World Development Vol. 40, No. 2, pp. 291–302, 2012� 2011 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevdoi:10.1016/j.worlddev.2011.06.003

Impact of Pension Privatization on Foreign Direct Investment

CHRISTOPHER REECEUniversity of Pennsylvania, USA

and

ABDOUL G. SAM *

The Ohio State University, USA

Summary. — We explore the causal effect of market-oriented pension reform on net foreign direct investment (FDI) inflows in LatinAmerica and among the transitional economies of Eastern Europe and Central Asia, both of which have experienced waves of pensionprivatization and FDI over the last two decades. With our balanced panel of 42 countries over the 1991–2006 period, we implement fixedeffects models, controlling for the decision to enact full or partial privatization of the public pension system and several other covariateswhose choice is informed by the rich empirical literature on FDI. Our econometric results indicate that pension privatization triggers asignificant increase in net FDI inflows and that the effect does not wane over time. We estimate that full privatization increases FDI as apercentage of GDP by about 57%, ceteris paribus.� 2011 Elsevier Ltd. All rights reserved.

Key words — pension privatization, foreign direct investment, Latin America, transitional economies

* We are grateful to the Editor-in-Chief of World Development, Oliver

Coomes, two anonymous referees, Sarah Brooks, Alfonso Flores-Lagun-

es, Alan Ker, Randall Aguilar, Doga Kerestecioglu, Claudio Gonzales-

Vega, seminar participants at The Ohio State University, the University of

Guelph, and the University of Florida for very helpful comments and

suggestions. The usual disclaimer applies. Final revision accepted: May 19,2011.

1. INTRODUCTION

In 1981, the military dictatorship of Augusto Pinochet un-veiled the centerpiece of its dramatic neoliberal agenda: theprivatization of Chile’s national pension system. With thisglobally unprecedented act, the public system was closed tonew workers, who instead began making mandated payrollcontributions to individualized, privately-managed retirementaccounts. In the subsequent decade, Chile experienced highgrowth rates while most Latin American countries stagnated.Moreover, solvency of the public pension system was seriouslyjeopardized in several countries due to a continual decline inthe workers to pensioners’ ratio, increasing longevity, cost-of-living adjustments, high evasion rates spurred by high pay-roll taxes, and built-in weaknesses of the public system, amongother factors (Kay, 2000). 1 Together, the apparent success ofthe Chilean privatization initiative and the looming bank-ruptcy of several public pension programs provided a strongimpetus for reform. Peru implemented a partial privatizationscheme in 1993, followed in 1994 by Argentina and Colom-bia. 2 In 1994, the World Bank launched an international cam-paign in support of a market-oriented “three-pillar model” forpension reform. 3 By 2009, at least 30 governments worldwidehad adopted some form of pension privatization (Holzmann,Mackellar, & Repansek, 2009).

Chile’s pioneering experiment and the World Bank’s advo-cacy for similar reform have spurred an active debate amongscholars about the macroeconomic effects of pension privatiza-tion (Arza, 2008; Catalan, 2004; Catalan, Impavido, &Musalem, 2001; Feldstein, 1997; Kay, 2000; Mesa-Lago,2002; Orszag & Stiglitz, 2001). For example, some economistsargue that by increasing savings and generating demand forfinancial instruments, privatization can develop capital mar-kets, attract foreign capital (Kay, 2000; Madrid, 1999) andstimulate growth (Catalan, 2004; Feldstein, 1997).

In this paper, we concern ourselves with an important facetof this debate, which pertains to the relationship between priv-atization and Foreign Direct Investment (FDI). The literature

291

on the determinants and effects of FDI is one of the most ac-tive in International Economics. Of particular interest to us, anumber of empirical papers find evidence of salutary effects ofFDI on growth (Alfaro, Chanda, Kalemli-Ozcan, & Sayek,2004a; Blomstrom & Kokko, 2003; Borensztein, De Gregorio,& Lee, 1998; Hermes & Lensink, 2003) and productivity(Alfaro, Kalemli-Ozcan, & Sayek, 2009; Xu, 2000; survey byLipsey (2004)) for host countries, contingent on adequatehuman capital and financial development.

We confine our empirical analysis to two regions where bothpension privatization and FDI inflows have been intense overthe last two decades: Latin America and transitional econo-mies of Eastern Europe and Central Asia. Of the 30 countriesthat adopted pension privatization between 1981 and 2009,nine are in Latin America 4 and 13 are in Eastern Europeand Central Asia 5 (Holzmann et al., 2009). Over the same per-iod, the average annual FDI flows into Latin American andCaribbean countries grew over 10-fold, from $6.04 billion(0.78% of GDP) in the 1980s to $62.71 billion (2.90% ofGDP) in the 2000s. Similarly, average annual FDI flows toEastern European and Central Asian countries increased over15-fold, from approximately $2.31 billion (.08% of GDP) inthe 1980s to $35.65 Billion (3.12% of GDP) in the 2000s.

Academics have advanced a number of arguments sugges-tive of a positive causal effect of pension privatization onFDI flows. Three distinct mechanisms stand out. First, privati-zation may function as a favorable signal to investors by indi-cating a state’s commitment to long-term fiscal responsibilityand macroeconomic stability (Kay, 2000; Madrid, 1999;

292 WORLD DEVELOPMENT

Maxfield, 1997). Second, in many cases, privatization wasaccompanied by reductions in payroll tax rates (Brooks,2007a; Dion, 2009; Madrid, 1999), potentially lowering wagecosts. Third, privatization may attract FDI indirectly throughits promotion of domestic financial development. For exam-ple, Catalan et al. (2001) find that the development of pensionfunds correlates positively with the development of financialmarkets. Financial development could in turn spur additionalFDI (Albuquerque, Loayza, & Serven, 2005).

On the other hand, privatization has a significant transitioncost in the short to medium terms, since the state must con-tinue to fund existing pensioners while forfeiting revenues thatwere previously used for this purpose (Brooks, 2007a; Cuevas,Gonzalez, Lombardo, & Lopez-Marmolejo, 2008). 6 The statemay choose to finance this transition cost in part through taxincreases and benefit reductions, as was the case in Chile, theDominican Republic, and El Salvador (Holzmann and Hinz,2005). However, these actions carry high political costs, andthus, some combination of debt and inflation may also be nec-essary. This may in turn spur capital flight as leery interna-tional investors divest. This argument suggests a negativelink between pension privatization and FDI flows.

The purpose of this research is to test the empirical validityof these competing theories. Among the countries that choseto privatize, was privatization associated with less FDI thanksto the hefty transition costs? Or, did foreign investors respondfavorably to the policy change, seeing beyond the transitioncosts and recognizing it as a favorable signal? To answer thesequestions, we gather a balanced panel of 42 Eastern European,Central Asian, and Latin American countries from 1991 to2006 and implement panel data econometric methods, control-ling for the effects of privatization and other relevant covari-ates from the literature. We use two variables to control forthe effects of pension reform on FDI flows. The first is a dum-my variable that captures the decision to privatize. The secondis a continuous variable that measures the intensity of privati-zation. This measure of intensity is desirable due to substantialvariation among countries that chose to enact privatization;some countries dismantled the public pension system alto-gether, while others left a substantial public component inplace. 7 In our model, intensity is measured as the percentageof pension income that is derived from an average pensioner’sprivate account following the enactment of reform. For thismeasurement, we use a simulation carried out by Brooks(2009) that weighs an average worker’s projected public bene-fits against her projected payments from her private account. 8

In countries where full privatization is enacted and the publicsystem is dismantled entirely, the intensity measure is 100%, aspensioners must rely exclusively on private accounts. 9 Incountries where the public system remains open to either com-pete with or supplement the private system, or where the pri-vate system is dominant but the state continues to make smallpayments from the public purse, 10 the intensity measure takeson a range of values. To account for the endogeneity of priv-atization, we use country-invariant effects as well as instru-mental variables.

Our study reveals that pension privatization spurs a statisti-cally and economically significant boost to FDI inflows; wealso find that this impact does not vanish over time. From apolicy standpoint, the link between pension privatizationand increased FDI is important, given the empirical evidencethat FDI inflows positively impact growth and productivity.

The paper is structured as follows. In Section 2, we providea brief literature review of pension privatization and discuss itspotential effects on FDI inflows. In Section 3, we discussdescriptive statistics of our data. In Section 4 we present the

econometric model. In Section 5 we discuss the empirical re-sults; Section 6 concludes.

