Trade-induced productivity gains reduce incentives to impose strategic tariffs

Economic Modelling (xxxx) xxxx–xxxx

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/econmod

Trade-induced productivity gains reduce incentives to impose strategic tariffs ⁎

Michael Hüblera,b, , Frank Pothena,b a b

Leibniz Universität Hannover, Königsworther Platz 1, 30167 Hannover, Germany Centre for European Economic Research (ZEW), Mannheim, Germany

A R T I C L E I N F O

A BS T RAC T

JEL classification: F12 F14 F17 O33 O40

Strategic tariffs, which raise an economy's welfare by restricting trade and improving the terms of trade, can create an obstacle to free trade. We evaluate how far trade-induced productivity gains (technology spillovers) reduce or remove this obstacle, because more intensive trade enhances these potential gains. Based on theory and the World Input-Output Database (WIOD) we estimate stronger import-induced than export-induced productivity gains. We feed the theory and the estimates into a global Computable General Equilibrium (CGE) model calibrated to WIOD. We find that the USA's, China's and the EU's optimal tariffs are reduced by less than 20%, Russia's and India's by around 25% and Brazil's by 40% when taking endogenous trade-induced productivity gains into account. Nonetheless, incentives for single economies to impose strategic tariffs persist. Particularly large, trade-intensive downstream sectors producing distinct goods incentivize high sectoral optimal tariffs. A global free trade agreement could overcome such incentives and maximize the trade-induced productivity gains.

Keywords: Trade policy Optimal tariff Technology spillovers Emerging economies CGE

1. Introduction Classic trade theory based on the Ricardian model of comparative advantages argues that economies benefit from international trade via specialization. Given that technology-related knowledge and ideas are embodied in traded goods, trade additionally fosters technology imitation and technology spillovers across firms (Grossman and Helpman, 1991). Furthermore, trade implies enhanced competition and on average more efficient production via the self-selection of heterogeneous firms (Melitz, 2003). Additionally, learning by exporting can enhance exporters’ productivity (Yang and Mallick, 2014). Via these channels, international trade is supposed to create a positive externality in the form of productivity and welfare gains on top of classic gains from trade. Significant welfare gains in turn create an incentive to reduce trade barriers in order to enhance trade and the resulting productivity gains in the home economy. To shed light on this issue, this paper studies how optimal tariffs (welfare-maximizing for the home economy), exogenously imposed by policymakers in the home economy, and the resulting welfare gains are affected by endogenous trade-induced productivity gains, anticipated by the policymakers. Unlike recent implementations of Melitz's theory of heterogeneous firms into numerical multi-region, multi-sector models (Balistreri et al., 2011; Akgul et al., 2016; Dixon et al., 2016) we model

⁎

productivity gains in a general stylized way following the broad econometric literature; i.e., we implicitly allow for all the abovementioned channels of productivity gains. The insights from this study are of particular relevance for the current trade policy controversy: on the one hand the theoretical argument for global free trade, on the other hand the creation of regional trade agreements, such as the trans-Atlantic and trans-Pacific agreements, for political reasons (The Economist, 2015). Referring to this policy controversy, we ask the question: are trade-induced productivity gains a valid theoretical and empirical argument for the reduction or removal of import tariffs? To answer this question, the paper first provides conceptual explanations building on classic trade theory (especially Markusen, 1975) and relating to the optimal tariff literature (Johnson, 1954; Hamilton and Whalley, 1983; Gros, 1987; Kennan and Riezman, 1988, 1990; Brown, 1987; Markusen and Wigle, 1989; Broda et al., 2008). The intuition for imposing optimal tariffs is as follows: if an economy exhibits power on international markets, it can increase domestic welfare by erecting trade barriers and thereby manipulating the terms of trade in its favor, similar to the behavior of a monopolist. The optimal tariff imposed on imports balances the marginal benefits from the improved terms of trade and the marginal losses from the resulting trade distortion. Surprisingly, it has hardly been investigated by this

Corresponding author at: Leibniz Universität Hannover, Königsworther Platz 1, 30167 Hannover, Germany. E-mail address: [email protected] (M. Hübler).

http://dx.doi.org/10.1016/j.econmod.2016.11.006 Received 9 June 2016; Received in revised form 29 August 2016; Accepted 9 November 2016 0264-9993/ © 2016 Elsevier B.V. All rights reserved.

Please cite this article as: Huebler, M., Economic Modelling (2016), http://dx.doi.org/10.1016/j.econmod.2016.11.006

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

concludes with policy implications.

literature how optimal strategic trade intervention is altered in the presence of trade-induced productivity gains, in particular technology spillovers. The following paper shows that import- and export-induced productivity gains reduce optimal tariffs and the resulting welfare gains. If the strength of productivity gains is sufficiently high, the incentive to impose tariffs will vanish because the achievable productivity gains via free trade exceed the strategic gains from restricting trade. The question is whether this theoretical finding holds empirically. Therefore, this paper then estimates trade-induced productivity gains econometrically. The econometric model is specified in a parsimonious way to keep it consistent with theory and the following CGE (Computable General Equilibrium) model implementation. The econometric estimation also uses the same recent global dataset called WIOD (World Input-Output Database; Dietzenbacher et al., 2013; Timmer et al., 2015; e.g. applied by Koopman et al., 2014)1 as the CGE model. The main advantage of using WIOD for this analysis of dynamic productivity gains is the availability of the global input-output matrix for 15 years. This allows us to carry out panel data estimations and to vary the benchmark year in the CGE analysis. Elasticities of substitution for the production functions, estimated with WIOD as well, are taken from Koesler and Schymura (2015). In this way, the paper reconciles theory-based estimation and numerical modelling consistently using the same dataset, which is often not possible in economic modelling due to lack of suitable data and technical constraints. The econometric results indicate, in accordance with the literature (summarized by Saggi, 2002; Keller, 2004), that statistically and economically significant import- as well as export-related productivity gains are measurable. Import-induced productivity gains appear to be larger than export-induced ones. For the numerical policy analysis, this paper utilizes a global multisector, multi-region CGE model, which is a prime implementation of the model tailored for the WIOD (Koesler and Pothen, 2013). This model enables the assessment of welfare effects of import tariffs, exogenously imposed on the model regions or specific sectors under the assumption of endogenous or fixed trade-induced productivity gains. Other multi-region studies on strategic tariffs, such as Ossa (2014), utilize GTAP (Global Trade Analysis Project)2 and do not take trade-induced productivity gains into account. As usually done in the literature, we utilize an Armington (1969) specification. After feeding the estimated productivity gains into the CGE model, we obtain optimal tariffs that range from 10 percent for Brazil to about 25 percent for the USA.3 Based on these simulation results, we show econometrically that optimal tariffs increase in the import and export intensity, in the relative sector size4 and the downstreamness5 of a sector. When taking the endogeneity of trade-induced productivity gains into account, the CGE analysis suggests that optimal tariffs should be reduced by less than 20 percent in the USA, China and the EU,6 around 25 percent in Russia and India and 40 percent in Brazil. Lower Armington elasticities representing more distinct goods drastically raise these tariffs. Overall, the results provide an argument for the reduction of trade barriers at the global level, which has been quantitatively opaque and is often overlooked in the policy debate and the trade literature. The paper proceeds as follows. Section 2 explains the theoretical background and formulates testable hypotheses (propositions). Section 3 estimates the magnitude of trade-induced productivity gains. Section 4 applies the estimated parameter values to a CGE model and performs various region- and sector-specific trade policy experiments. Section 5

2. Theoretical background This section explains the theoretical background as a basis for the following numerical analyses. We refer to the Markusen (1975) general equilibrium 2x2 trade model and the modified version by Jakob et al. (2013) who both describe trade policy in the presence of a negative trans-boundary, environmental externality created by foreign (and domestic) producers. Different to them, we assume that a positive productivity externality of trade occurs within the home country induced by imports as well as exports, albeit we do not look at productivity gains abroad induced by the home country's exports or imports. The full-fledged general equilibrium model in Section 4 will allow for simultaneous trade-induced productivity gains in all model regions. The assumption of a positive productivity externality induced by international trade builds on a broad literature (summarized by Saggi, 2002; Keller, 2004). The classic Ricardian gains from trade occur via comparative advantages and specialization (advanced by Eaton and Kortum, 2002). In addition, trade can create productivity gains by increasing the availability of differentiated, innovative intermediate goods and, thereby, improving productivity in final good production (Ethier, 1982; Hübler, 2015).7 In this context, imports are seen as a main source of international technology spillovers. They embody advanced knowledge that can be exploited via imitation, and they are often associated with international enterprises that exchange knowledge between their affiliates, which creates international technology spillovers. Knowledge can then spread further from foreign affiliates to local firms and create local technology spillovers. Like importing, exporting can spur innovation and enhance knowledge exchange between firms (learning by exporting). Furthermore, productivity gains can emerge due to increased competition and firm selection triggered by trade as described by Melitz (2003)8 and the vast literature based on this seminal work. Felbermayr et al. (2013) analyze strategic trade policy in a Melitz model. In their model, the optimal tariff addresses a mark-up distortion, an entry distortion and a terms of trade externality. Our approach is more general than other approaches in the literature by looking at any export- as well as import-related productivity gains including technology spillovers. The following framework illustrates the core economic mechanisms and formulates testable hypotheses (propositions) for the econometric and numeric analyses. 2.1. A stylized illustration This subsection illustrates a stylized mathematical representation of international trade with induced productivity gains that uses standard elements and specifications. We imagine a large open economy, called Home, producing two varieties, i = {X ; Y}, of one tradable good in one sector, s, given a constant total factor endowment Z . One variety is produced at home and exported, the other variety is imported from abroad and can also be produced at home. We do not look at the rest of the world and its strategic behavior or reaction explicitly and restrict the analysis to Home's unilateral trade policy. We neglect externalities from Home to the rest of the world and their impacts on production and trade patterns abroad, because we focus on policy-induced 7 Markusen et al. (2005) and Rutherford and Tarr (2008) implement Ethier (1982)'s mechanism in a CGE framework. Our strategic trade policy analysis abstracts from quality-differentiated goods and reputation spillovers (Das and DeLoach, 2003) as well as from intersectoral knowledge spillovers (Murat and Pigliaru, 1998). 8 In the Melitz model of heterogeneous firms, trade liberalization induces the exit of low-productivity firms and the expansion of the profits and the market share of highproductivity exporting firms. This reallocation of firms increases average productivity and welfare.

