Processing Trade Policy, Export VAT Rebate and International Trade: Evidence from Chinese Firm-level Data

Transcription

1 rocessing Trade olicy, Export VAT Rebate and International Trade: Evidence from Chinese Firm-level Data Jianpeng Deng March 12, 2016 Abstract rocessing trade policy is often used by developing countries, and export value-added tax rebate is common for countries at different stages of economic development. This paper extends Eaton, Kortum and Kramarz 2011 to investigate these policies. We use simulated method of moments estimate the parameters, and utilize the additional information from processing trade to identify the elasticity of substitution. In the counterfactual exercise of China shutting down the processing trade, its exports are reduced by 25%, but welfare is increased by 8%. 1 Introduction rocessing trade policy, allowing firms to claim import duty exemption for imported material used for producing exports, is commonly used in developing countries. Among them, China and Mexico are the two largest users of processing trade policy in the world, together accounting for about 85% of worldwide processing exports from 2000 to 2008, in which China alone accounted for about 67% while Mexico represents another 18% Maurer and Degain Moreover, for both China and Mexico, processing exports make up around half of their total exports 1

2 during this period Koopman, Wang and Wei 2012, Bergin, Feenstra and Hanson Value-added Tax VAT is a key component of the tax system in over 140 countries at different stages of economic development, raising about 25% of world tax revenue Harrison and Krelove Export VAT rebates are common in these countries having VAT, such as EU, Canada, China, etc. 1 How do these policies affect aggregate bilateral trade and production? What are the welfare implications? To answer these questions, we need a multi-country general equilibrium model with these policies embedded into, which also rationalizes key features at the firm-level data. This paper follows the approach of the seminal work of Eaton Kortum and Kramarz 2011 EKK2011, and extend their model by including processing trade policy and export VAT rebate. On the empirical side, EKK2011 found striking regularities associated with French manufacturing firms entry patterns and sales distributions across 113 destinations, including France itself. Since Chinese custom data separates the processing trade and ordinary trade at the firm-level 2, we redo the empirical exercises along these two dimensions as EKK2011, and we find that processing trade is almost identical to ordinary trade at the firm-level in terms of entry patterns and sales distributions across markets, which also coincides with those in French data. On the theory part, the structure of Melitz 2003, as augmented by Helpman Melitz and Yeaple 2004 and Chaney 2008, which is characterized by areto distribution of firms efficiency, Dixit-Stigliz demand, iceberg trade cost and fixed cost of entry, fails to term with some features in French manufacturing firms entry patterns and sales distributions: i Firms do not enter markets according to an exact hierarchy. ii Their sales in these penetrated markets deviate from the exact correlation that the basic model predicts. iii Exporting firms sell too much in their domestic market. iv Too many firms sell small amounts in a destination. EKK2011 introduces market and firm-specific heterogeneity in entry 1 I don t have the data about export VAT rebate share of total VAT revenue across countries, but Harrison and Krelove 2005 provided the VAT refund share of total VAT revenue for some countries, and export VAT rebates dominate the VAT refund. 2 To be precise, ordinary trade refers to transactions that firms sell their products in domestic market or foreign markets without claiming tariff exemption for imported material used to produce these products; processing trade refers transactions that firms sell their products in foreign markets with claiming tariff exemption. 2

3 costs and demand to reconcile the first two features and deals with the last two by incorporating Arkolakis s 2010 formulation of market access costs. This extended structure can connect more formally to the French firm-level data. The model in this paper replicates the extended structre in EKK2011 for processing trade. ut it differently, a firm independently solves an identical optimization problem regarding to enter a market via processing trade and ordinary trade, which only differ in some parameters and realizations of shocks. We also use the simulated method of moments to estimate all parameters in the model except the elasticity of substitution the same as EKK2011, but with two more moments related to processing trade. By utilizing the additional information of processing trade, we can identify and estimate the elasticity of substitution. With all the estimates of parameters, we compute the counterfactual equilibrium in which China shuts down the processing trade policy. China s export are reduced by 25%, and welfare is improved by around 8%. The most popular exporting destinations of Chinese firms, such as U.S., Japan and U.K. etc, have welfare loss but less 1%, while China s competitors like Vietnam, Thailand have welfare gain, but also less than 1%. The intuition behind these results is quite straightforward. With shutdown of processing trade policy in China, Chinese firms sell more varieties in domestic market and export less varieties to foreign markets, and the price index declines much more than the wage rate in China; rice indices in countries, such as U.S., Japan and U.K. etc, increases more than the wage rate since less varieties imported from China; Countries, such as Vietnam and Thailand etc, start to produce and export more varieties, driving up the real wage. This paper relates to a large literature of evaluating trade polices and is mostly related to studies that quantify potential welfare gains and costs from trade polices. In particular, Caliendo and arro 2015 extends Eaton and Kortum 2002 to multi-sectors to evaluate the trade and welfare effects of North American Free Trade Agreement. Relative to Caliendo and arro 2015, we investigate welfare implications of processing trade policy and extends EKK2011 to connect formally to Chinese firm-level data for processing and ordinary trade, respectively. Section 2 explores three empirical regularities as EKK2011, separating ordinary exports and processing exports. Section 3 presents the extended model which captures the empirical regularities described in section 2. Section 4 explains how 3

