Corruption, capital account liberalization, and economic growth: Theory and evidence

Author's Accepted Manuscript

Corruption, capital account liberalization, and economic growth: Theory and evidence Takuma Kunieda, Keisuke Okada, Akihisa Shibata

www.elsevier.com/locate/inteco

PII: DOI: Reference:

S2110-7017(14)00013-4 http://dx.doi.org/10.1016/j.inteco.2014.03.001 INTECO30

To appear in:

International Economics

Cite this article as: Takuma Kunieda, Keisuke Okada, Akihisa Shibata, Corruption, capital account liberalization, and economic growth: Theory and evidence, International Economics, http://dx.doi.org/10.1016/j.inteco.2014.03.001 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Corruption, Capital Account Liberalization, and Economic Growth: Theory and Evidence Takuma Kunieda∗ Department of Economics and Finance, City University of Hong Kong Keisuke Okada† Faculty of Human Environmental Studies, Hiroshima Shudo University Akihisa Shibata‡ Institute of Economic Research, Kyoto University

Abstract We investigate, both theoretically and empirically, how the negative effect of government corruption on economic growth is magnified or reduced by capital account liberalization. Our model shows that highly corrupt countries impose higher tax rates than do less corrupt countries, thereby magnifying the negative impact of government corruption on economic growth in highly corrupt countries and reducing the impact in less corrupt countries if capital account liberalization is enacted. Empirical evidence obtained from an analysis of panel data collected from 109 countries is consistent with our theoretical predictions, namely the interaction term of government corruption and financial openness has a significant and negative impact on economic growth, implying that financial openness magnifies the negative effect of government corruption on economic growth. Our theoretical and empirical results contribute to the recent policy debates on the merits and demerits of capital account liberalization.

JEL classification: O40; O43; F43; E62. Keywords: Economic growth, Government corruption, Capital account liberalization, Two-country model. ∗

Corresponding author. P7318, Academic Building, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon Tong, Hong Kong, Phone: +852 3442 7960, Fax: +852 3442 0195, E-mail: [email protected] † Faculty of Human Environmental Studies, Hiroshima Shudo University, 1-1-1 Ozuka-higashi, Asaminamiku, Hiroshima 731-3195, JAPAN, Phone: +81 82 830 1239, Fax: +81 82 830 1950, E-mail: [email protected] ‡ Institute of Economic Research, Kyoto University, Yoshida-honmachi, Sakyo-ku, Kyoto 606-8501, JAPAN, Phone: +81 75 753 7126, Fax: +81 75 753 7198, E-mail: [email protected]

1

1

Introduction

The impacts of corruption and financial globalization on economic growth have been investigated separately in many studies on empirical growth. On the one hand, the question of whether government corruption impedes economic growth and development has been long studied by many researchers. Since the pioneering work of Mauro (1995), many subsequent researchers, such as Ehrlich and Lui (1999), Glaeser and Saks (2006), and Mo (2001), have presented evidence indicating that government corruption hinders economic growth and development.1 On the other hand, the effect of financial globalization on economic growth has also been extensively studied by many researchers empirically (e.g., Chanda, 2005; Eichengreen and Leblang, 2003; Quinn, 1997; Quinn and Toyoda, 2008; Rodrik, 1998). However, it is still unclear whether financial globalization is necessarily beneficial to all countries. Since the early 1980s, many emerging East Asian and Latin American countries have enacted capital account liberalization (Lane and Milesi-Ferretti, 2007; Kose et al., 2009, 2010). At the same time, the World Bank has estimated that corruption costs the world economy more than one trillion dollars a year.2 Given these stylized facts of financial globalization and corruption, we investigate the impact of the combination of financial openness and government corruption on economic growth. As pointed out by Weil (2008), corrupt officials may waste public funds by, for instance, awarding contracts to private agents who pay the largest bribes, rather than to those who are the most efficient, or by putting the taxes collected directly into their own pockets. 1

Moreover, Wei (2000a, 2000b) studied the effect of corruption on foreign direct investment (FDI) and found that corruption is a significant obstacle to FDI. Additionally, Acemoglu and Verdier (1998), de Vaal and Ebben (2011), Ehrlich and Lui (1999), Mauro (2004), and Shleifer and Vishny (1993) among others have theoretically clarified that government corruption is harmful to economic growth and development. Yano (2009) emphasizes that high quality markets are essential to the healthy growth of an economy. Government corruption is one of the typical factors that reduce market quality. 2 See the News & Broadcast article of the World Bank at http://go.worldbank.org/LJA29GHA80. According to this article, one trillion dollars were paid worldwide in 2001-2002 as bribes in both developed and developing countries. Note that this amount does not include the embezzlement of public funds or theft of public assets. Daniel Kaufmann, the World Bank Institute’s former director for Governance and Anti-Corruption, says, “It is important to emphasize that this is not simply a developing country problem. Fighting corruption is a global challenge.”

2

Corruption takes many forms, such as bribes paid to government officials, trading of official contracts for cash, and embezzlement of public funds.3 As such, it is now widely accepted that government corruption is a serious obstacle to economic growth and development.4 Regarding the effect of financial globalization on economic growth, evidence is mixed, and it remains unclear whether all countries can benefit from financial globalization.5 Bekaert et al. (2005), Kraay (1998), and Quinn and Toyoda (2008) find no evidence that capital account liberalization positively affects economic growth, even though it has a high level of institutional quality. In contrast, Arteta et al. (2003), Durham (2004), and Klein (2005) find supportive evidence for its positive effect. However their results are not particularly robust, and depend upon the specifications and sample period. Along the same lines as these studies, we analyze panel data from 109 countries and examine whether capital account liberalization amplifies or mitigates the negative effect of government corruption on economic growth. Here, we obtain empirical evidence indicating that capital account liberalization is beneficial to less corrupt countries, but is disadvantageous to highly corrupt countries. Our study is not the first to investigate how the combination of financial openness and government corruption affects economic growth. Neeman et al. (2008) find empirical evidence that government corruption has a negative effect on the level of gross domestic product (GDP) in highly open countries, but has almost no effect in strongly closed countries. Our study departs from that of Neeman et al. (2008) in several respects. First, whereas they perform a cross-country analysis, we apply a panel-data analysis following Islam (1995) and Levine et al. (2000).6 Second, they study the impact of corruption on the level of GDP. We follow the traditional empirical specification that has been applied to growth regression since Barro (1991) and Mankiw et al. (1992) and assess the impact of corruption on the 3

The World Bank (1997, 2001) has identified the root causes of corruption. In contrast, some studies argue that corruption may actually help economic development (Leff, 1964). See also Bardhan (1997) for a survey of this branch of the corruption literature. 5 See Kose et al. (2009, 2010) for a comprehensive survey of the literature on financial globalization. 6 As pointed out by Islam (1995), the main benefit of applying the panel approach in empirical growth is that it makes it possible to control for the heterogeneity of production technologies across economies. 4

3

GDP growth rate. Third, they use the black market premium as the measure of financial openness, but we use both de jure and de facto measures of financial openness in our estimations. More specifically, we use the capital account openness index developed by Chinn and Ito (2006, 2008) as the de jure measure of financial openness. This index reflects the intensity of capital controls and the nature of capital policies. Following Kose et al. (2009, 2010), we also use the sum of total assets and total liabilities divided by GDP as the de facto measure of financial openness. Here, we compute the sum of total assets and total liabilities from the dataset created by Lane and Milesi-Ferretti (2007). Fourth, although Neeman et al. (2008) present a theoretical model that explains how government corruption hinders production, they do not explicitly model an open economy. In contrast, we explicitly consider financial openness in our model. We incorporate the behavior of a corrupt government into our model, as did Mauro (2004) and de Vaal and Ebben (2011). However, in contrast to these two studies, we develop a two-country model in which the two countries differ in their degree of corruption.7 We theoretically find that if government corruption is less prevalent in a country, capital account liberalization leads to a higher growth rate, and if government corruption is highly prevalent, capital account liberalization leads to a lower growth rate. In other words, capital account liberalization magnifies the negative effect of government corruption on economic growth in highly corrupt countries, but reduces the negative effect of government corruption on economic growth in less corrupt countries. These findings, obtained using our two-country model, are novel theoretical contributions to the existing literature on corruption and growth. These theoretical findings are also consistent with our empirical findings that capital account liberalization is beneficial to less corrupt countries, but is disadvantageous to highly corrupt countries. The mechanism behind these theoretical results is as follows. If a country is 7

The two-country setting simplifies the derivations of our main result. In Appendix A.2, we extend the two-country model to a many-country model, and obtain similar results. Note that if the number of countries goes to infinity, each country becomes a small open economy.

4

financially closed to the world market, government corruption affects economic growth only through a higher tax rate on labor income in equilibrium (including both the observed tax rate and unobserved bribes), and the magnitude of the effect of government corruption is thus relatively limited. However, if two countries are financially integrated, the negative effect of taxation on an investment project is higher in the highly corrupt country than in the less corrupt country. This is because the return on investment in the highly corrupt country becomes smaller than in the less corrupt country. Consequently, financial resources flow into the less corrupt country from the highly corrupt country. Accordingly, in the twocountry setting, the negative impact of government corruption is magnified in the highly corrupt country but reduced in the less corrupt country.8 The remainder of this paper proceeds as follows. In the next section, we develop a growth model for economies in the presence of government corruption. In section 3, we derive equilibrium growth rates in the case of both closed and open economies. In section 4, we discuss the mechanism through which the effect of government corruption on growth-rate differences is magnified in the two-country setting when compared to the closed economy case. In section 5, we provide empirical evidence for our theoretical findings, and in section 6, we present our concluding remarks.

