Literature on Inequality and Democratization

Nineteenth-century theorists saw changes in economic inequality as a determinant of changes in political regimes, but the direction of this relationship differed by theorists. De Tocqueville's core thesis was that rising economic equality—especially the equality of land ownership—was a necessary condition for France's first experiment with democracy and for America's democratic experience. Because “the gradual development of the principle of equality” was the “providential fact” of his time, de Tocqueville claimed that democratization was inevitable (de Tocqueville Reference de Tocqueville1986; Reference de Tocqueville1998).Footnote ² In contrast, Marx and his followers argued that rising economic inequality spawned revolution, which eventually would produce democratic socialism.

More recent scholarship emphasizes bargaining over tax rates on assets as a vehicle for democratization. Bates and Lien (1985) argue that democracy is the result of a bargain between monarchs and holders of tax-elastic physical assets. Because monarchs need revenue to fight wars and physical assets are mobile, they are forced to bargain with asset holders: “Elasticity of the tax yield made it necessary for [monarchs] to bargain with those who possessed property rights over the moveable tax base and to share with them formal control over the conduct of public affairs” (57). Bates and Lien rely on historical case studies to demonstrate the truth of their argument.Footnote ³

In Acemoglu and Robinson's (AR Reference Acemoglu and Robinson2000; Reference Acemoglu and Robinson2001) account of democratization, the threat of revolution looms large in the bargaining between elites and masses. AR (Reference Acemoglu and Robinson2000; Reference Acemoglu and Robinson2002) show how democratization can be part of the process that gives rise to the Kuznets (Reference Kuznets1955) curve—the idea that inequality rises in early phases of industrialization when national income is relative low but diminishes at later phases of industrialization when national income is relatively high. A rise in inequality under autocracy increases the threat of revolution by the poor. Elites weigh their gains from continuing to set tax rates in an autocracy while continuing to pay the costs of repression relative to their gains from granting democracy and allowing the poor (median) voters to set tax rates. For the poor, democracy represents a credible commitment by the elite to accept limited redistribution of income.

In the first half of their book, Economic Origins of Dictatorship and Democracy, AR (2006) develop a “workhorse,” two-group model of distributive politics. This model stresses the relative gains to the poor and elite from income redistribution under autocracy and democracy conditional on the poor's relative income after a revolution, the cost of repression to elites, and other parameters. It interprets democratization as a credible commitment to redistribution, but it is only one of several possibilities (Proposition 6.2). Empirically, the implication is that the relationship between equality and democracy is an inverted “U” or hump.

The second half of AR's book (2006, chap. 10) contains an open economy extension of their analysis. In it, they assume that, in most countries, labor is abundant and capital is scarce. They also assume that both trade and capital mobility produce global factor price equalization. The result is an increase in the income of the poor and, in turn, a reduction in the poor's (median voters’) preferred tax rate. In this way, trade and financial liberalization produce a lower income loss to the rich relative to what they experience in a closed economy. This makes it more likely that the rich prefer democracy instead of autocracy and repression. However, AR are quick to add that the assumptions underlying their open economy models are controversial, especially the assumption that globalization reduces inequality (346).

Boix (Reference Boix2003) argues that the origins of democracy depend on the interaction of inequality and asset specificity. He defines asset specificity as the “cost of moving capital away from its country of origin” (3).Footnote ⁴ An innovation in Boix's analysis is that he considers the possibility that the wealthy can earn income abroad. As in the Bates and Lien investigation, native elites have the ability to move some types of assets out of the political jurisdiction, depending on their “specificity.” This possibility of asset exit constrains the ability of the poor to tax elites.

Inequality's effects on democratization are then contingent on the degree of asset specificity in an economy. By definition, in closed economies, the degree of specificity is high, and the threat of exit is therefore not credible. Boix (Reference Boix2003) gives land assets as an example. If, in this case, income is relatively equally distributed, Boix contends that the wealthy will agree to democratize, which is in contrast to the argument of AR but similar to that of de Tocqueville. If, in contrast, in this closed economy, income is unequally distributed, Boix's prediction is for autocracy. If the economy is open—asset specificity is low—the wealthy have an “exit option,” which is what constrains the ability of the poor (median) voter to tax them. In this case, as in the Bates and Lien argument, democratization is the likely outcome: “The decline in the extent to which capital can be either taxed or expropriated as a result of its [specificity] fosters the emergence of a democratic regime” (12). Boix uses a combination of case studies and data analysis to support his arguments.

Recently, another work in this genre has appeared: Ansell and Samuels’ (2010) “contractarian explanation” for democratization. Their argument is similar to that of Bates and Lien, but Ansell and Samuels distinguish land from industrial assets. They derive different expectations for how land (positive) and income (negative) equality affects democratization and use both case studies and data analysis to defend their thesis.

Critique

Most existing analyses assume either a financially closed economy or an economy in which only some types of assets can exit. In addition, most do not include modern financial integration in the opportunities and constraints facing actors. For instance, it is only at the end of their article that Bates and Lien call for inclusion of capital inflows and outflows as an extension to what is otherwise a closed economy formal model.Footnote ⁵ Ansell and Samuels also assume a closed economy: They make no provision in their contractarian approach for the increased sales of land to foreigners and for the new forms of inequality that financial integration produces. Finally, the models in all the AR articles cited here as well as in the first half of their book assume the economy is closed. In these models and in their workhorse model, they assume that assets are owned only by native poor and native elites; there is no capital inflow and exit occurs only into “informal markets.”