4 Figure 1: Entry and sales by market size we estimate the model. Section 5 defines the general equilibrium and calculates counterfactual. 2 Empirical Regularities We show the entry patterns and sales distributions of Chinese firm-level data in 2005 as EKK2011 does it for French firm-level data, but separating ordinary exports and processing exports. 2.1 Market Entry anel A and B in Figure 1 plot the number of Chinese manufacturing firm selling to a market against total manufacturing absorption in that market via ordinary trade and processing trade, respectively. The number of firms selling to a market tends to increase with size of the market regardless of trade mode. 4

5 Table 1: Chinese Firms Selling to Strings of Top 7 Countries Number of Chinese Exporters Export String Ordinary rocessing US US-J US-J-KR US-J-KR-DE US-J-KR-DE-UK US-J-KR-DE-UK-CA US-J-KR-DE-UK-CA-AU We normalize the number of firms by the share of China in the market for both ordinary trade and processing trade, and plot the normalized entry against the total manufacturing absorption in that market, which are shown in panel C and D of Figure 1. Now the relationship is not only very tight, but linear in log especially for ordinary trade. Table 1 lists each of the strings of top-seven destinations that obey a hierarchical structure with ordinary trade and processing trade. 2.2 Sales Distribution We plot the sales of each firm in a particular market relative to mean sales there against the fraction of firms selling in the market who sell at least that much. Figure 2 plots results of ordinary trade and processing trade for U.S., Japan, Korea and Germany. Graphs for ordinary export and processing export of all four countries almost identical, which implies that we can introduce a different fixed cost for processing export to match this fact. 5

6 Figure 2: Ordinary and rocessing Sales Distribution by Country 3 Model The quantitative general equilibrium model builds on Eaton Kortum and Kramarz There are N countries. Countries are denoted by i or n. Labor is the only factor of production. Labor is immobile across countries. 3.1 Technology A continuum of intermediate goods is produced. The production technology of intermediate good ω from country i with efficiency z i ω is given by q i ω = z i ω [l i ω] β [m i ω] 1 β, where l i ω is labor and m i ω is the composite intermediate good used for production of intermediate good ω. The parameter β 0 is the share of value added 3. 3 Composite intermediate goods are put into the production function above only describing input usage of an average firm in a country, hence, no heterogeneity on the share of input usage across firms, but it is necessary to include input usage to model processing trade policy. 6

7 3.2 olicies Two policies are exogenously chosen by countries. The first is processing trade policy: a firm can claim import duty exemption for imported inputs used for producing exported goods, and no import duty exemption for imported inputs for domestic sales. In the model, firms can claim import duty exemption for imported intermediate goods aggregated in the composite intermediate goods used as inputs for producing exports. The second is export value-added tax VAT rebate policy: a firm can claim a full or partial refund of VAT applied to exported goods. To simplify the notation, let O and denote ordinary trade and processing trade, respectively, and let denote trade mode that can be either O or. Let I i denote the choice of processing trade policy in country i, where I i equals to 1 if country i chose processing trade policy, and 0 otherwise. Let τ denote 1 plus the ad-valorem flat-rate tariff of intermediate goods imported by country n from country i. Let t denote the applied VAT rate for the sales generated by firms from country i selling products to market n, particularly, { VAT rate in country i, if n = i; t = VAT rate minus export VAT rebate rate in country i, if n i. A firm from country i can potentially sell its products in market n through ordinary trade, processing trade or both. Suppose a firm from country i selling its products in market n through both ordinary and processing trade. We consider products sold through ordinary and processing trade to be two different intermediate goods, labeled as ω O and ω, respectively, because the composite intermediate goods used for producing ω O is different from that for ω. To be more specific, the compositions of intermediate goods in these two composite intermediate goods are different since the prices of the imported intermediate goods are different between ordinary and processing trade. Moreover, assume that z i ω O = z i ω = z i ω, which simply means that a firm has identical efficiency level to produce intermediate goods sold through ordinary and processing trade. 7

8 3.3 Ut Costs and Firm Heterogeneity Since production of intermediate goods is at constant return to scale, the cost of a bundle of inputs used for trade mode in country i is given by c i = Υ w i β i 1 β, = O or, 1 where w i is wage, i is the price of the composite intermediate good used for producing intermediate goods sold through trade mode, and Υ is a constant. If country i did not choose processing trade policy, there is no processing exports. The ut cost to a potential firm producing good ω from country i with efficiency z i ω delivering 1 ut to country n through trade mode is given by c d ω = ct i M z i ω, = O or, where d describes the iceberg cost of delivering intermediate goods from country i to country n. The measure of potential firms in country i with efficiency higher than z is µ z i z = T i z θ, z > 0, where θ and T i are parameters. Hence the measure of intermediate goods that can be delivered from country i to country n through trade mode at ut cost below c is µ c = Φ c θ, = O or, where Φ = T i c i d θ, = O or. 3.4 Demand, Market Structure and Entry A market n contains a measure of potential buyers. To sell to a fraction f of them, a firm from country i selling good ω in country n through trade mode must incur a fixed cost E ω = ε n ω E M f, = O or, 8

