The Impermanence of Democracy: Endowed Wealth Versus Wealth Creation William Gerken∗, Naveen Khanna†, and Tianpeng Zhou‡ August 14, 2017

Abstract In this paper we argue that countries are constantly struggling between autocratic and democratic forces and which regime form is dominant at which time is determined by whether the country’s prosperity is dependent more on endowed assets or on human capital. Since endowed assets can be expropriated by a ruling entity without much loss in value, countries more dependent on them are likely to be less democratic. However, countries more dependent on human capital have much stronger incentives to establish credible democratic institutions to limit their ability to expropriate future wealth to induce wider participation in wealth creation. Our empirical tests strongly support this hypothesis. We document that democracy levels are time-varying because the relative contributions from the two sources while countryspecific are time-varying due to macro forces, and are, thus, not captured by country fixed effects. We further document that countries that are more dependent on endowed wealth are more likely to have stronger militaries even in the absence of outside threats and are more likely to be autocratic. Finally, our model provides a new explanation for why there are democracy waves, why countries move in and out of democracy, and why some rich countries are not democratic while some poor ones are.

∗ University of Kentucky, Department of Finance and Quantitative Methods, 335H Gatton Building, Lexington, KY 40506; email: [email protected]. † Michigan State University, Department of Finance, 313-314 Eppley Center, East Lansing, Michigan 48823; email: [email protected]. ‡ Michigan State University, Department of Finance, 306 Eppley Center, East Lansing, Michigan 48823; email: [email protected].

1.

Introduction The determinants of democracy have been a subject of debate as far back as Aristotle.1

The view ascribed to him is that countries that are rich in wealth and human capital are more likely to develop democratic institutions needed to support wider political participation. This view has strong empirical support from scholars like Barro (1999), Barro and Lee (2001), Glaeser, Porta, de Silanes, and Shleifer (2004), Gennaioli, Porta, de Silanes, and Shleifer (2013) among others, and is consistent with the belief that building strong independent institutions requires both resources and a credible commitment that these institutions will be kept independent. To Aristotle, such a commitment could come from an enlightened populous that values democracy. On the other end of the spectrum is another strongly held view, that democratic institutions are a pre-requisite to economic prosperity and not the result. In a series of papers, such as Acemoglu and Robinson (2001), Acemoglu, Johnson, Robinson, and Yared (2005) and Acemoglu and Robinson (2012), the authors argue that democratic institutions, by constraining governmental power, strengthen property rights and reduce the risk of excessive expropriation. Such constraints encourage wider participation in profitable long term investments like education, entrepreneurship, research and development etc., which require credible and lasting assurance that the participants will be able to benefit from such activities. Without credible constraints on governmental ability to expropriate future rents, wider participation is unlikely to occur, making society less prosperous. Support for this view also comes from Huntington (1968), Fukuyama (2011, 2014), North and Thomas (1973), and North, Wallis, and Weingast (2009). However the debate remains unsettled as some important questions still need to be answered. If prosperity is the driver, why are there significant cross-country difference and time series variation in the level of democratization even after controlling for prosperity? What drives some prosperous governments to voluntarily constrain their authority while others 1

See for example Lipset (1959).

1

choose not to? Why do even prosperous democratic countries slip back into autocratic or single-party rule and vice versa? What explains democracy waves, and do democracies evolve organically or outside intervention can speed up the democratization process? Acemoglu and Robinson (2001) suggest that the observed cross-sectional variation is likely driven by crosscountry differences in geography, making some exogenously better candidates for democratic institutions. While that is a compelling explanation for cross-country variation, it is not obvious how a time-invariant variable like geography can also be consistent with within-country time variation in democracy levels and/or democracy waves. In this paper we argue that a number of these questions can be answered and the differences between the two views reconciled if, in addition to cross-sectional differences in prosperity levels and in geography, we also condition on the sources of a country’s prosperity and their cross-sectional and time variation. If some sources are more likely to push a country toward democracy while others toward autocracy, then the relative value of contributions coming from each source may determine how democratic a country is likely to be. Also if over time, the relative importance of the sources vary then so will the nature of the regime. This would suggest that countries are constantly struggling between autocratic and democratic forces, and which regime-form is dominant at which time is determined by which source of prosperity is relatively more valuable at that point. This dynamic would help explain why countries move in and out of democracy, why prosperous countries are not necessarily democratic, and why macro shocks to the relative value of a particular source of wealth across countries can cause either democracy or autocracy waves. In the economic growth literature, Sachs and Warner (1999) report that economies with abundant natural resources have tended to grow less rapidly than economies with scarce natural resources. Building on that we argue that an important characteristic of natural resources is that while their value is impacted by macro events, it is relatively independent of who controls the asset and can thus be expropriated by different parties without signifi-

2

cant loss in value.2 This contrasts strongly with assets such as human capital that require generally unobservable effort from an agent and is harder to expropriate without significant loss in wealth creation. For instance, physical assets like oil can be expropriated by a regime in power with limited loss in value,3 and if expropriation is affected through a tax, the tax is more like a lump-sum tax with mostly non-distortionary effect on the amount of existing wealth. Similarly, if expropriation occurs after a forcible regime change, these assets are likely to retain their value in the hands of the new regime. On the other hand, value of human/intellectual capital lies in its ability to create new wealth and depends on the quality and incentives of existing or potential producers in the economy. Producers can improve their quality through long term investments in education, technological know-how, apprenticeships, research and development, etc. but whether they choose to do so depends on how much of the created wealth they expect to keep. Here the possibility of expropriation of wealth through either taxes or other means reduces the incentive of producers to invest in these activities, reducing the amount of wealth they create. Thus, for created wealth the possibility and extent of potential expropriation has significant distortionary effects. The ruling regime is likely to control (or get control of) some/most of the physical assets of the country. The regime has the incentive to also expropriate as much of the new wealth being created while conditioning on the distortionary effects of such expropriation. Taxing human/intellectual output has two competing effects for the regime. On the one hand it reduces the incentive of producers to produce and, potentially also the amount of tax revenue accruing to the ruling entity. However, it also reduces the amount of wealth remaining with the producers, reducing the probability that they will become powerful enough to take power away from the current regime. Thus higher taxes, while reducing the amount of new wealth created (and possibly lower tax revenues), ensure a higher survival probability for the existing 2 While we use natural resources as a measure of endowed wealth, more generally it could also include existing physical assets like factories, real estate, etc. that can be expropriated by a ruling entity without much loss in their value. 3 This will depend on the strength on existing institutions ability to prevent expropriation. For mature democracies, these could be significant, reducing incentives of autocratic powers to take over.

3

regime. Given the long gestation of many activities connected with the creation of new wealth, credible institutions need to be in place to limit the current regime’s ability to change expropriation/tax rates ex-post. Absent such institutions, producers are unlikely to produce. Thus, if the potential for wealth creation is substantial, the regime has an incentive to voluntarily create institutions to constrain the amount it can expropriate after the wealth has been created. Absent an independent guarantor, the constraints are credible only to the extent they are self-enforcing. Such self-enforcing institutions serve another important function which also impacts the amount of wealth created. By limiting the ruling entity’s incentive to abuse power, they also reduce the potential cost to the ruling regime of losing power as the new regime too will be constrained by the same institutions. A reduction in expected cost from losing power should induce the ruling regime to lower taxes and increase wealth creation even though that makes producers richer and more dangerous adversaries. This argument is in line with Rajan and Zingales (2004), who propose an “interest group” theory of financial development. They argue there are time varying factors which determine when interest groups are likely to support financial development through arm-length financial markets as this is likely to create competition for power. They show that countries with open trade are more likely to develop such markets and the level of worldwide open trade is both time varying and exogenous to individual countries. This explains economic and political reversals over time. We document the following results. A higher level of endowed wealth correlates with a lower future level of democratic rights. One percentage increase in endowed wealth decreases future democracy levels by 2 percentage and is statistically significant. Conversely, higher levels of wealth creating opportunities (value of human capital) correlate with higher levels of future democratic rights. We are able to justify this causal relationship by instrumenting country-specific human capital by worldwide wealth creating opportunities as that satisfies the exclusion restriction and is correlated with country-specific opportunities to create

4

wealth. We also show that higher levels of human capital dampen the negative effect of endowed wealth on democracy. Additionally countries that are away from their predicted level of democracy based on the relative value of country’s endowed wealth to potential wealth appear to move towards the predicted level over the next 5-year period. Finally, we test for whether the strength of a country’s military impacts the level of democracy. We document that not only is military expenditure positively correlated with the proportion of endowed wealth and negatively correlated with democracy, but that these relationships exist even when surrounding countries have low military expenditures. This suggests strong militaries are more likely in countries with larger endowed wealth to enable the ruling entity to retain control of country-specific assets by lowering the level of democracy. Our main regression results are stable under several robustness checks. The results are robust against the choice of democracy proxies, and we get similar results using Vanhanen and Freedom House Democracy Indexes as the dependent variable in addition to Policy IV democracy index. Results are also robust against the choice of endowment and human capital proxies. Concerning the possibility that a small subset of countries could bias our results, we perform three subsample tests by excluding African countries, current and former socialist countries and stable democracies, and we get similar results. By regressing democracy on the ratio of human capital and endowed wealth, we are able to show that the results are not driven by unrelated factors simultaneously driving both endowment and human capital. The paper is organized as follows. Section 2 develops a simple theoretical model and gives the main hypotheses to be tested. Section 3 describes the data and provides the summary statistics. Section 4 discusses the empirical testing results. And section 5 concludes.