Boix's model and the model extension in the second part of AR's book do usefully recognize the possibility of capital or labor outflows. However, these models do not explain how the rise of international financial integration facilitates these flows or what are these flows’ political implications.Footnote ⁶

In view of the omission of global financial integration as an analytic consideration, it is not surprising that recent empirical investigations have produced mixed support for leading theories on the inequality–democracy relationship.Footnote ⁷ Boix's (2003) study covered regime transitions from 1950 to 1990; he found support for his arguments. Using data up through 2002, Houle (Reference Houle2009) recently tested the claims of Boix and AR. He used a dynamic probit model to predict transitions in Przeworski et al.'s (2000) measure of regime and used Rodriquez and Ortega's (2006) data to measure inequality. Houle's main finding is that inequality only affects democratic consolidation, not democratization: If there is any relationship between inequality and democracy, it is U-shaped, not hump-shaped (Houle Reference Houle2009, 610, 615). Neither empirical investigation found support for the AR argument.

Houle (Reference Houle2009), however, makes no provision for financial openness in his empirical analysis, even though a test of Boix's theory would require some controls for capital mobility and the interaction between capital mobility and inequality. Several of the same design problems plague other recent tests such as those reported by Ansell and Samuels (Reference Ansell and Samuels2010).Footnote ⁸

For these reasons, we need a new account of the economic origins of democracy, an account that incorporates the effects of modern financial integration. We offer one in the next section.

Overview

A key insight in the Bates and Lien (1985) and Boix (Reference Boix2003) studies is that the ability of elites to move assets within and out of political jurisdictions strengthens their bargaining position over tax rates. Our argument builds on this insight, but applies it to financially integrated autocracies.

We contend that financial integration alters the calculus of autocratic elites in a way that makes democratization likely. It is through three mechanisms that financial integration ameliorates the adverse effect of inequality on the willingness of autocrats to democratize. First, financial integration gives domestic elites an incentive not only to exit the political jurisdiction but also to construct an internationally diversified portfolio. The diversified portfolio has higher returns than a domestic-only portfolio, which increases economic inequality. Second, the makeup of a country's capital stock—the identities of its owners and its value to those owners in an internationally diversified portfolio—changes with financial integration. Third, financial integration constrains, but does not eliminate, the ability of the median voter to tax capital income. We describe in greater detail later the workings of these mechanisms. Given these considerations, autocratic elites have less at stake regarding median voters’ tax policy preferences under financial integration and less capacity to resist those policy preferences. In turn, median voters are able to capture some of the revenue they need to finance the production of public goods, but redistribution is constrained by financial integration. Hence the poor and the elite opt for democratization rather than for revolution or repression. This, we argue, is the new bargain that underlies democratization in the modern, financially integrated world economy. Empirically, the new bargain means that, as unequal autocracies move from financially closed to financially integrated economies, these autocracies will tend to democratize. Extant democracies will be unaffected by the new phase of world financial integration because they are based on the older, consolidated bargains that are already described in the literature.

International Portfolio Diversification and Democratization

Modern portfolio theory sheds light on how international financial integration helps investors create more diversified portfolios and therefore higher levels of wealth with lower risk. Native investors lower their risks relative to returns by diversifying into foreign equities through two mechanisms. First, international equity market price correlations are lower than intra-country, inter-industry equity price correlations. Second, international markets, by definition, have a bigger “investment opportunity set” (see Bekaert and Harvey Reference Bekaert and Harvey2000, Grubel Reference Grubel1968, and Quinn and Voth Reference Quinn and Joachim Voth2008, for discussions). The risk reduction achievable by a representative investor in the United States through investing in broadly based international portfolios compared to investing in, for example, the “core markets” of the United States, United Kingdom, Germany, and France was estimated to be 65% in 2000 (Goetzmann, Li, and Rouwenhorst Reference Goetzmann, Li and Rouwenhorst2005, 31). For an average U.S. investor between 1970 and 1994, the risk-adjusted return of an internationally diversified portfolio was 28% higher than that of a U.S.-only equity portfolio (DeSantis and Gerard Reference DeSantis and Gerard1997, 1907).

For native investors in a closed emerging economy, the potential returns from international portfolio diversification are much higher. As explained earlier, the asset price correlations internationally will be lower than the within-country asset correlations, and the investment opportunity set will be higher, offering domestic investors the opportunity to decrease risk and increase return. With liberalization of capital account outflows and the international diversification it allows, the domestic investor is able to ensure that his or her assets are not too “specific” (or, more correctly, are not too idiosyncratic in risk; see Bechtel Reference Bechtel2009, for a discussion of several types of investment risks).

Earlier studies emphasized the threat of asset exit or “asset elasticity” if rulers either adopt unfavorable policies or fail to commit credibly to favorable policies through democratic reforms. What is new in our argument is that modern portfolio theory recommends international diversification, even in a context in which governments have adopted policies favorable to holders of mobile assets. Exit is the rational investment strategy, not only a response to confiscatory taxation or deleterious policies.

The incentives for international diversification of elite assets have two political implications. First, international diversification lessens the threat to the autocratic elite from democratization and, more specifically, from the redistribution of capital income. Second, by design, portfolio diversification decreases the concentration of assets held by the native elite in their home economy. Moreover, with decreased elite asset ownership concentration comes a likely decrease in the elite's ability to solve “collective action” problems. Among these collective action problems are repression of democracy and control of tax rates. That is, with greater dispersion of asset ownership, free riding in bearing the costs of maintaining autocratic rule is likely to increase.Footnote ⁹

These two political implications will influence only autocratic democratizations and will not be associated with democratic reversals. The costs to an elite of organizing to reverse democratization are higher than the costs of maintaining an existing autocratic regime. Given that the forces we identify lessen elite stakes in, and organizing capabilities around, political action, the more costly task of reversing democratization will occur infrequently.

