Unemployment in an Interdependent World. by Gabriel Felbermayr, Mario Larch and Wolfgang Lechthaler

Transcription

1 Unemployment in an Interdependent World by Gabriel Felbermayr, Mario Larch and Wolfgang Lechthaler No August 2009, revised April 2012

2 Kiel Institute for the World Economy, Düsternbrooker Weg 120, Kiel, Germany Kiel Working Paper No August 2009, revised April 2012 Unemployment in an Interdependent World Gabriel Felbermayr, Mario Larch and Wolfgang Lechthaler Abstract: How do changes in labor market institutions like more generous unemployment benefits in one country affect labor market outcomes in other countries? We set up a two-country Armingtonian trade model with frictions on the goods and labor markets. Contrary to the literature, higher labor market frictions increase unemployment at home and abroad. The strength of the spillover depends on the relative size of countries and on trade costs. It is exacerbated when real wages are rigid. Using panel data for 20 rich OECD countries, and controlling for institutions as well as for business cycle comovement, we confirm our theoretical predictions. JEL classification: F11, F12, F16, J64, L11 Gabriel Felbermayr Economics Department, University of Stuttgart-Hohenheim, Stuttgart, Germany gfelberm@uni-hohenheim.de Mario Larch University of Bayrueth and Ifo Institute for Economic Research, Bayreuth, Germany mario.larch@uni-bayreuth.de Wolfgang Lechthaler Kiel Institute for the World Economy Kiel, Germany wolfgang.lechthaler@ifw-kiel.de The responsibility for the contents of the working papers rests with the author, not the Institute. Since working papers are of a preliminary nature, it may be useful to contact the author of a particular working paper about results or caveats before referring to, or quoting, a paper. Any comments on working papers should be sent directly to the author. Coverphoto: uni_com on photocase.com

3 Unemployment in an Interdependent World Gabriel J. Felbermayr, Mario Larch and Wolfgang Lechthaler May 9, 2012 Abstract How do changes in labor market institutions like more generous unemployment benets in one country aect labor market outcomes in other countries? We set up a two-country Armingtonian trade model with frictions on the goods and labor markets. Contrary to the literature, higher labor market frictions increase unemployment at home and abroad. The strength of the spillover depends on the relative size of countries and on trade costs. It is exacerbated when real wages are rigid. Using panel data for 20 rich OECD countries, and controlling for institutions as well as for business cycle comovement, we conrm our theoretical predictions. Keywords: Spillover eects of labor market institutions; unemployment; international trade; search frictions; heterogeneous rms JEL-Codes: F11, F12, F16, J64, L11 Acknowledgments: We are grateful to Kerem Cosar, Peter Egger, Christian Haefke, Oleg Itskhoki, Eckhard Janeba, Philip Jung, Wilhelm Kohler, Peter Neary, Jan van Ours, Julien Prat, Priay Ranjan, Stephen Redding, Hans-Joerg Schmerer, Hans-Werner Sinn, Cristina Terra, Lars Vilhuber, Christoph Weiss and Zhihong Yu as well as to seminar participants at the GEP-IFN Workshop in Stockholm 2008, the SED Annual Meeting in Istanbul 2009, the NOeG conference 2009 in Linz, the CEA Meetings 2009 in Toronto, the CESifo Institute Munich, the Kiel Institute for the World Economy, the Leverhulme Centre for Research on Globalisation and Economic Policy (GEP), the University of Aarhus, the University of Cergy-Pontoise, the University of Erfurt, the University of Innsbruck, the University of Mannheim, and the University of Tuebingen for comments and remarks. All remaining errors are ours. Ifo Institute for Economic Research at the University of Munich, LMU Munich, CESifo, and GEP at the University of Nottingham, Poschingerstrasse 5, Munich, Germany. felbermayr@ifo.de. Felbermayr acknowledges nancial support from the Fritz Thyssen Foundation under grant no University of Bayreuth, CESifo, Ifo Institute for Economic Research at the University of Munich and GEP at the University of Nottingham, Universitaetsstrasse 30, Bayreuth, Germany. mario.larch@unibayreuth.de. Larch acknowledges nancial support from the German Research Foundation (DFG) under grant no Kiel Institute for the World Economy, Duesternbrooker Weg 120, Kiel, Germany. wolfgang.lechthaler@ifw-kiel.de.

4 1 Introduction In a world, where countries are increasingly linked via trade in goods, labor market outcomes become more strongly interdependent: one country's institutions will not only aect its own rate of unemployment but also that of its trading partners. This has important implications for labor market reforms and for the normative eects of trade liberalization. In this paper we therefore shed light on the direction and magnitude of spillover eects on unemployment due to changes of labor market institutions in a trading partner country. Our contribution is twofold. First, we develop a very tractable two-country, asymmetric general equilibrium framework that combines an Armingtonian model of national product dierentiation with the canonical search and matching approach of Mortensen and Pissarides (1994). In our setup, structural (i.e., non-cyclical) unemployment rates are positively correlated across countries. The strength of the correlation depends on the size and openness of countries as well as on the degree of real wage exibility. Relatively larger and less open countries are harmed more by own labor market frictions and less by foreign ones, whereas small and open countries are hit relatively harder by foreign labor market frictions and less by own frictions. Second, we empirically investigate cross-country labor market linkages using panel data for 20 rich OECD countries for Using instrumental variable panel estimators and controlling for institutions as well as for business cycle comovement, we conrm our theoretical ndings and reject alternative approaches based on multi-sector comparative advantages. There is an emerging consensus in the macroeconomic labor literature that institutions matter for structural unemployment; in particular, pervasive product market regulation increases unemployment. 1 One may therefore conjecture that trade barriers also foster unemployment. Recent econometric evidence supports this view, see Dutt, Mitra, and Ranjan (2009). Moreover, to the extent that labor market institutions aect the volume and pattern of trade between countries, it is likely that trade acts as a vehicle through which institutional features of one country also aect labor market outcomes in other countries. 1 See for example Layard, Nickell, and Jackman (1991); Nickell (1997); Nickell and Layard (1999); Blanchard and Wolfers (2000); or Ebell and Haefke (2009). 1