2

Theoretical Framework for Corruption and Capital Account Liberalization

Although the current model is based on the growth model developed by Kunieda and Shibata (2011) and Kunieda et al. (in press), we extend their model economy to include government 8

Our theoretical model also contributes to the literature on the Lucas paradox, because the theoretical consequences are consistent with existing empirical evidence indicating that weak institutional circumstances in poor countries are obstacles to capital inflow from rich countries and may even induce capital flight from the poor countries (e.g., Alfaro et al., 2008; Wei, 2000a, 2000b). Capital does not flow from rich countries to poor countries as much as the neo-classical growth models predict (Lucas, 1990; Obstfeld and Taylor, 2004; Stulz, 2005). The volume of capital that moved to poor countries in recent years is a very small proportion of the total capital flow in the international financial market (Mishkin, 2007). In contrast to the standard neo-classical growth models, our model shows that, when government corruption is prevalent and a country opens its financial market to the world market, it is highly likely that capital will flow out. This will dampen the country’s economic growth significantly, more so than when the country is a closed economy.

5

corruption. The model economy consists of the government, an infinitely lived representative firm, individuals in overlapping generations, and bureaucrats who are potentially corrupt. Each individual in a generation lives for two periods, meaning that young and old agents always coexist in each period. We denote the population of young agents born at time t by Lt . One need not impose any assumption on the lifetime of bureaucrats because they consume in a hand-to-mouth manner as explained later, but we assume that the population of bureaucrats is γLt , for all t ≥ 0, where 0 < γ < 1.9 Time is discrete, expanding from 0 to ∞. Each individual born at time t obtains utility exclusively from her second-period consumption, ct+1 . Because she is risk neutral, her utility function is given by u(ct+1 ) := ct+1 .10 The timing of events from time t to time t + 1 is as follows. • Individuals are born at the beginning of time t. • Production at time t occurs, and individuals earn wages. The government collects sales tax from the representative firm. • Individuals make decisions on how much they invest, borrow, and/or lend. • At the end of time t, the government makes decisions on corruption and the sales tax rate imposed on the representative firm’s production for time t + 1. • Production at time t + 1 occurs, and the government collects sales tax from the representative firm. Individuals receive their returns on investment and lending. They repay their obligations if they borrowed at time t. Individuals consume the whole of their income. 9

This assumption is imposed for simplicity of exposition. Matsuyama (2004) also assumes that individuals consume only in the second period of their lifetimes. However, even if we assume that individuals consume in both the first and second periods of their lifetimes, the essence of the model does not change, although the investigation becomes complicated. In particular, if we use a Cobb-Douglas utility function or a log-linear utility function, the main result is identical to the one derived from the model in the main text. 10

6

As will be addressed later, the government’s choice variables are the misappropriation rate and the sales tax rate. The government is sub-benevolent, because its decisions are biased toward corrupt bureaucrats. Without corrupt bureaucrats, the government would choose the sales tax rate that would maximize per capita consumption in the economy. Note that, because the collective decisions of the government are made at the end of time t, neither the old agents at time t nor the young agents at time t+1 are involved in the decision making political process regarding time t + 1. The government commits to its policy decisions made at the end of time t with regard to time t + 1.

2.1

Production Sector

The final goods are produced from capital and labor. Following Barro (1990) and Futagami et al. (1993), we incorporate public capital into the production function. Specifically, the production function is given in a Cobb-Douglas form as follows: Yt = AZtα (gt Lt )1−α ,

0 < α < 1,

(1)

where Yt is the output, Lt is the aggregate labor, Zt is the aggregate capital, and A is the technology level of the production function.11 We denote the public spending per young agent by gt . Since the government produces public capital based on public spending with a one-for-one technology, we may also call gt the public capital per young agent. Both the capital and public capital depreciate entirely in one period. Since the government imposes a sales tax on the final product, the representative firm maximizes its net profit, (1 − τ t ) Yt −wt Lt −qt Zt , where τ t is the tax rate on the final product, wt is the wage rate, and qt is the price of the capital. Note that the tax rate, τ t , includes the observed tax rate and unobservable bribes. In this sense, bribes are considered to be tax from the perspective of the private sector. More specifically, τ t is divided into two parts: τ t = τ˜t + θt τ t , where 0 < θt < 1. Here, τ˜t Yt is the observed and collected tax revenue, and 11

A is a certain positive constant. The technology level should be large enough to ensure positive growth as in the traditional AK model.

7

θt τ t Yt is the amount that the representative firm has to pay to corrupt bureaucrats as bribes, which is unobservable. In other words, as assumed in Shleifer and Vishny (1993), we consider that corrupt bureaucrats sell the representative firm the license to operate its business. The bribes are embezzled by corrupt bureaucrats, and we refer to θt the misappropriation rate, which is determined in accordance with the government’s collective decision. Note that the representative firm maximizes its profit, taking the government’s behavior as given. In other words, the government and the representative firm are in a game-theoretic situation, where the government is a Stackelberg leader and the representative firm a Stackelberg follower. Given the misappropriation rate, θt , and the tax rate, τ t , the production factors in the competitive markets are paid their respective marginal products as follows:

2.2

qt = α (1 − τ t ) Yt /Zt

(2)

wt = (1 − α) (1 − τ t ) Yt /Lt .

(3)

Individuals

Each individual faces respective budget constraints in the first and second periods as follows: kt + d t ≤ w t ,

(4)

ct+1 ≤ qt+1 φkt + rt+1 dt ,

(5)

and

where kt is the investment in a project and dt is lending when positive and borrowing when negative. If an individual starts an investment project when young, she produces capital φkt in the second period, which is sold to the representative firm at price qt+1 , and φ is the productivity of the capital production. If the individual lends financial resources in the first period, she receives a gross return of rt+1 in the second period. If she borrows financial resources in the first period, she pays a gross interest rate of rt+1 in the second period, the same rate as the lending rate. 8

Because of agency problems in the financial market, investors face borrowing constraints. Following Aghion et al. (2005), the credit constraint facing each individual is given by dt ≥ −νwt ,

(6)

where ν ∈ [0, ∞) measures the degree of credit constraints. We note that individuals can borrow financial resources up to ν×wt ; here, wt can be considered a down-payment for the investment project. In Appendix A.1, we provide two kinds of microfoundations for Eq. (6).12 The non-negativity constraint for the investment project is given by kt ≥ 0.

(7)

Now, we introduce the heterogeneity of individuals with respect to the productivity of capital creation. Specifically, the productivity, φ, varies between individuals and is distributed uniformly over [0, 1]. Each individual knows her own productivity at birth, but does not know the productivity of any other individuals. In addition, φ is considered an idiosyncratic productivity shock as assumed in Angeletos (2007). As we will see later, owing to the heterogeneity of individuals’ talents and the credit constraints they face, two financially integrated countries can experience sustainable growth, even though there may be a difference in the equilibrium interest rates when each country is a closed economy.13 Each individual maximizes ct+1 subject to inequalities (4)-(7). The maximization problem is rewritten as max (rt+1 − φqt+1 ) dt dt

subject to −

μ wt ≤ dt ≤ wt , 1−μ

12

We implicitly assume the existence of a financial intermediary for the loan contracts between savers and borrowers. See Appendix A.1 for the microfoundations for the credit constraint (6). This type of assumption for credit market imperfections is often imposed in the literature (e.g., Aghion and Banerjee, 2005; Aghion et al., 1999; Aghion et al., 2005). Even if we replace the inequality (6) with bt ≥ −μkt , where μ ∈ [0, 1), this alternative credit constraint is equivalent to inequality (6), and the same results are obtained. 13 Our model is comparable to the AK model. In the standard AK model, if two countries with different equilibrium interest rates are financially integrated, sustainable growth cannot be achieved in the country with the lower interest rate. However, in our model, sustainable growth can be achieved in countries owing to the heterogeneity of individuals’ talents and the credit constraints.

9

where μ :=

ν . 1+ν

When rt+1 − φqt+1 > 0, it is optimal for an individual to choose dt = wt

μwt and kt = 0, whereas when rt+1 − φqt+1 < 0, it is optimal for her to choose dt = − 1−μ and

kt =

wt . 1−μ

Formally, we obtain the following lemma:

Lemma 1 Let φt :=

rt+1 . qt+1

Then, the following hold.

• If φt > φ, then kt = 0 and dt = wt . • If φt < φ, then kt =

wt 1−μ

μwt and dt = − 1−μ .

Lemma 1 implies that each individual chooses either to start an investment project or to lend funds to investors, depending upon her entrepreneurial talent. Note that φt is the cutoff that divides agents into savers and borrowers.