5

2.

Hypothesis Development

2.1.

A Simple Theoretical Model

To identify testable implications, we develop a very simple model with features we wish to emphasize.4 The economy consists of just two homogeneous risk neutral groups, the ruling entity and producers. the ruling entity can control and manage existing assets but cannot create new wealth. Producers have the technological know-how, expertise and capability to develop new products and services and can grow the economy by creating new wealth.5 The ruling entity can expropriate existing wealth through high taxes or other means without much loss in its value. However, the ruling entity can expropriate potential wealth only through future taxes which significantly reduce producers’ incentive to participate in creating wealth. If producers expect the wealth they create to be expropriated, they are unlikely to make long term investments in education, technological know-how, apprenticeships, specializations, etc. This reduces the value of what the ruling entity can expect to expropriate. To encourage producers to participate, the ruling entity needs to credibly constrain their own ability to expropriate in the future. This requires establishing durable and independent institutions to ensure property rights, rule of law, plurality, etc. However, establishing such institutions is a double-edged sword. While they provide the incentives to increase output, they also allow the producers to retain more of their wealth, increasing their political power and their ability to displace the existing ruling entity. These institutions, though, provide some protection also to these ruling entities when they are out of power, as they constrain the new ruling entity as well. Thus, the ruling entity has to trade off higher tax revenues from an increase in created wealth if it strengthens independent institutions, against the increased probability it will lose power and that a portion of the existing wealth will then be expropriated from them. As long as the relative contribution from created wealth is 4

There are many well developed and enlightening models about governments that deal in much greater depth with many of the issues we gloss over for the purpose of the paper. 5 Maybe producers also have similar management skills. However, given their productive inclinations and skills, their desire for control may be less intense, and their ability to control less honed.

6

high enough; there is a self-enforcing incentive for the ruling entity to keep the institutions independent and strong. However, when the contribution from new wealth decreases, there is an incentive for the ruling entity to surreptitiously (or even openly) reduce the capability of these institutions to provide protection to the producers. Assume an economy has W of existing/endowed wealth which is being controlled by the ruling entity. Assume that without the possibility of future expropriation, producers can produce, in expectation, I of new or future wealth. However, if the strength and autonomy of institutions in place cannot prevent τ of future wealth from being expropriated, future wealth expected to be created reduces to I(1 − f1 (τ )), where f10 > 0. The stronger the autonomy of the institutions is, the lower τ can be.6 A lower τ , while increasing the expected total wealth of the economy, leaves more wealth in the hands of the producers making them politically more powerful and a greater threat to the ruling entity. Thus, we assume the probability the ruling entity retains power is an increasing function of τ , f2 (τ ). Finally, when choosing an optimal τ (tax rate and/or strength of autonomous institutions), the current ruling entity conditions on how much of its existing wealth it expects to lose to the new ruling entity in the event of regime change. The higher this expected loss, the less likely they are to risk losing power and more likely to set higher taxes and/or weaker institutions. Thus, as long as adequate new wealth is being created, both the existing regime and the producers (who would form the new regime in the event they gain power) have an incentive to keep the institutions strong (and τ low). Also, higher strength of institutions under current regime is likely to translate into higher strength also under a new regime, limiting the amount the new regime can expropriate from members of the displaced regime. To reflect this, let the expected portion of loss of endowed wealth to the new regime in the event of a regime change be f3 (τ ), with f30 > 0. For simplicity, assume the discount rate is zero. Thus, the ruling 6 An underlying assumption here is that for the existing regime to credibly signal that it would expropriate only a small percent in the future requires the democratic institutions, rule of law, and other check and balance on arbitrary abuse of power to be much stronger.

7

entity chooses the τ to

max τ

f2 (τ ) [W + τ I(1 − f1 (τ ))] + (1 − f2 (τ )) [W (1 − f3 (τ ))]

(1)

given that fi0 (τ ) > 0 , i = 1, 2, 3. The first term is the expected wealth the ruling entity gets if they stay in power, while the second term is the expected wealth if they lose power. While in power they get the existing/endowed wealth, W , and the taxes on the new wealth.7 However, since the taxes are distortionary, they get only the tax on I(1 − f1 (τ )). In the event they lose power, they expect to keep the existing/endowed wealth less proportion of f3 (τ ), the tax (nondistortionary) the new regime is expected to impose on them. There are many good papers which focus on what each fi (·) should be. For the limited purpose of this model, we simply assume that each fi (·) is equal to τ .8 Thus the new optimization is: τ [W + τ I(1 − τ )] + (1 − τ ) [W (1 − τ )]

(2)

Therefore, the optimal τ solves the first-order condition of τ :

W + 2τ I − 3τ 2 I − 2W + 2τ W = 0

Dividing across by W and letting β =

I W

makes the first-order condition:

3βτ 2 − 2τ (1 + β) + 1 = 0 7

Admittedly new wealth will be created in the future and since the tax τ is on future wealth this is also an expected tax. We dispense with time subscripts given a zero discount rate. 8 While τ is probably an appropriate approximation for f3 (τ ) under the assumption that if existing institutions restrain the current regime to tax at most at τ ; these institutions should impose similar constrains on the new regimes ability to tax existing regime’s assets; τ is unlikely to be the best choice for either f1 (τ ) or f2 (τ ). However, the results that follow can be shown to also extend to fi that are affine transformations.

8

Since this is a quadratic, it provides the following two solutions for the optimal τ , s

" τ ∗ = (1 + β) ±

1 − β + β2 3β

q

# (3)

2

Lemma: It can be shown the optimal solution (1 + β) − 1−β+β gives the minimum, while 3β q 2 the (1 + β) + 1−β+β gives the maximum to (2). So the ruling entity will choose the second 3β solution as their optimal τ ∗ bounded above by 1. Proof. In appendix.  The following theorem gives us the main result for this section. Note, tax rates are inversely related to level of democratization. Theorem: The optimal tax rate is decreasing in the expected amount of wealth that can be created relative to existing wealth; i.e.,

dτ ∗ dβ

≤ 0.

Proof. In appendix.  Thus, as the opportunity to create new wealth increases, the existing regime would like to encourage producers to produce more and does so by decreasing the amount of expected tax on created wealth. However, as the amount of new wealth created increases, so does the incentive to increase taxes ex-post. Thus, to convince producers that the existing regime will not increase taxes ex-post, it has the incentive to self-impose checks and balances through increased autonomy of existing institutions, stronger rule of law, stronger property rights etc. In other words, tax rate is negatively correlated with strength of restraining rules/laws/institutions. If level of democratization is a measure of the strength of checks and balances on the ruling regime, then the ruling entity have a vested interest in strengthening democratic institutions as the economies ability to create new wealth increases. Of course, the reverse is also true. We can see the results graphically in Figure 1 for each agent. Figure 1 shows each agent’s wealth at the optimal τ and wealth at the optimum for each agent over a range of values of β. For low levels of wealth generating human capital, β < 1, 9

the ruling entity’s benefit to retain power dominates, and a corner solution is obtained (i.e. τ = 1). When beta exceeds one, the benefits to lowering the tax rate start to exceed the costs of sharing power and the optimal tax rate drops as beta increases. The overall production of the agents is still lower than the socially optimal level since the tax is non-zero and distortionary.9

2.2.

