Remittances Foster Government Corruption Faisal Z. Ahmed∗ University of Chicago December 2010

Abstract I harness a natural experiment of oil price driven remittance flows to poor, non-oil producing Muslim countries to demonstrate that remittances empower governments, especially those in less democratic polities, to reduce their provision of public goods in favor of higher corruption. The results indicate that a one standard deviation increase in remittances raises corruption by 1.5 index points (on a 6-point scale). This is equivalent to a $600 decrease in per-capita GDP. Concomitantly, remittances enable a government to reduce its provision of welfare goods. The results suggest that political institutions may mediate the potentially beneficial socio-economic effects of remittance inflows.

∗

Post-doctoral fellow, Woodrow Wilson School, Princeton University, E-mail: [email protected]. I would like to thank Scott Ashworth, Chris Berry, Carles Boix, Ethan Bueno de Mesquita, William Howell, Nikolas Mittag, Brent Neiman, Ralph Ossa, Emily Oster, Kyung Park, Kris Ramsay, Alberto Simpser, David Singer, Duncan Snidal, Daniel Treisman, and seminar participants at Princeton and Chicago for useful suggestions.

1

1. Introduction Excessive patronage is frequently a tactic governments in many developing countries employ to remain in power. In these countries, this “misuse of government office for private gain” often paves the way for rampant government corruption (Bardhan 1997), lower economic performance (e.g., Mauro 1995; World Bank 2004) and worse social and health conditions (e.g., Gupta et al 2002).1 In light of this, existing studies frequently find that rising household income (e.g., Treisman 2000, 2007), achieved in part through the tremendous growth of remittances, may serve as a conduit for mitigating government excess and improving the quality of governance (e.g., G8 Center 2004; Obama 2009; Pfutze 2009).2 This sentiment is misguided. This paper harnesses a natural experiment to demonstrate that remittances foster government corruption in poor countries with weak democratic institutions. The natural experiment uses plausibly exogenous variation in the price of oil interacted with a Muslim country’s distance to Mecca as an instrument for remittances received in poor Muslim countries. This instrument allays major worries about endogeneity bias arising from reverse causality and non-random measurement error. The instrumental variables results demonstrate that a one standard deviation increase in remittances corresponds to a more than one standard deviation increase in government corruption. This is equivalent to a 1.5 point jump in the 6-point index of government corruption (which amounts to a $600 decrease in per-capita GDP). The mechanism through which remittances can foster mis-governance is not obvious. While scholars have long recognized that direct financial transfers to governments, such as foreign aid, can generate rent-seeking behavior and fund corruption (e.g., Friedman 1958; 1

According, to the World Bank (2004) every year corruption costs at least $1 trillion worldwide.This US$1 trillion figure is an estimate of actual bribes paid worldwide in both rich and developing countries. This figure not does include the embezzlement of public funds or theft of public assets. According to World Bank officials, it is extremely difficult to assess the extent of worldwide embezzlement of public funds, “but we do know it is a very serious issue in many settings.” For example Transparency International estimates that former Indonesian leader Suharto embezzled anywhere between $15 to $35 billion from his country, while Ferdinand Marcos in the Philippines, Mobutu in Zaire and Abacha in Nigeria may have embezzled up to $5 billion each. 2 In his survey of the cross-national “causes” of corruption, Treisman (2000, 2007) finds that higher per capita income is consistently a robust attribute of lower measures of corruption. Remittances represent transfers in household income from foreign economies to the migrant’s home country. These financial flows arise from the temporary or permanent movement of people to those foreign countries. According to the IMF, from 2001 to 2007, remittances receipts more than doubled to $336 billion. Official recorded remittances sent home by migrant workers represent the second most important source of external funding in developing countries and are about twice as large as the level of aid-related inflows to these countries (Chami et al 2008).

2

Bauer 1972; Alesina and Weder 2002), the relationship between financial transfers to households (e.g., remittances) and government corruption is not so direct. Governments do not directly “observe” these transfers since a large share of remittances are sent through backchannels and via technologies (e.g., automated teller machines) that bypass their tracking by international development agencies and predominantly poor governments. Given these problems, remittances are largely untaxed by governments (de Luna Martinez 2005; Chami et al 2008) and thus cannot directly finance corruption. Instead, given their political incentives, governments in more autocratic polities may harness remittance inflows to substitute resources from the provision of welfare goods (e.g., government transfers, public health care) to the supply of patronage (e.g., corruption).3 This “substitution effect” supports existing theories and empirical evidence that governments in autocracies divert expenditures to engage in corruption to reward key supporters and stay in power (Bueno de Mesquita et al 2003; Acemoglu and Robinson 2006). Gauging the effects of remittances on governance (as well as economic outcomes), however, will suffer from endogeneity bias related to both reverse causality (i.e., countries with more corrupt governments and inferior socio-economic conditions tend to attract higher remittances) and measurement error (i.e., officially recorded flows of remittances tend to under-report actual flows). To combat these concerns, I harness a natural experiment of oil price driven remittance flows from the Persian Gulf to construct an innovative cross-country and time-varying instrument for remittances. < Figure 1 around here > For largely cultural and religious reasons, Gulf oil producers have tended to “import” a large share of their workforce from other Muslim countries (Choucri 1986). As figure 1 shows remittances to these poor, non-oil producing Muslim countries have tracked the price of oil. The price of oil provides plausibly exogenous variation in remittances that is uncorrelated with the internal economic and political conditions in poor, remittance receiving countries. Moreover, Muslim countries closest to oil producers in the Persian Gulf tended to receive 3

Appendix B provides a parsimonious model formalizing this mechanism. There are two agents in the model, a representative household and a government. Households care about the consumption of private goods and welfare goods. Welfare goods, such as basic health care services, can be provided by either the household or the government. The government derives utility from the provision of patronage goods to its key supporters and the provision of welfare goods to everyone. The model’s equilibrium posits that remittances enable a government to reduce its provision of welfare goods in favor of higher expenditures on patronage. Moreover, this substitution effect is increasing in a government’s level of institutionalized autocracy. That is, governments in polities with weak democratic institutions will favor remittances because it permits the government to engage in greater patronage, measured in terms of higher government corruption.

3

more remittances. The inclusion of distance is key in generating a statistically strong instrument, and differentiates this study from Werker et al’s (2009) examination of oil price driven foreign aid flows.4 These two facts underlie the instrument. Specifically, I use exogenous variation in the price of oil interacted with a Muslim country’s distance from Mecca as a time-varying instrument for remittances. The instrument, therefore, identifies the average treatment effect for poor, non-oil producing Muslim, and predominantly non-democratic countries.5 For a sample of 57 poor, non-oil producing countries between 1984-2004, the instrumental variables (IV) results show that remittances raise government corruption. The results imply that a one standard deviation increase in remittances is equivalent to moving from a low corruption country like Costa Rica (with corruption on par with Germany and the United States) to a moderately corrupt nation, such as Niger or Sri Lanka. These findings are robust to outliers, alternate econometric specifications, differential trends and potential violations of the exclusion restriction. There are three plausible channels through which oil prices could affect corruption independently of remittance inflows: foreign aid, prices (inflation, exchange rate), and trade flows. The findings are robust to specifications that take these other channels into account. Finally, I provide “micro” evidence from Jordan and cross-national analysis that the link between remittances and corruption plausibly operates through the reduction of welfare goods, such as health care and social spending. This paper’s findings counter the prevailing view that remittances are a conduit for improved governance. Many studies tend to focus on the potential democratization effect of remittances in home countries, particularly in Latin America. For instance, Mexican migrants in the United States have had a sizeable impact on the domestic Mexican political process, through home town associations that provide financial assistance to their home communities. Home town associations are often involved in financing public infrastructure activities, such as the construction of roads, schools, and health facilities (Orozco and Lapointe 2003) as well as in political mobilization (de-la Garza and Hazan 2003). Such mobilization may contribute to improved public policies, governance, and the ousting of an incumbent government (Pfutze 2009).6 4

This instrumentation strategy is similar to that employed by Werker et al (2009) to gauge the macroeconomic impact of foreign aid. Werker et al use the price of oil interacted with a whether a country is Muslim to gauge the impact of foreign aid sent from the Persian Gulf on short-run macroeconomic behavior in poor, non-oil producing Muslim countries. They do not investigate how oil price driven flows of foreign money (aid, remittances) affects political behavior in recipient countries. 5 The average POLITY score for these Muslim countries is -2.7 which falls far below the well-known +6 cutoff for democracy. 6 One such mechanism for political change is through the weakening of clientistic relationships in some

4

This sentiment has surfaced at the upper echelons of public policy. Leaders of the G8 countries, for instance, have officially acknowledged that remittances promote development and committed resources to policy initiatives to attract remittance inflows (G8 Centre 2004). More recently, in a speech promoting human rights and democracy in Cuba, President Obama (2009) declared “measures that decrease dependency of the Cuban people on the Castro regime and that promote contacts between Cuban-Americans and their relatives in Cuba are means to encourage positive change in Cuba. The United States can pursue these goals by facilitating greater contact between separated family members in the United States and Cuba and increasing the flow of remittances and information to the Cuban people.” These studies and views of policymakers, however, frequently ignore the political incentives faced by public officials in countries with weak democratic institutions to engage in patronage (and out-right theft) that fosters corruption and unaccountable governance. The findings from this paper also introduce a new explanation for cross-national variation in corruption. These studies consistently find that that countries with Protestant traditions, histories of British rule, more developed economies, and (probably) higher imports are less corrupt (for an overview see Treisman 2000, 2007). Scholars have also associated cross-national differences in corruption to economic transactions involving firms engaged in international trade (e.g., Ades and Di Tella 1999) and foreign direct investment (e.g., Wei 2000) and governments receiving foreign aid (e.g., Alesina and Weder 2002). To my knowledge, there are no published works documenting whether international financial transfers to households, in the form of remittances, deteriorate the quality of governance. Thus, this is the first paper to do so. The rest of the article is structured as follows. The next section provides the conceptual framework linking remittances and regime type to patronage and corruption. Section 3 discusses the empirical strategy, data, and trends. Section 4 presents the results. Section 5 discusses various mechanisms linking remittances to corruption. Section 6 concludes. nascent democracies. Using municipal-level election data from Mexico, Pfutze demonstrates that remittances increase the probability that opposition parties win municipal level elections, thus weakening the political control of the dominant national party (PRI) and raising the overall prospect for better governance. By fostering electoral turnover and greater demands for political change, a natural implication is that remittances are likely to contribute to improved governance.