5 A large strand of the theoretical literature uses the Heckscher-Ohlin model to study the eect of labor market institutions on the pattern of trade. Brecher (1974) studied minimum wages; Davis (1998) has provided a generalization. Due to factor price equalization, if a capital-abundant country has a binding minimum wage, trade exacerbates job losses there, while the labor-abundant country with perfect labor markets benets from higher wages without the incidence of unemployment. The precise nature of labor market imperfections matters little for the direction of linkages: Models with search frictions and wage bargaining, such as Davidson, Martin, and Matusz (1988, 1999) and Helpman and Itskhoki (2010) come to similar conclusions as Davis (1998). 2 Models explaining trade based on comparative advantages predict that an increase in labor market frictions at home leads to higher unemployment at home. The eects on the trading partner depend on the relative capital-labor endowments: Higher domestic unemployment increases capital-labor abundance at home. A relatively capital-rich home economy specializes more strongly on capital-intensive goods while the foreign country produces more of the labor-intensive goods. Labor demand in the foreign country goes up and the marginal value product of labor increases. Firms create more vacancies, which leads to a fall in unemployment. The opposite logic applies if the home country is labor-rich. Hence, the sign of the correlation of unemployment rates between countries depends crucially on the comparison of capital-labor ratios across countries. We therefore include capital-labor ratios in our empirical specications. It will turn out that the predictions of the models based on comparative advantages are not supported by our empirical analysis. The more recent literature focuses on rm-level increasing returns to scale and product dierentiation (Krugman, 1980) and the generalization to heterogeneous rms (Melitz, 2003). Some papers draw on the fair wage or eciency wage approach, others use the search and matching approach. A well-known limitation of Krugman-type models with asymmetries is their lack of analytical tractability. Hence, Egger and Kreickemeier (2008, 2009), Eckel and Egger (2009) or Felbermayr, Prat, and Schmerer (2011) concentrate on symmetric countries, so that institutional asymmetries and their cross-country implications cannot be studied. Helpman and Itskhoki (2010) have maintained analytical tractability by introducing a numéraire sector that remains 2 More recently, Cuñat and Melitz (2007) study the eect of cross-country dierences in ring restrictions on patterns of comparative advantage in a Ricardian setting, but they do not address the issue of unemployment. 2

6 unaected by monopoly power and trade costs. 3 This strategy blends the comparative advantage channel with Krugman/Melitz mechanisms. Often income eects are ruled out as preferences are quasilinear in the numéraire good. However, when studying macroeconomic implications of trade, such as on unemployment, it is important to take income eects into account. Generally, higher unemployment at home reduces demand for imports from foreign countries. This leads to a positive correlation of unemployment rates. Eects of this type operate in the new economic geography literature 4 but have hardly been explored in models of trade and unemployment. 5 The eect relies crucially on the use of a full-edged general equilibrium model. Our focus lies in deriving empirically testable predictions on observable market characteristics, like unemployment benets, size, geography, and real wage exibility and their conditioning eect on labor market spillovers. No existing model provides predictions about all these facts. Hence, we develop a simple, full-edged general equilibrium trade model, in which we are able to obtain analytical results for all hypotheses of interest. More precisely, we use a one-shot matching version of Pissarides (2000) in an Armingtonian trade model. Recent theoretical work by Arkolakis, Costinot and Rodriguez-Clare (2011) compares the Armington model to more elaborate setups such as the Krugman (1980) or Melitz (2003) model. Their work suggests that the Armingtonian model provides a good reduced form of more complicated models. In particular, the link between welfare and trade volume has the same functional form. Moreover, the model can be easily extended to allow for trade costs and gives rise to a gravity equation, implying that the predictions of bilateral trade ows of the model will do a reasonable good job in explaining observed bilateral trade ows. 6 3 Moreover, they assume quasi-linear preferences. 4 See for an overview Fujita, Krugman, and Venables (1999) or Baldwin, Forslid, Martin, Ottaviano, and Robert-Nicoud (2003). 5 Egger, Egger, and Markusen (2010) form an exception. They use a two-country heterogeneous rms model with homogeneous workers, no trade costs, and a xed number of potential entrants. Their model also yields a positive correlation between countries' unemployment rates resulting from binding real minimum wages. In their set-up rm heterogeneity is crucial for their result, as minimum wages aect the composition of producers and allows for binding minimum wages in both countries. With homogeneous rms in a Krugman (1980)-type model without trade costs and diversied production, minimum wages could only be binding in one country. 6 Helpman and Itskhoki (2010) and Egger, Egger and Markusen (2010) use versions of the Melitz model. They achieve tractability by either ignoring trade costs altogether or disallowing them in the perfect-competition nofrictions sector. Neither Helpman and Itskhoki (2010) nor Egger, Egger, and Markusen (2010) do investigate the role of economic geography or relative country size, nor do they carry out any empirical investigations. 3