1

URL http://www.wiod.org/home (accessed Aug. 2016). URL https://www.gtap.agecon.purdue.edu (accessed Aug. 2016). 3 The United States of America. 4 Economy-wide sectoral output share. 5 Sectoral output share in final consumption. 6 European Union. 2

2

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

gains. We assume that productivity gains are sector-specific so that a higher export intensity in the X sector expands X production, given the sectoral factor input ZX. γ E governs the strength of export-induced productivity gains, which are supposed to capture Melitz-type firm selection effects, productivity gains from competition on export markets and technology spillovers through contacts with trading partners. Likewise, a higher import intensity expands Y production, given the sectoral factor input ZY. γ M governs the strength of import-induced productivity gains, which are supposed to capture technology spillovers from knowledge embedded in imported goods and productivity gains from competition within import-competing markets. In this typical 2x2 trade setting with homogeneous products, each variety, i, is either (net) exported (X) or (net) imported (Y), and there is only one sector, s. This simplification will be relaxed in the econometric estimation and in the numerical model calibration in order to fit theory to empirical data with a number of traded goods and bidirectional trade. Yet, the insights obtained from this stylized illustration do not depend upon the number of goods under consideration.

productivity spillovers in the Home economy. We further assume that goods X and Y are produced by one representative firm per sector. Each representative firm characterizes the behavior of a large number of atomistic firms. Neither atomistic firms nor representative firms can influence prices, i.e. firms do not have power on domestic markets.9 Hence, in the absence of trade policy, firms cannot exploit Home's power on international markets. We define p0 =

pY

pX

as the domestic

price for good Y relative to good X and choose X as the numeraire with pX=1. Home's production pattern can be characterized by a concave, decreasing production possibility frontier: Q X = F (QY ), F QY <0, F QY QY < 0 . Q denotes produced quantities. Quantities are measured in constant currency values throughout the paper. A lower index represents a derivative with respect to the corresponding variable throughout the paper. F determines the quantity of X that can be generated when producing a certain quantity of Y by allocating the given factor endowment, Z , to the two sectors with perfect intersectoral factor mobility and Z X + ZY = Z . The output is absorbed by the representative consumer or exported. Home's consumption pattern can be characterized by a concave utility function, increasing in the consumption of X and Y, U (C X , CY ), UC X > 0, UCY > 0, with homothetic preferences: UC XC X < 0, UCY CY < 0 . C X and CY denote demand in the form of consumption values.10 Home's trade pattern can be described as follows. Without loss of generality, we assume that Home is a net exporter of X and a net importer of Y. We assume a balanced trade budget closure so that the following condition holds: E X = p* MY . E symbolizes exports, whereas pY * M symbolizes imports. International prices are expressed as p* = X . p

2.2. The argument for reduced optimal tariffs This subsection formally derives the argument for tariff reductions in the presence of trade-induced productivity gains. It formulates and proves several propositions, which will serve as testable hypotheses for the numerical analyses in the subsequent sections. To derive Home's welfare-optimal solution (which deviates from the globally optimal solution), we need to maximize the utility of Home's representative consumer, U (C X , CY ), given the relations detailed in the previous subsection and taking into account trade-induced productivity gains. From this solution, we will derive an optimal tariff that Home's policymaker exogenously imposes in order to internalize the endogenous trade-induced productivity gains in such a way that Home's welfare is maximized. Whereas the policymaker anticipates the impact of the tariff on trade flows and hence induced productivity gains, the ∂U ∂U atomistic producers do not anticipate it. Solving Y = Y = 0 yields ∂Q ∂M the following optimality condition (for details see Appendix):

*

In general p* differs from the domestic price ratio p0. Home's terms of trade improve when p* declines. The following expressions characterize the influence of Home's exports and imports and international prices: p *Y > 0, p *X > 0 . Higher imports of Y in Home raise the world market M E demand for Y and hence the relative price for Y, signified by p*. Conversely, higher exports from Home raise the world market supply of X, reduce the world market price for X, and hence raise p*. Following the empirical literature on trade-induced productivity gains (see above and Section 3), we assume that the sectoral intensities Y X of exports, E X , and imports, MY , augment (total factor) productivity Q

1 − γE p0 = ( p* + MY ·p *Y )· M M 1 + γ

   θ strat

Q

l i , can be proportionately so that the new output levels of each variety, Q expressed as: X⎞ Y⎞ ⎛ ⎛ l X = ⎜1 + γ 0 + γ E E ⎟ Q X ; Q lY = ⎜1 + γ 0 + γ M M ⎟ QY Q X Y ⎝ ⎝ Q ⎠ Q ⎠

where

p0

=

UCY UCX

θ prod

(2)

= −F QY , i.e. the domestic relative price for the import

good Y equals the consumer's marginal rate of substitution (the ratio of marginal utilities), and it equals the producers' negative marginal rate of transformation (the slope of the production possibility frontier). Hence, the domestic consumer price ratio equals the domestic producer price ratio. On the contrary, as illustrated by the above equation, there is in general a wedge between domestic prices, p0, and international prices, p* in Home's optimum. Home's optimal tariff creates exactly this wedge between domestic and international prices,12 which is welfare optimal from Home's point of view (but in general not from a global point of view due to the beggar-thy-neighbor character of strategic trade policy). Hence, there is an incentive for single countries to make use of strategic trade policy. The price wedge can be decomposed into two parts. θ strat represents the strategic term that we know from the literature cited in the Introduction: by imposing a tariff at the rate θ strat = MY ·p M*Y , Home optimally exploits its power on international markets.13 A higher term p *Y implies a stronger reaction of international prices to changes in

(1)

Qi signifies the benchmark output levels without trade policy intervention, which are taken into account by the producers. The intensity form is used in order to make the trade-induced productivity gains independent of the sector and economy size. γ 0 captures exogenous growth raising productivity equally for both sectors, X and Y.11 This corresponds to a proportional outward shift of the production possibility frontier by the factor γ 0 without sector bias. The focus of our analysis, however, is on trade-induced productivity gains. We assume that tradeinduced productivity gains add to exogenous productivity gains and are strictly separable from them. Notably, the atomistic producers in each sector cannot anticipate or internalize the trade-induced productivity gains. Their choice of X and Y production volumes is – in the absence of policy intervention – not affected by trade-induced productivity

M

Allowing for firm-level product differentiation (Balistreri and Markusen, 2009) or other forms of market power will reduce the optimal tariff quantitatively. 10 For simplicity, let C X and CY be connected in a multiplicative way, U X (C X )·UY (CY ), e.g. in form of a Cobb-Douglas function (C X )α ·(CY ) β with α + β = 1. 11 Exogenous productivity growth is irrelevant for this stylized illustration. Nonetheless, it will be required in the numerical analyses in the subsequent sections. 9

12 Although exports and imports are linked via the balanced budget condition, export and import volumes can be driven apart by tariffs with the help of a wedge between the prices for export and import goods. 13 The import-dependency of the international price creates a term that is typical for a maximization problem with monopoly power, in this case MY ·p* (MY ) .

3

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

volumes. The productivity-related term, on the opposite, expands import and export volumes in order to exploit induced productivity gains. The potential for expanding the externality in absolute terms increases in power on international markets. This can easily be seen in Eq. (2). The productivity-related term (θ strat ) reduces any given price wedge in relative terms, i.e. by a factor θ prod < 1. In absolute terms, the effect depends upon the magnitude of p* + MY p M*Y . Since p *Y rises in M Home's market power (i.e. with lower Armington elasticities), θ prod 's absolute effect also rises in Home's market power. The intuition is that with higher market power, Home has a higher potential for boosting trade by manipulating international prices so that foreign producers intend to enhance trade with Home. □

Home's imports (and exports given the balanced trade budget assumption). As a consequence, Home's optimal tariff rises in order to exploit the market power reflected by a larger term p M*Y . In an Armington specification of international trade as used in Section 4, a lower elasticity of substitution between traded goods (a lower Armington elasticity) reduces substitution possibilities across goods in the sense that traded goods are more heterogeneous. Therefore, a lower Armington elasticity translates into a larger term p *Y . In the absence M of market power, it is p M*Y = 0 so that θ strat collapses to p*. θ prod represents the term associated with trade-induced productivity gains, which is the novel aspect that we will discuss in more detail in the following paragraphs. In the absence of trade-induced productivity gains, it is γ E = γ M = 0 so that θ prod collapses to one. Let us call this scenario ExoSpill for exogenous or not trade-induced productivity gains. On the contrary, if trade-induced productivity gains are endogenous, then 1 > γ E > 0 ,14 γ M > 0 and 0 < θ prod < 1. Let us call this scenario EndoSpill. These scenarios will be used in the simulations in Section 4.2. Only in the absence of market power and the absence of (endogenous) trade-induced productivity gains, optimal domestic prices equal international prices, i.e. p0 = p*, and there is no incentive to make use of strategic trade policy intervention.