5

2. Corruption as political survival To stay in office, all governments supply welfare goods to the masses and targeted transfers in the form of patronage (e.g., Acemoglu and Robinson 2006; Bueno de Mesquita et al 2003). However, the relative distribution of welfare goods to patronage goods supplied by the government tends to differ by regime type. For instance, in their theory of democratization, Acemoglu and Robinson (2006) argue that a country democratizes as a credible commitment to future redistribution. By design, governments in democracies will therefore spend a larger fraction of their revenue on the provision of welfare goods. Similarly, Bueno de Mesquita et al’s (2003) selectorate theory demonstrates that given their political institutional constraints, democratic governments tend to provide a greater share of welfare goods than their authoritarian counterparts. Doing so better ensures the political survival of democratic governments. This tradeoff between patronage goods and welfare goods will enter the government’s utility function.7 Specifically, governments in autocracies will place greater weight on expenditures for patronage (compared to democracies) in their utility functions.8 Since governments in autocratic polities derive greater utility from the provision of patronage, they have a greater incentive to engage in practices conducive for patronage. For instance, governments in autocracies may optimally choose to reduce their provision of welfare goods as households receive remittance inflows and divert those expenditures on patronage (see Appendix B for a model that formalizes this mechanism).9 Indeed, empirically patronage is highly correlated with government corruption in autocracies. Figure 2 plots the average corruption scores by regime type.10 Corruption is measured on a 6-point scale where higher values correspond to greater corruption. Over the period 1984-2004, democracies on average tend to engage in less corruption than do autocracies. On average, democracies receive a low corruption score of around 2, whereas autocracies are 7

The government’s utility function can also be interpreted as its survival function. The government will choose the optimal bundle of patronage and welfare goods to maximize its probability of staying in power. 8 This correlation between regime type and the provision of welfare goods is evident in the data. Figure A1 shows that countries with more democratic governance (as measured with a higher POLITY index value) spend more on health care. Higher POLITY scores translate to countries with stronger institutions and behavior of democratic governance. For instance, there is greater competition in the recruitment and selection of the chief executive, greater citizen participation, and more stringent constraints on the chief executive. 9 Appendix B provides a parsimonious model linking remittances to patronage via a reduction in a government’s provision of welfare goods. Moreover, this “substitution effect” is magnified in more autocratic countries. 10 Democracies are countries that receive a POLITY score greater than +6. Thus, autocracies are countries are countries that receive a score less than or equal to 6 (i.e., on the range -10 to +6).

6

at least a full point higher. Moreover, for the sample of poor remittance recieving countries examined in this paper, more corrupt governments tend to last longer in office.11 For instance, governments in the most corrupt countries (with the maximal value of 6) stay in power for almost 15 years. In contrast, governments in the least corrupt polities (with corruption scores equal to 0 or 1) last around 3.6 years. < Figure 2 around here > For a variety of reasons, governments in autocracies are more likely to engage in corruption. According to the selectorate model of political survival (Bueno de Mesquita et al 2003), autocracies are likely to foster corruption for at least three reasons. First, eliminating corruption and encouraging institutions that promote the rule of law are public goods. Leaders in autocracies have few incentives to find and eliminate corruption. Second, leaders can provide benefits by granting the supporters the right to expropriate resources from themselves. Thus, autocratic leaders might encourage corrupt practices as a reward mechanism. Third, the prevalence of kleptocracy in autocracies frequently allows leaders to siphon off resources for pet projects. “Ruling to steal” constitutes a form of corruption. These reasons support the model’s prediction that governments in autocratic political systems are likely to harness remittances to engage in corruption.

3. Empirical strategy 3.1. Baseline specification To evaluate the impact of remittances on government corruption, I estimate variants of the following regression specification: CORRU P T IONi,t = α0 + α1 Ri,t + α2 Ai,t + α3 (Ri,t × Ai,t ) + α4 Xi,t + α5 Yt + α6 Di + i,t where the dependent variable measures government corruption for each country i in year t. This variable ranges on 0 to 6 scale, where higher values correspond to greater corruption. Ri,t is each country’s officially recorded inflows of remittances (% GDP) and Ai,t measures each country’s level of institutionalized autocracy. Since governments in more autocratic polities are incentivized to engage in corruption, α3 is expected to be positive. Xi,t is a set of time-varying (e.g., log GDP per capita, log population) and time-invariant (e.g., legal origin) 11

Figure A3 presents a bar chart describing the tenure of governments by corruption level.

7

variables, Yt is a year trend, and Di are dummies for each country.12 The inclusion of country fixed effects will account for observable and unobservable time-invariant country specific characteristics that may explain corruption. These observable characteristics include legal origin, colonial heritage, religious tradition, ethnic composition, geography (e.g., proximity, whether the country is a natural resource exporter). Consistent with existing cross-national studies of corruption (e.g., Alesina and Weder 2002), I estimate the baseline specification with OLS and conservatively cluster the standard errors by government.

3.2. Endogeneity Attempts to gauge the causal impact of remittances on governance will suffer from endogeneity stemming from reverse causality and potential measurement error. The direction and magnitude of this bias, however, is likely to be influenced by the relative and potentially offsetting effects of reverse causality and measurement error. With respect to reverse causality, if a country’s underlying degree of government corruption is positively correlated with its receipts of remittances, this will tend to upward bias the effect of remittances on government corruption. The decision to migrate and remit earnings are often driven by a dearth of economic opportunities in the home country, which often tends to be correlated with the country’s quality of governance. Indeed, for a sample of poor non-oil producing countries, there is a positive correlation between corruption and remittance inflows. For instance, countries with a “low” corruption score (0 to 2) receive remittances equal to 2 percent of GDP compared to remittances equal to 2.5 percent of GDP for countries with a “moderate” corruption score of 3 and 4. “High” corruption countries with a corruption index score of 5 and 6, on average, receive remittances equal to 4.9 percent of GDP. The second form of endogeneity relates to non-random measurement error. As already noted, officially recorded inflows of remittances tend to under-report actual flows. Moreover, the mis-measurement of remittances does not seem to be random: poorer countries, presumably with governments that have lower tracking capacities, are more prone to mis-measure remittances inflows (de la Martinez 2005). This fact is well acknowledged by practitioners and policymakers. For example, improving the measurement of remittances is a major concern and stated goal of development agencies and governments in remittance receiving and as well as sending countries (G8 centre 2004). From an econometric standpoint, the preva12

In models that include country dummies, the time-invariant measures of legal origin, colonial heritage, religious tradition, and ethnic composition are omitted to avoid multicollinearity.

8

lence of under-reporting and the existence of systematic measurement error will tend to attenuate the coefficient estimate of remittances on economic and political outcomes. Thus, the existence of non-random measurement error will tend to downward bias the coefficient estimates.

3.3. Natural experiment One strategy to mitigate this endogeneity problem is to identify an instrument for remittances. I employ a natural experiment linking fluctuations in the world price of oil and a country’s distance to Mecca to construct an instrument for remittances sent from Gulf oil producing countries to poor, non-oil Muslim producing countries.13 In the aftermath of the 1973 oil crisis, labor from different countries in North Africa, South Asia, and Middle East, migrated in great numbers to the oil-exporting countries in the Middle East. The first wave of workers (totaling about 500,000) migrated from non-oil producing Gulf states, such as Jordan, Palestine, and Yemen.14 In the latter part of the decade, Gulf states began to recruit a large number of South Asian workers from India, Pakistan, and Bangladesh. For example, it is estimated that the number of Pakistani workers jumped from roughly 500,000 in 1975 to over 1.25 million in 1979. By the early 1980s, there may have been some 3.5 million to 4.65 million migrants, in a combined labor force of 9 million to 10.2 million workers (Choucri 1986). This large movement of labor generated large capital flows in the form of worker remittances from Gulf oil producers to a variety of non-oil producing labor exporting countries, such as Jordan, Mali, and Pakistan.15 Two stylized facts make this an interesting natural experiment. First, the amount of aggregate remittances received by poor, non-oil producing Muslim countries tracks the world price of oil. As figure 1 shows, as the price of oil began to rise in 1974, remittance inflows to 13

Recently, “experimental” settings have been used to study the determinants of corruption. For instance, Fisman and Miguel (2007) analyze the parking behavior of United Nations officials (in response to a change in the law removing diplomatic immunity from parking tickets) to study the effect of corruption norms. The authors find a strong effect of corruption norms, as diplomats from high-corruption countries (on the basis of exiting survey-based indices) accumulated more unpaid parking violations after the implementation of the legal change. In another study, Olken (2007) presents evidence from a randomized experiment from Indonesian villages finding that monitoring (via government audits of projects) can play an important role in reducing corruption, even in a highly corrupt environment. 14 As Choucri (1986) observes, the trends indicate that the magnitude of migration was much greater than indicated by reports based on data collected in 1975 by the World Bank and the International Labor Office. 15 Figure 2 in Appendix A plots aggregate inflows of remittances (% GDP) for Jordan, Mali, and Pakistan. There is substantial variation both in aggregate levels and across time for these three poor, non-oil producing countries.

