Financial Liberalization and Neighbor Coordination

Financial Liberalization and Neighbor Coordination Arvind Magesan and Jordi Mondria January 31, 2011 Abstract In this paper we study the economic and strategic incentives for a country to financially liberalize its equity markets. While it has been established empirically that a country s level of financial liberalization has implications for its growth, we provide new evidence that the financial liberalization of other countries in the same region also has an economically and statistically significant effect on growth. Specifically, the effect of other countries liberalization on growth is not constant and depends on the level of regional liberalization and whether or not the country is itself liberalized. Moreover, we document that being the first country in a region to financially liberalize is not beneficial, and in fact the effect on GDP growth is significantly negative. Based on these empirical findings, we propose a structural model of financial liberalization decisions which explicitly allows for strategic effects between countries. We estimate the political and economic costs and benefits of liberalization, and find that a country s probability of financially liberalizing is increasing in the financial liberalization of its neighbors. Our estimates imply that it might be optimal for emerging markets to wait for neighbors to liberalize first, and suggests emerging markets should coordinate when considering financial liberalization decisions. JEL Codes: F30, F36, F43, G15, G18, G28 Keywords: Equity market liberalization; Financial development; Quality of institutions; GDP growth, Structural game University of Calgary, e-mail: anmagesa@ucalgary.ca University of Toronto, e-mail: jordi.mondria@utoronto.ca 1

1 Introduction 2 Data Our data spans from 1981-1997, and covers 89 countries. The key variable in our study is the Official Liberalization indicator devised by Bekaert et al (2004). For a detailed description of how the variable is constructed, see Bekaert et al (2004). Briefly, for each country, for countries that liberalize in year t the liberalization variable takes a value of one from the year of liberalization on, and is zero otherwise. The year of liberalization is defined to be the year of formal regulatory change after which foreign investors officially have the opportunity to invest in domestic equity securities. Country gdp data is measured in thousands of international Geary-Khamis dollars, and is taken from Maddison (2003). The data on country level democracy over time comes from the Polity IV data set (Marshall and Jaggers, 2004). Each of the democracy and autocracy indexes in the Polity data set, which range from 0-10 (0 being the lowest level of democracy (autocracy) and 10 being the highest level of democracy (autocracy)) are composites of other political variables. First, democracy is conceived as the composite of three things: the degree to which citizens can freely express preferences over political leaders and policies, the constraints on the exercise of power by the executive, and the guarantee of civil liberties to citizens. Autocracy on the other hand is determined by how sharply political participation and competition is restricted, and how freely the executive, once selected, exercises power. We follow the literature and use the difference between these two scores (the Polity Composite Index) as our measure of a country s level of democracy. Our measure of political stability comes from the durable variable in the Polity IV data set. This variable simply measures the number of years since a major political regime change in the country. Any missing data from the polity data set is imputed using the suggestions of the authors. Summary statistics of the key variables are given in table 1. We include a measure of regional financial liberalization in the table. This measure is the sum of (the natural log of) gdp of liberalized countries in the region. We discuss this variable in more detail below. 2

Table 1: Summary Statistics Variable Obs Mean Std. Dev. Min Max Measurement Units gdp per capita 1504 6.32 6.15 0.34 26.05 Thousands of Geary-Khamis International Dollars 5 year average growth rate 1504 0.01 0.03-0.15 0.15 Number of Years Financial Liberalization 1504 0.35 0.47 0.00 1.00 Binary Indicator Regional Liberalization 1504 55.9 73.37 0.00 262.31 ln of Thousands of Geary-Khamis International Dollars Democracy 1504 0.60 0.37 0.00 1.00 Discrete Index Normalized to lie between 0 and 1 Political Stability 1504 0.14 0.17 0.00 1.00 Number of Years Normalized to lie between 0 and 1 3 Regression Analysis: Growth and Financial Liberalization In this section we establish the importance of both a country s own liberalization as well as its regional neighbor liberalization for growth. Bekaert et al (2004) are among the earliest studies to establish the importance of equity market liberalization for economic growth. They show that on average, equity market liberalization leads to a 1% increase in annual real growth, and that this effect is robust to several different specifications. We first confirm the general findings of Bekaert et al (2004), and then provide evidence that the liberalization decisions of other countries in the same region have significant implications for growth as well. The general specification we consider in this section is: g b it = γ 0 d it + γ n Ne it + γ 0 nd it Ne it + β Z it + u i t (1) where: g b it is the five year average growth rate starting at year t: git b = 1 5 ( ln(yit+k ) ln(y it+k 1 ) ) (2) 5 k=1 3

