Revisiting Price-based Controls on Capital Inflows in a “Sophisticated” Emerging Market

World Development Vol. 35, No. 8, pp. 1329–1340, 2007� 2007 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

doi:10.1016/j.worlddev.2007.04.001www.elsevier.com/locate/worlddev

Revisiting Price-based Controls on Capital Inflows

in a ‘‘Sophisticated’’ Emerging Market

ANTONIO C. DAVID *

University of Essex, UK

Summary. — Large and volatile capital flows pose serious challenges for macroeconomic manage-ment in developing countries. In this paper, we examine the price-based capital controls adopted byBrazil during the 1990s with the objective of facing some of those difficulties. We use data on capitalaccount transactions at a monthly frequency in order to evaluate the effects of those policies on netfixed-income capital flows and on total portfolio flows using GMM techniques. Our results suggestthat the controls could have been effective in reducing both types of flows.
� 2007 Elsevier Ltd. All rights reserved.
Key words — international capital flows, capital controls, Latin America, Brazil

* I thank CAPES (Brazilian Ministry of Education) and

the Cambridge Political Economy Society for financial

support during this research. Carmen Li, Edouard Ch-

alle, Gabriel Palma, Felipe Tamega and four anonymous

referees provided very useful comments. The usual cav-

eats apply.

1. INTRODUCTION

The excessive liquidity in international capi-tal markets during the 1990s posed several chal-lenges to policy makers in emerging economies.Among numerous issues, the links betweenfinancial volatility and short-term capital flows,even if the direction of causality between thetwo is not always clear-cut, were a particularsource of concern. In addition, problems asso-ciated with moral hazard, fragility in financialintermediaries’ and corporate balance sheets,asset price bubbles, consumption booms, andreal exchange rate overvaluation were inti-mately connected to the boom–bust cycle incapital flows to developing countries followingthe financial liberalization process.

In this context, it became evident that fullcapital account openness can be associated withincreased macroeconomic volatility and limita-tions in degrees of freedom for domestic policymakers, which prompted a certain number ofcountries to (re)-adopt capital account regula-tions. In fact, the experience of developing econ-omies with the adoption of capital controls (oninflows, outflows or both) in the last 15 years isdiverse and encompasses countries such as Bra-zil, Chile, China, Colombia, India, Malaysia,Taiwan, Singapore among others (see Ariyoshiet al., 2000; Epstein, Grabel, & Jomo, 2003).

Our interest at present lays on examining theimposition of price-based capital account man-

132

agement polices in Brazil since the early 1990s.The aims of those regulations included reduc-ing the costs of sterilization policies and theappreciation of the real exchange rate associ-ated with large capital inflows and increasingthe autonomy of domestic monetary policy.This paper intends to evaluate the effectivenessof those regulations in attaining some of thesegoals.

In particular, we will analyze monthly datacompiled by the Brazilian Central Bank on cap-ital account transactions in fixed-income securi-ties and more generally on portfolio flows. Toour knowledge this specific dataset has not beenused previously to evaluate the Brazilian expe-rience. Firstly, we will present a description ofthe macroeconomic environment in Brazil dur-ing the 1990s and the capital account regula-tions adopted. Subsequently, we will proceedto a brief critical review of the current literatureon capital controls in Brazil. Finally, we willanalyse the effects of the price-based capital ac-count policies on net fixed-income inflows and

9

1330 WORLD DEVELOPMENT

on total net portfolio flows using GMM tech-niques.

2. BACKGROUND AND CAPITALACCOUNT POLICIES

Brazil undertook the first significant mea-sures toward the liberalization of its capital ac-count in 1987 with the creation of investmentchannels that gave foreigners exemptions fromdomestic income tax on capital gains (the so-called Annexes I–III). In May 1991, an impor-tant resolution concerning the regulation ofportfolio inflows to the country was imple-mented (the so-called Annex 4 legislation),profits relating to those investments were freeof tax and income remitted abroad only hadto pay 15% of income tax. Throughout the1990s, the process of deregulation of both cap-ital inflows and outflows took momentum,deposits in foreign currency, international is-sues of commercial papers, profit remittancesrelated to FDI, ADRs and IDRs, issues ofcertificates of deposits by Brazilian banks inforeign currency among others financial trans-actions were liberalized.