Financial Integration and the Changed Nature of Capital Assets in Autocratic Countries

Paradoxically, when a country with immobile or illiquid assets liberalizes inward capital account transactions, specific assets (or those that have idiosyncratic risks that are uncorrelated with returns in global capital markets) become valuable to foreign investors as components of their global, diversified portfolio. This is because, as explained earlier, the aims of international investors are to diversify risk both by investing in assets whose prices do not co-move with international prices and by expanding their investment opportunity set; (see Dellas and Hess Reference Dellas and Hess2005, for a discussion of stock market synchronization increasing with increasing liquidity and depth equity markets). As Goetzmann, Li, and Rouwenhorst (2005, 1) note, “the benefits from diversification rely increasingly on investment in emerging markets,” which contain assets with more idiosyncratic risk. They show that “the risk reduction from diversifying across all markets is more than double the risk reduction that can be achieved by diversifying across the core [United States, United Kingdom, Germany, and France] markets only” (31).

Capital account openness changes the meaning and economic value of asset specificity. Assets that previously were nonredeployable or immobile are now globally traded in world financial markets. In this way, capital assets—including land—are no longer “specific” in the same way as in the past. Owners of land are able to sell property rights to foreigners seeking diversified portfolios. In turn, these foreigners can trade property rights in secondary markets; with the proceeds from these sales, former (native) landowners are able to purchase new, often highly liquid assets in foreign markets. For these reasons, contrary to the arguments in Ansell and Samuels, the distinction between land and other assets dissolves in a context of international financial integration.Footnote ¹⁰

For example, through American Depositary Receipts, Global Depositary Receipts, and other instruments, Argentine landowners now can sell their assets to overseas investors in international equity markets, retain the proceeds from those sales, and buy international assets. Of the public offerings in the American Depositary Receipt (ADR) markets by industry through 2008, nearly 35% of the $175 billion in offerings sold outside home countries were in so-called fixed or immobile industries such as mining and agriculture.Footnote ¹¹ Of the $6.5 trillion in market capitalization value for the top 15 emerging markets, nearly 25% of the value of those markets traded in New York and not in the home market.Footnote ¹² In addition to these markets, international investors are buying and leasing immobile assets such as large tracts of land in Africa, Central Europe, and other parts of the world.Footnote ¹³ Prior to these developments in international financial markets (roughly before 1980), elites in underdeveloped countries had limited ability to convert domestic assets into fungible overseas assets. Financial openness—the ability to exit a country's economy, without the corresponding capability to diversify portfolios and engage in asset swaps—was of limited benefit to autocratic elites.

Through domestic asset sales to foreigners, native capital owners accrue large earnings, which increase inequality between native poor and native elites. Empirical studies show this positive correlation between inequality and financial openness. Financial globalization is a robust correlate of rising income inequality in a cross-section of countries examined in Quinn (Reference Quinn1997). Claessens and Perotti (Reference Claessens and Perotti2007) show that financial liberalization's benefits are highly skewed in favor of small groups of elites, especially in developing countries. Jaumotte, Lall, and Papgeorgiou (Reference Jaumotte, Lall and Papgeorgiou2008) find that, although trade has the effect of reducing income inequality, inward foreign direct investment (FDI) flows increase income inequality (cf. AR 2006, chap. 10, sect. 5.1). A study in 2008 by the International Labor Organization (ILO) also documents the correlation between rising income inequality and stock of FDI (ILO 2008). With an increase in wealth from capital asset sales comes an increased incentive for further international diversification by the native elite.Footnote ¹⁴

Exchange of Assets and Exchange of Political Risks

In a sense, native elites in a financially closed economy are holders of a highly undiversified investment portfolio with undiversified political risk. After financial integration, native elites are able to form internationally diversified portfolios, which diversify their political risk. Apart from a few core countries (e.g., the United States), the economic risks inherent in any one investment market are therefore small, and the stakes of the foreign elite in the politics of any given country are correspondingly diminished.

In these ways, modern financial integration potentially has much more complex effects on democratization than the exit option described in the literature on the economic origins of democracy. For one, modern portfolio theory recommends international diversification (or exit) even in cases where governments adopt ‘“virtuous” domestic policies and institutions. For another, the identity of holders of domestic assets changes: Both foreign and native elites hold these assets. And foreign investors holding diversified portfolios are less likely to respond in politically repressive ways to unfavorable domestic tax policies than the undiversified native holders of immobile assets in closed economies. International investors simply will not invest, which constrains the tax rate that can be imposed.

Financial Globalization, Capital Taxation, and Democracy

Many political scientists and economists argue that capital taxation in smaller economies with open capital accounts is difficult to sustain; such taxation is prone to a “race to the bottom” (for models see Devereux, Lockwood, and Redoano Reference Devereux, Lockwood and Redoano2007, and Tanzi Reference Tanzi1995; see Haufler Reference Haufler2001, for a review). For example, the predicted effect of open capital accounts models is that a government's revenue from capital taxation disappears, even if it persists in maintaining tax rates. Note also that AR's open economy extension (2006, 339) assumes that capital inflows and outflows either are not taxed or are taxed at a relatively low rate. Paradoxically, in this model, the unsustainability of high levels of taxation on mobile capital with open capital accounts is good for democratization. In the AR extension, capital inflows increase wages, making the income of the median voter higher, and therefore, ceteris paribus, reducing the redistributive pressures on elites: Democracy becomes less costly.