9 where E is the constant component of the cost faced by all sellers from country i in destination n through trade mode 4 ; ε n ω is the fixed cost shock specific to good ω in market n. The function M f = 1 1 f1 1/λ, the same across destinations, relates a seller s cost of entering a market to the share of consumers 1 1/λ it reaches there. The demand of good ω from country i in market n is given by X ω = α n ω fp 1 σ A, where A = X O n O n /τ σ 1 + I n X n n σ 1, which is an index of market demand of country n that proportionally scales the residual demand of every firm from country i; Xn O is the spending on goods consumed by households in country n or absorbed by firms in country n to produce domestic sales or ordinary exports; Xn is the spending on goods absorbed by firms in country n to produce processing exports, thus, X n = Xn O + I n Xn is the total absorption in country n. α n ω is the demand shock specific to good ω in market n 5. The demand of good ω from country i in market n is composed of two parts: the first part is for domestic use and producing ordinary exports, and the second for producing processing exports if country n chose processing trade policy, otherwise the second part is zero. Since the tariff rates are country-specific, the index of market demand in country n is different for firms from different countries. A firm producing good ω in country i selling in market n via trade mode with a ut cost of c n ω chooses price p and a fraction of f buyers to maximize its profit in the market p cn ω max p,f p p 1 β c n ω t p α n ω fp 1 σ A ε n ω E M f, = O or. 4 To simplify the expositon, we set E nn equal to infity, therefore, no firms sell products in domestic market through processing trade in equilibrium, which is consistent with the fact processing trade policy does not allow firms to claim tariff exemption for imported inputs used to generate domestic sales. 5 The demand shock or fixed cost shock are only specific to the good variety, so we don t need the superscripts to label these shocks. 9

10 We now describe a firm s behavior in market n in terms of its ut cost c n ω = c, demand shock α n ω = α, and entry shock η n ω = αnω ε nω = η. By solving the maximization problem, we have where p n ω = mζ c, where m = σ σ 1 and ζ = 1 1 β t 1 t 0, if c > c T f M η ; η, c = λσ 1 c 1, if c c η, c η c T M η = 1 η 1 t 1/σ 1 A mζ σe. Firms charge a higher markup with VAT than without, if not, then the marginal revenue is smaller compared to the case of no VAT since some of it is taxed away, so firms have to raise the price to increase the marginal revenue in order to maximize profits. 6 At the moment, we can take a glance at how these two policies to affect the cutoff in domestic market. c O nn η = 1 η 1 t 1/σ 1 nn A σenn O nn 1 [ ] 1/σ 1 η A m σenn O nn mζ nn O n m [ η ] 1/σ 1 X σenn O n, the first inequality holds because ζ nn 1 and 1 t nn 1, and the second inequality is from the fact that n n O, which will be very clear in the following subsection. The expression on the second line is the cutoff only with processing trade policy, which is bigger than the cutoff also with VAT because the firm with c O nn η will make positive profit for not paying VAT. The expression on the third line is the cutoff without both VAT and processing trade policy, and it is bigger than the cutoff only with processing trade policy because the processing trade policy makes the input price higher for firms selling in domestic market or foreign markets 6 In a simple closed economy, VAT can cause the welfare loss of the representative agent, because it induces less varieties, and make the price index higher. 10

11 through ordinary trade than firms through processing exports. The total sales via trade mode can be written as [ X α, η, c = α ] λσ 1 σ 1 c c σe 1, η c T M η c T M η 1 t = O or. Some of the profits is eaten up by its fixed cost, which are E α, η, c = α η E 3.5 rice Index 1 c/c T M η λ 1σ 1 1 1/λ, = O or. Each buyer in market n has access to the same measure of goods even though they are not necessarily the same goods. Every buyer faces the same probability of f η, c to purchase a good sold via trade mode from country i with cost c and entry shock η for any value of α. Hence we can write the price index faced by a representative buyer in market n as O n [ = m O 1 σ n = m 1 σ where κ 0 = i c η 0 ακ 0 αi T i M f η, c ζ τ c 1 σ dµ T M c i I i θ θ, θ σ 1 θ+σ 1λ 1 ζ τ 1 σ Φ [ c g α, η dαdη η ] θ σ 1 g α, η dαdη O 1 σ N n = m θ κ 1 Ψ [A τ σ 1] θ/σ where Ψ = ΨT M, and i=1 ] 1/σ 1 7 Ψ = I T i M Φ T M θ/σ ζ τ θ t, = O or, σe 7 I O i = 1 for all i, which is included to make the mathematical equations shorter. 11

