + a3 GDPit + a4 Polity2it ∗ GDPit + a5 X it + eit
(1)
where the X ’s are the economic control variables explained above. To our end, we want to evaluate the derivative function ∂Volatilityit = a2 + a4 GDPit . ∂Polityit
(1.1)
∂Volatility
Of particular importance is at what level of GDP is ∂Polity it < 0, and if this level it is similar for each specification. Statistically, then, it is imperative to evaluate at what level of GDP does the upper bound of the marginal effect cross zero. The standard error of interest is calculated as σˆ ∂Volatility = var(aˆ 2 ) + GDPit2 var(aˆ 4 ) + 2GDPit cov(aˆ 2 aˆ 4 ), and our confidence inter∂Polity
val will be 90%. A fourth valid concern for any researcher doing this sort of regression analysis would be to address any feedback that may occur between volatility and the determinants. It is well-known in the economics literature that the Mankiw et al. (1992) form of a growth regression has endogeneity issues that can cause significant problems when drawing inference. For instance, a regression of growth on investment implies that investment causes growth. Yet, high levels of growth can also spur investment. By the same token, different levels of growth volatility should also cause changes in investment. The same problem exists with lagged volatility in a dynamic panel format such as this one. The problem is obvious when viewing a simple form of (1), but without the polity and GDP variables; e.g., take the regression Volatilityit = αi0 + α1 Volatilityit−1 + ωit . A within estimation takes the first difference of the equation to wipe out the αi0 ’s. This leaves Volatilityit − Volatilityit−1 = α1 (Volatilityit−1 − Volatilityit−2 ) + (ωit − ωit−1 ). The problem here is that the term (Volatilityit−1 − Volatilityit−2 ) is correlated with (ωit − ωit−1 ). To make corrections for all relevant variables in our model, we use a two-stageleast-squares (2SLS) system where the first-stage generates the estimated forms for the second-stage. In a nutshell, we generate first-stage estimates of Volatilityit−1 , GDP, Investment, Inflation, Population Growth, Openness, GCF, and FDI, from their lags to be used in the second-stage estimation. We do not instrument the schooling, world and oil volatility variables, and the Polity2 variable, for different reasons. The schooling variable is constructed from a single point estimate every 5 years; therefore, we simply use the prior 5 year estimate for the subsequent 5 years of data. In other words, a point estimate for Schooling from 1980 would be used as observations for 1980–1984. The yearly observations for the world and oil volatility variables are the same for each country so any country-specific feedback is unlikely. Lastly, in our primary analysis,
123
Growth volatility and the interaction between economic and political development
191
we do not directly instrument for Polity2, even though cases have been made that it may also be endogenous to levels of volatility (e.g., Mobarak 2005). The nature of Polity2 does not allow for its estimation while maintaining its discreet characteristic. Any variable(s) other than binary dummies that exactly coincide with the time-specific properties of Polity2, will result in a continuous Polity2 estimate—not at all the intention of the framers of Polity2. Having said this, we do realize that any argument that volatility causes political development must include the thesis that a highly unstable economy could cause transitions from one type of political regime to another. We do control for these transition points by lagging Polity2 by one year. For instance, assume volatility in an economy in a particular year did actually cause a regime to transition from a level −3 to −4 in the Polity scale. Lagging Polity2 would require a feedback argument be equivalent to stating that volatility this year caused a transition last year. On the contrary, it does make sense that a transition in political development this year could cause changes in volatility next year. The first-stage regression for estimating the relevant RHS variables for use in the second-stage will be performed using a system GMM methodology (see Edwards et al. 2006b). For example, the first-stage regression for lagged volatility will take the general form Volatilityit−1 = λi0 + λ1 Volatilityit−2 + νit−1 . Taking the first difference as suggested by