7 The remainder of the paper is structured as follows. Section 2 outlines the theoretical model. Section 3 derives theoretical predictions. Section 4 provides empirical evidence for these predictions. The last section concludes. The Appendix provides analytical proofs to all theoretical propositions and robustness checks for the empirical analysis. 2 Model Setup We use a simple, analytically tractable theoretical framework that is able to deliver our key predictions. Since our focus is on international spillovers of labor market institutions we need a model with at least two large, potentially asymmetric countries. Countries can dier with respect to labor market institutions and their relative size. To allow for geography, we include trade costs. We rely on the static perfect competition Armingtonian model with one-shot matching frictions on the labor market. 2.1 Goods markets Denote Home by a subscript H and Foreign by a subscript F, and let i, j {H, F }. According to the Armington assumption, each country produces a single intermediate input good under perfect competition one-to-one from labor, the only factor of production. Labor supply is exogenous and given by L j. Domestic and imported intermediate inputs are assembled to a nal output good using a constant elasticity of substitution production function Q j = [ i (y ij ) σ 1 σ ] σ σ 1, (1) where the elasticity of substitution is σ > 1. 7 y ij denotes the use of an input produced in country i in country j s nal good assembly. The total value of output (GDP) in economy j is Y j = Q j P j, where P j = ( ) 1/(1 σ) i p1 σ ij is the aggregate price index. Prices of domestic and imported varieties dier due to iceberg 7 In contrast to monopolistic competition models, the assumption σ > 1 is not required for the existence of a prot-maximizing choice of output by rms. It is needed to ensure that foreign sales fall in trade costs. 4

8 transportation costs τ ij 1; hence, p ij = τ ij p i. 8 Expenditures of country j for an intermediate input from country i are denoted by X ij. By the aggregate income constraint, total expenditure is equal to GDP, Y j = i X ij which is in turn equal to the production value of the only active sector, hence Y j = p j S j, where S j is output of country j. Prot maximization implies the following demand (D ij ) and expenditures (X ij ) for intermediate inputs ( ) σ pij Y j D ij = τ ij, X ij = P j P j ( pij P j ) 1 σ Y j. (2) 2.2 Labor market Our description of the labor market is a static version of Pissarides (2000). 9 The search-andmatching setup has the advantage that it generates a rather parsimonious link between labor market institutions and the equilibrium rate of unemployment. Moreover, the simultaneous existence of unlled vacancies and searching workers is one of the most pervasive and welldocumented features of modern labor markets. We assume that at the beginning of a period, all potential workers L j in country j search for jobs. Firms post V j vacancies. The number of realized matches is given by the constant returns to scale matching function M j = m j L µ j V 1 µ j, where µ (0, 1) is the elasticity of the matching function and m j measures the eciency of the labor market. 10 Let θ j V j /L j denote the degree of labor market tightness in country j. Then, the share M j /V j = m j (V j /L j ) µ = m j θ µ j all vacancies is lled at the end of the period, when all output is realized. Similarly, the share M j /L j = m j (V j /L j ) 1 µ = m j θ 1 µ j of all workers is employed; the remainder is without a job at the end of the period, so that the unemployment rate is given by u j = 1 m j θ 1 µ j. We assume that the matching eciency is suciently low to assure that job nding rates and job lling rates are strictly within zero and one at any time. Let γ j (0, 1) denote the unemployment benet replacement rate and w j the wage rate in of 8 To save notation we write p i instead of p ii. Similarly for other variables. 9 See Keuschnigg and Ribi (2009) or Helpman and Itskhoki (2010) for recent models using a similar static approach. 10 Constant returns to scale is a common assumption in the search matching context. It has received ample support in empirical studies; see Petrongolo and Pissarides (2001). 5

9 country j. Then, unemployment benets are given by b j = γ j w j. As usual, wages are determined in a generalized Nash bargaining process after matching has occurred. The total surplus from a successful match is given by p j b j, while the rm's rent is given by p j w j and the worker's by w j b j. Let the worker's bargaining power be β j (0, 1). Nash bargaining implies w j b j = β j (p j b j ).Together with b j = γ j w j, 11 one obtains the wage equation w j = β j 1 (1 β j ) γ j p j. (3) This equation illustrates the advantage of assuming one-shot matching as it does not depend on θ j. 12 The worker appropriates a portion of the value of output and w j < p j. The bargained wage increases in the value of output p j, in the worker's bargaining power β j (since γ j (0, 1)) and in the replacement rate γ j. Firms create vacancies until all rents are dissipated. The free entry (zero prot) condition is given by M j /V j (p j w j ) = P j c j, where c j is the cost of posting one vacancy, which is in terms of the nal good. It states that expected rents (the probability of lling a vacancy times the rent per lled vacancy) have to be equal to the upfront cost of creating the vacancy. Rewriting, one nds the job creation curve w j = p j P jc j m j θ µ. (4) j As the wage curve, this curve is increasing in the value of output. It is decreasing in the expected costs of lling a vacancy P j c j /(m j θ µ j ), which is, in turn, an increasing function of θ j : as labor market tightness goes up, the probability of lling a vacancy falls. Using the job creation curve and the wage equation, one can express labor market tightness as a function of relative output prices and model parameters ( ) 1/µ ( ) 1/µ θ µ j = pj cj Ω j, (5) P j m j 11 We assume that workers and rms take the level of unemployment benets as exogenously given; i.e., they do not take into account that a higher wage would imply higher unemployment benets. 12 In the dynamic specication, the wage rate would be a positive function of labor market tightness as forwardlooking agents factor in a lower duration of unemployment in the case negotiations break down. 6