12 and κ 1 = αη θ/σ 1 1 g α, η dαdη. Repeat the same calculation, and we can get the price of the composite intermediate good used as input to produce processing exports if country n chose processing trade policy, 1 σ N n = m θ κ 1 τ σ 1 Ψ A τ σ 1 θ/σ i=1 Since τ σ 1 1, O n n. If I n = 0, then n is not defined, therefore, the price index O n where Ψ n = N i=1 Ψ. is simplified to the expression of price index in EKK2011 O n = m κ 1 Ψ n 1/θ X n 1/θ 1/σ 1, 3.6 Sales, Entry and Fixed Costs The total absorption in country n can be divided in the following way X n = X O n + I n X n = X O,O n + X O, n + I n X,O n + Xn,, where Xn O,O is the spending on goods via O consumed by households in country n or used by firms in country n to generate sales via O; Xn O, is the spending on goods from all other countries chose processing trade policy via consumed by households in country n or used by firms in country n to generate sales via O; Xn,O is the spending on goods via O only used by firms in country n to generate sales via ; Xn, is the spending on goods from all other countries chose processing trade policy via only used by firms in country n to generate sales via. To obtain X,, we calculate the total amount with a given values of α 12

13 and η, then integrate across the joint density g α, η to get X, = = X, c η = π, Xn, 0 α, η g α, η dαdη = O or and = O or, X, α, η, c dµ c where π, = Ψ T M [A τ σ 1 ] θ/σ 1 1 N i=1 Ψ [A τ σ 1 ] θ/σ 1 1, if = O, τ σ 1 Ψ T M [A τ σ 1 ] θ/σ 1 1 N i=1 τ σ 1 Ψ [A τ σ 1 ] θ/σ 1 1, if =. 4 Therefore, the spending on goods from country i consumed by households in country n or used by firms in country n to produce O is given by X O = X O, = π O, Xn O = π O X O n, where π O = Ψ [ A τ σ 1] θ/σ 1 1 N i=1 Ψ [ A τ σ 1] θ/σ 1 1 ; 5 the spending on goods from country i used by firms in country n to produce processing exports is given by X = X, = π X n, = π, Xn 13

14 where π = τ σ 1 Ψ [ A τ σ 1] θ/σ 1 1 N i=1 τ σ 1 Ψ [ A τ σ 1] θ/σ With equations and 6, we have π = π O n O n τ σ 1, 7 and the market demand index can be written as A = X O n = O n /τ σ 1 + I n X n Xn DO + I n Xn n σ 1 π O σ 1 π O n /τ. Comparing π O nn and π nn, it implies that the domestic sales or ordinary exports uses less imported inputs than processing exports. This is key point made by the accounting exercise of Koopman, Wang and Wei For firms from country i with a given value of η, a measure µ T M c η pass the entry hurdle via. To obtain the total measure of firms from country i that sell in n via, denoted by J, we integrate across the marginal density g 2 η, J = µ = κ 2 κ 1 1 t σe where κ 2 = η θ/σ 1 g 2 η dη. c η g 2 η dη = Xn O π O, Φ + I n X n π, c η θ g2 η dη, = O or, To obtain an expression for the fixed costs incurred in market n by firms from 14

15 country i via, given their values of α and η, we integrate E α, η = c η 0 = α η ΦT M E α η ET M 1 c/c T M 1 1 1/λ σκ 1 = αη θ/σ t = O or. [ 1 η λ 1σ 1 1 1/λ θ θ + σ 1 λ 1 X O n π O, dµ + I n X n π, c ] c η θ λ θ/ σ 1 + λ 1 Integrating across the joint density g α, η, total fixed cost incurred in market n by firm from country i via are E = 1 t σ X O n π O, + I n X n π, θ σ 1, = O or. 8 θ Summing across sources i, the total entry costs incurred in market n is E n = 1 σ θ σ 1 θ 3.7 Connecting the model to data 1 t Xn O π O + I n Xn π. i Entry patterns and sales distributions are almost identical in the Chinese firmlevel data for processing trade and ordinary trade, and French firm-level data also reveals these robust features that can be rationalized by the model proposed in EKK2011. Model in this paper replicates the structure in EKK2011 for processing trade and ordinary trade. In particular, firms idenpendently face the same optimization problem to sell in a market via processing trade and ordinary trade, only differing in some parameters and realizations of shocks, therefore, identical entry patterns and sales distributions are generated, which is consistent with what we observed in the data. Details are delegated in appendix., 15

16 4 Estimation We estimate the elasticity of substitution by utilizing the additional information from processing trade. We estimate other parameters as the estimation procedure proposed by EKK2011, adding two more sets of moments related to processing exporters. 4.1 Estimation of the Elasticity of Substitution Using equation 7, the ratio of China s processing import share of goods from country n to China s ordinary import share of goods from country n is given by π Cn π O Cn σ 1 = C τ Cn. O C This ratio only depends on the price indices of composite intermediate goods used by Chinese ordinary exporters and processing exporters respectively, the import tariff rate specific to country n and value of elasticity of substitution. Since Chinese firms independently the identical optimization problem to sell to a market via processing trade and ordinary trade, we can think that the processing trade policy replicates the manufacturing sector in China. Two identical sectors use different composite intermediate goods as inputs, but these two composite intermediate goods only differ in composition of intermediate goods because the the identical set of intermediate goods are aggregated in them, but prices for imported intermediate goods are different for ordinary import and processing import, and the difference is the import tariff. For the sector doing processing trade, it receives the import duty exemption and pays less for the imported intermediate goods, so the ratio of these two import shares reveals the substitutability among varieties. ick another country h, and we have the similar expression π Ch π O Ch σ 1 = C τ Ch. O C Dividing both equations above and taking log, the ratio of price indices is canceled out, and we obtain a relationship between observables and the parameter 16