In the late 1980s, Brazil was cut-off frominternational financial markets and faced a per-iod of high exchange rate volatility (particularlyafter the default on external debt). After a largedevaluation in September 1991, the Central

-20000

-15000

-10000

-5000

0

5000

10000

15000

20000

25000

30000

1Q:7891 1

4Q:789 1

Q:889

3891

Q:92

0991:

1Q 4Q:0991 1

3Q:199 9129:

2Q9913:Q

13991:

4Q 3Q:4991 1

2Q:599

Figure 1. Brazilian capital account 1987–2003 (in Millions

Bank chose to adjust the nominal exchange ratefollowing a PPP rule, therefore reducing ex-change rate instability and opening up a posi-tive interest rate differential with industrializedcountries; hence stimulating capital inflows. Infact, as shown in Figure 1, in the first quarterof 1992 Brazil was already experiencing largecapital flows, even in a period of instabilityassociated with high inflation. As far as thecomposition of those flows is concerned, oneshould note the predominance of portfolio in-flows until 1997 (see Figure 2). FDI onlystarted to become important from 1996 onward(essentially associated with the privatization ofpublic utilities) and showed resilience duringthe international financial crises of the late1990s, contrary to the other categories of capi-tal flows. Finally, it is notable that the residualcategory ‘‘other’’ experienced a significant de-crease in 1994, arguably linked to the difficul-ties experienced by the Brazilian bankingsector in that year, which culminated in a gov-ernment bail out program.

It is of crucial importance to emphasize thatthe return of capital flows to Brazil was accom-panied by an effort by Central Bank authoritiesto sterilize those inflows, which contributed toan increase in public debt. As shown by Palma(2004), sterilization policies were responsiblefor a significant part of the rapid increase indomestic public debt verified in the 1990s. Infact, the second half of the 1990s was marked

91

Q:691

6991:

4Q 7991:

3Q 2Q:8991 91

1Q:99 9919:Q

4002

:03Q 2Q:1002 2

1Q:200 2

Q:200

4002

Q:33

of US Dollars of 2000). Source: Brazilian Central Bank.

-60000

-40000

-20000

0

20000

40000

60000

1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

FDI

Portfo lio

Other

Figure 2. Composition of Capital Flows to Brazil 1987–2003 (Millions of US dollars of 2000). Source: Brazilian

Central Bank.

REVISITING PRICE-BASED CONTROLS ON CAPITAL INFLOWS 1331

by the conflict between the domestic objectiveof maintaining price stability and the externalobjective of maintaining an open capital ac-count in the context of a pegged exchange rateregime.

By 1993, the surge in foreign capital was al-ready provoking difficulties for the manage-ment of the exchange rate regime andmonetary policy. Hence, in November of thatyear the National Monetary Council (CMN),created a specific channel for fixed-incomeinvestments: the Foreign Capital Fixed YieldFund and a tax of 5% was imposed on capitalentering the country through this fund (thistax was denominated IOF, Imposto sobreOperacoes Financeiras). Furthermore, an en-trance tax of 3% was also imposed on financialloans in foreign currency.

According to the 1993 Brazilian CentralBank’s annual report, the main objective ofthe taxes consisted in reducing short-term in-flows toward fixed-income securities given theunwillingness of authorities to reduce interestrate differentials in order to ‘‘attenuate themonetary impact of the foreign sector withoutinterrupting the process of integration withinternational markets.’’ Given the expansionof capital inflows to Brazil, in March 1994 theIOF tax levied on foreign capital fixed-yieldfunds was extended to other portfolio inflows.

The changes on the taxes on capital inflowsthrough the 1990s are described in Appendix

A. One can infer that the Brazilian policy mak-ers seem to have attempted to pursue a counter-cyclical policy of controls on inflows, that isincreasing the taxes when inflows were highand reducing them in periods of scarcity. Forinstance, when Brazilian financial markets wereunder stress in the aftermath of the MexicanPesos crisis of December 1994, given the pres-sures associated with the ‘‘tequila effect,’’ theIOF tax on portfolio inflows was reduced inMarch 1995. Nevertheless, when the level ofcapital entering the country increased signifi-cantly in mid-1995, the authorities decided toincrease the tax on portfolio inflows.

Hence, Brazilian policy makers attempted toalleviate the tensions linked to large capital in-flows by compensatory measures such as thetaxes on portfolio flows, rather than decreasesin domestic interest rates. Indeed, the persis-tence of high real interest rates in Brazil to dateis a puzzle that spun a large literature empha-sizing a number of factors, in particular institu-tional ones (see for instance Arida et al., 2004;Favero & Giavazzi, 2002), but a full discussionof this puzzle lies beyond the scope of thispaper.

We may conclude that the taxes on portfolioinflows and foreign currency loans essentiallyintended to mitigate the negative effects associ-ated with the very high interest rates needed tosustain the nominal anchor and maintaindomestic price stability. With the end of the


quasi-fixed exchange rate regime in January1999 and the shift to a floating rate regime cou-pled with an inflation targeting system, the needfor sterilization policies was reduced and thecapital account regulations lost one the theirmajor rationales. According to Garcia and Val-passos (2000), throughout the 1990s some re-labeling of capital inflows in order to escapethe entrance tax occurred, but there is little evi-dence of how widespread the process was (seebelow for a more detailed discussion of issuesrelated to the evasion of capital controls).

