Long-Run Effects of the Korea-China Free-Trade Agreement

This paper uses a 53-country 15-industry computable general equilibrium model of trade to analyze the effects of the Korea-China free trade agreement on the Korean economy, the manufacturing sector in particular. The model is based on Yaylaci and Shikher (2014) which uses the Eaton-Kortum ethodology to explain intra-industry trade. The model predicts that the Korea-China FTA will increase Korea-China manufacturing trade by 56%, manufacturing employment in Korea by 5.7% and China by 0.55%. The model also predicts significant reallocation of employment across industries with the Food industry in Korea losing jobs and other industries there gaining jobs, with the Medical equipment industry gaining the most. There will be some trade diversion from the ASEAN countries, as well as Japan and the United States.


I. INTRODUCTION
In recent years, Korea has successfully signed free-trade agreements (FTA) with a number of partners including the U.S., EU and ASEAN. Korea and China institutional framework in China to protect Korean firms and people working in China. It is believed that more than 22,000 Korean firms are currently operating in China. Third, using the already-implemented FTA with ASEAN, Korea can play a crucial role in leading potential Asian economic cooperation including East Asia and ASEAN.
This paper examines the potential effects of the Korea-China FTA on the Korean economy, in particular industry structure and employment, using a computable general equilibrium (CGE) model. 3 The model is based on solid economic theory and has been tested and evaluated in previous studies (Yaylaci and Shikher, 2014). Specifically, the model has been found to accurately predict the effects of NAFTA (Shikher, 2012a). Our forecast predicts which Korean industries would grow as the result of the FTA. In those industries, existing firms would increase sales and new firms would enter business. Our forecast can also predict which Korean industries would experience a decline in sales and, therefore, employment.
The model in this paper covers 53 countries and 15 industries. Trade in the model is affected by technology, trade costs, cross-industry supply of intermediate goods, and tastes. For each industry and country, the model can predict changes in trade, output, employment, prices, cost of production, wages, and welfare. We also plan to quantify the magnitude of the trade diversion that would occur as the result of the FTA. Countries in our model follow the Ricardian assumption and trade with each other to exploit their comparative advantages. Following the gravity model, trade costs in our model are an obstacle to international trade and create differences in goods prices in different countries (Eaton and Kortum, 2012 Compared to other models of trade, the major innovation of our model is how it explains intra-industry trade. Other models use the Armington (1969) assumption, while our model is based on the methodology of Eaton and Kortum (2002)-noted as EK hereafter. In our methodology, each industry is populated by a multitude of producers making a variety of goods, with each producer wanting to be the 120 Sunghyun Kim, Serge Shikher ⓒ Korea Institute for International Economic Policy least-cost supplier in the market. The model explicitly incorporates trade costs and uses them to explain the cross-country price differentials. Therefore, it is well suited to study the effects of changes in trade costs, such as trade wars or trade liberalizations (Costinot and Rodriguez-Clare, 2014).
The model that we propose to use to study the effects of Korea-China FTA has been extensively tested and evaluated. The model shows an extremely close fit to the data used to parameterize it, with the correlation of 0.99. More challenging evaluations of the model in Shikher (2011) andShikher (2012a) looked at the model's ability to make accurate predictions outside of the sample used to parameterize it. The first paper evaluated the ability of the model to forecast changes in specialization that occurred during 1975-95. The second paper asked the model to forecast the effects of NAFTA. It found the correlation between the actual and predicted changes in trade to be 0.95. Therefore, we have considerable confidence that the model would be able to accurately forecast the effects of Korea-China FTA. While most studies on the Korea-China FTA are focused on the policy related issues, there are several studies that use CGE type model or econometric analysis to examine the effects of the Korea-China FTA including Lim (2011), Park and Choi (2012), etc. 4 II. MODEL The model in this section is based on Yaylaci and Shikher (2014) and Shikher (2012a). The model includes N countries and J industries. We use i and n to denote countries and j and m to denote industries. We focus on the manufacturing industries in this paper. The first J − 1 industries produce manufacturing products, while the last industry produces non-manufactures.