It is difficult to think of an argument in international political economy research that is more at odds with the observed behavior of governments. Consider Figure 1.Footnote ¹⁵ It reports OECD corporate tax collections and rates for 1970 and 2005, both years of world business cycle expansion.Footnote ¹⁶ For the average OECD country, corporate tax revenues as a percentage of GDP rose in those 35 years from 2.5% to 3.6% of GDP.Footnote ¹⁷ The 35 years between 1970 and 2005 were a period of financial globalization among OECD countries, with no significant capital controls remaining by 2005. Top corporate tax rates fell on average during the same period. However, the tax base was broadened through reductions in investment incentives and other deductions, which contributed to the steep rise in corporate tax collections.Footnote ¹⁸ Emerging market corporate tax collections circa 2005 have grown modestly, in contrast to tax collections for OECD member countries, yet the respective governments’ capital tax collections have remained relatively stable.

FIGURE 1.

Corporate Tax Revenue Collections as % of GDP 1970 vs. 2005 (OECD); 1990 vs. 2004 (Emerging Only)

Addressing the discrepancy between theory and evidence, Plümper, Troeger, and Winner (Reference Plümper, Troeger and Winner2009) argue that fiscal rules and equity norms (measured by Gini coefficients) put upward pressure on capital tax rates and revenue. Their results confirm that countries with open capital accounts do not converge on capital tax policies. These findings are consistent with the “system of constraints” results in Swank and Steinmo (Reference Swank and Steinmo2002) and the “tournament” model in Basinger and Hallerberg (Reference Basinger and Hallerberg2004). (See also Hays Reference Hays2003.) Governments, although not free in these analyses to tax capital at confiscatory rates, are able to capture income from capital taxation under conditions of capital account openness. Financial integration thus does not eliminate the tax burden on capital. The foreign capital that flows into and out of countries is taxable, although usually not at rates higher than those in the United States and major OECD countries.Footnote ¹⁹ The poor (median) voter therefore has some possibility to choose a tax rate that allows for some redistribution and production of some public goods.

Implications

As AR in their extension, Boix, and others argue, the open economy linkage between inequality and autocracy is likely to differ from the relationship in a closed economy. Where our argument differs is that we contend that modern financial integration allows for portfolio diversification by holders of domestic assets, which includes exchanging of assets with foreigners, who also hold diversified international portfolios. This swapping of assets mitigates the risks of adverse political events, such as confiscatory tax policies, for both native and foreign elites. In this way, financial integration limits the risk of democratization to autocratic elites in highly unequal societies.

We predict the greatest increase in democratization for financially open, unequal autocracies. In this context, the native rich have little to gain from resisting democratization because they can diversify their assets. Incoming international (diversified) investors are unlikely and unable to resist democracy to the same degree as their counterparts in financially closed economies (native elites with internationally undiversified portfolios). The dispersion of asset ownership implies a diminished interest by domestic elites in domestic politics and a lower capacity for collective action regarding its policies and institutions. Because capital asset taxation of foreign and native asset holders is still partly feasible, the poor (median voters) will opt for democratization rather than revolution. Therefore, overall, the prospects for democratization are brighter in financially open economies with high inequality than in financially closed economies with high inequality. Hence, the new phase of financial integration links high levels of inequality in open autocracies with democratization.

In sum, our causal chain is as follows (see Implications 1, 2, and 3). Domestic financial openness with high levels of international financial integration produces portfolio diversification by native elites with resulting increased domestic inequality. This diversification reduces the autocratic elites’ stake in domestic tax policies and their capacity for collective action. Financial openness with financial integration, in turn, reduces the median voter's preferred tax rate to somewhere between “safe haven” and the global average tax rate but still sufficient to make democratization more attractive than revolution. Therefore democracy results in less net redistribution than in the financially closed economy, and elite repression is less attractive in the financially open than in the closed case. Hence there is a greater probability of transition to democracy in financially integrated autocratic economies with high degrees of income inequality in comparison to unequal financially closed autocracies. (See Implications 1, 2, and 3 on the next page).

Data and Measures

Democracy. Our core dependent variable in this investigation is change in democracy, which we measure by using both Polity IV and Regime from 1955–2004. (The latter measure now is also known as DD.Footnote ²⁰) We estimate models using both measures to demonstrate robustness of our results. In using the 21-point Polity measure, we allow for both minor and major changes in democratic institutions. In using the dichotomous Regime variable, we focus on large changes in political institutions. For reasons explained later in this section and in the Appendix, we use five-year panels. For these panels, change in Regime is transformed into an interval variable with values ranging between -1 and 1; change in Regime represents the difference in five-year average values of DD for each country. A positive value of change in Regime thus indicates greater levels of democratization. We show later that the choice of the democracy indicator does not change our results.

An important question is whether the effects we propose are found for autocracies that are democratizing (or retreating deeper into autocracy) or for democracies that are consolidating (or reversing into autocracies). Boix (Reference Boix2003) also distinguishes between these two types of cases in his empirical estimations. We follow the Polity coders and treat countries with combined average Polity scores of 6 and higher in the five-year period prior to the period studied as democracies; we treat countries with Polity scores lower than 6 as autocracies.

Inequality. Cross-national inequality indicators are plagued by measurement difficulties. We use a single indicator of inequality for each country, a Gini coefficient,Footnote ²¹ based on three standard data sets: Deininger and Squires (D&S; Reference Deininger and Squire1996), Milanovic (Reference Milanovic2005), and United Nations University-WIDER's World Income Inequality Database (WIID; 2008). Because the D&S and WIID data contain information from various sources using diverse methods on diverse populations, they must be adjusted before using them in cross-national, time-series analyses.Footnote ²² The Milanovic data are comparable across time and space, but are limited in time to at most three observations per country.

Dollar and Kraay (DK; Reference Dollar and Kraay2002) develop a method for turning these different Gini measures into a single indicator that can be used in comparative research. We use their transformation algorithm for this purpose. In the appendix we explain in more detail how this is done and why, in the end, our measure of income inequality is sounder than that used by Houle (Reference Houle2009) and others.