Anderson and Hsiao (1982), we get Volatilityit−1 − Volatilityit−2 = λ1 (Volatilityit−2 − Volatilityit−3 ) + (νit−1 − νit−2 ). They recommend using Volatilityit−3 or (Volatilityit−3 − Volatilityit−4 ) as instruments for (Volatilityit−2 − Volatilityit−3 ) which is correlated with (νit−1 −νit−2 ). Arellano and Bond (1991) proposed using the former in a GMM methodology that could start at time period zero. This makes use of a larger number of moments and is sometimes called a difference GMM method. Arellano and Bover (1995) extend this for a highly persistent series whereby the series is differenced and estimated using levels as instruments, as well as a levels series estimated using differences as instruments; this is the system GMM (Blundell and Bond 1998). The instrumentation for the other determinants is just as simple, but shifted one period to the present because it is the contemporaneous values that must be estimated in the first-stage, not the lagged values. Finally, there may be one additional specification issue that arise when estimating (1) using a 2SLS method. While all of our standard errors will be robust to heteroscedasticity, a 2SLS regression will in itself cause inference problems due to the fact that the standard errors of the coefficients will have a finite sample bias. An adjustment of the estimators’ standard errors is needed to correct this problem. The typical way to make this adjustment is to multiply each estimator’s standard error bythe adjustment factor σˆ ε /σˆ ε,Est , where σˆ ε is obtained from the expression σˆ τ2 =
(Volitilityit −aˆ i0 −aˆ 1 Volatilityit−1 )2 , and σˆ ε,Est is obtained n−2 (Volitilityit −aˆ i0 −aˆ 1 EstimatedVolatilityit−1 )2 . This we do for n−2
from the expression
= each variable in each regression. Even though the discussion above outlines the primary empirical methodology that we believe is appropriate, we do attempt to satisfy other robustness curiosities of the reader. A reader could make an argument that volatility can in fact determine institutional structure. In particular, this makes sense for economies that are highly dependent
2 σˆ τ,Est
123
192
J. A. Edwards, F. C. Thames
on their export markets, especially natural resource exports. Economies such as these are quite susceptible to commodity price fluctuations, and their volatility levels would reflect such movements in these prices. Many of these economies are also known to have less democratic regimes. When times are good and world resource prices are high, these economies may actually become increasingly autocratic, hence driving the value of the Polity2 variable down. To this end, we instrument political development by running an OLS regression of the Polity2 variable on country-specific fixed effects, food exports, fuel exports, manufactures exports, and metals exports—all export variables as a fraction of a country’s GDP. The export variables subtract out that part of Polity2 attributable to these specific export markets, and the fixed effects control for a country’s geographical proximity to natural resources. (It is commonly believed that country-specific fixed effects are exogenous, hence subtracting their effects out of Polity2 may be overkill. However, it can be argued that they would serve as a proxy for those natural resource effects that are endogenous, but for which we simply do not have data. To this end, we want to err on the side of overkill to more substantively prove the quality of our model. Regardless, the exclusion of fixed effects from the first-stage regression did not alter our results in any significant way.) Since exports can be highly influenced by global economic conditions, we also include world oil price volatility and world GDP growth volatility. We then use the residuals from this regression as our political development instrument. Another robustness argument hinges on the fact that the volatility and GDP data has a rightward skew, and therefore our results could be driven by outliers. To check the robustness of our results in the face of possible outlier problems, we use a natural log form of volatility, and perform an algorithm that attempts to locate the per capita GDP transition point in the relationship between political development and volatility. The reader will find the results of both robustness tests below in the section just before the conclusion.