d it is the liberalization decision of country i at time t Ne it is our measure of neighbor liberalization. Z it is a vector of control variables. u it is the unobserved error term. We assume: u it = ω i + δ t + ν r(i),t + ũ it (3) where ω i is unobserved country specific heterogeneity, δ t is unobserved time-specific heterogeneity, ν r(i),t is an unobserved time-region specific shock, and ũ it is an idiosyncratic error that satisfies the usual OLS assumptions. The measure of neighbor liberalization we use is: Ne it = d jt y jt (4) j i,j R i That is, the liberalization of country i s regional neighbors at time t is given by the sum of the (ln) gdp per capita of the countries in the same region as country i who have liberalized by year t. We are thus assuming that a country who s liberalized neighbors are more economically important than another country s liberalized neighbors, is in a more liberalized region, even if they have the same number of liberalized neighbors. We estimate several versions of the equation 1. First, to confirm the results in Bekaert et al (2004), we estimate equation 1 under the assumption that γ n = γn 0 = 0. In table 3 we include estimates of the model with and without allowing for various forms of unobserved heterogeneity, and with and without political and economic control variables for illustrative purposes. The version of the model that we estimate in column 3 of table 3 is closest to the one considered by Bekaert et al (2004), the difference being that we allow for country specific unobserved heterogeneity while Bekaert et al do not, but they use more controls, including country level time-invariant variables such as initial gdp. As our primary purpose is not to uncover the specific determinants of growth but rather to understand the effect of own liberalization and neighbor liberalization on 4

Table 2: The effect of liberalization with no neighbor effects Column 1 Column 2 Column 3 Column 4 Column 5 Own Liberalization 0.015957** 0.007239** 0.010511 ** 0.002463 0.001693 (0.0017) (0.0022) (0.0021) (0.0022) (0.0022) gdp per capita - 0.003264-0.081241-0.085483-0.084860 (0.0013) (0.0070) (0.007) (0.007) Democracy - 0.017751 0.012802-0.002777-0.003586 (0.0033) (0.0036) (0.0036) (0.0036) Political Stability - -0.019130 0.017445-0.008566-0.011559 (0.0048) (0.0192) (0.0171) (0.018 ) Controls NO YES YES YES YES Country FE NO NO YES YES YES Time FE NO NO NO YES YES Region*Time FE NO NO NO NO YES N 1504 1504 1504 1504 1504 R-squared 0.072 0.135 0.610 0.642 0.653 Notes: Estimated by simple (OLS). Standard errors estimated using Newey-West (allows for autocorrelation and heteroskedasticity) growth, we choose to allow for permanent unobserved country level heterogeneity instead of allowing for several additional time invariant or weakly time varying controls. Bekaert et al find that own liberalization increases growth by about 1%. In column three, in spite of the difference in specification and difference in sample, we obtain almost the exact same result, we find that liberalization increases growth by 1.05%. Notice, however that as we move across to columns four and five, and allow for time specific unobserved heterogeneity and region/time unobserved heterogeneity, the effect diminishes in both economic and statistical significance, however. This suggests the possibility that economic growth and liberalization are jointly explained by time-specific shocks which are constant across all countries and time specific shocks that are constant across all countries within a given region, but which vary across regions. We now consider the possibility that a country s growth depends not only on its own financial liberalization but also on the liberalization of other countries in the same region. There are several reasons to expect that a country s growth depends positively on the financial liberalization of its neighbors. By the definition of the liberalization indicator we use, if country i s neighbors are financially liberalized, foreign investors (including those originating in country i) have the opportunity to invest in the liberalizing countries domestic securities. Thus there is clearly an opportunity for direct positive benefits for country i. There is also the opportunity for indirect spillover effects, as more economic attention is paid to the region. Finally, it has been established theoretically and empirically (Aizenmann and Noy, 2003 and Aizenmann, 2004) that there is a positive association between financial liberalization and trade openness/commercial development. 5