Financial Integration. Our measures of financial integration are based on work by Quinn (Reference Quinn1997) and Quinn and Toyoda (Reference Quinn and Toyoda2007). CAPITAL is the main element of capital account openness created from the text published in the International Monetary Fund's Annual Report on Exchange Arrangements and Exchange Restrictions. The indicator is an index for a government's policy stance toward capital account liberalization; it is scaled 0–100. Global and regional averages of capital account openness are also calculated based on CAPITAL (see the Appendix for further details).

In our critique of the literature we emphasized the difference between the effects of a threat of exit of the native elites’ mobile assets and the effects of native and foreign elites’ portfolio management. We argue that the threat of exit is credible now because (a) global financial markets are much more fully developed than in the past and (b) foreign elites seek to exchange assets (risks) with native elites. To capture this part of our argument, we create an interaction term between the indicator of liberalization of capital account openness and the indicator of global liberalization of capital. This indicator is a proxy for the native elites’ capacity for portfolio diversification. We expect the interaction term to have a positive and statistically significant coefficient, which implies that higher domestic financial openness in the context of increased global financial openness allows for increased portfolio diversification, with its attendant effects on democratic prospects in autocracies (see the Appendix for more details).

One additional, influential binary indicator of financial integration is Bekaert, Harvey, and Lundblad's (Reference Bekaert, Harvey and Lundblad2005) EQUITY measure. It dates equity liberalization episodes for 95 countries from 1980 to 1999. The measure takes the value of “0” before the date of financial liberalization and “1” after it. The data are based on Bekaert and Harvey's (Reference Bekaert and Harvey2004) A Chronology of Important Financial, Economic and Political Events in Emerging Markets. EQUITY indicates the first date from which nonresidents are able to conduct transactions frequently in a country's equity market through initial ADR listings. In a sense, EQUITY can be considered an early indicator of a country's financial integration into world markets. We use it as an alternative to CAPITAL to test the robustness of our results.Footnote ²³

Panels. Finally, we use five-year average panels instead of annual observations because our argument is that changes in inequality have effects on democratic institutions in periods longer than a year. We also use five- year averages because of the uncertainty in measurement caused by economic fluctuations. Economic cycles are usually accounted for in the economics and finance literature using five-year averages (because business cycles are normally three to five years in duration; see Beck and Levine Reference Beck and Levine2004). In addition, we use economic variables as controls (see later discussion), which are drawn from the Penn World Tables (Heston, Summers, and Aten Reference Heston, Summers and Aten2006).

Measurement error and measurement frequency also affect the inequality, democracy, and financial integration variables. Our use of five-year averages addresses these problems. Inequality data are rarely measured on an annual basis in most countries in our sample, thus creating significant structural breaks in the annual data series.Footnote ²⁴ The annual Polity data during revolutionary or coup episodes lasting more than a year are scored with “interruption” codes, leaving investigators the choice of interpolating the Polity scores, omitting the data in question, or averaging over a longer period of time (see the discussion in the Appendix). The financial integration variables are point-in-time variables (as of December 31 of a given year for CAPITAL and a single date in time for EQUITY). Averaging over five years reduces measurement error while allowing for an estimation of long-run effects. Our specifications use five-year nonoverlapping measures, with the units denoted by i = 1, 2,. . ., x and the index s representing five-year intervals, starting at 1955–59 and continuing onward. For instance, Democracy_i,s for the s = 1985–89 period is analyzed using data for the independent variables from the s - 1 = 1980–84 period.

Models and Methods

In this investigation, we are interested in analyzing the separate and joint effects of financial integration and income inequality on democratization. Pooled, cross-section, time-series (PCSTS) models are useful for this purpose because the variation in the dependent variables comes from both the dynamic and cross-sectional factors.

Political economists such as Acemoglu, Johnson, Robinson, and Yared (2008; hereafter AJRY) estimate a simple model with country and time-fixed effects, with Polity levels as the dependent variable. They add the key variable of interest in their investigation (in that case, log of income, lagged once). We adapt their model, adding inequality and inequality squared as the key independent variables conditioned on levels of financial integration. To be more specific, on the basis of our argument and its implications, we add to our statistical specification financial integration measures and interaction terms between income inequality and financial integration. We find, as AJRY (2008) do, persistent serial correlation in some models. We overcome the serial correlation by estimating the dependent variable in changes and by amending the AJRY model with additional lags of the level of the dependent variable where appropriate (see also Barro Reference Barro1999). Because we include lagged levels of the dependent variable on the right side of our equation, we no longer include fixed effects (because the inclusion of unit effects induces serial correlation due to the correlation between the unit effects and the lagged dependent variables). Despite the dynamic nature of our functional form, we focus on marginal effects in the next five-year period.

We used this simple AJRY model to explore the potentially nonlinear relationship derived in AR between inequality and democratization. A hump-shaped relationship, as derived by AR in their Corollary 6.1, implies that intermediate levels of inequality facilitate democratization and higher or lower levels impede democracy. This relationship will appear as a statistically significant positive coefficient on the level of Gini and a statistically significant negative coefficient on Gini squared. A U-shaped relationship between the two variables, as possibly found in Houle Reference Houle2009, would have the opposite and statistically significant signs on the respective coefficients. Contrary to AR's Corollary 6.1, this U-shaped relationship implies that low and high levels of income inequality facilitate democratization. A linear relationship is indicated when the coefficient on the squared inequality measure is not statistically significant, and the coefficient on the base (level) inequality measure is statistically significant when the quadratic term is omitted.

Our argument is that, once we include financial integration's effects, high levels of inequality will be associated with autocratic political reform. The “right-side” of the AR hump will shift up.