2. BACKGROUND AND CAUSAL MECHANISMS OFPENSION PRIVATIZATION

(a) Background

The PAYGO model remains the prevailing form of publicpension provision across the globe, and thus, the object ofmarket-oriented pension reform. Under the PAYGO system,current workers make regular payroll contributions to a publicpension fund, managed by the government, which is used tosupport current retirees. Despite cross-country variation inmarket-oriented reform programs, two common features de-fine all privatization schemes: (1) the (full or partial) place-ment of pension funds under private management and (2)the establishment of mandated savings programs and individ-ualized accounts, whereby some element of financial risk andsavings responsibility is transferred from the state to the indi-vidual (Brooks, 2007b).

According to the World Bank (1994) and other advocates(see e.g., Orszag & Stiglitz, 2001; Singh, 1996 for a review),privatization offers three types of potential advantages overthe PAYGO system. First, privatization may improve thefinancial performance of pension schemes and the reliabilityof old-age benefits. Performance may be improved in two re-spects: administrative costs are reduced and private portfolioaccounts yield potentially higher earnings (Palacios & White-house, 1998). Reliability is improved by removing the fiscalchallenges associated with demographic flux as the pen-sioner-to-worker ratio increases. Furthermore, mandatorysavings programs, along with incentives to contribute beyondmandated levels, generate new retirement savings and encour-age long-term planning (World Bank, 1994).

Second, by linking each individual’s contributions to thebenefits she will receive, the World Bank (1994) claims thatprivatization eliminates the “perverse redistributions”—bothintragenerational and intergenerational—of the PAYGOmodel. Intergenerational redistributions become problematicwhen demographic flux forces a relatively small number ofworkers to support a relatively large number of pensioners;intragenerational redistributions can cause labor-market dis-tortions, such as an increased demand for jobs in the informalsector, where payroll taxes may be avoided. In contrast, priv-atization reduces incentives for evasion, resulting in expandedcoverage and increased savings.

Third, proponents argue that pension privatization mayspur economic growth. By redirecting the flow of retirementsavings into local securities markets, pension reform stimu-lates financial innovation at the local level, leading to increasesin both the aggregate supply of capital and the efficiency of itsuse (Gill, Packard, & Yermo, 2005). According to advocatesfor privatization, these conditions facilitate growth (Feldstein,1997; World Bank, 1994) as an empirical link between capitalmarket development and economic expansion is well-estab-lished in the literature (see, e.g., Beck & Levine, 2001; Levine& Zervos, 1998). Indeed, empirical evidence suggests that pen-sion privatization can stimulate financial innovation as well asfinancial deepening: for example, Holzmann (1997) finds thatthe implementation of the Chilean reform correlated positivelywith several indicators of financial market development, suchas the market capitalization to GDP and financial liabilities toGDP ratios, as well as with the emergence of increasinglysophisticated financial instruments. Furthermore, Catalan

IMPACT OF PENSION PRIVATIZATION ON FOREIGN DIRECT INVESTMENT 293

et al. (2001) find empirical evidence of a positive link betweenthe ratio of pension funds to GDP and the ratio of stock mar-ket capitalization to GDP for several countries. 11

Finally, proponents assert that privatization can help attractFDI. In the next section, we review the purported mechanismsthrough which this relationship occurs. We identify three dis-tinct channels: signaling, taxation, and financial development.

(b) Causal mechanisms

The first mechanism by which privatization may attract FDIlies in its credible signaling capacity. Privatization is alleged tofunction as a favorable signal to foreign investors because itdemonstrates a commitment to fiscal responsibility and mar-ket-friendly policy, both of which decrease investment risk(Kay, 2000; Maxfield, 1997). Because direct investments areless mobile than other types of capital, direct investors maybe particularly interested in present indicators of future statebehavior. Indeed, Campos and Kinoshata (2008) find evidencethat market-friendly structural reforms such as financial liber-alization and the privatization of state-owned enterprises canact as favorable signals and promote FDI. Pension privatiza-tion may be a particularly credible signal because of its mag-nitude (pension schemes are often a state’s largest publicprogram) and its high short-term political and fiscal costs(Brooks, 2007a), thus indicating a state’s willingness to makeshort-term sacrifices for long-term macroeconomic stability.

Second, privatization may attract FDI indirectly by cuttingpayroll taxes. As the competition to attract FDI intensifiedduring the 1990’s, the importance of reducing wage costswas seen as paramount (Kay, 2000). In contrast to PAYGOsystems, which are sometimes characterized by large contribu-tions from employers, privatization often reduced employertax rates. 12 Advocates argued that this would bring downwages and attract investment (Brooks, 2007a; Kay, 2000).However, reductions in payroll tax rates for employers weresometimes coupled with higher rates for workers. 13 In a com-petitive labor market, these increases may drive wages upwardand offset the gains from lower taxes on employers. Further-more, while several studies have suggested a negative effectof corporate tax rates on FDI (Gastanga, Nugent, & Pasham-ova, 1998; Wei, 2000), these investigations focus exclusively oncorporate income tax rates. Easson (2004) suggests that thisnarrow focus may be justified: based on a survey of Fortune500 executives, the author concludes that corporate incometaxes significantly influence direct investment decisions, butthat payroll taxes are generally less important. These consider-ations cast some doubt on the plausibility of the taxationchannel.

Third, privatization may boost FDI indirectly through itsaforementioned promotion of financial development. In theabsence of local financial markets, foreign firms operate in“isolated enclaves with no links whatsoever to the domesticeconomy” (Alfaro, Chanda, Kalemli-Ozcan, & Sayek, 2004b,p. 92), whereas a developed financial system can help domesticentrepreneurs accommodate foreign firms. The authors pro-vide anecdotal evidence from the experience of a Suzuki man-ufacturing plant established in India: thanks to accessible localfinancing, a robust ancillary parts industry emerged to supplythe foreign venture, so that the importation of parts from Ja-pan was no longer necessary (p. 92). For this reason, foreigninvestors may care about financial development even wherethey need not rely on domestic financing. Campos andKinoshata (2008) find evidence for this claim. Similarly, Albu-querque et al. (2005) find that the ratio of credit to the privatesector as a percentage of GDP positively impacts FDI inflows,

particularly in developing countries. Thus, the financial devel-opment channel may provide an important mechanismthrough which privatization enhances FDI. For all these rea-sons, we posit that pension privatization positively impacts in-ward FDI.

Conversely, Brooks (2007a) notes that pension privatizationis extremely costly in the short-term. Upon implementation,the state must continue to fund existing pensioners who relyon the previous system while forfeiting revenues that were pre-viously used for this purpose, as privatization generally re-duces contributions to the public system (Cuevas et al.,2008). Though the state may finance this transition throughtax increases and benefits cuts, these options often carry highpolitical costs. Therefore, transitional costs may also entailsome combination of debt and inflation. Large fiscal deficitsbrought about by pension reform and resultant economicshocks could in principle both deter capital inflows and triggercapital flight, as leery investors turn elsewhere. If so, pensionreform could reduce net FDI inflows, contrary to our mainpremise.

We note, however, that the potential for capital flight fol-lowing privatization may be tempered by a final consideration:domestic policymakers may strategize in order to limit the eco-nomically disruptive effects of reform. Based on a series of casestudies, Brooks (2007a) concludes that the decision of whetherto privatize was driven by a cost-benefit analysis, where long-term benefits were weighed against the policy’s transitioncosts. In cases where these costs presented an unacceptablylarge fiscal shock, governments chose to retain the PAYGOmodel. Where short-term costs were high but not prohibitive,policy-makers chose to privatize, but limited the degree ofprivatization in proportion to the magnitude of transitioncosts. In these cases, a large public fund was often maintainedalongside the new individual accounts (Brooks, 2007a). Final-ly, in cases where a pre-existing budget surplus could readilyprovide for short-term costs, policy-makers chose to enactmore comprehensive forms of pension privatization.