17 describing the substitutability among intermediate goods: π log Cn / π Ch = σ 1 log τ πcn O πch O Cn /τ Ch. We can estimate σ from the following regression π log Cn / π Ch = σ 1 log τ πcn O πch O Cn /τ Ch + ε nh, where ε nh captures the measurement error in bilateral trade flows, assumed to be orthogonal to tariff rates. 4.2 arametrization We Assume that g α, η is joint log normal. Specifically, ln α and ln η are normally distributed with zero means and variances σ 2 α and σ 2 η, and correlation ρ. Let θ = θ, then we can write κ σ 1 1 and κ 2 as κ 1 = [ θ θ 1 ] θ θ + λ 1 2 σ α + 2ρσ α σ η θ 1 + σ η θ 1 exp 2, and 2 θση κ 2 = exp 2. We are left with only five parameters to estimate: } Θ = { θ, λ, σα, σ η, ρ. These parameters are estimated by simualted method of moments. Details for simulation algorithm, moments constructed and estimation procedure are delegated in appendix since we closely follow EKK2011 and add information regarding to processing trade. 17

18 Table 2: Estimate of Elasticity of Substitution Year dif log porat 2005 dif log tariff Constant Observations 2850 t statistics parentheses Table 3: Estimation Results θ λ σ α σ η ρ EKK Ordinary and rocessing Trade Ordinary Trade Results Table 2 reports the estimate of σ, 4.07, while EKK2011 calibrated σ as Table 3 reports estimates for all other parameters, values of which are not far away from those in EKK2011, except the correlation between the entry shock and demand shock. The results suggest that if a French firm s product is more popular bigger demand shock, then it is more likely for him to pay a larger fixed cost a small entry shock and generate a bigger sales. While the opposite is true for Chinese firms, one possible explanation is that almost 60 percent of Chinese total exports are done by foreign firms, parents of these foreign affiliates may provide connections and resources to facilitate entry. 18

19 5 General Equilibrium and Conterfactuals Additional assumpsions are required before formally defing the general equilibrium. A1 Non-manufacturing is as in Alvarez and Lucas Final output, which is non-traded, is a Cobb-Douglas combination of manufactures and nonmanufactures, with manufactures having a share γ. Labor is the only input to manufactures. Hence the price of final output in country n is proportional to n O γ 1 γ w n. A2 Fixed costs pay for labor in the destination, We thus decompose the countryspecific component of the entry costs E O = w n F O and E = w n F, where F O and F reflect the labor required for entry for seller from i in n through domestic sales or ordinary exports and processing exports, respectively. A3 Each country n s manufacturing deficit D n and overall trade deficit Dn A held at their 2005 values. are We write country n s total absorption of manufactures is the sum of final demand and use as intermediates as: X O n + X n = γi n + N i=1 1 β mζ in πin O Xi O τ in + I i π inx i, 9 and the total absorption of manufactures by processing exporters is given by X n = 1 β mζ nn i n π O, Xi O in τ in + I i π, in X i, 10 where I n = w n L n + R n + D A n + Π n, represents final absorption in country n, as the sum of labor income, tariff revenues, trade deficit and pre-tax profit. In particular, R n = N i=1 τ 1 πin O Xn O, τ in 19

20 and Π n = N i=1 mζ in 1 mζ in πin O Xi O τ in + I i Xi πin N i=1 E O in + E in. Finally, using the defition of expenditure and trade deficit we have that N i=1 π O Xn O τ + I n πx n D n = N i=1 πin O Xi O τ in + I i π inx i. 11 This condition reflects the fact that total expenditure, excluding tariff payments, in country n minus trade deficits equals the sum of each country s total expenditure, excluding tariff payments, on tradable goods from country n. We now define formally the equilibrium under tariff {τ }, applied VAT rate {t } and process trade policy choice { } I n in this model. Defition: Given L n, D n, T n, d and fixed entry costs, an equilibrium under tariff structure, applied VAT rate and processing trade policy choices, is a wage vector w R N ++ and prices O, R N ++ that satisfies equilibrium conditions and 11 for all n. 5.1 Compute Counterfactual Outcomes It will be good to have Mexican processing trade data, unfortunately, I don t have it now. Due to the data limitation, I treat China to be the only country chose processing trade policy. We consider that China shuts down the processing trade policy, and we apply the method used in Dekle, Eaton and Kortum 2008; henceforth DEK to calculate counterfactuals. Denote the counterfactual value of any variable x as x and define x = x /x as its change. equilibrium is characterized by the following equations: π O = π O,O ŵ i θβ O i θ1 β 1 + I n N k=1 πo,o nk ŵ k θβ O k θ1 β 1 + I n X θ/σ 1 1 X O X nk X O nk The counterfactual θ/σ 1 1, 12 20