Testable Implications

To take our theory to the data, we roughly categorize assets into two distinct classes: those that retain their value regardless of who controls them, which we refer to as endowed wealth and those that require specific inputs, such as effort and technological know-how, to make them valuable, which we will call potential wealth. As more of the country’s wealth is endowed the benefits from seizing and retaining power become greater and the cost lower, so that the optimal level of democracy for the existing ruling entity is lower. Given this categorization of assets, we can formulate the following hypothesis from our model: Hypothesis 1 (H1): The higher the proportion of a country’s endowed assets, the lower the equilibrium level of democracy. As more of the country’s wealth depends on producers’ effort/intellectual contribution, the benefit to voluntarily creating credible institutions to convince producers to contribute their talent increases. This constitutes our second hypothesis: Hypothesis 2 (H2): The higher the value of potential wealth that requires involvement of producers, the higher the equilibrium level of democracy. Our third hypothesis follows immediately from the prior arguments: 9

The agents in our model face trade offs similar to those faced by an entrepreneur contracting with a venture capitalist. In the venture capital case, the entrepreneur receives capital and in exchange the venture capitalist receives the right to seize the assets of the firm and fire the entrepreneur. The division of voting rights between the entrepreneur and venture capitalist can create a wide variety of incentives: if the entrepreneur has little power, a hold-up problem will ensue as the entrepreneur will not work hard as she knows that the venture capitalist can seize all her assets. See Khanna and Mathews (2016).

10

Hypothesis 3 (H3): The ratio of potential wealth to endowed wealth will determine the level of democracy in equilibrium. These hypothesis suggest a particular time dynamics that can be used as a consistency test. We should observe that ceteris paribus, countries that are at significantly different democracy levels than forecasted by our model tend to move toward that level over time.10 This gives our fourth hypothesis: Hypothesis 4 (H4): Countries with a level of democracy higher (lower) than supported by its composition of wealth will experience a fall (rise) in democracy towards the forecasted level. An important implication of this hypothesis is that since a democracy is more likely to be stable when the country has the capacity to create wealth, if intervention “gives” democracy to a country whose source of wealth supports autocracy, either those installed in power or a rival third party will exploit this opportunity to weaken independent institutions (if they even exist) and move towards an exploitative regime. Recent examples of such unsuccessful attempts to establish democracies are Afganistan, Iraq and Libya. Our hypothesis suggests that lack of success should not be surprising as these countries are rich in natural resources and candidates for exploitative regimes. This is also consistent with why most democracies in Africa failed within a few years after gaining independence even though they started with established judiciaries, rule of law and functioning government capacity. The same appears to support recent developments in Venezuela and continuing developments in Russia which are both dependent on their oil wealth has slid or are sliding towards autocracy. Hypothesis 5 (H5): Outside intervention is unlikely to establish sustainable democracies in countries that are more dependent on endowed wealth for their prosperity. 10

Given that the ratio is time variant because of macro as well as country specific effects, so is the predicted level of democracy, making this a difficult hypothesis to develop a clean test for.

11

Additionally, our paper provides an explanation for why there may be democracy waves as documented by Huntington (1991) and Markoff (2014). New technologies and new discoveries increase the value of human capital across all countries, while that of endowed wealth is not affected to the same extent, if at all. This gives incentive to move countries to develop democratic institutions to induce producers to create new wealth. Hypothesis 6 (H6): When new technologies emerge, the incentive to develop credible, independent institutions increases across all countries. Thus, there will be a general movement towards democracy. Once these new opportunities are exploited, there will be a movement towards autocracy. We also attempt to understand the relationship between military spending and democracy. Since military expenditures are determined either by the level of external threats or against internal threats that threaten the continuation of the existing regime, we need to identify a proxy which would enable us to distinguish between these differential threats. The proxy we believe is reasonable is the level of military expenditures of surrounding countries. If that is high and so is the country’s, it is difficult to distinguish whether the high expenditure is to safeguard against outside threats or to support an exploitative regime. However, if the country’s military expenditure is significantly higher than its neighbors, then it is more likely that it supports an exploitative system. If it turns out that the country is also more dependent on endowed wealth, it provides further support for our claim that the component of prosperity that is relatively more important determines the political structure of a country.11 Hypothesis 7 (H7): Countries that are more dependent on endowed wealth are more likely to have a stronger military even in the absence of external threats. These countries are also 11

A relevant strand of existing literature emphasizes the role that inter-state warfare played in driving the emergence of modern countries. According to this literature, costly new military weaponry forced countries to build modern fiscal systems to fund armies to survive interstate competition because warfare became much more expensive. See Bean (1973), Hintze (1975), Herbst (2000), and Fukuyama (2014). However, they have not focused on the role that endowed wealth plays in shaping the military power of a country.

12

likely to be more autocratic.

3.

Data To test our theoretical predictions we examine country level data which are available

starting in 1975 through 2010. The data availability varies but our sample includes between 100-180 countries depending on the specific variables used. The data are gathered from publicly available sources cited in Panel A of Table 1. Figure 2 graphs the worldwide average of three key variables over the sample data period. To facilitate comparing them on the same graph, we standardize them. From 1975 to 2010, the worldwide democracy level is increasing persistently as also the worldwide number of patents. In contrast, the contribution of natural resources to GDP on average has been decreasing during this period of time. As our theoretical arguments center around self-enforcing constraints that regimes can credibly impose on themselves, we use several measures of institutions that have been used in previous studies. Glaeser et al. (2004) point out that measures of institutions used by prior works may not be appropriate and likely reflect both constitutional constraints as well as good policies by dictators. This works to our advantage as we want to identify the credible level of constraint (the outcome).12 To measure the extent to which political power is shared, we take our primary measure of democracy from the Polity IV database. This democracy index is an additive eleven-point (0-10) scale where 10 indicates the highest level of democratic institutions (see Table 1 Panel B). The index is comprised of three main components: the presence of effective institutions, constraints on the power of the executive, and guarantee of civil liberties including free and fair election. For a robustness check, we obtain the Political Rights Index from Freedom House. This index measures the degree of freedom in the electoral process, political pluralism and participation, and functioning government capacity. In this database a value of 1 indicates countries 12

We assume that on average a country’s leadership is enlighted. And this assumption tends to bias against finding our results.

13

with free and fair elections, political competition, and autonomy for citizens, while a value of 10 indicates countries in which political rights are basically nonexistent due to extremely oppressive regimes, civil war, extreme violence or rule by warlords. To be consistent with the Polity IV democracy index, we take the opposite sign of the original Freedom House index so that a large value represents a higher level of democracy. We also use the Polyarchy dataset compiled by Tatu Vanhanen, which covers 187 countries over the period 1810 to 2000. The Vanhanen democracy index is computed by multiplying the percentage share of total votes cast in parliamentary elections for smaller parties and independent candidates (competition) and the percentage of population that voted in elections (participation) and dividing the outcome by 100. A larger number of this index indicates a higher democracy level. We find both Freedom House and Vanhanen democracy indexes have high correlations with the Polity IV index (see Table 2). We construct a panel dataset with five year intervals where each cross section unit represents a unique country. Our key explanatory variables are the two different sources of wealth of each country. Since we are trying to explain how the existence/emergence of wealth creating opportunities encourages power sharing, we use a country’s natural resources as a proxy for existing/endowed wealth. From the United Nations Statistics Division (unstats.un.org), we download the GDP and its breakdown at current prices in U.S. dollars. We then estimate a variable we call endowment ratio by aggregating the country’s income from economic activities listed under Section II (Mining and Quarrying) of International Standard Industrial Classification of All Economic Activities, Revision 3.1, and dividing by the country’s GDP.13 Section II comprises of mining of coal and lignite, extraction of peat, extraction of crude petroleum and natural gas, service activities incidental to oil and gas extraction excluding surveying, mining of uranium and thorium ores, mining of metal ores, and other mining and quarrying. As the endowment variable is a ratio, the choice of numeraire does not affect 13

We are implicitly assuming that the country’s annual flows from these activities are reflective of the value of the stock of these resources which would be the proper measure of the relative value of wealth from each source. As a robustness check, we use value of oil reserves to proxy for the stock of endowed wealth in a country.

14

our results. As seen in Table 1 Panel B, for our entire sample, the mean level of contribution from endowed wealth to GDP is 23.11%. As a robustness check, we obtain British Petroleum Statistical Review of World Energy (June 2014) to determine the historical value of oil reserves by country from 1975 to 2010 and use it as the proxy for endowed wealth. Likewise, we use number of patents as a proxy for a country’s dependence on (or ability to exploit) wealth creating opportunities. Unlike mineral or energy resources patents require the participation of human capital in creating wealth and is harder to expropriate owing to the moral hazard issue connected with it. From the U.S. Patent and Trademark office, we obtain the number of patents awarded by country of origin which is determined by the place of residence of the first-named investor.14 To make this measure less sensitive to extreme outliers, we add one and then take the natural logarithm. Like patents, overall educational level of a country’s residence is another measure of its ability to generate wealth and has been effectively used in previous works like Barro (1999).

