Regional GDP per Capita Convergence: The Role of
Regional GDP per Capita Convergence: The Role of Financial Development and Trade Openness Francisco A. Gallego Ponti cia Universidad Catolica de Chile, Department of Economics and Economic History and Cliometrics Lab (EH Clio Lab) Mar a Valentina Konow MIDEPLAN, Chile This version: September 2011
Abstract Several papers present evidence of the existence of regional convergence in GDP per capita for di erent countries and time periods. However, no paper has studied whether this convergence process depends upon characteristics of the economy. In this paper, we attempt to ll this gap by taking advantage of the Chilean experience from 1960 to 2005, a country that has implemented a series of economic reforms. We nd evidence of both sigma- and beta-convergence processes taking place, but also nd that di erent economic reforms have di erent e ects on this convergence process: while nancial development a ects positively convergence across regions (consistent with a simple growth model with borrowing constraints), trade openness seems to decrease convergence (consistent with a simple trade model). Results are robust to the inclusion of controls for other economic and political reforms. We also check for the external validity of our results and provide cross-country evidence for a small sample of countries that is consistent with our results for Chile. Keywords: Regional convergence, nancial development, trade openness, Chile, sigma convergence, beta convergence JEL codes: O18, R11, G2, O16.
Authors' email address: [email protected], [email protected]. We would like to thank Rodrigo Cerda, Jose Diaz, Rolf Luders, Raimundo Soto, Gert Wagner and participants in the 2010 LACEA meetings and a PUC-Chile seminar for useful comments and Felipe Gonzalez, Francisco Mu~ noz, and Diego Verdugo for superb research assistance. We have bene ted from nancial support from the Millenium Nuclei Research in Social Sciences, Planning Ministry (MIDEPLAN), Republic of Chile. The usual disclaimer applies.
1
Introduction
Levels of economic development vary almost as widely across states or regions within a country as they do across countries. For example, Bruhn and Gallego (2011) document that for a sample of 16 countries in Latin America around 2000, while the richest country (Argentina) has 5:5 times the GDP per capita of the poorest country (Honduras), at the same in average the ratio of the maximum to minimum GDP per capita within regions of the same country is above six. Several papers have identi ed the speed of GDP per-capita convergence across regions of the same country{evidence in favor of convergence{ nding typically a relatively slow convergence process (eg., Barro and Sala-i-Martin, 1991, 2003 for evidence on regions of Europe, Japan, and the US; and D az and Meller, 2003, Fuentes, 1997, Morande et al., 1995, Soto and Torche, 2004, Duncan and Fuentes, 2006, and Badia-Miro, 2008 for Chile). In parallel a number of recent papers study how some institutions and policies a ect the levels of economic development across and within countries (eg., see the literature review in Nunn, 2009) and also the speed of convergence across countries (eg., Aghion et al, 2005 for the e ect of nancial development on the country-level speed of convergence). To our knowledge, this paper provides the rst attempt to combine both literatures, by empirically studying how the evolution of a particular set of institutions and policies at the country level a ect the speed of convergence across regions of the same country. We study the case of Chile over the 1960-2006 period. Chile is an interesting case because over the analyzed period (i) there was a signi cant increase in country-level GDP per capita (PPP) from about US$900 in 1960 to about US$10,400 in 2000 (Penn World Tables, revision 6.1), (ii) there was a signi cant decrease in the standard deviation of regional (log) GDP per-capita from 0.68 in 1960 to 0.48 in 2006 (ie. evidence in favor of convergence concept), (iii) there seems to be a relatively slow process of convergence across di erent regions of the country (documented in several papers), and (iv) there was a deep process of institutional and policy reform: the country moved from a closed, heavily regulated, and nancially underdeveloped economy from the 1960s to the mid 1970s, to an open, deregulated and nancially developed economy in the 1990s. We hypothesize that these economic and policy reforms a ect the speed of convergence across regions of the country. Using the Chilean case, we nd that di erent policy and institutional dimensions have di erentiate e ects on the regional convergence process. While, nancial development has a positive and economically and statistically signi cant e ect on both
and
convergence{in a consistent way with a simple growth model with nancial constraints{, 1
trade openness decreases
and
convergence {consistent with a simple trade model in
which di erent regions have di erent endowments. Other economic and institutional dimensions, such as infrastructure development, schooling, democracy and macroeconomic stability, have less robust and economically relevant impacts on the convergence process. We check for the external validity of our results and run cross-country regressions for 26 countries for which we have data on regional GDP per capita. Our results are consistent with the evidence for Chile: nancial development has a negative and economically and statistically signi cant e ect on the standard deviation of GDP per capita. The e ects for trade openness are also consistent with our evidence for Chile, but results are only marginally signi cant and economically less relevant. The results in this paper are interesting from several perspectives and contribute to several literatures. First, our result that nancial development a ects the convergence process is present in the cross-country literature (eg., Aghion et al., 2005). In particular, Aghion et al. (2005) present a similar empirical result using cross-country data and develop a simple theoretical model with nancial frictions in which nancial development reduces the extent of nancial constraints and allows countries to take advantage of profitable projects. A similar rationale could be applied in our case, in which entrepreneurs in di erent regions bene t from nancial development by being able to implement the pro table projects already available and, therefore, move faster to the steady state GDP per capita. Second, our results also relate with the literature that studies the e ect of trade openness on regional development. Kim (2008) argues that international trade a ects regional patterns of production thus potentially a ecting regional inequality. The argument could be rationalize using a simple Heckscher{Ohlin argument: when there is regional heterogeneity in endowments, trade openness may potentiate the patterns of comparative advantage of di erent regions and, therefore, create regional di erences in development. However, from a conceptual point of view, the e ect of trade on regional development would depend on the initial level of distortions, under some conditions opening up to trade may reduce inequality by decreasing distortions that tend to concentrate production in some areas. From an empirical point of view, Kanbur and Venables (2005), Kanbur and Zhang (2005) and Rodr guez-Pose and Sanchez-Reaza (2005) provide evidence of a positive e ect of trade openness on regional inequality for di erent countries.1 1
We want to stress that our theoretical and empirical focus is not on the e ect of policies and institutions on the level of GDP per capita, but on the e ects of these policies and institutions on the convergence process.
2
The paper is organized as follows. Section 2 discusses the theoretical background and previous literature related to this paper. Section 3 describes the data. Section 4 analyzes the relationship between economic reforms and both sigma- and beta-convergence across Chilean regions in the 1960-2006 period. Section 5 presents some cross-country evidence consistent with our results and Section 6 concludes. 2
Motivating Theory
We present the theoretical motivation for studying the potential e ects of institutions and policies at the country level on regional GDP convergence. 2.1
On
and
convergence
To x ideas, we start de ning ways of measuring GDP per capita convergence across regions. There are two di erent, but related, concepts of convergence: (Barro and Sala-i-Martin, 2003).
and
convergence
convergence relates to the situations in which, ce-
teris paribus, poorer regions grow faster than richer regions. The process of
convergence
could be unconditional, when all regions converge to the GDP per capita in steady state, or conditional, when regions converge to di erent steady state levels. This
convergence process is typically associated to estimating equations of the
form: yit
yit
1
=
yit
1
+
t
+
i
+
it :
(1)
Subscript i refers to regions and t to years. y is the log of GDP per capita, time e ects that are common to all regions,
are region-speci c e ects, and
i
idiosyncratic shocks.2 Unconditional convergence implies that from 0. Estimating
i
are are
should not be di erent
convergence yields to estimating a positive value for
equation (1). To get a more structural explanation for
t
in the
notice that given the discrete
nature of the data (Barro and Sala-i-Martin, 2003), =1 where
e
;
is the speed of convergence to the steady state. Thus, from equation (1) we
can recover . Our claim is that
(and, therefore, ) is not constant and depends on
some institutional features that vary over time at the country level. In particular, in our regressions we model it as t
= + p0t ;
2
(2)
We abstract from region characteristics that vary over time because we do not have a long spam of data including them.
