Financial liberalisation and the relationship between finance and growth

UNIVERSITY OF CAMBRIDGE Centre for Economic and Public Policy

How and Where Capital Account Liberalization Leads to Economic Growth

Dennis P Quinn Professor McDonough School of Business Georgetown University Washington, D.C 20057

quinnd@gunet.georgetown.edu

Carla Inclan Ernst & Young, LLP Washington, D.C 20036

carla.inclan@ey.com

A Maria Toyoda Research Scholar Institute for International Studies Stanford University Stanford, California

(SBR-ABSTRACT

We utilize an empirical model of growth as a platform for examining the effects

of capital account liberalization on growth While we test for the direct effects of liberalization,

we are equally interested in another facet of liberalization: sequencing We ask what prior

political, social, or economic conditions were required for capital account liberalization to have led to subsequent growth Our key independent variable is a measure of capital account

openness that comes in the form of five-year time-series, cross-sectional observations for 80 nations, 1950 (or independence) to 1997 Our focus is on change indicators of liberalization, as

we argue that level indicators of government policies in political economic research are generally too imprecisely specified to exclude the influence of other collinear political economic variables

We focus not simply on the economic preconditions for beneficial liberalizations, but the

political and social preconditions as well We find that capital account liberalization has a robust and direct effect on subsequent economic growth in most countries Capital account

liberalization does not, however, lead to higher growth in emerging market democracies that have weak welfare states We conclude that policymakers in emerging market democracies have ample reason to be cautious about full capital account liberalization.

Policymakers in emerging market nations are routinely urged to liberalize their international economic transactions Scholars, journalists, and experts from international development agencies laud the growth and development benefits of freer trade in goods and services in particular (See Rodrik 2000 for a critical review of this advice.)

Is this advice sound regarding international capital flows? That is, if a government of an

emerging market nation were to liberalize its restrictions on inward and outward capital flows, what would be the growth effects in the liberalizing economy?

We offer a partial answer to these questions in this paper We proceed as follows After a brief review of the relevant literature, we utilize an empirical model of growth as a platform for examining the effects of capital account liberalization on growth While we test for the direct effects of liberalization,

we are equally interested in another facet of liberalization: sequencing What prior political, social, or economic conditions were required for capital account liberalization to have led to subsequent growth?

Our paper makes several contributions We introduce a measure of capital account openness that comes in the form of five-year time-series, cross-sectional observations for 80 nations, 1950 (or

independence) to 1997 We focus on change indicators of liberalization, arguing that level indicators of government policies in political economic research are generally too imprecisely specified to exclude the influence of other collinear political economic variables We focus not simply on the economic

preconditions for beneficial liberalizations, but the political and social preconditions as well We have new results, which we discuss at the end of the paper

I A Brief Review of the Capital Account Liberalization and Growth Literature

The starting point of this paper on the growth effects of capital account liberalization is the recent and comprehensive review essay on the topic by Barry Eichengreen (forthcoming) Eichengreen notes that various theoretical models drawn from economics imply inconsistent effects from capital account liberalization One strand of literature, drawing on traditional “frictionless” factor market models, proposes that capital account liberalization produces growth effects for many of the reasons trade

liberalization does (See also Sweeney 1997.) A second strand of theory, however, proposes that many

types of policy-based distortions in a nation’s economy lead to suboptimal outcomes from capital account liberalization as the distortions create “second-best” conditions (See Stiglitz 2000, e.g.) To a very large extent, the differences in the two perspectives derive from different starting assumptions about the policy and economic environments under which liberalization occurs

Given the conflicting starting assumptions and resulting theory, we might reasonably hope that empirical studies of the many countries that have liberalized their capital accounts might narrow the theoretical discourse Unfortunately, the first round of “large n” empirical studies of the direct effects of capital account liberalization produced indecisive results Grilli and Milesi-Ferretti 1995 found no

association between the levels of capital account openness (hereafter Openness) and economic growth, a finding that Rodrik 1998 replicated and extended Quinn 1997 showed that changes in capital account

openness (hereafter Liberalization) were associated with higher long-run growth Focusing on emerging markets, Bekaert, Harvey, and Lundblad 2000 also find that incidences of financial liberalizations were associated with subsequent economic growth

Much of the difference is in the data used Grilli and Milesi-Ferretti and Rodrik use a binary 0,1 indicator of the presence or absence of capital controls found in a table at the back of the International

Monetary Fund’s annual publication, Exchange Arrangements and Exchange Restrictions The IMF 0,1

data contains too little information for it to be used to study Liberalization per se, but only levels of Openness The Quinn study is based on a coding of the text of the laws governments used to regulate

capital accounts, which are reported in the text section of Exchange Arrangements (The Quinn/Toyoda

measure is described below.) These data contain ample information to generate a study of Liberalization, but the range of countries and years for which data were available was limited The Bekaert, Harvey, and Lundblad 2000 study also used a 0,1 measure, but this measure, unlike the IMF’s, is linked to the date of the liberalization Eichengreen forthcoming reviews the data and methodological differences among most

of these and other studies (For discussions of the different data measures, see also Edwards 2001 and Quinn 1997.)