While the correlation between democracy and education has been well docu-

mented, Acemoglu et al. (2005) point out that the causal relationship is less obvious. For comparability with prior work, we obtain the primary years schooling in the total population of age 25 and above, which is available in five year intervals from Barro and Lee (2001). Previous studies have also suggested that cross-country differences in legal systems may explain difference in political systems. To control for this, we use legal origin dummies from la Porta, de Silanes, Shleifer, and Vishny (1998, 1999). For a country with predominance of endowed wealth, the ruling regime would have the incentive to support a strong military for two reasons. It could be necessary to defend the country against external threats. However, it could also be used by the ruling regime to safeguard its power from internal threats. In our analysis we use the percentage of a 14

Since patent rights do not extend beyond country borders, investors must apply for patent rights in other countries to prevent foreign infringement. The result for these reciprocal patent filings are governed by international patent treaties. The U.S. is a signatory country to most major international patent treaties, such as the Paris Convention and the Parent Cooperation Treaty. The Paris Convention now has 172 contracting member countries. Given the prominence of U.S. in the patent field, the filings listed by the U.S. patent office should be relatively comprehensive.

15

country’s military expenditure to GDP reported by the World Bank as a proxy for the military strength of a country. To measure the likelihood that the military is likely to protect against foreign intervention, we compute the surrounding countries’ average military expenditure and use that as a proxy for level of outside threats. If the neighboring countries also have strong militaries, we are unable to distinguish between these two motives. However, if the surrounding countries have weak militaries then it is more likely the country has a strong military to enable the regime to retain power, especially if the country is also more reliant on endowed wealth. Table 2 lists the pairwise correlations. As expected, democracy is strongly negatively correlated with the percentage of GDP that represents endowed wealth (endowment ratio). It is also negatively correlated with the value of oil reserves in that country. Besides, democracy is positively correlated with number of patents awarded to the country. So for a country with more wealth creating opportunities, its citizens tend to have more political freedom and better ability to keep a larger share of wealth. This is also consistent with our hypothesis H2. Military expenditure’s negative correlation with democracy level and its positive correlation with endowed wealth suggest that higher military expenditures appear to exist in countries which are less democratic and whose prosperity is more dependent on natural resources. All three democracy indexes give similar correlation coefficients mentioned above. We find that the number of patents registered in a country is positively correlated with education which is consistent with the commonly held belief that better education helps build human capital. We also see that a country’s number of patents is positively correlated with the aggregate number of patents registered worldwide, which suggests that a number of countries take advantage of these opportunities. There is cross sectional variation in how much a country exploits these opportunities and that is driven by both geography and the country’s ability/resources to be able to build/maintain/guarantee credible democratic institutions.

16

4.

Results

4.1.

Benchmark Results

We begin by testing hypothesis H1 and H2. We define our dependent variable as the Polity IV measure of democracy five years after we measure each of our independent variables. To examine H1, we first regress the level of democracy on the lagged endowment variable in Table 3 column (1).15 Prior work, such as Lipset (1959), Barro (1999) and others, suggests that the prosperity level can be an important determinant of democracy. So we also include the country’s GDP level in our first specification. We find that one percentage increase in endowment ratio decreases the level of democracy five years ahead by around 0.0982 points. This relationship is statistically significant, which lends support for H1.16 We add the country’s domestic number of patents as a country’s human capital measure to our regression in column (3). The coefficient is positive and significant while the coefficient of endowment ratio remains significantly negative. The magnitude of the effect suggests that a one percentage increase in human capital increases the level of future democracy by about 0.0078 points, all else equal. Other authors, such as Acemoglu et al. (2005), suggest that other long-standing institutions jointly cause both the level of democracy and education or level of human capital. To address the omitted variable problem we add legal origin dummies into our regression in column (5) and find no significant changes in the coefficients of endowment and other measure of human capital. As a robustness check, using the value of oil reserves in a country instead of income from natural resources as a proxy for endowed wealth gives similar results as earlier (columns (2), (4) and (6) of Table 3). In total, our empirical results reported in Table 3 support hypotheses H1 and H2. 15

Because the aggregate GDP in a country moves slowly, we cannot control for the country-level fixed effects additionally in our regression. 16 As reported in Table 8 column (1) instead of using the endowment ratio, using the dollar value of mining and quarry in one year still gives similar results.

17

4.2.

Causation

Despite using a five year lead on our democracy variable to address the issue of causality, we cannot preclude the possibility that the relationship between democracy and human capital is both long term and jointly determined. That is while an increase in a country’s level of human capital apparently increases it’s democracy level five years later, that effect may be spurious if a relatively permanent move towards democratization is the starting point to induce producers to develop human capital but this move towards democratization is gradual as it takes time to build credible institutions. In other words, the causation may well be from democracy to human capital which raises the question as to what caused the move towards democratization in the first place. To address this apparent circularity, we appeal to an instrumental variables approach. We look for an instrument which is independent of a country’s level of democracy, but impacts the value of human capital for all countries. An instrument that appears reasonable is a worldwide change in the value of innovation which should translate into a change in the relative value of human capital to endowed wealth cross-country. We use the aggregate number of patents registered each year worldwide as the instrument for changes in the relative importance of human capital for each country. Whether a country chooses to (or is able to) take advantage of this opportunity for increasing wealth by motivating producers to develop/improve/use their human capital will then be driven by the relative value of its endowed wealth. Countries with relatively larger endowments are less likely to become democratic. We run a two-stage least squares analysis, regressing a country’s number of new patents every year on aggregate number of worldwide patents for the corresponding year. We find that country patent numbers are strongly correlated with worldwide levels and the adjusted R2 of 95% suggesting this has potential to be a valid instrument. In the second stage of the first specification, we find that the results are consistent with our earlier estimates reaffirming our prior conclusions with regard to H1 and H2. The point estimates of the effect for endowment has a decrease of 0.0712 in democracy level for every percentage increase in 18

endowment ratio, while the point estimates of the effect for human capital has an increase of 0.01 in democracy level for a 1% increase in number of patents. As a robustness check, in the second specification, we use oil reserves as the proxy for the endowed wealth. The results are similar to the first specification.17

4.3.

Robustness of Main Result

4.3.1.

Alternative Specifications

To handle the possibility of omitted variables and possible misspecification, we estimate the relationship using alternate functional forms. As a benchmark, column (1) in Table 5 is identical to the column (3) of Table 3 in which we regress Policy IV democracy index on endowment ratio, number of patents (human capital), and log(GDP). In column (2) we include an interaction term between endowment ratio and human capital. The direction of the effects of endowment and human capital on democracy reported in Panel B of Table 5 are still statistically significant and consistent with our earlier results, and the coefficient of endowment ratio is highly significant (p-value is less than 1%). Notably, the coefficient of the interaction term is negative and significant, which implies that a higher level of human capital dampens the negative effect of endowment ratio on democracy. In column (3), we include a lagged dependent variable. Though Plumper, Thoeger, and Manow (2005) caution against the use of lagged dependent variables and unit dummies when not suggested by theory, we use these alternate specification to alleviate concerns that deep seeded institutions may be driving both the dependent and independent variables. While the signs of the endowment ratio and human capital coefficients are consistent with the prior findings, the estimates are not directly comparable in magnitude to the previous ones. The drop in the magnitudes of endowment and human capital coefficients is not surprising given 17

Number of patents generated in countries such as US, Japan, and Germany takes a major share of the aggregate world patents, and therefore they might makes the world patent IV less exogenous to the human capital variable if we include all the countries in the sample. As another robustness check, though not reported here, we take out the countries with the top 10% patent numbers and redo our regression. We obtain similar results.

19

the slow moving nature of the democracy variables. The statistical significance of these coefficients after controlling for lagged democracy is important because it provides evidence that the remaining noise affecting democracy after controlling for past democracy levels cannot be fully captured by the latitude variables and continent dummies as in Acemoglu and Robinson (2001). That is because here the omitted variable, the relative value of endowed and potential wealth is time-varying both because of the variation in wealth creating opportunities and the value of natural resources. Adding an endowment ratio and human capital interaction term to our third specification still gives results which are consistent with those reported in column (2).

4.3.2.