2.3

Government’s Behavior

Bureaucrats are potentially corrupt. It is implicitly assumed that bureaucrats affect the government’s collective decision. As assumed in section 2.1, the representative firm pays bribes to bureaucrats to operate its business smoothly. Other than affecting the government’s collective decision and receiving bribes, the bureaucrats do not actively impact the economy because they only consume bribes that they receive in a hand-to-mouth manner. As assumed in Futagami et al. (1993) among others, the government runs a balanced budget, given by14 gt+1 Lt+1 = τ˜t+1 Yt+1 = τ t+1 (1 − θt+1 )Yt+1 .

(8)

From this equation, we can rewrite the production function (1) as follows: (1−α)/α

Yt+1 = A1/α (1 − θt+1 )(1−α)/α τ t+1

Zt+1 .

(9)

Thus, the equations of capital price (2) and wage (3) can be written respectively as (1−α)/α

qt+1 = αA1/α (1 − τ t+1 ) (1 − θt+1 )(1−α)/α τ t+1

(1−α)/α

wt+1 = (1 − α) A1/α (1 − τ t+1 ) (1 − θt+1 )(1−α)/α τ t+1 14

(10) zt+1 ,

(11)

Our interest is not in public debt and the assumption for a balanced budget simplifies our investigation.

10

where zt+1 := Zt+1 /Lt+1 . From Lemma 1, we compute the aggregate capital supplied by investors as 

1

Zt+1 =

φkt Lt dφ = φt

wt (1 − (φt )2 ) Lt . 2 (1 − μ)

(12)

For simplicity, we assume that Lt is constant, namely Lt+1 = Lt for all t ≥ 0. From Eqs. (11) and (12), we obtain the law of motion of capital as follows: (1−α)/α

zt+1 =

(1 − τ t ) τ t

(1 − α) A1/α (1 − θt )(1−α)/α (1 − (φt )2 ) zt . 2 (1 − μ)

(13)

The dynamic behavior of capital is subject to the tax rate τ t , the misappropriation rate θt , the degree of credit constraints, and the number of savers (borrowers). Collective decisions on both the tax rate (τ t+1 ) and the degree of political corruption (θt+1 ) are affected by corrupt bureaucrats, and accordingly, the geometric average between consumption per capita in the generation and bribe per corrupt bureaucrat is maximized as follows: 1−β β bt+1 , max c¯t+1

τ t+1 ,θ t+1

where c¯t+1 is the consumption per capita in the generation and bt+1 the bribe per corrupt bureaucrat; β is a measure of the degree of political corruption.15 Since the government is a Stackelberg leader, it solves its maximization problem by considering the representative firm’s first-order conditions, namely, Eqs. (10) and (11). The setting for the governmental maximization problem is specific to the current model, but it simply describes the behavior of the corrupt government. The consumption by individuals with φ < φt is given by ct+1 = φt qt+1 wt , and that with φ > φt+1 is given by ct+1 = (φ − μφt )qt+1 wt /(1 − μ). Therefore, it follows that  c¯t+1 Lt =



φt 0

φt qt+1 wt Lt dφ +

15

1 φt

φ − μφt qt+1 wt Lt dφ, 1−μ

This specification for political corruption is specific to the current model, but is convenient for characterizing the behavior of the corrupt government. This maximization is equivalent to the maximization of the geometric average between the total disposable income and the total bribe at time t + 1; that is, 1−β β [θt+1 τ t+1 Yt+1 ] . maxτ t+1 ,θt+1 [(1 − τ t+1 ) Yt+1 ]

11

or equivalently,  1 1 dφ + (φ − μφt )dφ, 1 − μ φt 0   1 (φt )2 − 2μφt + 1 . = 2 (1 − μ)

c¯t+1 = φt qt+1 wt



φt

(14)

Moreover, from Eqs. (9) and (12) it follows that (1−α)/α

Yt+1 = A1/α (1 − θt+1 )(1−α)/α τ t+1



 1 − (φt )2 wt Lt / (2 (1 − μ)) .

Hence, from Eq. (8), the bribe per corrupt bureaucrat is given by bt+1 =

2 θt+1 τ t+1 Yt+1 1/α (1 − (φt ) ) wt . = A1/α θt+1 (1 − θt+1 )(1−α)/α τ t+1 γLt 2 (1 − μ) γ

Since the choice variables of the collective decisions are τ t+1 and θt+1 , and qt+1 is given by Eq. (10), the government’s maximization problem is converted into two independent maximization problems: (1−α)(1−β)+β α

max (1 − τ t+1 )1−β τ t+1 τ t+1

,

(15)

and max (1 − θt+1 ) θ t+1

1−α α

θβt+1 .

(16)

These two maximization problems are solved as follows: τ ∗ := (1 − α) (1 − β) + β,

(17)

αβ . (1 − α) (1 − β) + β

(18)

and θ∗ :=

These solutions imply that the sales tax rate and the misappropriation rate are increasing functions with respect to the degree of political corruption. If there are no corruption motives, that is, β = 0, the tax rate will be equal to 1 − α, the same rate that Barro (1990) derives. Note that the tax rate that can be actually observed is computed as τ˜t+1 = (1−θ∗ )τ ∗ = 1−α, which is not affected by the degree of corruption β. Although higher corruption results in a 12

higher tax rate, τ ∗ , this rate includes unobserved bribes, making it practically impossible to empirically test the relationship between τ ∗ and the degree of corruption. Since (1 − θ∗ )τ ∗ = 1 − α, the capital price and wage rate become, respectively, q¯ : = αA1/α (1 − α)(1−α)/α (1 − τ ∗ ) w¯t : = (1 − α)1/α A1/α zt (1 − τ ∗ ) . Defining the growth rate of an economy as Γt+1 :=

zt+1 , zt

(19) (20)

from Eq. (13) and (1 − θ∗ )τ ∗ = 1 − α

we obtain Γt+1

3

(1 − τ ∗ ) (1 − α)1/α A1/α (1 − (φt )2 ) . := 2 (1 − μ)

(21)

Equilibrium Growth Rates

The equilibrium growth rate is derived from Eq. (21) such that the domestic financial market (in a closed economy model) or the world financial market (in a two-country model) clears.

3.1

Closed Economy Model

From Lemma 1, the excess supply of financial resources at time t in the country is given by  Bt :=

φt 0

 wt Lt dφ −

φt

= w t Lt φ t − =

1

μwt Lt dφ 1−μ

μwt Lt (1 − φt ) 1−μ

φt − μ w t Lt . 1−μ

(22)

In a closed economy, the financial market clears within a country. Therefore, the financial market clearing condition is given by Bt = 0, or equivalently, φt = μ.

(23)

From Eqs. (21) and (23), we obtain the equilibrium growth rate of the closed economy as follows: Γt+1

(1 − τ ∗ ) (1 − α)1/α A1/α (1 + μ) . := 2 13

(24)

Note that the growth rate in a closed economy declines as the corruption parameter, β, increases because of the higher tax rate on wage income.

3.2

Two-Country Model

Although we perform a panel data investigation of 109 countries in section 5, the use of a two-country model simply exposes the interactive relationship between corruption, financial openness, and economic growth. In Appendix A.2, we show that a many-country model provides similar results to the two-country model. Suppose that the world economy consists of two countries, country 1 and country 2. To investigate the effect of government corruption on economic growth, we assume that the two countries are identical in terms of degree of credit constraints, technology, and size of population, but not in terms of degree of government corruption. These assumptions seem to be strong, but enable us to focus on the effect of government corruption on economic growth, all other things being equal.16 We impose a parameter condition, Assumption 1, as shown below. Assumption 1 μ (1 − β 1 ) ≤ 1 − β 2 < 1 − β 1 , where β i (i = 1,2) is the corruption parameter of country i. In the second inequality of Assumption 1, we assume that the degree of corruption in country 1 is less than that of country 2, that is, β 1 < β 2 . From Eq. (17), the equilibrium tax rate of country 1 is less than that of country 2, that is, τ 1 < τ 2 , and from Eq. (18), the misappropriation rate of country 1 is less than that of country 2; that is, θ1 < θ2 . As will be shown later, the first inequality guarantees that there are always agents who create capital in both countries. In fact, if the degree of credit constraints, μ, is large enough for μ (1 − β 1 ) > 1 − β 2 , then no one will produce capital in country 2 for a sufficiently large t. We avoid this case for simplicity. 16

In particular, the same degree of financial friction between the two countries makes our computations simple.

14

Since the two countries are financially integrated, they face a common world interest rate. Thus, it holds that rt+1 = q¯1 φ1,t = q¯2 φ2,t , where q¯i is the capital price and φi,t is the cutoff in country i. From Eq. (19) and 1 − τ i = α (1 − β i ), we can rewrite q¯1 φ1,t = q¯2 φ2,t as (1 − β 1 ) φ1,t = (1 − β 2 ) φ2,t .

(25)

From β 1 < β 2 it follows that φ1,t < φ2,t , implying that when the two countries are financially integrated, the number of savers (borrowers) is greater (smaller) in country 2 than in country 1. From Eq. (22), the net foreign assets at time t in country i, which is the excess supply of financial resources from the perspective of the domestic market, is given by Bi,t =

φi,t −μ wi,t Lt . 1−μ

The international financial market clears over the two countries, and it must hold that B1,t + B2,t = 0,. Thus, φ1,t < μ < φ2,t for all t ≥ 0.17 The inequalities B1,t < 0 and B2,t > 0 indicate that country 1 is always a net borrower and country 2 is always a net lender in the international financial market. Note that, in each period, country 1 borrows resources from country 2, and in the next period it pays off its interest-bearing debt to country 2. In contrast, in each period, country 2 lends resources to country 1, and in the next period, receives the proceeds of investing in country 1. From B1,t + B2,t = 0, 1 − τ i = α (1 − β i ), and Eq. (20), we have 

   φ1,t − μ (1 − β 1 ) z1,t + φ2,t − μ (1 − β 2 ) z2,t = 0.