A second round of studies moves beyond the direct effects of Openness on growth to a focus on the channels through which openness might produce growth Kraay 1998 finds scant evidence of the

effects of Openness on investment Klein and Olivei 1999 show Openness leads to financial “deepening,” but only for advanced industrial nations Klein and Olivei propose that emerging market nations lack some key political economic institutions through which Openness might act beneficially, which implies that Openness has a contingent relationship to growth Klein and Olivei use a summed version of the 0,1 IMF indicator for 1986-95 Levine and his coauthors, in a series of papers (King and Levine 1993; Levine and Zervos 1998; Beck, Levine, and Loayza 2000), find that financial or stock market

development leads to growth, though whether Openness or Liberalization contributes to that process was not addressed in those papers

Scholars undertaking a third round of studies test a version of the proposition that Openness or Liberalization’s effects are contingent upon various economic and policy preconditions Diaz-Alejandro

1985 notes that financial liberalization in emerging market nations often produced poor results, leading to the question of what prior states distinguished emerging market nations from other nations Kraay 1998 finds little evidence that Openness’ effects are contingent on various economic preconditions Edwards

2001 finds that Liberalization lead to growth in middle to high-income countries Arteta, Eichengreen, and Wyplosz 2001 revisit Edwards’s study, and while they reject his findings on methodological and other grounds, they suggest that Liberalization does indeed have a contingent relationship with growth The contingency that matters, they believe, is macroeconomic imbalances – as exemplified by black market premia Chandra 2001, investigating sociological contingencies, finds that countries with higher levels of ethnic heterogeneity were actually harmed by Openness More homogenous societies, in

contrast, benefited Edwards and Arteta et al used the same indicator as Quinn 1997, whereas Chandra used the IMF 0,1 data

These studies have advanced our understanding of capital account openness and its effects The search for contingencies under which liberalization produces beneficial effects is a particularly promising avenue of research

Even so, the prior studies have some limitations For one, the 0,1 measure of capital account openness taken from the IMF tables is of limited further use Another lesson we can take from the prior studies is that purely cross-sectional research designs are unlikely to reconcile some of the differences

among the studies because of the necessarily limited range of information that can be entered as

independent variables in purely cross-sectional models

Perhaps the salient problem found in the prior studies is a central problem of empirical political economic scholarship generally – nations at similar levels of political, social, and economic development have similar “clusters” of political economic policies and structures Of relevance to this study, the levels

of Openness are, without question, part of a broader cluster of policies and processes, and are therefore collinear with many other variables (We will develop this point below as we describe the “repression syndrome” and the “liberalization cycle.”) For related discussions, see Arteta, Eichengreen, and Wyplosz

2001, 11; Rodríguez and Rodrik 2000, 28-34; and Eichengreen forthcoming, 6

We explicitly address the first and third round of studies of capital account liberalization and growth We use annual indicators of capital account restrictions for 77 countries to construct five-year panel averages, 1960-98 based on the same data and coding rules as Quinn 1997 The range of

information allows us to compute meaningful indicators of change in capital account regulation, and it also allows us to enter a wide array of information into the analysis We also use a robust method - pooled-cross-section, time-series analysis To guard against simultaneity, we use long lags of the key independent variables

We overcome in part the policy cluster problem by focusing on changes in capital account regulation, while including levels (Openness), along with many other variables as control variables in a pooled regression model Both levels and changes in capital account regulation carry relevant

information, but in a time-series, cross-section research design, changes in Openness are less likely to exhibit collinearity with other “cluster” variables A further advantage of focusing on Liberalization rather than Openness is that Liberalization is a topic of greater policy relevance for many emerging market nations

II The Repression Syndrome and the Liberalization Cycle

Groups of countries have hard-to-measure, common attributes for many reasons Nations share strategic interests, citizens of nations at similar levels of development have similar tastes, and many nations share common linkages from emigration and colonization Other groups of nations, even with

otherwise different cultures and traditions, have shared ideological or religious beliefs Of importance to this investigation is that these common experiences and values lead nations to adopt similar – and related – political economic policies

Let us develop a relevant example If the political elite of a group of nations were to share a belief in the efficacy of markets, we might see these countries adopt similar institutions and policies A floating exchange rate, open trade accounts, an independent central bank, anti-inflationary policies, and open capital accounts might jointly characterize these economies As policies are hard to measure precisely, the effects of mutually related policies are sometimes conflated For example, capital account openness might proxy for the above list of related policies across countries Moreover, even if economic policies could be measured precisely, because these economic policies are frequently collinear, the econometric difficulty of estimating the effects of any one policy is great

