IMF Staff Papers (2010) 57, 427–449. doi:10.1057/imfsp.2009.19; published online 25 August 2009

The Impact of Trade Liberalization on the Trade Balance in Developing Countries

Abstract

JEL Classifications:

F11; F14

Many developing countries have substantially liberalized their trade regime over the past three decades, either unilaterally or as part of multilateral initiatives. Nevertheless, trade barriers remain high in many developing countries. One of the concerns that attributes to the reluctance of many of these countries to liberalize their trade regime is the possible worsening of the trade balance. 1 This is the question we want to investigate in this study: did past liberalization episodes in developing countries lead to a deterioration of their trade balance?

On the theoretical ground, Ostry and Rose (1992) offer an extensive survey of the macroeconomic effects of trade tariffs based on different theoretical frameworks, including the income-expenditure approach, the monetary approach, and the intertemporal approach. The authors conclude that there is no clear conclusion about the effect of a tariff change on the trade balance. The effect depends on the behavior of real wages and exchanges rates, on the values of a variety of elasticities, the degree of capital mobility, and whether the tariff shock is perceived as temporary or permanent.

Using a simple two-period intertemporal trade model, we analyze the effect of trade liberalization on the import, the export, and the trade balance in a small country. The effects rely on the interactions among the real income effect. the intratemporal substitution effect between importing goods and exporting goods, and the intertemporal substitution effect across time periods. The intertemporal substitution effect is negligible, as the tariff reductions are permanent, and the small country takes the world prices and the interest rate as exogenous. Tariff reductions increase the real income and decrease the price of the import good. Thus, both the real income effect and the intratemporal substitution effect increase imports. The intratemporal substitution effect decreases the domestic consumption of the exportable good, whereas the real income effect increases it. Assuming the former effect dominates, trade liberalization will increase exports. As trade liberalization increases both exports and imports, the difference of these two, the trade balance, may increase or decrease due to tariff reductions. The impact of trade liberalization on the trade balance, therefore, needs to be investigated empirically.

One stream of the related empirical literature attempts to find out how trade liberalization affects a country’s imports, and generally finds a positive impact ( Melo and Vogt, 1984 ; Bertola and Faini, 1991 ; and Santos-Paulino, 2002a ). There are also empirical researches focusing on the effects of trade liberalization on exports, where the findings are more mixed. Some of them show that countries which embarked on liberalization programs have improved their export performance ( Thomas, Nash, and Edwards, 1991 ; Ahmed, 2000 ; and Santos-Paulino, 2002b ), whereas others have found little evidence of such a relationship ( Greenaway and Sapsford, 1994 ; Jenkins 1996 ).

For policymakers, the impact of trade liberalization on the overall balance would be the more important question. There have been, however, surprisingly few cross-country empirical studies on the subject. Ostry and Rose (1992) studied the impact of tariff changes on the trade balance using five different data sets, mostly data from the Organization for Economic Cooperation and Development countries, and found no statistically significant effect. UNCTAD (1999) studied the effect of trade liberalization on the trade balance for 15 developing countries over the period of 1970–1995, and found a significant negative relationship. In a more recent paper, Santos-Paulino and Thirlwall (2004) studied the effect of trade liberalization on imports, exports, and on the overall trade balance using a sample of 22 developing countries for the period of 1972–97. They found that liberalization stimulated export growth but raised import growth by more, leading to a worsening of the overall trade balance.

One constraint researchers on the subject often face is the lack of systematic data measuring the dates of trade liberalization. Indeed, due to data limitation, most of the empirical studies on the subject are constrained to country case studies. In this article, we use two recently compiled data sets establishing trade liberalization dates that cover a large sample of developing countries for a long period of time. In particular, our two samples cover 39 and 77 developing countries for the period of 1970–2004, and 1970–2001, respectively. Our study focuses on the impact of trade liberalization for developing countries, for which the policy relevance of this question remains especially high. We find strong evidence that trade liberalization leads to faster import and export growth. The evidence on the overall trade balance, however, is mixed. Using our first measure of trade liberalization dates, we find little evidence that trade liberalization worsens the trade balance. There is some evidence that liberalization leads to a deterioration of the trade balance when we use our second measure of liberalization dates, although the finding is not robust to alternative estimation specifications.