(26)

From Eqs. (25) and (26), we obtain the cutoffs of the two countries as follows: φ1,t = μ +

μ(β 1 − β 2 )z2,t (1 − β 1 )(z1,t + z2,t )

(27)

φ2,t = μ +

μ(β 2 − β 1 )z1,t . (1 − β 2 )(z1,t + z2,t )

(28)

and

17

Although we do not analyze the dynamic behavior of φi,t , it suffices for us to know that φ1,t < μ < φ2,t when the growth rates are compared.

15

Since φ2,t <μ + μ(β 2 − β 1 )/(1 − β 2 ) from Eq. (28), it follows from Assumption 1 that φ2,t < 1, for all t ≥ 0, This implies that there are always agents who create capital in both countries.18 From β 1 < β 2 and Eqs. (27) and (28), we can confirm that φ1,t < μ < φ2,t for all t ≥ 0. Note that from Eqs. (21) and (24) the effect of government corruption on the country’s economic growth is reflected only in the tax on wage income in the closed economy model. However, the effect is reflected in both the tax rate and the cutoff, φi,t , in the two-country model. The taxation reduces the return on investment, and financial resources flow out from the country with the higher tax rate to the country with the lower tax rate. We can now compare the growth rates of country 1 and country 2. Proposition 1 Under Assumption 1, suppose that country 1 and country 2 are identical in terms of their degree of credit constraints, technology, preferences, and size of population, but are different in their degree of corruption. Then, we can rank the growth rates as follows: Γ1,o > Γ1,c > Γ2,c > Γ2,o , where Γi,o is the growth rate when country i is an open economy and Γi,c is the growth rate when country i is a closed economy. Proof: If both countries are closed economies, it follows from Eq. (24) that Γ1,c > Γ2,c because τ 1 < τ 2 . If both countries are open economies, it follows from φ1,t < μ < φ2,t and Eq. (21) that Γ1,o > Γ1,c and Γ2,c > Γ2,o .  We have provided an analysis of the cases in which two countries are perfectly open or closed. Although it is difficult to investigate the case in which two countries are semi-open using the current model, we can predict that the growth rate of a less-corrupt, semi-open country falls between Γ1,o and Γ1,c and that of a more-corrupt, semi-open country falls between Γ2,c and Γ2,o . 18

Since φ1,t < φ2,t holds in equilibrium, φ1,t < 1 is guaranteed.

16

4

Discussion

While Γ1,o and Γ2,o in Proposition 1 are time-variant, the ranking of the growth rates does not change. Proposition 1 generically implies that the differences in growth rates between the two countries are magnified by capital account liberalization. Capital account liberalization is beneficial to a less corrupt country relative to other countries, but it is unfavorable to a highly corrupt country. Our model provides a possible mechanism through which capital account liberalization amplifies the negative effect of corruption on economic growth. The equilibrium interest rate is greater in a less corrupt country than in a highly corrupt county when each is a closed economy because the return on an investment project is greater in the less corrupt country than in the highly corrupt country. The growth rate in a closed economy is independent of interest rates, and the negative effect of corruption on the economic growth of a country is reflected only in the tax rate on wage income. Once the two countries are financially integrated, however, financial resources will flow from a country with a low interest rate to a country with a high interest rate, as is well known in international economics, and in equilibrium, both the countries will face the common world interest rate. The effects of the inflow and outflow of financial resources on the economic growth of the two countries will be reflected in the equilibrium cutoff, φi,t . The number of savers in a highly (less) corrupt country is greater (less) than that in a less (highly) corrupt country. The effects of inflow and outflow of financial resources on economic growth magnify the negative effect of government corruption when the capital account is liberalized. As a result, the difference in the growth rates between the two countries is enlarged. Although the equilibrium interest rate plays a crucial role in our theoretical model, it is beyond the scope of our study to investigate empirically whether the equilibrium interest rate is really greater in a less corrupt country than in a highly corrupt country when each is a closed economy.19 19

It is very difficult to investigate this question because we must collect real interest rates that are reliable

17

Capital restrictions are exogenous in our model. As investigated by Dreher and Siemers (2009), however, capital restrictions may be a function of corruption. In other words, it is more likely that corrupt countries impose capital controls because corruption prevents the government from collecting tax revenue. This situation is absent in our theoretical model. Nevertheless, the testable implication regarding corruption, financial openness, and economic growth presented in Proposition 1 is a contribution of our theoretical model. One might argue that the government of a highly corrupt country can impose a tax on the net foreign assets held by its domestic agents and/or the profits from the net foreign assets. In this way, it could correct the distortion created by taxation on the domestic product, and capital account liberalization would thus still be beneficial to the country. In reality, however, introducing such an international taxation regime would be difficult because the private agents of a country may take an international tax avoidance strategy.20 To avoid being taxed, multinationals often resort to transfer pricing, tax havens, money laundering, treaty shopping, and thin capitalization.21 For example, by analyzing the income shifting through transfer pricing in a large selection of OECD countries, Bartelsman and Beetsma (2003) conclude that more than 65% of the marginal revenue arising from a country’s tax increase is lost through income shifting.22

5

Empirical Evidence

We have provided a theoretical argument that capital account liberalization is beneficial to less corrupt countries, but is disadvantageous to highly corrupt countries. In this section, we verify this proposition empirically . and comparable across strongly closed countries. 20 It is beyond the scope of this study to discuss the international tax avoidance strategy. See chapter 3 in Kobetsky (2011) for a discussion on the shortcomings of international taxation. 21 Treaty shopping means that by establishing a subsidiary in some country, firms utilize the country’s treaties on international double taxation to reduce their tax burdens. Thin capitalization means that firms raise the share of debt financing in high tax countries to deduct interest payments from the sales tax base. 22 See also Bloomberg’s article at http://bloom.bg/SQGG5G on December 10, 2012. According to this article, big multinationals significantly avoided being taxed in 2011 and the European Union incurred a loss of more than one trillion US dollars in one year as a result of tax avoidance

18

5.1

Data

The data for this study are drawn from various databases. Depending upon the availability of datasets, we collected the annual data of 109 countries over the period 1985-2009. The countries in our sample are listed in Table A4 of Appendix A.4. The per capita real GDP growth rates for our dataset are calculated using the per capita real GDP obtained from the World Development Indicators of the World Bank (2012). We use two types of indicators for financial openness, both of which must be associated with capital account liberalization. The first is the capital account openness index developed by Chinn and Ito (2006, 2008). The Chinn-Ito index is a de jure measure of capital account openness, reflecting the intensity of capital controls and nature of capital control policies (henceforth, we refer to this as “de jure financial openness”). The Chinn-Ito index values range from −1.8556 to 2.4557, with the larger values indicating greater capital account liberalization. Following Kose et al. (2009, 2010), we also consider the sum of total assets and total liabilities divided by GDP as a measure of financial openness. As discussed in Kose et al. (2009, 2010), this indicator is a de facto measure of capital account openness (henceforth, we refer to this as “de facto financial openness”). This indicator was computed from the dataset for total assets and liabilities created by Lane and Milesei-Ferretti (2007).23 It is not easy to measure corruption, because corruption is illegal and so unreported. We use the corruption index created by the PRS Group (2011), which publishes the International Country Risk Guide. This corruption index is perception-based and subjective, and assesses the corruption taking place within a political system. While the original index ranges from most corrupt (0) to least corrupt (6) in an index scale, we rescaled this index to range from least corrupt (0) to most corrupt (6) to enable us to interpret the index more easily. We incorporate various control variables used in the literature on empirical growth in 23

The data points in the dataset created by Lane and Milesi-Ferretti (2007) are available only up to 2007, so we averaged the indicators from 2005 to 2007 for the last period.

19

our estimation. We include the natural logarithm of initial real GDP per capita to control for the stage of economic development. The initial real GDP per capita data are taken from the World Development Indicators of the World Bank (2012). We also include education, the private credit/GDP ratio, and investment/GDP ratio. Education is a proxy for human capital, measured by the average years of total schooling of the population aged over 15 years, as developed by Barro and Lee (2013). The private credit/GDP ratio is often used as a proxy for financial development in the literature on finance and growth.24 The private credit/GDP ratio data are taken from Beck et al. (2010), entitled “private credit by deposit money banks and other financial institutions to GDP” in their database. The data for investment/GDP ratios are collected from the “investment share of GDP per capita” in the Penn World Table (Heston et al., 2011). In our theoretical model, investment induces the growth rate, as seen in Eq. (12), and reflects various growth determinants such as government corruption, credit constraints, and total factor productivity. However, there may be other channels through which investment affects a country’s growth rate. To capture the effects of investment that do not appear in our theoretical model, we incorporate investment as a control variable. For robustness checks, we also control for trade, inflation, government expenditures, population growth, and life expectancy. The data for these control variables were obtained from the World Development Indicators of the World Bank (2012). With the exception of the initial per capita GDP and education, we averaged each variable, including control variables, for five years, following the procedure used in the literature on empirical growth (e.g., see Levine et al., 2000).25 Accordingly, we create our dataset for the following non-overlapping five periods: 1985-1989, 1990-1994, 1995-1999, 2000-2004, and 2005-2009. The use of five-year averaged data enables us to mitigate the noises associated with short-run economic fluctuations. In all estimations, all variables other than the 24 25

See Levine (2005). The data for education can be collected every five years. See Barro and Lee (2013).