4 Results The results in Table 1 from the top are those of the coefficient estimates for the control variables from the second-stage regressions of each control group permutation. (All first-stage results are available upon request.) Near the bottom of this table are the estimates for the variables of greatest importance, i.e., Polity2, GDP, and their interactions. In this table we find that high world growth volatility, high oil price volatility, and countries with low population growth rates and schooling, will all increase a country’s level of volatility. For the most part, these outcomes would fit any reasonable priors that one might have regarding their effects on volatility. Most concerting, and in agreement with almost all other studies, is that regardless of the model, an increase in the level of political development does indeed lower a country’s level of economic volatility. Although the interaction with political development and GDP is not significant, what is more important is the statistical significance of the complete marginal effect of Polity2 on volatility across all levels of GDP. Simply evaluating the point
123
Growth volatility and the interaction between economic and political development
193
Table 1 Second-stage within regression coefficient estimates of the variables used as controls Variable
Model (1)
Model (2)
Volatilityit−1
0.106*
0.081
0.0707
0.0018
(0.069)
(0.283)
(0.338)
(0.978)
Investmentit Population growthit Schoolingit Opennessit Inflationit
Model (3)
Model (4)
−0.0114
−0.0168
−0.0168
(0.583)
(0.421)
(0.435)
−1.5067**
−1.5196**
−1.6060**
(0.025)
(0.024)
(0.017)
−0.5552**
−0.3893**
−0.4047**
(0.000)
(0.016)
(0.014)
−0.0004
−0.0018
0.0009
(0.949)
(0.790)
(0.903)
−0.0003
−0.0002
−0.0003
(0.458)
(0.579)
(0.497)
0.1172**
0.1469**
World volatilityt Oil volatilityt
(0.014)
(0.016)
0.0165**
0.0155**
(0.004)
(0.006) −0.0137
FDIit
(0.550) −0.0056
GCFit
(0.202) Polity2 GDP Polity2* GDP Constant
−0.0817**
−0.0688**
−0.0598**
−0.0774**
(0.000)
(0.007)
(0.018)
(0.005)
−0.0581
0.0292
0.0373
0.048
(0.484)
(0.756)
(0.690)
(0.675)
0.0063
0.007
0.0077
0.0107
(0.412)
(0.423)
(0.373)
(0.322)
2.781**
7.2110**
6.0190**
6.0119**
(0.000)
(0.000)
(0.000)
(0.000)
# Countries
73
73
73
73
# Observations
2117
2117
2117
1896
R2
0.017
0.093
0.102
0.115
Correction factor
1.062
1.202
1.201
1.157
* Indicates significance at 10%, ** indicates significance at 5%. All p values were calculated using robust standard errors, and were corrected for first-stage estimation bias using the correction factor in the bottom row
estimates in Table 1 tells us nothing of the range of incomes over which Polity2 is significant. Table 2 lists these marginal effects for per capita real income levels from zero to $15,000, in $1,000 increments. (We did not table the results beyond $15,000 simply because, as will be obvious, the marginal effect continues to be insignificant beyond
123
194
J. A. Edwards, F. C. Thames
Table 2 Marginal effects of Polity2 on volatility for income levels up to $10,000, including 90% confidence interval GDP Model (1) L.B.
Model (2)
ME
U.B.
L.B.
Model (3) ME
U.B.
L.B.
Model (4)
ME
U.B.
L.B.
ME
U.B.
0
−0.115 −0.081 −0.047
−0.110 −0.068
−0.026 −0.101 −0.059 −0.018 −0.123 −0.077 −0.031
1
−0.103 −0.075 −0.047
−0.098 −0.061
−0.024 −0.088 −0.052 −0.015 −0.106 −0.066 −0.026
2
−0.097 −0.068 −0.040
−0.091 −0.054
−0.017 −0.080 −0.042 −0.008 −0.097 −0.055 −0.014
3
−0.095 −0.062 −0.029 − 0.090 −0.047
4
−0.098 −0.056 −0.014 −0.092
5
−0.101 −0.049
6
−0.005 −0.077 −0.036
0.005 −0.094 −0.045
0.004
−0.040
0.011 −0.079 −0.028
0.021 −0.096 −0.034
0.027
0.002
−0.096 −0.033
0.029 −0.082 −0.020
0.040 −0.100 −0.027
0.052
−0.107 −0.043
0.020
−0.101 −0.026
0.048 −0.086 −0.013
0.060 −0.105 −0.012
0.079
7
−0.112 −0.036
0.038
−0.107 −0.019
0.068 −0.092 −0.005
0.081 −0.111 −0.002
0.106
8
−0.118 −0.030
0.057
−0.113 −0.012
0.089 −0.097
0.002
0.102 −0.117
0.008
0.134
9
−0.124 −0.021
0.075
−0.120 −0.005
0.110 −0.103
0.010
0.123 −0.123
0.019
0.162
10
−0.130 −0.017
0.094
−0.127
0.001
0.131 −0.109
0.018
0.145 −0.130
0.030
0.190
11
−0.136 −0.011
0.113
−0.134
0.009
0.152 −0.115
0.025
0.167 −0.136
0.040
0.218
12
−0.142 −0.004
0.132
−0.141
0.016
0.173 −0.121
0.033
0.189 −0.143
0.051
0.246
13
−0.148
0.001