Proposition 3. For every world market price, there exists a certain strength of productivity gains through imports and exports such that the incentive to manipulate the terms of trade vanishes. Proof. Setting p0 = p* in (2) yields

1 + γM 1 − γE

−1=

MY ·p * Y M

p*

. If this

condition is fulfilled, there will be no difference between the original world market price p* and the one manipulated via Home's optimal tariff. The incentive to use beggar-thy-neighbor policies is perfectly offset by the incentive to internalize the productivity spillovers. More market power and lower international prices require stronger tradeinduced productivity gains to fulfill this condition. Although this outcome is straightforward in theory, it is an open question whether it holds empirically. Section 4.2.1 will address this question. □

Proposition 1. In the presence of trade-induced productivity gains, there is an incentive to expand trade even without market power on international markets. The potential for expanding trade with the aim to exploit trade-induced productivity gains increases in international market power.

Proposition 4. The welfare gain achievable for a large open economy via a given tariff rate is lower in the presence of productivity gains through imports and exports than in their absence.

Proof. Consider Eq. (2) for a small open economy. Without power on international markets, p *Y is zero. Hence, the possibility to manipulate M the terms of trade (θ strat ) vanishes. The incentive to internalize the productivity effect of trade is nevertheless present, represented by the last term (θ prod ). Home attempts to export and import more in order to exploit the trade-induced productivity gains that occur within its boundaries. If international prices stay constant and cannot be influenced by Home, Home can nevertheless influence domestic prices relative to the given international prices. This mechanism differs from the Markusen (1975) model, in which the environmental externality occurs abroad and Home requires market power to mitigate the environmental externality in the foreign country by influencing international prices. In our model, on the contrary, the externality occurs within the home country so that power on international markets is not essential. This result also differs from Brown's (1987) model, in which no externality is taken into account so that the terms of trade effect will vanish, when traded commodities become perfectly substitutable, i.e. when market power disappears (see our numerical analysis in Section 4.2.4). □

Proof. Whereas it is straightforward that trade-induced technology spillovers create and incentive to enhance trade and thus to reduce tariffs, it is an open question, how they affect welfare. More market power expressed by a higher p *Y or a lower Armington elasticity M creates a stronger impact of Home's imports on international prices and therefore magnifies the potential for welfare gains through the manipulation of international prices. A reduction in imports, MY, reduces consumption, CY , which is detrimental for Home, and simultaneously reduces exports valued by international prices, MY ·p* (MY ), which raises consumption, C X , which is beneficial. The more potent Home's market power is, the stronger the latter beneficial effect is. As a consequence, the welfare gain that can be achieved by compressing imports is higher under more market power. It is obvious in Eq. (2) that the productivity gain factor, θ prod , reduces p *Y and hence M the effective market power and thus counteracts the use of strategic tariffs. This in turn attenuates the welfare gain generated by a tariff (the optimal tariff or any other tariff). □ The following Section 3 finds empirical estimates for the tradeinduced productivity gains underlying the propositions. The estimates are required for the policy analysis in Section 4, which validates the propositions in a general equilibrium model.

Proposition 2. Productivity gains through imports and exports reduce the optimal tariff that manipulates the terms of trade in favor of a large open economy. Proof. In Eq. (2), θ strat attenuates the price for Y imports and elevates the price for X exports relative to each other. This improves the terms of trade in Home's favor but hampers trade in absolute volumes. Stronger productivity gains via exports, expressed by γ E , and stronger productivity gains via imports, expressed by γ M , both counteractthe effect of θ strat . This converse effect of productivity gains from trade on the terms of trade is summarized by the productivity-related term, θ prod . θ prod < 1 has the form of an ad-valorem subsidy that multiplies the world market price plus the strategic tariff by a factor smaller than one. The intuition is straightforward: the strategic term improves the international price in Home's favor, but diminishes import and export

3. Estimation of productivity gains This section estimates the coefficients governing the strength of import- and export-induced productivity gains. Our econometric analysis builds upon a broad literature on international productivity (technology) spillovers as summarized by Saggi (2002); Keller (2004) and Havranek and Irsova (2011). Coe and Helpman (1995) and Coe et al., (1997) are seminal papers on North-South productivity spillovers. We follow the common approach of using TFP (total factor productivity) growth as the dependent variable (Jalles and Tavares, 2015). Our search for export-induced productivity gains also follows the literature seeking for productivity gains from firm selection and learning by exporting. Girma et al. (2004), for example, find that

14 1 < γ E describes an odd situation, in which export-induced productivity gains are so extensive that Home is willing to pay customers to take more of the export good.

4

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

⎛ Qrs (t +1) ⎞ M rst E rst dlog ⎜ rs (t +1) ⎟ = γ 0r + γ M rst + γ E rst + εrst ⎝Z ⎠ Q Q

manufacturing firms in the United Kingdom which export are initially more productive than other firms and additionally become more productive through exporting. Yang and Mallick (2010) find similar results for Chinese firms and Mallick and Yang (2013) for Indian firms. Although the results of these literature strands are diverse and sometimes ambiguous, the bottom-line is that trade-induced (and more significantly foreign direct investment-induced) international productivity gains can be detected. Different to most studies, we contrast import-induced with export-induced productivity gains within one estimation. Since a broad literature exists, it is not our primary goal to find robust evidence for international productivity spillovers. Our econometric analysis is mainly an intermediate step that illustrates the theoretical background detailed in Section 2 and provides parameter values for the numerical policy analysis in Section 4. We run a standard fixed-effects panel data regression as well as a GMM (Generalized Method of Moments) robustness check regression with lagged variables used as instruments. We abstain from including control variables besides fixed-effects, because we want make the estimations as consistent as possible with the CGE implementation in terms of the economic model and the data source. The following subsections derive the econometric model and interpret the estimation results.

(4)

We estimate the parameters γ by using the novel World Input Output Database (WIOD) panel data for 40 countries,15 34 sectors16 and the years 1995 to 2009. It is to our knowledge the first database providing bilateral and bisectoral input-output relations for a sequence of years within one consistent dataset. Monetary quantities are measured in 1995 US-$. The TFP values and the import and export intensities are computed with WIOD. Inputs of labor and energy are also taken from WIOD and measured in physical units (million hours worked, Terajoule). Elasticities of substitution used for computing input values based on the production function, Eq. (6), are taken from Koesler and Schymura (2015). They estimate the elasticities with WIOD based on a non-linear econometric model. In this way, we utilize consistent data and parameter values throughout the econometric and numerical modelling analyses. One caveat regarding WIOD is that the region Rest of the World (ROW) represents a residual region which incorporates remaining trade flows not attributed to other countries or regions. Hence, the transfer of results for ROW to the real world is difficult. 3.2. Econometric results

3.1. Econometric approach

The estimated slope coefficients are reported in Table 1 and the fixed-effects in Table 2. We always report heteroscedasticity-robust standard errors clustered at the country-level. The regressions include 40 country-specific fixed-effects. Anticipating the regional structure of the modelling exercise in the following section, we summarize the 40 country-specific results in form of eight region-specific results. To this end, we aggregate the country-specific growth rates by computing GDP-weighted averages.17 Table 2 displays the eight model regions and the corresponding estimated aggregate annual TFP growth rates. The estimates range from 0.5 per cent for the EU to 3.0 per cent for China. We find three main results:

This subsection derives the econometric model referring to the theoretical background exposed in the previous Section 2. Eq. (1) implicitly assumes that output expands while all inputs stays constant. Now we let the sector-specific input, Zs, which captures all inputs of production factors as well as intermediate goods, enter the equation explicitly, because it is not kept constant anymore. Furthermore, we generalize the model to a number of goods or sectors, s. Each good s appears as one imported variety and one domestically produced and exported variety, i. This requires varieties of each good produced in different countries, r. Imports are an aggregate of intermediate goods from all available countries. Exports, on the contrary, encompass only the specific good that a sector produces and are in general exported to all countries available in the dataset. In each sector, imports and exports of a variety, i, create sector-specific productivity gains. In addition, we introduce time, t, measured in years. This results in a panel specification. It allows us to exploit the time dimension, which is crucial for the estimation of productivity gains. With these modifications, the combination of the two equations labeled Eq. (1) results in the following generalized specification, defined over countries, r, sectors, s, and time, t:

⎛ Qrs (t +1) M rst E rst ⎞ Qrst = ⎜1 + γ 0r + γ M rst + γ E rst ⎟ rst ⎝ Z rs (t +1) Q Q ⎠Z

Result 1. The WIOD-based estimates indicate import- and exportinduced productivity gains (technology spillovers). Both, the coefficients of import intensity and export intensity, are statistically significant and positive. Accordingly, importing and exporting raise TFP.18 The results are qualitatively in accordance with the econometric literature summarized above. The magnitude of the 15 Australia (ROW), Austria (EUR), Belgium (EUR), Canada (ROW), Czech Republic (EUR), Denmark (EUR), Estonia (EUR), Finland (EUR), France (EUR), Germany (EUR), Greece (EUR), Hungary (EUR), Ireland (EUR), Italy (EUR), Japan (EAS), Luxembourg (EUR), Mexico (ROW), Netherlands (EUR), Poland (EUR), Portugal, Slovak Republic (EUR), South Korea (EAS), Spain (EUR), Sweden (EUR), Turkey (ROW), United Kingdom (EUR), United States of America (USA), Bulgaria (EUR), Brazil (BRA), China (CHN), Cypress (EUR), India (IND), Indonesia (ROW), Latvia (EUR), Lithuania (EUR), Malta (EUR), Romania (EUR), Russia (RUS), Slovenia (EUR), Taiwan (EAS); for the region codes and explanations see Table 2 and Section 4.1. 16 Agriculture, Hunting, Forestry and Fishing; Mining and Quarrying; Food, Beverages and Tobacco; Textiles and Textile Products; Leather, Leather and Footwear; Wood and Products of Wood and Cork; Pulp, Paper, Printing and Publishing; Coke, Refined Petroleum and Nuclear Fuel Chemicals and Chemical Products; Rubber and Plastics; Other Non-Metallic Mineral; Basic Metals and Fabricated Metal; Machinery, Nec; Electrical and Optical Equipment; Transport Equipment; Manufacturing, Nec, Recycling; Electricity, Gas and Water Supply; Construction; Sale, Maintenance and Repair of Motor Vehicles and Motorcycles, Retail Sale of Fuel; Wholesale Trade and Commission Trade, Except of Motor Vehicles and Motorcycles; Retail Trade, Except of Motor Vehicles and Motorcycles; Repair of Household Goods; Hotels and Restaurants; Inland Transport; Water Transport; Air Transport; Other Supporting and Auxiliary Transport Activities; Activities of Travel Agencies; Post and Telecommunications; Financial Intermediation; Real Estate Activities; Renting of M & Eq and Other Business Activities; Public Admin and Defence, Compulsory Social Security; Education; Health and Social Work; Other Community, Social and Personal Services. 17 For the country-region matching see footnote 15. 18 Note that these results are estimated based on observed trade volumes without explicitly taking into account trade policy (tariffs).