3. A BRIEF SURVEY OF THEECONOMETRIC LITERATURE

The seminal study by Cardoso and Goldfajn(1997) created an index of capital controls withthe objective of analyzing the determinants ofcapital flows to Brazil. Changes in the capitalaccount regulations were used to build indica-tors of restrictions on capital inflows and out-flows. Those indexes are constructed such thata change in legislation that aims to reduce cap-ital inflows, receives a value of 1 and changesthat aim to attract capital inflows receive a va-lue of �1. For each month the changes in reg-ulations are added and a total value is used tobuild the index. The same holds for restrictionson capital outflows.

Those authors estimate OLS regressionsusing monthly data of the period from January1989 to December 1995, using total net capitalflows as the dependent variable. The coeffi-cients for the capital control indexes provedto be positive and significant, which is some-what counterintuitive and rejects the hypothesisthat capital controls reduced total net capitalflows. Evidently, this type of OLS regressionis plagued with problems and as the authorsthemselves acknowledge, capital controls arelikely to be endogenous, that is respond to cap-ital flows.

For these reasons, Cardoso and Goldfjanalso estimated VAR models and found thatshocks (i.e., a tightening) to the controls oncapital inflows do reduce total net capital flows,shocks to controls on capital outflows actuallyincrease net capital flows and shocks to thecomposite index (both controls on inflows andoutflows) have a negative effect on net capitalflows. As far as the composition of flows is con-cerned, shocks to the capital control measures

reduce debt flows and portfolio investment,but net FDI does not seem to be affected. Final-ly, capital controls seem to react only to portfo-lio flows, not responding to debt and FDIflows.

Soihet (2002) extends the Cardoso and Gold-fajn regressions until December 2000 and in-cludes a variety of other regressors such as acovered interest parity term. The results ob-tained are similar to the ones outlined above.Nevertheless, this author only considers univar-iate models, therefore not taking account of theendogeneity issues already discussed. Unfortu-nately, both studies fail to present completediagnostic statistics that would detect misspeci-fication of their OLS regressions and mightpossibly explain counterintuitive results suchas positive coefficients for the control on capitalinflows variable.

Goldfjan and Minella (2005) extend the anal-ysis of capital flows and capital controls in Bra-zil to a more recent period, restating the resultthat the controls on inflows affected the volumeand composition of flows, but were ineffectivein the long run. Nonetheless, their regressionsand VAR models for the determinants of capi-tal flows do not include a variable measuringdirectly the capital account regulations thatare of interest for our paper.

Garcia and Valpassos (2000) discuss theeffectiveness of the controls arguing that theprivate sector managed to circumvent the regu-lations due to the relative sophistication of cap-ital markets in Brazil (hence the title of ourpaper) and therefore the policies did not pres-ent lasting effects on net capital flows. Theirmain argument is that sophisticated marketscan easily evade regulations, rendering themineffective. In fact, one has to acknowledge thatdespite the imposition of controls, there was asignificant increase in domestic debt associatedwith the large sterilization effort of the Brazil-ian Central Bank. In fact, Palma (2004) esti-mates that sterilization costs amounted to 30Billion US Dollars (of 1998) in the period fromJuly 1994 to December 1998. This evidence im-plies that the taxes were arguably not applied inan optimal way, at least in terms of significantlyreducing sterilization costs. Nevertheless, aninteresting counterfactual question seems tobe what would have been the level of flows inthe absence of the taxes?

In a more recent paper, Carvalho and Garcia(2006) skilfully demonstrate a number of chan-


nels or strategies that could have been used bymarket players to evade the controls on capitalinflows during the 1990s. Their interviews withfinancial market agents revealed that thosestrategies involved mainly disguising short-terminvestment as FDI, labeling fixed-incomeinvestment as equity investment and usingderivatives to circumvent the controls. It isimpossible to deny that those channels existed,but what really matters for our purposes is theextent to which this evasion activity (which alsoinvolves costs) took place, that is whether thecontrols were binding or not. One could arguethat the Brazilian Central Bank should proba-bly have been more active in closing loopholesto optimize the effectiveness of controls, butthis does not necessarily mean that those regu-lations were irrelevant.

Carvalho and Garcia (2006) claim usingVAR models that the effectiveness of the con-trols lasted for 6 months or so and those poli-cies would have been ineffective thereafter,hence concluding that the controls were ineffec-tive in the long run. Nonetheless, the impulseresponse functions estimated were not signifi-cant in most cases (as pointed out by theauthors themselves). In addition, they use anindex rather than the tax rate itself to measureimpact of the capital controls.

To sum up, a re-evaluation of the impact ofprice-based capital controls in Brazil seemswarranted in the light of the mixed conclusionsreached by the rather sparse previous researchefforts. In addition, a number of importantshortcomings as far as econometric techniquesare concerned still seem to be present, particu-larly in reference to endogeneity and Unit Rootissues.