21 n O N = where and i=1 π O,O θ1 β ŵ i θβ O 1 i 1 + I n X O n = N i=1 1 β mζ in X X O π O in θ/σ 1 1 I n = ŵ n w n L n + R n + D A n + Π n, R n = N i=1 τ 1 τ π O X O n, 1/θ X O n X O n ŵ n 1/θ 1/σ 1, 13 Xi O + γi τ n, 14 in [ Π n = N mζ in 1 π O X O N ] i θ σ 1 in 1 t Xi O πin O mζ in τ in σθ i=1 finally, the counterfactual trade balance is given by N i=1 π O X O n τ D n = N i=1 π O in We present a step by step procedure to solve the model. X O i τ in. 15 Step 1 Guess a vector of wages ŵ and a vector of price indices O. Step 2 Use equation 12 to calculate the bilateral trade shares given by ŵ and O. Step 3 Use equation 15 to solve the total absorption of manufactures in each country given by the bilateral trade shares from step 2, ŵ and O. Step 4 Use equation 14 to obtain the new vector of wages ŵ. Step 5 Use equation 13 iteratively to calculate the new price indices given the new vector of wages ŵ until it converges. Go back to Step 1 with the new 21

22 vector of wages ŵ and the new vector of price indices O, and stop until the vector of price indices converge. Table 4 reports the counterfactual results. China s welfare is improved by around 8%. The most popular exporting destinations of Chinese firms, such as U.S., Japan and U.K. etc, have welfare loss but less 1%, while China s competitors like Vietnam, Thailand have welfare gain, but also less than 1%. The intuition behind these results is quite straightforward. With shutdown of processing trade policy in China, Chinese firms sell more varieties in domestic market and export less varieties to foreign markets, and the price index declines much more than the wage rate in China; rice indices in countries, such as U.S., Japan and U.K. etc, increases more than the wage rate since less varieties imported from China; Countries, such as Vietnam and Thailand etc, start to produce and export more varieties, driving up the real wage. A different way to understand the welfare gain of China is that processing trade policy is a destination based distortion, and removing this distortion is welfare improving for China. Then a natural question is that why countries like China adopt processing trade policy. Reasons could be attracting foreign direct investment, accumulating foreign reserve and creating jobs etc, which are not included in the model. We can interpret the value of China s welfare gain as a long-run result, and an upper bound in the short run, because all these incentives will be weakened in the long-run, and the policy will become less effective as import tariff declines 8. 8 As tariff decline, less firms in China choose processing trade, which is empirically verified by Brandt and Morrow

23 Table 4: Counterfactual Countries Wage Change rice Change Welfare Change Algeria Argentina Australia Austria Bangladesh Belgium Brazil Canada Chile China Colombia Czech Republic Denmark Finland France Germany Greece Hungary India Israel Italy Japan Korea, Rep Mexico Morocco Netherlands New Zealand Norway akistan eru oland ortugal Russian Federation Slovak Republic Vietnam South Africa Spain Sweden Switzerland Thailand Turkey Uted Kingdom Uted States ROW

24 References [1] Alvarez, Fernando, and Robert Jr. Lucas 2007: General Equilibrium Analysis of the Eaton-Kortum Model of International Trade, Journal of Monetary Economics, Vol. 54, No. 6, [2] Bergin, aul R., Robert C. Feenstra and Gordon H. Hanson 2009: Offshoring and Volatility: Evidence from Mexico s Maquiladora Industry, American Economic Review, 994: [3] Caliendo, Lorenzo, and Fernando arro 2014: Estimates of the Trade and Welfare Effects of NAFTA, Review of Economic Studies, forthcoming. [4] Eaton, Jonanthon, Samuel Kortum, and Francis Kramarz 2011: An Anatomy of International Trade: Evidence from French Firms, Econometrica, Vol. 79, No. 5, [5] Harrison, Graham, and Russell Krelove 2005: VAT Refunds: A Review of Country Experience, IMF Working aper, W/05/218. [6] Koopman, Robert, Zhi Wang, and Shang-Jin Wei 2012: Estimating Domestic Content in Exports When rocessing Trade is ervasive, Journal of Development Economics, Vol. 99, No. 1, [7] Maurer, Andreas, and Christophe Degain 2005: Globalization and Trade Flows: What You See Is not What You Get, Staff Working aper ERSD