(3)

γ 0r

The growth factor is region-specific and to be estimated. It represents country fixed-effects. The trade-related growth factors γ M and γ E are assumed to be identical in all sectors and regions and also to be estimated. The ratio of the output value relative to the input value Qrst is interpreted as TFP (total factor productivity; cf. Jalles and Z rst Tavares, 2015, p. 801). We compute the input value with the help of the production function defined by Eq. (6) in the Appendix. The function depicts the constant elasticity of substitution (CES) nesting structure that will be used in the numerical model. The output value is given by rst rst the data. Mrst and E rst are the import and export intensities. Setting Q

Q

Z rs (t +1) = Z rst in Eq. (3) and multiplying by Zrst on both sides leads back to Eq. (1). This means, the above equation describes total factor rs (t +1) productivity growth. The growth rate of Qrs (t +1) can be rewritten in dlog Z form (differences of natural logarithms over time). Adding an error term εrst, which captures deviations not explained by the model, yields: 5

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

and exports).

Table 1 Panel estimation for 40 countries, 34 sectors and 15 years including country-specific fixed-effects (robust standard errors in parentheses).

Import intensity

Mrst Qrst

γ M = 0.0359 ***

(0.0137)

Result 3. When taking endogenous trade-induced productivity gains into account, the strategically optimal international price ratio declines by about five per cent. According to (2), the strategically optimal international price ratio, which policymakers could exogenously create to restrict trade and hence maximize their home country's welfare, can be computed as

Export intensity

Erst Qrst

γ E = 0.0160 **

(0.0078)

θ prod =

Annual growth rate of total factor productivity ⎛ Qrst +1 ⎞ dlog ⎜ rst +1 ⎟ ⎝Z ⎠

F R2 Number of observ. *** **

19.44 0.0380 15,678

1 − γE 1 + γM

once the magnitudes of the import- and export induced

productivity gains are known. The estimations of γ E and γ M reported in Table 1 yield θ prod ≈ 0.95. Using the delta method, we find that this factor is statistically highly significant. The resulting 95 percent confidence interval is [0.93; 0.97]. This result provides a first rough estimate. A more precise and regionally and sectorally differentiated evaluation, however, requires a complex general equilibrium model, which will be set up in the next section.

(0.0000)

p < 0.01. p < 0.05.

Table 2 Aggregated region-specific fixed-effects taken from the panel estimation.

4. Trade policy analysis

Region-specific exogenous annual growth rate of total factor productivity

γ 0r European Union United States of America Russia Brazil India China East Asia Rest of the World

EUR USA RUS BRA IND CHN EAS ROW

This section represents the main contribution of the paper. It implements the trade-induced productivity gains that have been econometrically studied in the previous Section 3 into the WIOD CGE (World Input Output Database Computable General Equilibrium) model. It then evaluates the propositions formulated in Section 2 in a number of policy experiments. The results provide numerical estimates for the effects of trade policy in the presence of trade-induced productivity gains. Our analysis is related to numerical studies of trade liberalization as critically reviewed by Ackerman and Gallagher (2008). Ackerman and Gallagher highlight the crucial role of Armington (1969) elasticities, which we will also address in our robustness checks (the sectoral benchmark values of the Armington elasticities can be found in Table 6 in the Appendix). Ossa (2014) presents an advanced multi-region analysis based on GTAP 8. Instead to GTAP, we calibrate our model to the novel WIOD20 providing bilateral and bisectoral21 trade data plus production and consumption data for 40 countries and 35 sectors for the years 1995 to 2009. Furthermore, this literature does not take trade-induced productivity gains into account, which is in the spotlight of our modelling exercise. Finicelli et al. (2013) show how the specialization via Ricardian selection (based on Eaton and Kortum, 2002) raises sectoral productivity. They find that when allowing for trade, the average productivity in manufacturing increases by about ten percent compared to autarky due to Recardian selection. Recently, modellers have implemented the Melitz (2003) mechanism of heterogeneous firms and productivity gains from firm selection into CGE models (Balistreri et al., 2011; Balistreri and Rutherford, 2012; Akgul et al., 2016; Dixon et al., 2016). With the Melitz approach, Balistreri et al. (2011) find gains from trade liberalization that are four times larger than with the standard Armington approach. These approaches, however, do not take into account technology spillovers through exporting and importing from a general perspective as our model does. In the domain of development economics, a few model studies include international technology spillovers. Diao et al. (2005), for example, build a general equilibrium model of Thailand, in which trade-related international technology spillovers enhance economic growth. They demonstrate that trade liberalization creates a strong short-term stimulus, but a smaller long-term stimulus. Accordingly,

0.005 0.008 0.014 0.000 0.017 0.030 0.009 0.009

estimates is in accordance with other studies as well (cf. Hübler and Keller, 2009).19 When instrumenting the import and export intensities with their one and two year time lagged values and estimating with GMM, their coefficients slightly decrease to 0.0322 and 0.0154 without losing their significance. Notwithstanding, the econometric results need to be interpreted with caution, taking into account the results of the broad related empirical literature for comparison. Result 2. The strength of trade-related productivity gains is asymmetric: imports entail higher productivity gains than exports. This result is in accordance with the econometric literature. In our results, the coefficient of import intensity is more than twice the coefficient of export intensity. Consequently, fostering imports via trade policy entails larger productivity gains than fostering exports. The estimated import-related, γ M , and export-related, γ E , coefficients can be interpreted in the following way: Taking China as an example, we find a fixed (not trade-induced) growth rate of 0.030 for the Chinese economy reported in Table 2. The import intensity, M , of Q

the Chinese sector say Electrical and Optical Equipment in the year 2009 computed with WIOD is 0.1860. Multiplying γ M taken from Table 1 with M yields about 0.0067 as the import-induced component Q

of the sectoral growth rate. The export intensity,

E , Q

for the same

γE

Chinese sector and year is 0.2258. With the reported in Table 1, we E obtain γ E · Q ≈ 0.0036 as the export-induced component. Summing up the fixed, import- and export-induced components yields about 0.04 for the exemplary overall TFP growth rate of the Chinese sector Electrical and Optical Equipment in 2009. Accordingly, in this Chinese sector three quarters of the growth rate are determined by economic progress of the overall Chinese economy while one quarter depends on the sectoral trade intensity (taking into account imports

20 Unlike GTAP, WIOD does not provide information on policy instruments like tariffs or subsidies. In the following policy analysis, however, we vary tariff rates exogenously and do not require this information. Hence, the analysis is not subject to the tariff aggregation problem either (see Himics and Britz, 2016). 21 Trade originates from a specific country as well as a specific sector and flows in general to another country and another sector.

19

Hübler and Keller (2009) regress energy intensities of 60 developing countries between 1975 and 2004 in dlog form on import intensity. They find a negative, yet insignificant coefficient of −0.017 for import intensity (in regression B1, which is most similar to our estimation). This result comes close to the coefficient of 0.016 for total factor productivity (the inverse of factor intensity) that we find for export intensity.

6

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

output and are generally exported to all other regions. In each sector, imports and exports create sector-specific productivity gains. Sectoral shifts that occur in addition to sector-specific productivity gains are captured by the general equilibrium. For computational reasons and for a better regional focus, we aggregate the WIOD dataset to eight regions, r: EU, USA, China, India, Brazil, Russia, East Asia (without India and China) and Rest of the World (a residual region). In addition, we aggregate the original 34 WIOD sectors to 18 sectors24 denoted by s. We choose 2007 as the benchmark year representing period 1. This means, we calibrate our model to the global WIOD input-output table for the year 2007.25 Period 2 is generated by expanding the production of each region and sector according to the above equation.

our analysis concentrates on short-term effects. In the domain of environmental economics, model-based assessments have scrutinized the influence of international technology spillovers on climate policy costs (e.g. Bosetti et al., 2008; Leimbach and Baumstark, 2010; Hübler, 2011). This literature strand finds a limited influence of international technology spillovers on climate policy costs. 4.1. The general equilibrium model This subsection summarizes the setup of the CGE model introduced by Koesler and Pothen (2013). Details of the mathematical model formulation can be found in the Appendix. The model is formulated as an MCP (Mixed Complementarity Problem) in GAMS (General Algebraic Modelling System). Our model specifies that imports and exports simultaneously create endogenous productivity gains in each model region (a country or a set of countries) and sector. The model assumes perfect competition without market power with one representative firm per sector in each region. Firms require capital and labor as well as energy- and nonenergy intermediate goods as inputs. The model contains one representative consumer in each region. All goods and factor markets clear, and the consumer's income balances the expenditures. International trade is modeled with an Armington specification that combines varieties from different countries to an import bundle and the import bundle with the domestically produced variety. Within this specification, lower elasticities of substitution between varieties imply more market power (see Section 2). We implement Eq. (3) from Section 3 with the estimated parameter values. Since we do not analyze long-term growth paths22 but today's economic situation, we project the dynamic process to two model periods, t = {1; 2}. The first period, denoted by ‘1’, endogenously determines the growth patterns that in turn determine the tradeinduced productivity gains. The exogenous and endogenous tradeinduced productivity gains are realized in the second period, denoted by ‘2’. The resulting time delay between trade and the induced productivity gains is in accordance with empirical evidence as summarized in Section 3.23 Firms do not anticipate productivity gains as in Section 2. Hence, policy intervention is required to exploit tradeinduced productivity gains and power on international markets.

