Laura Alfaro Maggie X. Chen

1. Selection, Reallocation, and Spillovers:Identifying the Sources of Gains from Multinational Production Laura Alfaro Maggie X. Chen Harvard Business School George Washington University & NBER & World Bank May 2012

2. Introduction • Nations with greater multinational activity exhibit, on average, higher productivity. (Borensztein et al. 1998, Alfaro et al. 2004, Harrison and Rodriguez-Clare, 2010, Kose et al. 2010;).

3. Introduction • This positive correlation, likely conditional on other factors, is often attributed to knowledge spillovers whereby foreign multinationals generate positive productivity externalities to domestic firms: • Knowledge transfer through partnerships, interaction and movement in labor markets, etc. • There is a less stressed, alternative explanation, centering on firm selection. • Selection of multinational firms: Countries with greater openness to multinational activity to attract firms that are, by selection, more productive (Helpman et al., 2004). • Selection of domestic firms: Greater multinational activity leads to tougher competition and market reallocation, only the most productive domestic firms to survive (Melitz, 2003).

4. Introduction • These mechanisms all imply a positive relationship between multinational production and greater host-country productivity but different causalities. • Selection of multinational firms: Higher host-country productivity reflect the productivity of self-selected multinational firms. • Selection of domestic firms and knowledge spillovers imply multinational activity causes higher aggregate domestic productivity. • Tougher selection means a contraction of domestic production, • Technology spillovers represent positive externalities.

5. Objective of thePaper • Disentangle and quantify the relative magnitude and statistical significance of the firm selection and knowledge spillover effects: • Crucial for identifying the sources of productivity gains and setting effective FDI and industrial policies. • Ignoring sources can lead to significant bias in understanding the nature of welfare and productivity gains and anover-estimation of the importance of knowledge spillover. • If increases in productivity are due to knowledge spillovers, special treatment to foreign multinationals may be justified; • If increases in productivity are due to tougher selection on domestic firms, policies should aim at improving human capital, and financial market conditions, facilitating gains from competition and resource reallocation.

6. Methodology • Distinguishing these mechanisms is difficult by simply examining the relationship between multinational activity and average productivity. • We develop a model of monopolistic competition and heterogeneous firms, adapted from Melitz (2003) and Helpman et al. (2004), and a structural empirical framework exploring the distinct predictions for the distributions of domestic firms: • Selection of multinationals: Firms with greater ex-ante productivity self-select into multinational production. • Selection of domestic firms: Competition from multinationals leads to market reallocations and an increase in the cutoff productivity and revenue (greater left truncation of the distributions); • Knowledge spillovers: rightward shift of the productivity distribution, while the revenue distribution sees a weaker, or even leftward, shift.

7. Data and Estimation • Worldwide firm-level dataset: Orbis • Comprehensive financial, operation, and ownership information for over one million public and private manufacturing firms, 2002-2007. • Identify multinational activity across countries and explore the heterogeneous effect of foreign investments (manufacturing firms). • Structural estimation consists of two steps: • Estimate the selection of multinational firms as a function of their ex-ante productivity, host-country and bilateral factors. • Evaluate the effect of multinational activity on various properties of domestic production and the productivity distribution.

8. PreliminaryFindings: Selection, Reallocation, Knowledge Spillover Effects • Tougher selection for the domestic firms. • Entry of multinationals raises the cutoff productivity of domestic firms; least productive domestic firms more likely to exit. • Increased competition and market reallocations. • Increase in the minimum revenue of continuing domestic firms (implying increase in fixed production costs and financing constraints). • Domestic firms’ revenue distribution shifts leftward (25th and 50th). • Domestic firms’ productivity distribution shift rightward; distribution becomes more left truncated due to selection. • Knowledge Spillovers: • Surviving domestic firms productivity increases at different percentiles.

9. PreliminaryFindings: ProductivityGains • Aggregate productivity increases by 1.4 percent across countries when the probability of entry by new multinational firms increases by 100 percent. • The productivity of domestic firms increases by 0.9 percent, • Knowledge spillover and domestic selection account for 69 and 31 percent, respectively. • The relative importance of each source exhibits significant country heterogeneity. • Firm selection and market reallocation are, on average, more important in developed nations.

10. RelatedLiterature: Spillovers and Selection • Studies assessing the productivity spillovers from multinationals to domestic firms: • Aitken and Harrison (1999) find evidence of negative spillovers in Venezuelan manufacturing enterprises (market-stealing effect). • Keller and Yeaple (2009) show strong evidence of positive spillovers from foreign multinational to domestic firms in the United States. • Javorcik (2004) shows multinational activity leads to positive externalities via backward production linkage. • Arnold and Javorcik (2009), and Guadalupe at al. (2001) find foreign multinationals to acquire best performing domestic firms, explaining some but not all of the performance differential.