25 Appendix A: Connecting the Model to Data Transform the efficiency of any potential firm in China as EKK2011 u ω = T C z C ω θ. We refer to u ω as firm ω s standardized ut cost, and standardized ut costs have a uform measure that does not depend on any parameter. Associated with the entry hurdles c satisfying u c T M η = Φ η is standardized entry hurdles u η, 1/θ η, = O or. Firm ω enters market n via if its u ω and η n ω satisfy and we have u u ω u = u η n ω η = η θ/σ 1 1 t κ 1 σe η = Φ c η, π O,O XO n + I n π,o X n. Conditional on firm ω s passing on this hurdle, its sales in market n via in terms of u ω is given by X [ ω = ε n ω 1 u ] λσ 1/θ u ω u ω η n ω η n ω We equate the measure of Chinese firms J in each destination with the actual in u σ 1/θ σe 1 t. through trade mode selling teger number N M, and equate XT their average sales, while X in the model. = πo, Xn O+I n π, Xn J 25

26 Entry We have that is, N = κ 2 1 t κ 1 σe π O, σe 1 t = κ 2 κ 1 X. Xn O + I n π, Xn, Using equations above, the standardized cost hurdle can be written as u η n ω = N η n ω θ, κ 2 where θ = θ. This equation is the same as the equation 29 in EKK2011. With σ 1 or without import tariff, VAT, export VAT rebate and processing trade policy, the cutoff of the standardized cost only depends on the number of entries and the entry shock. Sales in a Market Conditional in a firm s entry into market n through domestic sales or ordinary exports, the term v ω = u u ω η n ω is distributed uformly on [0, 1]. Replacing u ω with v ω, we can write the sales as X [ ω = ε n ω 1 v ω ] λ/ θ v ω 1/ θ κ 2 X, κ 1 which is the same as the equation 31 in EKK

27 Appendix B: Simulated Method of Moments In this section, we describe the simulation algorithm, moments constructed and estimation procedure in detail. Simulation Algorithm We denote an artificial Chinese exporter by s and the number of such exporters by S. rior to runng any simulations, i we draw S realizations of v s independently from the uform distribution U [0, 1], putting them aside to construct standardized ut cost below, and ii we draw two sets of S N realizations of a n s and h n s independently from N 0, 1, putting them aside to construct αn O s and ηn O s, and αn s and ηn s below. A given simulation of the model requires a set of parameters Θ, data for each destination n on average domestic sales or ordinary exports X O by Chinese ordinary exporters, average processing exports X by Chinese processing exporters, and the numbers N O and N of Chinese firms selling there via domestic sales or ordinary exports and processing exports, respectively. It involves ne steps: Step 1 Calculate κ 1 and κ 2. Step 2 Calculate σet M 1 t for each destination according to σe 1 t = κ 2 κ 1 X. Step 3 We use the a n each of ln α n [ s s and h n s and ln η n ln αn ln ηn where = DO,. s s ] = s as [ s s to construct S N realizations for σ α 1 ρ 2 σ α ρ 0 σ η ] [ a T n M h T n M Step 4 We construct the S N entry hurdles for domestic sales and ordinary exports u O n s = N O κ 2 η O n s θ, s s ], 27

28 and S N entry hurdles for processing exports Step 5 We calculate u n s = N κ 2 η n s θ. { u s = max u O n s, u n s }, n the maximum u consistent with selling somewhere through domestic sales, ordinary exports or processing exports. Step 6 To simulate ordinary and processing exporters that sell in China, u s should be a realization from the uform distribution over the interval [0, u s]. Therefore, we construct u s = v s u s. Step 7 In the model, a measure of u firms have standardized ut cost below u. Our artificial Chinese exporter s therefore gets an importance weight u s. This importance weight is used to construct statistics on artificial Chinese exporters that relate to statistics on actual Chinese exporters. Step 8 We calculate δ O s, which indicates whether artificial exporters s enters market n through domestic sales or ordinary exports, as determined by the entry hurdles δ O s = { 1 if u s u O n s, 0, otherwise. Wherever δ O s = 1, we calculate sales as [ λ/ θ] 1/ θ X O s = αo n s u s u s σe O 1 ηn O s u O n s u O n s 1 t, Step 9 We calculate δ s, which indicates whether artificial exporters s enters market n through domestic sales or ordinary exports, as determined by the entry hurdles { δ 1 if u s u n s, s = 0, otherwise. 28

29 Wherever δ s = 1, we calculate sales as [ λ/ θ] 1/ θ X s = α n s u s u s σe 1 ηn s u n s u n s 1 t, Moments For a candidate value Θ, we use the algorithm above to simulate the sales of S artificial Chinese exporting firms in N markets. From these artificial data, we compute a vector of moments m Θ analogous to particular moments m in the actual data. Here we choose six sets of moments. The first four sets of moments are related to ordinary exports: We compute the proportion m k 1; Θ of simulated ordinary exporters selling to each possible combination k of the seven most popular ordinary export destinations. One possibility is exporting yet selling none of the top seven via ordinary exports, giving us 2 7 possible combinations. The corresponding moments from the actual data are simply the proportion m k 1 of exporters selling to combination k. Stacking these proportions gives us m 1; Θ and m 1 with 128 elements subject to 1 adding up constraint. For firms selling in each N 1 export destinations n, we compute the qth percentile of ordinary export sales s q n 2 in that market for q = 50, 75, 95, from actual data. Using these s q n 2, we assign firms that sell in n into four mutually exclusive and exhaustive bins determined by these three ordinary export sales levels. We compute the proportion m n 2; Θ of artificial firms falling into each bin analogous to the actual proportion m n 2 = 0.5, 0.25, 0.2, Stacking across N 1 countries gives us the proportion m 2; Θ and m 2, each with 4 N 1 elements subject to N 1 adding up constraints. For firms selling via ordinary exports in each N 1 destinations n, we compute the qth percentile of sales s q n 3 in Chinese excluding firm with no sales in China for q = 50, 75, 95, from actual data. roceeding as above, we get m 3; Θ and m 3, each with 4 N 1 elements subject to N 1 adding up constraints. 29