⎛ M rs1 E rs1 ⎞ Ars2 = ⎜1 + γ 0r + γ M rs1 + γ E rs1 ⎟ Ars1 ⎝ Q Q ⎠

4.2. Policy experiments This subsection first explains the results of the trade policy experiments, in which an import tariff is introduced and exogenously varied in one region. Trade flows and induced productivity gains emerge endogenously in the CGE. The subsection begins with studying European tariffs. It then examines how the results vary across different countries, benchmark years, Armington elasticities (and thus different degrees of market power) as well as sectors. It identifies determinants of optimal tariffs and their welfare effects. It ends with a resume and a critical discussion. 4.2.1. European trade policy We first choose the European Union (EUR) in the year 2007 as the region r that makes use of trade policy. Throughout all analyses, trade policy affects both model periods and is not changed between model periods. We focus on the second model period, in which the tradeinduced productivity gains are incorporated in addition to the distortion created by trade policy. We exogenously vary the tariff rate τ imposed on Europe's imports and set it equal for all European sectors. We examine the effect on European welfare and identify the optimal tariff with endogenous versus exogenous trade-induced productivity gains for Europe. We also investigate, how the other model regions are affected by the European tariff. We first solve a benchmark run without trade policy intervention. Then, we impose tariffs at various rates on European imports. In the exogenous spillover scenario, denoted by ExoSpill, productivity gains are fixed at their benchmark run values independent of changes in imports and exports. In the endogenous spillover scenario, denoted by EndoSpill, productivity gains are a function of the import and export intensity following our theoretical and empirical model. Without policy intervention and thus without deviations of the trade pattern, both scenarios generate the same benchmark growth rates between periods one and two. When trade patterns change due to policy intervention, productivity growth will be unaffected in scenario ExoSpill, but will react in scenario EndoSpill. Likewise, Eq. (2) and the propositions formulated in Section 2.2 compare a situation, in which productivity gains depend on imports and exports (EndoSpill) with a situation in which they do not (ExoSpill). Fig. 1 illustrates Europe's relative welfare change referring to the second model period, measured in per mill (per thousand or a tenth of per cent), for the scenarios EndoSpill and ExoSpill relative to the benchmark run without a tariff, always plotted over various tariff rates.

(5)

As before, this specification implies that each sector, s, in a model region, r, imports and exports varieties, i, of each good. In relation to rs1 Eq. (3), Ars1 describes total factor productivity, Qrs1 , given by the Z benchmark situation. The γ -parameter values are taken from the econometric estimation in Section 3. We plug the estimated magnitudes of endogenous import- and export-induced productivity gains reported in Table 1 into γ M and γ E and the region-specific rates of exogenous growth reported in Table 2 into γ0r. Output, Q, imports, M, exports, E, and inputs, Z, are endogenous variables resulting from the general equilibrium. This means, they can deviate from the benchmark situation. In accordance with the econometric Eq. (3), imports are an (Armington) aggregate of intermediate goods from all sectors and regions (countries) in the model. Exports are taken from a sector's 22 A realistic long-term assessment would require the definition of detailed growth scenarios with assumptions on population growth, technical progress, structural change and regional integration. We leave such a scenario analysis for future research and refer the following short-term analysis to today's economic situation. 23 In this short-term analysis, we abstain from discounting in order to keep the results independent of the choice of the discount rate, which would influence the results and which is highly disputed (cf. Gollier and Weitzman, 2010). The stronger the future were discounted, the lower would be the importance of trade-induced productivity gains and hence trade policy from today's point of view, because trade-induced productivity gains occur with time delay.

24 Agriculture/forestry/fishing, chemicals, construction, coke/petroleum/nuclear, electrical/optical equipment, electricity/gas/water supply, food/beverages/tobacco, machinery, metals, mining/quarrying, other non-metallic minerals, other manufacturing/ recycling, paper/printing/publishing, services, transport equipment, textiles, transportation, wood. 25 We choose 2007 as a compromise between using the newest data and using data that are not affected by the economic crisis from 2008 onward. Other benchmark years will be discussed in a robustness check.

7

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

Europe's optimal tariff by more than one per mill, whereas the other regions lose to different extents. The USA absorb part of the imports which, prior to the introduction of the optimal tariff, went to Europe. Hence, the USA benefit from an inverse trade diversion effect. In other words, lower European import demand attenuates world market prices so that the USA can import at lower prices. Russia as an energy exporter to Europe, on the contrary, loses up to 15 per mill, and China as a major exporter to Europe up to 9 per mill of welfare due to Europe's tariff. India, on the contrary, is hardly affected by Europe's trade policy. In most cases, the magnitudes of the (negative or positive) policy effects are larger in ExoSpill than in EndoSpill. The first reason is that the optimal tariff under EndoSpill is lower than under ExoSpill so that the (negative or positive) trade impacts have smaller magnitudes. The second reason is that Europe can achieve higher productivity growth under EndoSpill. Consequently, it will demand more and supply more and/or cheaper exports, which is beneficial for the other regions and mitigates negative welfare effects. Yet, it is not beneficial for the USA, because they benefit from higher European trade barriers due to inverse trade diversion as explained above. The optimal tariff rates of 13 or 16 per cent that we find have realistic magnitudes: Europe's average tariff rate on products from the USA was 7.3 per cent in 200726; it reached 9.1 per cent in 1990 and 12.0 per cent in 1995; it declined to 4.6 in 2010.

Fig. 1. Europe's per mill welfare change in EndoSpill and ExoSpill relative to the benchmark run without a tariff measured over various tariff rates.

The curve has an inverted U-shape, which is typical for optimal tariff analyses. We recall Proposition 2 stating that productivity gains through imports and exports reduce the optimal tariff manipulating the terms of trade in favor of a large open economy. Fig. 1 shows that the optimal, i.e. the welfare-maximizing, tariff rate under EndoSpill is about 13 per cent, whereas the optimal tariff rate under ExoSpill is about 16 per cent, which corroborates and evaluates Proposition 2. We recall Proposition 3 stating that there exists a certain strength of productivity gains from trade such that the incentive to manipulate the terms of trade vanishes. In our simulations, the spillover strength of exports and imports is given by the econometric estimates of the previous section. Apparently, the estimated spillover strength is by far too low to completely enervate the incentive to use a tariff for strategic (terms of trade) reasons. Hence, Proposition 3 describes mainly a theoretical possibility rather than an empirical observation. We recall Proposition 4 stating that the welfare gain for a large open economy achieved via a given tariff rate is lower in the presence of productivity gains through imports and exports than in their absence. Fig. 1 illustrates that the welfare curve for EndoSpill always lies below the welfare curve for ExoSpill in accordance with Proposition 4. The maximum relative welfare gain reached by the optimal tariff is about 3.2 per mill under ExoSpill and only about 2.1 under EndoSpill. This leads us to conclude:

4.2.2. Region-specific trade policy We carry out the same tariff analysis for the other main model regions, i.e. the USA or the BRIC countries (Brazil, Russia, India and China) impose tariffs. Fig. 2 in the Appendix puts the European results depicted by Fig. 1 in perspective to the corresponding results for the other model regions. Table 4 summarizes the same results for the other r regions in the form of optimal tariffs τopt and corresponding welfare effects Wr. In the second row, for example, the USA introduce a tariff, while the other regions do not impose any trade policy. The numbers in this row refer to the United States. The numbers are reported for each scenario, ExoSpill and EndoSpill compared to the benchmark situation without trade policy, and in parentheses as deviations in EndoSpill relative to ExoSpill. All optimal tariffs are significantly greater than zero. This outcome is in line with Brown (1987) who argues that strong terms of trade effects exist independent of the size of the model regions in an Armington specification. We find that the United States' optimal tariffs and welfare gains are higher than Europe's, but their relative changes between ExoSpill and EndoSpill are smaller than in Europe. Russia's resultsand their changes are slightly higher than Europe's. Brazil's numbers are relatively small, but the relative change in welfare and the optimal tariff between the scenarios is highest among all regions. India's optimal tariffs are lower than Europe's, yet its welfare gains compared to the baseline are higher; and the relative change in welfare and the optimal tariff between the scenarios is second highest among the regions. Finally, China's optimal tariffs are the highest among the regions, whereas the changes in the optimal tariff and welfare between ExoSpill and EndoSpill are small and similar to those of the USA. The regional diversity of the results is surprising considering that we assume the same strength of trade-induced productivity spillovers and the same sectoral Armington elasticities for all regions. Thus, region-specific characteristics affect the potential of endogenous tradeinduced productivity gains. They are determined by the region-specific input-output structure including existing productivity levels, the sectoral composition of production, trade patterns, and growth rates as reported in Table 2. In the following subsections we will strive for insights in the drivers of optimal tariffs and their welfare effects.