9

poor non-oil producing Muslim countries rose sharply.16 This level of remittance remained high through the early 1980s and then began to fall as the price of oil tanked. Since the 1990s remittance flows have tended to be less volatile but still tend to co-move with the price of oil. As the world price of oil is largely determined by supply decisions in oil producers and demand conditions in large (industrialized and rich) economies, the price of oil provides a plausibly exogenous source of variation in remittance flows that is unrelated to the economic, political, and social conditions in remittance-receiving countries. The second stylized fact is that remittances inflows to non-oil producing Muslim countries is inversely related to each country’s distance from the Persian Gulf. Countries closer to oil producing Gulf economies experienced greater outward migration and subsequently higher remittance inflows. Table 1 shows that non-oil producing Muslim countries closer to Mecca tended to receive higher remittances (as a share of GDP).17 For instance, over the sample period Jordan (765 miles from Mecca) on average received remittances equal to 17.7 percent of GDP. In contrast, Bangladesh (3212 miles from Mecca) on average received remittances equal to 3 percent of GDP. < Table 1 around here > These two stylized facts underly the construction of the instrument. Specifically, I interact the price of oil with a Muslim country’s distance from Mecca as an instrument for remittances. This instrument varies across both time (i.e., annual fluctuations in oil prices) and across countries (e.g., distance to Mecca differs across countries). Thus, the instrument will identify within and across country variation in corruption. This instrument improves upon existing ones which are predominantly time-invariant and limited to explaining crosssectional variation in corruption, such as colonial settler mortality rates (Treisman 2000).18 This instrumentation strategy is similar to that employed by Werker et al (2009) to gauge the impact of oil price driven foreign aid flows on macroeconomic outcomes (e.g., growth, consumption, investment, inflation). This paper’s instrument differs from Werker et al on two key dimensions. First, it explains remittance flows rather than foreign aid flows. Second, the instrument requires a country’s distance from the Persian to precisely measure 16

The Y-axis on the left scale measures the total remittances as a share of total GDP for a set of 19 non-oil producing Muslim countries 17 The table also provides evidence that these countries tended to be autocratic-leaning since their POLITY scores are well below +6. 18 For instance, per-capita income and corruption tend to endogenous. To tackle this, Treisman (2000) uses settler mortality at time of first colonization to instrument for income. This instrument can help explain the effect of income on corruption at one point in time, but is limited in its ability to explain within-country changes in corruption.

10

remittance flows. As the results will show, variations in oil price alone are insufficient to explain remittance inflows to poor non-oil Muslim countries (i.e., the instrument is extremely “weak”). Armed with this instrument, the reduced form two-stage regression setup is: First Stage: REM ITit = α + βDISTi × p(oil)t + γXit + δYt + κDi + it Second Stage: CORRU P T IONit = a + b × REM ITit + c × Xit + d × Yt + f × Di + uit where the dependent variable is government corruption for each country i in year t, REM ITi,t is each country’s officially recorded inflows of remittances (% GDP). DISTi is the distance of a poor, non-oil producing Muslim country from Mecca (measured in 1,000 of miles). A country is defined as Muslim if at least 70% of its population identifies with the Islamic faith. Xi,t is a set of time-varying (e.g., log GDP per capita, log population), Yt is a year trend, and Di are dummies for each country.19 Both stages are estimated via OLS and the standard errors are conservatively clustered by government. In the second stage regression, the coefficient on remittances will measure the average treatment effect for a group of poor, non-oil producing Muslim countries that tend to have autocratic-leaning politics. The average POLITY score for the treatment group over the sample period is -2.7, which falls far below the standard “+6” threshold of democratic governance. Within the treatment group, these countries exhibit variation in the quality of governance. For instance, Jordan is a monarchy with an average POLITY score of -4 over the period 1984-2004.20 In contrast, Bangladesh and Turkey have fluctuated between episodes of autocracy and weak democracy. Bangladesh swings from a POLITY score of -7 from 1975-1985 (roughly) to 6 since 1991, while Turkey moves between a low POLITY score of -5 to a high of 9.

3.4. Data Measuring “patronage” corruption Corruption is usually understood to mean the “misuse of public office for private gain,” where the “private gain” may accrue either to the individual public official or to groups or parties 19

Some specifications may use continent fixed effects (instead of country dummies) to mitigate problems with incidental parameters prevalent in estimating non-linear models, such as probit. 20 Prior to 1988, Jordan’s POLITY score was -9. From 1988 to 1992 it increased to -4. Since 1992, the POLITY score has been -2.

11

to which be belongs, such as her political party or governing coalition (Bardhan 1997). This definition is quite broad and can capture various forms of corruption, such as payments to public officials (bribery) and transfers from the government to key groups (patronage). To measure government corruption in the form of patronage, I use the International Country Risk Guide (ICRG) corruption index. While it is extremely difficult to directly observe and assess the extent of worldwide embezzlement and misuse of public funds, officials at the World Bank (2004) “do know that it is a very serious issue in many settings.” Fortunately the ICRG corruption index, unlike other measures of corruption, explicitly takes into account government patronage. The ICRG’s official documentation makes clear that its corruption measure “is more concerned with actual or potential corruption in the form of excessive patronage, nepotism, job reservations, ‘favor-for-favors’, secret party funding, and suspiciously close ties between politics and business.” To the extent that government patronage cannot be reliably quantified for a wide set of developing countries, the ICRG corruption index does capture the tradeoff between government patronage and the provision of welfare goods. For instance, the corruption index is negatively correlated with government expenditures on public health care, education, and social contributions.21 The published ICRG corruption index ranges from a score of 0 (high corruption) to 6 (low corruption) and has been released on a monthly basis since 1984 for around 140 countries. This index also spans the longest period (from 1984) and is the most prevalent measure of corruption in cross-country panel analysis (e.g., Alesina and Weder 2002; Fisman and Gatti 2002; Treisman 2007). I calculate annual averages of these corruption scores and re-scale the index such that higher values correspond to higher corruption (i.e., a score of 6 implies highest corruption and 0 implies lowest corruption). Independent variables The key independent variable, workers’ remittances (% GDP), is from the World Bank’s World Development Indicators (WDI). It measures officially recorded flows of remittances and will tend to understate actual remittances since a large share of these capital flows go through back-channels and/or are difficult to monitor by poor governments (de Luna 21

Across all countries (as well as the 57 poor, non-oil producing countries in this study) the ICRG corruption index exhibits negative correlations with public health expenditure (%GDP), total public spending on education (%GDP), and social contributions (% government revenue) of -0.53, -0.32, and -0.34 respectively. Other scholars (Treisman 2000; Alesina and Weder 2002) have documented that the ICRG corruption measure is highly correlated with corruption measures from alternate sources (e.g., Transparency International) and is correlated with factors indicative of mis-governance such as public health expenditures.

12

Martinez 2005). Variables measuring economic, demographic, country-specific factors (e.g., colonial heritage), and the quality of political institutions are also likely to affect government corruption. The economic (real GDP per capita, real GDP per capita growth) and demographic (log population) variables are from the WDI. Measures of legal origin, colonial heritage, religious tradition, and ethnic-linguistic fractionalization are drawn from Treisman (2000). To gauge a country’s underlying type and quality of governance, I invert the conventional POLITY score of democratic governance (Marshall and Jaggers 2006) to create a continuous and time-varying measure of a country’s level of institutionalized autocracy.22 This institutionalized autocracy score lies on a [0,1] interval, where values closer to 1 correspond to more autocratic governance. Tables 2 and 3 summarize these variables for the estimating sample. Table 2 describes the variation in average corruption score, quality of democratic governance, and remittance inflows for each country over the sample period, 1984-2004. Table 3 provides additional summary statistics for all the variables. < Table 2 around here > < Table 3 around here >

3.5. Trends Government corruption exhibits wide cross-national and temporal variation. Table 4 shows that in any given year, governments may become more or less corrupt around 14 percent of the time from the previous year. Most of the time, corruption increases or decreases by 1 index point from the previous year, although some countries do undergo a 2 point annual change in corruption. This pattern is similar across both the control group of non-Muslim countries and the treatment group of Muslim countries (columns 2 and 3). < Table 4 around here > 22

The POLITY score measures the manner in which a country’s chief executive is “recruited”, the “constraints” to her power as well as the nature in which individuals can participate in the process of choosing a leader. Empirically, the POLITY score is frequently use as a measure of both formal institutions (e.g., constitution) and political behavior institutions. The POLITY score ranges on a 21 point index (-10 to +10), where scores closer to +10 correspond to a higher quality of democratic governance. The convention is to categorize country’s with scores greater than +6 as democratic and those with scores below that threshold as non-democratic. I add 11 to the index so that it lies on a 1 to 21 point scale, where values closer to +21 imply a higher quality of democratic governance. I invert this transformed index so that it lies on a [0,1] where scores closer to 1 correspond to more authoritarian governance.

13

There is additional evidence that annual movements in corruption do not differ across Muslim and non-Muslim countries. Figure 3 plots the percentage of Muslim and non-Muslim countries that undergo change in corruption index from the previous year. In some years, very few countries experience a change in their corruption index (e.g., 1995). While in other years, a larger share of countries experience changes (e.g., 1992, 1997). This temporal variation should not hide the fact that in any given year, around 15 percent of Muslim and non-Muslim countries undergo a change in government corruption.23 In general, the two series co-move in the same direction with a correlation of 0.60. To the extent this captures global trends (e.g., end of Cold War, Washington Consensus emphasizing good governance), these trends tend to affect government corruption in poor Muslim and non-Muslim countries in similar ways. < Figure 3 around here >

4. Results 4.1. Baseline results Table 5 presents the baseline OLS results. The model in column 1 regresses corruption on annual receipts of remittances (% GDP) and a standard set of covariates existing studies identify to explain cross-national variation in corruption. Remittances exhibit a positive and highly statistically significant effect on corruption. A two standard deviation increase in remittances will raise the corruption index by over 0.30 points. The effect of the control variables on corruption tend to collaborate existing studies (coefficients not reported). Former British colonies, countries with larger Protestant populations, more ethnically homogenous societies, and wealthier countries (measured with log GDP per capita) exhibit a negative (and often statistically significant) effect on corruption. Many of these control variables (e.g., colonial legacy) are time-invariant and can be subsumed with country fixed effects. In a fixed effects model that regresses corruption on remittances and a set of time varying covariates (column 2), remittances exhibits no effect on corruption. However, in a model that introduces the interaction of remittances and a country’s level of institutionalized autocracy (column 3), the interaction term exhibits a positive (coefficient=0.13) and statistically significant effect on corruption. < Table 5 around here > 23

The sample means for the Muslim and non-Muslim series in figure 2 are 15.2% and 15.9% respectively.