As in the Ricardian and EK models, the only factor of production is labor. The total stock of labor is fixed in each country but labor is mobile across industries within a country. Each industry has its own Cobb-Douglas cost function: ⓒ 2015 Journal of East Asian Economic Integration where w is the wage, ρ is the price of the intermediate goods, and β is the share of labor. The intermediate good is produced by a Cobb-Douglas production function that uses goods from all industries. Therefore, the price of ρ is also a Cobb-Douglas function of p as in: where η is the share of industry m in the input of industry j, such that ∑ η = 1, ∀j. Following EK, we assume that some nonmanufacturing output can be traded without incurring any costs and use it as the numeraire: p ≡ 1. We use the same framework in EK to model intra-industry production, trade, and prices. Each industry j < J has a continuum of goods indexed by l ∈ [0,1], produced with productivity z (l). The price of good l of industry j produced in country i and delivered to country n is p (l) = c d /z (l), where d is the iceberg transportation cost. 5 The distribution of prices p is described by the following cdf: G (p) = 1 − F (c d /p) = 1 − e . Consumers in country n purchase goods from supplier with the lowest cost, and therefore the price of good l in country n can be denoted as p (l) = min p (l), i = 1, . . . , N .
The price index for the CES objective function within industry is Parameter T denoted productivity at the industry level which governs the comparative advantage among industries. Parameter θ determines the comparative advantage across goods within an industry. When the value of θ is low, it means 122 Sunghyun Kim, Serge Shikher ⓒ Korea Institute for International Economic Policy that more dispersion of productivities among producers exists, which leads to a higher comparative advantage within industry.
Next step is to derive the expressions for the industry-level bilateral trade volumes. The probability that a producer from country i has the lowest price in country n for good l is π ≡ T γc d /p . Because of the assumption that a continuum of goods exists on the unit interval, this probability is also same as the fraction of industry j good that country n buys from country i. It is also the fraction of country n's expenditure on industry j good from i : X /X . Therefore, We complete the model by describing the market clearing conditions. We have where Z is the spending on intermediate goods and Y is the spending on final goods made by industry j. We use the assumption in EK that each country consumes a constant fraction of its income on goods from each industry, α = Y /Y . We also have where Z is the industry m's spending on intermediate goods made by industry j and M is the amount that industry m spends on all intermediate inputs. Therefore, the market clearing equation is where the nonmanufacturing industry's consumption is treated as final rather than intermediate consumption.
The final model can be described by equations (1) -(5). In the model, β , η , γ, ⓒ 2015 Journal of East Asian Economic Integration θ, α , w , d , T , and Y are the parameters, and p , c , π , and L are the endogenous variables. The solution of the model is achieved by solving for the production costs using equations (1), (2), and (3), which requires solving a system of N × (J − 1) equations. 7 Once we derive the trade costs, π can be calculated from (4) and industry employments L can be derived from (5).
Combining (1), (2), and (3), we obtain the following equation for trade costs: We take logs of this equation and we get which is easier to solve numerically than (6).

III. OBTAINING MODEL PARAMETERS
The model is parameterized following a procedure first described in Shikher (2012b). The parameters are obtained as follows. Labor shares β are obtained from output and value added data. Industry shares η are obtained from input-output tables. Demand parameters α are calculated from production and trade data, as explained in this section. Wages w and country incomes (GDPs) Y are taken directly from data. The data sources are described in Section 4. Parameter θ is taken from EK, where it is estimated to be 8.28. 8 Technology parameters T and trade costs d are estimated using methodology 7 In our case, there are 53 × 15 = 795 equations with 795 unknowns (53 countries and 15 manufacturing industries). We solve the system of equations using numerical algorithm in Matlab. 8 They also obtain a second estimate of 3.6, but 8.28 is their preferred estimate since = 3.6 results in unreasonably high trade costs.

124
Sunghyun Kim, Serge Shikher ⓒ Korea Institute for International Economic Policy similar to EK's, but modified to account for multiple industries. Specifically, the price of inputs ρ is an index of industry prices p and we cannot substitute it out as done in EK.
From (4): We define S ≡ T c as an index for international competitiveness of industry j in country i. Using the definition of S , taking logs of (1) leads us to log X X = −θ log d + log S − log S .
As in EK, trade costs are proxied by where d (k = 1, . . . ,6) is the effect of distance in the kth interval, b is the effect of sharing a common border, l is the effect of common language, f is the effect of the free trade area, m is the overall destination effect, and δ represents the geographic barriers.
As in EK, we combine equations (9) and (10) to obtain the estimating equation for S and trade costs: where D = log S is the exporter dummy and D = −θm − log S is the dummy for importer. The overall destination effect is m = −(1/θ) D + D .