10 where Ω j := 1/(1 β j) γ j 1 γ j 1 summarizes the eective bargaining power of workers. Ω j is increasing in the worker's bargaining power β j and in the replacement rate γ j. A decrease in m j, an increase in Ω j or an increase in the cost of vacancy posting c j lowers labor market tightness and increases the unemployment rate. An increase in the relative price of output p j /P j makes vacancy creation more worthwhile and therefore leads to higher labor market tightness and lower unemployment. Note that equilibrium unemployment in country j is not directly aected by variables of country i but only indirectly via the relative price p j /P j. 2.3 Terms of trade Employing the denition of the aggregate price index, dening Home's terms of trade as π p H /p F and assuming τ = τ ij = τ ji if i j and τ = 1 if i = j, it is straightforward to express the relative price p H /P H as a function of terms of trade π: p H /P H = [ 1 + (π/τ) σ 1] 1/(σ 1). (6) Hence, Home's terms of trade π and the relative price p H /P H are positively related. The larger trade costs τ are, the weaker is that link as the share of imported goods in Home's price index goes down. A similar relation exists for p F /P F, which is of course negatively related to π. Using equation (5) one can express Home's labor market tightness as θ H = [1 + (π/τ) σ 1] ( ) 1 1/µ µ(σ 1) ch Ω H. (7) m H This shows that, ceteris paribus, an increase in Home's terms of trade (π) yields a higher degree of labor market tightness θ H and hence lower unemployment. This eect is lower, the higher trade costs τ are. If Home increases the eective bargaining power of workers (Ω H ) or the costs of vacancy creation (c H ), its labor market tightness falls and, thus, the equilibrium unemployment rate goes up. A similar relation exists for Foreign: θ F = [1 + (πτ) 1 σ] ( ) 1 1/µ µ(σ 1) cf Ω F. (8) m F 7

11 Again, an increase in Foreign's terms of trade (1/π) improves labor market outcomes there; the eect is lower, the higher τ. 2.4 Relative demand and relative supply In the next step of the analysis, we discuss how Home's labor market institutions aect the terms of trade. For this purpose, we derive the relative demand and the relative supply schedules for the world economy. Using S j = m j L j θ 1 µ j relative to the supply of Foreign's input is given by and equations (7) and (8), the supply of Home's input RS (π) S H S F [ = λb (1 µ) µ 1 + (π/τ) σ (πτ) 1 σ ] 1 µ µ(σ 1), (9) where λ L H /L F is relative country size and B (c H Ω H / (c F Ω F ))(m F /m H ) 1/(1 µ) measures the relative importance of labor market frictions at Home as compared to Foreign. Clearly, relative supply (RS) is strictly increasing in π since σ > 1; see Appendix A1 for detailed derivations. Relative demand (RD) can be found by using (2) and imposing the balanced trade condition p F H D F H = p HF D HF. The RD locus is given by RD (π) D H D F 1 2σ 1 + (π/τ)σ 1 = π. (10) 1 + (πτ) 1 σ Clearly, due to identical and homothetic preferences, relative demand depends only on relative prices and trade costs. It does not directly depend on variables that aect Home's income relative to Foreign's. In particular, labor market related variables do not play a role here. Relative demand takes a particularly simple form if there are no trade costs (i.e., if τ = 1). Then, RD (π,.) = π σ, and the slope of the locus in π is straightforwardly signed as negative. Strictly positive trade costs complicate the picture, but only reinforce the negative slope of the RD schedule as detailed in Appendix A1. 8

12 3 Theoretical Insights 3.1 Trade liberalization and unemployment Before we turn to the comparative statics with respect to labor market institutions, it is worthwhile to investigate the eects of trade liberalization. Proposition 1 (Trade liberalization and unemployment.) In a neighborhood of the symmetric equilibrium, a decrease in iceberg trade costs leads to a decrease in equilibrium unemployment rates in both countries. Proof. In Appendix A2. Trade liberalization modeled as a reduction in iceberg trade costs has fairly complicated eects on the terms of trade. However, starting in the symmetric equilibrium where π = 1, a change in τ leaves π unchanged, which makes perfect sense because a cut in trade costs aects both countries equally and thus cannot imply a deviation from symmetry (the proof is in Appendix A2). In that case, Proposition 1 follows directly from equations (7) and (8). The intuition for this result is simple. When τ falls, given π, the price of the nal good (the price index) in both countries has to fall, since imports become cheaper. Therefore, p j /P j goes up and rms in both countries nd that the price of output relative to the costs of vacancy creation has increased. As a consequence, they create more vacancies (see equation (5)). The equilibrium unemployment rate goes down. If countries are asymmetric so that π > 1 in the initial equilibrium (e.g., because Home is smaller than Foreign, λ < 1), both relative demand RD and relative supply RS shift upwards with lower τ. Making the reasonable assumption that σ > 1/µ, the exponent in equation (9) is smaller than one, implying that relative demand increases by more than relative supply. 13 In order to establish a new equilibrium, π has to increase, as relative demand is decreasing and relative supply increasing in π. In that case, Home (the smaller country) benets by more from falling trade costs than in the symmetric case (π = 1) as terms of trade move in its favor. Foreign 13 The condition σ > 1/µ requires that the matching elasticity µ be larger than the elasticity of inverse demand 1/σ. For the standard value of µ = 1/2 (see Hosios, 1990), σ needs to exceed 2, which is a very realistic assumption (see Anderson and van Wincoop, 2004). 9