Result 4. In accordance with Propositions 2 and 4, optimal tariffs and the resulting welfare gains always become lower when accounting for endogenous trade-induced productivity gains. How does the optimal European tariff affect the other regions' welfare? Table 3 answers this question by setting the European tariff to the optimal rate within scenario ExoSpill (16 per cent) and thereafter to the optimal rate within scenario EndoSpill (13 per cent) as depicted by Fig. 1. Positive numbers indicate welfare gains, whereas negative numbers indicate welfare losses. Table 3 reveals that the USA gain from Table 3 Regional welfare effects of Europe's optimal tariffs under the scenarios ExoSpill and EndoSpill in per mill (compared to the benchmark without tariffs). Welfare effects of EU trade policy Wr

European Union United States of America Russia Brazil India China East Asia Rest of the World

EUR USA RUS BRA IND CHN EAS ROW

ExoSpill 3.2 1.5 −15.1 −3.6 −0.4 −9.2 −2.6 −6.8

EndoSpill 2.1 1.1 −12.8 −3.2 −0.6 −7.7 −2.1 −6.2

26 UNCTAD (United Nations Conference on Trade and Development), TRAINS (Trade Analysis Information System) data, URL http://databank.worldbank.org/data/reports. aspx?source=UNCTAD-~-Trade-Analysis-Information-System-(TRAINS) (accessed Jul. 2013).

8

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

high value of 25.27 This mimics the situation with almost no power on international markets as marked by Proposition 1. In accordance with the proposition, we find a negative optimal tariff, i.e. an import subsidy under EndoSpill.

Table 4 r Optimal tariffs τopt of the main model regions in per cent and the corresponding welfare affects Wr in per mill under the scenarios ExoSpill and EndoSpill (compared to the benchmark without tariffs); relative changes of Endospill relative to ExoSpill in per cent in parentheses.

Result 5. The numerical simulations corroborate Proposition 1 stating that trade-induced productivity gains can also be exploited without power on international markets to raise welfare, resulting in a negative optimal tariff. In this case, the import subsidy induces productivity gains that overcompensate the induced deterioration of the terms of trade. However, the welfare gain achieved through this optimal subsidy is small and appears to be negligible for practical trade policy.

Region-specific optimal tariffs and welfare effects r τopt , Wr ExoSpill

European Union United States of America Russia Brazil India China

EUR USA RUS BRA IND CHN

r τopt

Wr

16 24 17 10 15 18

3.2 4.8 3.6 1.3 4.0 12.1

EndoSpill r τopt

Wr

13 (−19%) 20 (−17%) 13 (−24%) 6 (−40%) 11 (−27%) 15 (−17%)

2.1 3.6 2.3 0.4 2.4 8.8

(−34%) (−25%) (−36%) (−69%) (−40%) (−27%)

4.2.5. Sectoral productivity effects In the following, we strive for deeper insights in the effects of optimal tariffs by focusing on European sectors. To illustrate sectoral effects, Fig. 4 in the Appendix plots forgone total factor productivity (total factor productivity loss) in the policy scenarios with reduced trade compared to the benchmark run. For the time being, we keep the restriction that tariffs are equally chosen for all sectors in the European economy as in Section 4.2.1. We explore European trade policy, while the other regions do not engage in trade policy. Clearly, reduced trade results in reduced productivity gains. We run scenario EndoSpill twice: once by setting the tariff to its optimal level as before, and once by setting the tariff to the optimal level given by the ExoSpill scenario. We signify the latter setup by EndoSpillExoTariff. In EndoSpill-ExoTariff, the tariff is set to a rate above the optimal level. Thus, it generates higher forgone total factor productivity than EndoSpill in all sectors as illustrated in Fig. 4. This forgone productivity is solely driven by the trade-induced productivity spillover channel since the tariff rate and all other model parameters are kept constant. Note that the difference in forgone productivity between the two scenarios represents the forgone welfare by not taking into account that trade induces endogenous productivity gains. The figure illustrates that services, construction and electricity/gas/water supply suffer the highest forgone total factor productivity in both scenarios, whereas agriculture/ forestry/fishing, mining/quarrying and other non-metallic minerals suffer to the smallest extent. Notably, the economy-wide welfare effect of the trade policies under scrutiny is positive, as shown in the previous analyses, because the government collects the revenues from the tariffs and redistributes them to the representative consumer in a lump-sum way. Furthermore, the tariffs shift demand from imports to domestic supply, which is beneficial for domestic producers. These positive effects overcompensate the forgone sectoral TFP losses and are not visible in Fig. 4.

4.2.3. Variation of the benchmark year It is a strength of WIOD to offer benchmark data for the years 1995 to 2009. We exploit this by calibrating the model to other benchmark years for comparison. Fig. 3 in the Appendix shows the outcome for the European Union (EUR). Besides the year 2007 (which is also available in the GTAP 8 data), we report results for the year 2004 (which is also available in the GTAP 7 data) and for the most recent available years 2008 and 2009, which goes beyond GTAP 8. We report the results in parentheses in the following form: (optimal tariff in per cent/welfare change with respect to benchmark in per mill under ExoSpill | optimal tariff in per cent/welfare change with respect to benchmark in per mill under EndoSpill). In 2004, the optimal tariffs and the corresponding welfare gains for Europe under ExoSpill and EndoSpill (14/2.3|11/1.5) are significantly smaller than for 2007 (16/3.2|13/2.1). In 2008, the optimal tariffs are the same as in 2007, whereas the welfare gains are slightly higher (16/3.6|13/2.4 ). In 2009, the values are again smaller (15/2.5|11/1.6 ), similar to the result for 2004. This robustness check demonstrates that the choice of the benchmark year can play a role in absolute terms, i.e. for some years the results are very similar, whereas they differ for some other years. Notwithstanding, the results are very robust when deviations are measured relative to benchmark data, which is the usual case in policy analysis. A clear time trend of the results across different benchmark years is not evident. We conclude that the sensitivity of the results to the choice of the benchmark year is limited and does not affect the qualitative interpretation of the results.

4.2.6. Sector-specific trade policy Now we relax the assumption of an identical tariff for all goods imported into a region in order to explore the sector dimension in greater detail. To this end, we introduce sector-specific tariffs. For each EURs European sector s, we calculate the optimal tariff τopt that maximizes Europe's welfare WEUR. Tariffs on all goods except s are fixed to zero EURs . The results of these simulations are reported when determining τopt by Table 6 in the Appendix.28

4.2.4. Variation of Armington elasticities In another robustness check, we vary the Armington elasticities, i.e. the elasticities of substitution between foreign varieties as well as between the import bundle and the domestic variety. The original Armington elasticity values (see Table 6 in the Appendix) are standard values taken from GTAP. We refer to Europe calibrated to 2007 data. Higher Armington elasticities make varieties from different countries more similar and reduce market power (compare Section 2.2). Hence, the optimal tariffs and the resulting welfare gains decline in higher Armington elasticities. Fig. 3(e) in the Appendix presents the results for all Armington elasticities set to a high value of eight, whereas Fig. 3(f) presents the results for all Armington elasticities set to a low value of two. In the high Armington case, the optimal tariffs and welfare gains drop to (10/2.6|7/1.3). In the low Armington case, the values soar to (63/17.6|59/15.9 ). We conclude that the sensitivity of the results to the choice of Armington elasticities is high. Moreover, a lower (higher) Armington elasticity represents lower (higher) substitutability between varieties and thus higher (lower) market power and vice versa. Against this background, the optimal tariffs and corresponding welfare effects rise in market power in accordance with Proposition 1. Furthermore, we set the Armington elasticities of Europe to a very

27 Perfect substitutes and perfect competition on international markets would require an infinite Armington elasticity, which is not feasible for this type of model. 28 The first column lists the 18 sectors. Columns 2 to 5 display the sectoral optimal EURs tariff τopt in per cent and the corresponding European welfare effects WEUR in per mill, both in the ExoSpill and EndoSpill scenario. The percentage changes in parentheses (6th and 7th column) show the differences between the two scenarios in per cent. All further columns display parameters potentially explaining the results. The Armington elasticity MEURs1 between foreign and domestic varieties σsa′, the import and export intensities, EURs1 and

EEURs1 Q EURs1

, in per cent, the sector size

Q EURs1 Q EUR1

Q

measured as the share of Europe's total output CEURs1

in per cent, and the share of good s consumed by final demand EURs1 in per cent as a Q measure for the position in the value chain. A higher final demand share implies more downstreamness. The index ‘1’ indicates that the values are taken from the first model period.

9

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

The export intensity has a significant and positive coefficient in part of the regressions. A larger sector share has a positive, though weakly significant impact on welfare. Its impact on the optimal tariff is not significant. A larger consumption share, interpreted as a measure for downstreamness, significantly increases the optimal tariff, but not welfare. In summary, the results point to positive effects of the import and export intensity as well as the sector and the consumption share on the potential of trade policy. In accordance with the previous estimates presented in Table 1, imports entail a clearer effect than exports.