3. DATA AND DESCRIPTIVE STATISTICS

We gather a panel data with annual observations for 42 La-tin American and transitional economies from 1991 to 2006.Allowing for lagging, we have a balanced panel of 42 countriesand 15 observations per country for a total of 630 observa-tions. Table 1 displays the countries used in the study, indicat-ing the year of reform for those countries that have enactedprivatization. The countries considered in our analysis thathave adopted privatization did so within the time span ofthe dataset (with the exceptions of Chile (1981), the SlovakRepublic (2007), and Romania (2008)); consequently they ap-pear in the data first as non-privatized. We have two sourcesof identification of the “treatment” effect; pre-reform datafor 22 countries that ultimately privatized, and data on 20countries that chose not to privatize. In addition, for the priv-atized countries, the table displays the intensity of reform,measured as the percentage of an average worker’s pension in-come derived from her private account following the enact-ment of reform. A 100% in this column generally indicateseither full privatization or, in some cases, the enactment of aparallel model that incentivizes workers to participate exclu-sively in the private system. 14 Descriptive statistics are pre-sented in Table 2, which displays mean values of FDI as apercentage of GDP and the twelve time-varying control vari-ables for privatized and non-privatized countries. The first seg-ment compares values for countries that retain public systems

Table 1. Countries included in the study with year and intensity ofprivatization

Country Currentstatus ofpensionsystem

Year ofprivatization

Intensity ofprivatization

Armenia PAYGO N/A N/ABelarus PAYGO N/A N/ABrazil PAYGO N/A N/ACzech Republic PAYGO N/A N/AEcuador PAYGO N/A N/AGeorgia PAYGO N/A N/AGreece PAYGO N/A N/AGuatemala PAYGO N/A N/AGuyana PAYGO N/A N/AHonduras PAYGO N/A N/AKyrgyz Republic PAYGO N/A N/AMoldova PAYGO N/A N/ANicaragua PAYGO N/A N/APanama PAYGO N/A N/AParaguay PAYGO N/A N/ATurkey PAYGO N/A N/ATurkmenistan PAYGO N/A N/AUkraine PAYGO N/A N/AUzbekistan PAYGO N/A N/AVenezuela PAYGO N/A N/AArgentina Privatized 1994 54Bolivia Privatized 1997 92Bulgaria Privatized 2002 37Chile Privatized 1981 100Colombia Privatized 1994 100Costa Rica Privatized 2001 20Croatia Privatized 2002 34Dominican Republic Privatized 2003 100El Salvador Privatized 1998 100Estonia Privatized 2002 42Hungary Privatized 1998 39Kazakhstan Privatized 1998 100Latvia Privatized 2001 45Lithuania Privatized 2004 23Macedonia Privatized 2005 31Mexico Privatized 1997 91Peru Privatized 1993 100Poland Privatized 1999 49Romania Privatized 2008 25Russia Privatized 2004 25Slovak Republic Privatized 2007 38Uruguay Privatized 1996 48

294 WORLD DEVELOPMENT

in the given year to values for countries that have privatized bythe given year, while the second segment compares values onlyfor countries that have privatized, pre- and post-privatization.In a statistical sense, most of the variables have significantlydifferent means for privatized than for non-privatizedcountries. In particular, the sample average of FDI inflowsas a percentage of GDP is larger for reformers than for non-reformers.

In addition, as shown in Table 2, privatized and non-priv-atized countries have significantly different mean values forseveral control variables, with privatized countries experienc-ing a higher mean political risk rating (implying less risk), alower mean level of trade openness, a higher mean level ofGDP growth, a higher mean GDP per capita, a lower meanlevel of inflation, a higher number of per-capita telephonelines, and a higher level of net capital inflows. Furthermore,

privatized countries have a lower mean payroll tax rate anda lower pensioner-to-worker ratio. Both of these differenceslikely result from the tax reductions and retirement age in-creases that often accompany privatization.

In sum, a coarse look at our data provides some preliminaryconfirmation of our conjecture in that privatization leads toincreased FDI flows, but more importantly, motivates a morecareful econometric analysis of the data, a task to which wenow turn.

4. ECONOMETRIC MODEL

We examine the impact of pension privatization on FDI byestimating the following reduce form equation

Y it ¼ b1P it þ b2X it þ ui þ kt þ eit ð1Þwhere the dependent variable is net FDI inflows as a percent-age of GDP—as commonly done (Asiedu, 2002). As a net fig-ure, the dependent variable may be negative; it is possible forde-investment outflows to outweigh investment inflows in a gi-ven year. However, this is not frequently the case in our data-set; the overwhelming majority of the figures carry positivevalues, indicating net inflows of FDI in most years. Pit is ameasure of pension privatization for country i in year t, Xit

is a vector of covariates anticipated to explain FDI flows, b1

and b2 are model parameters, eit is an error term, ui representscountry-invariant effects, and kt represents time-specific effects.We capture the effects of privatization in two ways. First, weinclude a binary variable (Private) set equal to one for a givenobservation if a country has implemented some form of pen-sion privatization by the given year of the observation. Sec-ond, we use the degree of privatization (Intensity), whichallows us to examine whether the relationship between pensionreform and FDI, if any, is subject to the degree of privatiza-tion enacted. This second variable measures the projected per-centage of pension income that will be derived from theaverage pensioner’s private account following the implementa-tion of reform. Because pension reform effects may wane overthe course of time, we attempt to distinguish between short-term and mid-to-long-term effects of privatization. This isdone by including a variable measuring the number of yearsthat have elapsed since the implementation of reform (Age).

We have proposed that beyond its direct signaling effect,privatization may impact FDI indirectly through its effectson financial development and payroll taxes; we explore thesetwo channels in our analysis. Many studies use bank creditor stock market capitalization to measure financial develop-ment (Durham, 2003). However, large numbers of missing val-ues prevent us from using either of these indicators. Instead, tomeasure financial development, we use net inflows of portfolioinvestment as a percentage of GDP (Portfolio), based on theintuition that large inflows of portfolio investment reflect awell-developed capital market. We measure payroll taxes asthe sum of social security payroll tax rates for employersand employees (Payroll Tax). 15 We expect the relationship be-tween payroll tax and FDI to be negative, and the relationshipbetween portfolio and FDI to be positive.

In addition to the effects of pension reform, financial devel-opment, and payroll tax rates, we include several additionalcovariates whose selection is informed by a rich empirical lit-erature on FDI (see e.g., Amaya and Rowland, 2003; Asiedu,2002; Biswas, 2002; Chakrabarti, 2001). First, previous empir-ical work includes GDP per capita as an explanatory of FDIinflows; we do likewise. This variable serves as a proxy forboth economic development and market size (as GDP per

Table 2. Mean Values for privatized and non-privatized countries

Group Variable Mean value,non-privatized

Mean value,privatized

Significance test forequality of means

All countries FDI, net inflows (% of GDP) 3.071140 4.703888 .000Trade openness (imports + exports/GDP) 88.947965 71.146769 .000Political risk .342792 .425105 .000GDP growth (%) 1.9890888 4.7273739 .000GDP per capita 2741.578926 4051.195440 .000Proceeds ($ millions) .489687 .644513 .349Inflation (%) 1.874059 .080123 .000Telephone 16.963743 20.413705 .000Capital controls .531669 .566791 .247Portfolio (% GDP) .0017020 .0017259 .977Payroll tax (%) 22.38 17.94 .000P/W 17.12 14.76 .003Capital flows 3.14 4.88 .000

Privatizing countries FDI, net inflows (% of GDP) 2.928579 4.703888 .000(Pre and post-privatization) Trade openness (imports + exports/GDP) 84.912584 71.146769 .001

Political risk .384674 .425105 .026GDP growth (%) 1.169633 4.7273739 .000GDP per capita 2884.807645 4051.195440 .000Proceeds ($ millions) .527947 .644513 .497Inflation (%) 1.162543 .080123 .000Telephone 19.410680 20.413705 .357Capital accounts restrictions .547261 .566791 .603Portfolio (% GDP) .0021342 .0017259 .665Payroll tax rate (%) 24.32 17.94 .000P/W 18.91 14.76 .000Capital flows 3.01 4.88 .000

aThe first segment compares all country-year values for non-privatized systems, regardless of whether the country eventually privatized, to the post-privatization values for countries that have enacted reform.bSignificance tests do not assume equal variance.cAll variables obtained from the World Development Indicators database, with following exceptions Political Risk Rating (Henisz, 2002), Payroll Tax Rate(Social Security Programs Throughout the World, SSA).

IMPACT OF PENSION PRIVATIZATION ON FOREIGN DIRECT INVESTMENT 295

capita is highly correlated with GDP). 16 Empirical evidenceon the impact of per capita income is mixed; Schneider andFrey (1985), Tsai (1994), Lipsey (1999), Root and Ahmed(1979) and Amaya and Rowland (2003) find that a higherGDP per capita spurs FDI inflows—suggesting that foreigninvestors prefer larger, more developed economies—whileAsiedu (2002), Edwards (1990), and Jaspersen, Aylward, andKnox (2000) find the opposite. 17

Second, in addition to a high level of economic develop-ment, foreign investors stand to reap higher returns in a grow-ing economy. Amaya and Rowland (2003) and Edwards(1990) find a positive relationship between GDP growth andFDI. We, therefore, control for the effects of economic growthon FDI by including GDP growth as an annual percentage.

Third, researchers commonly use inflation as a proxy formacroeconomic stability (Amaya and Rowland, 2003; Asiedu,2002; Grosse, 1997), fiscal responsibility (Schneider & Frey,1985), and wage costs (Grosse, 1997). As high inflation ratesare associated with macro-economic instability, high budgetdeficits, and increasing wage costs, we expect inflation to benegatively correlated with FDI. We note, however, that empir-ical studies have generally not found inflation to be a signifi-cant predictor of FDI (Amaya and Rowland, 2003; Asiedu,2002; Grosse, 1997).