I. Theoretical Analysis

This section develops a two-period intertemporal trade model to analyze the effect of trade liberalization on the import, the export, and the trade balance. Since our empirical analysis investigates permanent tariff reductions in developing countries, we study a permanent tariff reduction in a small country in this theoretical analysis. The key insight relies on the interactions among the real income effect. the intratemporal substitution effect between importing goods and exporting goods, and the intertemporal substitution effect across time periods.

The life-time utility function for the representative consumer in the home country is defined as

where C _t is the consumption in period t (t = 1,2) and β is a time-preference factor. With respect to technology, the home country specializes in producing a single good, labeled as good 1, and the foreign country specializes in producing good 2. Let the output of good 1 in period t be y _{t 1}. The linear homogeneous production function for good 1 is y _{t 1} = F (k _t . L ) where k _t and L are capital and labor used in production, respectively. We assume that labor supply, L. is fixed, and is normalized as 1 from here on. The capital stocks evolve according to

where I _t is the investment in period t and the depreciation rate is assumed to be zero.

Consumption and investment are composite of foreign and domestic goods:

where G (x _{t 1}. x _{t 2} ) = (x _{t 1} ρ + x _{t 2} ρ ) 1 / ρ is an Armington aggregator and 0ρ ≤1. The elasticity of intratemporal substitution between foreign and domestic goods is σ = 1 / (1−ρ ) and 1 σ ∞. This setup is standard in the literature of international real business cycle (IRBC) (see Backus, Kehoe, and Kydland, 1992 and 1994 for more discussions). While the IRBC literature uses an infinite horizon model for calibrations, we use a two-period model to get closed form solutions, in order to provide some intuition for our empirical investigations.

Similar to the argument in Ostry (1988). the consumer may be viewed as solving a two-stage optimization problem. In the first stage, the consumer chooses x _{t 1} and x _{t 2} to minimize her expenditure for a given level of consumption and investment, C _t + I _t . That is, she solves

where p _ti is the domestic price of good i. Letting τ be an ad valorem tariff rate on imports, we have p _t1 = p _t1 * and p _t2 = (1 + τ )p _t2 * where p _ti * is the world price. The solution to this problem yields the expenditure function

where q _t = (p _{t 1} 1− σ + p _{t 2} 1− σ ) 1 / 1− σ. To simplify the analysis, we assume that the world prices do not change. That is, p _1j * = p _2j * for j = 1,2, and therefore, we have q ₁ = q ₂ = q. Using the envelope theorem, we have

where r is the world interest rate that the small country takes as exogenous. The government redistributes the tariff revenue, τp _{t 2} * x _{t 2}. back to the consumer in every period. Note that capital, k ₂. accumulated in period 1 will be consumed at the end of period 2 and k ₃ will be zero, implying that I ₂ = k ₃ −k ₂ = −k ₂. In the second stage, the consumer chooses C ₁. I ₁. and C ₂ to maximize lifetime utility (1) subject to the intertemporal budget constraint (3) (k ₁ is given by history and is not subject to choice on date 1).

and

where Q = q ₂ −τp ₂₂ * q ₂ σ p ₂₂ − σ is the aggregate price index, excluding the tariff revenue effect. Equation (4) is the standard Euler equation, and equation (5) states that the marginal value product of capital equals the interest rate. C ₁. I ₁. and C ₂ are solved by equations (3). (4). and (5). The import value M _t . the export value X _t . and the trade balance TB _t are correspondingly written as

Note that the intertemporal budget constraint (3) implies that TB ₁ + TB ₂ / (1 + r ) = 0.

We are now ready to discuss the effect of trade liberalization. With some computations, we can show that ∂Q / ∂τ >0. Hence, the aggregate price index declines as tariff rate τ decreases. Equation (5) then indicates that k ₂. and therefore I ₁ must increase, since now the real price of the domestic product, p ₂₁ / Q. becomes higher. Rewriting the intertemporal budget constraint (3), we have

The value of the right-hand side of equation (6) increases as k ₂ increases. Therefore, C ₁ + I ₁ must increase. The proof is straightforward: if C ₁ + I ₁ were smaller, then C ₁ would be smaller since I ₁ is larger, then C ₂ would be smaller using equation (4). so that the value of the left-hand side of equation (6) would decline, and that would be a contradiction.