Alternative Proxies

Next, we check to see if our results are driven by our choice of proxies. The first column is identical to the second column of Table 5. We employ the two alternative measures of democracy and our results are qualitatively the same (the second and third columns of Table 6). As patents are an outcome variable and we would ideally want a measure of future value of human capital, we try using a measure of education. Notably, there is a simultaneity issue between democracy and education. Barro and Lee (2001) argues that improvement in education helps the development of democracy while Acemoglu et al. (2005) argues the reverse causality. Therefore we use a five-year lagged education here to disentangle and alleviate this simultaneity issue. In the fourth column, we use the years of education as in Barro and Lee (2001) and again find similar result.18 Since oil is a specific but very politically and economically meaningful commodity worldwide, in the last specification, we use the value of oil reserves instead of the endowment ratio as the proxy for the endowed wealth. Again, we find results consistent with earlier findings. In sum, we do not find evidence that our results are driven by our choice of proxy. 18 We prefer patents over education in our primary specifications because the quality of education can vary greatly form country to country, and our theory makes predictions based on the value of the human capital, which we believe is better captured through patents.

20

4.3.3.

Subsample Testing

We would like our results to be generalizable to as many countries as possible, so we need to be concerned with the possibility that a small subset of countries are driving our results. In Table 7, we check to see if our results are being driven by any particular subsample by excluding sets of likely suspect groups. First, African countries had a high degree of variation in level of democracy during the sample period. In the first specification, we exclude African countries and find that our results are qualitatively similar. Second, given that the socialist countries have distinctly different political systems, we exclude the current and former socialist countries and rerun our main specification. Again, we find no significant changes in our estimates. Third, to ensure that our results are not being driven by wealthy and established democracies such as the U.S., we exclude all countries that stay completely democratic during the sample period. Again, our results are generally unchanged.

4.4.

Equilibrium Results

So far we have documented support for H1 and H2. H3 is a novel prediction. As a benchmark regression, in the first specification we regress democracy on the lagged dollar value of endowment (log of economic activity under Section II (Mining and Quarrying)) and human capital and we find similar results to our earlier regressions: democracy level is negatively related to the value of endowed wealth while it is positively related to the country’s wealth creating opportunities. Both coefficients are statistically significant at 1% significance level. Then, to address the concern that our results could be caused by unrelated factors simultaneously driving both endowments and human capital, we construct a ratio of potential wealth to endowed wealth, β.19 In the second specification in Table 8, the coefficient of β is positive and strongly statistically significant indicating that a country relying more on 19

By construction, we implicitly assume that the average value of each country’s patents are the same. To the degree this is not true, the discrepancy will bias against us finding a relationship.

21

producers’ human capital and less on endowment tends to be more democratic. This result supports H3 suggesting that the relative importance of the two different sources of wealth is an important omitted variable in precious work. Since it is time varying, it is not captured country dummies which is commonly used in the literature.

4.5.

Disequilibrium Results

Another novel empirical prediction, H4, is that countries that are not at their equilibrium level of democracy as predicted by their relative sources of wealth will move towards their equilibrium level overtime. To test H4, we use our model in column (3) of Table 3 with the data from year 1975 to 2005 to predict the level of democracy in five-years (year 2010) and then compare it to the level of democracy that exists in that country in 2010 as measured by Polity IV. We divide the sample into three groups: “Underdemocractic” or those countries at least 3 points out of the zero to ten scale below their predicted level, “Overdemocractic” or those countries at least 3 points above their predicted level, and “Initially Correct” or those countries within 3 points of their predicted level. In each group, we then look at the number of countries that start as Autocratic (0-3), Neutral (4-6), and Democratic (7-10) and examine their transition to the predicted levels at the end of the next five years. In Panel A of Table 9, we see that 86.96% out of 95.65% of the “underdemocratic” countries that were listed as autocracies in Polity IV stay as autocracies in five years. However, 8.70% of them become either neutral or democratic in next five years while only 4.35% changes from democracies to autocracies. In Panel B of Table 9, we see that out of 95.35% of the countries that were listed as “Overdemocractic” countries only 83.72% stay as democracies while 11.63% become either neutral or democratic in five years. In Panel C of Table 9, 92.3% of the “initially correctly predicted” countries keep the same levels of democracy in five years. There are no abrupt jumps or drops in democracy levels for countries in this category. For instance, if a country was initially correct and started as autocracy, it either stays as autocracy or increases its democracy level to neutral state. Similarly a country starting 22

as democracy either stayed as democracy or dropped to the neutral state. These findings are consistent with H4 in that countries in disequilibrium tend to move towards their equilibrium democracy level which is predicted by relative importance of their endowed wealth and wealth creating potentials.20 Table 10 lists some countries with lowest and highest democracy rankings and six other countries that are of interest. We report their current democracy level ( “democ” column), the predicted equilibrium level (“demochat” column), and the difference between them (“error” column). A negative number under “error” column indicates that the country is underdemocratic and, thus, has the potential to become more democratic while a positive number indicates that the country is overdemocratic and has the potential to become less democratic. For instance, countries that come out as significantly underdemocratic include China, Cuba, Iran, Singapore and Venezuela. These countries have relatively lower endowed wealth and are more likely to promote human capital development which would require democratic institutions. To stimulate the producers to invest in education and technological know-hows, their ruling regime should grant more political freedom to their citizens. So we predict these countries to improve their level of democracy. On the other hand, countries that are significantly “overdemocratic” include Greece and Turkey, whose democracy levels are predicted to deteriorate after 2010, and we observe that to be the case for Turkey which appears to be becoming more autocratic. In addition to the statistical support documented for our model, the current geopolitical environment lends further evidence. Our model predicts that positive shocks to the value of endowments can have a destabilizing effect on democracies. This is very evident in the spike in oil prices and the reactions in oil-rich countries like Venezuela and Russia. In Russia, Vladmir Putin has exploited the strong growth in Russian GDP due primarily to increases in oil prices to tighten his control on power. Human Rights Watch has noted that “Putin cracked down on civil society and freedom of assembly” and compared Putin to other despot 20

We look forward only five years as the predicted democracy levels are time varying. Thus any longer window will create additional measurement problems.

23

leaders in Pakistan and Zimbabwe. Our work also has implication for the United States’ role in reconstructing Iraq. Given the large amounts of oil and relatively poorly educated populace, our work suggests that implementing democracy may take a long-time to engineer and then too is unlikely to hold. The same appears true for Afghanistan. Simply put, democracy is likely to hold only in countries where a larger contribution from human capital either exists or is expected to happen because of an increase in the aggregate value of new created wealth given there are basis resources to develop/maintain independent democratic institutions. This is consistent with our hypothesis H5 and H6.

4.6.

Military, Endowed Wealth, and Democracy

In this section we investigate the reason why some countries surrounded by weak neighbors still hold strong military forces and the impact of such an imbalance on democracy. On the one hand, the military power could be built up for the purpose of defending the country against foreign intervention. On the other hand, for a country with more existing/endowed wealth, the ruling regime has the incentive to build a strong military to protect its wealth and retain power. In another word, the regime relies on its army to protect itself from its own citizens. As a proxy for a country’s military strength, we use the percentage of a country’s military expenditure each year. We also compute the surrounding country’s average military expenses and use that as the proxy for external threats. We first test the effect of a country’s military on its democracy in Table 11. In the first specification, we include a five-year lagged military expenditure along with the lagged endowment ratio, human capital and GDP level. The coefficients of endowment and human capital variables are still statistically significant and their signs are consistent with the predictions in H1 and H2. We also find a negative, though insignificant, coefficient of military expenditure, indicating that a country with high military expenditure usually has a low democracy level. In the second specification, we include the interaction term between endowment ratio and 24

military expenditure in addition to the explanatory variables used in our first specification. And we find a negative and strongly significant coefficient of the interaction term. That indicates that the negative effect of endowed wealth on democracy is amplified for a country with higher military expenses. This is consistent with a stronger military strengthening the ruling entity’s power enabling it to expropriate the producers’ wealth. The reported marginal effects of military expenditure on democracy in the second column of Panel B reveal that stronger military tends to be associated with lower democracy level, which is statistically significant at 5% significance level. Lastly, we control for the average military expenditure of the surrounding countries additionally in the third specification. We find that a country’s democracy level is lower if its surrounding countries’ military expenditures are higher. However, after controlling for the surrounding countries’ military expenditures, the country’s own military expenditure’s impact on its democracy level appears to be statistically and economically insignificant. Perhaps a better test to differentiate between the two potential objectives of having a strong military is to sort all countries into two groups depending on the comparison between their own military expenditure and the surrounding countries’ average. In this way, we directly control for the share of military used to enable the ruling entity to expropriate internal producers’ wealth. Table 12 Panel A reports results for countries whose own military expenditure is higher than the average of their neighbors. In our first specification, we regress the democracy level on endowment ratio, number of patents (human capital), log(GDP), and the the variable Difference which is defined as own military expenditure minus surrounding countries’ average. Note that the values of the variable Difference is then non-negative. We find that the coefficients of endowment ratio and human capital are still consistent with our hypotheses H1 and H2, and statistically significant at 5% significance level. Moreover, the negative coefficient of the variable Difference suggests that as a ruling regime’s incentive to have a stronger military appears to be to retain power internally democracy in the country deteriorates. We also get similar results in the second specification after replacing human