0.151
−0.146
0.023
0.194 −0.128
0.041
0.210 −0.150
0.062
0.274
14
−0.155
0.007
0.170
−0.155
0.030
0.216 −0.134
0.049
0.232 −0.157
0.072
0.303
15
−0.161
0.014
0.189
−0.163
0.037
0.237 −0.140
0.056
0.254 −0.164
0.083
0.331
this income level; hence, the pertinent results seem to be only in the zero to $15,000 range.) The first column lists the respective per capita GDP levels; the first column under each model lists the lower-bound estimates of the 90% confidence interval for function (1.1); the second column lists the estimates of the mean marginal effect, and column three lists the upper bound of the 90% confidence interval. It is the upper bound that we are most interested in. For all models, the upper-bound becomes insignificant between $2,000 and $5,000. The marginal effect for the uncontrolled regression (1) statistically becomes zero between $4,000 and $5,000. The model (2) effect becomes insignificant between $3,000 and $4,000, while the model (3) and (4) effects become insignificant between $2,000 and $3,000. The lower bound estimate remains below zero over all income levels. The mean of the marginal effect crosses zero between $12,000 and $13,000 for model (1), $9,000 and $10,000 for two, and between $7,000 and $8,000 for models (3) and (4). But, extra insight can be gained by looking at a plot of these effects. Like the estimates in Table 2 and Fig. 1 shows that the lower bound estimate remains below zero over all income levels, while the upper bound estimate crosses zero into positive levels between $2,000 and $5,000. The mean of the marginal effect does not cross zero until nearly $10,000 per capita income. One interesting thing to note about the plots in Fig. 1 is that no matter what the conditioning set, the lower bound never converges to zero—in fact, just the opposite occurs. This indicates that never will political development significantly lower economic volatility for relatively wealthy nations.
123
Regression of Model (2)
.3 0 -.3
Marginal Effect on Volatility
-.6
.3 0 -.3
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Per Capita Real GDP in Thousands
Per Capita Real GDP in Thousands
Marginal Effect of Polity2 on Growth Volatility
Marginal Effect of Polity2 on Growth Volatility
Regression of Model (3)
Regression of Model (4)
.3 0
0 -.3
-.6
-.3
Marginal Effect on Volatility
.6
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
.3
.6
.6
Marginal Effect of Polity2 on Growth Volatility
Regression of Model (1)
-.6
Marginal Effect on Volatility
195
Marginal Effect of Polity2 on Growth Volatility
-.6
Marginal Effect on Volatility
.6
Growth volatility and the interaction between economic and political development
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Per Capita Real GDP in Thousands
Per Capita Real GDP in Thousands
Fig. 1 Marginal effects estimated over income levels
Table 3 More precise income measures for the insignificance of Polity2 on volatility Model (1)
Model (2)
Model (3)
Model (4)
Maximum income
$4,850
$3,340
$2,660
$2,800
# Below max. inc.
44
39
36
36
Percent of sample (%)
60.2
53.4
49.3
49.3
Maximum income is the level of income above which, Polity2 has no effect on volatility. # Below max. inc. is the number of countries at and below this level, and percent of sample is the percentage of countries at and below this level
Table 3 provides measures of the level of income that are more precise than the delineation above. What this table shows is that the actual income spread between (1) and (4) is only a little over $2,000. The marginal effect for the uncontrolled regression (1) statistically becomes zero at $4,850. Model (2)s effect becomes insignificant at $3,340, while the model (3) and (4) effects become insignificant $2,660 and $2,800, respectively. The difference between models (3) and (4) is less than $150, even though the sample for (4) contains nearly 10% fewer observations than for (3). The percentage of countries that would benefit from political development ranges from 49 to 60%–only an 11% difference. Furthermore, there is only about a 14% difference between model (2), and models (3) and (4). Overall, nearly 40 countries in our sample of 73 would have more stable economies by further political development.