When tariff rate τ is reduced, the real price of the domestic good and therefore the real income increases. This is labeled as the real income effect, which increases both consumption demand and investment demand. The intertemporal substitution effect across time periods is negligible. Even if C ₁ declines, C ₁ + I ₁ must be higher after tariff reductions.

The effect of trade liberalization on the import value in current period is

It is easy to show that the first derivative in the right-hand side of equation (7) is ∂ (q ₁ σ p ₁₂ − σ ) / ∂τ 0. This is called the intratemporal substitution effect; the tariff reduction reduces the price of the import good and therefore increases the import demand. As we have argued above, the real income effect implies that ∂ (C ₁ + I ₁ ) / ∂τ 0. Thus, both the intratemporal substitution effect and the real income effect increase the value of imports.

Noting that y ₁ = F (k ₁ ) does not change, the effect of trade liberalization on the value of exports in current period is

Now, the first derivative in the right-hand side of equation (8) is ∂ (q ₁ σ p ₁₂ − σ ) / ∂τ >0. That is, the intratemporal substitution effect decreases the domestic consumption of the exportable good and therefore increases the export value, whereas the real income effect does the opposite. Assuming the former effect dominates, we have ∂X ₁ / ∂τ 0. Hence, trade liberalization increases the export value.

As both X ₁ and M ₁ increase, the difference of these two, the trade balance, may increase or decrease due to tariff reductions. More precisely, with some computations, we have

and its sign may be positive or negative. Summarizing we have:

Proposition 1

Tariff reductions increase the value of imports in the current period. If the intratemporal substitution effect dominates the real income effect, tariff reductions increase the value of exports in the current period. The effect of tariff reductions on the trade balance is ambiguous.

Since y ₁ = F (k ₁ ) does not change, our results also hold for the ratio of the import value, the export value, and the trade balance to GDP, which we will use as the dependent variables in our empirical study. As we will show next, the theoretical results we derived above are consistent with our empirical investigations.

II. Two Measures of Trade Liberalization Dates

Our first measure of trade liberalization dates is based on Li (2004). who has individually documented trade liberalization episodes in 45 countries between 1970 and 1995. We extended the liberalization measure for the 39 developing countries 2 in her data set to 2004 using the tariff data from the United Nations Conference on Trade and Development’s Trade Analysis and Information System (TRAINS) database (supplemented by data from the IMF’s Trade Policy Information Database—TPID). In doing so, a trade liberalization episode is identified if there is a continuous and accumulated tariff reduction by at least 35 percent (for example, a tariff reduction from 15 to 9.75 percent). 3 However, once a country’s overall tariff level reaches 10 percent or lower, we regard it as open and a further tariff cut, even by more than 35 percent, will no longer be considered as a liberalization episode. 4 The IMF’s TPID database also rates a country’s nontariff barrier level into three categories (open, moderate, and restrictive). In addition to looking at tariff reductions, we also take the reductions in nontariff barriers into consideration when defining a liberalization episode. However, it turns out that reductions in nontariff barriers are usually accompanied by large tariff cuts.

Table 1 reports our first measure of liberalization dates covering the period between 1970 and 2004, with the years of liberalization episodes highlighted (tariff reductions typically spread over several years). Two observations are worth mentioning. First, the period of 1985–95 seems to be the “opening-up decade” for developing countries. Almost all the countries in our sample experienced one or more episodes of liberalization during this period. Secondly, many countries experienced multiple episodes of liberalization (this is the case for 20 of the 39 countries in the sample). Indeed, trade liberalization is still an ongoing process for many developing countries.