Fourth, a high level of integration into the global trade sys-tem makes for a more favorable investment climate andshould correlate positively with FDI. Past models measuretrade openness as the total value of imports and exports as apercentage of GDP, and we have done likewise. The relation-ship between FDI and openness has been found to be both

positive (Edwards, 1990) and insignificant (Amaya and Row-land, 2003; Wheeler & Mody, 1992). 18

Fifth, previous work has sought to control for political riskfactors, such as regime type, contract repudiation, and govern-ment corruption as determinants of FDI based on the hypoth-esis that investors heed political risk. For example, Biswas(2002) finds that countries with institutions that protect prop-erty rights are more able to attract FDI. In the same realm,Jensen (2003) finds that countries with democratic institutionsattract as much as 70% more FDI than countries without. Wecapture political risk using the Political Constraint Index(Henisz, 2002), which measures the degree to which politicalactors are constrained in their ability to enact policy change.Henisz argues that investors value policy stability, and, there-fore, prefer political structures characterized by constraints onpolicymakers. Indeed, Henisz (2002) finds that the PoliticalConstraint Index significantly predicts foreign investments ininfrastructure, whereas attempts to predict FDI using conven-tional measures of political risk have been less successful(Jaspersen et al., 2000; Kolstad & Villanger, 2008). The indexranges from 0 (which characterizes a risky investment climatewhere autocratic regimes may make policy changes at whim)to 1 (maximum stability due to strong legislative checks andbalances).

Sixth, pension privatization occurred during an era whenmany states were relaxing capital controls in an attempt to at-tract investment. Therefore, we control for capital accountsrestrictions (Capital Control) in order to differentiate the im-pact of pension privatization from that of financial liberaliza-tion. While several studies (Asiedu & Lien, 2004; Desai, Foley,

296 WORLD DEVELOPMENT

& Hines, 2006) find a negative relationship between capitalcontrols and FDI, others find an insignificant relationship(Noy & Vu, 2007). Furthermore, Monteil and Reinhart(1999) find that capital controls have no effect on overallinvestment flows, but shift their composition away fromshort-term investment and toward FDI, resulting in a positiverelationship between capital restrictions and FDI. This findingis congruent with the assertion that capital controls are direc-ted primarily toward short-term investment flows rather thanFDI, and are useful in reducing dangerous over-reliance onshort-term capital investments (Stiglitz, 2000). In light of thesemixed findings, we anticipate that the relationship betweenCapital Control and FDI may be positive, negative, or insig-nificant. We measure capital accounts restrictions using infor-mation from the IMF’s Annual Report on ExchangeAgreements and Exchange Restrictions, constructing an indexfrom 0 (least restrictive) to 1 (most restrictive) according to themethod described in Miniane (2004).

Seventh, privatization of state-owned enterprises (SOEs) is acentral tenet of economic liberalization and provides invest-ment opportunities for foreign investors. It also sends a signalto investors that the government is dedicated to market-friendly reform. Privatization of SOEs was prevalent in manyLatin American countries in the 1990s. Between 1990 and1997, privatization of SOEs in Latin America totaled $116.5billion, dwarfing similar privatization proceeds in Africa, Eur-ope, and East and Central Asia over the same period(UNESCAP, 2000). Many of the privatized companies wereacquired by foreign direct investors; for example, theUNESCAP report indicates that in 1997, $11.4 billion of the$62 billion of FDI into Latin America and the Caribbeanwas related to privatization of SOEs. Theoretically, Mukherjeeand Suetrong (2009) show that privatization spurs FDIinflows. We use the proceeds from privatization of SOEs—which we denote proceeds—to control for the impact ofprivatization of public assets on FDI. This variable isobtained from the World Bank’s privatization database. 19

Eighth, following Asiedu (2002) and Biswas (2002) and forlack of a better control, we include the number of telephonesper 100 inhabitants as a proxy of the availability of infrastruc-ture. Better infrastructure is expected to positively affect FDIas it facilitates the conduct of business operations. Finally,we include a dummy variable (equal to 1 for Latin Americaand 0 for transitional economies) to control for regional differ-ences in FDI flows.

Two issues arise in the econometrics. First, pension privati-zation may be thought of as a treatment, with the treatmentgroup comprising those countries that have implementedprivatization schemes. However, a country’s decision to pri-vatize is likely endogenous because there may be unobservedfactors simultaneously affecting both the privatization deci-sion and FDI flows. Political will is an example of an unob-served characteristic (Vreeland, 2002); a government with thediscipline and cohesiveness that is necessary to pass compre-hensive pension reform may also be more likely to success-fully attract FDI. Given the panel nature of the data,sample selection effects can be mitigated by including fixedeffects in the regressions in order to control for omittedtime-invariant and context-specific characteristics that maybe correlated with the decision to privatize (see e.g., Duffalo,2005; Sanyal & Menon, 2005). However, endogeneity mayalso emanate from the omission of time and country-varyingrandom factors that impact both the decision to privatize andFDI flows. We, therefore, use an instrumental variable ap-proach to account for the endogeneity of the privatizationvariables within our fixed effects framework. The first-stage

model predicts a country’s likelihood of implementing pen-sion privatization or intensity of privatization, dependingon the privatization variable used; the second-stage fixed ef-fects model includes both the privatization variable and theresidual from the first-stage regression to control for sampleselection bias (Hausman, 1978; Terza, Basu, & Rathouz,2008).

In specifying our first-stage model, we draw upon a numberof previous studies that explore the determinants of pensionprivatization using data and methods similar to our own(Brooks, 2005, 2007b; Brooks & James, 2001). Like the pres-ent study, these investigations explore the factors that predictboth the likelihood of structural pension reform and the de-gree of reform enacted. While it is conceivable that a countrythat implements pension privatization can reverse course, suchmove is extremely unlikely because of institutional inertia; sofar only Argentina has done so. 2 Therefore, followingprevious studies, our first stage estimation drops a country’sobservations from the sample the year following theimplementation of pension reform. This reflects a diminishing“risk” group (Brooks, 2007b). 20

We consider the following three time-varying variables asinstruments: the pensioner-to-worker ratio (P/W), net capitalinflows (portfolio and FDI) as a percentage of GDP (CapitalFlows), and the number of regional peers, specific to eitherLatin America or Eastern Europe and Central Asia, that haveenacted privatization (Peers). As the ratio of pensioners-to-workers grows, the transitional cost of enacting privatizationbecomes increasingly prohibitive (Brooks, 2007b). Indeed,Brooks (2007a) finds that states enact privatization wherebudgetary challenges are surmountable and existing obliga-tions can be met in the wake of reform. Thus, as a country’spopulation ages and obligations to existing pensionersincrease, the state should grow less receptive to pensionprivatization. We proxy the pensioner-to-worker ratio by thenumber of individuals over statutory retirement age to thenumber of working-age individuals. 21 The variable P/W isexpected to carry a negative coefficient in the first-stage model,indicating that an increase in the old-age dependency ratiodecreases the likelihood/intensity of privatization.

Moreover, a growing body of political science literature oninstitutional reform advocates the importance of peer adop-tion as a means of policy diffusion across regions (e.g., Brooks,2005). For a given state, as the number of states in the sameregion that have implemented a certain policy increases, theuncertainty associated with the policy decreases, and the statebecomes more receptive to the policy (Brooks, 2005). OurPeers variable measures the number of peer countries thathave enacted some form of privatization by the given yearof the observation. We expect it to carry a positive coefficientper the discussion above.

Finally, we include lagged net foreign capital inflows(Capital Flows) as a first-stage predictor. We do so on thegrounds that states with healthy capital inflows in the pastmay anticipate easy access to financing in the future, therebyfacilitating transitional costs. We, therefore, expect thatCapital Flows will positively predict privatization.

A second econometric issue is that the causal links betweenpension reform and several of our remaining regressors suchas non-pension privatizations and trade openness may bebi-directional. To circumvent potential reverse causality bias,we lag all of our regressors. Implementing the Wooldridge testfor autocorrelation in panels (Drukker, 2003; Wooldridge,2002), we fail to reject the null of no autocorrelation in ourmodels at all conventional levels. Consequently, we implementthe Prais–Winsten procedure to correct the autocorrelation.

IMPACT OF PENSION PRIVATIZATION ON FOREIGN DIRECT INVESTMENT 297

Standard errors are robust to heteroskedasticity and cross-sec-tional correlation (Beck & Katz, 1995).

