Trade and Income We discusses the study by Frankel and Romer (1999). Does trade cause growth? American Economic Review 89(3), 379-399. Frankel and Romer examine the impact of trade on real income using an instrumental variable approach. Preview We first discuss the potential (theoretical) effects of trade on real income. We then describe the empirical approach used by Frankel and Romer (and many others). Then, we present the main results obtained by Frankel and Romer. Finally, we summarize the main arguments and results of other studies on the relationship between trade and income 1-1
Gains from Trade Producers and consumers allocate resources most efficiently when governments do not distort market prices through trade policy. In standard theory, gains from trade arise from a reallocation of resources from import-competing sectors to specific export sectors in which a country has a comparative advantage, implying a contraction in the activity of the former and an expansion of the latter. National welfare of a small country is highest with free trade. With restricted trade, consumers pay higher prices. With restricted trade, distorted prices cause overproduction either by existing firms producing more or by more firms entering the industry. 1-2
Gains from Trade 1-3
X 2 World market price III X 2 E Gains from Trade Country A Country B Imports X 2 C B Ex A G F J 3 J 2 H Im D J 1 J 3 Exports X 1 I X 1 II J 1 X 1 1-4
The Cases for Free Trade Trade may promote specialization in sectors in which a country has comparative advantage and lead to a reallocation of resources from the relatively inefficient nontrade sector to the more productive export sector. Opening to trade can increase productivity and thus real income by offering larger economies of scale. Since combining the international market with the domestic market facilitates larger-scale operations than does the domestic market alone, an expansion of exports allows countries to benefit from economies of scale Free trade provides opportunities for innovation. 1-5
The Cases for Free Trade The export sector may generate positive externalities on the non-export sector. The sources of these knowledge spillovers include incentives for technological improvements, labor training, and more efficient management due to increased international competition and, direct access to foreign knowledge through relationships with foreign buyers (learning by exporting) Capital goods imports from high-income countries are typically associated with higher productivity, since capital goods embody technological know-how, implying that countries can acquire foreign knowledge (for example, in the form of research and development) through (capital) goods imports (learning by importing, reverse engineering). 1-6
Reasons Why Some Countries Do Not Gain from Trade An important source of gains from trade is the existence of cross-border knowledge spillovers. The ability to absorb foreign knowledge and technology depends on absorptive capacity. Several authors hypothesize that primary exports may be an obstacle to attaining a higher standard of living. Increased primary exports can lead economies to shift away from the competitive manufacturing sectors in which many externalities necessary for growth are generated, while the primary export sector itself does not have many linkages with, and spillovers into, the economy; Primary exports are subject to large price and volume fluctuations. Increased primary exports may therefore lead to increased GDP variability and macroeconomic uncertainty. 1-7
Reasons Why Some Countries Do Not Gain from Trade In a scenario of severe factor-market imperfections that limit both the mobility of factors between sectors and the flexibility of factor prices, increased trade may be associated with unemployment and, as a consequence, with income losses. The income effect of trade may also depend on the level of regulation The income effect of trade may depend on the quality of institutions. Institutions, such as property rights, lower transaction costs by reducing uncertainty and establishing a stable structure to facilitate interactions, thus helping to allocate resources to their most efficient uses. 1-8
General Approach in Cross-Country Studies Frankel and Romer (1999): ln(y i )=α+βt i +cs i +e i (1) ln(y i ) is the natural logarithm of income per person or income per worker in country i T i is the trade share of GDP (measured in logarithms or levels), and S i is country size, usually proxied by the logarithm of population and the logarithm of area. 1-9
General Approach in Cross-Country Studies ln(y i )=α+βt i +cs i +e i (1) Country size is included in the regression model for two reasons. (1) Country size serves as a proxy for the amount of trade within a country; the estimate of c can be used to assess whether countries also benefit from within-country trade. (2) Because larger (smaller) countries tend to have more (less) opportunities for trade within their borders, and therefore lower (higher) trade shares, it is necessary to control for country size in estimating the impact of international trade on income. Otherwise, S i would enter the error term, thereby inducing a negative correlation between e i and T i and thus a downward bias in the estimate of β. 1-10
General Approach in Cross-Country Studies ln(y i )=α+βt i +cs i +e i (1) Eq. (1) cannot be estimated by OLS, because of the likely endogeneity of trade, and because of omitted variables due to unobserved country-specific effects. 1-11