13 (the larger economy), in turn, benets by less and it is theoretically possible that the terms of trade deterioration osets the gains from lower trade costs. 3.2 Institutional spillovers We now come to the core interest of this paper, namely how a change in labor market institutions in one country aects the other country. To start, note that the parameter B in RS summarizes all labor market variables in the model: search costs c j, bargaining power β j, the eciency of the labor market m j and the generosity of unemployment insurance γ j. An increase in B signals that labor market frictions in Home have increased relative to Foreign. So, without loss of generality one can focus on any labor market variable, or on the summary statistic B. The following proposition summarizes the international spillover eects of labor market frictions. Proposition 2 (Institutional spillovers). An increase of labor market frictions in Home leads, ceteris paribus, to a decrease in labor market tightness in both Home and Foreign. Hence, unemployment goes up in both countries. Proof. In Appendix A3. In terms of elasticities, the labor market eects of Home's institutions are given by: ε θh,b = 1 µ + ε θ H,πε π,b < 0, (11) ε θf,b = ε θf,πε π,b < 0, (12) where ε π,b measures the eect of Home's institutions on the terms of trade and ε θj,π measures the eect of the terms of trade on labor market tightness. The intuition for the results is as follows. For given relative prices, an increase in the importance of labor market frictions in Home relative to Foreign leads to a reduction in Home's labor market tightness and hence to higher unemployment (equation (7)). At given π, the elasticity of Home's tightness with respect to B is just 1/µ, the inverse of the elasticity of the matching function. Higher unemployment in Home obviously reduces the supply of Home's good. This, in turn, implies that Home's terms of trade improve π/ B > 0, which works towards an increase in labor market tightness 10

14 (equation (7)), thus counteracting the direct eect. However, this positive indirect eect can never dominate the direct negative eect. Foreign is not directly aected by a change in Home's labor market institutions. The indirect eect is negative since its terms of trade deteriorate. 14 The important insight from Proposition 2 is that, in an open economy setup, changes in labor market institutions in one country have eects on labor market outcomes not only in that economy but in its trade partners as well. Through the adjustment of terms of trade, the unemployment costs of more severe labor market frictions are partly exported, so that the trading partner takes over part of the unemployment increase. Conversely, the benets of labor market reforms are exported as well so that the trading partners share the benets of reforms. This fact has immediate implications for the political economy of labor market institutions: without coordination, countries will have insuciently low incentives to reform institutions. In this paper we do not pursue this possibility further; see Felbermayr, Larch and Lechthaler (2012) for an analysis in a Melitz (2003) environment with search frictions on the labor market. 3.3 The roles of relative country size, trade costs, and wage rigidity In this section, we are interested in the eect of a change in relative country size (λ) and transportation costs (τ) on the spillovers caused by labor market reforms in the Home country, i.e., we want to understand how the elasticities ε θi,b depend on λ and τ. Moreover, we also highlight the role of wage exibility for the magnitude of spillovers The role of country size Proposition 3 (Institutional spillovers and country size.) In a neighborhood of the symmetric equilibrium, the higher the relative size of Home, the stronger is the increase in unemployment in all countries due to more severe labor market frictions in Home. Proof. In Appendix A4. 14 The ipside of the terms of trade eect we have just described is an income eect. The increase in unemployment in Home decreases income in Home and thus reduces demand for Foreign's products. Foreign's unemployment goes up and, thus, Foreign's income goes down too. 11

15 To understand the proposition, we discuss the elasticities in (11) and (12) separately. The elasticities ε θh,π and ε θf,π both become unambiguously smaller when Home becomes bigger. The reason is that Home becomes more important for demand in both countries and thus Home's unemployment depends less on the terms of trade, while Foreign's unemployment depends more on the terms of trade (remember that ε θf,π < 0). However, it is not easy to tell whether ε π,b becomes bigger or smaller with country size. From the expression given in Appendix A3 it is immediately clear that country size does not have a direct eect on ε π,b. However, country size can have an indirect eect via its eect on the terms of trade. It is easy to show that larger countries have lower terms of trade, i.e., π/ λ < 0. But the eect of the terms of trade on the elasticity ε π,b is ambiguous and depends on country size and other model parameters. However, in the neighborhood of the symmetric equilibrium (where π = 1), it can be shown that ε π,b / π = 0 and, thus, ε π,b / λ = 0: A change in the terms of trade (and thereby country size) has no rst-order eect on the elasticity ε π,b. This does not mean that labor market institutions do no longer have an eect on the terms of trade, but that this eect is constant with respect to changes in the terms of trade at the symmetric equilibrium. It follows that in a neighborhood of the symmetric equilibrium, the eects of country size on ε θi,π dominate. Thus, if Home becomes bigger, more severe labor market frictions in Home trigger a larger increase in unemployment in both countries The role of geography Proposition 4 (Institutional spillovers and geography.) In a neighborhood of the symmetric equilibrium, and if µ > 1/σ, higher trade costs decrease the importance of spillovers. So, more severe labor market frictions in Home increase Home's unemployment rate by more and Foreign's unemployment rate by less. Proof. In Appendix A5. It is easy to show that the eect of a change in π on Home's labor market tightness (ε θh,π > 0) is weaker the higher trade costs are (i.e., ε θh,π/ τ < 0). This is intuitive since a change in 12