123
196
J. A. Edwards, F. C. Thames
5 Robustness checks In this section, we employ two additional checks of robustness for the results depicted above. The first robustness check involves subtracting-out the effect that exportdependent economies with relatively concentrated export sectors may have on political development. As explained earlier, doing this should to some extent purge the influence that volatility generated from these markets would have on political development. To this end, as an instrument for political development, we use the residuals from a regression of Polity2 on proxies for a nation’s resource wealth, sector-specific export dependency, and volatility in global demand. A case can be made that for these countries, volatility can feed back into political development. If this is true, we would perhaps find significant changes in our results. The results of the Polity2 regression with p-values in parentheses are (coefficient estimates for the constants for each country are available upon request) Polity2 = 0.010 × Food Exports + 0.052 × Fuel Exports (0.581)
(0.003)
+ 0.099 × Manufacturing Exports − 0.049 × Metals Exports (0.000)
(0.073)
− 0.328 × World Volatility − 0.052 × Oil Price Volatility . (0.000)
(0.000)
What we find from the first-stage estimates of the Polity2 variable is that in a statistically significant sense, countries with high fuel and manufacturing exports as a percentage of GDP tend to have more democratic regimes, while those that export metals have less democratic regimes. Furthermore, volatility in world growth rates and world crude oil prices tends to reduce democratization. The R-squared of this regression is quite high. The fixed effects and these six variables explain 73% of the variation in Polity2. To this end, 27% is left unexplained, and it is this 27% that we use to instrument for political development. The second robustness check involves reducing the rightward skew in the distribution of volatility and perhaps per capita GDP. Of course the method of choice would be to take the natural logs of each variable, rerun a regression of (1), and evaluate the marginal effects using this log–log form. However, a problem here arises with the functional form for the marginal effect of volatility with respect to political development. If we just simply took natural logs of volatility and GDP in (1), we would have the form ln(Volatilityit ) = bi0 + b1 ln(Volatilityit−1 ) + b2 Polity2it + b3 ln(GDPit ) + b4 Polity2it ∗ ln(GDPit ) + b5 X it + eit .
(2)
The marginal effect of volatility with respect to political development would then be, ∂Volatilityit = Volatilityit ∗ (b2 + b4 ln(GDPit )). ∂Polityit
123
(2.1)
197
.3 .2
Density
.2 0
0
.1
.1
Density
.4
.3
Growth volatility and the interaction between economic and political development
0
10
20
30
Linear form of Volatility
40
50
-6
-4
-2
0
2
4
Logarithmic form of Volatility
Fig. 2 Histograms of volatility
The function (2.1) is highly nonlinear and not very practical when trying to find a cutoff point in per capita GDP whereby political development becomes ineffective. However, as one purpose of this study is to give applied economists the wherewithal to model GDP-related structural changes in the volatility/political development relationship, a more parsimonious application of this robustness check is possible. To this end, we assume the simpler form of (2) ln(Volatilityit ) = bi0 + b1 ln(Volatilityit−1 ) + b2 Dτ − Polity2it +b3 Dτ + Polity2it + b4 X it + eit
(3)
In this form, Dτ − is a dummy variable that equals one up to a chosen breakpoint in per capita GDP level τ , and zero otherwise, while Dτ + is a dummy variable that equals one after per capita level τ and zero otherwise. (This form is similar to that used for a Chow-type test for time series heterogeneity, only applied here in panel data and using per capita GDP as the breakpoint determinant instead of year.) In an algorithmic fashion, we simply move per capita GDP levels upward by $1,000 increments in an attempt to locate the GDP breakpoints as shown in Table 2. Therefore, we will track the coefficient b2 across these income levels and locate the level of per capita GDP whereby b2 becomes insignificant. Like the results found earlier for the purely linear model, economists using the natural log form of volatility will be able to construct a dummy variable that represents the breakpoint we find and simply incorporate an interaction with political development into their model(s). Furthermore, if the breakpoint(s) found using this algorithm coincide with those found earlier, we can conclude that for the most part, our results are robust to this alternate volatility specification. Figure 2 below tells us the potential importance for assuming that outliers may actually be affecting our estimates. The left-hand panel shows the distribution of volatility in its linear form, while the right-hand panel shows the distribution of the natural log of volatility. And, even though the linear form is quite common in the literature, it certainly pays to investigate the estimates generated from the logarithmic form as well.
123
198
J. A. Edwards, F. C. Thames
Table 4 Marginal effects of Polity2 on volatility using the natural log of volatility and instruments GDP Natural log models
Instrumented models
Model (2) L.B.
ME
Model (3) U.B.
L.B.
ME
Model (2) U.B.
L.B.
Model (3)
ME
U.B.
L.B.
ME
U.B.