Regarding international financial liberalization, we find two distinct types of political economic clusters (This section draws on Quinn 2000.) Low levels of capital account openness are associated with lower levels of per capita income, lower levels of trade openness, weaker financial development, higher levels of inflation, fixed exchange rates, and higher premia on the black market for foreign currencies These are characteristics of what McKinnon 1973 called a financially repressed economy These

economically repressed countries are frequently also politically repressed (Quinn 2000), and highly vulnerable to political instability Politically repressed economies are characterized also by low rates of investment in human capital (Helliwell 1994) and high birthrates (Feng, Kugler, and Zack 2000) These joint and cumulative attributes of political and economic repression we will henceforth call the

“repression syndrome” of which capital account closure is only one part In an econometric

cross-sectional investigation, one of the indicators of the repression syndrome is highly likely to capture part of the influence of the other indicators

Let us turn to a “liberalization cycle.” From the 1950s onward, most democratic governments have liberalized finance and liberalized trade (Quinn 2000) These economically open countries

developed strong financial sectors, produced low levels of inflation (comparatively speaking), and had very limited black markets in currencies These democracies also invested in human capital and had

lower birthrates than similar countries, and per capita income was higher than in authoritarian countries Democratic countries were far less vulnerable to revolutions, coups, and other forms of political

instability than other nations, in part because democratic nations followed economic policies that

decreased economic risk even at the expense, at times, of higher growth (Quinn and Woolley 2001) These generalizations apply at least partially to both emerging market as well as advanced industrial democracies

We present some evidence on the repression syndrome and the liberalization cycle in Tables 1 and 2 where we report the simple correlations between variables, 1960-98 (using five year averaged data, which will be described below)

[Tables 1 and 2 about here]

The pairwise contemporaneous correlations do not capture change over time, do not take into account the effect of other variables, and tell is little about the direction of relationships But, they do help

us make the point that political economic variables are correlated in line with the repression syndrome and the liberalization cycle Note that the repression syndrome and liberalization cycles also present themselves, albeit in a weaker form, in the emerging market data

An empirical implication for our project is that political economic clustering exposes to challenge estimating the direct effects of Openness We will show below that the levels of capital account openness, lagged two periods, usually have a statistically significant and positive coefficient when entered in a growth regression But, because of the clustering of variables, we cannot directly tell whether the levels

of Openness, or some other aspects of the repression syndrome or the liberalization cycle, are what is actually linked to subsequent growth

In the second part of our analysis, we make use of the information found in these and other political economic variables to examine whether Liberalization has a direct or contingent (or both) relationship to economic growth We turn now to a discussion of the possible contingencies

III Preconditions for the Benefits of Liberalization?

The Liberalization cycle and the repression syndrome are not simply econometric problems More importantly, these political economic states potentially reinforce, enable, or impede other policies

Are features of the liberalization cycle necessary preconditions for capital account liberalization to generate growth? Do features of the repression syndrome impede the positive effects of capital account liberalization?

Economic States as Preconditions for Liberalization This section draws on the prior work of Edwards

2001 and Arteta, Eichengreen, and Wyplosz 2001 The authors of these papers see Liberalization as having a contingent, rather than direct, effect on growth The core conclusion of the Edwards 2001 study

is that economic development, proxied by per capital income, was a precondition for the benefits of Liberalization The Arteta, Eichengreen, and Wyplosz 2001 paper examines a number of policy

preconditions for the benefits of Liberalization to occur They use indicators of trade liberalization from Sachs and Warner 1995 and the premiums paid in black markets for foreign currencies as indicators of imbalanced macroeconomic conditions They advance the theory that the benefits from Liberalization are likely to occur only after key distortions have been eliminated to prevent either misdirection of the resulting inflows or capital flight

Two additional questions then emerge about related policies… The first is whether the prior liberalization of international financial transactions associated with the underlying current account transactions might produce benefits different from liberalizing the underlying transactions themselves The second is whether financial development is a precondition for the positive effects of capital account liberalization Arteta, Eichengreen and Wyplosz 2001 do not find such an effect, but we test for it

Political and Legal States as Precondition for Liberalization We draw on Quinn 2000 and Quinn and

Woolley 2001 for this section Wealthy democracies have been relentless liberalizers of both capital and current accounts See Figure 1, which shows median levels of Openness over time for three groups of countries: democratic OECD nations, continuously democratic emerging market nations,1 and