Consequently, the estimation produces the relative competitiveness measure S /S , . Taking logs of the definition of the relative competitiveness measure S ij , we have Note that it is necessary to remove both wages and prices from S to get technology parameters T from S , which is different from EK where only wages needed to be removed. Using the relationship = T derived from equation From (12) and (13), we can derive the expression for industry prices. We then combine that expression with (2) to get the following relationship for input prices: Finally, putting together equations (12) and (1) with the above equation, we derive the expression for the technology parameters: We use a two-step procedure in order to estimate the technology parameters. First, we estimate the gravity equation (11) to obtain exporter dummy S /S , . Then we use these estimates to derive technology parameters T /T , according to (14). We calculate the demand share parameters α using the production and trade data and α is expressed as Then, α is calculated as the average of α across the countries in the dataset.

IV. ESTIMATED TRADE COSTS AND TECHNOLOGY PARAMETERS
The model is parameterized using 2005 data for 15 industries and 53 countries. The industries are based on the 2-digit ISIC rev. 3 classification and are described in Table 1. The countries included in the dataset can be seen in Table 2.
The data necessary to estimate the gravity equation (11) was presented in Yaylaci (2012). That paper shows the evolution of trade cost between a large number of countries over a span of several decades. The data sources are as follows.
Sectoral output data comes from the United Nation's Industrial Statistics database (INDSTAT2-2010, Rev.3). The corresponding bilateral trade data is obtained from the COMTRADE database of the UN which uses the 4-digit SITC (Rev. 1) classification. Using a concordance, the 4-digit SITC Revision 1 trade data was aggregated to the 2-digit ISIC data. Missing data was filled from nearby years. The gravity data (distance, common border, common language, currency union, regional trade agreements) comes from the Gravity Database compiled by CEPII. The distance is divided into 6 intervals, as in EK. Data on the existing tariffs between the U.S. and Korea come from WITS online database of the World Bank. We derive imports from home X as output minus exports, and spending X as output minus exports plus imports. Share of labor in output, β , is calculated as the average of labor shares from all countries in our database. Table 1 presents the parameters α and β . We use the OECD input-output tables to obtain the data for industry shares η . 9 The trade costs d and technology parameters T are ⓒ 2015 Journal of East Asian Economic Integration estimated following the methodology described in Section 3. The average estimated trade cost (averaged across country pairs and industries) is 2.84, which is equivalent to 184% ad-valorem tariff. 10 The average (across country pairs) 10 According to Anderson and van Wincoop (2004), the average international trade cost among rich OECD countries is around 1.7. This is lower than the (non-weighted) average trade cost of 2.84 estimated in this paper. However, our dataset includes many less-developed countries that have much higher trade costs than the rich OECD countries. If these countries are excluded from the dataset, the average trade cost for the remaining rich OECD countries is 1.76, which is much closer to the number reported in Anderson and van Wincoop (2004). The mean productivity draws, measured by T / , are estimated for each industry j and country i. The results are presented in Table 3 for selected countries and selected industries. The mean productivity draws are measured relative to the US. Table 4 shows the rankings of the countries in these selected industries according to their mean productivity draw (i.e. "state of technology"). The U.S. has the highest or second-highest state of technology in all industries. Other developed countries have top rankings as well while the least developed countries are near the bottom of the rankings. Korea has the 7th place according to the cross-industry average of presented industry rankings (shown in the last column of Table 4). It is ahead of such countries as Spain, Australia, and Sweden. The numbers shown in Tables 3 and 4 represent the absolute advantages of each country in different industries.
Comparing mean productivity draws across industries tells us the comparative advantages of countries. The comparative advantages in turn affect the pattern of trade between countries. Since in this paper we are analyzing the trade between China and Korea, we will compare mean productivity draws of China and Korea across industries.
Korea has absolute advantage in all industries except for the Wood industry in which China has a tiny absolute advantage. Korea has a much higher variability of

V. COUNTERFACTUAL SIMULATIONS
We will now use the model described in the previous sections to predict the effects of a free-trade agreement between China and Korea. The exercise entails the removal of tariffs currently in place between the two countries. 11 The model will be solved with the tariffs removed and the results will be compared to the baseline model, which has the tariffs in place. We will especially focus on the changes in trade and employment.