For countries that experienced multiple liberalization episodes, a subsequent liberalization is often implemented either because the earlier one was limited in scope or was later reversed (at least partially). We, therefore, define a trade liberalization dummy, which takes the value of one after the end of the last recorded liberalization episode for a country and zero beforehand. 5

Our second measure of trade liberalization dates is from Wacziarg and Welch (2003). Wacziarg and Welch define the liberalization date as the date after which all of the Sachs and Warner (1995) openness criteria are continuously met. In particular, Wacziarg and Welch classify a country as closed if it displays at least one of the following characteristics: (1) average tariff rates of 40 percent of more; (2) nontariff barriers covering 40 percent or more of trade; (3) a black market exchange rate that is depreciated by 20 percent or more relative to the official exchange rate, on average; (4) a state monopoly on major export; and (5) a socialist economic system. However, data limitations often forced them to rely on country case studies of trade policy. One advantage of the Wacziarg-Welch data set is that it covers a substantially larger sample of developing countries. The Wacziarg-Welch liberalization dates are also reported in the last column of Table 1 (only for the overlapping countries).

We note in many cases the identified dates are very close across the two measures. For example, our first measure would identify 1992 as the year that Argentina liberalized its trade regime, compared with 1991 in Wacziarg and Welch (2003). For multiple liberalization episodes identified by our first measure, in several cases the Wacziarg-Welch date is closer to the first episode. For example, our first measure suggests that Chile had two episodes of liberalization, during 1974–79 and 1985–92, respectively. Thus, our first liberalization dummy will be one starting from 1993. The Wacziarg-Welch liberalization measure, instead, identifies 1976 as the year after which the economy has been open. This misses the reversal afterwards and the second liberalization during 1985–92. 6 Finally, in a few cases, the identified liberalization dates are quite different across the two measures. For example, Li (2004) identifies a liberalization era lasting from 1985 to 96 for Indonesia (average nominal tariff more than halved), whereas Wacziarg and Welch classify Indonesia as open from 1970. Nevertheless, the two measures are significantly and positively correlated, with a correlation coefficient of 0.57 (for countries in which they overlap).

Table 2 tabulates the average import-, export-, and trade-balance-to-GDP ratios using our first measure of trade liberalization for the periods before and after liberalization. Reported at the bottom of the table are cross-country averages. In general, countries not only import but also export more after they liberalized their trade regimes. The cross-country average import-to-GDP ratio increased from 23.8 to 30.6 percent, with 33 countries seeing their import-to-GDP ratio increased vs. four countries experiencing a decline. The average export-to-GDP ratio increased from 19.5 to 24.1 percent, with the ratio increased in 28 countries and reduced in nine countries. The average increase in exports, however, is smaller than that of imports, as the average trade deficit slightly increased from 4.3 to 6.5 percent. However, the picture is not uniform across countries—22 countries experienced a deterioration of the trade balance after liberalization and 15 countries actually had an improved trade balance.

Table 2 — Import, Export, and Trade Balance-to GDP-Ratios Before and After Trade Liberalization

(Extended Li Trade Liberalization Measure, 1970–2004) .

Table 3 reports the summary statistics using the Wacziarg-Welch measure of trade liberalization dates. 7 The average import-to-GDP ratio increased from 25.1 percent before liberalization to 29.9 percent afterwards. In all, 47 of the 62 developing countries that experienced trade liberalization during the period had higher import-to-GDP ratios. The average export-to-GDP ratio increased from 18.5 to 20.4 percent, with 40 countries experiencing an increase in the average ratio and 22 countries a decrease. Finally, the average trade deficit increased from 6.5 to 9.5 percent, with 41 out of 62 countries experienced a worsening of their trade balance.

III. Regression Analysis

Specification and Data

We follow Santos-Paulino and Thirlwall (2004) to use trade balance over GDP as the dependent variable and estimate the following dynamic panel equation:

where TB denotes the trade balance (the lagged dependent variable is included in the equation to control for adjustment dynamics); lib is the trade liberalization dummy; ŷ _it and ŷ _it * are domestic and foreign real GDP growth respectively; rêer _it and TÔT denote the change in (log) real exchange rate and terms of trade respectively. We also include fiscal-balance-to-GDP ratio (fisr ) to control for the impact of government fiscal policy on the trade balance. Finally, u _i represents time in varying country-specific effects, and v _it is a well-behaved disturbance term.