4. EMPIRICAL MODELS

The data used for the estimation of our mod-els of capital flows to Brazil was mainly ob-tained at the Brazilian Central Bank websiteand at the online dataset maintained by theInstitute for Applied Economic Research(IPEA). The cyclical component of the real ex-change rate was obtained by estimating a struc-tural time series model for the real exchangerate series (through Kalman filtering tech-niques). The definitions and sources for thevariables used in the regressions are outlinedin Appendix B.

The monthly data on capital flows to Brazilapplied in this section is of special interest. Itis a series compiled by the Central Bank usingbalance of payments data that allows us to ana-lyze at a relatively disaggregated level the capi-tal flows entering the country. In particular, itpermits the separate consideration of netfixed-income capital inflows, which were a spe-cific target of the capital account policies. Asfar as we know this has not been done in the lit-erature before. We will also consider the moregeneral effect of the policies on net portfolioflows and on total capital flows.

Unit root tests for the series used are pre-sented in Table 1. All the series seem to be sta-tionary, with the possible exception of theinterest differential series that was included infirst differences (see justifications below). Lagselection was based on the Schwartz informa-tion criterion. We chose to follow the sequen-tial testing strategy proposed by Perron (1988)when modeling deterministic components(trends and intercepts). All the variables werestationary in first-differences, that is there wereno I(2) variables.

The theoretical foundations of the regres-sions can be found in the models of capitalflows to developing countries surveyed in Fitz-gerald (2002) and Mody and Taylor (2002 and2003). We included in our models variablesidentified as important in the traditional litera-ture on the determinants of capital flows focus-ing on push and pull factors (see Agenor &Montiel, 1999), such as domestic and foreigninterest rates and the country risk premium.

In addition, we tried to capture the importantrole of information asymmetries in determiningcapital flows, as emphasized by Mody and Tay-lor (2002), by including the US Junk bondspread in the regressions. In fact, those authorsargue that the high yield spread is important inexplaining the supply of flows to emerging mar-kets, partly because default risk in the US couldbe correlated to default risk in emerging mar-kets, but also because the empirical evidenceshows that a rise in the high-yield spread iscapable of predicting a slowdown in real eco-nomic activity, which is consistent with finan-cial accelerator models (see Gertler & Lown,1999). This predicted slowdown in economicactivity could lead to a reduction in net capitalflows. Moreover, and from the viewpoint of mi-cro level portfolio choice, Fitzgerald (2002) ar-gues that the high-yield spread may capture

Table 1. Unit-root tests for selected variables

ADF-GLS test statistic Ng–Perron test statistic

Net portfolio flows �3.312*** �2.665***

Net total flows �5.614*** �3.354***

Net fixed-income flows �4.763*** �3.175***

Interest differential �2.036** �1.093*

Real exchange rate gap �4.180*** �4.152 ***

World oil prices �2.755*** �2.847***

High-yield spread �2.198** �2.205**

Trade balance �3.493 *** �2.691 ***

* Denotes significance at the 10% level.** Denotes significance at the 5% level.*** Denotes significance at the 1% level.


endogenous cycles in foreign investor’s risk ten-dencies.

We also introduce lags of capital flows as anexplanatory variable with the objective of cap-turing herding, inertia or bandwagon effectsthat may arise in international capital marketsin the face of asymmetric information andbounded rationality. It is clear that endogeneityproblems are very likely to arise in the estima-tion of equations describing capital flows. Inaddition, right-hand-side variables are alsomeasured with error. In this context, it is likelythat the residuals of the regression will be cor-related with the explanatory variables, whichimplies that OLS estimates would be biasedand inconsistent.

We will tackle this problem by estimatingregressions using the Generalized Method ofMoments (GMM) technique. In order to applythis technique, it is necessary to impose orthog-onality conditions between the explanatoryvariables and the error term. In our case theorthogonality conditions will consist of a setof instruments, let us call them Zt, that are as-sumed to be orthogonal to the error term in ourmodels for capital flows (Ut), such thatE[ZtUt] = 0 (see Favero, 2001). Note thatGMM is a robust estimator and does not re-quire assumptions about the exact distributionof data generating process (unlike MaximumLikelihood estimators such as two-stage-least-squares).

When implementing our GMM models wecorrect for heteroscedasticity and autocorrela-tion of unknown form by using Barlett weightsto ensure that the variance-covariance matrixestimated is positive definite. In addition, we

use the J-statistic obtained from the regressionsto test for the validity of the overidentifyingrestrictions imposed, in order to ensure thatthe instruments used are adequate.

We will try to evaluate the effect of the taxesby estimating log-linear models for capitalflows as a function of the interest rate differen-tial, the junk bond spread and the tax rates.The models are estimated over the period goingfrom January 1995 to December 1998. Thesample period was chosen because of dataavailability (the capital inflows series only be-comes available at a monthly frequency fromJanuary 1995) and to ensure that a single policyregime was included in the estimation in orderto avoid structural breaks. One should bear inmind that the Real stabilization plan wasimplemented in mid-1994 and the domestic cur-rency suffered a severe crisis in January 1999,forcing the government to abandon the ex-change rate peg.