20

initial per capita GDP and education are treated as endogenous variables.26 The detailed definitions and data sources are provided in Table A5 of Appendix A.4, and the descriptive statistics of the variables are shown in Table A6 of Appendix A.4.

5.2

Estimation Method

When examining the effects that financial openness and its interaction with corruption have on economic growth, we must address the endogeneity problem associated with reverse causality from economic growth to explanatory variables. For instance, if the growth rate is low, the salary of government officials might be very low owing to low tax revenue. In this case, the government officials will have a greater incentive to accept bribes, and corruption will be more prevalent in such an economy. Financial openness also seems to be an endogenous variable because as an economy develops, the government is more likely to open the financial market to the world. To control for such simultaneity bias, we conduct a dynamic panel data analysis with country-specific fixed effects.27 Arellano and Bond (1991), Arellano and Bover (1995), and Blundell and Bond (1998) develop the linear generalized method of moments (GMM) estimators for dynamic panel models. In our analysis, we specifically use the system GMM estimators.28 We estimate the following equation: yi,t − yi,t−1 = (α1 − 1) yi,t−1 + α2 Financial opennessi,t + α3 Corruptioni,t +α4 Financial opennessi,t × Corruptioni,t + Xi,t β + t + η i + ui,t , (29) where i and t represent a country and time, respectively. In addition, is a time-specific 26

The initial per capita GDP and education are treated as predetermined variables. Another possible estimation strategy is to perform simple cross-country regressions with instrumental variables. However, in our estimations, we have too many endogenous variables, including financial openness, government corruption, their interaction term, private credit, and investment to perform simple cross-country regressions. Since it is very difficult to identify external instrumental variables to mitigate the endogeneity problems associated with all these endogenous variables, we conduct a dynamic panel data analysis in this study. As proposed by Henry (2007), it would be also convincing to develop an estimation strategy using firm or industry level data or using the policy-experiment approach. Nevertheless, one of the merits of following the traditional growth regression approach is that we can compare our result with those of many existing studies on empirical growth. 28 See Arellano and Bover (1995) and Blundell and Bond (1998) for the moment conditions. 27

21

effect, η is a country-specific effect, and u is an error term. Then, y is the logarithm of per capita real GDP. Other than the endogenous variables discussed previously, we have good reason to use the system GMM because yi,t−1 is not strictly exogenous but pre-determined. Note that we can regard yi,t −yi,t−1 as the average growth rate in the t th period. Furthermore, X is the set of control variables including a constant. Our theory predicts that α4 is negative because financial globalization leads to a decrease in the growth rate of a highly corrupt country, and that α2 is positive, because financial globalization leads to an increase in the growth rate of a less corrupt country. To obtain consistent estimates, we need to address the validity of the instruments and therefore we consider two specification tests. The first test examines the hypothesis that the error terms are not serially correlated. For this, we test whether the differenced error terms are serially correlated with respect to the second order.29 The second test is the Hansen test of over-identifying restrictions. This tests the orthogonality conditions of the instruments. We must also address the small sample bias associated with estimating the variancecovariance matrix because the number of countries in our dataset is at most 109, and thus our sample size is limited. In the second-step estimation, when performing the two-step system GMM, the residuals from the first-step estimation are used to produce a consistent estimate of the variance-covariance matrix. However, this estimate will be severely downward-biased if the sample size is small. Windmeijer (2005) has developed corrected standard errors to correct such a small sample bias. We report Windmeijer’s (2005) corrected standard errors in our estimation results. Given the size of our sample, we must also consider the number of moment restrictions, because if we use too many instruments, the system GMM estimators would be unable to eliminate endogenous components. Moreover, using too many instruments would reduce the power of the Hansen test of over-identifying restrictions (See Bowsher, 2002; Roodman, 29

Since we examine the differenced error terms, the examination of the first-order serial correlation makes no sense.

22

2009; and Ziliak, 1997). Therefore, we use only two-to-three-period lagged variables as instrumental variables.

5.3

Results

Columns (1) and (2) in Table 1 present the estimation results without the interaction term between financial openness and corruption as a benchmark. In columns (1) and (2), we use the de jure financial openness and the de facto financial openness, respectively. [Table 1 here] The coefficients of financial openness are negative in both columns (1) and (2), although they are not significant. Both education and investment are significant and positive in both estimations, just as they are in the prior literature on empirical growth (e.g., Barro, 1991). However, a somewhat surprising result is that private credit, a proxy for financial development, has a significant negative impact on economic growth.30 While this result contradicts the traditional literature on finance and growth (Aghion et al., 2005; Levine, 2005; Levine et al., 2000), it is consistent with the empirical results recently obtained by Loayza and Ranci`ere (2006), Rousseau and Wachtel (2011), and Saci et al. (2009).31 To investigate the validity of the instrumental variables in columns (1) and (2), we perform the Hansen test of over-identifying restrictions. The Hansen test does not reject the orthogonality conditions at the conventional significance level in the estimation in column (1). However, the Hansen test does reject the orthogonality conditions for the estimation in column (2) at the 10% significance level. The instrumental variables in column (2) may not 30

This result is obtained even if we eliminate investment from our estimations. We include private credit in all estimations as a control variable. This procedure may not be standard, but without it, the Hansen test does not reject the orthogonality conditions in almost all the estimations. The relationship between finance and growth is an ongoing debate. Recent evidence shows that the development of financial intermediation has a negative impact on growth in the short run and a positive impact in the long run (Loayza and Ranci`ere, 2006). Since each data point of our dataset is created by averaging the original data points for only five years, the short-run effect of financial development on economic growth probably dominates the long-run effect. Existing evidence also indicates that the development of financial intermediation has a negative impact on growth if the data points are extended until quite recently (Rousseau and Wachtel, 2011). Therefore, a relook into the empirical claims on the relationship between finance and growth has become a necessary task for future research. 31

23

be valid. For the system GMM estimator to be consistent, no serial correlations of the error terms should exist. We examine whether the differenced error terms in columns (1) and (2) are serially correlated with respect to the second order (the AR (2) test). The p-values in both estimations are so large that the AR (2) tests cannot reject the null hypothesis of no second-order serial correlation of the error terms. Columns (3) and (4) in Table 1 present our main estimation results, including the interaction term between financial openness and corruption. In column (3), we use the de jure financial openness, and in column (4), we use the de facto financial openness. In both columns (3) and (4), the coefficients of financial openness and its interaction with corruption are statistically significant at the conventional significance level. The signs of the coefficients of financial openness are positive, and the signs of its interaction terms with corruption are negative. The negative signs of the interaction terms imply that financial openness amplifies the negative impact of corruption on economic growth, which is consistent with our theoretical predictions in Proposition 1. The coefficients of the other control variables are stable compared to the estimations in columns (1) and (2). Fig. 1 (left) illustrates the 95% confidence interval for the marginal effect of financial openness at various levels of corruption, based on the result in column (3).32 This figure suggests that the merits of capital account liberalization are reduced by corruption. In column (3), the partial impact of financial openness is (0.0144 − 0.0049 × corruption), which provides the threshold value of corruption, 2.9388, that divides countries according to the partial impact of financial openness on economic growth. If the degree of corruption is below this threshold, the partial effect of financial openness on economic growth is positive, whereas if the degree of corruption is above this threshold, its partial effect is negative. This threshold value is an approximate median value in our sample. See also Fig. 1 (right), which shows 32

At the bottom of each table for the estimation results, we also report the point estimate and F -statistics of the marginal effects of financial openness (corruption) on economic growth given the minimum, mean, and maximum value of corruption (financial openness).

24

the 95% confidence interval for the marginal effect of corruption at various levels of financial openness. If capital account liberalization exceeds the threshold value of financial openness, 0.2449, the negative impact of corruption becomes large as capital account liberalization proceeds. However, this figure suggests that, if financial openness is very small, corruption promotes economic growth. Although this result is surprising as opposed to our theoretical prediction of the closed economy model, it is consistent with the assertion of Leff (1964) who points out that corruption would be desirable to economic development because it may perform the valuable function of a safeguard against the losses of bad economics regulations when the government is proceeding in the wrong direction.33 The coefficients of corruption in the estimations in columns (1) and (2) are insignificant. This insignificance may reflect the presence of some strongly closed countries in which economic growth is promoted by corruption. Nevertheless, the negative and significant coefficients of the interaction terms indicate that the negative impact of corruption on economic growth is magnified by financial openness as suggested by Proposition 1. [Fig. 1 here] In the estimations in columns (3) and (4), the Hansen tests do not reject the orthogonality conditions at the conventional significance level and the AR (2) tests do not reject the null hypothesis of no second-order serial correlation of the error terms, either. However, if the instrumental variables that we used in our estimations weakly identify endogenous variables, our results indicating that financial openness amplifies the negative impact of corruption may not be robust. Unfortunately, the procedure proposed by Bazzi and Clemens (2013) cannot reject under-identification or weak instruments in the estimations in columns (3) and (4).34 Because there are too many endogenous variables, we cannot directly tackle the 33

Our theoretical model does not capture the assertion of Leff (1964) because we assume that corrupt bureaucrats consume bribes in a hand-to-mouth manner. Our model focuses more on the effect that the combination of corruption and financial openness has on economic growth. 34 We test for under-identification by applying the Keibergen-Paap LM test and for weak instruments by applying the Cragg-Donald and Kleibergen-Paap Wald tests (Cragg and Donald, 1993; Keibergen and Paap, 2006) for both difference and level equations. See Bazzi and Clemens (2013) for details of the test procedure.