5. REGRESSION RESULTS

Our first-stage data (time series cross section data with abinary dependent variable (BTSCS)) are grouped durationdata with one year intervals. Beck, Katz, and Tucker (1998)have shown that by adding a series of temporal dummies ora smooth function of time (they suggest splines) to ordinaryLogit estimation of BTSCS data, the coefficient estimates ob-tained are identical to estimates of the Cox proportional haz-ard model. They also argue that a Probit model that includes asmooth function of time (as in Brooks, 2007b) is also anappropriate alternative for BTSCS data. For several reasons,we opt for linear probability and OLS models with countryfixed effects to explain a country’s decision to privatize its pen-sion system and the intensity of privatization. 22 FollowingBrooks (2007b) and Box-Steffensmeier and Jones (2004), weadd the log of time to capture duration dependence, if any,in the likelihood of implementing pension privatization.

Robust standard errors are reported in parentheses. In bothmodels, two of the three instruments (P/W and Peers) are sta-tistically significant predictors of the decision to privatize.Capital Flows is statistically significant with the predicted sign

Table 3. Linear probability and OLS models of pension privatization withfixed effects

Variables (Model 1) (Model 2)

Dependent variable:private

Dependent variable:intensity

P/W �0.0146*** �0.383*

(0.00395) (0.222)Peers 0.0295*** 1.157***

(0.00471) (0.333)Capital flows 0.00310 0.334***

(0.00200) (0.105)Trade openness 0.00107*** 0.0218*

(0.000278) (0.0117)Political risk 0.0418 �0.710

(0.0307) (2.037)GDP growth �1.11e-05 �0.121**

(0.000637) (0.0555)GDP per capita 6.59e-05*** 0.00165*

(1.32e-05) (0.000994)Proceeds �0.00174 �0.0372

(0.00239) (0.134)Inflation 4.49e-05 0.00393

(0.000432) (0.0247)Capital control 0.503*** 51.21***

(0.0452) (2.805)Latin America �0.00554 15.69***

(0.151) (5.633)Log (time) 0.00244 3.104**

(0.0192) (1.415)Constant �0.555*** �51.34***

(0.122) (4.513)Observations 490 490Adjusted R2 0.379 0.421Cross-sections 41 41

Robust standard errors in parentheses.* p < 0.1.

** p < 0.05.*** p < 0.01.

in the second model. Both GDP per capita and trade opennesscorrelate positively and significantly with the likelihood ofpension reform, indicating that higher levels of economicdevelopment/market size and economic integration imply agreater openness to the preferences of global economic actors,and, therefore, a greater likelihood of enacting market-friendlyreform. GDP growth carries a negative and significant sign inmodel 2, suggesting that countries with higher growth rates areless inclined to implement significant pension reform. Thismay be because economic growth impels higher levels of fiscalstability, reducing both the budgetary challenges associatedwith the PAYGO system and the incentives for reforming it.Our results also suggest that stronger capital controls increasethe likelihood of privatization; this finding is consistent withthe argument that pension privatization may be driven bythe need to increase domestic savings so as to reduce the reli-ance on volatile foreign capital (Vittas, 2000). Finally, the La-tin America dummy variable indicates a regional effect onpolicy intensity, increasing the expected value of policy inten-sity by 15.7 points. 23

In Tables 4 and 5, we present the results of six variants ofour reduced-form equation in order to (i) account for the end-ogeneity of pension reform, (ii) ascertain the robustness of ourresults, (iii) explore the mechanisms via which privatization

Table 4. Estimation of the determinants of FDI with time and country fixedeffects

Variables Model 1 Model 2 Model 3

Private 1.686*** 1.714*** 1.673***

(0.495) (0.494) (0.490)Age �0.0645 �0.0739 �0.0631

(0.0813) (0.0805) (0.0797)Trade openness 0.0177** 0.0176** 0.0172**

(0.00695) (0.00697) (0.00695)Political risk 0.157 0.208 0.241

(0.939) (0.942) (0.936)GDP growth 0.0595*** 0.0596*** 0.0555***

(0.0203) (0.0203) (0.0200)GDP per capita �0.000214* �0.000213* �0.000197

(0.000125) (0.000126) (0.000121)Proceeds 0.204*** 0.205*** 0.205***

(0.0342) (0.0340) (0.0343)Inflation �0.0194* �0.0193 �0.0203*

(0.0117) (0.0117) (0.0118)Telephone 0.0659 0.0686 0.0648

(0.0444) (0.0445) (0.0433)Capital Control �1.074 �1.067 �0.948

(0.785) (0.783) (0.795)Portfolio 0.000946 0.00103

(0.00203) (0.00206)Payroll tax 0.0275 0.0283

(0.0223) (0.0224)Latin America 0.140 0.163 0.0574

(0.985) (0.981) (0.962)Residual 0.111 0.137 0.145

(0.486) (0.483) (0.484)Constant 1.874 1.798 2.437**

(1.459) (1.460) (1.240)Observations 630 630 630Adjusted R2 0.453 0.459 0.450Cross-sections 42 42 42

Robust standard errors in parentheses.* p < 0.1.

** p < 0.05.*** p < 0.01.

298 WORLD DEVELOPMENT

affects FDI, and (iv) test whether FDI flows are related notonly to the decision to privatize, but also to the degree of priv-atization enacted. All models include time- and country-fixedeffects. The only difference between Tables 4 and 5 results isthat Table 4 results are obtained using the privatization dum-my to capture the effects of pension reform, while Table 5 re-sults measure privatization using the intensity variable. In allmodels, the residuals of the first-stage regression (models 1and 2) are included in our second-stage equations (residualsfrom model 1 of Table 3 are used in Table 4; residuals frommodel 2 of Table 3 are used in Table 5) in order to accountfor potential endogeneity bias of pension reform not capturedby the country and time dummies. In all of our estimations,the Wald test strongly rejects the null that all slope coefficientsare equal to zero with p-values close to zero.

We hypothesized in Section 2(b) that, beyond its signalingeffect, privatization may positively affect FDI via its effectson financial development and payroll taxes. To explore thesetwo channels, we proceed as follows. First, we add both finan-cial development and payroll tax rates as covariates in our sec-ond-stage Eqn. (1) which removes the possibility that the effectof privatization operates via changes in these variables (model1). We then remove, alternately, the payroll tax variable (mod-

Table 5. Estimation of the determinants of FDI with time and country fixedeffects

Variables Model 1 Model 2 Model 3

Intensity 0.0188*** 0.0192*** 0.0184***

(0.00599) (0.00596) (0.00589)Age �0.0388 �0.0492 �0.0366

(0.0769) (0.0755) (0.0757)Trade openness 0.0183*** 0.0181*** 0.0178***

(0.00683) (0.00686) (0.00684)Political risk 0.343 0.387 0.367

(0.941) (0.943) (0.945)GDP growth 0.0591*** 0.0592*** 0.0555***

(0.0207) (0.0207) (0.0204)GDP per capita �0.000138 �0.000137 �0.000125

(0.000116) (0.000117) (0.000115)Proceeds 0.200*** 0.201*** 0.201***

(0.0338) (0.0336) (0.0338)Inflation �0.0205* �0.0204* �0.0213*

(0.0119) (0.0119) (0.0120)Telephone 0.0757* 0.0786* 0.0720*

(0.0423) (0.0425) (0.0416)Capital control �1.052 �1.036 �0.929

(0.769) (0.768) (0.783)Portfolio 0.00113 0.00119

(0.00209) (0.00212)payroll tax 0.0246 0.0255

(0.0223) (0.0225)Latin America 0.104 0.121 �0.0190

(0.957) (0.954) (0.946)Residual 0.0102 0.0104 0.0103

(0.00657) (0.00652) (0.00652)(0.831) (0.833) (0.847)

Constant 1.528 1.464 2.165*

(1.416) (1.420) (1.189)Observations 630 630 630Adjusted R2 0.453 0.457 0.450Cross-sections 42 42 42

Robust standard errors in parentheses.* p < 0.1.

**p < 0.05.*** p < 0.01.

el 2) and the financial development variable (model 3) frommodel 1 to examine how omitting each of these variables af-fects the size of the pension privatization coefficient. If, forexample, the coefficient of privatization increases significantlywithout the inclusion of the payroll tax variable that wouldsuggest that the positive effect of pension reform operates inpart through the payroll tax channel.