0
−0.039 −0.023 −0.006 −0.038 −0.021 −0.003 −0.115 −0.066 −0.018 −0.109 −0.063 −0.016
1
−0.033 −0.019 −0.004 −0.031 −0.017 −0.002 −0.097 −0.053 −0.010 −0.099 −0.056 −0.014
2
−0.033 −0.019 −0.004 −0.030 −0.016 −0.001 −0.080 −0.041 −0.001 −0.091 −0.050 −0.009
3
−0.034 −0.020 −0.005 −0.031 −0.017 −0.002 −0.066 −0.028
4
−0.029 −0.015
0.000 −0.026 −0.012
0.002 −0.053 −0.015
0.022 −0.081 −0.037
0.006
5
−0.029 −0.015
0.000 −0.026 −0.012
0.002 −0.042 −0.002
0.037 −0.080 −0.031
0.017
0.009 −0.085 −0.044 −0.002
6
−0.028 −0.014
0.000 −0.025 −0.011
0.003 −0.032
0.010
0.053 −0.080 −0.024
0.030
7
−0.027 −0.013
0.001 −0.025 −0.011
0.003 −0.024
0.023
0.070 −0.080 −0.018
0.043
8
−0.027 −0.013
0.001 −0.025 −0.011
0.003 −0.017
0.035
0.089 −0.082 −0.012
0.057
9
−0.027 −0.013
0.001 −0.025 −0.011
0.003 −0.011
0.048
0.108 −0.084 −0.005
0.072
10
−0.027 −0.013
0.001 −0.025 −0.011
0.003 −0.005
0.061
0.128 −0.086
0.087
0.000
11
−0.027 −0.013
0.001 −0.025 −0.011
0.003
0.000
0.074
0.148 −0.088
0.007
0.102
12
−0.027 −0.013
0.001 −0.024 −0.010
0.004
0.005
0.087
0.169 −0.091
0.013
0.118
13
−0.027 −0.013
0.001 −0.025 −0.011
0.003
0.010
0.100
0.189 −0.094
0.019
0.133
14
−0.028 −0.014
0.000 −0.025 −0.011
0.003
0.015
0.112
0.210 −0.097
0.026
0.149
15
−0.028 −0.014
0.000 −0.025 −0.011
0.003
0.020
0.125
0.231 −0.100
0.032
0.165
The coefficient estimates listed in this table are those of parameter b2 in (3). Since the general dynamic of the marginal effect depicts an upward trend as GDP increases and becomes insignificant early on (as shown in Fig. 1), the latter part of the relationship with GDP depicted by the parameter b3 would always be insignificant. To this end, we only evaluate estimates of b2 in order to capture the “significant” portion of this relationship
Listed in Table 4 below are the estimates of the mean, upper, and lower, 90% confidence intervals for regressions of models (2) and (3), but one set of estimates are attained from the natural log form of volatility, and the other from the use of the Polity2 instrument. Essentially, this table is a reflection of Table 2. We do not include estimates from models (1) and (4) simply for brevity, because they add nothing to analysis, and because most applied researchers use a standard “base” conditioning set such as those reflected in models (2) and (3), to which they add variables of interest. If we hark back to Table 2, we found that for model (2), the upper bound crosses zero between income levels 3 and 4 (in thousands), and the mean crosses zero between income levels 9 and 10. If we use the natural log specification of volatility, the upper bound results are the same, but the mean effect never crosses zero—i.e., it stays negative over the entire range of income levels. Regardless, like the Table 2 results, this still implies that it would be necessary to delineate political developments’ impact on volatility in the $3,000–$4,000 income range. For the model (3) results, both the mean and upper bounds differ from those in Table 2, but again, not a drastic difference. While the mean continues to stay negative over the range of income levels, the upper bound now crosses between income levels 3 and 4, while those in Table 2 crossed between levels 2 and 3. Nevertheless, changing the structural form of volatility from
123
Growth volatility and the interaction between economic and political development
199
a linear form to a logarithmic form does not seem to have produced results that are much different in the sense that it remains necessary to account for a structural break in the Polity2 effect at these lower levels of income per capita. The opposite is the case if we instrument the Polity2 variable. The upper bound of the instrumented model (2) regression now crosses zero between income levels 2 and 3, while the upper bound of model (3) crosses at levels 3 and 4; this outcome is the reverse of that in Table 2. Even so, these breaks are largely consistent in that regardless of the instrumentation of the Polity2 variable, a significant change in the marginal effect does in fact occur between income levels of $2,000 and $4,000. Having said this, the mean certainly crosses zero thousands of dollars away from the results in Table 2, but again, it is the upper bound that is the relevant statistic for our purposes. In general, however, we