14

The positive coefficient on the interaction term in table 5 is unlikely to be driven by a country’s level of institutionalized autocracy. First, countries have not become more autocratic over time. Over the sample period the proportion of authoritarian countries in the estimating sample has steadily fallen and the average quality of democratic governance has risen.24 Second, the total marginal effect of remittances on corruption is greater than zero in column 3. Thus, it is the inflow of remittances that may be sustaining corrupt practices despite the apparent global trend of improved governance.

4.2. Instrumental variable results First stage Table 6 reports the results of the first stage regression describing the interactive effect of oil prices with a Muslim country’s distance from Mecca on remittances inflows. The regression includes a year trend and country fixed effects and explains nearly 80 percent of the variation in remittance inflows. In column 1, the coefficient on the instrument is 0.026 and is significant at the 1 percent level. For the typical poor, non-oil producing Muslim country this coefficient implies that a 10 dollar increase in the world price of oil raises inflows of remittances equal to 0.6 percent of GDP.25 Between 1984-2004, oil prices ranged from $15 to $49 per barrel. Such a movement in oil prices represents a shock to remittances equal to about 1.9% of GDP. < Table 6 around here > The control variables have the expected signs. More autocratic countries attract a higher share of remittances. Richer and more populated countries tend to receive less remittances. Growth exhibits a positive and statistically effect on remittances, which is surprising since remittances tend to be counter-cyclical. The F-test on the instrument yields a value of 11.9 which exceeds the conservative threshold of weak instruments of 9.6 suggested by Stock et al (2002). Figure 1 suggests that remittances lag oil prices (by about a year). The effect of one a year lag in oil prices interacted with a Muslim’s country from Mecca generates an identical impact on remittances (column 2). The interaction of oil price and whether a country is 24

See figure A4. In 1984, around 80 percent of the sample was authoritarian (i.e., POLITY score below +6). By the early 2000s, the proportion had more than halved to around 33 percent. This has been accompanied by a steady rise in the average quality of democratic governance (as measured with the POLITY score). Over the sample period, the average POLITY score has nearly tripled from an average value of 4.5 in 1984 to around 13 by 2004. 25 The typical Muslim country is about 2,200 miles from Mecca. For the treatment group, on average, remittances inflows total about 4 percent of GDP.

15

Muslim is not a strong predictor of remittances (column 3), as the F-stat on this particular instrument is extremely low. This indicates that a country’s distance to the Persian Gulf is required to improve the precision of the instrument. Finally, the instrument is also a strong predictor of real remittances per capita (column 4). Remittances and corruption With this strong instrument, I evaluate the impact of instrumented remittances on government corruption (table 7). Column 1 reports the effect of remittances on corruption from an OLS regression restricted to the countries in the treatment group. This serves as a benchmark to compare the instrumental variables result. In this OLS model, the coefficient on remittances is small (around 0.04). For the same restricted sample, the coefficient on the interaction of remittances with a country’s institutionalized autocracy score is around 0.06 (column 2). Column 3 reports a much larger coefficient (0.3) and statistically significant effect (p-value=0.06) of instrumented remittances on government corruption. The estimated effect on corruption is similar with remittances inflows from the previous year (column 4). These IV estimates imply that a three percentage point increase in remittances raises government corruption by almost one full index point. For the treatment group with average remittances inflows equal to 6 percent of GDP, this implies that remittances raise government corruption by around 1.8 index points. This coefficient estimate is substantive: a one standard deviation increase in remittances raises corruption more than one standard deviation (equivalent to 1.5 index points). In the estimating sample, this is equivalent to moving from a low corruption country like Costa Rica (which has average corruption on par with Germany and the United States) to a moderately corrupt country like Niger or Sri Lanka.26 This increase in corruption represents a substantial welfare loss. According to estimates by Dreher and Herzfeld (2008), who estimate the impact of corruption on growth, a 1.5 index point increase in the ICRG corruption index translates to around a $600 decrease in per-capita GDP. The IV coefficient estimates are larger than the OLS estimates, suggesting that they correct for the attenuation bias (attributable to measurement error) in the OLS models. This implies that the baseline OLS results are downward biased and should be viewed as a lower bound for the effect of remittances on corruption in autocracies (table 5, column 3). Finally, since the IV estimates represent the average treatment effect for a group of largely 26

Costa Rica, Germany, and the United States have an average corruption score around 1 over the sample period. Niger and Sri Lanka have an average corruption score around 2.5.

16

autocratic-leaning countries, these results provide additional evidence that remittances fosters government corruption in countries with authoritarian politics. Existing theories positing that corruption is a conscious government policy in autocracies predict large effects for the interaction of remittances with autocratic governance. Instrumenting directly for this interaction term (column 5) generates a large positive and highly significant effect. This coefficient estimate captures the heterogeneous effect of remittances in autocracies on government corruption. Moreover in specifications that also control for each constitutive part of the interaction term, the instrumented interaction term continues to exhibit a positive and highly significant effect on government corruption (results not reported). Finally, instrumenting for alternate measures of remittances of yields similar results. For instance, instrumenting for log remittances (column 5) raises government corruption.27 The coefficient estimate is positive (0.14) and highly significant (p-value=0.03). A one standard deviation increase in log remittances (equal to 10.8) raises government corruption by 1.5 index points. Remittances measured in terms of per capita (column 7) and log per capita (column 8) also raise government corruption. < Table 7 around here > Remittances and government patronage The ability of a government to engage in corruption hinges on its capacity to engage in patronage. While it is difficult to objectively observe government patronage across developing countries, patronage is likely to be highly correlated with a government’s compensation of employees. In autocratic regimes, a large portion of these workers are likely to be within the government’s “inner circle.” Thus, higher government employee compensation provides an objective and observable measure of government patronage across non-democratic countries. The estimated effects of remittances on government compensation are reported in table 8, columns 1-3.28 Contemporaneous and one-year lagged remittance inflows (% GDP) as well as remittances per capita raise government employee compensation (% government expenditures). The estimated effects in columns 1 and 2 imply that a one percentage point increase in remittances (% GDP) shifts government expenditures toward employee compensation by about 3.6 percentage points. Since employee compensation is measured as a share of total 27

This is not the preferred measure of remittances as the first stage regression is weak, with a F-stat equal to 6). 28 These specifications control for population, per capita GDP growth, and average per capita income. The sample size is smaller (315 observations) as data on government compensation is available for a subset of countries from 1990 onwards.

17

government expenditures this represents a reduction in other government outlays, which is consistent with a substitution effect. < Table 8 around here > The ability of a government to allocate more resources towards patronage and ultimately corruption, allows it to expand its political authority to the potential detriment of a country’s aggregate socio-economic and political welfare. To gauge the effect of remittances on a government’s political authority, I use the executive constraints index from the POLITY data set. The executive constraints index measures the extent of institutionalized constraints on the decision-making powers of the chief executives, whether they are individuals or collectives. This index has been used to “unbundle” cross-country differences in the quality of governance and property rights (Acemoglu and Johnson 2005) and according to Gledistich and Ward (1997) is the most important feature differentiating autocracies from democracies.29 The executive constraints index lies on a 7-point scale where a lower value corresponds to greater executive control.30 The results in table 8, columns 4-6 provide evidence that remittances expand a government’s political authority. In these specifications remittances exhibit a robust negative effect on the government’s political constraints. A one standard deviation increase in remittances (% GDP) lowers the executive constraints index by 1.5 points. This downward movement in the constraints index corresponds to less secure property rights and an environment conducive to lower long-run economic growth, investment, and financial development (Acemoglu and Johnson 2005).

4.3. Sensitivity analysis Specification checks The main finding that remittances fosters government corruption is robust to a number of specification concerns (table 9). Residual plots for the IV estimates do exhibit a weak correlation between high remittance inflows (>15% GDP) and the regression residuals.31 For instance, certain countries in the treatment group receive large amounts of oil price driven remittance inflows (e.g., between 1975-2004, annual remittances into Jordan averaged 29

Gledistich and Ward claim that “although the degree of executive constraints accounts for only 4 of the possible 10 democracy scale points, all our analyses point strongly to the conclusion that this variable virtually determines the democracy and autocracy scale values” (380). 30 The executive constraints index takes 7 values: 1=unlimited authority, 2=interediate category between 1 and 3, 3=slight to moderate limitations, 4=intermediate category between 3 and 5, 5=substantial limitations, 6=intermediate category between 5 and 7, and 7=executive parity of subordination. 31 See figure A5.

18

17.7% of GDP). Instrumented remittances continue to exhibit a positive and significant effect on government corruption in a model that drops observations from Jordan (column 1), as well as other individual countries in the treatment group (not reported). Of course outliers in the control group of non-oil producing, non-Muslim countries could also influence the findings. The finding that remittances raises government corruption is robust to excluding observations from all countries with remittance inflows exceeding 15% of GDP (column 2). Another concern is the potential endogeneity of the control variables with remittance inflows, which could introduce a form of selection bias in the baseline specification. For instance, remittances represent additional income to households which will raise per capita GDP and per capita GDP growth. Higher average income, independently, tends to lower corruption (Treisman 2000, 2007). Controlling for these variables may therefore bias the estimated effect of remittances on corruption. I address these concerns in two ways. First, I gauge the effect of remittances in a model without any of the control variables (column 3). Second, I re-estimate the baseline specification with one year lags of the control variables (column 4). As remittances received in year t will not affect the value of these controls in year t-1, this specification addresses the concern that remittances are contemporaneously correlated with values of the control variables. In both models, remittances exhibit a positive and significant (p-value An additional worry relates to the linear specification of the ordinal measure of corruption. To check that the underlying method of estimation is not driving the results, I estimate a two stage probit model. The first stage is estimated via OLS and the second stage uses probit.32 Instead of using the 6-point measure of government corruption as the dependent variable, I construct a “high corruption” indicator variable that is equal to 1 if a country receives a corruption score of 5 or 6, and zero otherwise. In this model, instrumented remittances exhibit a strong positive and highly significant (p-value