Furthermore, following a strand of the litera-ture on capital flows to emerging markets (seeAgenor et al., 2002); we include the interest dif-ferential, the junk bond spread and the tax rateseries in first differences. We justify this choicefor two reasons. The first one relates to theneed to get rid of unit root problems. In addi-tion, since we are dealing with capital flows; itis intuitive that changes in those series are morerelevant than simply looking at the levels.

The results obtained from estimating aGMM regression for the log of net fixed-in-come capital inflows (mainly bonds) are pre-sented in Table 2. We included lagged capitalflows, the first difference of interest rate differ-ential, the first difference of the tax on inflows

Table 2. GMM estimation of net fixed-income capital flows to Brazil (January 1995 to December 1998)

Variable Coefficient [P-values] R2 J-test Normality

C 0.686 [0.00] 0.110 0.209 [0.96] 0.345 [0.84](Capital flows)t�1 0.358 [0.00]Interest differential 4.296 [0.00]Tax (IOF) �16.765 [0.03]High-yield spread �62.988 [0.00]

J-statistic refers to the test for the validity of instruments included for the estimation of the model. Normality refersto the Jarque–Bera test for the normality of the residuals in the regression. P-values for all the test statistics arepresented in brackets.


and the first difference of the junk bond spreadas explanatory variables. In addition, we choseto include as instruments, lags of capital flows,lags of the interest differential, the world oilprices, lags of the high-yield spread, lags ofthe trade balance and lags of the real exchangerate gap (deviations from trend).

The diagnostic statistics confirm the ade-quacy of the model, there is no evidence of non-normality of the residuals and the J-statisticdoes not reject the validity of the instrumentsused. All the coefficients are statistically signif-icant and have the expected sign. The coeffi-cient of the tax on capital inflows is indeednegative and significant at the 3% level. Sincethe model is estimated in logs, the coefficientscan be interpreted as elasticities, hence the esti-mated elasticity of fixed-income flows to the taxis �16.8. One should also note that a one-offshock to the first difference of the tax rate isequal to a permanent shock to the level of thetax rate.

The results indicate that the taxes were effec-tive in reducing fixed-income flows despite theclaims made in part of the literature that theywere widely circumvented (Garcia & Valpassos,2000). Hence, our results are more in line with

Table 3. GMM estimation of net portfolio capital fl

Variable Coefficient [P-values]

C 0.266 [0.00](Capital flows)t�1 0.429 [0.00]Interest differential 4.073 [0.00]Tax (IOF) �6.967 [0.07]High-yield spread �37.755 [0.00]

J-statistic refers to the test for the validity of instruments into the Jarque–Bera test for the normality of the residualspresented in brackets.

the ones obtained by Cardoso and Goldfajn(1997). Arguably the use of data on net fixed-income inflows, rather than total flows mightexplain the different conclusion obtained here.

We will now relax our focus on fixed-incomeflows and analyze more generally the effect ofthe taxes on total net portfolio capital flows,in order to address the possibility that someform of re-labeling to circumvent the regula-tions occurred. We consider the same sampleperiod as before for the reasons already out-lined. We included the lagged portfolio flows,the tax, the interest differential, and the junkbond spread as explanatory variables. The setof instruments includes the lagged explanatoryvariables, the trade balance, world oil prices,and the real exchange rate gap. The estimationoutput is presented in Table 3.

The R2 suggests a good fit, but there is evi-dence of nonnormality of the residuals accord-ing to the Jarque–Bera test. Nonetheless this isnot very preoccupying since the properties ofGMM estimators are asymptotic and do notrely on an underlying normality assumptionto obtain asymptotic normality of the estima-tor. The J-test also fails to reject the validityof the instruments used.

ows to Brazil (January 1995 to December 1998)

R2 J-test Normality

0.204 0.218 [0.91] 12.511 [0.00]

cluded for the estimation of the model. Normality refersin the regression. P-values for all the test statistics are


All the coefficients are statistically significantand present the expected signs, albeit the coef-ficient of the tax on capital inflows is only sig-nificant at the 7% level, which is acceptablegiven the small sample size. We can confirmthat the IOF tax seems to have reduced netportfolio capital inflows to Brazil during theperiod in which it was implemented, hence gen-eralizing the results obtained previously. Theestimated elasticity for portfolio capital flowsto changes in the tax rate is approximately �7and it is only smaller in magnitude than theelasticity with respect to changes in the high-yield spread.

Therefore, it seems that the IOF tax helped toreduce in some way the sterilization costs facedby the Brazilian monetary authorities in the1990s. By reducing the amount of portfoliocapital flows that would have entered the coun-try in the absence of taxes, the price-based con-trols have contributed somewhat to alleviatethe tension between the maintenance of thepegged exchange rate regime, domestic pricestability and a liberalized capital account (theso-called policy trilemma).