25

weak identification problem following the procedure proposed by Bazzi and Clemens (2013). Therefore, we adopt a strategy of estimating another specification to check the robustness of our results.35 We split the sample into two subsamples according to the degree of financial openness and estimated the following equation, which does not contain the interaction term between financial openness and corruption: yi,t − yi,t−1 = (α5 − 1) yi,t−1 + α6 Corruptioni,t + Xi,t β + t + η i + ui,t .

(30)

We prepared three pairs of split subsamples depending on differing cutoffs of financial openness; the cutoffs are 0, 0.2449, and 0.5 on the Chinn-Ito index, where strongly open countries have greater index values. See Table 2 for the results of the split-sample estimations. If financial openness amplifies the negative impact of corruption on economic growth, the coefficient of corruption α6 in strongly open countries should be negative as well as less than that in less open countries. In the estimations in columns (1), (2), and (3) for strongly open countries, the coefficients of corruption are negative and significant. Although the Hansen tests in columns (1) and (2) reject the orthogonality conditions at the 10% significance level, the test in column (3) does not. In contrast, in the estimations in columns (4), (5), and (6) for strongly closed countries, the coefficients of corruption are not significantly different from zero. The implications from these results are consistent with those of the estimations in columns (3) and (4) in Table 1, implying that the negative effect of corruption is magnified by financial openness. [Table 2 here] The initial level of GDP per capita insignificantly enters in all the estimations in Table 1, but in the estimations for strongly open countries in Table 2, it significantly enters and has a negative impact. This outcome may appear because the subsamples of strongly open economies are likely to exclude less developed countries, many of which may fall in a poverty 35

As mentioned, in all estimations, all variables but the initial per capita GDP and education are treated as endogenous variables. It is very difficult to apply a weak-instrument-robust test, such as the Anderson and Rubin (1949) test, because we have too many endogenous variables.

26

trap in which economic growth stagnates. In such a case, even if accumulated capital is very small in less developed countries, the growth rate can be very low. Additionally, note that corruption enters insignificantly in the estimations in columns (1) and (2) in Table 1. We estimated the same specification by limiting the sample countries to those used in Mauro (1995). However, corruption is still insignificant in most cases within our estimation period, although the result is not reported here. Given the estimation results of columns (3) and (4) in Table 1, and of strongly open countries in Table 2, we note that it is important to investigate the impact of corruption on economic growth by considering it together with capital account liberalization. We perform robustness checks by adding various control variables that have been considered important determinants of economic growth in previous studies (e.g., Levine and Renelt, 1992) to estimations (3) and (4) in Table 1. Tables 3 and 4 provide the results with the de jure financial openness and the de facto financial openness, respectively. All estimation results in Table 3 are robust and similar to that in column (3) in Table 1 in terms of both the magnitudes of the coefficients and their significance. This implies that the estimation results in Table 3 are consistent with our theoretical hypothesis. The Hansen tests for all estimations in Table 3 do not reject the orthogonality conditions at the conventional significance level. Furthermore, the AR(2) tests for all estimations in Table 3 do not reject the null hypothesis of no second-order serial correlation of the error terms. [Table 3 here] Comparing the estimation results in Table 4 with that in column (4) in Table 1, we find that the estimation result in column (5) is similar to that in column (4) in Table 1. In columns (1) through (3) in Table 4, although the signs and magnitudes of all the coefficients, other than the logarithm of initial GDP per capita, are stable, the coefficients of financial openness are not significant. Moreover, the coefficient of the interaction term of financial openness and corruption is not significant in column (2). In the estimation in column (4), the magnitudes 27

of the coefficients of both financial openness and its interaction term with corruption are smaller than in other estimations. The Hansen tests of over-identifying restrictions and the serial correlation tests show the validity of the instrumental variables in all estimations except in column (3). In the estimation in column (3), although the AR (2) test does not reject the null hypothesis of no second-order serial correlation of error terms, the Hansen test rejects the orthogonality condition at the 10% significance level. [Table 4 here] Although the estimation results in Table 4 are consistent with our theoretical hypothesis in the sense that the signs of the coefficients of financial openness and its interaction term with corruption support our hypothesis, the effects of the de facto financial openness on economic growth are likely to be mitigated by such control variables as trade, inflation, government expenditure, and population growth. There is a caveat to our estimation results even though they are consistent with our model’s prediction. We applied the system GMM estimators by using two-to-three-period lagged variables, including the lagged corruption index, as instrumental variables. However, government corruption seems to have a long-run impact on economic growth. If this is so, the lagged corruption index cannot be a valid instrumental variable, because it must be correlated with the differenced error terms and the moment conditions proposed by Arellano and Bond (1991), Arellano and Bover (1995), and Blundell and Bond (1998) do not hold. To examine whether this problem does seriously affect our estimation results, we performed further robustness checks. Specifically, we estimated Eq. (29) by removing all the lagged corruption terms and all the lagged interaction terms between financial openness and corruption from the set of the instrumental variables, still assuming that corruption, financial openness and their interaction term in the right-hand side of Eq. (29) are endogenous variables. As shown in Tables A1, A2, and A3 in Appendix A.3, the estimation results are similar to those in Tables 1, 3, and 4, respectively. 28

6

Concluding Remarks

We investigated how the interaction between capital account liberalization and government corruption affects economic growth from both a theoretical and empirical point of view. Our theoretical results are as follows. Highly corrupt countries tend to impose a higher tax rate (including the observed tax rate and unobserved bribes) than do less corrupt countries. This magnifies the negative impact of government corruption on the economic growth of highly corrupt countries if the countries liberalize their capital accounts. On the other hand, the negative impact of government corruption on the economic growth of less corrupt countries is reduced if the countries liberalize their capital accounts because of the inflow of financial resources from the highly corrupt countries. As a result, the least corrupt, financially open countries experience the highest growth rates; the least corrupt, financially closed countries experience the second-highest growth rates; highly corrupt, financially closed countries experience the third-highest growth rates; and highly corrupt, financially open countries experience the lowest growth rates. Empirical evidence obtained from our analysis of panel data collected from 109 countries supports our theoretical predictions. In other words, the interaction term of government corruption and financial openness has a negative and significant impact on economic growth. This implies that financial openness magnifies the negative effects of government corruption in the growth regression. Our theoretical model and empirical evidence also suggest that capital account liberalization is beneficial to less corrupt countries and unfavorable to highly corrupt countries. This study analyzed the interactive effects of capital account liberalization and government corruption on economic growth. We have made a novel contribution to the existing literature on corruption and growth and on financial globalization and growth. Furthermore, we believe that our research contributes to the recent policy debates on the merits and demerits of capital account liberalization.

29

Appendix A.1 Microfoundations for Credit Constraints We describe two types of microfoundations for a credit constraint in this appendix. Microfoundation I We follow the model settings of Aghion and Banerjee (2005), Aghion et al. (1999), and Aghion et al. (2005) to provide a microfoundation. When borrowing, each agent is endowed with her own wealth wt , which is wages earned when she is young. Therefore, her total resources to invest are kt = wt − dt . As seen in the main text, qt+1 φ is the return on one unit of investment. If a borrower faithfully repays her obligations, she earns a net income of qt+1 φkt +rt+1 dt . On the other hand, if the borrower does not repay her obligations, she incurs a cost, δkt , to conceal her revenue. If this happens, a financial intermediary monitors the borrower and captures her with probability pt+1 . Hence, if she decides to default, she obtains the expected income, qt+1 φkt − δkt + pt+1 rt+1 dt . The incentive compatibility constraint under this lending contract, which leads the borrower not to default, is as follows: qt+1 φkt + rt+1 dt ≥ (qt+1 φ − δ) kt + pt+1 rt+1 dt ,

(A.1)

which is rewritten as dt ≥ −

δ kt , rt+1 (1 − pt+1 )

(A.2)

The borrower acquires the revenue given in the left-hand side of Eq. (A.1) when she starts a project and faithfully repays her obligations. The borrower’s gain when she defaults is given in the right-hand side. Note that Eq. (A.2) is independent of the return on one unit of investment. The financial intermediary selects the optimal probability pt+1 to detect the borrower’s deception; however, it incurs an effort cost, dt Φ(pt+1 ), so as to attain the optimal probability. 30