5. Mechanisms 5.1. Evidence of a substitution effect The empirical analysis thus far demonstrates the reduced form relationship between remittances and government corruption. Analysis of expenditure side data from Jordan and cross-national evidence seems to support the claim that remittances empower governments to fund patronage via a reduction in its provision of welfare goods. Evidence from Jordan Ever since the Kingdom’s inception, the Heshemite rulers of Jordan have remained in power by constructing a series of patronage institutions - usually at the expense of economic development - to hold together a highly disparate coalition of business elites and Transjordian tribes (Brynen 1992; Peters and Moore 2009). While foreign aid has been a staple source of external assistance to Jordan’s monarchy since the collapse of the Ottoman Empire, increasingly “indirect” external rents in form of remittances inflows have enabled the government to finance policies to win the support of coalition members. As Peters and Moore note, “authoritarian regimes adapt as different sources of external rent decrease or increase, seeking out new sources of external rent and devising new ways to deliver it to coalition members” (258). Indeed through periods of abrupt demographic changes and intense political violence (both domestically and regionally), Peters and Moore (2009) argue that Jordan’s “monar22

chy, in concert with geopolitically motivated donors, has met these demands by modifying old distributional mechanisms and institutionalizing new venues to take advantage of the international system’s provision of economic rents” (257). One such observable channel is through the reduction of welfare provision and the expansion of government compensation to coalition members (many of whom work for the government) in response to remittance inflows. Table 12 presents evidence that higher inflows of remittances may permit a government to shift its expenditures on welfare payments to government patronage. Between 1990-1994, remittances to Jordan averaged 15 percent of GDP. Over this period government employee compensation and welfare payments comprised 63 and 13 percent of the government’s expenditures respectively. Over the next decade, remittances rose by 5 percentage points to around 20 percent of GDP. Over this period, the government’s share of expenditures on patronage rose by 5 percentage points while welfare transfers declined by almost 3 percentage points. This shift in budgeting priorities towards patronage concomitantly raised Jordan’s corruption score by one index point between 1990 to 2004. < Table 12 around here > Cross-national evidence The re-allocation of expenditures from welfare payments to increased government patronage in response to higher remittance inflows is not unique to Jordan, nor its high level of aggregate remittance inflows. Small increases in remittances can shift the allocation of government expenditures to patronage.34 For instance, in countries that receive remittances less than 2 percent of GDP, government’s on average allocate 27 and 38 percent of their budget to employee compensation and government transfers respectively. As remittance inflows rise, governments tend to allocate a greater share of their budget to employee compensation. In countries that receive moderate inflows of remittances (between 2 to 4 percent of GDP), for instance, governments allocate 30 percent of their expenditures on employee compensation and 26 percent to government transfers. In countries that receive inflows of remittances exceeding 4 percent of GDP, around 33 percent of government expenditures are spent on patronage and 31 percent is transferred to the population. This re-allocation of government resources to patronage as remittance inflows rise is robust to the inclusion of variables that capture a country’s economic growth, average income, population, and underlying degree of autocratic governance. In a 2SLS specification that 34

Figure A6 charts the allocation of government expenditures on patronage and welfare goods provision at low (less than 2 % GDP), moderate (2-4 % GDP) and high (greater than 4%) levels of remittance inflows.

23

controls for these effects, a one percentage point in remittances reduces the share of expenditures a government allocates to subsidies and transfers by 4.5 percentage point (table 13, column 1). This is consistent with the earlier finding that remittances raise government expenditures on employee compensation (patronage). < Table 13 around here > The ability of governments to reduce expenditures on public goods in response to remittances has tangible harmful effects on the population. For example, remittances reduce childhood immunizations to measles in the population (column 2). This type of health service represents a welfare good a government or household provides on a regular basis (and the class of substitutable welfare goods envisioned in the formal model in Appendix B). The IV coefficient suggests that a 1 percentage point increase in remittances (% GDP) reduces the percentage of infants immunized to measles by 5.7 percentage points. Remittances also tend to shift expenditures on health care between households and the government. Remittance inflows exhibit a negative effect public health care expenditures (column 3), but tend have a positive effect on private health care expenditures (column 4). While these estimated effect on remittances on public and private health care expenditures are not statistically significant at conventional levels, the direction of the effects are informative and provide additional evidence that remittances may re-orient a government’s willingness to spend funds on welfare goods. Together, the results in table 13 suggest that governments may reduce the provision of welfare goods in response to remittance inflows.

5.2. Discounting other mechanisms Direct rent extraction Given the incentives of governments in autocracies to extract resources for patronage, governments may opt to directly extract rents from rising household wealth. That is, rather than diverting expenditures from the provision of welfare goods to patronage, governments in autocracies may choose to directly extract rents from households. This alternate mechanism is not borne out in the data. For instance, suppose a government could directly “observe” remittances inflows and extract (tax) some of this additional income from households. This would imply that remittances exhibit a positive effect on government tax revenue. Such an effect is not empirically identified, as remittances exhibit a negative relationship with various measures of tax revenue (table 14, columns 1-3). Of course these results may just reflect the inadequate and inefficient tax collection systems prevalent in many developing 24

countries. Thus, an alternate way for governments to extract rents from rising household income is from consumption based taxation.35 The expected positive effect of remittances on consumption taxes is not evident. Remittances tend to lower the share of tax revenues collected from goods and services (column 4). < Table 14 around here > Political discontent Remittances may foster corruption by affecting internal political discontent. There are two plausible channels through which this could happen. The first is that remittance inflows necessarily require outward migration of citizens, some of whom may be dissatisfied with the incumbent government. Remittances may therefore lower internal political dissent and permit the government to engage in greater corruption. The second channel is that remittances inflows raise household income, which might lower the opportunity cost to rebel (e.g., Collier and Hoeffler 2004). If this is the case, then governments may respond to rising discontent by increasing its provision of patronage (corruption) to keep its governing coalition intact. I evaluate these competing effects in two ways. First, I assess whether remittances affect political dissatisfaction by evaluating its effect on the number of anti-government demonstrations. Second, I gauge the impact of remittances on corruption while controlling for this internal dissent.36 Remittances exhibits a weak negative but statistically insignificant effect on the number of anti-government demonstrations (column 5). Remittances also exhibit a negative relationship with other measures of political discontent, such as the number of riots and strikes (results not reported). Moreover, accounting for the number of anti-government demonstrations does not diminish remittances affect on raising government corruption (column 6). Consistent with the baseline estimates, a three percentage point increase in remittances raises government corruption by over one index point.

6. Conclusion Given their political institutional setting, governments in more autocratic-leaning countries frequently engage in corruption as means to stay in power. Indeed, existing studies find that 35

This channel is plausible as the extant literature documents that a large fraction of remittances are consumed on goods and services (e.g., Chami et al 2008). 36 To measure internal dissent directed exclusively at the government and its policies, I use the antigovernment demonstration variable from Banks (2004). This variable counts the number of public gatherings (of at least 100 people) for the primary purpose of displaying or voicing their opposition to government policies or authority, excluding demonstrations of a distinctly anti-foreign nature. The variable has wide coverage and extends back to the early 1960s for many developing countries.

25

governments that receive unearned government revenue through such channels as natural resource rents (e.g., extraction of oil) and foreign aid tend to engage in greater patronage which is conducive to corruption and broader political institutional decay. Thus, the prevalence of government corruption in many developing countries may reflect the conscience decision of these governments to engage in these practices as means of remaining in power. Building off this logic, I model a process by which shocks to unearned household income in the form of remittances empower a government to engage in government corruption by substituting funds from the provision of welfare goods to finance government patronage. This argument counters some existing case studies and the views of many prominent policymakers in developed countries that remittances can engender good governance. The argument linking remittances to government corruption is substantiated with robust cross-country empirical evidence. To allay concerns related to endogeneity bias, I harness a natural experiment of oil-price driven remittance inflows to construct an innovative crosscountry and time-varying instrument for remittances. The instrumental variables results demonstrate that remittances raise government corruption (via a reduction in the provision of welfare goods). These findings are robust to a battery of specification checks, considerations of differential trends, and potential violations to the exclusion restriction. Moreover, the most likely mechanism through which remittances foster corruption is via a shift in government expenditures from the provision of welfare goods to the supply of patronage. Competing explanations are ruled out. Together, these findings demonstrate that the quality of political institutions mediates whether higher household income necessarily paves the way to improvements in governance.

26

Bibliography Acemoglu, Daron and Simon Johnson. 2005. “Unbundling Institutions.” Journal of Political Economy, 113(5): 949-996. Acemoglu, Daron and James Robinson. 2006. The Economic Origins of Dictatorship and Democracy. Cambridge University Press: Cambridge, MA. Ades, Alberto and Rafael Di Tella. 1999. “Rents, competition, and corruption.” American Economic Review, 89(4): 982-893. Alesina, Alberto and Alexander Wagner. 2006. “Choosing (and Reneging On) Exchange Rate Regimes.” Journal of the European Economic Association, 4: 770-799. Alesina, Alberto and Beatrice Weder. 2002. “Do corrupt governments receive less foreign aid?” American Economic Review 92(4): 1126-1137. Banks, Arthur. 2004. Cross-National Time Series Data Archive [Computer file]. Jerusalem: Databanks. Bardhan, Pranab. 1997. “Corruption and development: A review of issues.” Journal of Economic Literature, 35(3): 1320-1346. Bauer, Peter T. 1972. Dissent on Development: Studies and Debates in Development Economics, Cambridge MA: Harvard University Press. Bueno de Mesquita, Bruce, Alastair Smith, Randolph M. Siverson, and James D. Morrow. 2003. The Logic of Political Survival, Cambridge MA: MIT Press. Brynen, Rex. 1992. “Economic Crisis and Post-Rentier Democratization in the Arab World: The Case of Jordan.” Canadian Journal of Political Science, 1: 69-97. March. Chami, Ralph, Adolfo Barajas, Thomas Cosimano, Connel Fullenkamp, Michael Gapen, and Peter Montiel. 2008. The Macroeconomic Consequences of Remittances. International Monetary Fund: Washington D.C. Choucri, Nazli. 1986. “The Hidden Economy: A New View of Remittances in the Arab World.” World Development, 14(6): 697-712. De-La Garza, Rodolfo and Myriam Hazan. 2003. “Looking Backward, Moving Forward: Mexican Organizations in the US as Agents of Incorporation and Dissociation.” The Tomas Rivera Policy Institute. de Luna Martinez, Jose. 2005. “Workers’ remittances to Developing Countries: A Survey of Central Banks on Selected Public Policy Issues.” Policy Research Working Paper, No. 3638, World Bank: Washington D.C. Dreher, Axel and Thomas Herzfeld. 2008. “The Economic Costs of Corruption: A Survey.” In Annals of the IVth Global Forum on Fighting Corruption. Office of the Comptroller General of Brazil, pp. 263-292. Fisman, Raymond and Roberta Gatti. 2002. “Decentralization and corruption: Evidence across countries.” Journal of Public Economics, 83(3): 325-345. Fisman, Raymond and Edward Miguel. 2007. “Corruption, Norms, and Legal Enforcement: Evidence from Diplomatic Parking Tickets”, Journal of Political Economy, 115(6): 1020-1047.