In general, optimal tariffs and welfare gains are smaller in the presence of trade-induced productivity gains than without them. This confirms both the theoretical considerations and the numerical findings for economy-wide optimal tariffs. Sectoral optimal tariffs are generally lower than the economy-wide European tariff (according to Table 4, 16 per cent in ExoSpill and 13 per cent in EndoSpill). Only the optimal tariff for textiles is greater than the economy-wide one in the ExoSpill scenario, while three sectors exhibit optimal tariffs above the economy-wide one in the EndoSpill scenario: food, transport equipment and textiles. The largest welfare gain of 0.7 per mill is achieved by the optimal tariff on mining products. Despite its high Armington elasticity of about 8.5, the mining sector's huge import intensity of 150 per cent allows it to exert substantial market power. The comparison of the chemicals and metals sectors is illuminating. Both account for about four per cent of European production. Chemicals, however, exhibit a higher import intensity than metals, 15 per cent compared to 13 per cent, and a lower Armington elasticity, 3.3 compared to 3.6. Consequently, sectoral optimal tariffs are higher for chemicals than for metals, i.e. 15 per cent compared to 13 per cent in the ExoSpill and 12 per cent compared to 10 per cent in the EndoSpill scenario. Welfare effects are stronger as well. In the EndoSpill scenario, the welfare gain is 0.19 per mill for chemicals and 0.12 per mill for metals. Without trade-induced productivity gains, expressed by ExoSpill, the welfare gain is 0.29 per mill for chemicals and 0.22 for metals. Accounting for 53 per cent of total production, the services sector is the largest sector in the European economy. Its Armington elasticity is low (1.9). The import intensity is low, too (2.5 per cent). As a consequence, the sectoral optimal tariffs (5 per cent under ExoSpill and 2 per cent under EndoSpill) as well as the corresponding welfare effects (0.06 per mill in the ExoSpill and 0.02 per mill in the EndoSpill scenario) are small. Notwithstanding, services is the sector for which neglecting trade-induced productivity gains is most detrimental to Europe's welfare. Considering endogenous trade-induced productivity gains reduces the sectoral optimal tariff by 60 per cent and the welfare gain by 75 per cent. A larger sector size implies that any trade-induced productivity gain affects a larger part of the economy and thus has a stronger impact on the economy. Economic intuition is less clear-cut with respect to downstreamness. If trade policy is induced on an upstream sector, it will not only affect this sector, but also consecutive downstream sectors. This mechanism enlarges the impact of productivity gains, but also the welfare-reducing distortion induced by tariffs.

4.3. Summary and discussion We summarize the results of the numerical policy experiments as follows: Result 6. Trade-induced productivity gains generate regionally and sectorally diverse optimal tariffs and induced welfare effects. Lower Armington elasticities (lower substitution possibilities between traded goods and hence more distinct goods), higher import or export intensities as well as higher sector or consumption shares (more downstreamness) raise the optimal tariff and its welfare effects. On the one hand, our sensitivity analysis identifies Armington elasticities as a key parameter that creates discrepancies in trade policy results across models (cf. Ackerman and Gallagher, 2008). This result calls for more research into Armington and other elasticities of substitution (as Koesler and Schymura, 2015; Olekseyuk and Schürenberg-Frosch, 2016). On the other hand, the sensitivity analysis indicates a limited effect of choosing different benchmark years for the model calibration on the results, especially when measuring deviations between the policy scenario and the benchmark scenario in relative from. We are able to obtain this new insight with WIOD, because it provides consistent benchmark data for a number of different years. Nonetheless, the WIOD are subject to limitations such as assumptions on constant trade shares and the Rest of the World (ROW) representing a residual region. Thus, results for ROW lack a direct real-world correspondence. The WIOD also lack information on actually implemented policy instruments like tariff or subsidy rates. For an assessment of these policies, the modeller would need to transfer them from other sources like GTAP. Notably, the magnitude of the trade-induced policy effects depends on the econometric estimates. Our estimates used in the model are in line with other studies (e.g. Hübler and Keller, 2009). Notwithstanding, the related econometric literature is very broad and finds diverse results, which indicates statistical uncertainty in trade-related policy effects, independent of the model implementation. Compared to existing trade policy studies, we find the following differences and congruencies. As our analysis, Balistreri et al. (2011) find limited welfare effects of tariff variations. The welfare effects in our policy experiments have magnitudes of several per mill. This means, they are of limited importance for today's trade policy. This result is in line with Ackerman and Gallagher (2008) who review model-based trade policy studies and conclude that the gains from free trade have limited magnitudes. For example, Itakura (2014) studies a scenario with a free trade agreement of China, Japan and Korea and finds policy effects with the same order of magnitude that we detect. However, we assess tariffs imposed by single model regions but no free trade agreements, (non-)cooperative bargaining or strategic retaliation (as Itakura, 2014; Li and Whalley, 2014; Ossa, 2014), which can result in more pronounced effects. Compared to the optimal tariff analysis by Ossa (2014), which neglects international productivity gains, we find significantly lower optimal tariffs and lower welfare gains. Our results are close to those by Markusen and Wigle (1989). They find an optimal tariff for the USA of about 18 percent in a multi-region Nash equilibrium without tradeinduced productivity gains, whereas we find an optimal tariff for the

4.2.7. Determinants of optimal tariffs In order to obtain deeper insights in the determinants of optimal tariffs and welfare gains, we extend the sectoral analysis in the following way. First, we run the sectoral tariff experiment for the year 2007 for all model regions instead for Europe only. In each experiment, one region chooses optimal sectoral tariffs, whereas the remaining regions do not engage in trade policy. As a result, we obtain a version of Table 6 for each model region (available upon request). We combine the tables and focus on the scenario EndoSpill that takes the endogeneity of productivity gains into account. We obtain in total 144 observations. rs Second, we run regressions with the optimal tariff τopt or the resulting welfare gain Wr generated by the policy experiments as the dependent variable. The explanatory variables are the import and export intensities, the sector share and the consumption share. The explanatory variables are directly given by WIOD independent of the policy experiments. Details of the estimation procedure can be found in the Appendix together with the estimation results in Table 5. The results show that the import intensity creates a highly significant and positive effect on the optimal tariff and the welfare effect in most regressions. The magnitude is higher for the welfare effect than for the optimal tariff. 10

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

the policies under scrutiny today because productivity gains require considerable time to emerge.

USA of 20 per cent with trade-induced productivity gains but without a Nash game. Our policy conclusions are also in accordance with studies that examine the influence of international technology spillovers on climate policy costs (e.g. Bosetti et al., 2008; Leimbach and Baumstark, 2010; Hübler, 2011). These studies find a significant, but small influence of technology spillovers on climate policy. Finally, we use a two-period setup, because we are not looking at long-term growth paths. Running the model over a long time horizon would result in a higher cumulated welfare gain. Rutherford and Tarr (2002) simulate a 54-years time horizon and find a remarkable average welfare gain of ten per cent induced by a ten per cent tariff cut.

Acknowledgement We thank two anonymous reviewers and Editor Sushanta Mallick for very helpful comments as well as participants of the International Conference on Economic Modelling in Bali in 2014 and a research seminar at University Heidelberg in 2013. We also thank Simon Koesler and Alexander Glas for their great help. We gratefully acknowledge financial support by the state of Baden-Württemberg within the programme SEEK (Strengthening Efficiency and Competitiveness in the European Knowledge Economies) carried out at ZEW (Centre for European Economic Research) in Mannheim, Germany. We thank Andreas Löschel and Victoria Alexeeva-Talebi for making this project possible. This work has also benefited from financial support for setting up the basic WIOD CGE model within the WIOD (World Input-Output) project, funded by the European Commission as part of the 7th Framework Programme, Theme 8: Socio-Economic Sciences and Humanities.

5. Conclusion The results of the trade policy analysis are relevant for the current policy controversy about regional trade agreements, such as the transAtlantic and trans-Pacific agreements, versus global free trade. The results suggest a considerable reduction of trade barriers in order to optimally exploit trade-induced productivity gains. Ignoring tradeinduced productivity gains in trade policy results in limited, but measurable welfare losses. Nevertheless, when taking trade-induced productivity gains into account, single countries still have incentives to choose strategic non-zero tariffs to raise their own welfare. The optimal tariff rates that we find have similar magnitudes as tariffs rates in practice. Such strategic policies, however, have a beggar-thy-neighbor character, this means, they are harmful for trading partners. As an exception, we find that EU tariffs are beneficial for the USA, because Europe's reduced demand lowers world market prices and shifts imports to the USA. Regional trade agreements, however, hinder potential productivity gains for countries within the scope of the agreement from trade with countries outside the agreement as well as productivity gains for the countries outside. Consequently, trade-induced productivity gains provide an argument for free trade with a global scope additionally to conventional gains from trade. The WTO (World Trade Organization) is the institution with the capability to foster the path to global free trade. Notably, our analysis deals with conventional tariffs which are nowadays in many sectors relatively low. The reduction of non-tariff trade barriers is of high economic relevance but due to missing data not captured by our model analysis. This means, if a free-trade agreement reduces non-tariff barriers as well, the resulting trade-induced productivity and welfare gains will come on top of those measured in this paper. In this context, social and environmental standards are a crucial and controversial topic. They are, however, beyond the scope of a stylized trade model like ours. The model simulations show that the incentive to reduce tariffs in order to exploit trade-induced productivity gains widely differs across countries and sectors. We find stronger tariff reductions for Brazil than for the USA, the EU or China. Notably, small open economies without power on international markets can also benefit from trade-induced productivity gains by reducing tariffs, because the positive productivity spillover occurs at home, not in a foreign country as in the case of trans-boundary (environmental) externalities (Markusen, 1975). On the contrary, strategic tariffs are particularly powerful instruments for raising welfare when imposed on large, trade-intensive downstream sectors producing distinct goods. Thus, free traders and deregulators should keep an eye on such sectors. Despite critical empirical uncertainties in Armington elasticities and trade-induced productivity gains as well as the limitations of the WIOD, we hope that this study is a relevant step towards the consistent integration of numerical modelling with theory and econometrics. While our analysis has focused on productivity gains from trade today, further research may study the impact of trade on future growth paths in long-term scenarios (as indicated by empirical evidence by Shahbaz, 2012). Attributing a lower value to the future reduces the relevance of