do believe that these results provide strong support that our main conclusions are robust to both outliers and endogeneity of the Polity2 variable. To this end, it now becomes much more comfortable for a future researcher investigating the volatility/political development relationship to dichotomize their data at about $3,000 per capita real GDP when using a more comprehensive conditioning set. 6 Conclusion While scholars remain divided upon the effect of democracy on growth, there appears to be a consensus that democracy reduces growth volatility. In fact, scholarly research has clearly identified both economic and political development as critical factors that negatively affect the level of volatility. This article sought to determine whether these two causes worked together. Using a novel empirical approach to studies of growth volatility, we found that the effect of higher levels of democratic institutionalization on growth volatility is conditioned by the level of economic development. Specifically, we find that increasing the strength of democratic political institutions will have an independent effect on growth volatility at lower levels of development. There is no effect of increasing democratization in wealthier, more economically developed countries. While the independent effect of democracy is not continuous across all levels of development, our results show that well over half of the countries in our sample would benefit from increased democratization. References Acemoglu D, Zilibotti F (1997) Was prometheus unbound by chance? Risk, diversification, and growth. J Polit Econ 105(4):709–751 Aghion P, Banerjee A, Piketty T (1999) Dualism and macroeconomic volatility. Q J Econ 114(4):1359– 1397 Almeida H, Ferreira D (2002) Democracy and the variability of economic performance. Econ Polit 14(3):225–257 Anderson TW, Hsiao C (1982) Formulation and estimation of dynamic models using panel data. J Econom 18(1):47–82 Arellano M, Bover O (1995) Another look at the instrumental-variable estimation of error-component models. J Econom 68(1):29–52 Arellano M, Bond S (1991) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev Econ Stud 58(2):277–297
123
200
J. A. Edwards, F. C. Thames
Baba S (1997) Democracies and inefficiency. Econ Polit 9(2):99–114 Barro R, Lee J (2000) International data on educational attainment: updates and implications. CID Working Paper 42 Baum M, Lake D (2003) The political economy of growth: democracy and human capital. Am J Polit Sci 47(2):333–347 Bekaert G, Harvey CR, Lundblad C (2006) Growth volatility and financial liberalization. J Int Money Finance 25(3):370–403 Black D (1948) On the rationale of group decision-making. J Polit Econ 56(1):23–34 Blundell R, S Bond (1998) Initial conditions and moment restrictions in dynamic panel data models. J Econom 87(1):115–143 Brambor T, Clark WR, Golder M (2005) Understanding interaction models: improving empirical analyses. Polit Anal 14:63–82 Calderon C, Loayza N, Schmidt-Hebbel K (2005) Does openness imply greater exposure? The World Bank, Policy Research Working Paper Series 3733 Chandra K (1999) The ethnification of the party system in Uttar Pradesh and its consequences. In: Ramashray R, Paul W (eds) Indian politics and the 1998 election. Sage Publications, New Delhi Dahl R (1972) Polyarchy: participation and opposition. Yale University Press, New Haven Downs A (1957) An economic theory of democracy. Harper Collins, New York Dreze J, A Sen (1989) Hunger and public action. Clarendon Press, Oxford Durham JB (1999) Economic growth and political regimes. J Econ Growth 4(1):1381–4338 Edwards J (2007) Dynamics of the cross-country growth/volatility relationship. Glob Econ J 7(2) Edwards J, Sams A, Yang B (2006) A refinement in the specification of empirical macroeconomic models as an extension to the EBA procedure. Top Macroecon 6(2) Edwards J, Thames F, Edwards M (2006) Measuring the dissemination of volatility across levels of development. Top Macroecon 6(2) Feng Y (1997) Democracy, political stability and economic growth. Br J Polit Sci 27(3):391–418 Gerring J, Bond P, Barndt W, Moreno C (2005) Democracy and economic growth: a historical perspective. World Polit 57(3):323–364 Guiso L, Sapienza P, Zingales L (2004) The role of social capital in financial development. Am Econ Rev 