The values of currently existing tariffs in Korea and China in each industry are obtained from the World Bank's WITS database. The tariffs are shown in Table 5. The level of protection varies significantly across industries. By far, the most protected industry in both countries is the Food industry where the tariffs are 24% in China and 34% in Korea. The Transport and Nonmetals industries are protected in China while the Textile industry is protected in both countries. 12 In order to simulate the China-Korea free-trade agreement, we will reduce the 11 Note that the level of tariff reduction currently described in the final agreement is not a full tariff reduction. Instead it is rather a low quality reduction over a long period of time. However, since the detailed schedule and speed of tariff reduction in each industry is still undecided when this paper was written, we analyze the most basic case with full tariff reduction. Therefore, the numbers predicted in this paper can be considered as an upper limit of the effects of actual FTA. 12 It is also interesting to compare openness of different industries in China and Korea. Openness can be measured by a ratio of exports to output. This ratio tells us what fraction of output is exported. By this measure, the food industry is fairly closed. The openness ratio in the Food industry is 0.04 in Korea and 0.10 in China. By comparison, the openness ratio in the Medical industry is 0.89 in Korea and 0.71 in China. This is despite the fact that the Food industry has a higher share of intermediate goods than the Medical industry (see Table 1). The Transport industry is more open in Korea than China: the openness ratio is 0.35 in Korea and 0.09 in China. The openness ratio is typically higher in smaller countries.  Table 5 and solve the model for industry employments, output, prices, and trade. Several factors will determine the magnitudes of trade changes. The size of the existing tariff, which is being removed, will affect trade changes. Removing bigger tariff will tend to produce bigger effects on trade. For example, since the food industry has large existing tariffs, we should expect trade to increase significantly if these tariffs are removed.
It is also important to look at the size of the tariff being removed in relation to the total trade cost in an industry. If the total trade costs are small, then removing a tariff will have a greater effect. On the other hand, if trade costs are large, then removing a tariff that only constitutes a small portion of all trade costs will not have a very large effect on trade.
The pattern of comparative advantages will also affect changes in trade. Reducing trade costs allows comparative advantages to play a bigger role in determining the pattern of trade. The pattern of comparative advantages will have an especially strong effect on employment due to trade liberalization. Generally, a country with the comparative advantage will gain employment while the other country will lose employment.
Finally, with trade liberalization there will be trade diversion. For example, let's consider three countries, A, B, and C, with A importing good X from C before liberalization. If A reduces tariffs on X coming from B, then B may become a cheaper source for X in A, so trade will divert from C to B. There can potentially be large trade diversion due to Korea-China free-trade agreement because China currently has a free-trade agreement with ASEAN countries. It means that ASEAN countries, such as Malaysia, Philippines, and Indonesia currently enjoy low trade barriers in China while Korean goods are covered by tariffs. With Korea-China FTA, Korean goods will compete on a level playing field with the ASEAN countries in China. So at least some of the goods that China currently sources from the ASEAN countries will be sourced from Korea once the FTA is implemented. In addition, with Korea-China FTA, some of the goods that China currently buys from the U.S., especially Machinery goods, may be sourced from Korea once the FTA is implemented, since Korea and the U.S. are close competitors in Machinery. Table 6 shows the effects of the Korea-China FTA on the bilateral manufacturing trade between the two countries. The model predicts that, everything else equal, the Korea-China FTA would increase Korea's manufacturing exports to China by 61.6% and China's manufacturing exports to Korea by 48.4%. The greatest increase in trade would occur in the Food industry. This is because the Food industry had the highest level of tariffs before the FTA. The second-highest trade increases would occur in the Textile industry, which was also heavily protected by tariffs before the FTA. Table 7 shows the specialization before the FTA. Tables 8 shows what happens to specialization, measured by industry shares in total manufacturing employment, and welfare as the result of the FTA. Table 9 presents percent changes in industry employments. Looking at Table 7, we note that the current pattern of specialization is different in Korea and China. In Korea, the Electrical and Communications Machinery industry has the greatest share of manufacturing workers, 20.9%. The largest industry in China by this measure is Textile. Tables 8 and 9 show that industry-level changes that occur due to the FTA are also different in Korea and China. For example in Korea, the Food industry shrinks significantly, while the Medical industry expands. In China, the Food industry grows while the Medical industry shrinks. To understand what happens in the Food industry, we need to look at Tables 3  and 4, which show comparative advantages, and Table 5, which shows existing (pre-FTA) tariffs. From Table 5, we know that the Food industry has high existing tariffs. Therefore, we should expect a lot of new trade after FTA is implemented. This is what we see in Table 6. Tables 3 and 4 tell us that China has comparative advantage in Food: the productivity in Chinese Food industry is just a bit below that of the Korean Food industry, while generally, China's productivity is much lower than Korea's. Since China has comparative advantage in Food, production shifts to China when trade is liberalized. The employment in Chinese Food industry grows together with its share in Chinese manufacturing. Korean Food industry shrinks in terms of absolute employment as well as share of manufacturing. Now, let's take a look at the Textile industry. Both Korea and China have relatively high tariffs in this industry, so there is a significant post-FTA increase in trade. Table  3 and especially Table 4 tell us that Korea has a comparative advantage in the Textile industry, though the advantage is moderate. Therefore, post-FTA Korea increases its specialization in Textile and employment in that industry grows. China decreases its specialization in Textile, but not much. Despite the decline of the share of Textile in total Chinese manufacturing, the employment in China's Textile industry grows a little because of the growth of total manufacturing employment.