Our base model is:

The conditioning effects of financial integration are embodied in the coefficients β₆ and β₇; these coefficients test Implications 1 and 2. In some models, we substitute the EQUITY measure for CAPITAL. (The procedures for establishing the main effects and the confidence intervals are discussed in the methods section of the Appendix.)

To produce a test of Implication 2, we create an interaction for the financial integration of a domestic economy in world markets: CAPITAL_i,s*Global_CAPITAL_∑j-i,s. (See the data section of the appendix for a description of the construction of the variables.) We predict that the easing of restrictions on an autocratic elite's ability to construct diversified portfolios increases the likelihood of democratization, which implies a positive and statistically significant coefficient on this interaction term. In regard to Implication 3, our argument suggests that only autocracies will be affected by financial integration. We therefore separately analyze autocracies and democracies.

AR (2006, 338–42) argue that “Capital_In” and “Capital_Out” (their terms) potentially have different effects on inequality and therefore on the transition to democracy. CAPITAL distinguishes between restrictions on residents and nonresidents, which correspond to restrictions on capital outflows (CAPITAL_Out) and inflows (CAPITAL_In), respectively (see IMF 1993, 80–81).Footnote ²⁵ Indicators for a country's restrictions on inflows and outflow are created, as are global averages of the indicators (see the Appendix for details). The interaction term between the indicator of liberalization of resident exports of capital and the indicator of global liberalization of capital imports is a proxy for the native elite's capacity for diversification. We expect a positive and statistically significant coefficient on the interaction term, which implies that an elite's ability to export capital into world markets is associated with democratic reforms.

Recent scholarship stresses the importance of investigating and controlling for unobserved cross-sectional or spatial correlation in time-series panel studies; (see, for instance, Franzese and Hays Reference Franzese and Hays2007). Gleditsch and Ward (Reference Gleditsch and Ward2006) find that a country's prospects for democracy are influenced by regional forces, as measured by regional averages for democracy. To assess the influence of the behavior of regional neighbors, we compute the average level of democracy for a given country's region (removing the value for that country).Footnote ²⁶ We also add an indicator for a region's integration into world financial markets by computing the country's regional average for capital account openness (net of that country), as well as another indicator of global trends in financial openness. Still another source of spatial correlation is a country's inclusion into the Soviet Bloc. We include a 0, 1 indicator of Soviet Bloc membership. We add this additional information to the base model.

We begin with ordinary least squares (OLS) estimations. OLS estimations are potentially plagued by methodological problems, including the explanatory variables being correlated with the error term. One reason for this correlation may be that democracy and economic inequality are endogenous, which potentially biases the estimates. To address the biases introduced by the possible correlation between the right-hand side variables and the error term, we also use the generalized method of moments (GMM) estimation method of Blundell and Bond (Reference Blundell and Bond1998); it is the same estimator used by Eichengreen and Leblang (Reference Eichengreen and Leblang2003), Milner and Mukherjee (2009), and Quinn and Toyoda (2007), among others. The validity of the instruments is assessed through the Sargan test.Footnote ²⁷ (Details of the GMM model and estimation procedures are found in the methods section of the Appendix.)

To control for the possible influence of economic variables on democracy, we add some standard variables to the GMM model. These include growth in purchasing power parity (PPP) adjusted per capita income and log of levels of investment (as a share of GDP), as used in Boix (Reference Boix2003). To distinguish between trade effects and financial globalization effects, we add the log of levels of trade openness (imports + exports as a percentage of gross domestic product [GDP]).Footnote ²⁸ To control for the separate effects of domestic investment, we add an indicator of investment as a percentage of a country's GDP.

Inequality, Democracy, and Financial Integration

We begin with the results that are most similar to what has appeared in the literature: The simple full sample results from the base model without interaction variables, but with the financial openness measures: See Models 1.1 and 1.2 in Table 1 (CAPITAL and EQUITY, respectively). The signs on the estimated coefficients on the Gini terms are consistent with the AR hump, but, as has been reported previously in the literature, the coefficient estimates are far from statistically significant.

TABLE 1.

Change in Democracy (∆Y = ∆Polity), 1955–2004 (Polity and Capital) or 1980–1999 (Polity and Equity)

Notes: OLS estimates. *< 0.1 significance level; **< 0.05 significance level; ***<.01 significance level. Standard errors are listed below the coefficients.

Implication 1 is that financial integration conditions the relationship between inequality and political change in autocracies. These results are reported in Model 1.3 and Model 1.5. The key results are embodied in the coefficients in rows 4–7 of Table 1.Footnote ²⁹ The coefficient in row 6 for Model 1.3 (GINI*CAPITAL) is positive and statistically significant. This indicates that, for autocracies, the combination of a large Gini (more inequality) and larger value of CAPITAL (more financial openness) produces a greater propensity for democratization in the next five-year period.

To assess the magnitude as well as the statistical significance of the effects of inequality conditioned by financial openness on democracy through the full range of Gini scores, we use a customized simulation program based on CLARIFY. (This simulation program was kindly supplied to us by Michael Tomz. See the methods section of the Appendix for details.) Figure 2 shows the results of the simulations. In a closed economy, denoted by the red line, autocracies with intermediate levels of inequality (Gini scores of 30 to 50) experience a statistically significant, mean increase in Polity scores in the +5 range (because the confidence intervals for this case are centered on + 5 and, between Gini values of 30 and 50, do not contain zero). In contrast, the more unequal a financially open autocracy is (Gini values of 30 to 45), the larger the increase in Polity scores it experiences: In this case, the mean increases from +8 to +13 points on the Polity scale in the subsequent five-year period. The mean democratic gains flatten out, but remain statistically significantly different from zero (the confidence intervals in the open case do not include zero from a Gini value of 25 upward).