Trade, GDP, and fiscal balance data are from the IMF’s International Financial Statistics (IFS) database. Terms of trade data are from the IMF’s World Economic Outlook database. Foreign (real) GDP growth is the weighted growth rates of a country’s export market countries, where the weight is the market country’s 1990 share of the home country’s total exports. Bilateral trade data used to calculate the weights are from the IMF’s Direction of Trade Statistics database. Finally, the real exchange rate is calculated as a geometric weighted average of bilateral real exchange rates between home country and its trading partners:

where i indicates home country and j indicates trading partner countries. E _i,us is the nominal exchange rate of country i in U.S. dollar per local currency unit, and W _ij is the share of country j in country i ‘s total trade with its major trading partners. Countries whose trade share in home country is larger than 10 percent are included as major trading partners in calculating reer except China, because of incomplete consumer price index (CPI) data (both CPI and bilateral exchange rate data are from the IFS). An increase in reer indicates a real appreciation.

Before studying the impact of trade liberalization on the overall trade balance, we first analyze its impact on imports and exports separately. The standard trade equation would use the log of import and export volume as the dependent variable to derive income and price elasticities. This, however, will dramatically reduce our sample size due to missing import / export price data for many countries. Because income and price elasticities are not our primary interests, we use the import- and-export-to-GDP ratio (in log) 8 as the dependent variable in the import and export analyses to maintain our sample size and for consistency between import-export regressions and the trade balance regressions (where trade balance over GDP is the dependent variable).

Impact of Trade Liberalization on Imports

The regression results using our first measure of liberalization dates are reported in Table 4. The sample covers 39 countries with 1,202 observations. Column (1) reports the fixed effects panel regression as a benchmark. The trade liberalization dummy is positive and significant at the 1 percent level, indicating that liberalization leads to higher import growth. In addition, higher domestic growth also leads to a higher import-to-GDP ratio, suggesting an income elasticity larger than 1. Both real exchange rate appreciation and improved terms of trade (through lower import prices) lead to lower imports (in value), suggesting a price elasticity lower than 1. 9 Finally, the positive sign on the fiscal balance is a bit puzzling, as we would expect that an improvement in the fiscal balance lowers the import demand.

Table 4 — Trade Liberalization and Imports

(Extended Li Trade Liberalization Measure, 1970–2004) .

However, under the dynamic panel setting fixed effects estimates, even if the country fixed effects assumption is correct, will be consistent only if the time series dimension of the panel goes to infinity. We, therefore, use the system-generalized method of moments (GMM) developed by Blundell and Bond (1998) to get consistent estimates. 10 As a robustness check, we report both one-step and two-step estimates. The two-step procedure involves the additional computation of an optimal weight matrix, but is theoretically more efficient. We first follow the standard procedure to use all available lags of the dependent variable and the exogenous regressors in levels dated t – 2 to all earlier years as instruments in the estimation. 11 However, too many instruments can “overfit” endogenous variables and bias coefficient estimates, as well as weaken Hansen test of instrument validity ( Ziliak, 1997 ; Bowsher, 2002 ), and it has been suggested that shorter lags of instruments be used ( Arellano, 2003 ; Roodman, 2007 ). We, therefore, also report GMM estimates only using lags dated t – 2 and t – 3 as instruments (labeled as GMM (2, 3) in the tables). The GMM estimates are reported in columns (2)–(5) of Table 4 .

The results are broadly similar to the fixed effects regression 12 except that the fiscal balance now becomes insignificant and domestic GDP growth becomes insignificant when shorter lags are used as instruments. In all specifications, trade liberalization is shown to lead to higher imports. The Arellano-Bond test confirms the absence of second order correlation of the disturbance term required for consistency, and the Hansen test also does not reject the null hypothesis of joint validity of instruments. 13

Table 5 reports the import regressions using the Wacziarg-Welch measure of trade liberalization dates that covers a larger sample of 77 developing countries (62 of which “opened up” during the sample period) with 2,039 observations. The results are broadly similar to those reported in Table 4 except that the fiscal balance now becomes negative as expected, although insignificant. The trade liberalization dummy is positive and significant at the 1 percent level in all specifications. The estimated coefficients are larger than those reported in Table 4. For example, for one-step GMM (2, 3), the coefficient on the trade liberalization dummy is 0.074 vs. 0.047 in Table 4 .

Table 5 — Trade Liberalization and Imports

(Wacziarg-Welch Trade Liberalization Measure, 1970–2001) .