16 relative world market prices has a smaller domestic impact when trade barriers are strong. The same logic holds true for Foreign, where ε θf,π < 0. That elasticity is reduced in absolute value, i.e., terms of trade movements have a lower impact on Foreign's unemployment (i.e., ε θf,π/ τ > 0). It is more dicult to ascertain the eect of higher trade costs on the elasticity ε π,b, the eect of B on the terms of trade. Two eects have to be distinguished. On the one hand, trade costs directly aect the elasticity ε π,b. On the other hand, trade costs potentially have an eect on the terms of trade, which in turn can have an eect on the elasticity. As in the previous subsection, we focus on the symmetric equilibrium (where π = 1) to get clear-cut results. In this case the second, indirect eect vanishes altogether because trade costs do not aect the terms of trade. The rst, direct eect depends on the relative size of the parameters σ and µ. The crucial question is whether trade costs make relative supply (equation (9)) or relative demand (equation (10)) relatively more sensitive to changes in the terms of trade. Assuming again that σ > 1/µ, rising trade costs imply a smaller impact of labor market institutions on relative supply than on relative demand. This in turn implies that the terms of trade react less strongly to changes in labor market institutions, i.e., ε π,b / τ < 0. It follows that Foreign is less aected by Home's labor market institutions, if trade costs are larger. On the one hand, an increase in labor market frictions at Home has a smaller impact on the terms of trade. On the other hand, the terms of trade become less important for Foreign's unemployment rate. Exactly the same relationship holds for Home, but for Home this means that unemployment depends more strongly on its own labor market institutions and so an increase in labor market frictions increases unemployment by more The role of wage rigidity Finally, we take a look at the role of wage rigidity for the magnitude of the spillover eect. We do not intend a fully edged comparative statics analysis, because, due to the one-shot nature of our labor market, the theoretical model does not feature a simple exogenous parameter that governs the sensitivity of wages with respect to, say, unemployment. However, one can simply assume that real wages do not adjust at all in both countries. 13

17 Proposition 5 (Institutional spillovers and real wage rigidity.) The more rigid wages are, the more moderate is the adjustment in the terms of trade π. However, for a given change in π, the reaction in Foreign's tightness and unemployment is stronger if real wages are rigid. Proof. In Appendix A6. In this price-constrained economy, a cost shock yields a stronger reaction in quantities than in prices. So relative employment falls by more than in the exible case. While the terms of trade eect of a change in Home's labor market institutions is more moderate under rigid wages, it is also clear that the quantity adjustment in each economy needs to be larger. Any shock be it the initial B shock in Home or the resulting terms of trade shock in Foreign translates into a stronger quantity adjustment on labor markets. Under exible wages the increase in the terms of trade will lead to a lower wage in Foreign, thus dampening the negative eect of the terms of trade movement on unemployment. If wages are rigid, this dampening eect does not take place and thus the reaction in unemployment has to be stronger the more rigid wages are. 4 Empirical evidence In this section, we use panel data on labor market institutions and unemployment rates for 20 rich OECD countries for Our aim is to check whether the empirical evidence is in line with the ve key predictions of our model, namely: (1) lower trade costs (higher openness) lead to lower unemployment in Home and Foreign; (2) higher search costs or equivalently more generous unemployment benets or a higher bargaining power of workers in one economy leads to an increase in equilibrium unemployment rate in all economies; (3) the larger the relative size of Home, the stronger are the spillovers of Home's labor market institutions on Foreign; (4) the larger trade costs, the smaller are spillovers; (5) the more rigid real wages are, the stronger are spillovers. Note that our theoretical model implies that changes in search costs or unemployment benets have similar eects both domestically and in trading partners. Because the generosity of unemployment benets is more easily measured and data is available for a larger number of countries, we focus on them in the subsequent exercise. 14