3
where p is a vector of institutions or policies that vary over time (including nancial development, trade openness, infrastructure, and democracy, among others). A test of 6= 0.
our hypothesis {which we explain in more detail below{is that The concept of
convergence, ie. whether the volatility of y decreases or not as
time passes, is related to this idea. To x ideas, for simplicity, let us assume that there v (0;
are no time and region-speci c xed e ects and that 2 t
where
2 t
=
2
+
2 0 2
is the variance of yi at time t,
which is equal to that when
2 u
> 0,
, and
2
1 e 2 t
2 0
2
e
2 t
). Then,
;
(3)
is the variance of yi at the steady state,
is the initial variance of yi . Thus, this equation implies 2 0
decreases over time only when
convergence implies
2
>
2
. Putting it di erently,
convergence only when the country starts with a distribution
of GDP per capita across regions that is "high" (ie. it is above its steady state level). If this is the case, as t increases,
2 t
decreases.
The previous analysis implies that the existence of and su cient condition for the existence of tence of
convergence is a necessary
convergence, but, in contrast, the exis-
convergence is a necessary but not su cient condition for the existence of
convergence. This shows the connection between both measures of regional GDP convergence. If we extend the analysis of
convergence to the inclusion of time varying
it is implicit in equation (2) ; we may expect that if (decrease) the speed of convergence decrease (increase)
2 0
>
2
, as
, variables that increase
convergence.
We use this basic measurement framework and the relationship between both convergence concepts in the empirical section of the paper. Now we move to discuss theories and previous research that suggest a relationship between trade openness and nancial development and GDP convergence across regions. 2.2
Theoretical Motivation and Previous Literature
A simple version of the Solow-Swan growth model gives us a basic motivation for our idea.3 Under the basic assumptions of this model, regions have di erent initial capital stocks per-capita and, therefore, di erent marginal productivities.4 These di erences in productivities imply di erences in returns and create incentives for factor movements 3
Most models related to regional convergence process assume that regions (a partially arbitrary de nition in most cases) are di erent economies (in terms of fundamental determinants of income and growth) and, therefore, it makes sense to study a convergence process. We make the same assumption. 4 For simplicity, we abstract from a distinction between population and workers.
4
that at the steady state will equalize capital per capita across regions. Empirical estimates of the regional speed of convergence tend to identify relatively low rates of around 2% (eg., see Barro and Sala-i-Martin, 2003 for evidence on regions of Europe, Japan, and the US). Several papers identify similar annual rates of convergence for Chile, with point estimates going from 1% (D az and Meller, 2003) to 2:7% (Fuentes, 1997), with other estimates in between (eg., Morande et al., 1995; Soto and Torche, 2004; Badia-Miro, 2008). Going to the topic of our paper, notice that the transitional dynamics in the SolowSwan is result of an implicit assumption related to the existence of borrowing constraints that make the investment level constrained by the current level of savings thus creating, jointly with the assumptions on the production function, transition dynamics. Thus, when nancial development takes place this borrowing constraint is less binding and, therefore, regions should converge more quickly to the steady state. An analogous result could emerge in an endogenous growth model, along the lines of Aghion et al. (2005), in which the speed of convergence of di erent results to the steady state is also related to the existence of borrowing constraints of entrepreneurs to invest in new technologies. Aghion et al. (2005) present empirical estimates for countries in which nancial development increase the speed of convergence towards the steady state. Guiso et al. (2002) present evidence that nancial development at the regional level has a positive e ect on growth. Notice that this is conceptually and empirically di erent than our argument: we claim that nancial development at the country level increases the speed of convergence. To our knowledge, no paper studies empirically this point. When we move away from one-sector frameworks, as the Solow-Swan model, we may have than the e ect of national policies and institutions depend upon the complementarity between these policies and region endowments and characteristics. For instance, one may take the case of the Hecksher-Ohlin model and the impact of trade openness on the levels and the variability of regional output. In this case, when there are di erences in factor endowments at the regional level, trade openness may increase regional output disparities when trade have di erent e ects on regions with di erent endowments. From a more general point of view, in multi-sector models, policy and institutional changes that have di erent degrees of complementarity with di erent regions may increase regional disparities and, therefore, decrease the speed of convergence. Kim (2008) presents a review of the empirical and theoretical literature about the impact of di erent institutions and policies on patterns of regional output. He argues that trade openness tend to be associated with increases in the degree of concentration of economic activity
5
and, therefore, of GDP per-capita (Kanbur and Venables, 2005; Kanbur and Zhang, 2005; Rodr guez-Pose and Sanchez-Reaza, 2005).5 . In the case of nancial development, this literature does not discuss it explicitly, but it could also be the case that di erent regions of the country tend to have di erent nancial needs (due to, for instance, to di erences in the production structure, along the lines of the cross-country papers by Levchenko, 2004 and Nunn, 2007) and, therefore, decrease the regional process of convergence. To our knowledge, no paper has studied this point. In sum, the main hypotheses we test in this paper are the following: The higher the level of nancial development at the country level the faster both convergence (ie. the lower the variance of log GDP per-capita) and -convergence (ie. the higher the speed of convergence in log GDP per-capita) The higher the level of trade openness at the country level the slower both convergence and
-
-convergence (ie. the smaller the speed of convergence in log
GDP per-capita) 3
Data
We use several datasets in this paper. Table 1 presents the variables used, the level at which each variable is collected, and the main descriptive statistics of each variable. National account estimates of GDP at the regional level are only available since 1985. Thus, we combine the o cial data from the Central Bank of Chile with data from Bonacic and Vial (1994), which is available since 1960 to 1992. We use population series from the Chilean Statistics Bureau (INE) to generate the per capita GDP series. We measure nancial development with the ratio of nancial intermediation to GDP from D az et al. (2009), which includes total bank deposits, domestic public debt, total mortgage debt, and the value of the stock market.6 We measure trade openness as the ratio of exports plus imports to annual GDP (also from D az et al., 2009). In addition, we also include, as control variables, other variables related to policies and institutions at the country level: infrastructure development (the annual percentage of total roads that are paved, the ratio of total roads to total population and the number of telephones every thousand inhabitants, from D az et al., 2009), a democracy index 5
He also presents arguments for the potential impact of other variables, such as, democracy, decentralization, and infrastructure. In this case, however, the theoretical rationale for most of these variable is not as clear as in the case of trade openness. 6 This index is closely related to the traditional proxies for nancial development. See Levine (2005) for a detailed discussion on the measurement of nancial development.