Nevertheless, it is evident that these capitalaccount policies were not sufficient to avert vul-nerability to external shocks, as evidenced bythe way that Brazil was affected by the Russiancrisis of 1998. Taxes on capital inflows couldhave the potential for reducing external vulner-ability as discussed in David (2007). In particu-lar, taxes on inflows could reduce thereversibility of capital flows by changing thematurity structure from short- to long-termflows and/or by altering the composition offlows toward forms more resilient to crises such

Table 4. GMM estimation of capital flows to Brazil e

Variable

Fixed-inc

C 0.397(Capital flows)t�1 0.509Interest differential �5.50Tax (IOF) �39.94High-yield spread �60.78R2 0.J-test 0.171Normality 0.306

J-statistic refers to the test for the validity of instruments innormality of the residuals in the regression. P-values for a

as FDI. In addition, taxes could mitigate exces-sive real exchange rate appreciation (and itsnegative effects in terms of current account sus-tainability) and reduce currency mismatchesand therefore the real costs of largedevaluations. The Brazilian controls on inflowsprobably had very limited impacts on theseareas.

We also attempted to estimate similar modelsto the one described above for total net capitalflows to Brazil; nonetheless the results obtainedwere not robust to the instruments used, whichprobably reflects the relative volatility of thisaggregate (particularly associated with theturmoil linked to the crises in Mexico, Asia,and Russia) and the fact that this variable is af-fected by factors not captured in our modelsuch as institutional quality, political economyissues, among others. Those factors are difficultto model empirically in the time series dimen-sion using data for just one country. Theestimation output will not be reported withthe objective of saving space and is availableupon request.

In order to check the robustness of the con-clusions inferred from the previous models,we decided to estimate models for portfolioflows and fixed-income flows with an extendedsample going from January 1995 to December2001, that is ignoring the fact that the exchangerate crisis and move toward a floating regimefrom January 1999 could lead to structuralbreaks. The estimation output is reported inTable 4 (the first column presents the resultsfor fixed-income flows as the dependent vari-able and the second column presents resultsfor portfolio flows as the dependent variable).

xtended sample (January 1995 to December 2001)

Coefficient [P-values]

Dependent variable

ome flows Portfolio flows

[0.00] 0.185 [0.00][0.00] 0.429 [0.00]

2 [0.00] 7.883 [0.00]3 [0.09] �41.212 [0.06]6 [0.00] �25.272 [0.02]

130 �0.784[0.88] 0.133 [0.87][0.81] 152.997 [0.00]

cluded. Normality refers to the Jarque–Bera test for thell the test statistics are presented in brackets.

Table 5. GMM estimation of capital flows to Brazil: alternative definition of interest rate differential(January 1995 to December 1998)


Dependent variable

Fixed-income flows Portfolio flows

C �0.387 [0.01] 0.072 [0.15](Capital flows)t�1 0.485 [0.00] 0.895 [0.00]Interest differential �3.423 [0.05] 8.463 [0.00]Tax (IOF) �33.210 [0.01] �3.775 [0.57]High-yield spread 23.228 [0.24] �20.572 [0.04]R2 �0.795 �1.177J-test 0.253 [0.95] 0.337 [0.58]Normality 0.699 [0.70] 42.257 [0.00]

J-statistic refers to the test for the validity of instruments included. Normality refers to the Jarque–Bera test for thenormality of the residuals in the regression. P-values for all the test statistics are presented in brackets.

Table 6. GMM estimation of capital flows to Brazil: alternative definition of high-yield spread(January 1995 to December 1998)


Dependent variable

Fixed-income flows Portfolio flows

C 0.855 [0.00] 0.291 [0.00](Capital flows)t�1 0.257 [0.00] 0.382 [0.00]Interest differential 4.883 [0.00] 4.642 [0.00]Tax (IOF) �8.512 [0.20] �3.055 [0.29]High-yield spread �85.64 [0.00] �37.796 [0.00]R2 0.110 0.208J-test 0.224 [0.98] 0.298 [0.73]Normality 0.320 [0.85] 8.380 [0.01]

J-statistic refers to the test for the validity of instruments included. Normality refers to the Jarque–Bera test for thenormality of the residuals in the regression. P-values for all the test statistics are presented in brackets.


The coefficients for the tax on inflows are stillnegative and significant, although the magni-tudes of the estimated elasticities are muchhigher in this case. The negative R2 obtainedfor the portfolio flows regression indicatesthat the diagnostic statistics for that specifica-tion are weaker than the ones present previ-ously.