1 We define continuously democratic emerging market nations as those whose summed democracy/autocracy Polity

98 scores were continuously above 7 on the –10 to 10 scale, 1960 (or after independence) to 1995 These nations are Botswana, Colombia, Costa Rica, India, Israel, Jamaica, and Trinidad If we were to stretch the definition of democracy to include nations whose Polity 98 scores are continuously above zero, we would include Malaysia, South Africa, Sri Lanka, and Venezuela

continuously autocratic nations, 1950-97 Emerging market democracies, while generally more open than authoritarian emerging market nations, were characterized by far lower levels of Openness than OECD democracies

The lower levels of Openness might be explained in part by a finding in Quinn 2000, which was that capital account liberalization was a risk factor contributing to democratic reversals in emerging market democracies Capital account liberalization is robustly associated with subsequent increased income inequality (Quinn 1997) Dixon and Boswell 1996 find that foreign investment “penetration,” which follows from capital account liberalization, also increases income inequality (cf Firebaugh 1996) Increased income inequality, in turn, has deleterious effects on polities, particularly emerging market polities (Muller and Seligson 1987)

Democracies, therefore, tend to compensate losers from market competition These

“side-payments” take the form, in wealthier democracies, of direct transfer payments In poorer democracies, however, the ability to make direct transfer payments is limited by their ability to tax and fund a welfare system Consequently, emerging market democracies use other mechanisms of compensation, such as employment in state-owned enterprises or trade protection Even private firms in emerging market democracies frequently provide social welfare benefits to society (Khanna 2000) Capital account liberalization, however, constrains the ability of these democratic countries to maintain these other forms

of compensation We might therefore expect that the transactional costs of dismantling traditional methods of compensation and establishing new ones are very high Non-democratic nations presumably are less concerned with compensation Hence, the growth benefits of Liberalization might be less for emerging market democracies than other nations

A second possible explanation for lower levels of Openness in emerging market democracies is that voters in all forms of democracy are characterized by a high degree of risk aversion Quinn and

2 We define continuously autocratic nations as those whose summed democracy/autocracy Polity 98 scores are

continuously zero or below, 1960 (or shortly after independence) to 1995 These nations are Algeria, Bahrain, China (PRC), Egypt, Ethiopia, Indonesia, Iran, Iraq, Jordan, Kenya, Morocco, Rwanda, Syria, and Tunisia

Woolley 2001 find that voters everywhere punished incumbent governments for increased growth

volatility In advanced industrial democracies, Liberalization might lead to lower growth volatility from portfolio diversification effects In emerging market democracies, in contrast, it might lead to increased growth volatility as capital account liberalization in financial repressed economies have tended to result in speculative bubbles with subsequent financial crashes (See Reinhart and Kaminsky 1998 and

Williamson and Mahar 1998 for a review of the evidence that international financial liberalization is associated with subsequent economic crises.)

Third, democracies tend to be associated already with “virtuous” economic policies As

democratic governments tend to reform their economies, the marginal economic benefits of capital

account liberalization are smaller Autocratic countries, in contrast, tend not to reform, since

Liberalization might reduce an autocrat’s freedom to impose arbitrary policies

Another possible political precondition for beneficial effects from Liberalization is investor

protection La Port, Lopez-de-Silanes, Shleifer, and Vishny 1998 propose that nations with English common law traditions will have deeper financial markets because that tradition offers investors strong property rights protection Liberalization might result in higher rates of growth in common law countries

Political volatility, revolutions and coups, and other forms of social unrest are also part of the fabric of many emerging market societies In settings with higher levels of political uncertainty and violence, a natural response of investors to Liberalization could well be capital flight Moreover, in the face of political uncertainty, foreign investors will avoid delayable investments (Rivoli and Salorio 1997) Hence, liberalizing against the backdrop of violence and uncertainty is plausibly economically

counterproductive

Social Development A central question in the growth literature has been whether ethnically or

linguistically fractionalized societies suffer lower rates of growth because of the economic consequences

of social tensions (See Easterly and Levine 2001.) Chandra 2001 tests directly for the contingent effects

of ethnic fragmentation on Openness in a growth regression Her pessimistic findings suggest that

heterogeneous societies do not benefit from financial openness

Another strand of growth theory has examined the growth consequences of social

underdevelopment (See Feng forthcoming; Feng, Kugler, and Zack 2000.) Nations suffering from a poverty trap are less able to use policy reforms to escape their situation Lower levels of education and higher rates of birth proxy for the existence of a poverty trap, which might prevent capital account liberalization from producing growth

Summary We make three points above 1) The level of capital account openness is part of a “clustering”

of political economic variables, which can be characterized as a “repression syndrome” or a liberalization cycle “Openness” as such is a less reliable regressor in a growth regression Capital account

liberalization’s effects on growth, in contrast, are less vulnerable to the “clustering” problem, and offer a more policy relevant study 2) Liberalization’s effects on growth might be contingent on prior states 3) Prior economic states are not the only relevant contingencies: political, legal, and social conditions might also matter