6
(restrictions to executive power and the level of political competition, from the updated version of Vanhanen, 2000), the average years of schooling of the labor force (D az et al., 2009), terms of trade (from D az et al., 2009), an index of structural reforms (from Fuentes, Larra n and Schmidt-Hebbel, 2004), and a proxy for macroeconomic stability (measure as
1 1+
;where
is the in ation rate). In the cross-country exercises, we use
information on GDP per capita from Bruhn and Gallego (2009) several other local sources. Information on policies and other controls come from Beck et al. (2002) and Gallup et al. (1999). We now take a preliminary look to the Chilean data. Figure 1 presents the Kernel estimates of the distribution of log GDP per-capita for 13 the Chilean regions for decennial observations from 1960 to 2000. It is interesting to notice how the distribution starts moving to the right from the 1980s and also how the distribution tends to become more concentrated in the recent periods. This is consistent with the idea of having convergence across regions and that the convergence process is not constant. Figure 2 complements this with the evolution of the standard deviation of log GDP per-capita from 1960 to 2006. The gure presents both the simple calculation of the standard deviation and the standard deviation weighted by regional population (in two versions: weighted by contemporaneous and initial population). There is a negative trend with some uctuations in both deviations (which, despite di erences in levels, are highly correlated among them). It is worth noting three main periods, (i) one that goes until 1975 with a strong decrease in the standard deviation, (ii) one that goes between 1983 with an increase in the variability of regional GDP, and (iii) a nal period with a strong decline in the standard deviation from 1983 to 2006 (with a small increase between 1997 and 2001). This suggests that from a uni-variate perspective the convergence process does not seem to be uniform. Finally, the similarity between
weighted by
initial and contemporaneous population makes clear that population movements are not a signi cant source of convergence in Chile (a fact already noted and discussed in detail by Soto and Torche, 2004). Finally, Figure 3 presents the evolution of our key variables: trade openness and nancial development. It is important notice that while both variables are correlated their evolution is di erent both in terms of levels and timing. While trade openness increases from about 0.20 in the 1960s to a peak around 0.70 in the 1990s, nancial development increases from about 0.10 to values around 1.50. In addition, the timing di ers, while both variables increase from the mid-1970s and have a small decrease in the early 1980s, the big increase in nancial development occurs from the mid 1980s.
7
This di erences in timing are important from an empirical point of view to isolate the di erential impact of both policy dimensions. 4
The Effects of Economic Reforms on Regional Convergence
In this section we present estimates on how nancial development and trade openness a ect both convergence concepts. We start presenting estimates of the determinants of convergence since a failure to identify an impact of nancial development and trade openness implies that they do not have an impact on nd an impact of these variables on on
convergence. Similarly, if we
convergence, we have to nd a consistent impact
convergence. Thus, this approach is useful to double-check the implications of
our theory using two di erent approaches. 4.1
Estimates for -Convergence
Our main estimating equation is t
where
=
+
o Ot 1
+
F Ft 1
+
T Tt
+ Xt0
1
+
t:
(4)
is the standard deviation of log GDP per capita, O is trade openness, F is
nancial development, T is a linear trend, X is the vector of controls mentioned in section 3, and
is a shock. Our main hypothesis is that
o
> 0 and
F
< 0. In
addition, we expect that if there is a convergence process that does not depend upon other variables,
T
< 0, as suggested by equation (3). We estimate using lags of the
right-hand side variables to avoid mechanical potential reverse causality.7 Estimating this equation in the case of Chile is complicated because many policy and institutional dimensions move in the same direction, so there is a non-trivial level of multi-colinearity. Thus our approach is to run di erent estimation in which we include di erent combinations of variables to study the robustness of the results. As previously discussed, the fact that both the levels and especially the timing of the policy reforms make it possible to estimate di erential policy e ects. Table 2 presents the results of estimating equation (4). In several columns we estimate with di erent combinations of control variables to test for robustness. The main results imply that F tend to have a negative, statistically signi cant (except in one case), and economically relevant e ect on . The (short-run) point estimates vary depending 7
We inlcude also the lag of the dependent variable in the equations given (i) the level of inertia observed in the data and (ii) the fact that by construction is present an AR(1) process when there is convergence. See Barro and Sala-i-Martin (2003) for more details. Results on the e ects O and F do not change in a signi cant way when we exclude the lag.
8
on the speci cation, but our estimations imply that the increase in nancial development we observe between 1960 and 2000 would explain between 8 and 12% of the big decrease in
we observe over the same period. In contrast, trade openness seem to have a positive, statistically signi cant (except
in one case too), and economically relevant e ect on . Our estimates in this case imply that the increase in trade openness we observe between 1960 and 2000 would increase by between 13 and 28% of the level we observe in this period. Regarding the e ect of control variables, they tend to have smaller and non-robust impacts on . For instance, the level of terms of trade shocks, capturing cyclical uctuations, seems to have a small negative impact on , with periods with positive term of trade shocks having negative impact on
. In turn, infrastructure and schooling
variables have a non-clear pattern (both in terms of signs and economic and statistical signi cance). 4.2
Estimates for -Convergence
The results in the previous section present initial evidence that convergence across regions is related to trade openness and nancial development in ways consistent with our theoretical discussion. Now, we move to an additional test of these hypotheses. In this case, we study how
convergence is related to institutions and policies. To do so, our
main estimating equation is (a transformation of): yit yit 5 + Ft F + Ot O + Xt0 yit 5 + t + i + it ; (5) = 5 which combines equations (1) and (2). We are interested in the term in parenthesis, ie. how
convergence is a ected by a vector of policies and institutions. Our previous
results imply that
F
of time dummies (ie.
< 0 and t)
O
> 0. In addition, notice that the inclusion of a vector
imply that we are controlling for any direct e ect of these
policies and institutions on income levels. So equation (5) only identi es any e ect of F and O operating through the interaction with initial development. As customary, we estimate (5) using a 5-year panel. In order to deal with traditional endogeneity problems in dynamic models with xed e ects, we estimates the regressions using the GMM methodology suggested by ArellanoBond. As known, this methodology uses lags as instruments (ie. internal instruments) and, therefore, in all the speci cations we present two speci cation tests: (i) the Sargan test that tests whether moment conditions for identi cation are supported by the data and (ii) the AR(2) test that studies whether lags of the dependent variables are valid instruments. 9
Table 3 presents the results. We start by just presenting the typical regional convergence regression{ie., without including the interaction terms in (5) but including region and time e ects. Estimates imply a speed of convergence of about 1:7%, in line with previous studies for Chile and for other countries. This is a small speed of convergence that implies that half of the gap between current and steady state output per capita will be closed in about 41 years. The results in next columns follow the same organization as in Table 2. We present several speci cations trying to how robust the estimates are to the inclusion of other control variables. Consistently with results in Table 2, nancial development presents a negative, statistically signi cant (expect in one case), and economically relevant e ect on the speed of convergence. The increase in nancial development between 1960 and 2005 implies that the convergence speed would increase to between 4% and 20% depending on the speci cation. In contrast, trade openness seems to decrease convergence. The interaction between trade openness and initial output gap is signi cant in almost all columns (except in one case). Again the e ect is economically signi cant: the increase in trade openness observed between 1960 and 2005 would have decreased the speed of convergence by between 4 and 7 p.p. This is also an economically relevant e ect. Regarding the other determinants of
convergence, we observe a robustly negative
impact of terms of trade on the speed of convergence, probably implying that the speed of convergence increases in good economic cycles. However notice that this e ect is (i) economically irrelevant, with speed of convergence increase around 2% when terms of trade increase by one standard deviation and (ii) not consistent with results in Table 2 for
convergence. Something similar occurs in relation with the democracy index and
with improvements in the availability of phone lines. In all, results regarding nancial development and trade openness are consistent with our previous theoretical discussion and present a pattern of economically relevant e ects, in particular in the case of trade openness. 5
Some Cross-Country Evidence
In the previous evidence we focus in Chile, as a particular example of a country having undertaken a signi cant process of trade and nancial development in the last 50 years and having at the same time experienced a process of regional convergence. In this section we explore, as a check for "external" validity, whether the same pattern appears in a sample of 26 countries for which we have data on regional GDP. We constructed
10
measures of the standard deviation of log GDP per capita for these countries and run the following regression: i
=
+
o Oi
+
F Fi
+ +Xi0 + %t :
where i refers to country. We include the log of GDP per capita at the country level, a vector of soil characteristics (from Gallup et al., 1999), and the log of the number of regions in the country in the vector of controls. If the pattern we found using time series evidence for Chile is also present in the cross section of countries, then we expect that o
> 0 and
F
< 0.