In addition, we also decided to estimate mod-els for portfolio flows and fixed-income flowsincluding alternative definitions of the explana-tory variables. Table 5 presents the results ob-tained when removing the country riskpremium from the construction of the interest

rate differential. The coefficients obtained forthe tax on flows are still negative, albeit insig-nificant in the portfolio flows regression. None-theless, the weak diagnostic statistics suggestthat it might be misleading to use this measureof the interest rate differential rather than onethat includes the country risk premium, whichseems to be an important variable in determin-ing capital flows. Table 6 presents the resultsobtained when using an alternative definitionof the high-yield spread series (see AppendixB for details). Once again the estimated coeffi-cients for the tax rate on capital flows are neg-ative, but they do not seem to be significant at


conventional levels; which seems to contradictthe results reported previously. Therefore, itappears that the conclusions obtained aresomewhat sensitive to the definition of thehigh-yield spread series.

5. CONCLUSION

Overall, we can conclude that the results ob-tained here indicate that there is no evidencethat the Brazilian taxes on capital inflows wereeasily evaded and that capital accountregulations could not possibly be effectivelyimplemented because Brazil has relativelysophisticated financial markets when comparedto other emerging markets. The taxes on capitalinflows targeted portfolio (particularly fixed-in-come) flows. It is apparent from our modelsthat they were effective in diminishing netfixed-income and net total portfolio inflows toBrazil, although the statistical significance ofthe results seems to depend on the definitionof the high-yield spread series used.

The ultimate effect of the price-based regula-tions on the appreciation of the real exchangerate, monetary policy autonomy, and externalvulnerability among other important goalswas limited. This might help to explain whyBrazil failed to avert significant financial dis-tress by the end of the 1990s. It is arguablythe case that the taxes were not optimally ap-plied (too small and narrow in scope for someperiods of time, perhaps excessively large forothers). Nevertheless, this claim can only be

evaluated through a fully-fledged theoreticalmodel, which lies beyond the scope of thispaper.

In addition, it is important to note that limi-tations associated with the sample size did notallow us to estimate adequate co-integrationmodels for the impact of the taxes on inflowson the real exchange rate. In fact, the Brazilianeconomy was marked by deep structuralchanges in the 1990s (stabilization plan, cur-rency crises, etc.), which considerably affectedthe number of observations that could be in-cluded in a long-run model for real exchangerates.

What could policy-makers have done in or-der to strengthen the effectiveness of the taxesand achieve some of the objectives outlinedabove, in particular in terms of reducing vul-nerability to financial shocks? The response tothis complex question certainly involves in-creased efforts to counter evasion strategies,but more crucially the regulations need to beapplied in a counter-cyclical manner, respond-ing actively to surges and droughts in capitalinflows. Transparency and consistency of thecriteria used in decisions to alter the regulationsare also essential to maintain confidence in thesystem. The imposition of restrictions on capi-tal flows is not a panacea, they need to be com-bined with appropriate supervision andregulation of domestic financial intermediaries.Hence, it becomes evident that capital controlshave to be considered as a part of a wider reg-ulatory system designed to enhance financialstability in developing countries.

REFERENCES

Agenor, P. R., & Montiel, P. (1999). Developmentmacroeconomics (2nd ed.). New Jersey: PrincetonUniversity Press.

Agenor, P. R. et al. (2002). Cyclical fluctuations inBrazil’s real exchange rate: the role of domestic andexternal factors (1988–95). Revista Brasileira deEconomia, 56(1), 47–74.

Arida, P. et al. (2004). High interest rates in Brazil:Conjectures on jurisdictional uncertainty. NUPE/CdGRio de Janeiro, Mimeo.

Ariyoshi, A. et al. (2000). Capital controls: Countryexperiences with their use and liberalisation’’. IMFOccasional paper no. 190. Washington, DC.

Cardoso, E., & Goldfajn, I. (1997). Capital flows toBrazil: The endogeneity of capital controls. IMFWorking Paper no. 115 October. Washington, DC.

Carvalho, B., & Garcia, M. (2006). Ineffective controlson capital inflows under sophisticated financial

markets: Brazil in the 1990s. NBER Working Paper12283, June. Cambridge, MA.

David, A. C. (2007). Controls on capital inflows andexternal shocks. World Bank Policy Research Work-ing Paper 4176, March. Washington, DC.

Epstein, G., Grabel, I., & Jomo, K. (2003). Capitalmanagement techniques in developing countries: Anassessment of experiences from the 1990s and lessonsfor the future. PERI Working Paper Series 56,University of Massachusetts, Amherst.

Favero, C. A. (2001). Applied macroeconometrics. NewYork: Oxford University Press.

Favero, C. A., & Giavazzi, F. (2002). Why are Brazil’sinterest rates so high? IGIER/Bocconi Working Paperno. 224, July, Milan.

Fitzgerald, V. (2002). The instability of the emergingmarkets asset demand schedule. UNU/WIDER Dis-cussion Paper DP2002/80, September, Helsinki.


Garcia, M., & Valpassos, M. (2000). Capital flows,capital controls and currency crises: The case ofBrazil in the 1990s. In F. B. Larrain (Ed.), Capitalflows, capital controls & currency crises: Latin Amer-ica in the 1990s. Ann Arbor: University of MichiganPress.