The effort cost is increasing and convex with respect to pt+1 . As in Aghion and Banerjee (2005), we assume Φ(pt+1 ) = κ log(1 − pt+1 ), where κ is strictly greater than δ such that all borrowers face credit constraints more severe than their natural debt limits. The financial intermediary solves the following maximization problem: max − pt+1 rt+1 dt − κ log (1 − pt+1 ) dt . pt+1

Since −dt > 0, this maximization problem is equivalent to max pt+1 rt+1 + κ log (1 − pt+1 ) . pt+1

From the first-order condition, we have κ . 1 − pt+1

rt+1 =

(A.3)

The increase in the interest rate rt+1 leads to a high probability of detecting defaulting borrowers. From Eqs. (A.2) and (A.3), we obtain δ d t ≥ − kt , κ or equivalently, dt ≥ −

δ wt . κ−δ

(A.4)

Although the financial intermediary does not impose agent-specific credit constraints because the agent’s productivity φ is unobservable, it must know the agent’s wealth, wt . None will default in equilibrium, provided that the financial intermediary imposes the credit constraint given by inequality (A.4) on all agents. Since δ < κ, we can let ν := δ/ (κ − δ) ∈ [0, ∞), and thus dt ≥ −νwt . This is the credit constraint in the main text. Here, δ and κ are associated with a default cost and a monitoring cost, respectively. As δ increases or κ decreases, a financial market is considered to be fully developed. 31

Microfoundation II Antr`as and Caballero (2009) develop a microfoundation for a credit constraint. We extend their microfoundation in a manner suitable to our model. We consider the participation constraint faced by the financial intermediary and the incentive compatibility constraint of borrowers that leads them not to default. At the end of the first period of each borrower’s lifetime and after investment has occurred, any borrower can walk away without carrying out her investment project. She takes some fraction of her investment with no cost, (1 − μ) (wt − dt ), where 0 < μ < 1, without repaying her obligations to the financial intermediary. In this case, the borrower will be engaged in capital production somewhere, and sell the capital in a market. If a borrower absconds at the end of the first period, the financial intermediary can withdraw the remainder of the investment, μ (wt − dt ). We assume that the financial intermediary can relend the remainder of the investment in the financial market. This implies that when making a financial contract with a borrower, the financial intermediary faces a participation constraint, such that rt+1 μ (wt − dt ) ≥ −rt+1 dt , or equivalently, dt ≥ −

μ wt . 1−μ

The incentive compatibility constraint for a borrower, which leads her not to abscond from her project, is given by φqt+1 (wt − dt ) + rt+1 dt ≥ φqt+1 (1 − μ) (wt − dt ) .

(A.5)

Eq. (A.5) always holds for agents with φ such that rt+1 − μφqt+1 ≤ 0. Hence, we focus on agents with φ such that rt+1 − μφqt+1 > 0. Then, Eq. (A.5) becomes dt ≥ −

μ w. (φt /φ) − μ 32

(A.6)

Since it follows that φt /φ ≤ 1 in equilibrium, we obtain −μ/ [(φt /φ) − μ] ≤ −μ/ (1 − μ), which implies that Eq. (A.6) is redundant. In summary, borrowers never default if the financial intermediary imposes a credit constraint of dt ≥ −μwt / (1 − μ), which is the participation constraint of the financial intermediary. Letting μ/ (1 − μ) := ν, we have the credit constraint bt ≥ −νwt as shown in the main text. As μ or ν increases, it becomes more difficult for borrowers to withdraw their investment without repaying their obligations. If we consider these variables as being associated with the legal protection of the lenders, a financial market develops fully as the variables increase.

A.2 Many-Country Model In this appendix, we show that a many-country model provides the same results as those of the two-country model. Suppose that a finite set, N , consists of n countries. As in the two-country model in section 3.2, to focus on the effect of government corruption on economic growth, we assume that all countries are identical in terms of their degree of credit constraints, technology, and size of population, but not in terms of their degree of government corruption. As in section 3.2, the ranking of corruption is assumed as follows: β 1 < · · · < β i < · · · < β n.

(A.7)

For the same reason as in Assumption 1, we impose the following assumption: Assumption 2 μ(1 − β 1 ) ≤ 1 − β n . Since all countries face the same world interest rate, we obtain a similar equation to Eq. (25) as follows: (1 − β 1 ) φ1,t = · · · = (1 − β n ) φn,t .

33

(A.8)

As in the two-country model, the international financial market clears over all the countries and thus we have

 i∈N

Bi,t = 0, or equivalently, 

 φi,t − μ (1 − β i ) zi,t = 0.

(A.9)

i∈N

From Eqs. (A.8) and (A.9), we obtain: φi,t (1 − β i ) =

μ



− β j )zj,t , j∈N zj,t

j∈N (1



(A.10)

for all i ∈ N . As the right-hand side of Eq. (A.10) is common across all countries, we obtain the ranking of the cutoff φi,t . More corrupt countries have a higher value of φi,t . Additionally, from Eq, (A.9), there are highly corrupt countries whose cutoff is greater than μ and there are less corrupt countries whose cutoff is smaller than μ. Therefore, a simple manipulation using Eq. (21) yields a similar result to the one in Proposition 1. Of course, if n goes to infinity, each country becomes a small open economy.

A.3 Further Robustness Checks Table A1 shows the estimation results when the specifications are the same as in Table 1, except that we eliminate all the lagged corruption terms and the lagged interaction terms between financial openness and corruption from the set of instrumental variables. Each column in Table A1 shows estimation results very similar to those in Table 1. The AR(2) tests do not reject the null hypothesis of no second-order serial correlation for any estimations. The Hansen tests of over-identifying restrictions do not reject the orthogonality conditions at the conventional significance level in the estimations using the de jure financial openness in columns (1) and (3). However, the tests do reject the orthogonality conditions in the estimations using the de facto financial openness in columns (2) and (4) at the 10% significance level. [Table A1 here] In Tables A2 and A3, we carry out estimations using the same specifications as in Tables 3 and 4 respectively, removing all the lagged corruption terms and the lagged interaction 34

terms. All columns in Tables A2 and A3 indicate estimation results very similar to those in Tables 3 and 4, respectively. The AR(2) tests do not reject the null hypothesis of no second-order serial correlation for any estimations. The Hansen tests of over-identifying restrictions do not reject the orthogonality conditions at the conventional significance level in all estimations in Table A2, which uses the de jure financial openness. However, when we use the de facto financial openness and compare the results to columns (4) and (5) in Table 4, the Hansen tests show significance in columns (4) and (5) in Table A3. Other than these Hansen tests, the coefficients of financial openness and its interaction term with corruption in Table A3 are very close to those in Table 4. By carrying out further robustness checks, we can confirm that there is no endogeneity problem associated with the lagged corruption index affecting our estimation results. [Table A2 here] [Table A3 here]

A.4 Data Description See Tables A4-A6. [Table A4 here] [Table A5 here] [Table A6 here]

Acknowledgments This work was supported by Grant-in-Aid for Specially Promoted Research (No. 23000001), Grant-in-Aid for Research Activity Start-up (No. 24830119), the Joint Research Program of KIER, and City University of Hong Kong (No. 7200307). The authors are grateful to two anonymous referees, Shin-ichi Fukuda, Yasuyuki Sawada, Kosuke Aoki, Tomohiro Hirano, Junichi Fujimoto, and the seminar participants at the University of Tokyo and Hiroshima 35

University for their invaluable comments and discussions. The first author would like to express his special thanks to Bharat R. Hazari for his personal discussions. Of course, all remaining errors are ours.

References [1] Acemoglu D, Verdier T. Property rights, corruption and the allocation of talent: a general equilibrium approach. Economic Journal 1998;108(450):1381-403. [2] Aghion P, Banerjee A. Volatility and growth. New York: Oxford University Press; 2005. [3] Aghion P, Banerjee A, Piketty T. Dualism and macroeconomic volatility. Quarterly Journal of Economics 1999;114(4):1359-97. [4] Aghion P, Howitt P, Mayer-Foulkes D. The effect of financial development on convergence: theory and evidence. Quarterly Journal of Economics 2005;120(1):173-222. [5] Alfaro L, Kalemli-Ozcan S, Volosovych V. Why doesn’t capital flow from rich to poor countries?

An empirical investigation. Review of Economics and Statistics

2008;90(2):347-68. [6] Anderson TW, Rubin H. Estimation of the parameters of a single equation in a complete system of stochastic equations. Annals of Mathematical Statistics 1949;20(1):46-63. [7] Angeletos G-M. Uninsured idiosyncratic investment risk and aggregate saving. Review of Economic Dynamics 2007;10(1):1-30. [8] Antr`as P, Caballero RJ. Trade and capital flows: a financial frictions perspective. Journal of Political Economy 2009;117(4):701-44. [9] Arellano M, Bond S. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 1991;58(2):277-97. 36

[10] Arellano M, Bover O. Another look at the instrumental variable estimation of errorcomponents models. Journal of Econometrics 1995;68(1):29-51. [11] Arteta C, Eichengreen B, Wyplosz C. When does capital account liberalization help more than it hurts? In: Helpman E, Sadka E. (Eds.), Economic policy in the international economy: essays in honor of Assaf Razin, Cambridge: Cambridge University Press; 2003, pp. 177-206. [12] Bardhan P. Corruption and development: a review of issues. Journal of Economic Literature 1997;35(3):1320-46. [13] Barro RJ. Government spending in a simple model of endogenous growth. Journal of Political Economy 1990;98(5):S103-25. [14] Barro RJ. Economic growth in a cross section of countries. Quarterly Journal of Economics 1991;106(2):407-43. [15] Barro RJ, Lee J-W. A new data set of educational attainment in the world, 1950-2010. Journal of Development Economics 2013;104:184-98. [16] Bartelsman EJ, Beetsma RMWJ. Why pay more? Corporate tax avoidance through transfer pricing in OECD countries. Journal of Public Economics 2003;87(9-10):222552. [17] Bazzi S, Clemens MA. Blunt instruments: avoiding common pitfalls in identifying the causes of economic growth. American Economic Journal: Macroeconomics 2013;5(2):152-86. [18] Beck T, Demirg¨ uc¸-Kunt A, Levine R. Financial development and structure dataset (updated November 2010). Washington, DC: World Bank; 2010.