FIGURE 2.

The Effects of Inequality on Democracy Conditional on Openness (Open – Capital = 100; Closed – Capital = 0; Model 1.3)

The results for the model with the EQUITY measure for financial integration, Model 1.5, are even more striking. Note the signs on the coefficients for the variables Gini, Gini², Gini*EQUITY, and Gini²* EQUITY. The first pair of coefficients has the signs associated with a hump relationship between inequality and democratization, whereas the latter pair has the signs associated with an inverted U. This means that if the autocratic economy is financially closed (EQUITY = 0), then, as predicted by AR, there indeed is a hump in the relationship between inequality and democracy. But, if the autocratic economy is financially open (EQUITY = 1), the relationship between inequality and democratization is a combination of four terms including Gini*Equity, Gini²*Equity. These last two terms embody the conditioning effect of financial integration on the relationship between inequality and democracy.

The simulation results for Model 1.5 are shown in Figure 3. The overall relationships for the closed and open cases are consistent with Implication 1: Financial openness transforms the shape of the relationship between inequality and democracy from a hump in a closed autocracy to an upward sloping line at higher levels of inequality in an open autocracy. In the financially closed autocratic economy (EQUITY = 0), intermediate levels of inequality (Gini scores of 35 to 40) are associated with +7 to +8.5 mean increases in Polity scores in the next period. The effects of Gini flatten out until Gini scores reach 50 and higher; thereafter the marginal increase in Polity scores decreases. The confidence intervals in the financially closed case do not contain zero from Gini scores of 35 to 58. The financially open autocratic economy case (EQUITY = 1) shows marginal gains in changes in Polity from increasing inequality from a Gini score of 45 upward. At the upper end of the observed Gini scores in our sample (62), a financially open autocratic economy is associated with a +17-point mean increase in Polity. At lower ranges of inequality in financially open autocratic economies, the confidence interval for Gini's effects also does not contain zero.

FIGURE 3.

The Effects of Inequality on Democracy Conditional on Openness (Open – Equity = 1; Closed – Equity = 0; model 1.5)

As Implication 3 suggests, the other models in Table 1 show that there are no such relationships for the democracy-only samples (Model 1.4 and Model 1.6) alone or for the full set of countries (Model 1.7 and Model 1.8); the coefficients in rows 4–7 for these other models are all statistically insignificant. More important, for the models for the democracy-only samples, the Wald statistic for the joint statistical significance of all regressors (except the constant) is not close to statistical significance. The democracy-only models therefore have no explanatory power.

Exchange of Assets and Exchange of Risks

The second major implication of our argument is that the combination of a country's openness to capital inflows and outflows and the development of global financial markets makes it possible for native and foreign elites to diversify their portfolios and, in turn, for native elites in autocracies to more readily accept the consequences of democracy. The key results for this implication are in rows 5, 7, and 8 in Table 2. Model 2.1 shows that changes in the interaction term between the financial openness of autocratic economies (CAPITAL) and Global Markets (Global_CAPITAL) affect the prospects for autocracies becoming more democratic. The overall next-period effect is a combination of three terms including the interaction of these variables. Figure 4 (blue lines) displays this marginal overall effect; it depicts the change in Polity for a one-unit change in domestic capital openness at each observed level of Global Capital. Figure 4 shows that, once Global_CAPITAL reaches 62.5 (roughly three-quarters of a standard deviation above its mean), these effects are all positive, statistically significant, and substantively large. Consider a move by an autocracy from being completely closed (CAPITAL = 0) to being fully open (CAPITAL = 100) under conditions of high Global Capital openness. Such a move would be associated with a subsequent mean +6.3 increase in Polity. A move from complete closure (CAPITAL = 0) to partial openness (CAPITAL = 50) under conditions of high Global Capital openness would be associated with a subsequent +3.2 increase in Polity. Under conditions of low financial integration (Global Capital openness equals 45), moves from a closed to partially open to a fully open autocracy are associated with subsequent mean changes of -1.5 and -3 in Polity, respectively.Footnote ³⁰ Financial exit without financial integration therefore does not appear to be associated with subsequent democratization.

TABLE 2.

Change in Democracy (∆Y = ∆Polity), 1955–2004 – OLS Estimations

Notes: OLS estimates. *< 0.1 significance level; **< 0.05 significance level; ***<.01 significance level. Standard errors are listed below the coefficients.

FIGURE 4.

The Marginal Effects of Global Financial Integration on Policy Scores in Autocracies [Model 2.1 (Capital) and A2.1 (Capital_out)]

Model 2.3 refines these results by adding into the specification the GINI*CAPITAL (or β₆) interaction term from Model 2.1.Footnote ³¹ The coefficients for both the GINI*CAPITAL term and the CAPITAL*Global_CAPITAL terms in Model 2.3 are positive and statistically significant. Adding the GINI*CAPITAL term increases the explanatory power of Model 2.3 in comparison to Model 2.1. Both effects on democratization—of financial integration and of increasing inequality in the context of financial integration– are found in autocracies. With the addition of the GINI*CAPITAL interaction term to the model, the Gini and Gini² coefficient estimates for Model 2.3 are again statistically significant: They indicate a hump in the context of a financially closed autocratic economy (CAPITAL = 0). Finally, as suggested by Implication 3, these effects are weaker to nonexistent for democracy-only samples (Models 2.2 and 2.4). The Wald statistic for the joint statistical significance of all regressors (except the constant) is not close to significance in the democracy-only samples. The results for the full sample are reported for reader comparison (Models 2.5 and 2.6); they resemble those of the autocracy-only sample (Models 2.1 and 2.3), although these models for the full samples are characterized by weaker substantive effects.