IV DESIGN

Our research design is simple We begin with a base growth model in which we test for the direct effects of capital account liberalization on growth In the second stage of our analysis, we estimate

models including the interaction between Liberalization and prior levels of various independent variables

– e.g., when studying the effect of Black Market Premium we include the following three terms in the model:

Capital Account Liberalizations-1 + Black Market Premiums-2 + CALs-1*BMPs-2

If Liberalization's effect is contingent, and the contingent variable is a dummy (indicator) variable, the interaction of capital account liberalization with the "contingent" variable will be statistically significant, but the base capital account liberalization term will no longer be statistically significant If neither the interaction term nor the prior levels variable is statistically significant, then the direct effect of capital account liberalization is maintained (assuming its initial statistical significance) When the contingent variable is a measurement (like black market premium, democracy, population growth, etc.) and the interaction term is positive and statistically significant, than the interpretation is that the effect of capital account liberalization is greater when preceded by higher values of the contingent variable On the other

hand, if the interaction term is negative and statistically significant, it means the effect of capital account liberalization on growth is smaller when preceded by higher values of the contingent variable

The prior political economic variables are these:

A) prior economic states - national income; current account liberalization; trade liberalization (the

Sachs-Warner trade openness measure); black market premium; and financial depth;

B) prior political and legal states - democracy (Polity 98 scores); continuously democratic or

continuously autocratic countries; property rights protection (English common law tradition); and

political stability (revolutions and coups; volatility of democracy scores);

C) prior social states - educational attainment (secondary schooling); population growth; and ethnic

fractionalization (ELF60 or ETHFRAC)

Let us note three possible design problems The first is possible endogeneity in the relationships between growth and various independent variables We focus on lagged changes in capital account liberalization, which should not be influenced by subsequent growth A second problem is that data limitations for many of the prior state variables lead to large reductions in the sample In particular, the experiences of the 1960s are sometimes lost with the use of the prior state variables: for the financial depth and black market premium measures, we lose up to 40% of the available sample We can only interpret cautiously some of the results of the prior state variables A third problem, to which we have already spoken, is the extensive collinearity among variables When collinear variables are used to create interaction terms, the collinearity problem is exacerbated To reduce this problem, we “recenter” the variables used to create the interaction terms, which has the effect of reducing the Variance Inflation Factor in most cases.3

V METHODS, DATA, and MODELS

Methods

3 We subtract the means of the variables in question from the observed values of the variables For example, when considering X1=democracy(s-2) and ∆ capital(s-1), the three terms included in the model are: (X1-mean of X1), ( ∆ capital – mean of ∆ capital), and (X1-mean of X1)*( ∆ capital – mean of ∆ capital) Centering before obtaining the interaction terms reduces the correlation between each of the variables and the interaction term without affecting the coefficients of interest

The dependent variable in this investigation is per capita ppp-adjusted economic Pooled, cross-section, time-series (PCSTS) models are useful in evaluating the question of why, over time, some nations grow quickly and others do not That is, the variation in the dependent

variables comes from both the time series and the cross-sections, and some pooling of data is necessary to address the questions We estimate PCSTS models using five-year averaged data

An explicit assumption in estimating PCSTS models is that the relationship between independent and dependent variables are not simultaneously determined We use very long lag periods to guard against simultaneity The pooled equations are estimated by ordinary least squares using panel corrected standard errors, as suggested by Beck and Katz 1995.4 All models are fixed effects models5 in which country dummy variables are used (The coefficient estimates of the country dummy variables are not reported, but are available from the authors.)

Data

International Financial Regulation We operationalize international financial regulation through two

indicators of change in international financial openness or closure, which are described in Quinn 1997

CAPITAL and CURRENT are the main components of OPENNESS created from the text of an annual volume published by the International Monetary Fund (IMF), Exchange Arrangements and Exchange Restrictions This IMF text reports on the laws governments use to govern international financial

transactions The measure is available from 1950 to 1997 for 58 countries, and for a shorter period for an additional 23 (See Eichengreen forthcoming for a review of this and other measures.)

4 All PCSTS estimations use the POOL command with HETCOV option in Shazam 9.0 Because of a matrix error

in the code for HETCOV, all residuals analysis is done with the OLS option in POOL, which gives accurate

residuals

5 An alternative is to use random effects models We replicate key results using S-Plus These data are not from a

random sample, but are the universe of that which is available For discussions, see Hsiao 1986, chapter 4, and Pesaran, Shin, and Smith 1998, 4

CAPITAL is scored 0-4, in half integer units, with 4 representing a fully open economy

CURRENT is scored 0-8, in half integer units, which represents the sum of the two components of current account scores: trade (exports and imports) and invisibles (payments and receipts for financial and other services) We transformed each measure into a 0 to 100 scale taking 100*(CAPITAL/4) and