Table 4 presents the results. We rst estimate the regression only including nancial development and
nd a negative, statistically, and economically signi cant negative
correlation of this variable with
convergence. The correlation implies that a one
standard deviation increase in nancial development is associated with a decrease of of about 0.57 standard deviations. Next, we just include trade openness in the regression and nd a positive, but statistically and economically insigni cant correlation with (the standardized e ect is just about 0.06 standard deviations of ). In the next column, we include jointly both variables and nd estimates with similar signs, but with bigger e ects of both variables. The standardized e ect of nancial development increases to about 0.80 and the standardized e ect of trade openness increases to about 0.25, but the latter is just marginally signi cant (p-value of 0:11). Figure 4 presents added variable plots for the two variables of interest for this speci cation.8 In the nal two columns of the Table, we study the robustness of the previous results by including GDP per-capita and a vector of controls for soil characteristics. In both cases, nancial development remains signi cant with a big standardized e ect.9 In all, the correlations presented in this section suggest the pattern of an association between
and nancial development and trade openness which is, especially in the
case of nancial development, consistent with our previous evidence, thus providing an additional check of the external validity of our results. Given the small sample we use in these regressions, we take this evidence as additional supporting evidence that should be studied in more detail in further research. 8 Excluding Ecuador from these regressions changes only marginally the estimated e ects and their statistical and economic signi cance. 9 We also experimented with IV regressions to deal with potential endogeneity problems using (i) the legal origins variables from Laporta et al.(2008) as instruments for nancial development and (ii) the predicted trade variable from Frankel and Romer (1999) as an instrument for trade openness. In both cases the Hausman tests do not reject the null hypotheses that the OLS estimates are equal to IV estimates (with p-values of 0.39 and 0.90 for nancial development and trade openness, respectively) and therefore OLS estimates are preferred.
11
6
Concluding Remarks
The potential e ects of policies and institutions on development at the country level has been a much debated topic in the literature. Our study of the e ects of policies on the convergence process at the regional level can help us to understand the e ects of policies and institutions on other dimensions of the development process. Previous research has mainly focused on identifying the existence or not of convergence processes at the regional level, while leaving the study of the determinants of this convergence process open. We document that two particular country-level policies that have a theoretical motivation a ect the convergence process as expected: while nancial development makes the convergence process faster, trade openness produces the opposite e ect. We nd these results using two di erent, but related, views of the convergence process: the study of both
and
convergence. In both cases the impact of both trade openness and
nancial development are statistically and economically relevant. These results imply that economic policies do not only a ect the level of development, but also may a ect the heterogeneity of development within countries. Given the big di erences in development within countries we observe, our results give an additional role to nancial policies. We focus in Chile, as a particular example of a country having undertaken a significant process of trade and nancial development in the last 50 years and having at the same time experienced a process of regional convergence. However, we also show in an exploratory way that the same pattern of e ects appear when we analyze a sample of 26 countries for which we have measures of the standard deviation of log GDP per capita within country. Further research should study this hypothesis including more countries and estimating the impact also on
convergence. References
[1] Aghion P, Howitt P and Mayer D (2005) The E ect of Financial Development on Convergence, Quarterly Journal of Economics. [2] Badia Miro, Marc. 2008. La localizacion de la actividad economica en Chile,18701973. PhD thesis, Universitat de Barcelona. [3] Barro, R., Sala-i-Martin, X., 1991, Convergence across States and Regions, Brookings Papers on Economic Activity. [4] Barro, R., Sala-i-Martin, X., 2003, Economic Growth, MIT Press. 12
[5] Beck, Thorsten & Levine, Ross & Loayza, Norman, 2000. "Finance and the sources of growth," Journal of Financial Economics, Elsevier, vol. 58(1-2), pages 261-300. [6] Bonacic, C., Vial, J., 1994, "Evolucion del Producto por Regiones 1960-1992", Coleccion de estudios CIEPLAN Nro.39. [7] Bruhn, M. and F. Gallego (2011) "Good, Bad, and Ugly Colonial Activities: Do They Matter for Economic Development?" The Review of Economics and Statistics. [8] D az, R., Meller, P., 2004, "Crecimiento Economico Regional en Chile: >Convergencia?", Universidad de Chile, Departamento de Ingenier a Industrial, Working Paper. [9] D az, J, R. Luders, and G. Wagner, 2009. La Republica en Cifras. Dataset available at the Economic History and Cliometrics Lab: http://www.economia.puc.cl/cliolab. [10] Duncan, Roberto; Rodrigo Fuentes. 2006. "Regional Convergence in Chile: New Tests, Old Results" Cuadernos de Econom a, Vol. 43, N 127, pp. 81-112. [11] Frankel, J. and Romer. 1999. "Does Trade Cause Growth?" American Economic Review. [12] Fuentes, R., 1996, ">Convergen las Regiones en Chile? Una Interpretacion" in Analisis Emp rico del Crecimiento en Chile, Centro de Estudios Publicos, Santiago. [13] Fuentes,R., Larra n, M.,Schmidt-Hebbel, K., 2004, Fuentes del crecimiento y comportamiento de la productividad total de factores en Chile, Working Paper 287, Banco Central de Chile. [14] Gallup, JL, J.D. Sachs and A. Mellinger (1999), "Geography and economic development", CID Working Paper Vol. 1, Harvard, MA. [15] Guiso, L., Sapienza, P., Zingales, L., 2002, Does Local Financial Development Matter?, Working paper # 8923, NBER. [16] Kanbur, R. Vanables, A., 2005, "Spatial Inequality and Development", Review of Development Economics. [17] Kanbur, R., Zhang, X., (2005), \Fifty Year of Regional Inequality in China: a Journey through central planning, reforms and opennes", Review of Development Economics. 13
[18] Kim, S., 2008, "Spatial Inequality and Economic Development: Theories, Facts, Policies", Working Paper nr. 16, Comission on Growth and Development. [19] La Porta, R., F. Lopez-de-Silanes and A. Shleifer (2008) "The Economic Consequences of Legal Origins", Journal of Economic Literature, June. [20] Levchenko, Andrei, (2004) \Institutional Quality and International Trade," IMF Working Paper WP/04/231, International Monetary Fund. [21] Levine, R. 2005. "Finance and Growth" in Handbook of Economic Growth. [22] Morande, F., Soto, R., Pincheira, P., 1996, Achilles, the Tortoise, and Regional Growth in Chile" in \Analisis Emp rico del Crecimiento en Chile, Centro de Estudios Publicos, Santiago. [23] Nunn, Nathan (2007) "Relationship-Speci city, Incomplete Contracts and the Pattern of Trade," Quarterly Journal of Economics, Vol. 122, No. 2, 569-600. [24] Rodriguez-Pose, A., Sachez-Reaza, J., 2005, \Economic Polarization through trade liberalization and regional growth in Mexico", in Spatial Inequality and Development, Oxford University Press. [25] Soto, R., Torche, A., 2004, "Spatial Inequality, Migration, and Economic Growth in Chile", Cuadernos de Econom a, Vol. 41. [26] Vanhanen, T. (2000) "A New Dataset for Measuring Democracy, 1810-1998". Journal of Peace Research, Vol. 37, No. 2, 251-265.
14
"& "% '()*(+,-(*./012 "$ "# ! !3
! 3 Figure 1: Distribution of Log GDP per capita, different years
#!!!
.55
.7
.35 .4 .45 .5 Sigma weighted by population
.65 .6 Sigma .55 .5
.3
.45 .4 1960
1970
1980
1990
2000
2010
year Sigma
Sigma weighted by population
Sigma weighted by 1960 population
Figure 2: Standard Deviation of Log GDP per capita, 1960-2005
.6 .5 Openness .4 .3 .2
1970
1980
1990
2000
2010
1960
1970
1980
1990
2000
2010
0
50
Financial Development 100 150
200
1960
Figure 3: Trade Openness and Financial Development, 1960-2000
14! .-1
$-. /*24' +2& +%& !*0
$%& !"#
&3. /-"
/.3 +.(
+-'
#&5
'() 5*(
,-" .,#
2() 12. 3*( ,#.