30 For firms selling via ordinary exports in each 112 destinations n, we compute the qth percentile ratio of ordinary export sales s q n 4 in n to sales in China excluding firms with no sales in China for q = 50, 75, from actual data. roceeding as above, we get m 4; Θ and m 4, each with 3 N 1 elements subject to N 1 adding up constraints. The last two sets of moments are related to processing exports: We compute the proportion m k 5; Θ of simulated processing exporters selling to each possible combination k of the seven most popular processing export destinations. One possibility is exporting yet selling none of the top seven via processing exports, giving us 2 7 possible combinations. The corresponding moments from the actual data are simply the proportion m k 5 of exporters selling to combination k. Stacking these proportions gives us m 5; Θ and m 5 with 128 elements subject to 1 adding up constraint. For firms selling in each N 1 export destinations n, we compute the qth percentile of processing export sales s q n 6 in that market for q = 50, 75, 95, from actual data. Using these s q n 6, we assign firms that sell in n into four mutually exclusive and exhaustive bins determined by these three processing export sales levels. We compute the proportion m n 6; Θ of artificial firms falling into each bin analogous to the actual proportion m n 6 = 0.5, 0.25, 0.2, Stacking across N 1 countries gives us the proportion m 6; Θ and m 6, each with 4 N 1 elements subject to N 1 adding up constraints. Stacking the six sets of moments gives us a 15N element vector of deviations between the moments of the actual and artificial data: m 1 m 1; Θ m 2 m 2; Θ m 3 m 3; Θ y Θ =. m 4 m 4; Θ m 5 m 5; Θ m 6 m 6; Θ 30

31 Estimation rocedure We base our estimation procedure on the moment condition E [y Θ 0 ] = 0, where Θ 0 is the true value of Θ. We thus seek a Θ that achieves where W is the weighting matrix. [ Θ = arg min y Θ W y Θ ], Θ 31

32 Appendix C: Closing the Model In this part, we will write down the market clearing conditions in detail. The composite intermediate goods produced in country n, used as inputs to produce domestic sales and ordinary exports, is equal to the final consumption of composite intermediate goods and inputs demanded by firms in country n to produce domestic sales and ordinary exports. The equation is given by Q O n = Q O nf + m O n ω dω, where Q O n is the total quantity demanded in country n, and Q O nf is the quantity demanded for final consumption. Since the share of expenditure on composite intermediate goods is fixed for firms in country n, m O n ω = N 1 β i=1 mζ in q n O in ω p in ω, and we have Q O n = Q O nf + N 1 β q mζ in n O in ω p in ω dω, i=1 Since q in ω p in ω dω = π O,O in i OQO i τ in + I i π,o in i Q i, we have Q O n = Q O nf + N i=1 π O,O in O i Q O i τ in + I n π,o in i Q i, that is, X O n = γi n + N i=1 1 β mζ in π O,O in X DO i τ in + I i π,o in X i The composite intermediate goods produced in country n, used as inputs to produce processing exports, is equal to the inputs demanded by firms in country n to produce processing exports. The equation is given by Q n = m n ω dω.. Since the share of expenditure on composite intermediate goods is fixed in country n, m n ω = 1 β i n q inωp in ω mζ in, and we have n 32

33 We can rewrite the equation as 1 β Q i n n = q in ω p in ω dω. mζ in n Q n = i n 1 β mζ in n q in ω p in ω dω. Since q in ω p in ω dω = π O, in X n = i n X O i τ in 1 β mζ in + I i π, in X i, we have π O, Xi O in τ in + I i π, in X i. 33

34 Appendix D: Data Source There are two data sources used in this paper to construct the Chinese firmlevel data in year The first is Annual Survey of Chinese Manufacturing ASCM, which contains the balance sheets information for manufacturung firms in China with sales more than 5 million Chinese Yuan about 0.6 million dollars. The second is the Chinese Custom Records CCR which covers the uversal transactions, containg information, such as firm name, 8-digit HS code, trade mode, import or export, value, quantity, destination, etc. We aggregate the custom data by firm, trade mode, import or export and destination, then merge it with the data of ASCM. Data for counterfactual exercise includes: bilateral trade flows data comes from UN COMTRADE database, tariff data is from UNCTAD-TRAINS database. GD, manufacturing trade deficit and total trade deficit are from the World Bank World Development Index database. Manufacturing output and value-added are obtained from UNIDO Industrial Statistical Database. 34