Several important findings emerge from Tables 4 and 5. Se-ven of the thirteen control variables carry significant coeffi-cients: the pension privatization variable, Trade Openness,GDP Growth, Proceeds and Inflation; GDP per Capita is onlysignificant, in Table 4 while Telephone is only significant inTable 5. The coefficient signs for trade openness and GDPgrowth are positive, and the signs for inflation are negative,indicating that investors prefer stable, open, growing econo-mies. In Table 4, there is a negative and significant relationshipbetween GDP per capita and FDI, perhaps due to the lowerrate of return on capital in higher income countries (Asiedu,2002). The coefficient on Proceeds is also positive and highlysignificant as expected (see discussion above). The coefficienton the first-stage residual is statistically insignificant in allmodels, suggesting that endogeneity of the decision to privat-ize may be driven by country-fixed effects which are alreadycontrolled for.

Turning to the effects of our key variable, our results indi-cate that pension reform has a highly statistically significantand positive effect on net FDI inflows in all three modelsand in both Tables 4 and 5. However, the coefficient doesnot change significantly whether we control for or omit thePayroll Tax and Portfolio variables; hence, the empirical anal-ysis does not lend support to the hypothesis that the effect ofprivatization operates via financial development and the taxchannels. Consequently, we focus on the results from model1 for the ensuing analysis. In Table 4, the privatization effectis quite robust and economically impressive; on average, priv-atization increases the share of FDI as a percentage of GDPby an additional 1.68% (per model 1). Given the mean ofthe FDI/GDP for the “treatment” countries before privatiza-tion of 2.93%, the marginal effect implies that pension reformincreased FDI inflows as a share of GDP by 57.4%. 24 Theseempirical findings reinforce our coarse statistics presented inTable 2, which indicate that pension reform is associated withan increase in FDI inflows of 60% on average. 25 The Age var-iable is not significant at any conventional level, indicatingthat the estimated positive effect of privatization on FDI didnot systematically wane or strengthen over time.

Foreign investors may not only respond to pension reform,but also to the intensity of privatization, defined as the post-reform percentage of an average pensioner’s income thatderives from her private account. Of the 21 countries that priv-atized their pension systems, only six opted for privatizationschemes whereby the entirety of a pensioner’s income was re-placed by the private system (see Table 1). The remainingcountries maintain some element of public funding to supple-ment income from the private accounts. We examine the rela-tionship between the degree of privatization (Intensity) andFDI in Table 5. In all three models, the degree of privatizationis significant and positive, indicating that more comprehensiveprivatization schemes are more effective in boosting FDI in-flows. Given the average intensity of privatization in our sam-ple of 70%, privatization is estimated to boost net FDI inflowsby an additional 1.32% (per model 1). 26 This effect translatesto a proportional marginal impact of 45%. This effect is lowerthan the effect found in Table 4, where we did not account forthe degree of privatization, and suggests that the intensity ofreform matters beyond the mere decision to enact some form

IMPACT OF PENSION PRIVATIZATION ON FOREIGN DIRECT INVESTMENT 299

of privatization. The results mirror those in Table 4 for ourremaining controls both in statistical significance and sign ofcoefficients, with the exception of GDP Per Capita (which goesfrom marginally significant in Table 4 to insignificant inTable 5) and Telephone (which becomes significant at 10%level with the expected sign).

Our failure to find evidence for the financial developmentand taxation mechanisms suggest that the estimated effect ofpension privatization on FDI lies primarily in its direct impactas a policy signal rather than its potential indirect impact viatax reduction or financial development.

Finally, several of the covariates in the second-stage equa-tion differ statistically between privatizing and non-privatizingcountries; for example, GDP growth is almost 300 basis pointshigher for privatizing countries on average. It is possible thatpension privatization accelerates GDP growth, which in turnpositively impacts FDI inflows. If so, the true effect of pensionreform is underestimated. To explore this possibility, we re-estimated model 1 of Table 4 but omitting all control variablesthat are statistically different between privatizing and non-privatizing countries as per Table 2. Doing so, we find thatthe coefficient of the privatization dummy drops significantlyto 1.45, which translates to a proportional marginal effect of49%. However, because these omitted variables could also af-fect net FDI inflows for a variety of reasons unrelated to pen-sion privatization, their omission may lead to biased andinconsistent estimates.

6. CONCLUSION

Our econometric results provide robust evidence that pen-sion privatization led to a sizable and statistically significantboost in FDI flows in Latin America and among transitionaleconomies of Eastern Europe and Central Asia. The empiricalresults lend support to the hypothesis that privatization acts asa policy signal: investors appear to respond to the decision it-self, rather than to its potentially favorable economic and fis-cal concomitants, which take time to materialize. Indeed, wefind no evidence that privatization attracts FDI by developingcapital markets or by reducing payroll tax rates, althoughthese mechanisms may become influential over a longer timehorizon and should be revisited in the future. Our results arealso congruent with Brook’s finding that countries do not en-act privatization when doing so would substantially increasethe risk of capital flight. Finally, the results suggest that the de-gree of privatization matters—more comprehensive privatiza-tion schemes appear more effective in boosting FDI inflows.

We expect that these results will be relevant to two broaderdiscussions. First, policymakers who confront the problem ofpension reform may consider the potentially positive relation-ship between privatization and FDI inflows as an additionalincentive to consider the measure. Second, our findings suggestthat credible policy decisions matter to foreign investors—afinding with important implications for the ongoing academicdiscussion around the determinants of FDI.

NOTES

1. For example, it was estimated that for the Argentinean PAYGOsystem to pay the benefit rates of 1995 at the legal 70% replacement rate, itwould require an increase of the participant workforce by 159% (Kay,2000).

2. Unlike Chile, which closed its public system, these three countriesimplemented multi-pillar reform, whereby public schemes were noteliminated but reduced in size to function alongside new private programs.Furthermore, Argentina announced plans in 2008 for the government tore-nationalize the $25 billion that had been placed in privatized pensionfunds, transferring these back into the public system.

3. The model’s three pillars comprise a public component to provide abasic pension (the public pillar), a mandated individual savings programto provide supplementary pension income (the private pillar), and avoluntary individual savings program for those who wish to contributebeyond the mandated rate (See World Bank, 1994 for details).

4. Argentina, Bolivia, Chile, Columbia, Costa Rica, Dominican Repub-lic, El Salvador, Mexico, Peru, and Uruguay.

5. Kazakhstan, Poland, Latvia, Bulgaria, Croatia, Estonia, Russia,Kosovo, Lithuania, Slovak Republic, Macedonia, Romania, and Ukraine.

6. The transition period can be very long and costly; for example, Chile’stransition costs average about 6% of GDP over the 1981–1999 period andare expected to reach 4.3% of GDP from 1999 to 2037 (Devesa-Carpio &Vidal-Melia, 2002) in large part because of a minimum pension programfor individuals who, over 20 years or more of contribution, fail toaccumulate a state-determined minimum pension fund. In Argentina,transition costs were responsible for almost half of the increase in publicdebt between 1993 and 2000 (Matijascic & Kay, 2006).

7. Mesa-Lago and Muller (2002) identify three distinct models ofmarket-oriented pension reform. Under the substitutive model (under-

taken in Bolivia, Chile, the Dominican Republic, El Salvador, Kazakh-stan, and Mexico), the public system is closed to new entrants; all newworkers participate exclusively in the private system. Under the parallelmodel (undertaken in Columbia, Lithuania, and Peru), the public systemis reformed and remains open alongside the new private system; the twocompete with one another, and workers are allowed to choose betweenthem. The public system also remains open under the mixed or “multi-pillar” model (undertaken in Argentina, Bulgaria, Costa Rica, Croatia,Slovakia, and Uruguay), whereby pensioners draw a basic pension fromthe public system and a supplementary pension from the private system.

8. This figure pertains only to workers who will retire after the transitionto the privatized system is complete, several decades after the reform isinitially implemented.

9. Due to the incentives typically built into reform programs, thesimulation assumes that a worker will choose to opt into the privatesystem where possible, as has been generally observed. For example,approximately 96% of covered Peruvian workers have opted for theprivate pillar (Mesa-Lago, 2005).

10. As is the case in Mexico and Bolivia (Brooks, 2009).

11. Skeptics have challenged many claims about the purported macro-economic benefits of pension privatization. In particular, Orszag andStiglitz (2001) refute many of these alleged benefits which they call“myths”. Among their list of myths are the notions that (i) privatizationspurs higher national savings and that (ii) private pension accounts earna higher rate of return than the PAYGO system. Using data from 8 LatinAmerican countries that have adopted pension privatization, Mesa-Lago(2002) finds empirical support for many of Orszag and Stiglitz’sarguments. In the same realm, Arza (2008) indicates that pensionprivatization in Argentina has failed to achieve many of its keyobjectives.