27

Friedman, Milton. 1958. “Foreign Economic Aid: Means and Objectives.” Yale Review, 47(4): 500-516. G8 Centre. 2004. “G8 Action Plan: Applying the Power of Entrepreneurship to the Eradication of Poverty.” Published on June 4, 2004. Accessed: February 13, 2010 URL: http://www.g7.utoronto.ca/summit/2004seaisland/poverty.html. Gleditsch, Kristian S. and Michael D. Ward. 1997. “Double Take: A Re-Examination of Democracy and Autocracy in Modern Polities.” Journal of Conflict Resolution, 41: 361-383. Gupta, Sanjeev, Hamid R. Davoodi, and Erwin R. Tiongson. 2002. “Corruption and the Provision of Health Care and Education Services.” In Governance, Corruption, and Economic Performance, ed. Abed George, T. and Sanjeev Gupta. International Monetary Fund: Washington, D.C., pp. 254-279. Hefeker, Carsten. 2009. “Taxation, Corruption, and the Exchange Rate Regime.” CESifo Working Paper Series, No. 2561, CESifo Group Munich. Kaufman, Daniel, Art Kraay, and Pablo Zoido-Lobaton. “Governance Matters.” Policy Research Working Paper, No. 2196, World Bank, Washington D.C. Lambsdorff, Johann G. 1998. “An Empirical Investigation of Bribery in International Trade.” European Journal of Development Research, 10(1): 40-59. Marshall, Monty G. and Keith Jaggers. 2006. PolityIV Dataset. [Computer file; version p4v2006]. College Park, MD: Center for International Development and Conflict Management, University of Maryland. Mauro, Paolo. 1995. “Corruption and Growth” Quarterly Journal of Economics, 60(3): 681-712. Neumayer, Eric. 2003. “What Factors Determine the Allocation of Aid by Arab Countries and Multilateral Agencies?” Journal of Development Studies, 39 (4): 134-147. Obama, Barack. 2009. “Memorandum on Promoting Democracy and Human Rights in Cuba.” August, 13 2009. Olken, Benjamin A. 2007. “Monitoring Corruption: Evidence from a Field Experiment in Indonesia.” Journal of Political Economy, 115 (2): 200-249. Orozco, Manuel and Michelle Lapointe. 2003. “Mexican Hometown Associations and Their Development Opportunities.” Inter-American Dialogue Research Paper Series, September. Peters, Anne Mariel and Peter W. Moore. 2009. “Beyond Boom and Bust: External Rents, Durable Authoritarianism, and Institutional Adaption in the Hashemite Kingdom of Jordan.” Studies in Comparative International Development, 44: 256-285. Political Risk Services. 2005. “International Country Risk Guide Methodology.” Pfutze, Tobias. 2009. “Does Migration Promote Democratization: Evidence from the Mexican Transition.” Working Paper, Oberlin College. Schleifer, Andrei and Robert Vishny. 1993. “Corruption.” Quarterly Journal of Economics, 108(3): 599-617.

28

Singer, David. 2009. “Migrant remittances and exchange rate regimes in the developing world.” MIT Working Paper. Stock, James H., Jonathan H. Wright, and Motohiro Yogo. 2002. “A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments.” Journal of Business and Economic Statistics, 20(4): 518529. Svensson, Jakob. 2000. “Foreign aid and rent-seeking.” Journal of International Economics, 51: 437-461. Tanzi, Vito and Hamid Davoodi. 2002. “Corruption, Public Investment, and Growth.” In Governance, Corruption, and Economic Performance, ed. Abed George, T. and Sanjeev Gupta. International Monetary Fund: Washington, D.C., pp. 280-299. Treisman, Daniel. 2000. “The causes of corruption: a cross-national study.” Journal of Public Economics, 76: 399-457. Treisman, Daniel. 2007. “What Have We Learned About The Causes of Corruption From Ten Years of Cross-National Empirical Research?” Annual Review of Political Science, 10: 211-244. Wei, Shang-Jin. 2000. “How Taxing is Corruption on International Investors?” Review of Economics and Statistics, 82(1): 1-11. Werker, Eric, Faisal Z. Ahmed and Charles Cohen. 2009. “How is aid spent?: Evidence from a Natural Experiment.” American Economic Journal - Macroeconomics, 1(2): 225-244. World Bank. 2004. “The Costs of Corruption” Published: April 8, 2004. Accessed: November 7, 2009. URL: http://web.worldbank.org/WBSITE/EXTERNAL/NEWS. World Bank. 2005. “World Development Indicators, 2005.” World Bank: Washington D.C. Yang, Dean. 2008. “International Migration, Remittances and Household Investment: Evidence from Philippine Exchange Rate Shocks.” The Economic Journal, 118, 591-630.

29

Figures Figure 1: Price of oil and remittances (% GDP) in poor, non-oil producing Muslim countries

Percent of GDP 6

2002 US dollars 90

80 5 70 Remittances (left scale) 60

4

50 3 40 Oil price (right scale)

2

30

20 1 10

0 1972

1976

1980

1984

1988

1992

1996

2000

0 2004

Notes: Total remittances (% GDP) to poor, non-oil producing Muslim countries. Based on author’s calculations.

30

Figure 2: Corruption by regime type

Corruption Index (0=lowest, 6=highest) 4

3 Autocracies

2

Democracies

1

0 1984

1988

1992

1996

2000

2004

Year

Notes: Average corruption score for autocratic and democratic countries. Autocracies are countries with POLITY scores less than +6. Democracies are countries with POLITY scores more than or equal to +6. The POLITY index ranges from -10 to +10 where higher values correspond to greater democratic governance.

31

Figure 3: Percentage of countries that undergo change in corruption from the previous year

Percent 45 Non-Muslim countries

40

Muslim countries

35

30

25

20

15

10

5

0 1985

1987

1989

1991

1993

1995 Year

1997

Notes: Sample of 57 poor, non-oil producing countries from 1985-2004.

32

1999

2001

2003

Tables Table 1: Remittances and Distance from Mecca Distance from Mecca Countries Remittances (% GDP) Under 1000 miles Djibouti, Jordan, Lebanon, Sudan 12.1 1000-2000 miles Afghanistan, Morocco, Somalia, Turkey 3.8 Comoros, Niger, Pakistan 3.2 2000-3000 miles More than 3000 miles Bangladesh, Guinea, Mali, Mauritania, Senegal 2.1 Notes: Sample of 16 poor, non-oil producing Muslim countries between 1972-2004.

33

POLITY -4.3 -2.2 -1.7 -3.9

Table 2: Sample of countries Country name Albania Armenia Bangladesh Belarus Bolivia Botswana Bulgaria Burkina Faso Chile Costa Rica Cote d’Ivoire Dominican Republic El Salvador Estonia Ethiopia Gambia, The Ghana Guatemala Guinea Guinea-Bissau Guyana Haiti Honduras Hungary Jamaica Jordan Kenya Latvia Lebanon Liberia Madagascar Malawi Mali Moldova Mongolia Morocco Mozambique Nicaragua Niger Pakistan Panama Paraguay Philippines Poland Senegal Serbia and Montenegro Sierra Leone South Africa Sri Lanka Sudan Tanzania Togo Turkey Uganda Ukraine Uruguay Zambia Zimbabwe

Remittances (% GDP) 9.2 0.5 3.5 0.1 0.4 0.0 0.0 4.7 0.0 0.5 0.0 6.4 9.2 0.0 0.2 0.0 0.3 2.1 0.3 1.7 3.2 7.5 3.5 0.0 6.7 18.2 0.0 0.0 20.1 4.6 0.2 0.0 3.8 3.8 1.4 7.0 0.1 3.2 0.5 4.1 0.3 0.8 0.6 0.4 2.6 9.5 1.0 0.0 6.0 3.1 0.0 1.8 2.1 1.2 0.3 0.0 0.0 0.0

Corruption 2.8 4.0 4.9 3.0 3.8 2.4 2.1 2.8 3.0 1.2 2.9 2.9 3.2 2.0 3.6 3.0 3.3 3.5 2.6 4.0 3.4 4.9 4.0 1.6 3.6 2.7 3.3 3.4 4.3 3.8 2.0 2.7 4.0 4.0 2.6 3.2 2.4 1.6 2.5 3.9 4.0 4.7 3.8 1.8 3.0 4.1 3.9 0.5 2.7 4.4 2.0 4.0 3.3 3.7 4.7 3.0 4.1 2.7

POLITY 0.4 5.0 2.0 -7.0 8.9 8.2 -1.0 -4.6 2.1 10.0 -6.6 6.7 6.7 6.0 -2.0 5.9 -1.5 4.4 -3.4 -1.9 3.4 -3.1 6.1 5.7 9.5 -4.1 -4.6 8.0 0.0 0.0 1.4 -1.2 1.4 7.6 9.7 -7.1 -0.1 4.8 -0.8 2.0 4.8 2.5 6.3 8.8 0.8 -0.6 -3.2 4.0 5.2 -4.5 -7.0 -4.0 7.6 -4.5 7.0 8.8 -7.1 -4.1

Muslim

X

X

X

X

X

X X

X

X

X

Average remittances inflows, corruption, and POLITY score for the sample of countries from the baseline OLS model. “Muslim” indicates whether the country is in the treatment group.