25

capital and endowment ratio by the ratio β. We repeat our regressions in Panel A using the countries whose own military expenditure is less than the surrounding countries, and report the results in Panel B. Our regression results are still consistent with the results reported in Panel A. Notably the coefficient of the Difference is less statistically significant than the one reported in Panel A because for a country surrounded by neighbors with strong militaries is more likely to depend against external threats. Then it is not clear as to whether the military is being also used to retain power against internal threats. Lastly, we look into the reason why some countries surrounded by weak neighbors still hold strong military forces and the role the country’s endowed wealth plays in it. We label a country’s military expenditure level to be high (low) if it belongs to the top (bottom) 50% after sorting all countries by their military expenditures. Based on the relation between the own-military expenditure and the surrounding countries’ military expenditure, we separate the countries into 4 groups: (high,high), (high,low), (low, high), and (low, low). The first element in each parenthesis refers to the country’s own military expenditure level and the second element refers to the surrounding countries’ average expenditure level. We are interested in countries that are in the (high, low) state because these countries are surrounded by weak-military neighbors but still maintain a strong army. In Table 13, we use Probit models to analyze the effects of endowment and β on the probability that a country falls into the (high, low) state. The positive coefficient of endowment ratio in the first specification and the negative coefficient of β in the second specification indicate that countries with higher levels of endowed wealth are more likely to maintain a strong military even when surrounded by weak neighbors. Both coefficients are strongly significant at 1% significance level. This is consistent with hypothesis H7.

26

5.

Conclusion We find strong support for our theory that the time varying composition of wealth is an

important determinant of the level of democracy. Specifically, expropriable endowed/existing wealth encourages autocracy, while non-expropriable sources of wealth like human capital encourage democracy. Empirically, we find that a higher level of endowment wealth correlates with a lower future level of democratic rights as predicted by our model Conversely, higher levels of wealth-creating human capital correlates with higher levels of future democratic rights. These results remain after using an instrument based on worldwide levels of innovation. That shows that democracy appears to be caused by the predominance of endowed wealth. We also show that these results are not driven by a particular choice of proxies, specification or subsample of countries. Countries that are away from their predicted level of democracy based on the sources of their wealth tend to move towards the predicted levels after five years. Finally, we show that military expenditure is positively correlated with the importance of endowed wealth and negatively correlated with democracy in a country, and these relationships remain even when surrounding countries are weak. This suggests countries with larger endowed wealth are more likely to maintain a strong army to enable the existing ruling entity to retain power and to be able to expropriate wealth from other entities within the country. As such these countries are likely to be less democratic. The results help explain why democracy is impermanent, why many rich countries are not democratic, why there are democracy waves and what role military may be playing in different countries.

27

Appendix Proof of Lemma Let L (τ ) = τ (1 + τ β(1 − τ )) + (1 − τ )(1 − τ ), which is a continuous function when τ ∈ [0, 1]. The first order derivative is ∂L = −3βτ 2 + 2τ (1 + β) − 1 ∂τ √ Notice that if τ < τH ,

∂L ∂τ

(1+β)−

1−β+β 2 3β

(4)

√ ≡ τL or τ >

(1+β)+

1−β+β 2 3β

≡ τH ,

∂L ∂τ

< 0; if τL < τ <

> 0.

Also notice that L (0) = 0, and L (1) = 1. Because β is defined as β =

I , W

β ∈ [0.1].

Therefore, τL > 0, and τH < 1. Since we have shown that L (τ ) is decreasing when τ ∈ (0, τL ) ∪ (τH , 1), and L (0) = 0, L (1) = 1, we know that

L (τH ) > 1 > 0 > L (τL )

Hence completes the proof of Lemma.

(5)



Proof of Theorem √ ∗

Take the first order derivative of τ (β) =

(1+β)+

1−β+β 2 , 3β

we have

p dτ ∗ ( 12 β − 1) − 1 − β + β 2 p = dβ 3β 2 1 − β + β 2 To show that

dτ ∗ dβ

≤ 0, we need to show that ( 21 β − 1) ≤

p 1 − β + β 2 , which is obvious when

we take the square of both sides. Hence completes the proof of the Theorem.

28

(6)



References Acemoglu, D., S. Johnson, J. A. Robinson, and P. Yared. 2005. From Education to Democracy? The American Economic Review 95:44–49. Acemoglu, D., and J. A. Robinson. 2001. A Theory of Political Transitions. The American Economic Review 91:938–963. Acemoglu, D., and J. A. Robinson. 2012. Why Nations Fail. 1st ed. New York: Crown. Barro, R. J. 1999. Determinants of Democracy. Journal of Political Economy 107:158–183. Barro, R. J., and J.-W. Lee. 2001. International Data on Educational Attainment: Updates and Implications. Oxford Economic Papers 53:541–563. Bean, R. 1973. War and the Birth of the Nation State. Journal of Economic History 33:203–221. Fukuyama, F. 2011. The Origins of Political Order: From Prehuman Times to the French Revolution. 1st ed. New York: Farrar, Straus and Giroux. Fukuyama, F. 2014. Political Order and Political Decay: From the Industrial Revolution to the Globalization of Democracy. 1st ed. New York: Farrar, Straus and Giroux. Gennaioli, N., R. L. Porta, F. L. de Silanes, and A. Shleifer. 2013. Human Capital and Regional Development. The Quarterly Journal of Economics 128:105–164. Glaeser, E. L., R. L. Porta, F. L. de Silanes, and A. Shleifer. 2004. Do Institutions Cause Growth? Journal of Economic Growth 9:271–303. Herbst, J. I. 2000. States and Power in Africa. 1st ed. Princeton: Princeton University Press. Hintze, O. 1975. Historical Essays of Otto Hintze. 1st ed. New York: Oxford University Press. 29

Huntington, S. 1968. Political Order in Changing Societies. 1st ed. New Haven: Yale University Press. Huntington, S. 1991. The Third Wave. 1st ed. Norman: University of Oklahoma Press. Khanna, N., and R. D. Mathews. 2016. Posturing and Holdup in Innovation. The Review of Financial Studies 29:2419–2454. la Porta, R., F. L. de Silanes, A. Shleifer, and R. Vishny. 1998. Law and Finance. Journal of Political Economy 106:1113–1155. la Porta, R., F. L. de Silanes, A. Shleifer, and R. Vishny. 1999. The Quality of Government. Journal of Law 15:222–279. Lipset, M. S. 1959. Some Social Requisites of Democracy: Economic Development and Political Legitimacy. The American Political Science Review 53:69–105. Markoff, J. 2014. Waves of Democracy: Social Movements and Political Change. 2nd ed. New York: Routledge. North, D. C., and R. P. Thomas. 1973. The Rise of the Western World. 1st ed. New York: Cambridge University Press. North, D. C., J. Wallis, and B. R. Weingast. 2009. Violence and Social Orders: A Conceptual Framework for Interpreting Recorded Human History. 1st ed. New York: Cambridge University Press. Plumper, T., V. E. Thoeger, and P. Manow. 2005. Panel Data Analysis in Comparative Politics: Linking Method to Theory. European Journal of Political Research 44:327–354. Rajan, R. G., and L. Zingales. 2004. Saving Capitalism from the Capitalists: Unleashing the Power of Financial Markets to Create Wealth and Spread Opportunity. 1st ed. Princeton, NJ: Princeton University Press. 30

Sachs, J. D., and A. M. Warner. 1999. The Big Rush, Natural Resource Booms And Growth. Journal of Development Economics 59:43–76.