Thus the financial liberalization of a country s neighbors may have other spillover benefits by facilitating trade activity. Yet there is also reason to believe that liberalization of a country s neighbors may have negative consequences for the country. To explore the possibility that neighbor liberalization has an effect on growth, and that this effect may depend on a country s own liberalization we estimate equation 1, but this time allowing γ n 0 and γ 0 n 0. In table 3 we include estimates of the analogous specifications from table??. In the interests of space we do not include the estimates on the political and economic control variables, as the estimates do not change much from the first specification. Table 3: The effect of liberalization with neighbor effects Column 1 Column 2 Column 3 Column 4 Column 5 Own Liberalization 0.018485** 0.012145** 0.010398** 0.006934** 0.005705** (0.0024) (0.0028) (0.0022) (0.0026) (0.0028) Neighbor liberalization 0.000149** 0.000042 0.000380** 0.000271** 0.000227** (0.000022) (0.000026) (0.000042) (0.000062) (0.000069) Own Lib Neighbor lib -0.000140** -0.000074** -0.000103** -0.000077** -0.000064** (0.000025) (0.000025) (0.000028) (0.000029) (0.000030) Controls NO YES YES YES YES Country FE NO NO YES YES YES Time FE NO NO NO YES YES Region*Time FE NO NO NO NO YES N 1504 1504 1504 1504 1504 R-squared 0.103 0.141 0.643 0.650 0.657 Notes: Estimated by simple (OLS). Standard errors estimated using Newey-West (allows for autocorrelation and heteroskedasticity) Notice first of all that the estimate of γ 0, the effect of own liberalization on growth, is economically and statistically significant across all five columns. In particular, once we account time specific and region-time specific heterogeneity the effect of own liberalization on growth remains economically large and statistically significant. This suggests that neighbor liberalization is an omitted variable in the general specification we considered in table 3. Moreover, since the bias on the estimate of γ 0 in table 3 is negative, and as we can see in table 3 the estimated effect of neighbor liberalization on growth is positive, the covariance between own liberalization and regional liberalization conditional on the controls and the various forms of unobserved heterogeneity is negative. Neighbor liberalization has a baseline positive effect on growth, but this effect is weaker for countries who have themselves liberalized. Looking at it another way, a country s growth payoff to 6

liberalization is positive, but decreasing in the liberalization of its neighbors. To quantify the effect, recall that neighbor liberalization is the sum of (the natural log of) gdp per capita of liberalized neighbors. If we consider the model in column 5 where we include controls and allow for all forms of unobserved heterogeneity, a standard deviation increase in neighbor liberalization results in a 1.7% increase in growth for an unliberalized country and an increase of 1.2% increase in growth for a liberalized country. The returns to growth from neighbor liberalization are economically significant whether or not a country has liberalized, but they are larger for liberalized countries. Looking at it from the perspective of an unliberalized country deciding whether or not to liberalize, a country whose neighbors are at the mean level of neighbor liberalization experiences just a 0.2% increase in average growth. Clearly, a country whose neighbors are one standard deviation above (below) the mean level of neighbor liberalization will experience a significant decrease (increase) in growth with liberalization. To further explore the relationship between growth, own liberalization and neighbor liberalization, we consider the possibility that liberalizing when no other country in the region has liberalized could have different implications for growth than liberalizing after other countries in the region have already liberalized. There are several reasons to expect this may be the case. Formally, we consider the effect of the following variable on growth: first it = d it 1{Ne it = 0} (5) This variable takes the value of one if country i is the only liberalized country in the region at time t, and takes the value of zero otherwise. γf 0 captures the effect of this variable on 5-year average growth. We interpret this as the effect on growth of being the first one in the water. We consider an equation of the form: g b it = γ 0 d it + γ n Ne it + γ 0 nd it Ne it + γ f 0 first it + β Z it + u i t (6) The estimation results are in table 3. Focusing on column 5, where we include controls and allow for unobserved heterogeneity of all forms, we see that the effects of own liberalization and neighbor liberalization are similar in 7

Table 4: The effect of liberalization with neighbor effects and First in effect Column 1 Column 2 Column 3 Column 4 Column 5 Own Liberalization 0.019229** 0.012946** 0.011214** 0.007854** 0.006783** (0.0025) (0.0029) (0.002241) (0.002731) (0.00285) Neighbor Liberalization 0.000149** 0.000043 0.000380** 0.000275** 0.000234** (0.000022) (0.000026) (0.000042) (0.000049) (0.0000628) Own Neighbor lib -0.000144** -0.000079** -0.000109** -0.000084** -0.000072** (0.000025) (0.000025) (0.000028) (0.000030) (0.000029) First in -0.010198-0.009727-0.010207* -0.009795** -0.010710 ** (0.0044) (0.0035) (0.0041) (0.0042) (0.0042) Controls NO YES YES YES YES Country FE NO NO YES YES YES Time FE NO NO NO YES YES Region*Time FE NO NO NO NO YES N 1504 1504 1504 1504 1504 R-squared 0.104 0.142 0.644 0.651 0.658 Notes: Estimated by simple (OLS). Standard errors estimated using Newey-West (allows for autocorrelation and heteroskedasticity) sign and magnitude to those in table 3. Additionally, the effect of being first in is negative and significant. A country in a region with no liberalized neighbors will experience negative growth if it chooses to be the first country to liberalize. Altogether, we have evidence of the following: 1. Liberalization has a positive effect on growth. 2. Neighbor liberalization has a positive effect on growth. 3. Neighbor liberalization has a positive effect on a country s growth, but the effect is smaller for liberalized countries than unliberalized countries. Viewed another way, the positive effects of own liberalization are smaller if the region is very liberalized. 4. Being the first to liberalize has a negative effect on growth. These findings suggest that the benefits to liberalization are first increasing and then decreasing in neighbor liberalization: when a country is first in, the effect is negative. As neighbors liberalize the effect becomes positive and grows, and then starts to decline again. In figure 1 we plot the effect of financial liberalization on growth as a function of neighbor liberalization. These results raise other questions of interest. Clearly there are gains to liberalization when the value of neighbor liberalization is not too large. Yet we often observe countries that do not liberalize for a long period of time, foregoing the positive growth benefits. This suggests that there are other costs to financial liberalization that we have not accounted for. In the next section we develop 8