37

[19] Bekaert G, Harvey CR, Lundblad C. Does financial liberalization spur growth? Journal of Financial Economics 2005;77(1):3-55. [20] Blundell R, Bond S. Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics 1998;87(1):115-43. [21] Bowsher CG. On testing overidentifying restrictions in dynamic panel data models. Economics Letters 2002;77(2):211-20. [22] Chanda A. The influence of capital controls on long run growth: where and how much? Journal of Development Economics 2005;77(2):441-66. [23] Chinn MD, Ito H. What matters for financial development? Capital controls, institutions, and interactions. Journal of Development Economics 2006;81(1):163-92. [24] Chinn MD, Ito H. A new measure of financial openness. Journal of Comparative Policy Analysis 2008;10(3):309-22. [25] Cragg JG, Donald SG. Testing identifiability and specification in instrumental variable models. Econometric Theory 1993;9(2):222-40. [26] de Vaal A, Ebben W. Institutions and the relation between corruption and economic growth. Review of Development Economics 2011;15(1):108-23. [27] Dreher A, Siemers LHR. The nexus between corruption and capital account restrictions. Public Choice 2009;140(1-2):245-65. [28] Durham JB. Absorptive capacity and the effects of foreign direct investment and equity foreign portfolio investment on economic growth. European Economic Review 2004;48(2):285-306. [29] Ehrlich I, Lui FT. Bureaucratic corruption and endogenous economic growth. Journal of Political Economy 1999;107(S6):S270-93. 38

[30] Eichengreen B, Leblang D. Capital account liberalization and growth: was Mr. Mahathir right? International Journal of Finance and Economics 2003;8(3):205-24. [31] Futagami K, Morita Y, Shibata A. Dynamic analysis of an endogenous growth model with public capital. Scandinavian Journal of Economics 1993;95(4):607-25. [32] Glaeser EL, Saks RE. Corruption in America. Journal of Public Economics 2006;90(67):1053-72. [33] Henry PB. Capital account liberalization: theory, evidence, and speculation. Journal of Economic Literature 2007;45(4):887-935. [34] Heston A, Summers R, Aten B. Penn World Table version 7.0, Philadelphia: Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania; 2011. [35] Islam N. Growth empirics: a panel data approach. Quarterly Journal of Economics 1995;110(4):1127-70. [36] Klein MW. Capital account liberalization, institutional quality and economic growth: theory and evidence. NBER Working Paper 11112, Cambridge: National Bureau of Economic Research; 2005. [37] Kleibergen F, Paap R. Generalized reduced rank tests using the singular value decomposition. Journal of Econometrics 2006;133(1):97-126. [38] Kobetsky M. International taxation of permanent establishments: principles and policy. Cambridge: Cambridge University Press; 2011. [39] Kose MA, Prasad E, Rogoff K, Wei S-J. Financial globalization: a reappraisal. IMF Staff Papers 2009;56(1):8-62.

39

[40] Kose MA, Prasad E, Rogoff K, Wei S-J. Financial globalization and economic policies. In: Rodrik D, Rosenzweig M. (Eds.), Handbook of development economics vol. 5, Amsterdam: Elsevier; 2010, pp.4283-359. [41] Kraay A. In search of the macroeconomic effects of capital account liberalization. Unpublished manuscript, Washington, DC: World Bank; 1998. [42] Kunieda

T,

globalization

Okada change

K,

Shibata

their

A.

Finance

and

relationship?

inequality:

Macroeconomic

how

does

Dynamics,

http://dx.doi.org/10.1017/S1365100512000843, in press [43] Kunieda T, Shibata A. Endogenous growth and fluctuations in an overlapping generations economy with credit market imperfections. Asia-Pacific Journal of Accounting & Economics 2011;18(3):333-57. [44] Lane PR, Milesi-Ferretti GM. The external wealth of nations mark II: revised and extended estimates of foreign assets and liabilities, 1970-2004. Journal of International Economics 2007;73(2):223-50. [45] Leff NH. Economic development through bureaucratic corruption. American Behavioral Scientist 1964;8(3):8-14. [46] Levine R. Finance and growth: theory and evidence. In: Aghion P, Durlauf SN. (Eds.), Handbook of economic growth vol. 1A, Amsterdam; Elsevier; 2005, pp.865-934. [47] Levine R, Loayza N, Beck T. Financial intermediation and growth: causality and causes. Journal of Monetary Economics 2000;46(1):31-77. [48] Levine R, Renelt D. A sensitivity analysis of cross-country growth regressions. American Economic Review 1992;82(4):942-63.

40

[49] Loayza NV, Ranci`ere R. Financial development, financial fragility, and growth. Journal of Money, Credit, and Banking 2006;38(4):1051-76. [50] Lucas RE Jr. Why doesn’t capital flow from rich to poor countries? American Economic Review 1990;80(2):92-6. [51] Mankiw NG, Romer D, Weil D. A contribution to the empirics of economic growth. Quarterly Journal of Economics 1992;107(2):407-37 [52] Matsuyama K. Financial market globalization, symmetry-breaking, and endogenous inequality of nations. Econometrica 2004;72(3):853-84. [53] Mauro P. Corruption and growth. Quarterly Journal of Economics 1995;110(3)681-712. [54] Mauro P. The persistence of corruption and slow economic growth. IMF Staff Papers 2004;51(1):1-18. [55] Mishkin FS. Is financial globalization beneficial? Journal of Money, Credit and Banking 2007;39(2-3):259-94. [56] Mo PH. Corruption and economic growth. Journal of Comparative Economics 2001;29(1)66-79. [57] Neeman Z, Paserman MD, Simhon A. Corruption and openness. The B.E. Journal of Economic Analysis & Policy (Contributions) 2008;8(1):Article 50. [58] Obstfeld M, Taylor AM. Global capital markets: integration, crisis, and growth. Cambridge: Cambridge University Press; 2004. [59] PRS Group. International country risk guide. New York: PRS Group; 2011. [60] Quinn D. The correlates of change in international financial regulation. American Political Science Review 1997;91(3):531-51. 41

[61] Quinn DP, Toyoda AM. Does capital account liberalization lead to growth? Review of Financial Studies 2008;21(3)1403-49. [62] Rodrik D. Who needs capital-account convertibility? In: Fischer S, Cooper RN, Dornbusch R, Garber PM, Massad C, Polak JJ, Rodrik D, Tarapore SS. (Eds.), Should the IMF pursue capital-account convertibility? Princeton: International Finance Section, Department of Economics, Princeton University; 1998, pp.55-65. [63] Roodman D. A note on the theme of too many instruments. Oxford Bulletin of Economics and Statistics 2009;71(1):135-58. [64] Rousseau PL, Wachtel P. What is happening to the impact of financial deepening on economic growth? Economic Inquiry 2011;49(1):276-88. [65] Saci K, Giorgioni G, Holden K. Does financial development affect growth? Applied Economics 2009;41(13):1701-07. [66] Shleifer A, Vishny RW. Corruption. Quarterly Journal of Economics 1993;108(3):599617. [67] Stulz RM. The limits of financial globalization. Journal of Finance 2005;60(4):1595-638. [68] Wei S-J. How taxing is corruption on international investors? Review of Economics and Statistics 2000a;82(1):1-11. [69] Wei S-J. Local corruption and global capital flows. Brookings Papers on Economic Activity 2000b;2000(2):303-46. [70] Weil DN. Economic growth, second edition. Boston: Addison-Wesley; 2008. [71] Windmeijer F. A finite sample correction for the variance of linear efficient two-step GMM estimators. Journal of Econometrics 2005;126(1)25-51. 42

[72] World Bank. World development report 1997: the state in a changing world, New York: Oxford University Press; 1997. [73] World Bank. World development report 2002: building institutions for markets, New York: Oxford University Press; 2001. [74] World Bank. World development indicators, Washington, DC: World Bank; 2012. [75] Yano M. The foundation of market quality economics. Japanese Economic Review 2009;60(1)1-32. [76] Ziliak JP. Efficient estimation with panel data when instruments are predetermined: an empirical comparison of moment-condition estimators. Journal of Business & Economic Statistics 1997;15(4)419-31.

43

Fig. 1: Marginal effect of financial openness (left) and corruption (right) Notes: These figures are illustrated based on the result of column (3) in Table 1. Dashed lines indicate 95% confidence interval.

Figure 1: Marginal effect of financial openness (left) and corruption (right)

44