300 WORLD DEVELOPMENT

12. This was the case in Bolivia, Chile, Bulgaria, Croatia, Hungary,Latvia, Mexico, Peru, Poland, and Uruguay.

13. This was the case in Bolivia, Bulgaria, Croatia, Hungary, Latvia,Poland, and Uruguay.

14. We are grateful to Sarah Brooks for generously providing us with herdataset on pension reform in Latin America.

15. Payroll taxes are obtained from the US Social Security Administra-tion’s Social Security Programs Throughout The World, a bi-annual report,accessible online at http://www.ssa.gov/policy/docs/progde-sc/ssptw/in-dex.html.

16. In our data, the correlation coefficient between GDP per capita andGDP is 0.57.

17. Asiedu (2002) argues that the return on capital is lower in high percapita income countries—where capital is abundant—than in less devel-oped countries which suggests an inverse relationship between GDP percapita and FDI.

18. The relationship between Trade openness and FDI could be negativeor positive depending on the type of FDI undertaken. Export-orientedforeign investors are likely more attracted to more open economies withfewer trade restrictions (lower transaction costs); if so a positive causallink between openness and FDI would emerge. On the other hand, in thepresence of significant trade barriers (low level of trade openness), market-seeking companies may have no choice but to invest in the host county inorder to obtain market share; if so, FDI and openness would be negativelycorrelated (Asiedu, 2002).

19. http://rru.worldbank.org/Privatization/.

20. Given that Argentina has re-nationalized its pension system and thatsimilar action is underway or under consideration in Hungary, Poland andBulgaria, we also estimated our first stage models by treating pensionprivatization as a reversible event, that is, not dropping countries from thedataset following implementation of privatization. See endnote 23 for asummary of our findings.

21. We construct a pensioner-to-worker ratio using demographic infor-mation as well as country-specific information on statutory retirementages (obtained from the Social Security Administration’s biannual reporton global pension programs). The ratio reflects the number of individualsabove the statutory retirement age relative to the number of working-ageindividuals. Thus, this measure varies according to both demographic fluxand changes in statutory retirement ages. However, due to data limita-tions, our pensioner-to-worker ratio contains a few significant weaknesses.First, the statutory retirement age does not necessarily reflect the effectiveretirement age, as many workers retire before they are eligible for state

pensions, and thus, do not contribute to PAYGO pension schemes at thesame rates as active members of the workforce as our ratio implies.Second, the ratio does not include information on the scope of coverage:not all individuals receive pensions nor contribute to public pensionschemes (i.e., informal workers and women who work in the home). Ourratio presumes coverage of all individuals who are working-or retirement-age when this is not the case, and cross-national variation in coverage mayimpact the pensioner-to-worker ratio in ways we do not capture. Finally,we do not include information on evasion rates: many informal workersavoid payroll taxes, but contribute long enough to qualify for a publicpension, placing additional strains on the PAYGO system. Theselimitations imply that in many cases, the true effect of pensioner-to-worker may be higher than our estimate.

22. We choose for the linear probability model (LPM) rather than aProbit model to generate the residual for the second stage for threereasons. First, the Probit does not lend itself well to the inclusion of fixedeffects because such effects cannot be removed (Greene, 2000). Addingfixed effects in the first stage is important in our analysis since we do so inthe second stage regressions to control for year and country-invariantomitted factors that may be correlated with our regressors. Second, if aProbit model is used, consistency of the second stage parameters hinges onthe correctness of the assumed normal distribution for the first-stage error(Angrist & Krueger, 2001); distributional misspecification is not a concernwith the LPM. Third, Probit cannot be used to model the Intensity ofpension reform, a continuous variable (our second model in Table 3).

23. We re-estimated the first stage models without dropping countriesthat have enacted pension reform. Doing so increases the sample size from490 observations to 630. The results (available from the authors uponrequest) are identical in terms of signs and statistical significance to thosein Table 3 with two exceptions: with the full sample, the variables GDPGrowth and Proceeds are negative and statistically significant in bothmodels. The negative sign on Proceeds suggests that higher proceeds fromprivatization of SOEs lower the likelihood and intensity of privation:aggressive privatization of assets leads to better fiscal fortunes for the stateand, therefore, less urgency to reform the public pension system.

24. The proportional marginal impact is found by dividing the marginalimpact of privatization 1.68% (per model 1), by the 2.93% average of FDI/GDP � 100 for pension reformers before they enacted reform.

25. Pension privatization may also have spurred FDI through otherchannels that our analysis does not control for. For example, privatepension funds often purchased shares in privatized state-owned enter-prises. FDI then flowed into invest in these privatized companies.

26. The proportional marginal impact is found by dividing the marginalimpact of privatization, for example 0.0188 � 70% (per model 1), by the2.93% average of FDI/GDP � 100 for pension reformers before theyenacted reform.

REFERENCES

Albuquerque, R., Loayza, N., & Serven, L. (2005). World marketintegration through the lens of foreign direct investment. Journal ofInternational Economics, 66(2), 267–295.

Alfaro, L., Chanda, A., Kalemli-Ozcan, S., & Sayek, S. (2004a). FDI andeconomic growth: The role of local financial markets. Journal ofInternational Economics, 64, 89–112.

Alfaro, L., Chanda, A., Kalemli-Ozcan, S., & Sayek, S. (2004b). FDI,productivity, and financial development. World Development, 32(1),111–135.

Alfaro, L., Kalemli-Ozcan, S., & Sayek, S. (2009). FDI, Productivity andFinancial Development. The World Economy, 32, 111–135.

Amaya G. C. A., & Rowland, P. (2003). Determinants of investment flowsinto emerging markets. Banco De La Republica, Borradores deEconomia No. 313.

Angrist, J., & Krueger, A. (2001). Instrumental variables and the searchfor identification: From supply and demand to natural experiments.Journal of Economic Perspectives, 15(4), 69–85.

Arza, C. (2008). The limits of pension privatization: Lessons fromargentine experience. World Development, 36(12), 2696–2712.

Asiedu, E. (2002). On the determinants of foreign direct investment todeveloping countries: Is Africa different?. World Development, 30(1),107–119.

IMPACT OF PENSION PRIVATIZATION ON FOREIGN DIRECT INVESTMENT 301

Asiedu, E., & Lien, D. (2004). Capital controls and foreign directinvestment. World Development, 32(33), 479–490.

Beck, N., & Katz, J. (1995). What to do (and not to do) with times-series—Cross-section data in comparative politics. American PoliticalScience Review, 89(3), 634–647.

Beck, N., Katz, J., & Tucker, R. (1998). Taking time seriously: Time-series-cross-section analysis with a binary dependent variable. Amer-ican Journal of Political Science, 42(2), 1260–1288.

Beck, T., & Levine, R. (2001). Stock markets, banks and growth:Correlation or causality? World Bank Policy Working Research Paper2670.

Biswas, R. (2002). Determinants of foreign direct investment. Review ofDevelopment Economics, 6(2), 494–504.

Blomstrom, M., & Kokko, A. (2003). The economics of foreign directinvestment incentives. NBER Working Paper 9489.

Borensztein, E., De Gregorio, J., & Lee, J. W. (1998). How does foreigndirect investment affect economic growth? Journal of InternationalEconomics, 45(1), 115–135.

Box-Steffensmeier, J. M., & Jones, B. S. (2004). Event history modeling: Aguide for social scientists. Cambridge: Cambridge University Press.

Brooks, S., & James, E. (2001). The political economy of structuralpension reform. In R. Holman, & J. E. Stiglitz (Eds.), New ideas aboutold age security: Toward sustainable pension systems in the twenty-firstcentury (pp. 133–170). Washington, DC: World Bank.

Brooks, S. M. (2005). Interdependent and domestic foundations of policychange: The diffusion of pension privatization around the world.International Studies Quarterly, 49(2), 273–294.

Brooks, S. M. (2007a). Globalization and pension reform in LatinAmerica. Latin American Politics & Society, 49(4), 31–62.

Brooks, S. M. (2007b). When does diffusion matter?. The Journal ofPolitics, 69(3), 701–715.

Brooks, S. M. (2009). Social protection and the market in Latin America:The transformation of social security institutions. Cambridge: Cam-bridge University Press.

Campos, N. F., & Kinoshata, Y. (2008). Foreign direct investment andstructural reforms: Evidence from Eastern Europe and Latin America.IMF Working Paper 08/26. Washington, DC: International MonetaryFund.

Catalan, M., Impavido, G., & Musalem, A. (2001). Contractual savings orstock market development: Which leads? Journal of Applied SocialScience Studies, (120), 445–487.

Catalan, M. (2004). Pension funds and corporate governance in develop-ing countries: What do we know and what do we need to know?Journal of Pension Economics and Finance, 2(3), 197–232.

Chakrabarti, A. (2001). The determinants of foreign direct investment:Sensitivity analysis of cross country regressions. KYKLOS, 54(1),89–114.