(6&
!78
#;@)ABC
Abstract Several papers present evidence of the existence of regional convergence in GDP per capita for di erent countries and time periods. However, no paper has studied whether this convergence process depends upon characteristics of the economy. In this paper, we attempt to ll this gap by taking advantage of the Chilean experience from 1960 to 2005, a country that has implemented a series of economic reforms. We nd evidence of both sigma- and beta-convergence processes taking place, but also nd that di erent economic reforms have di erent e ects on this convergence process: while nancial development a ects positively convergence across regions (consistent with a simple growth model with borrowing constraints), trade openness seems to decrease convergence (consistent with a simple trade model). Results are robust to the inclusion of controls for other economic and political reforms. We also check for the external validity of our results and provide cross-country evidence for a small sample of countries that is consistent with our results for Chile. Keywords: Regional convergence, nancial development, trade openness, Chile, sigma convergence, beta convergence JEL codes: O18, R11, G2, O16.
Authors' email address: [email protected], [email protected]. We would like to thank Rodrigo Cerda, Jose Diaz, Rolf Luders, Raimundo Soto, Gert Wagner and participants in the 2010 LACEA meetings and a PUC-Chile seminar for useful comments and Felipe Gonzalez, Francisco Mu~ noz, and Diego Verdugo for superb research assistance. We have bene ted from nancial support from the Millenium Nuclei Research in Social Sciences, Planning Ministry (MIDEPLAN), Republic of Chile. The usual disclaimer applies.
1
Introduction
Levels of economic development vary almost as widely across states or regions within a country as they do across countries. For example, Bruhn and Gallego (2011) document that for a sample of 16 countries in Latin America around 2000, while the richest country (Argentina) has 5:5 times the GDP per capita of the poorest country (Honduras), at the same in average the ratio of the maximum to minimum GDP per capita within regions of the same country is above six. Several papers have identi ed the speed of GDP per-capita convergence across regions of the same country{evidence in favor of convergence{ nding typically a relatively slow convergence process (eg., Barro and Sala-i-Martin, 1991, 2003 for evidence on regions of Europe, Japan, and the US; and D az and Meller, 2003, Fuentes, 1997, Morande et al., 1995, Soto and Torche, 2004, Duncan and Fuentes, 2006, and Badia-Miro, 2008 for Chile). In parallel a number of recent papers study how some institutions and policies a ect the levels of economic development across and within countries (eg., see the literature review in Nunn, 2009) and also the speed of convergence across countries (eg., Aghion et al, 2005 for the e ect of nancial development on the country-level speed of convergence). To our knowledge, this paper provides the rst attempt to combine both literatures, by empirically studying how the evolution of a particular set of institutions and policies at the country level a ect the speed of convergence across regions of the same country. We study the case of Chile over the 1960-2006 period. Chile is an interesting case because over the analyzed period (i) there was a signi cant increase in country-level GDP per capita (PPP) from about US$900 in 1960 to about US$10,400 in 2000 (Penn World Tables, revision 6.1), (ii) there was a signi cant decrease in the standard deviation of regional (log) GDP per-capita from 0.68 in 1960 to 0.48 in 2006 (ie. evidence in favor of convergence concept), (iii) there seems to be a relatively slow process of convergence across di erent regions of the country (documented in several papers), and (iv) there was a deep process of institutional and policy reform: the country moved from a closed, heavily regulated, and nancially underdeveloped economy from the 1960s to the mid 1970s, to an open, deregulated and nancially developed economy in the 1990s. We hypothesize that these economic and policy reforms a ect the speed of convergence across regions of the country. Using the Chilean case, we nd that di erent policy and institutional dimensions have di erentiate e ects on the regional convergence process. While, nancial development has a positive and economically and statistically signi cant e ect on both
and
convergence{in a consistent way with a simple growth model with nancial constraints{, 1
trade openness decreases
and
convergence {consistent with a simple trade model in
which di erent regions have di erent endowments. Other economic and institutional dimensions, such as infrastructure development, schooling, democracy and macroeconomic stability, have less robust and economically relevant impacts on the convergence process. We check for the external validity of our results and run cross-country regressions for 26 countries for which we have data on regional GDP per capita. Our results are consistent with the evidence for Chile: nancial development has a negative and economically and statistically signi cant e ect on the standard deviation of GDP per capita. The e ects for trade openness are also consistent with our evidence for Chile, but results are only marginally signi cant and economically less relevant. The results in this paper are interesting from several perspectives and contribute to several literatures. First, our result that nancial development a ects the convergence process is present in the cross-country literature (eg., Aghion et al., 2005). In particular, Aghion et al. (2005) present a similar empirical result using cross-country data and develop a simple theoretical model with nancial frictions in which nancial development reduces the extent of nancial constraints and allows countries to take advantage of profitable projects. A similar rationale could be applied in our case, in which entrepreneurs in di erent regions bene t from nancial development by being able to implement the pro table projects already available and, therefore, move faster to the steady state GDP per capita. Second, our results also relate with the literature that studies the e ect of trade openness on regional development. Kim (2008) argues that international trade a ects regional patterns of production thus potentially a ecting regional inequality. The argument could be rationalize using a simple Heckscher{Ohlin argument: when there is regional heterogeneity in endowments, trade openness may potentiate the patterns of comparative advantage of di erent regions and, therefore, create regional di erences in development. However, from a conceptual point of view, the e ect of trade on regional development would depend on the initial level of distortions, under some conditions opening up to trade may reduce inequality by decreasing distortions that tend to concentrate production in some areas. From an empirical point of view, Kanbur and Venables (2005), Kanbur and Zhang (2005) and Rodr guez-Pose and Sanchez-Reaza (2005) provide evidence of a positive e ect of trade openness on regional inequality for di erent countries.1 1
We want to stress that our theoretical and empirical focus is not on the e ect of policies and institutions on the level of GDP per capita, but on the e ects of these policies and institutions on the convergence process.
2
The paper is organized as follows. Section 2 discusses the theoretical background and previous literature related to this paper. Section 3 describes the data. Section 4 analyzes the relationship between economic reforms and both sigma- and beta-convergence across Chilean regions in the 1960-2006 period. Section 5 presents some cross-country evidence consistent with our results and Section 6 concludes. 2
Motivating Theory
We present the theoretical motivation for studying the potential e ects of institutions and policies at the country level on regional GDP convergence. 2.1
On
and
convergence
To x ideas, we start de ning ways of measuring GDP per capita convergence across regions. There are two di erent, but related, concepts of convergence: (Barro and Sala-i-Martin, 2003).
and
convergence
convergence relates to the situations in which, ce-
teris paribus, poorer regions grow faster than richer regions. The process of
convergence
could be unconditional, when all regions converge to the GDP per capita in steady state, or conditional, when regions converge to di erent steady state levels. This
convergence process is typically associated to estimating equations of the
form: yit
yit
1
=
yit
1
+
t
+
i
+
it :
(1)
Subscript i refers to regions and t to years. y is the log of GDP per capita, time e ects that are common to all regions,
are region-speci c e ects, and
i
idiosyncratic shocks.2 Unconditional convergence implies that from 0. Estimating
i
are are
should not be di erent
convergence yields to estimating a positive value for
equation (1). To get a more structural explanation for
t
in the
notice that given the discrete
nature of the data (Barro and Sala-i-Martin, 2003), =1 where
e
;
is the speed of convergence to the steady state. Thus, from equation (1) we
can recover . Our claim is that
(and, therefore, ) is not constant and depends on
some institutional features that vary over time at the country level. In particular, in our regressions we model it as t
= + p0t ;
2
(2)
We abstract from region characteristics that vary over time because we do not have a long spam of data including them.