18 4.1 Econometric specication Our starting point is the standard cross-country unemployment regression framework: 15 u it = λ LMR it + π P MR it + δ OP EN it + γ GAP it + F i + T t + S it + ε it, (13) where the vector LMR it collects labor market variables such as union density, the degree of corporatism, employment protection legislation (EP L it ), a proxy for the ow value of nonemployment (b it ) and a proxy for the degree of real wage exibility (F LEX it ). To cover product market characteristics, we include the variable P MR it which measures the intensity of product market regulations and an indicator of openness to trade (OP EN it ) computed as imports plus exports over GDP in constant prices. GAP it refers to the output gap. 16 F i and T t are country and year eects, respectively, while S it controls for exogenous shocks (total factor productivity (T F P ), real interest rates, terms of trade (T OT ), and labor demand shocks). The construction of the latter variables is detailed in Bassanini and Duval (2006) and is in line with common practice in the literature. Inclusion of macroeconomic variables and shocks is crucial to control for business cycles so that estimated coecients relate to structural (or equilibrium) unemployment rates. 17 The error term ε it is assumed to have the usual properties. Bassanini and Duval (2006) do not survey a single study which would address the possibility that the foreign rate of unemployment or foreign labor market regulations might matter for domestic labor market outcomes. The existing literature has found robust and quantitatively relevant unemployment eects only for a very limited number of labor market institutions. The most important is the participation tax rate, or tax wedge (i.e., the sum of the average wage tax burden and social benets; see Costain and Reiter, 2008). This will be our preferred measure of b it. Other measures relating to the nature of wage bargaining, employment protection legislation, or the prevalence of minimum wages receive mixed empirical support. This is not necessarily surprising, given the ambiguity of theoretical results (see, e.g., the discussion in Blanchard and 15 See the survey by Bassanini and Duval (2006, 2009). 16 Calculated as the dierence between actual output and the HP-ltered series. 17 Due to the data limitations, using ve-year averages instead of yearly data is dicult. In the robustness checks, we provide some estimates based on three-year averages. 15

19 Wolfers, 2000). Hence, we focus on the tax wedge (b it ) and include other controls in robustness checks. Costain and Reiter (2008) argue that the tax wedge can be considered as exogenous in the present context. We also work with a measure of real wage exibility based on Holden and Wulfsberg (2009). Using data on hourly earnings at the industry level, those authors have measured the frequency of real wage cuts for a sample of 19 OECD countries from They nd that real wage exibility is less prevalent in countries with strict employment protection legislation and high union density (p. 605). Unfortunately, their country and time coverage is not identical to ours. Since the data from Holden and Wulfsberg (2009) are the only recent source of comparable timeseries information on wage rigidity, we model the likelihood of real wage cuts as a function of observable variables (union density, employment protection legislation, the output gap) as well as time and country eects. Including year and country dummies into the regression, we infer the missing data. Our theoretical model gives rise to a gravity-type relation between bilateral trade volumes and explanatory variables related to countries' market size (Proposition 3) and bilateral trade costs (Proposition 4). It predicts that the eect of labor market regulations of some country j on country i s rate of unemployment is conditioned by the amount of bilateral trade between the two countries. We proxy the amount of bilateral trade between i and j by the strictly exogenous measure ω ijt = P OP α 1 it P OP α 2 jt DISTij δ, (14) where P OP it denotes population of country i, DIST ij is geographical distance; 18 α 1, α 2, and δ are parameters. ω ijt varies with time as population changes. 19 Standard gravity predictions suggest that α 1 = α 2 = 1. In their meta analysis, Disdier and Head (2008) nd that the mean elasticity of distance is about 0.9, with 90% of estimates lying between 0.28 and Hence, we choose δ = 1 as our benchmark case, but conduct robustness checks with respect to the 18 Great circle distance between the two countries' most populated cities 19 We have also worked with predicted bilateral trade volumes obtained by regressing observed bilateral trade on exogenous variables such as population, distance, and other typical covariates such as common language, contiguity, joint membership of countries in currency unions, free trade areas, and GATT/WTO. Results are qualitatively and quantitatively comparable. 16

20 assumptions on α 1, α 2, and δ. We calculate ω ijt for all 168 countries for which population and distance data are available (i.e., not only the 20 OECD countries for which we have reliable labor market data). There are several possible ways to normalize the data; the choice of normalization has interpretational consequences but should not aect our qualitative ndings. In our preferred setting, we normalize the weights such that 168 j=1 ω ijt = 1 for all 168 countries. Then, we construct the trade-weighted average of foreign unemployment rates, u it = 20 j=1 ω ijtu jt, where country i s rate of unemployment is excluded by denition (ω iit = 0) and the summation only involves the 20 OECD countries for which high-quality unemployment rates are available. Similarly, we construct the trade-weighted average tax wedge of all countries other than i as b it = 20 j=1 ω ijtb jt (and similarly for all other labor market variables LMR it, denoted by LMR it), and the average foreign output gap as GAP it = 20 j=1 ω ijtgap jt. 20 With the exception of the real wage exibility measure of Holden and Wulfsberg (2009), labor market and macro variables come from Bassanini and Duval (2006). This dataset reects intensive eorts at the OECD to come up with harmonized measures. Unfortunately, it covers only 20 countries 21 for the years 1982 to Data on geographical distance come from CEPII. 23 Population data is from the Penn World Tables mark 6.2. Table 1 provides summary statistics. 4.2 The role of domestic and foreign institutions As a rst step, we show that our data replicates the typical results found in the empirical literature. Column (1) in Table 2 shows the results of estimating (13) using OLS. The coecient 20 A natural alternative normalization would set the weights such that k K ω ikt = 1, where K is the set of our 20 OECD countries. One could also normalize weights by max j ω ijt. In a series of robustness checks, we will show that the choice of normalization has little qualitative eect on our results. 21 Australia (AUS), Austria (AUT), Belgium (BEL), Canada (CAN), Switzerland (CHE), Germany (DEU), Denmark (DNK), Spain (ESP), Finland (FIN), France (FRA), Great Britain (GBR), Ireland (IRL), Italy (ITA), Japan (JPN), Netherlands (NLD), Norway (NOR), New Zealand (NZL), Portugal (PRT), Sweden (SWE), and the United States of America (USA). 22 For larger country samples, data on unemployment rates and institutions are scarce and data lack comparability across time and space