Robustness

The results in Table 1 and 2 are based on the Polity measure of democracy, and they make no correction for possible correlation between the right-hand side variables and the error term. To address these issues, we performed additional analyses with Polity as the dependent variable and with the Blundell and Bond (Reference Blundell and Bond1998) GMM estimator. Table 3, Models 3.1, 3.2, and 3.5, report GMM estimation results that correspond to OLS Models 1.3, 1.4, and 1.7, respectively, from Table 1. The GMM estimation yields a statistically insignificant Sargan test statistic, which indicates that the instruments are not correlated with the error term and can be considered valid. There is no indication of serial correlation in any of the relevant models. Table 3, Models 3.3, 3.4, and 3.6, report the GMM estimation results using Regime as the dependent variable.Footnote ³² The relevant coefficient estimates for the autocracy-only sample for both the Polity (Model 3.1) and Regime (Model 3.3) for the Gini and CAPITAL variables are larger and estimated more precisely. If measurement error plagues the OLS models, and if the GMM instruments are sound, the accuracy of the estimates will increase.

TABLE 3.

Change in Democracy (∆Y = ∆Polity or ∆Regime), 1955–2004 – GMM Estimations

Notes: No serial correlation is indicated in GMM models when the Arellano-Bond test for second-order serial correlation is (ABm2) not significant at the .05 level and beyond, and where we expect to find from the ABm1 test evidence of statistically significant negative serial correlation in the differenced residuals. Standard errors are listed below the coefficients. *p value < 0.10; **p value < 0.05; ***p value < 0.01.

Substantively, the results of the robustness checks in Table 3 are reassuring. For example, for the autocracy-only sample, the coefficient on the interaction term for GINI*CAPITAL (s-1) is still positive and statistically significant, indicating that financial integration conditions the effect of inequality on democratization. Equally important, the results hold up for a different measure of democratization: The results for Regime are similar to those for Polity. For instance, for the autocracy-only sample, a highly unequal autocracy (at 62) that has a fully financially liberalized economy is associated with a mean .9 increase in Regime in the subsequent period (Model 3.3).Footnote ³³ The evidence is that unequal, financially open autocracies liberalize. Once again, the democracy-only sample shows a Wald statistic for the joint significance of the regressors that is far from statistical significance.

The companion GMM estimations for Table 2 are reported in Appendix Table A1. The Sargan test statistics for the estimations in Table A1 are statistically insignificant, indicating the instruments are not correlated with the errors. These GMM results are substantively similar to those for the OLS models in Table 3, although the coefficient estimates for the key variables are larger and have generally higher levels of statistical significance. The inclusion of both the interaction term for financial integration and the interaction term for inequality and financial openness improves the explanatory power of the respective models. Both effects are found in autocracies. As before, the models estimated on the democracy-only sample have Wald statistics for the joint significance of the regressors that are far from statistical significance. The democracy-only models have little to no explanatory power.

As a final robustness check, we explored the possible differential effects of resident exports of capital compared to nonresident imports of capital. We created disaggregated measures of outflow and inflow restrictions, respectively. The variables are substituted into the models from Table 2 (Models 2.1, 2.2, and 2.5), as well as from Table A1 (Models A1.1, A1.2, and A1.5). The outflow and inflow restriction measures replace the aggregate CAPITAL measure. The results for the disaggregated measures of domestic inward and outward capital flow restrictions in autocracies are reported in Appendix Table A2 (rows 5–10). Both OLS and GMM estimators are used. Again, the overall effect of financial integration is a complex combination of the indicated variables, including the interaction terms. Figure 4 (red lines) depicts the effects for one variable, CAPITAL_Out for Model A2.1, the autocracy-only sample using OLS conditioned on the variable CAPITAL_In (CAPITAL_Out and CAPITAL_In are rescaled from 0–50 to 0–100 so their effects can be compared to CAPITAL.) The statistically significant effects for this variable are evident after Global_Capital_In reaches a level of 56, which is approximately at the series mean. The substantive effect of full financial openness for resident capital exports joined to high levels of global openness for capital imports is large. Given a Global_Capital_In score of 70, a move by an autocracy from closure to openness for resident capital exports produces an estimated mean +8.9 change in its Polity score.Footnote ³⁴ The effects are robust across the OLS (A2.1) and GMM (A2.3) autocracy-only samples. In the democracy-only samples, the p value of the Wald statistic for the joint significance is far from statistical significance. From this evidence, most of the effects from autocratic integration into global markets come from the ability of the native elite to export capital rather than from the ability of nonresident elites to acquire domestic capital.

The control variables are not central to our analysis. Only a few have consistent next-period effects within the autocracy-only specifications. As expected, we find that regional forces affect autocratic prospects for democratization. In most autocracy-only models, the coefficient estimate on the variable for Regional Democracy is positive and statistically significant, as is consistent with the findings in Gleditsch and Ward (Reference Gleditsch and Ward2006). Change in income in the GMM estimations has a positive and statistically significant coefficient in most specifications, suggesting that increasing income is associated with subsequent democratization in these models.

In results available from the authors, generalized least squares (GLS) estimates for Table 1, in place of OLS estimates, are reported. The results are very similar to the OLS results. The estimated coefficient estimates of the relevant variables in GLS tend to be a little larger and the estimates of the standard errors a little smaller compared to the OLS estimates. In results available from the authors, we reanalyze the results from Table 1 using OLS and change in Regime in place of Polity. The results are very similar: Unequal financially open autocracies tend to democratize. Finally, we reanalyze the results of Table 3 using change in Regime as the dependent variable and OLS. The results are again very similar: Financially integrated autocracies tend to democratize. The explanatory power of the democracy-only models ranges from null to negligible.