3
where p is a vector of institutions or policies that vary over time (including nancial development, trade openness, infrastructure, and democracy, among others). A test of 6= 0.
our hypothesis {which we explain in more detail below{is that The concept of
convergence, ie. whether the volatility of y decreases or not as
time passes, is related to this idea. To x ideas, for simplicity, let us assume that there v (0;
are no time and region-speci c xed e ects and that 2 t
where
2 t
=
2
+
2 0 2
is the variance of yi at time t,
which is equal to that when
2 u
> 0,
, and
2
1 e 2 t
2 0
2
e
2 t
). Then,
;
(3)
is the variance of yi at the steady state,
is the initial variance of yi . Thus, this equation implies 2 0
decreases over time only when
convergence implies
2
>
2
. Putting it di erently,
convergence only when the country starts with a distribution
of GDP per capita across regions that is "high" (ie. it is above its steady state level). If this is the case, as t increases,
2 t
decreases.
The previous analysis implies that the existence of and su cient condition for the existence of tence of
convergence is a necessary
convergence, but, in contrast, the exis-
convergence is a necessary but not su cient condition for the existence of
convergence. This shows the connection between both measures of regional GDP convergence. If we extend the analysis of
convergence to the inclusion of time varying
it is implicit in equation (2) ; we may expect that if (decrease) the speed of convergence decrease (increase)
2 0
>
2
, as
, variables that increase
convergence.
We use this basic measurement framework and the relationship between both convergence concepts in the empirical section of the paper. Now we move to discuss theories and previous research that suggest a relationship between trade openness and nancial development and GDP convergence across regions. 2.2
Theoretical Motivation and Previous Literature
A simple version of the Solow-Swan growth model gives us a basic motivation for our idea.3 Under the basic assumptions of this model, regions have di erent initial capital stocks per-capita and, therefore, di erent marginal productivities.4 These di erences in productivities imply di erences in returns and create incentives for factor movements 3
Most models related to regional convergence process assume that regions (a partially arbitrary de nition in most cases) are di erent economies (in terms of fundamental determinants of income and growth) and, therefore, it makes sense to study a convergence process. We make the same assumption. 4 For simplicity, we abstract from a distinction between population and workers.
4
that at the steady state will equalize capital per capita across regions. Empirical estimates of the regional speed of convergence tend to identify relatively low rates of around 2% (eg., see Barro and Sala-i-Martin, 2003 for evidence on regions of Europe, Japan, and the US). Several papers identify similar annual rates of convergence for Chile, with point estimates going from 1% (D az and Meller, 2003) to 2:7% (Fuentes, 1997), with other estimates in between (eg., Morande et al., 1995; Soto and Torche, 2004; Badia-Miro, 2008). Going to the topic of our paper, notice that the transitional dynamics in the SolowSwan is result of an implicit assumption related to the existence of borrowing constraints that make the investment level constrained by the current level of savings thus creating, jointly with the assumptions on the production function, transition dynamics. Thus, when nancial development takes place this borrowing constraint is less binding and, therefore, regions should converge more quickly to the steady state. An analogous result could emerge in an endogenous growth model, along the lines of Aghion et al. (2005), in which the speed of convergence of di erent results to the steady state is also related to the existence of borrowing constraints of entrepreneurs to invest in new technologies. Aghion et al. (2005) present empirical estimates for countries in which nancial development increase the speed of convergence towards the steady state. Guiso et al. (2002) present evidence that nancial development at the regional level has a positive e ect on growth. Notice that this is conceptually and empirically di erent than our argument: we claim that nancial development at the country level increases the speed of convergence. To our knowledge, no paper studies empirically this point. When we move away from one-sector frameworks, as the Solow-Swan model, we may have than the e ect of national policies and institutions depend upon the complementarity between these policies and region endowments and characteristics. For instance, one may take the case of the Hecksher-Ohlin model and the impact of trade openness on the levels and the variability of regional output. In this case, when there are di erences in factor endowments at the regional level, trade openness may increase regional output disparities when trade have di erent e ects on regions with di erent endowments. From a more general point of view, in multi-sector models, policy and institutional changes that have di erent degrees of complementarity with di erent regions may increase regional disparities and, therefore, decrease the speed of convergence. Kim (2008) presents a review of the empirical and theoretical literature about the impact of di erent institutions and policies on patterns of regional output. He argues that trade openness tend to be associated with increases in the degree of concentration of economic activity
5
and, therefore, of GDP per-capita (Kanbur and Venables, 2005; Kanbur and Zhang, 2005; Rodr guez-Pose and Sanchez-Reaza, 2005).5 . In the case of nancial development, this literature does not discuss it explicitly, but it could also be the case that di erent regions of the country tend to have di erent nancial needs (due to, for instance, to di erences in the production structure, along the lines of the cross-country papers by Levchenko, 2004 and Nunn, 2007) and, therefore, decrease the regional process of convergence. To our knowledge, no paper has studied this point. In sum, the main hypotheses we test in this paper are the following: The higher the level of nancial development at the country level the faster both convergence (ie. the lower the variance of log GDP per-capita) and -convergence (ie. the higher the speed of convergence in log GDP per-capita) The higher the level of trade openness at the country level the slower both convergence and
-
-convergence (ie. the smaller the speed of convergence in log
GDP per-capita) 3
Data
We use several datasets in this paper. Table 1 presents the variables used, the level at which each variable is collected, and the main descriptive statistics of each variable. National account estimates of GDP at the regional level are only available since 1985. Thus, we combine the o cial data from the Central Bank of Chile with data from Bonacic and Vial (1994), which is available since 1960 to 1992. We use population series from the Chilean Statistics Bureau (INE) to generate the per capita GDP series. We measure nancial development with the ratio of nancial intermediation to GDP from D az et al. (2009), which includes total bank deposits, domestic public debt, total mortgage debt, and the value of the stock market.6 We measure trade openness as the ratio of exports plus imports to annual GDP (also from D az et al., 2009). In addition, we also include, as control variables, other variables related to policies and institutions at the country level: infrastructure development (the annual percentage of total roads that are paved, the ratio of total roads to total population and the number of telephones every thousand inhabitants, from D az et al., 2009), a democracy index 5
He also presents arguments for the potential impact of other variables, such as, democracy, decentralization, and infrastructure. In this case, however, the theoretical rationale for most of these variable is not as clear as in the case of trade openness. 6 This index is closely related to the traditional proxies for nancial development. See Levine (2005) for a detailed discussion on the measurement of nancial development.
6
(restrictions to executive power and the level of political competition, from the updated version of Vanhanen, 2000), the average years of schooling of the labor force (D az et al., 2009), terms of trade (from D az et al., 2009), an index of structural reforms (from Fuentes, Larra n and Schmidt-Hebbel, 2004), and a proxy for macroeconomic stability (measure as
1 1+
;where
is the in ation rate). In the cross-country exercises, we use
information on GDP per capita from Bruhn and Gallego (2009) several other local sources. Information on policies and other controls come from Beck et al. (2002) and Gallup et al. (1999). We now take a preliminary look to the Chilean data. Figure 1 presents the Kernel estimates of the distribution of log GDP per-capita for 13 the Chilean regions for decennial observations from 1960 to 2000. It is interesting to notice how the distribution starts moving to the right from the 1980s and also how the distribution tends to become more concentrated in the recent periods. This is consistent with the idea of having convergence across regions and that the convergence process is not constant. Figure 2 complements this with the evolution of the standard deviation of log GDP per-capita from 1960 to 2006. The gure presents both the simple calculation of the standard deviation and the standard deviation weighted by regional population (in two versions: weighted by contemporaneous and initial population). There is a negative trend with some uctuations in both deviations (which, despite di erences in levels, are highly correlated among them). It is worth noting three main periods, (i) one that goes until 1975 with a strong decrease in the standard deviation, (ii) one that goes between 1983 with an increase in the variability of regional GDP, and (iii) a nal period with a strong decline in the standard deviation from 1983 to 2006 (with a small increase between 1997 and 2001). This suggests that from a uni-variate perspective the convergence process does not seem to be uniform. Finally, the similarity between
weighted by
initial and contemporaneous population makes clear that population movements are not a signi cant source of convergence in Chile (a fact already noted and discussed in detail by Soto and Torche, 2004). Finally, Figure 3 presents the evolution of our key variables: trade openness and nancial development. It is important notice that while both variables are correlated their evolution is di erent both in terms of levels and timing. While trade openness increases from about 0.20 in the 1960s to a peak around 0.70 in the 1990s, nancial development increases from about 0.10 to values around 1.50. In addition, the timing di ers, while both variables increase from the mid-1970s and have a small decrease in the early 1980s, the big increase in nancial development occurs from the mid 1980s.