100*(CURRENT/8)

When using CAPITAL and CURRENT as independent variables, we need to model the potential

influences of changes and levels of these variables over many years We use five-year averages, which are calculated as follows:

CAPs=( Xt + Xt+1 + Xt+2 + Xt+3 + Xt+4 )/5 where Xt = 100*(CAPITALt/4) The subscript s represents a five year period: s=1960-64, 1965-69,…, and the subscript t identifies the first year in the five year period: t=1960, 1965,… Because we are interested

in isolating how changes in policy affect growth, and because we seek to avoid problems of endogeneity, our primary focus is on lagged changes of CAPITAL, or:

∆CAPITALs-1 = CAPs-1 - CAPs-2

We also create a contemporaneous change measure and a lagged levels measure, or, ∆CAPITALs and CAPITALs-2, respectively Corresponding variables for CURRENT are defined similarly In cases of missing values, the averages are obtained over the number of observations available

Economic, Political, and Social Data Our focus in this investigation is on international capital account

variables The other variables in the study are treated as control variables

In estimating the unbalanced panel models, we use a beta version of Penn World Tables Mark 6.0.6 The advantage of this data set is that we are able to estimate models using data from 1960 to 1998 The disadvantage is that the data may contain errors.7 To insure the robustness of the results, we repeat

6 Prof Alan Heston provided the data Email correspondence between Quinn and Prof Heston, 28 Feb 2001

7 The PWT 6.0 growth series for Syria contain some unlikely numbers, as does the trade series for India before

1970 The data for Syria and India before 1970 are excluded from the analyses The data for Argentina and

all analyses with data drawn from the Penn World Tables Mark 5.6 We report the results for key

equations using PWT 5.6 data, and note significant deviations between the PWT 6.0 and PWT 5.6 results The PWT 6.0 data contain more observations than the PWT 5.6 data, but when matched with the data for other variables the resulting data set also contains fewer countries (70 vs 76) Our core results are robust

to the choice of which PWT data set we use

The educational attainment measures are Barro/Lee indicators from the World Bank 2001 The data on revolutions, coups, etc are updated Cross-National Times Series data from Banks 2001

Educational data for Nigeria, Tunisia, and Morocco are unavailable, as are political data for Hong Kong

In order to use the widest range of countries, we omit education from the base model, but we use

educational measures in our interaction models Our core results are highly robust to the inclusion or exclusion of educational attainment measures The black market premium data and the Liquidity

measures are taken from Beck, Levine and Loayza 2001 Because the black market data have an

extremely skewed distribution but also contain negative numbers, we transform the series using a signLog transformation (Atkinson 1985).8

For a democracy indicator, we use the “Democracy” plus “Autocracy” indicators from the Polity

98 data set, which report data from 1800 to 1998 for the countries used in this investigation (See Gurr and Jaggers 2000.) Autocracy is scored on a –10 to 0 scale, Democracy is scored on a 0 to 10 scale, and

both are summed to produce the main political indicator of this investigation, DEMOCRACY (This

indicator is also used the World Bank 1997, 112.) Missing data are interpolated linearly

Singapore in PWT 6.0 contain a few missing data points In constructing the five year panel averages used here, we used the data from PWT 5.6 to interpolate the missing data points

8 Taking logarithms is a common practice when fitting linear regression models for several reasons However, when

X has negative values, or 0's, the logarithm is not defined One alternative is to use the following transformation: sign(x)log(abs(x)+1) This is a monotonic transformation which achieves the same objective of making the

distribution more symmetric and the relationship with Y better described by a straight line This transformation is like the power transformation with offset discussed in Atkinson (1985)

We use the levels of the index transformed into a 0 to 100 scale and lagged two five-year periods

as an explanatory variable, and we also consider its interaction with financial regulation We also use FULLDEM, a 0,1 Democracy indicator for countries that were continuously democratic during the period studied, following Quinn 2000. 9 The 0,1 indicator, when interacted with the financial liberalization variables (FULLDEM*∆CAPITALs-1), allows us to estimate precise differences in coefficients between continuously democratic and other nations We further distinguish the seven continuously democratic emerging market nations by (EMERGING*FULLDEM*∆CAPITALs-1) and the eighteen continuously democratic advanced industrial nations by (OECD*FULLDEM*∆CAPITALs-1) To examine whether the effects of capital account liberalization differ for continuously autocratic nations versus other nations we create AUTOCRAT* ∆CAPITALs-1

For social fragmentation, we use two different indicators One is the linguistic fractionalization (ELF60) index from Mauro 1995, which is used by Chandra 2001 The indicator is taken from a single

1960 survey ELF60 is therefore available for only one point in time, which creates estimation difficulties

in a fixed effect model.10 Krain 1997 offers a more nuanced measure of ethnic (rather than linguistic) fractionalization over a number of points in time (ETHFRAC), and we use it also (ELF60 and

ETHFRAC are highly correlated.)