General Approach in Cross-Country Studies ln(y i )=α+βt i +cs i +e i (1) To overcome these problems, Frankel and Romer (1999) suggest an instrumental-variable (IV) approach. A valid instrument is correlated with the endogenous variable (T i ), but uncorrelated with the error term (e i ) and thus not associated with the dependent variable (ln(y i )) through any channel other than the endogenous variable. 1-12
General Approach in Cross-Country Studies ln(y i )=α+βt i +cs i +e i (1) To construct such an instrument, Frankel and Romer propose the following two-step procedure. The first step is to estimate a gravity equation for bilateral trade shares using distance between trading partners and country size as explanatory variables (components of trade, which are assumed to be independent of income). The second step involves calculating a predicted aggregate trade share for each country on the basis of the estimated coefficients of the gravity equation. This predicted trade share is then used as a geographybased instrument for trade in regression (1). 1-13
The Trade Instrument In the first step, Frankel and Romer (1999) estimate a gravity model of the form: log(trade ij /GDP i )= β 0 +β 1 log(dist ij )+β 2 log(n i )+ β 3 log(a i ) +β 4 log(n j )+ β 5 log(a j )+β 6 (L i +L j )+ β 7 BORDER ij + β 8 BORDER ij log(dist ij )+β 9 BORDER ij log(n i )+ β 10 BORDER ij log(a i ) +β 11 BORDER ij log(n j )+β 12 BORDER ij log(a j )+β 13 BORDER ij (L i +L j ) + e ij, TRADE is bilateral trade between countries i and j, DIST is the distance between i and j, N is population, A is area, L is a dummy for landlocked countries, and BORDER is a dummy for a common border between two countries. 1-14
The Trade Instrument log(trade ij /GDP i )= β 0 +β 1 log(dist ij )+β 2 log(n i )+ β 3 log(a i ) +β 4 log(n j )+ β 5 log(a j )+β 6 (L i +L j )+ β 7 BORDER ij + β 8 BORDER ij log(dist ij )+β 9 BORDER ij log(n i )+ β 10 BORDER ij log(a i ) +β 11 BORDER ij log(n j )+β 12 BORDER ij log(a j )+β 13 BORDER ij (L i +L j ) + e ij, The equation includes two measures of size: log population and log area, dummy variables for landlocked countries and common borders, interaction terms of all of the variables with the common-border dummy, because a large part of countries trade is with their immediate neighbors and because the goal is to identify geographic influences on overall trade. 1-15
The Trade Instrument Frankel and Romer aggregate the fitted values from the bilateral trade equation. They first rewrite the above equation log(trade ij /GDP i )= a X ij + e ij, where a is the vector of coefficients, and X ij is the vector of right-hand side variables. The estimate of the geographic component of country i s overall trade share is then Instrument = Σe â Xij 1-16
Results, Frankel and Romer, IV estimates Table 1 (included observations: 98) Coefficient Std. Error t-statistic Prob. C 1.620810 3.484226 0.465185 0.6429 TRADE 2.960762 1.340939 2.207977 0.0297 LOG(POPULATION) 0.351181 0.145084 2.420532 0.0174 LOG(AREA) 0.201787 0.176904 1.140657 0.2569 1-17
Results, Frankel and Romer, IV estimates Table 2 (included observations: 150) Coefficient Std. Error t-statistic Prob. C 4.961235 2.035972 2.436789 0.0160 TRADE 1.966313 0.912417 2.155061 0.0328 LOG(POPULATION) 0.191747 0.088080 2.176961 0.0311 LOG(AREA) 0.086371 0.097830 0.882866 0.3788 1-18
Results of Other Studies Rodríguez and Rodrik (2001) argue that the Frankel and Romer findings simply reflect the impact of geography on income, rather than the impact of trade on income, since the geography-based instrument is correlated with other geographic variables that affect income through non-trade channels, such as morbidity, agricultural productivity, and institutions. Rodríguez and Rodrik (2001) re-estimate the Frankel- Romer regression, adding additional controls for geography (such as distance from the equator, the percentage of a country s land area that lies in the tropics, and regional dummies), and find that the IV coefficient estimates on trade become statistically insignificant once additional geography variables are included. 1-19
Results of Other Studies Several other studies also include institutional variables in the IV regression. These are intended to explicitly control for potential income effects of the geography-based trade instrument that can be associated with the effects of geography on income through institutions. Frankel and Rose (2002), as well as Noguer and Siscart (2005), for example, estimate Eq. (1) with and without additional controls for both geography and institutions. They detect a large and statistically significant effect of trade on income that is robust to the inclusion of additional control variables. 1-20
Study Frankel and Romer (1999) Hall and Jones (1999) Rodríguez and Rodrik (2001) Frankel and Rose (2002) Irwin and Tervio (2002) Dependent variable Trade/GDP nominal Independent variable ln(trade/gdp) ln(trade/gdp) nominal real Geographical controls Institutional controls worker) 1.97 / 2.96 No No worker) 0.185 Yes Yes capita) 1.97 No No capita) 0.21 / 0.34 Yes No capita) 1.59 / 1.96 No No capita) 1.13 / 1.28 Yes No capita) 0.68 Yes Yes capita) 0.65 / 4.91 No No capita) -7.19 / 1.30 Yes No 1-21
Study Alcalá and Ciccone (2004) Rodrik et al. (2004) Noguer and Siscard (2005) Felbermayr (2005) Dependent variable Trade/GDP nominal Independent variable ln(trade/gdp) nominal ln(trade/gdp) real Geographical controls Institutional controls worker) 0.394 / 1.013 Yes Yes worker) 1.002 / 1.482 Yes Yes capita) -0.87 / 0.02 Yes Yes worker) -0.42 / -0.30 Yes Yes capita) -0.94 / -0.77 Yes Yes capita) 2.59 / 2.96 No No capita) 0.89 / 1.22 Yes No capita) 0.82 / 1.23 Yes Yes capita) -0.344 Yes No 1-22