7
This di erences in timing are important from an empirical point of view to isolate the di erential impact of both policy dimensions. 4
The Effects of Economic Reforms on Regional Convergence
In this section we present estimates on how nancial development and trade openness a ect both convergence concepts. We start presenting estimates of the determinants of convergence since a failure to identify an impact of nancial development and trade openness implies that they do not have an impact on nd an impact of these variables on on
convergence. Similarly, if we
convergence, we have to nd a consistent impact
convergence. Thus, this approach is useful to double-check the implications of
our theory using two di erent approaches. 4.1
Estimates for -Convergence
Our main estimating equation is t
where
=
+
o Ot 1
+
F Ft 1
+
T Tt
+ Xt0
1
+
t:
(4)
is the standard deviation of log GDP per capita, O is trade openness, F is
nancial development, T is a linear trend, X is the vector of controls mentioned in section 3, and
is a shock. Our main hypothesis is that
o
> 0 and
F
< 0. In
addition, we expect that if there is a convergence process that does not depend upon other variables,
T
< 0, as suggested by equation (3). We estimate using lags of the
right-hand side variables to avoid mechanical potential reverse causality.7 Estimating this equation in the case of Chile is complicated because many policy and institutional dimensions move in the same direction, so there is a non-trivial level of multi-colinearity. Thus our approach is to run di erent estimation in which we include di erent combinations of variables to study the robustness of the results. As previously discussed, the fact that both the levels and especially the timing of the policy reforms make it possible to estimate di erential policy e ects. Table 2 presents the results of estimating equation (4). In several columns we estimate with di erent combinations of control variables to test for robustness. The main results imply that F tend to have a negative, statistically signi cant (except in one case), and economically relevant e ect on . The (short-run) point estimates vary depending 7
We inlcude also the lag of the dependent variable in the equations given (i) the level of inertia observed in the data and (ii) the fact that by construction is present an AR(1) process when there is convergence. See Barro and Sala-i-Martin (2003) for more details. Results on the e ects O and F do not change in a signi cant way when we exclude the lag.
8
on the speci cation, but our estimations imply that the increase in nancial development we observe between 1960 and 2000 would explain between 8 and 12% of the big decrease in
we observe over the same period. In contrast, trade openness seem to have a positive, statistically signi cant (except
in one case too), and economically relevant e ect on . Our estimates in this case imply that the increase in trade openness we observe between 1960 and 2000 would increase by between 13 and 28% of the level we observe in this period. Regarding the e ect of control variables, they tend to have smaller and non-robust impacts on . For instance, the level of terms of trade shocks, capturing cyclical uctuations, seems to have a small negative impact on , with periods with positive term of trade shocks having negative impact on
. In turn, infrastructure and schooling
variables have a non-clear pattern (both in terms of signs and economic and statistical signi cance). 4.2
Estimates for -Convergence
The results in the previous section present initial evidence that convergence across regions is related to trade openness and nancial development in ways consistent with our theoretical discussion. Now, we move to an additional test of these hypotheses. In this case, we study how
convergence is related to institutions and policies. To do so, our
main estimating equation is (a transformation of): yit yit 5 + Ft F + Ot O + Xt0 yit 5 + t + i + it ; (5) = 5 which combines equations (1) and (2). We are interested in the term in parenthesis, ie. how
convergence is a ected by a vector of policies and institutions. Our previous
results imply that
F
of time dummies (ie.
< 0 and t)
O
> 0. In addition, notice that the inclusion of a vector
imply that we are controlling for any direct e ect of these
policies and institutions on income levels. So equation (5) only identi es any e ect of F and O operating through the interaction with initial development. As customary, we estimate (5) using a 5-year panel. In order to deal with traditional endogeneity problems in dynamic models with xed e ects, we estimates the regressions using the GMM methodology suggested by ArellanoBond. As known, this methodology uses lags as instruments (ie. internal instruments) and, therefore, in all the speci cations we present two speci cation tests: (i) the Sargan test that tests whether moment conditions for identi cation are supported by the data and (ii) the AR(2) test that studies whether lags of the dependent variables are valid instruments. 9
Table 3 presents the results. We start by just presenting the typical regional convergence regression{ie., without including the interaction terms in (5) but including region and time e ects. Estimates imply a speed of convergence of about 1:7%, in line with previous studies for Chile and for other countries. This is a small speed of convergence that implies that half of the gap between current and steady state output per capita will be closed in about 41 years. The results in next columns follow the same organization as in Table 2. We present several speci cations trying to how robust the estimates are to the inclusion of other control variables. Consistently with results in Table 2, nancial development presents a negative, statistically signi cant (expect in one case), and economically relevant e ect on the speed of convergence. The increase in nancial development between 1960 and 2005 implies that the convergence speed would increase to between 4% and 20% depending on the speci cation. In contrast, trade openness seems to decrease convergence. The interaction between trade openness and initial output gap is signi cant in almost all columns (except in one case). Again the e ect is economically signi cant: the increase in trade openness observed between 1960 and 2005 would have decreased the speed of convergence by between 4 and 7 p.p. This is also an economically relevant e ect. Regarding the other determinants of
convergence, we observe a robustly negative
impact of terms of trade on the speed of convergence, probably implying that the speed of convergence increases in good economic cycles. However notice that this e ect is (i) economically irrelevant, with speed of convergence increase around 2% when terms of trade increase by one standard deviation and (ii) not consistent with results in Table 2 for
convergence. Something similar occurs in relation with the democracy index and
with improvements in the availability of phone lines. In all, results regarding nancial development and trade openness are consistent with our previous theoretical discussion and present a pattern of economically relevant e ects, in particular in the case of trade openness. 5
Some Cross-Country Evidence
In the previous evidence we focus in Chile, as a particular example of a country having undertaken a signi cant process of trade and nancial development in the last 50 years and having at the same time experienced a process of regional convergence. In this section we explore, as a check for "external" validity, whether the same pattern appears in a sample of 26 countries for which we have data on regional GDP. We constructed
10
measures of the standard deviation of log GDP per capita for these countries and run the following regression: i
=
+
o Oi
+
F Fi
+ +Xi0 + %t :
where i refers to country. We include the log of GDP per capita at the country level, a vector of soil characteristics (from Gallup et al., 1999), and the log of the number of regions in the country in the vector of controls. If the pattern we found using time series evidence for Chile is also present in the cross section of countries, then we expect that o
> 0 and
F
< 0.