Appendix A. Supplementary Information Supplementary information associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.econmod. 2016.11.006. References Ackerman, F., Gallagher, K.P., 2008. The shrinking gains from global trade liberalization in computable general equilibrium models – a critical assessment. Int. J. Polit. Econ. 37 (1), 50–77. Akgul, Z., Villoria, N.B., Hertel, T.W., 2016. GTAP-HET: introducing firm heterogeneity into the GTAP model. J. Glob. Econ. Anal. 1 (1), 111–180. Armington, P., 1969. A theory of demand for products distinguished by place of production. IMF Staff Papers. Washington, DC, USA. Vol. 16, pp. 159–178. Balistreri, E.J., Hillberry, R.H., Rutherford, T.F., 2011. Structural estimation and solution of international trade models with heterogeneous firms. J. Int. Econ. 83, 95–108. Balistreri, E.J., Markusen, J.R., 2009. Sub-national differentiation and the role of the firm in optimal international pricing. Econ. Model. 26, 47–62. Balistreri, E.J., Rutherford, T.F., 2012. Subglobal carbon policy and the competitive selection of heterogeneous firms. Energy Econ. 34 (Suppl. 2), 190–197. Bosetti, V., Carraro, C., Massetti, E., Tavoni, M., 2008. International energy R & D spillovers and the economics of greenhouse gas atmospheric stabilization. Energy Econ. 30 (6), 2912–2929. Broda, C., Limao, N., Weinstein, D.E., 2008. Optimal tariffs and market power: the evidence. Am. Econ. Rev. 98 (5), 2032–2065. Brown, D.K., 1987. Tariffs, the terms of trade, and national product differentiation. J. Policy Model. 9, 503–526. Coe, D., Helpman, E., 1995. International R & D spillovers. Eur. Econ. Rev. 39, 859–887. Coe, D., Helpman, E., Hoffmaister, A., 1997. North-South R & D spillovers. Econ. J. 107, 134–149. Das, J., DeLoach, S., 2003. Strategic trade policy in the presence of reputation spillovers. J. Int. Trade Econ. Dev. 12 (2), 101–116. Diao, X., Rattsø, J., Stokke, H.E., 2005. International spillovers, productivity growth and openness in thailand: an intertemporal general equilibrium analysis. J. Dev. Econ. 76, 429–450. Dietzenbacher, E., Los, B., Stehrer, R., Timmer, M., de Vries, G., 2013. The construction of world input-output tables in the wiod project. Econ. Syst. Res. 25 (1), 71–98. Dixon, P.B., Jerie, M., Rimmer, M.T., 2016. Modern trade theory for cge modelling: the armington, krugman and melitz models. J. Glob. Econ. Anal. 1 (1), 1–110. Eaton, J., Kortum, S., 2002. Technology, geography, and trade. Econometrica 70 (5), 1741–1779. Ethier, W.J., 1982. National and international returns to scale in the modern theory of international trade. Am. Econ. Rev. 72 (3), 389–405. Felbermayr, G., Jung, B., Larch, M., 2013. Optimal tariffs, retaliation, and the welfare loss from tariff wars in the Melitz model. J. Int. Econ. 89 (1), 13–25. Finicelli, A., Pagano, P., Sbracia, M., 2013. Ricardian selection. J. Int. Econ. 89, 96–109. Girma, S., Greenaway, D., Kneller, R., 2004. Does exporting increase productivity? A microeconometric analysis of matched firms. Rev. Int. Econ. 12 (5), 855–866. Gollier, C., Weitzman, M.L., 2010. How should the distant future be discounted when discount rates are uncertain? Econ. Lett. 107, 350–353. Gros, D., 1987. A note on the optimal tariff, retaliation and the welfare loss from tariff wars in a framework with intra-industry trade. J. Int. Econ. 23 (3–4), 357–367. Grossman, E., Helpman, E., 1991. Trade, knowledge spillovers, and growth. Eur. Econ. Rev. 35 (2–3), 517–526.

11

Economic Modelling (xxxx) xxxx–xxxx

M. Hübler, F. Pothen

bargaining: some numerical simulation results. Econ. Model. 37, 65–78. Mallick, S., Yang, Y., 2013. Productivity performance of export market entry and exit: evidence from Indian firms. Rev. Int. Econ. 21 (4), 809–824. Markusen, J.R., 1975. International externalities and optimal tax structures. J. Int. Econ. 5, 15–29. Markusen, J.R., Rutherford, T.F., Tarr, D.G., 2005. Trade and direct investment in producer services and the domestic market for expertise. Can. J. Econ. 38 (3), 758–777. Markusen, J.R., Wigle, R.M., 1989. Nash equilibrium tariffs for the united states and canada: the roles of country size, scale economies, and capital mobility. J. Polit. Econ. 97 (2), 368–386. Melitz, M.J., 2003. The impact of trade on intra-industry reallocations and aggregate industry productivity. Econometrica 71 (6), 1695–1725. Murat, M., Pigliaru, F., 1998. International trade and uneven growth: a model with intersectoral spillovers of knowledge. J. Int. Trade Econ. Dev. 7 (2), 221–236. Olekseyuk, Z., Schürenberg-Frosch, H., 2016. Are Armington elasticities different across countries and sectors? A European study. Econ. Model. 55, 328–342. Ossa, R., 2014. Trade wars and trade talks with data. Am. Econ. Rev. 104 (12), 4104–4146. Rutherford, T.F., Tarr, D.G., 2002. Trade liberalization, product variety and growth in a small open economy: a quantitative assessment. J. Int. Econ. 56, 247–272. Rutherford, T.F., Tarr, D.G., 2008. Poverty effects of Russia's WTO accession: modeling real households with endogenous productivity effects. J. Int. Econ. 75, 131–150. Saggi, K., 2002. Trade, foreign direct investment, and international technology transfer: a survey. World Bank Res. Obs. 17 (2), 191–235. Shahbaz, M., 2012. Does trade openness affect long run growth? Cointegration, causality and forecast error variance decomposition tests for Pakistan. Econ. Model. 29, 2325–2339. The Economist, 2015. Free exchange|game of zones – regional trade agreements aren't as good as global ones but they are still beneficial. March 21st, 65. Timmer, M.P., Dietzenbacher, E., Los, B., Stehrer, R., de Vries, G.J., 2015. An illustrated user guide to the world input-output database: the case of global automotive production. Rev. Int. Econ. 23 (3), 575–605. Yang, Y., Mallick, S., 2014. Explaining cross-country differences in exporting performance: the role of country-level macroeconomic environment. Int. Bus. Rev. 23 (1), 246–259. Yang, Y., Mallick, S., 2010. Export premium, self-selection and learning-byexporting: evidence from Chinese matched firms. World Econ. 33, 1218–1240.

Hamilton, B., Whalley, J., 1983. Optimal tariff calculations in alternative trade models and some possible implications for current world trading arrangements. J. Int. Econ. 15, 323–348. Havranek, T., Irsova, Z., 2011. Estimating vertical spillovers from FDI: why results vary and what the true effect is. J. Int. Econ. 85, 234–244. Himics, M., Britz, W., 2016. Flexible and welfare-consistent tariff aggregation over exporter regions. Econ. Model. 53, 375–387. Hübler, M., 2015. A model of endogenous growth that elucidates the complexity of South-North Convergence. J. Int. Trade Econ. Dev. 24 (5), 724–750. Hübler, M., 2011. Technology diffusion under contraction and convergence: a CGE analysis of China. Energy Econ. 33 (1), 131–142. Hübler, M., Keller, A., 2009. Energy savings via FDI? Empirical evidence from developing countries. Environ. Dev. Econ. 15 (1), 59–80. Itakura, K., 2014. Impact of liberalization and improved connectivity and facilitation in ASEAN. J. Asian Econ. 35, 2–11. Jakob, M., Marschinski, R., Hübler, M., 2013. Between a rock and a hard place: a tradetheory analysis of leakage under production- and consumption-based policies. Environ. Resour. Econ. 56 (1), 47–72. Jalles, J.T., Tavares, J., 2015. Trade, scale or social capital? Technological progress in poor and rich countries. J. Int. Trade Econ. Dev. 24 (6), 767–808. Johnson, H.G., 1954. Optimum tariffs and retaliation. Rev. Econ. Stud. 21 (2), 142–153. Keller, W., 2004. International technology diffusion. J. Econ. Lit. 42 (3), 752–782. Kennan, J., Riezman, R., 1988. Do big countries win tariff wars? Can. J. Economics 23 1, 70–83. Kennan, J., Riezman, R., 1990. Optimal tariff equilibria with customs unions. Int. Econ. Rev. 29 (1), 81–85. Koesler, S., Pothen, F., 2013. The basic WIOD CGE model: a computable general equilibrium model based on the world input-output database. ZEW-Dok., 13–104, (URL 〈ftp://ftp.zew.de/pub/zewocs/docus/documentation1304.pdf〉 (accessed Aug. 2016). Koesler, S., Schymura, M., 2015. Substitution elasticities in a constant elasticity of substitution framework - empirical estimates using non-linear Least Squares. Econ. Syst. Res. 27 (1), 101–121. Koopman, R., Wang, Z., Wei, S.J., 2014. Tracing value-added and double counting in gross exports. Am. Econ. Rev. 104 (2), 459–494. Leimbach, M., Baumstark, L., 2010. The impact of capital trade and technological spillovers on climate policies. Ecol. Econ. 69, 2341–2355. Li, C., Whalley, J., 2014. China's potential future growth and gains from trade policy

12