Korea's comparative advantage is much more pronounced in the Rubber industry, where its productivity is nearly twice as high as China's. As the results of the FTA, production in that industry shifts to Korea. On the producer level, the EK model implies the following. Korean producers that were not competitive in China pre-FTA can now out-compete the Chinese firms that make the same products. These Korean producers are more productive than the Chinese firms that they drive out of business, but less productive than the Korean producers that were exporting to China even before the FTA. As the results of the FTA, Korean exports to China of Rubber products increase, but exports become a smaller fraction of output. The mirror image of this happens in China: there is less output in the Rubber industry, but a greater fraction of output is exported.
In the Medical industry, the current total cost of importing goods from Korea to China is lower than the cost of importing from China to Korea. At the same time, tariff reductions that occur with the FTA are similar in both countries. This means ⓒ 2015 Journal of East Asian Economic Integration that the FTA reduces trade costs from Korea to China proportionally more than the trade costs from China to Korea. This is one reason why Korea's exports to China in this industry increase more than China's exports to Korea.
Korea has a comparative advantage in the Medical industry, so specialization in this industry increases in Korea and decreases in China as the result of the FTA. In fact, Medical industry in Korea benefits the most from the FTA -its employment grows 13.46%. There is also significant trade diversion in the Medical industry due to the Korea-China FTA. A big portion of the increase in Korean exports to China come at the expense of the exports of ASEAN countries and some developed countries, such as Japan. For example, the employment in the Medical industry in Philippines declines 6.58% as the result of the Korea-China FTA. In Japan, the employment in this industry declines 3.4%, in the U.S., 0.64%.
The FTA has positive overall effects on the Korean and Chinese economies. The last column of Table 9 shows that the total manufacturing employment grows in both countries, but more so in Korea. This means that labor shifts from agriculture and services to manufacturing. The last column of Table 8 shows the welfare effects of the FTA. Prices of manufacturing goods fall as the result of the FTA in both countries and, therefore, welfare increases. As typical in FTA analyses, our model predicts moderate (but permanent) welfare effects of the FTA: 0.18% in China and 0.27% in Korea. 13 In terms of the importance for the economies involved, the Korea-China FTA ranks above the Korea-U.S. FTA. The Korea-China FTA increases bilateral trade by 56% and increases manufacturing employment by 5.67% in Korea and 0.55% in China. The Korea-U.S. FTA is projected to increase bilateral Korea-U.S. trade by 31%, manufacturing employment in Korea by 0.97% and the U.S. by 0.26%. The Korea-China FTA may be compared to NAFTA, which increased U.S.-Mexico trade by about 60-70%.

VI. CONCLUSION
Korea-China free-trade agreement can potentially have a very significant impact on the economies of Korea, China, and even other countries. In this paper, we use a computable general equilibrium (CGE) model of the world economy to predict the economic effects of this agreement. Our model includes 53 countries and 15 industries and, unlike most other CGE models, uses the EK methodology to explain intra-industry trade instead of the Armington assumption. This means that our industries are populated by many different producers instead of the representative producer. Consumers choose to buy from a producer that can out-compete others, rather than basing their decisions on the national origin of the producers. Technology and trade costs play key roles in our model in determining the patter of trade and specialization. The model that we use to predict the effects of the Korea-China FTA has been previously evaluated in several historical simulations, including NAFTA, and found to make accurate predictions. We simulate the effects of Korea-China FTA by removing all existing tariffs on manufactured goods between the two countries. The simulation results show that the bilateral trade in manufactures between Korea and China increases by 56% as the result of the FTA. The largest trade increases occur in the Food industry which is currently the most protected.
There are also significant changes in specialization and industry employment mostly driven by the pattern of comparative advantages. In Korea, the Food industry contracts the most. Textile, Chemicals, Rubber, and Medical equipment industries expand. There is also trade diversion in some industries, especially from the ASEAN countries, but also from Japan and the United States. We find large effects on the Korea economy as the result of the FTA. Prices of traded goods decrease as the result of the FTA and welfare increases. Manufacturing employment increases by 5.7% and there is a large reallocation of workers across industries. The Food industry loses almost 12% of its workforce while Medical equipment industry increases its workforce by 13.5%. We find that the Korea-China FTA can have greater effects on trade and employment of Korea than the Korea-U.S. FTA.