Models

We use a panel variant of the standard Barro 1991 economic growth model The base model

includes per capita income measured at the beginning of the period, change in investment, lagged levels

of investment (as a share of GDP), annual rates of population growth, lagged levels of trade openness

9 These are: Botswana, Canada, Colombia, Costa Rica, the U.S., Iceland, India, Israel, Japan, Austria, Belgium,

Denmark, Finland, Germany, Ireland, Italy, Jamaica, the Netherlands, Norway, Sweden, Switzerland, Trinidad, the United Kingdom, Australia, and New Zealand The Polity 98 scores for France dropped from 10 to 5 in 1958 as the Algerian war of independence intensified and the 4th Republic collapsed (France’s score rises to 8 in 1969.)

Because of unobtainable data on financial regulation, all former Soviet-Bloc nations are excluded

(imports + exports as a percentage of gross domestic product), and change in trade openness The level of income is logged for standard econometric reasons We add to this model indicators of oil price shocks and political instability (revolutions, coups, assassinations, guerrilla wars, and political crises from Banks

∆GDPi,s = ß0 + ß1Incomei,t + ß2(∆Investmenti,s ) +ß3(Investmenti,s-1) +

ß4(Population Growthi,s-1 ) + ß5(∆Trade Opennessi,s ) + ß6(Trade Opennessi,s-1 )

+ ß7(RevolutionsCoupsi,s ) + ß8(∆OilPricei,s ) + ß9(OilPricei,s )

+ ß10(∆CAPITALi,s ) +ß11(∆CAPITALi,s-1 ) + ß12(CAPITALi,s-2 )

+ ß13, 14 (Country Dummy Variables) + εi,s i=1,2, ,70

Note that we are reducing the likelihood that the coefficient estimates of the independent

variables will be spuriously statistically significant by using fixed-effect models with lagged averages of the key independent variables Let us also note that the correlation between various indicators of

CAPITAL and CURRENT are very high (please see Appendix Table A3), which introduces the

possibility of an upward bias in the estimates of the standard errors when both are in a model Appendix

Table A1 lists the countries and years used Table A2 describes the variables

VI RESULTS Direct Effects In Table 3, we report the main results for capital account Liberalization, or ∆∆∆∆Capital(s-1)

Models 3.1 and 3.3 report the base model results for PWT 6.0 and PWT 5.6 data, respectively Models 3.2 and 3.4 add current account indicators to the models for PWT 6.0 and PWT 5.6 data, respectively All

10 We use the United States as the omitted country in the fixed effects model When we estimate models with

ELF60, we also omit Canada (in order to avoid an exact linear combination of variables)

four models perform well from an econometric perspective, and the control variables have coefficient estimates that conform to standard expectations

[Table 3 about here]

Capital account liberalization has a positive and statistically significant coefficient at beyond the 05 level in all four models, which suggests that it leads to growth Capital account Openness has a positive and statistically significant coefficient in three of four models (save the 5.6 model with the highly

collinear current variables entered)

In Table 4, we focus on emerging market nations, the risk of sample selection bias

notwithstanding In models 4.1 and 4.3, ∆∆∆∆Capital(s-1)’s coefficient is positive at statistically significant

at the 1 and 05 levels, respectively In the models with the CURRENT account variables added, the coefficient estimates of capital account liberalization are not statistically significant at conventional levels, though the estimates are larger than their standards errors Capital account openness has

statistically significant coefficients in models 4.1 and 4.3, but not in the models with the CURRENT variables The coefficient estimate of levels of current account openness is positive and highly

statistically significant in model 4.2

We find a preponderance of evidence that capital account liberalization has a direct effect on growth The estimated effects of Liberalization are smaller in the emerging market sub-sample than in the full sample, but are still positive and statistically significant in two of four models A separate analysis of OECD nations (not reported here to save space) shows that the coefficient estimates for capital account liberalization are nearly identical to those estimated in Table 3 We believe, therefore, that the coefficient estimates for capital account liberalization do not differ between OECD nations and emerging market nations

How well is the estimated relationship maintained in the second stage of our analysis? We next estimate the effect of the interaction between prior political economic states and capital account

liberalization on growth

Contingent Effects In Table 5a, we focus our attention on the possible contingent effects of prior

economic states on Liberalization To our surprise, none of the interaction terms between Liberalization

and the prior economic states was statistically significant

Many of the variables representing the prior economic states were, in contrast, statistically significant: current account liberalization (positive), black market premium (negative) and the Sachs-Warner trade openness measure (positive at the 1 level) are worthy of note Our findings are partly in accord with those of Arteta, Eichengreen, and Wyplosz 2001 in that the estimates for their most

comprehensive pooled model studying the effects of trade openness and capital account liberalization (their Table 5, model 8) roughly match ours (The design and data of the two studies and the Edwards