31

Figure 1. Optimal Tax Rate and Wealth

32

Figure 2. Worldwide Trend of Standardized Key Variables

33

Table 1: Data Description Panel A: Variable Definitions Variable Name Description Polity IV Polity IV democracy score Freedom Freedom House democracy score Vanhanen Vanhanen democracy score GDP log of per capita GDP (US$) Endowment ratio Annual income from natural resources divided by GDP Oil Value of oil reserves in 2015 dollar (billions) Human Capital log(1+ number of patents) Worldpatents log(1+number of the total world patents) Education Years of total schooling Legal Origin Legal Origin Dummies Military Military expenditure (% of GDP) Mili surrounding Bordered countries’ mean military expenditure

Source Polity IV Freedom Hall Vanhanen un.org un.org B.P. Report U.S. Patent Office U.S. Patent Office Barro and Lee (2000) LLSV (1999) World Bank World Bank

Panel B: Summary Statistics Variable Obs. Mean Std. Dev. Minimum Maximum Polity IV democracy 1067 4.57 4.16 0 10 Freedom 1204 −3.80 2.24 −7 −1 Vanhanen 821 11.71 13.14 0 44.25 GDP 1250 23.21 2.36 16.63 30.34 Endowment ratio 1250 0.23 0.14 0.005 1.17 Oil 287 1129.77 2764.40 0 23 550.25 Human Capital 986 1.90 2.58 0 11.59 Worldpatents 1263 11.54 0.44 10.93 12.24 Education 387 5.32 2.83 0.23 12.18 Military 649 0.03 0.03 0.0004 0.49 924 0.03 0.02 0.01 0.14 Mili surrounding

34

Table 2: Cross-Correlation Table Variables Polity IV Vanhanen Polity IV 1.00 Vanhanen 0.80 1.00 Freedom 0.84 0.81 GDP 0.45 0.53 Endow. Ratio (ER) −0.15 −0.09 Oil −0.11 −0.21 Human Capital (HC) 0.52 0.62 World Patents (WP) 0.27 0.26 Education (Educ) 0.60 0.70 Military −0.12 −0.11 −0.19 −0.23 Mili surrounding (MS)

Freedom

GDP

ER

Oil

HC

1.00 0.44 −0.18 −0.16 0.56 0.20 0.65 −0.15 −0.24

1.00 0.23 0.22 0.77 0.28 0.65 −0.07 −0.04

1.00 0.25 0.05 −0.04 0.07 0.10 0.25

1.00 0.06 0.22 −0.01 0.13 0.26

1.00 0.10 0.78 −0.06 −0.18

WP

Educ

Military

1.00 0.20 1.00 −0.15 −0.01 0.005 −0.12

1.00 0.006

MS

1.00

35

Table 3: Linear Model for Democracy The initial sample contains country-years for which both the Polity IV data and GDP figures were available.The dependent variable is the democracy score. All independent variables are lagged five years. Endowment Ratio is defined as the aggregation of economic activities under Section II (mining and quarrying) divided by total aggregate GDP. We also use value of oil reserves for robustness checks (the coefficients of oil value are 1000 times larger than the regressed coefficients). Human Capital is log of (1+number of patents). The legal origin dummies are taken from LLSV (1999), and the benchmark group are the countries with socialist origin. All the standard errors reported are clustered at country level. Notice that the coefficient of German Legal Origin dummy variable is omitted due to multicollinearity. The symbols ∗ , ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Variables Endowment Ratio

(1) −9.82 (1.83)∗∗∗

(2)

−0.68 (0.18)∗∗∗

Oil Human Capital GDP

(3) −7.62 (1.61)∗∗∗

0.81 (0.12)∗∗∗

1.59 (0.22)∗∗∗

0.78 (0.14)∗∗∗ −0.01 (0.19)

(4)

−0.54 (0.13)∗∗∗ 0.97 (0.23)∗∗∗ 0.26 (0.38)

UK Legal Origin French Legal Origin German Legal Origin Scandinavian Legal Origin Constant Observations R2

−11.64 −35.01 (2.96)∗∗∗ (5.45)∗∗∗ 737 216 0.22 0.45

36

5.87 (4.11) 737 0.30

−4.07 (8.97) 216 0.58

(5) −8.13 (1.50)∗∗∗

(6)

−0.56 (0.13)∗∗∗ 0.74 (0.20)∗∗∗ 0.47 (0.31) 2.12 (1.54) 1.74 (1.39) omitted

0.81 (0.14)∗∗∗ 0.02 (0.18) −0.50 (0.78) 0.32 (0.72) −1.59 (1.99) 1.69 4.80 (0.77)∗∗ (1.42)∗∗ 5.12 −10.45 (3.98) (6.75) 720 210 0.33 0.61

Table 4: Instrumental Variable Approach The definitions of the variables are the same as that in Table 3. All the standard errors reported are clustered at country level. We run a two-stage least squares regression in this section using aggregate number of world patents as the instrument variable. In the first specification, we use endowment ratio as the proxy for endowed wealth while in the second specification we use value of oil to proxy for the endowed wealth. We also report the adjusted R2 in the first stage. The symbols ∗ , ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Variables

(1) Human Capital Democracy Endowment Ratio 0.41 −7.12 (0.55) (1.55)∗∗∗ Human Capital 0.95 (0.16)∗∗∗ Oil World Patents GDP Constant Observations R2 Adjusted R2

0.23 (0.11)∗∗ 0.32 (0.08)∗∗∗ −8.52 (1.83)∗∗∗ 737 0.96 0.95

−0.20 (0.20) 9.83 (4.36)∗∗ 737 0.29

37

(2) Human Capital Democracy

−0.02 (0.05) 0.48 (0.22)∗∗ 0.43 (0.16)∗∗∗ −13.81 (4.27)∗∗∗ 216 0.96 0.95

1.14 (0.28)∗∗∗ −0.51 (0.12)∗∗∗ −0.03 (0.45) 1.28 (10.42) 216 0.57

Table 5: Alternative Specifications The definitions of the variables are the same as that in Table 3. Different from Table 3, we add in the interaction term between endowment ratio and human capital as an explanatory variable. All the standard errors reported are clustered at country level. Notice that fixed effects are inappropriate here because the variables (Endowment Ratio, Human Capital, and GDP) are slow-moving. The symbols ∗ , ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Variables Lagged Democracy

Panel A: Regression Coefficients (1) (2)

Endowment Ratio Human Capital

−7.62 (1.61)∗∗∗ 0.78 (0.14)∗∗∗

Endow. Ratio×Human Capital GDP Constant Observations R2

−0.01 (0.18) 5.87 (4.11) 737 0.30

−5.90 (1.67)∗∗∗ 1.34 (0.30)∗∗∗ −2.35 (1.01)∗∗ −0.02 (0.18) 5.53 (3.94) 737 0.31

(3) 0.81 (0.03)∗∗∗ −1.53 (0.54)∗∗∗ 0.13 (0.05)∗∗∗ −0.02 (0.06) 1.83 (1.25) 737 0.76

(4) 0.80 (0.03)∗∗∗ −1.37 (0.54)∗∗∗ 0.19 (0.09)∗∗ −0.24 (0.29) −0.02 (0.06) 1.81 (1.25) 737 0.76

Panel B: Marginal Effects on Democracy Variables (1) (2) (3) (4) Endowment Ratio −7.62 −10.73 −1.53 −1.86 ∗∗∗ ∗∗∗ ∗∗∗ (1.61) (2.18) (0.54) (0.74)∗∗∗ Human Capital 0.78 0.73 0.13 0.12 ∗∗∗ ∗∗∗ ∗∗∗ (0.14) (0.14) (0.05) (0.05)∗∗∗

38

Table 6: Alternative Proxies The definitions of the variables are the same as that in Table 3. Different from Table 3, we add in the interaction term between endowment ratio and human capital as an explanatory variable. All the standard errors reported are clustered at country level. The symbols ∗ , ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Variables Endowment Ratio

Panel A: Regression Coefficients Polity IV Vanhanen Freedom −5.90 −7.31 −2.83 ∗∗∗ (1.67) (4.63) (0.86)∗∗∗

Polity IV 1.39 (2.18)

−0.70 (0.19)∗∗∗ 0.91 (0.25)∗∗∗

Oil Human Capital

1.34 (0.30)∗∗∗

7.69 (1.24)∗∗∗

0.87 (0.16)∗∗∗

Education

1.25 (0.20)∗∗∗

Endow. Ratio×Human Capital

−2.35 −17.97 (1.01)∗∗∗ (4.03)∗∗∗

−1.66 (0.57)∗∗∗ −0.05 (0.05)

Oil×Human Capital Endow. Ratio×Education GDP Constant Observations R2

−0.02 (0.18) 5.53 (3.94) 737 0.31

0.11 (0.59) 7.79 (13.05) 496 0.45

−0.05 (0.10) −2.55 (2.15) 737 0.37

−2.40 (0.70)∗∗∗ 0.43 (0.20)∗∗ −8.66 (4.57)∗∗ 373 0.39

Panel B: Marginal Effects Variables Polity IV Vanhanen Freedom Polity IV Endowment Ratio −10.73 −43.11 −6.24 −11.59 ∗∗∗ ∗∗∗ ∗∗∗ (2.18) (7.97) (1.22) (2.68)∗∗∗ Oil Human Capital