Cuevas, A., Gonzalez, M., Lombardo, D., & Lopez-Marmolejo, A.(2008). Pension privatization and country risk. International MonetaryFund Working Paper 08/195. Washington, DC: International MonetaryFund.

Desai, M. A., Foley, C. F., & Hines, J. R. (2006). Capital controls,liberalizations, and foreign direct investment. Review of FinancialStudies, 19(4), 1433–1460.

Devesa-Carpio, J., & Vidal-Melia, C. (2002). The reformed pension systemsin Latin America. World Bank Social Protection Discussion PaperSeries, No 0209.

Dion, M. (2009). Globalization, democracy, and Mexican Welfare, 1988–2006. Comparative Politics, 42(1), 63–82.

Drukker, David. M. (2003). Testing for serial correlation in linear panel-data models. The Stata Journal, 3(2), 168–177.

Duffalo, 2005. Esther. Empirical methods. http://web.mit.edu/14.771/www/emp_handout.pdf.

Durham, J.B. (2003). Foreign Portfolio Investment, Foreign Bank Lending,and Economic Growth. Board of Governors of the Federal ReserveSystem International Finance Discussion Papers No. 757.

Easson, A. J. (2004). Tax incentives for foreign direct investment. TheHague: Kluwer Law International.

Edwards, S. (1990). Capital flows, foreign direct investment and debt-equityswaps in developing countries. National Bureau of Economic Research,Working Paper 3497.

Feldstein, M. (1997). The case for privatization. Foreign Affairs.Gastanga, V. M., Nugent, J. B., & Pashamova, B. (1998). Host country

reforms and FDI inflows: How much difference do they make? WorldDevelopment, 26(7), 1299–1314.

Gill, I. S., Packard, T., & Yermo, J. (2005). Keeping the promise of socialsecurity in Latin America. Washington, DC: World Bank.

Greene, William H. (2000). Econometrics analysis (4th ed.). NJ: PrenticeHall Inc..

Grosse, R. (1997). Foreign direct investment in Latin America. In R. E.Grosse (Ed.), Generating savings for Latin American development(pp. 135–146). Coral Gables, Florida: North-South Center Press,University of Miami.

Henisz, W. J. (2002). The institutional environment for infrastructureinvestment. Industrial and Corporate Change, 11(2), 355–389.

Hausman, J. A. (1978). Specification tests in econometrics. Econometrica,46(6), 1251–1271.

Hermes, Niels, & Lensink, Robert (2003). Foreign direct investment,financial development and economic growth. Journal of DevelopmentStudies, 40(1), 142–163.

Holzmann, R. (1997). Pension reform, financial market development, andeconomic growth: Preliminary evidence from Chile. IMF Staff Papers,44(2), 149–178.

Holzmann, R., & Hinz, R. (2005). Old-Age income support in the 21stcentury. Washington: The World Bank.

Holzmann, R., Mackellar, L., & Repansek, J. (2009). Pension reform insoutheastern Europe: Linking labor and financial market reforms.Washington DC: The World Bank.

Jaspersen, F. Z., Aylward, A. H., & Knox, A. D. (2000). The effects of riskon private investment: Africa compared with other developing areas.In P. Collier, & C. Patillo (Eds.), Investment and risk in Africa(pp. 79–95). New York: St. Martin’s Press.

Jensen, N. (2003). Democratic governance and multinational corpora-tions: Political regimes and inflows of foreign direct investment.International Organization, 57, 587–616.

Kay, S. (2000). Recent changes in Latin American welfare states: Is theresocial dumping?. Journal of European Social Policy, 10(2), 185–203.

Kolstad, I., & Villanger, E. (2008). Determinants of foreign direct investmentin services. European Journal of Political Economy, 24, 518–533.

Levine, R., & Zervos, S. (1998). Stock markets, banks, and economicgrowth. American Economic Review, 88(3), 537–558.

Lipsey, R. E. (1999). The location and characteristics of US affiliates inAsia. NBER Working Paper 6876.

Lipsey, R. E. (2004). Home and host country effects of foreign directinvestment. In R. E. Baldwin, & L. A. Winters (Eds.), Challenges toGlobalization: Analyzing the Economics (pp. 333–382). Chicago:University of Chicago Press.

Madrid, R. (1999). The new logic of social security reform: Politics andpension privatization in Latin America. Unpublished Doctoral Disser-tation, Stanford University, Stanford.

Matijascic, M., & Kay, S. (2006). Social security at the crossroads:Toward effective pension reform in Latin America. International SocialSecurity Review, 59(1), 3–26.

Maxfield, S. (1997). Gatekeepers of growth: Central banking in developingcountries. Princeton, NJ: Princeton University Press.

Mesa-Lago, C. (2002). Myth and reality of pension reform: The LatinAmerican evidence. World Development, 30(8), 1309–1321.

Mesa-Lago, C., & Muller, K. (2002). The politics of pension reform inLatin America. Journal of Latin American Studies, 34(1), 687–715.

Mesa-Lago, C. (2005). Evaluation of a quarter century of structuralpension reform in Latin America. In C. A. Crabbe (Ed.), A quartercentury of pension reform in Latin America and the Caribbean: lessonslearned and the next steps (pp. 43–82). Washington, DC: InteramericanDevelopment Bank.

Miniane, J. (2004). A new set of measures on capital account restrictions.IMF Staff Papers, 51(2).

Monteil, P., & Reinhart, C. M. (1999). Do capital controls andmacroeconomic policies influence the volume and composition ofcapital flows? Evidence from the 1990s. Journal of International Moneyand Finance, 18(4), 619–635.

Mukherjee, Arijit, & Suetrong, Kullapat (2009). Privatization, strategicforeign direct investment and host-country welfare. European Eco-nomic Review, 53(7), 775–785.

Noy, I., & Vu, T. B. (2007). Capital account liberalization and foreigndirect investment. The North American Journal of Economics andFinance, 18(2), 175–194.

Orszag, P., & Stiglitz, J. E. (2001). Rethinking pension reform: Ten mythsabout social security systems. In R. Holman, & J. E. Stiglitz (Eds.),New ideas about old age security: Toward sustainable pension systems inthe twenty-first century (pp. 17–56). Washington, DC: World Bank.

302 WORLD DEVELOPMENT

Palacios, R. & Whitehouse, E. (1998). The role of choice in the transition toa funded pension system. World Bank Social Protection DiscussionPaper 9812.

Root, F. R., & Ahmed, A. A. (1979). Empirical determinants ofmanufacturing foreign direct investments. Economic Development andCultural Change, 27, 751–767.

Sanyal, P., & Menon, N. (2005). Labor disputes and the economics of firmgeography: A study of domestic investment in India. EconomicDevelopment and Cultural Change, 53, 825–854.

Schneider, F., & Frey, B. S. (1985). Economic and political determinantsof foreign direct investment. World Development, 13(2), 161–175.

Stiglitz, J. E. (2000). Capital market liberalization, economic growth, andinstability. World Development, 28(6), 1075–1086.

Singh, A. (1996). Pension reform, the stock market, capital formation andeconomic growth: A critical commentary on the World Bank’sproposals. International Social Security Review, 49(3), 21–43.

Terza, J. V., Basu, A., & Rathouz, P. J. (2008). Two-stage residualinclusion estimation: Addressing endogeneity in health econometricmodeling. Journal of Health Economics, 27(3), 531–543.

Tsai, P. L. (1994). Determinants of foreign direct investment and its impactof economic growth. Journal of Economic Development, 19(1), 137–164.

United Nations Economic and Social Commission for Asia and thePacific (UNESCAP) (2000). Interregional Cooperation in Trade andInvestment Asia-Latin America. Studies in Trade and InvestmentNo. 43.

Vittas, D. (2000). Pension reform and capital market development:‘Feasibility’ and ‘impact’ preconditions. World Bank Policy WorkingResearch Paper 2414.

Vreeland, J. R. (2002). The effects of IMF programs on labor. WorldDevelopment, 30(1), 121–139.

Wei, S. J. (2000). How taxing is corruption on international investors?Review of Economics and Statistics, 82(1), 1–11.

Wheeler, D., & Mody, A. (1992). International investment locationdecisions: The case for US firms. Journal of International Economics,33, 77–86.

Wooldridge, J. M. (2002). Econometric analysis of cross section and paneldata. Cambridge, MA: MIT Press.

World Bank. (1994). Averting the old age crisis: Policies to protect the oldand promote growth. New York: Oxford University Press.

Xu, B. (2000). Multinational enterprises, technology diffusion, and hostcountry productivity growth. Journal of Development Economics, 62(2),477–493.

.sciencedirect.com

Available online at www