34

Table 3: Summary statistics Variable name

Observations

Mean

Std Dev

Min

Max

Corruption index 878 3.23 1.10 0 6 Remittances (% GDP) 878 3.05 5.57 0 64.03 Remittances per capita (2000 US$) 878 42.56 118.37 0 1459.56 Log remittances (2000 US$) 878 16.19 10.77 0.00 27.00 Log remittances per capita (2000 US$) 878 1.95 1.91 0 7.29 Dummy for British legal origin 878 0.30 0.46 0 1 Dummy for former British colony 878 0.31 0.46 0 1 Protestant (% of population in 1980) 878 7.68 12.43 0 66 Ethnic-linguistic fractionalization in 1985 878 0.53 0.27 0.01 0.92 Growth in GDP per capita, annual % 878 1.22 5.41 -43.65 35.73 Log GDP per capita (1995 US$) 878 6.62 1.09 4.31 8.72 Log population 878 15.99 1.14 13.52 18.82 POLITY score 863 1.79 6.63 -9 10 Autocracy score 863 0.13 0.11 0.05 0.5 Autocracy score x Remittances (% GDP) 863 0.34 0.88 0.00 12.00 Notes: Summary statistics for poor, non-oil producing countries from the estimating baseline model sample. Autocracy score is equal to 1/(POLITY+11). Liberia is a non-former British colony with British legal origin. Jordan is a former British colony with non-British legal origin (Jordan is based on Islamic law and French codes).

35

Table 4: Annual change in government corruption Percentage change Point change in corruption index All countries Non-Muslim countries Muslim countries -2 1.1 1.1 1.0 -1 5.1 4.4 7.2 0 86.1 86.6 84.5 +1 6.6 6.9 5.7 +2 1.1 1.0 1.6 NOTES: Summary statistics for poor, non-oil producing countries from the baseline estimation model sample. Excludes the initial year of 1984.

36

Table 5: Determinants of corruption Dependent variable:

Corruption index (1) (2) (3) Remittances (% GDP) 0.032 0.011 0.008 [0.011]*** [0.010] [0.024] Autocracy x Remittances (% GDP) 0.112 [0.061]* Autocracy score -1.059 [0.543]* Time varying controls variables Y Y Y Time invariant control variables Y Year trend Y Y Y Country fixed effects Y Y Number of observations 878 905 881 R-squared 0.12 0.66 0.67 Notes: Estimation via OLS. Robust standard errors, clustered by government reported in brackets. * = significant at 10%; ** = significant at 5%; *** = significant at 1%. Time-varying controls include growth in GDP per capita (% annual), log GDP per capita (1995 US$), and log population. Time invariant controls include dummies for former British colony, British legal origin, protestant (% of population in 1980), and ethnic-linguistic fractionalization in 1985. These coefficients and a constant are not reported.

37

Table 6: First stage regression Dependent variable:

Distance to Mecca x Oil price in t

Remittances (% GDP) (1) 0.026 [0.008]***

Distance to Mecca x Oil price in t-1

(2)

(3)

Remittances per capita (2000 US$) (4) 0.246 [0.078]***

0.026 [0.008]***

Muslim x Oil price

0.033 [0.030] POLITY Autocracy score 0.144 0.146 0.15 3.214 [0.062]** [0.062]** [0.061]** [1.145]*** Growth in GDP per capita, annual % 0.088 0.086 0.089 0.669 [0.033]*** [0.033]*** [0.033]*** [0.347]* Log GDP per capita (1995 US$) -6.549 -6.508 -6.638 -24.457 [2.219]*** [2.210]*** [2.244]*** [21.711] Log population -10.305 -10.117 -10.603 -209.855 [4.291]** [4.294]** [4.491]** [73.270]*** Constant -818.821 -825.231 -784.764 -12946.99 [175.999]*** [175.743]*** [164.026]*** [3364.055]*** Year trend Y Y Y Y Country fixed effects Y Y Y Y F-stat on instrument 11.9 11.4 1.2 9.94 Number of observations 863 863 863 863 R-squared 0.78 0.78 0.78 0.85 Notes: OLS regression. Sample restricted to poor, non-oil producing countries. Robust standard errors, clustered by government reported in brackets. * = significant at 10%; ** = significant at 5%; *** = significant at 1%. Distance measures the number of miles (in thousands) between the capital city of a poor, non-oil producing Muslim country to Mecca. A country is defined as Muslim if at least 70% of the population identify with Islam.

38

39

OLS (1) 0.044 [0.038]

OLS (2) 0.026 [0.040] 0.056 [0.105]

2SLS (3) 0.337 [0.176]*

0.306 [0.166]*

1.473 [0.564]***

Corruption index 2SLS 2SLS (4) (5)

0.143 [0.061]**

2SLS (6)

0.036 [0.019]*

2SLS (7)

2SLS (8)

0.951 [0.508]* Time-varying controls Y Y Y Y Y Y Y Y Year trend Y Y Y Y Y Y Y Y Country fixed effects Y Y Y Y Y Y Y Y Sample Treatment Treatment Entire Entire Entire Entire Entire Entire Number of observations 185 189 863 863 863 863 863 863 R-squared 0.69 0.7 0.31 0.38 0.21 0.3 0.3 0.29 Notes: OLS regression. Treatment sample refers to poor, non-oil producing Muslim countries. Entire sample refers to poor, non-oil (Muslim and non-Muslim) producing countries. Robust standard errors, clustered by government reported in brackets. * = significant at 10%; ** = significant at 5%; *** = significant at 1%. Time-varying controls include growth in GDP per capita (% annual), log GDP per capita (1995 US$), log population, and POLITY autocracy score. In columns 3, 5-8 the various measures of remittances are instrumented with Distance to Mecca x Oil price. Remittances (% GDP), 1 year is instrumented with Distance to Mecca x Oil price in year t-1.

Log remittances per capita (2000 US$)

Remittances per capita (2000 US$)

Log remittances (2000 US$)

Remittances (% GDP), previous year

Autocracy x Remittances (% GDP)

Remittances (% GDP)

Dependent variable: Method of estimation:

Table 7: Remittances and government corruption

Table 8: Remittances and government patronage Compensation of employees Executive constraints Dependent variable: (% govt expenditures) (1=low, 7=high) (1) (2) (3) (4) (5) (6) Remittances (% GDP) 3.537 -0.283 [1.065]*** [0.159]* Remittance (% GDP), previous year 3.653 -0.305 [1.073]*** [0.159]* Remittances per capita (2000 US$) 0.284 -0.023 [0.119]*** [0.013]* Time-varying controls Y Y Y Y Y Y Continent dummies Y Y Y Country dummies Y Y Y No. observations 315 315 315 1553 1553 1553 R-squared 0.01 0.01 0.01 0.86 0.84 0.83 Notes: OLS regression. Sample restricted to poor, non-oil producing countries. Robust standard errors, clustered by government reported in brackets. * = significant at 10%; ** = significant at 5%; *** = significant at 1%. In columns 1-3, the dependent variable is the government compensation to employees (% government expenditures). In columns 4-6, the dependent variable is the executive constraints index. Time varying controls include growth in GDP per capita (% annual), log GDP per capita (1995 US$), log population, and the POLITY autocracy score.

40

Table 9: Specification checks Dependent variable:

Corruption High corruption (1) (2) (3) (4) (5) Remittances (% GDP) 0.354 0.399 0.289 0.327 0.194 [0.193]* [0.215]* [0.129]** [0.163]** [0.038]*** Time-varying covariates Y Y Y Y Year trend Y Y Y Y Y Country fixed effects Y Y Y Y Y Number of observations 843 838 918 860 869 R-squared or Log Likelihood 0.3 0.33 0.2 0.32 -2798.3 Notes: In columns 1-4, estimation via 2SLS. In column 5, estimation via instrumental probit. Robust standard errors, clustered by government reported in brackets. * = significant at 10%; ** = significant at 5%; *** = significant at 1%. Column 1 excludes observations from Jordan. Column 2 excludes observations with remittances exceeding 15% GDP. Column 3 excludes all time varying covariates. In column 4 all time varying covariates are lagged one year. In column 5, the dependent variable is equal to 1 if a country receives a “high” corruption score of 5 or 6, and zero otherwise. Time-varying countries include growth in GDP per capita (% annual), log GDP per capita (1995 US$), log population, and POLITY autocracy score.

41

Table 10: Accounting for differential trends Dependent variable: Corruption (1) (2) (3) (4) (5) Remittances (% GDP) 0.222 0.359 0.295 0.541 0.33 [0.111]** [0.184]* [0.204]† [0.308]* [0.167]* Muslim*Autocracy 0.085 [0.049]* Cold War*Muslim 0.109 [0.312] Average income*Year -0.025 [0.014]* Time-varying covariates Y Y Y Y Y Year trend Y Y Y Y Y Country fixed effects Y Y Y Y Region*Year trends Y Country*Year trends Y Number of observations 863 863 863 863 863 R-squared 0.4 0.31 0.52 0.32 0.33 Notes: 2SLS regression. Robust standard errors, clustered by government reported in brackets. †= significant at 15%; * = significant at 10%; ** = significant at 5%; *** = significant at 1%. Time-varying countries include growth in GDP per capita (% annual), log GDP per capita (1995 US$), log population, and POLITY autocracy score.

42

Table 11: Tests of the exclusion restriction Dependent variable: Corruption (1) (2) (3) (4) Instrumented variables Remittances (% GDP) 0.332 0.36 0.338 0.402 [0.174]* [0.206]* [0.178]* [0.206]* Trade and remittances (% GDP) Additional controls Foreign aid (% GDP) Exchange rate, % annual

(5)

0.052 [0.027]*

0.01 [0.011] 0.001 [0.001]

Inflation, % annual

0.000 [0.000]

Trade (% GDP)

-0.01 [0.006]* Time varying controls Y Y Y Y Y Year trend Y Y Y Y Y Country fixed effects Y Y Y Y Y Number of observations 837 843 862 845 845 R-squared or Log likelihood 0.3 0.29 0.31 0.22 0.28 Notes: 2SLS regression. Robust standard errors, clustered by government reported in brackets. * = significant at 10%; ** = significant at 5%; *** = significant at 1%. Time-varying countries include growth in GDP per capita (% annual), log GDP per capita (1995 US$), log population, and POLITY autocracy score.