2001 study do not fully match, so strict comparability is not possible.) We do find strong negative effects

of high black market premia, but do not find, as Arteta et al did, a significant interaction effect

[Table 5a about here]

In Table 5b, we focus on the prior political and legal states In this table, many of the interaction terms are statistically significant

[Table 5b about here]

The interaction term for ∆∆∆∆Capital(s-1)*Democracy(s-2) is statistically significant and negative in

both the overall model and the emerging market model The ∆∆∆∆Capital(s-1) term is also statistically

significant but positive in both models This suggests that capital account liberalization does lead to higher growth but that the effect of liberalization is smaller in more democratic countries We will investigate this relationship more carefully below

The interaction term for common law property rights protection (or ENGLISH) is statistically significant at the 1 level in the overall model, but not in the emerging market model The interaction term for ENGLISH is positive and statistically significant in both models, which suggests that the effects of capital account liberalization is greater in common law countries than in others.11

11 In order to estimate fixed effects models, we omitted an additional dummy country dummy, which was Brazil

The interaction terms for political volatility were not statistically significant The interaction term for political violence and Liberalization was statistically significant, but positive The base term for Liberalization is no longer statistically significant, so the level of political violence joins the level of democracy as a possible contingency for liberalization and growth

In Table 5c, we undertake a “contest” among the statistically significant variables from the political and legal development analysis in Table 5b That is, when we enter all the statistically

significant prior condition variables and their interactions, which variables emerge with statistically significant effects?

[Table 5c about here]

The coefficient estimate of the baseline term, ∆∆∆∆Capital(s-1), is positive and statistically

significant in both the overall and the emerging market model The hypothesis that capital account liberalization leads to growth is maintained The level of capital account openness also has a positive and highly statistically significant coefficient Only ∆∆∆∆Capital(s-1)*Democracy(s-2), of the other interaction

terms, has a coefficient estimate that is statistically significant in both models, and it is negative This implies that democracies appear not to benefit as much from capital account liberalization as other nations

do The coefficient estimate for ∆∆∆∆Capital(s-1)*RevCoups(s-2) is positive and statistically significant at the 1 level in the emerging market nation model

In Table 5d, we examine the interaction effects between capital account liberalization and prior social conditions: educational attainment, population growth, and two measures of ethnic or linguistic fragmentation None of the interaction terms is statistically significant

Note, however, that ELF60 and ETHFRAC have opposite signs This seems to derive from how many Latin American countries are treated The linguistically derived ELF60 measure shows Latin America to be relatively homogenous The ethnically derived ETHFRAC shows Latin America to be diverse, in contrast Given the widespread use of ELF60 and the pessimistic implications that are derived from its analysis, scholars should replicate their ELF60 results with Krain’s 1997 ETHFRAC to insure robustness

In reviewing the results of the interaction analysis, we find that Liberalization has a direct effect generally on growth The level of a nation’s democracy appears to be the main mediator for

Liberalization’s effects We are unable to tell from this analysis, however, whether democracies are hurt

by capital account liberalization or are not helped

In light of the robustness of the democracy results, we next explore further the role of democracy

We distinguish among four groups of countries: continuously democratic advanced industrial nations; continuously democratic emerging market nations; continuously autocratic nations; and transitional polities (Please see footnotes 1, 2, and 8 for a list of the countries.) We undertake two types of

analyses First, we create dummy variables representing the groups of countries, and interaction terms for the capital variables for each grouping We enter these into the regression analyses, using the transitional polities as the base case Second, we undertake partial residual plots, and derive the coefficient estimates for capital account liberalization for each grouping of countries See Appendix B, which describes the procedure for partial residual plots, and which reports the results of the analyses

[Table 6 about here]

What we find, from Table 6, model 6.1-6.3 and from the partial residual plot analysis, is that capital account liberalization has different effects among the groupings The baseline coefficients for Liberalization, which represent the experiences of 34 countries that were neither continuously democratic nor autocratic, are positive and highly statistically significant In advanced industrial democracies, the coefficient estimates for Liberalization (Openness) are also positive and highly statistically significant, once proper adjustments are made.12 The coefficient estimate for Liberalization is larger for the advanced industrial democracies than it is for the baseline countries

12 The coefficient estimates for the interaction variables are added to the baseline coefficient, and we obtain the

corresponding standard error taking into account the covariance between the coefficients The Liberalization coefficient estimate for OECD democracies is 0.0437 with a t-stat of 3.511; the Liberalization coefficient estimate for emerging market democracies is –0.032 with a t-stat of –1.157; and the Liberalization coefficient estimate for continuous autocracies is 0.0344 with a t-stat of 0.992