Polity IV

0.73 (0.14)∗∗∗

2.92 (0.56)∗∗∗

Education

0.44 (0.07)∗∗∗ 0.65 (0.17)∗∗∗

39

0.30 (0.40) −5.04 (9.27) 216 0.58

Polity IV

−0.58 (0.16)∗∗∗ 0.95 (0.24)∗∗∗

Table 7: Subsample Tests The definitions of the variables are the same as that in Table 3. The stable democratic countries in the last column are the ones whose average democracy score in 1975-2010 is 10. The symbols ∗ , ∗∗ , and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Variables Endowment Ratio Human Capital Endowment Ratio×Human Capital GDP Constant 40

Observations R2

Excluding Africa −8.73 (2.28)∗∗∗ 1.16 (0.30)∗∗∗ −2.22 (1.01)∗∗ −0.11 (0.20) 9.48 (4.35)∗∗ 552 0.30

Excluding current and former socialist −6.96 (1.71)∗∗∗ 0.93 (0.27)∗∗∗ −0.72 (0.92) 0.04 (0.19) 4.60 (4.13) 645 0.33

Excluding stable democracies −5.39 (1.67)∗∗∗ 1.20 (0.41)∗∗∗ −2.87 (1.23)∗∗ 0.21 (0.19) 0.05 (4.15) 615 0.16

Table 8: Ratio Specification The definitions of the variables are the same as that in Table 3. log(mining) is calculated by taking the log of the economic activity under Section II (Mining and quarrying). β is constructed as a ratio of human capital wealth to endowment wealth (in billion dollars). The symbols ∗ , ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Variables log(mining) Human Capital

(1) −1.76 (0.57)∗∗∗ 0.81 (0.15)∗∗∗

β GDP Constant Observations R2

1.77 (0.65)∗∗∗ −34.37 (14.36)∗∗∗ 737 0.28

41

(2)

0.11 (0.03)∗∗∗ 0.54 (0.14)∗∗∗ −8.13 (3.30)∗∗ 737 0.19

Table 9: Disequilibrium Tests In this table we report the results of disequilibrium test. Countries are initially sorted by the model prediction and then sorted by the measured level of democracy in five years using the Polity IV data. First, we label a country to be in “underdemocratic” (“overdemocratic”) level if its Polity IV democracy score in 2010 is lower (higher) than the predicted level by at least 3 points out of the zero to ten scale. And we label a country to be “initially correctly predicted” if the difference is less than 3 points. Second, we define a country to be “autocracy” if its Polity IV score is less than or equal to 3, and we define it to be “democracy” if its Polity IV score is greater than or equal to 7. For the countries with a Polity IV score greater than 3 and less than 7 we define them to be “neutral”. Panel A: Initially ”Underdemocratic” In five years Autocracy Neutral Democracy Start as Autocracy 86.96% 4.35% 4.35% Start as Neutral 0% 0% 0% Start as Democracy 4.35% 0% 0% Panel B: Initially ”Overdemocratic” In five years Autocracy Neutral Democracy Start as Autocracy 0% 0% 0% Start as Neutral 0% 0% 4.65% Start as Democracy 6.98% 4.65% 83.72% Panel C: Initially Correctly Predicted In five years Autocracy Neutral Democracy Start as Autocracy 26.92% 1.92% 0% Start as Neutral 3.85% 19.23% 1.92% Start as Democracy 0% 0% 46.15%

42

Table 10: Predicted Versus Actual Values in the Year 2010 In this table we list the predicted democracy level based on Table (3) column (3) and the actual Polity IV democracy score as of the year 2010 for some countries of interest. In Panel A (B) we report the first ten lowest (highest) ranked countries based on the predicted democracy level. And in Panel C we report eight other countries of interest. Panel A: Lowest ranked countries Country Predicted Actual Error Libya -0.61 0 0.61 Qatar 0.26 0 -0.26 Angola 0.51 2 1.49 Gabon 1.21 4 2.79 Oman 1.51 0 -1.51 Chad 1.62 1 -0.62 Algeria 1.66 3 1.34 Azerbaijan 1.78 0 -1.78 Trinidad and Tobago 1.79 10 8.21 Kuwait 1.98 0 -1.98 Panel B: Highest ranked countries Country Predicted Actual Error United States 12.89 10 -2.89 Japan 11.80 10 -1.80 Germany 10.82 10 -0.82 France 10.61 9 -1.61 United Kingdom 10.60 10 -0.60 Canada 9.80 10 0.20 Italy 9.66 10 0.34 Israel 9.65 10 0.34 Netherlands 9.60 10 0.40 Sweden 9.42 10 -0.58 Panel C: Other countries of interest Country Predicted Actual Error China 3.82 0 -3.82 Cuba 5.82 0 -5.82 Greece 6.79 10 3.21 Iran 2.87 0 -2.87 Singapore 7.99 2 -5.99 Turkey 5.57 8 2.43 Venezuela 2.84 1 -1.84 Zimbabwe 3.26 3 -0.26

43

Table 11: Military Expenses Effect on Democracy The definitions of the variables are the same as that in Table 1. The symbols ∗ , ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Panel A: Regression Coefficients Variables (1) (2) (3) Endowment Ratio −9.31 −5.94 −7.46 (2.31)∗∗∗ (2.94)∗∗ (2.20)∗∗∗ Military −1.79 11.41 1.72 ∗∗ (3.70) (4.91) (1.95) Endowment Ratio×Military −72.35 (27.76)∗∗∗ Mili surrounding −47.45 (20.19)∗∗ −62.39 Military×Mili surrounding (176.62) Human Capital 0.81 0.81 0.71 (0.23)∗∗∗ (0.23)∗∗∗ (0.24)∗∗∗ GDP −0.32 −0.33 −0.26 (0.26) (0.27) (0.27) Constant 14.10 13.75 13.62 (5.92)∗∗ (5.97)∗∗ (6.05)∗∗ Observations 331 331 331 2 R 0.28 0.29 0.33 Panel B: Marginal Effects Variables (1) (2) (3) Endowment Ratio −9.31 −8.09 7.46 (2.31)∗∗∗ (2.42)∗∗∗ (2.20)∗∗∗ Human Capital 0.81 0.81 0.71 (0.23)∗∗∗ (0.23)∗∗∗ (0.24)∗∗∗ Military −1.79 −6.70 0.02 ∗∗ (3.70) (3.13) (3.17) Mili surrounding −49.30 (15.85)∗∗∗

44

Table 12: Military Expenses Effect on Democracy The definitions of the variables are the same as that in Table 1. The variable Difference is defined as Military-Mili surrounding. In Panel A we report regression results for countries whose military expenditure is larger than or equal to the average military expenditure of its surrounding countries while in Panel B we report the results for those whose military expenditure is less than or equal to the average military expenditure of surrounding countries. The symbols ∗ , ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Panel A: Own-Expenditure ≥ Surrounding-Expenditure Variables (1) (2) Endowment Ratio −9.41 (4.40)∗∗ Human Capital 0.57 (0.28)∗∗ β 0.11 (0.04)∗∗ Difference −8.91 −16.80 ∗∗ (3.66) (5.34)∗∗∗ GDP 0.07 0.53 (0.37) (0.27)∗ Constant 5.47 −7.38 (8.56) (6.56) Observations 108 108 2 R 0.38 0.24 Panel B: Own-Expenditure ≤ Surrounding-Expenditure Variables (1) (2) Endowment Ratio −7.68 (2.86)∗∗∗ Human Capital 0.95 (0.34)∗∗∗ β 0.19 (0.10)∗∗ Difference 3.31 4.62 (1.59)∗∗ (2.19)∗∗ GDP −0.55 0.11 (0.33) (0.26) Constant 18.96 2.62 ∗∗∗ (7.54) (6.13) Observations 220 220 2 R 0.23 0.09

45

Table 13: Effect of Endowment on Military Expenditure The definitions of the variables are the same as that in Table 1. Based on the relation between the own-military expenditure and the surrounding countries’ military expenditure, we separate the countries into 4 groups: (high,high), (high, low), (low, high), and (low, low). The first letter indicates whether the country’s own expenditure is high or low, while the second letter indicates whether the average military expenditure of the surrounding countries is high or low. We are interested in the probability of a country in (high, low) state. We use Probit model to analyze the effects of endowment and β on the probability that a country falls into the (high, low) state. The symbols ∗ , ∗∗ and ∗∗∗ denote significance at the 10%, 5% and 1% levels, respectively. The numbers in parentheses are standard errors. Variables Endowment Ratio Human Capital

(1) 2.20 (0.75)∗∗∗ 0.06 (0.06)

β GDP Constant

−0.004 (0.37) −1.72 (1.87)

46

(2)

−0.04 (0.01)∗∗∗ −0.02 (0.04)∗ −0.86 (1.07)