43

Table 12: Remittances and government expenditures in Jordan, 1990-2004 1990-1994 1995-1999 2000-2004 Remittances (% GDP) 15.0 20.3 20.2 Remittances per capita (2000 US$) 245.7 348.6 357.1 Government employee compensation (% expenditures) Government transfers (% expenditures) Corruption Notes: Average values for each sub-period.

44

62.9 12.8

67.1 11.6

68.4 10.1

2.4

2.2

3

Table 13: Remittances and government expenditures Dependent variable: Subsidies and transfers Immunizations Health care expenditures (% GDP) (% govt. expenditures) to measles Public Private (1) (2) (3) (4) Remittances (% GDP) -4.468 -5.659 -0.427 0.138 [2.253]** [1.644]*** [0.290] [0.283] Time-varying controls Y Y Y Y Year trend Y Y Y Y Continent dummies Y Y Y Y Number of observations 316 1299 298 298 R-squared 0.05 0.64 0.11 0.1 Notes: Estimation via 2SLS. Robust standard errors, clustered by government reported in brackets. * = significant at 10%; ** = significant at 5%; *** = significant at 1%. In column 2, “Immunizations” measures the percentage of children (aged 12-23 months) immunized to measles. In column 3, the dependent variable is the government’s expenditures on health care (% GDP). In column 4, the dependent variable is the household’s expenditures on health care (% GDP). Time-varying countries include growth in GDP per capita (% annual), log GDP per capita (1995 US$), log population, and POLITY autocracy score.

45

46

Tax revenue (% GDP) (1) -0.255 [0.487]

Income tax (% total tax) (2) -1.755 [1.115]†

Income tax (% revenue) (3) -1.617 [1.071]†

Taxes on goods & services (% revenue) (4) -1.592 [1.009]†

Anti-government demonstrations (5) -0.152 [0.187]

(6) 0.337 [0.175]* -0.001 [0.024] Y Y

Corruption

Time-varying covariates Y Y Y Y Y Year trend Y Y Y Y Y Continent fixed effects Y Y Y Y Country fixed effects Y Y No observations 391 391 382 376 1522 862 R-squared 0.21 0.16 0.16 0.17 0.26 0.31 Notes: Estimation via 2SLS. Robust standard errors, clustered by government reported in brackets. †= significant at 15%; * = significant at 10%; ** = significant at 5%; *** = significant at 1%. Income tax includes taxes on income, profits, and capital gains. Anti-government demonstrations counts the number of public gatherings (of at least 100 people) for the primary purpose of displaying or voicing their opposition to government policies or authority, excluding demonstrations of a distinctly anti-foreign nature. Time-varying countries include growth in GDP per capita (% annual), log GDP per capita (1995 US$), log population, and POLITY autocracy score.

Anti-government demonstrations

Remittances (% GDP)

Dependent variable:

Table 14: Discounting alternate mechanisms

Appendix A: Additional figures Figure 1: Public health expenditures (% GDP) by regime type

8

DEU

0

Health expenditure, public (% of GDP) 2 4 6

HRV E FRA SW DNK PANNOR PRT AUS CAN BEL JPN NZL SVN CZE ITA GBR USA CHE ISR CRI AUT ESP GRC SVK NLD MKDYUG FIN HUN BLR ARG LSO TGO NAM IRL LTU MNG EST GUY JOR TUR ZWE BGRPOL PNG ROM SLV NIC BOL SW ZBTN ZAF URY MOZPRY DJI MDALVA JAM VEN RUS GMB TKM BRA HND ERI RWA SAUBHR UKR MW I LBR ZMB BWA TUNTCD GNB UZB SLE ETH TZA MEX FJ I OMNAREKWT DZA KEN HTI CHL CYP IRN NERALB MLI GTM GHA SYR DOMTHA MUS KOR QAT KAZ KGZ BEN SEN CHN AGO LKA EGY UGABFA KHMMYS PHL MAR ARM CAF LAO MRT YEM COM TTO LBY VNM COGGNQGAB NGA CIV CMR IND SGP SOM GEOECU MDG PAK NPL TJK GIN AZE BGDIDN SDN BDI IRQ AFG MMR ZAR COL

CUB

-10

-5

0 POLITY score

5

10

Notes: Public health expenditures (% GDP) by regime type in 2000. The POLITY index measures the regime type. The POLITY index ranges from -10 to +10 where higher values correspond to greater democratic governance.

47

Figure 2: Remittances (% GDP) in poor, non-oil producing countries

Remittances (% GDP) 30

2002 US dollars 90

80 25 70

20

60

Jordan (left scale)

50 15 40

Pakistan (left scale)

Oil price (right scale)

10

30

20

Mali (left scale)

5

10

0 1972

1976

1980

1984

1988

48

1992

1996

2000

0 2004

Figure 3: Government tenure, by corruption level

Years in office 16 14.8 14

12

9.7

10

8 6.6 6

4

3.6

2

0 0-1

2-3

Corruption

49

4-5

6

Figure 4: Trends in autocracy

POLITY (0-21) score

Proportion

14

0.90

0.80 12

Average POLITY score (left scale) 0.70

10 0.60

Authoritarian countries (right scale) 8

0.50

0.40

6

0.30 4 0.20 2 0.10

0 1984

0.00 1988

1992

Year

50

1996

2000

-6

-4

Residuals -2 0

2

4

Figure 5: Residual plot from the baseline IV regression

0

20

10 Remittances (% GDP)

51

30

Figure 6: Allocation of government expenditures, by level of remittance inflows

Percent of government expenditure Compensation to government employees Government transfers

45 40

38.1

35

33.3 31.1

30.2 30 27.0

26.1

25 20 15 10 5 0 4

Appendix B: Model B1. Players and preferences There are two actors in the model: a representative household and government. For simplicity, there are two goods in the model. One is a private good that must be purchased by the household and the other good could be provided by the government or by the household (e.g., education and health services). The funding of this latter welfare good does not affect the good’s marginal utility. This means the quality of this welfare good is the same whether it is provided by the government or by the household, although households would prefer the government to supply these goods. Households have Cobb-Douglas preferences over these goods given by: U (c, p, g) = λlog(c) + (1 − λ)log(p + g)

(1)

where c is the representative household’s consumption of the private good, p is the household’s consumption of the welfare good, and g is the government’s provision of that good. The parameter λ is the weight households place on private goods relative to welfare goods (0 < λ < 1). Households finance their expenditures subject to their budget constraint: (1 − t)y + R = c + p

(2)

where y is the household’s income, t is the tax rate, and R is remittances. R is untaxed by the government. Governments also care about patronage goods and welfare goods, but do so in relation to their main objective of staying in power. They do this by redistributing economic and/or political rents to key individuals (e.g., party supporters, business elites, military officials) and groups (e.g., organized labor, the majority of the voting population) in return for their political support. The distribution of rents can be in the form of welfare and patronage goods. The relative importance of providing patronage goods to welfare goods is captured in the α parameter in the government’s utility function: φ(k, U ) = αlog(k) + (1 − α)U (c, p, g)

(3)

where, 0 < α < 1, k stands for whatever the government keeps for its own consumption. Given these parameters, the government chooses k to maximize its utility function subject to its budget constraint: ty = g + k (4) where t is the tax-rate and y is income (thus ty is government tax revenue).

B2. Equilibrium I model the interaction between the representative household and government as a Stackelberg game where the government moves first. Solving for the equilibrium therefore requires backward induction. The household’s provision of welfare goods is given by maximizing its 53

utility function (with respect to p) subject to its budget constraint. The households optimal provision of welfare goods is: p∗ = (1 − λ)[(1 − t)y + R] − λg

(5)

The first order condition given by equation (5) shows that the household’s optimal expenditures on welfare goods is increasing in total household income (i.e., after-tax income plus remittances) and decreasing in the government’s provision of the welfare good. This means that if the government increases its provision of the welfare good, the household will reduce its expenditures on that good. Since the household allocates its budget between p and c, if household expenditures on the welfare good rise then expenditures on private goods must decline (and vice-versa). In the first stage of the game, the government incorporates the household’s optimal provision of welfare goods in to its utility function. After making this substitution, the government’s optimal provision of welfare goods is determined by maximizing its utility function (with respect to g) subject to its budget constraint. The government’s optimal provision of welfare goods is given by: g ∗ = (t − α)y − αR

(6)

Whatever the government does not spend on welfare goods can be used in the provision of patronage goods. In equilibrium, the government allocates for itself: k ∗ = α(y + R)

(7)

B3. Interpreting the parameters The relative distribution of welfare goods to patronage goods supplied by the government tends to differ by regime type. For instance, in their theory of democratization, Acemoglu and Robinson (2006) argue that a country democratizes as a credible commitment to future redistribution. By design, governments in democracies will therefore spend a larger fraction of their revenue on the provision of public goods. Similarly, Bueno de Mesquita et al’s (2003) selectorate theory demonstrates that given their political institutional constraints, democratic governments tend to provide a greater share of welfare goods than their authoritarian counterparts. Doing so better ensures the political survival of democratic governments. This tradeoff between patronage goods and welfare goods will enter the government’s utility function. The government’s utility function can also be interpreted as its survival function. The government will choose the optimal bundle of patronage and welfare goods to maximize its probability of staying in power. Thus, governments in autocracies will place greater weight on expenditures for patronage (compared to democracies) in their utility functions. This implies that α gauges a country’s degree of institutionalized autocracy in the government’s utility function (equation 3), where values closer to one correspond to more autocratic governance. Based on this interpretation, equation (6) therefore demonstrates that the government’s optimal provision of welfare goods is increasing in income and decreasing in remittances. The reduction in the government’s provision of welfare goods associated with

54

higher remittances represents a substitution effect. Moreover, the interaction of R with α implies that the substitution effect is magnified by the country’s underlying level of institutionalized autocracy. This means that governments in more autocratic-leaning political systems will tend to reduce their provisions of public services the most. Reducing the provision of welfare goods, frees resources for governments to engage in corruption. Equation (7) demonstrates that the government’s optimal expenditure on patronage is increasing in income and remittances. Moreover, the overall effect is magnified by the government’s underlying level of institutionalized autocracy. Since k* represents expenditures on patronage goods that reward the government’s key supporters, equation (7) clearly shows a direct link between remittances, autocracy, and patronage.

55