21 Table 1: Summary statistics Variable Description Mean Std. Dev. Min Max u unemployment rate (percent) b tax wedge (percent) F LEX real wage exibility (index) Union density (percent) High corporatism (dummy) EP L employment protection legislation (index, 1-10) P MR product market regulation (index, 1-10) OP EN (percent) u W u b W b F LEX W F LEX Union density W Union density High corporatism W High corporatism EP L W EP L P MR W P MR OP EN W OP EN GAP output gap (percent) GAP W GAP P OP Population (in mio) P OP W P OP k /k (W (K/L))/(K/L) T F P shock T OT shock Real interest rate shock Demand shock All data are from Bassanini and Duval (2006), except F LEX which is from Holden and Wulfsberg (2009) and the weighting matrix W which comes from own calculations. Number of observations N = 397. Weights are based on α 1 = α 2 = 1 and δ = 1; standard normalization. Foreign variables are not to be interpreted as means, since weights do not add up to 1 (due to inclusion of rest of the world in calculation of weights). 18

22 on the tax wedge b, our key labor market variable of interest, implies that a 20 percentage point increase in b (approximately equal to one standard deviation of b in the data) increases the equilibrium rate of unemployment by about 1.8 percentage points, conrming parts of Proposition 2. Also the degree of real wage exibility (F LEX) is strongly correlated with unemployment. An increase in the exibility index by one standard deviation (0.35) lowers unemployment by about 1.4 percentage points. Union density, employment protection legislation (EP L), and the degree of corporatism do not have any measurable inuence on equilibrium unemployment. This is a standard nding; see Bassanini and Duval (2006) and the references therein. Finally, the output gap (GAP) is an important determinant of the unemployment rate. Note that country eects, year eects, the output gap, and macroeconomic shocks alone explain about 87 percent of the total variation (adjusted R 2 ) of unemployment rates in our sample (not shown). Hence, timevariation in labor market institutions has a limited role in explaining variance in unemployment rates. Next, we study the eect of trade openness on the rate of unemployment. Proposition 1, derived from our model, suggests a negative relation. Column (2) in Table 2 therefore introduces OP EN into the regression. It also introduces the OECD index of product market regulation which strongly correlates (negatively) with OP EN but accounts for domestic product market distortions as well as for foreign entry regulation. OP EN turns out statistically signicant and with a negative sign. The estimate of implies that a one standard deviation increase in OP EN (30.7) lowers the structural rate of unemployment by about 1.4 percentage points. Recent empirical literature discusses the potential endogeneity of OP EN in unemployment regressions. 24 This literature does not nd evidence that OLS overestimates the true causal eect of OP EN. They also come up with point estimates that very much resemble the one found in column (2). However, in the remainder of this paper we still choose to instrument OP EN. For this purpose, column (3) uses the fth to tenth lags of OP EN as well as similarly lagged rst dierences as instruments. The Sargan score test is insignicant, so that the IV strategy appears valid. The Durbin-Wu-Hausman test on exogeneity rejects that OP EN can be treated as an exogenous variable, but only at a very marginal level of signicance See Dutt, Mitra, and Ranjan (2009). 25 Instrumentation of openness has very little eect on the estimates of λ, our parameter of interest. Results of 19

23 The next regressions analyze the direct eect of foreign labor market institutions on the domestic rate of unemployment in order to check Proposition 2. We estimate an equation of the form: u it = λ LMR it +λ LMR i,t+π P MR it +δ OP EN it +γ 1 GAP it +γ 2 GAP it+f i +T t + S it +ε it, where LMR it collects foreign labor market variables and GAPit is the foreign output gap. Column (4) in Table 2 shows the most parsimonious specication, where we include only the domestic and the foreign tax wedges (b, b ) as well as the controls for the domestic and the foreign business cycles and the complete set of xed eects. We nd that the own and the foreign tax wedges help explain the domestic unemployment rate. Both have coecients with the signs predicted by our theoretical model and are accurately estimated: a marginal increase in the domestic wedge has an eect about 9 times as large as a marginal increase in the foreign wedge. A one-standard-deviation increase in the domestic or foreign wedge adds about 1.5 ( ) or 0.2 ( ) percentage points to the equilibrium unemployment rate, respectively. This conrms Proposition 2. Column (5) adds OP EN (instrumented as above) and the OECD measure of product market regulation. These additional variables show up with the expected signs and increase the adjusted R 2 by about one percentage point. Qualitatively, the unemployment increasing eects of the domestic and foreign wedges are unaected by this, but the relative importance of foreign institutions grows. This may be an indication that product markets matter for the transmission of foreign institutions. Columns (6) and (7) add more domestic and foreign labor market institutions to the regression. Not surprisingly, adding variables for which the direct eect on Home unemployment is already inconclusive (union density or EP L), does not improve accuracy of estimation. The coecients on b become insignicant, others seem implausibly large (see the coecient on EP L for the most striking case). Hence, the lack of a robust relationship between these variables in standard equations such as (13) also impairs inference when using their spatial lags. Moreover, (15) OLS regressions are available upon request. 20