Gertler, M., & Lown, C. (1999). The information in thehigh-yield bond spread for the business cycle:Evidence and some implications. Oxford Review ofEconomic Policy, 15(3), 132–150.

Goldfjan, I., & Minella, A. (2005). Capital flows andcontrols in Brazil: What have we learned? NBERWorking Paper 11640, September. Cambridge, MA.

Mody, A., & Taylor, M. P. (2002). International capitalcrunches: The time-varying role of informationalasymmetries. IMF Working Paper 43, February.Washington, DC.

APPENDIX A. TAXES ON CAPITA

Changes in capit

November 1993 Creation of Foreign CapitalAlso tax of 3% on financial

March 1994 IOF tax extended to other pOctober 1994 Tax on portfolio inflows inc

loans augmented to 7% in aMarch 1995 Pressures associated with the

the tax on foreign currency lAugust 1995 Increase on the tax on portf

loans tax to 5%September 1995 Tax on foreign currency loan

of loans. A tax of 5% was imtwo years, 4% for loans withof maturity between 3 and 4between 4 and 5 years

April 1997 Tax on portfolio inflows waDecember 1998 Increase in the IOF tax fromMarch 1999 IOF tax was reduced from 2June 1999 IOF Tax set to 0%

Source: Brazilian Central Bank, Soihet (2002), Cardoso an

Mody, A., & Taylor, M. P. (2003). The high yield spreadas a predictor of real economic activity. IMF StaffPapers, 50 (Vol. 3, pp. 373–402).

Palma, J. G. (2004). The 1999 Brazilian financial crisis:How exuberant monetarism led to banking fragilityand public sector Ponzi finance. Mimeo: University ofCambridge.

Perron, P. (1988). Trends and random walks in macro-economic time series: Further evidence from a newapproach. Journal Economic Dynamics and Control,12, 297–332.

Soihet, E. (2002). Indice de controle de capitais: Umaanalise da legislacao e dos determinantes do fluxo decapital no Brasil no periodo 1990–2000. UnpublishedMaster’s Degree Dissertation. FGV/EPGE, Rio deJaneiro.

L INFLOWS IN BRAZIL 1993–99

al account regulation

Fixed Yield Fund: IOF tax of 5% imposed.loans in foreign currencyortfolio flowsrease to 9% and the tax on foreign currencyn attempt to counter the surge in flows

‘‘tequila effect.’’ IOF tax reduced to 5% andoans to 0%olio inflows to 7% and on the foreign

s altered to increase the maturity structureposed on loans with a maturity of less thanmaturity between 2 and 3 years, 2% on loansyears and 1% on loans of maturity

s reduced from 7% to 2%2% to 2.38%

.38% to 0.5%

d Goldfajn (1997), and Goldfjan and Minella (2005).

(See Overleaf)

APPENDIX B. OVERVIEW OF DATA AND SOURCES

Series Description/notes Source

SELIC Reference domestic interest rate Central Bank of Brazil Website:http://www. bcb.gov.br/?SERIETEMP

Trade balance Exports of goods and servicesminus imports (in Logs)

Author’s Calculations based on rawdata from IPEADATA: http://www.ipeadata.gov.br/ipeaweb.dll/ipeadata?274539406

Nominalexchange rate

Real/US dollar marketexchangerate (average for the month)

IPEADATA Website: http://www.ipeadata.gov.br/ipeaweb.dll/ipeadata?274539406

Forwardexchange rate


Realexchange rate

Real effective exchange rateINPC exports index


Country riskpremium

EMBI Brazil spread over UStreasury bills

J.P. Morgan, obtained fromThomson’s Datastream

Junk bondspread

Difference in yields between UShigh-yield bonds and 10-yearUS treasury bonds. A number ofdifferent high-yield indexeswhere used such as Master II(H0A0) and high-yield 175 (X0A0)

Merrill Lynch, obtained fromThomson’s Datastream

World oil priceindex

IMF’s international financialstatistics database

US interest rate 90-Day US treasury bill IMF’s International FinancialStatistics Database

Net portfolioflows

Author’s Calculations based onraw data from Central Bank ofBrazil’s Website

Net fixedincome flows

Author’s calculations based onraw data from Central Bank ofBrazil’s Website

Net total flows Author’s calculations based on rawdata from Central Bank of Brazil’sWebsite

Interest ratedifferential

We experimented with a number ofdistinct differentials. Generallycalculated as difference between theSelic rate, the US interest rate, thenominal exchange rate devaluation(actual, past, and forward rates wereused alternatively) and thecountry risk spread

Author’s calculations


http://www.bcb.gov.br/?SERIETEMP

http://www.ipeadata.gov.br/ipeaweb.dll/ipeadata?274539406












Documents

Revisiting Price-based Controls on Capital Inflows in a “Sophisticated” Emerging Market