94(3):526–556 Henisz WJ (2000) The institutional environment for multinational investment. J Law Econ Organ 16(2):334–364 Helliwell J (1994) Empirical linkages between democracy and economic growth. Br J Polit Sci 24(2):225– 248 Heo U, Tan A (2001) Democracy and economic growth: a causal analysis. Comp Polit 33(4):463–473 Hoover K, Perez S (2004) Truth and robustness in cross-country growth regressions. Oxf Bull Econ Stat 66(5):765–798 Islam N (1995) Growth empirics: a panel data approach. Q J Econ 110(4):1127–1170 Knack S, Keefer P (1995) Institutions and economic performance: cross-country tests using alternative institutional measures. Econ Polit 7(3):207–227 Koren M, Tenreyro S (2007) Volatility and development. Q J Econ 122(1):243–287 Kormendi R, Meguire P (1985) Macroeconomic determinants of growth: cross country evidence. J Monet Econ 16(2):141–163 Kose M, Prasad E, Torrones M (2006) How do trade and financial integration affect the relationship between growth and volatility. J Int Econ 69(1):176–202 Leblang D (1996) Property rights, democracy and economic growth. Polit Res Q 49(1):5–26 Leblang D (1997) Political democracy and economic growth: pooled cross-sectional and time-series evidence. Br J Polit Sci 27(3):453–466 Levine R, Renelt D (1992) A sensitivity analysis cross-country growth regressions. Am Econ Rev 82(4):942–963 Lipset SM (1959) Some social requisites of democracy: economic development and political legitimacy. Am Polit Sci Rev 53(2):69–105 Lipset SM (1981) Political man: the social bases of politics. Johns Hopkins University Press, Baltimore Lohman S (1999) What price accountability? The Lucas Island model and the politics of monetary policy. Am J Polit Sci 43(2):396–430 Mankiw G, Romer D, Weil D (1992) A contribution to the empirics of economic growth. Q J Econ 107(1):407–437
123
Growth volatility and the interaction between economic and political development
201
Mobarak AM (2001) Why democracies are more stable: models of policy choice and political conflict (mimeograph). University of Colorado, Boulder Mobarak AM (2005) Democracy, volatility, and economic development. Rev Econ Stat 87(2):348–361 Munk G, Verkuilen J (2002) Conceptualizing and measuring democracy: evaluating alternative indices. Comp Polit Stud 35(1):5–34 Nordhaus W (1975) The political business cycle. Rev Econ Stud 42(2):169–190 Nooruddin I (2003) Credible constraints: political institutions and growth rate volatility. Doctoral Dissertation, Dept. of Political Science, Univ. of Michigan Olson M (1982) The rise and decline of nations: economic growth, stagflation and social rigidities. Yale University Press, New Haven Pastor M, Sung JH (1995) Private investment and democracy in the developing world. J Econ Issues 29(1):223–243 Polity IV Project (2004) Polity IV Dataset. (Computer file; version p4v2004) Center for International Development and Conflict Management, University of Maryland, College Park Przeworski A (1991) Democracy and the market; political and economic reforms in Eastern Europe and Latin America. Cambridge University Press, New York Przeworski A, Limongi F (1993) Political regimes and economic growth. J Econ Perspect 7(3):51–69 Przeworski A, Alvarez M, Cheibub JA, Limongi F (2000) Democracy and development: political institutions and well-being in the World, 1950–1990. Cambridge University Press, New York Quinn D, Woolley J (2001) Democracy and national economic performance: the preference for stability. Am J Polit Sci 45(3):634–657 Rodrik D (1997) Democracy and economic performance. Mimeo, John F. Kennedy School of Government, Harvard University Rodrik D (2000) Participatory politics, social cooperation, and economic stability. Am Econ Rev 90(2):140–144 Sala-I-Martin X (1997) I just ran two million regressions. Am Econ Rev 87(2):173–183 Sirowy L, Inkeles A (1990) The effects of democracy on economic growth and inequality: a review. Stud Comp Int Dev 25(1):126–158 Tavaresa J, Wacziarg R (2001) How democracy affects growth. Eur Econ Rev 45(8):1341–1378 Wittman D (1989) Why democracies produce efficient results. J Polit Econ 97(6):1395–1424 Wittman D (1995) The myth of democratic failure: why political institutions are efficient. University of Chicago Press, Chicago World Bank (2006) World Development Indicators [CD-Rom]. World Bank, Washington, DC
123