Figure 1: Effect of liberalization on Growth as a Function of Neighbor Liberalization and estimate a simple structural model of financial liberalization, which incorporates the fact that country liberalization decisions are interdependent. We use the model to impute the underlying cost of liberalization, and get a more broad idea of the strategic and economic incentives that drive the liberalization decision. 4 Model and Estimation Countries who have yet to liberalize make a binary decision to liberalize or not taking as given current values of state variables and the effects of their decisions on payoffs. I describe the model in detail here. At year t, region r {1,..., R} is comprised by Nt r countries (we allow the number of countries in a region to change over time for obvious reasons). Countries and regions are given exogenously in our model. Let x it {0, 1} indicate the openness status of country i at year t. If x it = 1, we say country i is open at the beginning of time t. Otherwise it is closed. We can represent the openness status of all the countries in region r by the vector x r t = {x it : i = 1, 2,..., Nt r }. Similarly, we represent the profile of openness decisions of countries in region r at period t by d r t {d it : i = 1, 2,..., Nt r }. Of course, d it = x it+1. It is assumed here that countries can not close their markets again after opening. That is, the decision to open is irreversible. Formally: x it = 1 d iτ = 1 τ t 9

Country i s payoffs in year t are the difference between economic growth g i and a market opening cost function C i : Π i (x t, z t, d t, ε it ) = g i (x t, z t ) C i (x t, z t, d it, ε it ) (7) where z t is a vector of exogenous political and economic variables and ε it is a vector of private information shocks of country i. Payoffs Based on the reduced form results above, we first assume that economic growth in country i at time t depends on it s own openness status in the following way: g b it = γ 0 d it + γ n Ne it + γ 0 nd it Ne it + β Z it + u i t (8) Then, by simple manipulation, the difference between growth under liberalization and growth without liberalization is given by: g b it(1) g b it(0) = γ 0 + γ 0 nne it + γ 0 f 1{ Ne it = 0 } (9) Costs The cost function C i (d t, z t, ν it ) is directly analogous to the concept of firm operating cost in the IO literature. It is the cost for country i of having an open market. For now, we specify the following cost function for country i at time t: ) C it = d it (1 x it ) (θc i + θ z z it + θ n Ne it + θ f first it + ν it (10) where θ i c is the liberalization cost of country i, 1 and z it is a vector of state variables that affect the cost of liberalization. ν it is private information of country i at time t. Equilibrium 1 Note that we cannot separately identify entry costs and fixed costs; this is due to the fact that the decision to liberalize is irreversible. We need to observe de-liberalizations to be able to identify entry costs and fixed costs separately. 10