Table 4 presents the results. We rst estimate the regression only including nancial development and
nd a negative, statistically, and economically signi cant negative
correlation of this variable with
convergence. The correlation implies that a one
standard deviation increase in nancial development is associated with a decrease of of about 0.57 standard deviations. Next, we just include trade openness in the regression and nd a positive, but statistically and economically insigni cant correlation with (the standardized e ect is just about 0.06 standard deviations of ). In the next column, we include jointly both variables and nd estimates with similar signs, but with bigger e ects of both variables. The standardized e ect of nancial development increases to about 0.80 and the standardized e ect of trade openness increases to about 0.25, but the latter is just marginally signi cant (p-value of 0:11). Figure 4 presents added variable plots for the two variables of interest for this speci cation.8 In the nal two columns of the Table, we study the robustness of the previous results by including GDP per-capita and a vector of controls for soil characteristics. In both cases, nancial development remains signi cant with a big standardized e ect.9 In all, the correlations presented in this section suggest the pattern of an association between
and nancial development and trade openness which is, especially in the
case of nancial development, consistent with our previous evidence, thus providing an additional check of the external validity of our results. Given the small sample we use in these regressions, we take this evidence as additional supporting evidence that should be studied in more detail in further research. 8 Excluding Ecuador from these regressions changes only marginally the estimated e ects and their statistical and economic signi cance. 9 We also experimented with IV regressions to deal with potential endogeneity problems using (i) the legal origins variables from Laporta et al.(2008) as instruments for nancial development and (ii) the predicted trade variable from Frankel and Romer (1999) as an instrument for trade openness. In both cases the Hausman tests do not reject the null hypotheses that the OLS estimates are equal to IV estimates (with p-values of 0.39 and 0.90 for nancial development and trade openness, respectively) and therefore OLS estimates are preferred.
11
6
Concluding Remarks
The potential e ects of policies and institutions on development at the country level has been a much debated topic in the literature. Our study of the e ects of policies on the convergence process at the regional level can help us to understand the e ects of policies and institutions on other dimensions of the development process. Previous research has mainly focused on identifying the existence or not of convergence processes at the regional level, while leaving the study of the determinants of this convergence process open. We document that two particular country-level policies that have a theoretical motivation a ect the convergence process as expected: while nancial development makes the convergence process faster, trade openness produces the opposite e ect. We nd these results using two di erent, but related, views of the convergence process: the study of both
and
convergence. In both cases the impact of both trade openness and
nancial development are statistically and economically relevant. These results imply that economic policies do not only a ect the level of development, but also may a ect the heterogeneity of development within countries. Given the big di erences in development within countries we observe, our results give an additional role to nancial policies. We focus in Chile, as a particular example of a country having undertaken a significant process of trade and nancial development in the last 50 years and having at the same time experienced a process of regional convergence. However, we also show in an exploratory way that the same pattern of e ects appear when we analyze a sample of 26 countries for which we have measures of the standard deviation of log GDP per capita within country. Further research should study this hypothesis including more countries and estimating the impact also on
convergence. References
[1] Aghion P, Howitt P and Mayer D (2005) The E ect of Financial Development on Convergence, Quarterly Journal of Economics. [2] Badia Miro, Marc. 2008. La localizacion de la actividad economica en Chile,18701973. PhD thesis, Universitat de Barcelona. [3] Barro, R., Sala-i-Martin, X., 1991, Convergence across States and Regions, Brookings Papers on Economic Activity. [4] Barro, R., Sala-i-Martin, X., 2003, Economic Growth, MIT Press. 12
[5] Beck, Thorsten & Levine, Ross & Loayza, Norman, 2000. "Finance and the sources of growth," Journal of Financial Economics, Elsevier, vol. 58(1-2), pages 261-300. [6] Bonacic, C., Vial, J., 1994, "Evolucion del Producto por Regiones 1960-1992", Coleccion de estudios CIEPLAN Nro.39. [7] Bruhn, M. and F. Gallego (2011) "Good, Bad, and Ugly Colonial Activities: Do They Matter for Economic Development?" The Review of Economics and Statistics. [8] D az, R., Meller, P., 2004, "Crecimiento Economico Regional en Chile: >Convergencia?", Universidad de Chile, Departamento de Ingenier a Industrial, Working Paper. [9] D az, J, R. Luders, and G. Wagner, 2009. La Republica en Cifras. Dataset available at the Economic History and Cliometrics Lab: http://www.economia.puc.cl/cliolab. [10] Duncan, Roberto; Rodrigo Fuentes. 2006. "Regional Convergence in Chile: New Tests, Old Results" Cuadernos de Econom a, Vol. 43, N 127, pp. 81-112. [11] Frankel, J. and Romer. 1999. "Does Trade Cause Growth?" American Economic Review. [12] Fuentes, R., 1996, ">Convergen las Regiones en Chile? Una Interpretacion" in Analisis Emp rico del Crecimiento en Chile, Centro de Estudios Publicos, Santiago. [13] Fuentes,R., Larra n, M.,Schmidt-Hebbel, K., 2004, Fuentes del crecimiento y comportamiento de la productividad total de factores en Chile, Working Paper 287, Banco Central de Chile. [14] Gallup, JL, J.D. Sachs and A. Mellinger (1999), "Geography and economic development", CID Working Paper Vol. 1, Harvard, MA. [15] Guiso, L., Sapienza, P., Zingales, L., 2002, Does Local Financial Development Matter?, Working paper # 8923, NBER. [16] Kanbur, R. Vanables, A., 2005, "Spatial Inequality and Development", Review of Development Economics. [17] Kanbur, R., Zhang, X., (2005), \Fifty Year of Regional Inequality in China: a Journey through central planning, reforms and opennes", Review of Development Economics. 13
[18] Kim, S., 2008, "Spatial Inequality and Economic Development: Theories, Facts, Policies", Working Paper nr. 16, Comission on Growth and Development. [19] La Porta, R., F. Lopez-de-Silanes and A. Shleifer (2008) "The Economic Consequences of Legal Origins", Journal of Economic Literature, June. [20] Levchenko, Andrei, (2004) \Institutional Quality and International Trade," IMF Working Paper WP/04/231, International Monetary Fund. [21] Levine, R. 2005. "Finance and Growth" in Handbook of Economic Growth. [22] Morande, F., Soto, R., Pincheira, P., 1996, Achilles, the Tortoise, and Regional Growth in Chile" in \Analisis Emp rico del Crecimiento en Chile, Centro de Estudios Publicos, Santiago. [23] Nunn, Nathan (2007) "Relationship-Speci city, Incomplete Contracts and the Pattern of Trade," Quarterly Journal of Economics, Vol. 122, No. 2, 569-600. [24] Rodriguez-Pose, A., Sachez-Reaza, J., 2005, \Economic Polarization through trade liberalization and regional growth in Mexico", in Spatial Inequality and Development, Oxford University Press. [25] Soto, R., Torche, A., 2004, "Spatial Inequality, Migration, and Economic Growth in Chile", Cuadernos de Econom a, Vol. 41. [26] Vanhanen, T. (2000) "A New Dataset for Measuring Democracy, 1810-1998". Journal of Peace Research, Vol. 37, No. 2, 251-265.
14
"& "% '()*(+,-(*./012 "$ "# ! !3
! 3 Figure 1: Distribution of Log GDP per capita, different years
#!!!
.55
.7
.35 .4 .45 .5 Sigma weighted by population
.65 .6 Sigma .55 .5
.3
.45 .4 1960
1970
1980
1990
2000
2010
year Sigma
Sigma weighted by population
Sigma weighted by 1960 population
Figure 2: Standard Deviation of Log GDP per capita, 1960-2005
.6 .5 Openness .4 .3 .2
1970
1980
1990
2000
2010
1960
1970
1980
1990
2000
2010
0
50
Financial Development 100 150
200
1960
Figure 3: Trade Openness and Financial Development, 1960-2000
14! .-1
$-. /*24' +2& +%& !*0
$%& !"#
&3. /-"
/.3 +.(
+-'
#&5
'() 5*(
,-" .,#
2() 12. 3*( ,#.
(6&
!78
#;@)ABC