Informally, the timing of the game is as follows. A country wakes up on the morning of year t, and compares the benefit to growth from liberalization with the cost. Note however that country i can only have an expectation over the benefit and cost, because it doesn t know Ne it. The expectation is conditional on the information available to country i at time t, namely z t and x t. Countries take as given the state x, z and the choice probabilities P of other countries in the same region and enter when the expected benefit to doing so is greater than the expected cost. Countries are not forward looking; they are myopic in their decision making. That is, country i liberalizes at time t if the current payoff to liberalization is greater than the current payoff to not liberalizing: Using equations 9 and 10, a country which enters period t unliberalized, liberalizes iff: (γ 0 θ i c) + (γ 0 n θ n )E [ Ne it x t, z t ] + (γ 0 f θ f )E [ 1 { Ne it = 0 } x t, z t ] θz z it E [ Ne it x t, z t ] νit (11) where: E [ ] ( Ne it x t, z t = P (x t, z t ) djt (1 P (x t, z t )) (1 djt) j i,j R i d i D i and D i is the set of all possible decision vectors of the other players. Define: ( ) ) djt y jt j i,j R i i ( xt, z t ) (γ 0 θ i c) + (γ 0 n θ n )E [ Ne it x t, z t ] + (γ 0 f θ f )E [ 1 { Ne it = 0 } x t, z t ] θz z it E [ Ne it x t, z t ] (12) Note that If the private information shock ν it has distribution Λ, then the ex-ante probability of country i liberalizing at time t, given it has yet to liberalize is given by: ( ( ) ) P i (x t, z t ) = Λ i xt, z t (13) and P i (x t, z t ) = 1 for any x where x it = 1 (irreversibility). game. A fixed point in the full system of equations defined by this P mapping is an equilibrium of the 4.1 Estimation The parameters of interest are :θ = {γ 0, γ 0 n, γ 0 f, θi c, θ n, θ f, θ z }. We estimate the full vector of parameters θ in two steps. In the first step the parameters of the function 9, γ 0, γn, 0 γf 0, are estimated 11

Table 5: Estimates of Structural Parameters Parameter Estimate Standard Errors α1 c -2.1818887** 0.31301618 α2 c -2.1585296** 0.29665184 α3 c -2.4074710** 0.40255155 α4 c -1.5459156** 0.38703482 α5 c -3.9789432** 0.87502768 α6 c -2.5796205** 0.39426740 α7 c -1.7475039** 0.33883520 α8 c -3.1380582** 0.40401496 gdp 0.30660052** 0.14453935 democ 0.66628949** 0.32022970 α n 0.034716409* 0.019855158 α f -0.71122836** 0.26792007 Log-Likelihood function -135.86881 Likelihood Ratio Index 0.14447828 Pseudo-R2 0.088506540 using simple Ordinary Least Squares (see above). In the second step we estimate the parameters of the choice probability function P i (x t, z t ): α c i = γ0 θ i c α n = γ 0 n θ n α f = γ 0 f θ f θ z using the Nested Pseudo Likelihood method of Aguirregabiria and Mira (2007). Using the estimates from the first stage we are then able to identify θc, i θ n, θ f. The first stage estimates we obtain (column 5 of table 3 are γ 0 = 0.006783, γn 0 = 0.000072, γf 0 = 0.010710. The estimates of the structural parameters α are in 5. The structural estimates suggest several interesting patterns. Firstly, there is significant variation across region in the cost of liberalization. 2 Permanent differences across region explain a significant amount of the variation in timing of the liberalization decision. We also see that countries with a higher level of current GDP percapita have a lower cost of liberalization than countries with a higher cost of liberalization. Countries with more democratic political institutions also have a lower cost of liberalization than countries with less democratic institutions. To understand the 2 We would ideally have liked to control for country level heterogeneity in cost as opposed to regional level heterogeneity, but there are several countries that either never liberalize (during the time horizon of our data) or liberalize in the first year, making identification of a country level cost infeasible. 12

Figure 2: Probability of Leader Liberalizing as Function of Neighbor Liberalization effect of neighbor liberalization on cost separately from its effect on growth, note that the implied estimates of θ n, θ f are 0.035 and 0.70 respectively. The cost of liberalization is decreasing in neighbor liberalization, and the cost of liberalization is larger for countries who are the first in the region to liberalize. 4.2 Neighbor Liberalization and the Decision to liberalize We have established that neighbor liberalization, and its interaction with a country s own liberalization status, has important implications for both gdp growth as well as for the non-gdp related costs of financial liberalization. In figures 2 and 3 we plot the equilibrium probability of liberalization as a function of the number of liberalized neighbors. Figure 2 plots the probability of liberalization for a regional leader, a country whose gdp per capita is larger than the other countries in the region, while figure 3 plots the probability of liberalization for a follower, a country whose gdp per capita is lower than the other countries in the region. As we can see in the figures, a leader s liberalization probability is relatively insensitive to neighbor behavior, while a follower s is very sensitive. In both cases the probability of liberalization is increasing in neighbor liberalization. This finding suggests that the positive effect of neighbor liberalization on the cost of liberalization outweighs the negative effect of neighbor liberalization on a liberalized country s gdp growth that we found evidence of above. 13

Figure 3: Probability of Leader Liberalizing as Function of Neighbor Liberalization 5 Conclusion 6 References 1. Abiad et al 2. Aguirregabiria and Mira (2007) 3. Aizenmann (2004) 4. Aizenmann and Noy (2000) 5. Bekaert et al (2004) 6. Magesan (2010) 7 Appendix 14