11. RelatedLiterature: Factor Markets/ CompetitionEffect • Evidence on the domestic selection effect of multinational activity and the relative importance of spillovers and selection is scarce. • Ramondo (2009) finds foreign plants' entry to be negatively correlated with the market shares of domestic firms and positively correlated with the productivity of domestic incumbents in Chile. • A few studies have taken the step to evaluate MNCs’ factor market effects . • Labor: Aitken, Harrison, and Lipsey (1996) and Feenstra and Hanson (1997) find foreign multinational activity to increase industry wages and share of non-production workers in wage. • Capital: Harrison and McMillan (2003) find borrowing by foreign firms exacerbates the credit constraints of domestic firms. Harrison, Love and McMillan (2004) find FDI inflows to be associated with a reduction in financing constraints using a larger cross-country dataset.

12. RelatedLiterature • Recent studies evaluating the welfare effects of multinational production (Ramondo and Rodriguez-Clare , 2010; Irrarazabal, Moxnes, Opromolla, 2011), and focus on the productivity effect of resources allocation across establishments (see, Hsieh and Klenow, 2009; Alfaro et al, 2009). • Our paper is an effort to distinguish and quantify the roles of selection, reallocation and knowledge spillovers in the aggregate productivity effect of multinational production. • Micro theoretical foundation to develop an empirical strategy that is able to distinguish their relative importance; • Structural framework to quantify the magnitude of productivity gains associated with each effect; • Cross-country analysis to quantify and evaluate how the knowledge spillover and selection effects may vary systematically across nations.

13. Outline • Introduction and Motivation • Theoretical Framework • Data • EmpiricalMethodology • Results • Conclusions

14. Theoretical Framework: Setup Standard model of monopolistic competition with heterogeneous firms (Melitz (2003) and Helpman et al. (2004)). • Two identical countries, H and F; two sectors, homogeneous (numeraire) and differentiated. • Continuum of firms in each country. Each firm produces a different variety of the differentiated product and has a distinct productivity level θ. • Given a CES utility function, the demand function for each variety of the differentiated product: • x(θ) = the quantity of demand; p(θ) = price of the product variety, • E = the aggregate expenditure; P = the aggregate price

15. Theoretical Framework: Production,DomesticFirms • If firm θ of country H chooses to produce and sell at home, it must employ one unit of labor for each unit of output and incur marginal cost w/θ, (w= common wage rate.). • The firm must pay a per-period fixed cost cfD, • c denotes unit capital price • fDdenotes the units of capital required in the production. • The profit-maximizing strategy is to set p(θ)= w/(αθ)

16. Theoretical Framework: Production, ForeignFirms • Firms of country F may invest and produce in country H to serve country H's consumers via multinational production. • Foreign multinational firms must pay a fixed costcfMin each period, • fM> fD. • The profit earned by the foreign firm in country H:

17. Theoretical Framework: ProductionDecisions, CutoffProductivity • Domestic firms produce in the domestic market if πD(θ) ≥ 0, which gives the cutoff productivity level θD for domestic firms to survive: • Foreign firms invest and produce in the domestic market if πM(θ) ≥ 0; which gives the cutoff productivity level for foreign firms: • θM> θD, minimum productivity to survive is higher for foreign firms (fM>fD).

18. Theoretical Framework: KnowledgeSpillovers • Productivity of domestic firms is assumed to be a function of two components: • Raw productivity θa drawn from a distribution function G(θa) • Slope parameter τθ(zM) • zM: indicator variable that denotes the existence of foreign multinational production.

19. Model: TheImpact of MultinationalProductionTheSelection of DomesticFirms • Openness to FDI induces an increase in the domestic cutoff productivity level θD , the least productive firms no longer earn positive profits and exit. • Domesticcutoffproductivity, (A denotes ex-ante, before MNC entry) • ExitsforA<<D • Theeffectsoperatethroughdomestic factor markets: increased factor demandbymultinationalfirmsbids up the real wage and capital price. • Revenue: • Increase in the average productivity and in the number of firms serving the market: a decrease in the aggregate P and on revenues, while the spillover from foreign firms exerts a positive effect. • In the absence of productivity spillovers, all firms incur a loss in domestic sales in the open economy.

20. Data • We employ a cross-country firm-level panel dataset, drawn from Orbis, that contains comprehensive financial, operation, and ownership information for over 1 million manufacturing firms in 60 countries. • We estimate total factor productivity using the semiparametric estimator developed by Olley and Pakes (1996). • The production function is estimated for each country and each NAICS 4-digit industry; • We consider two sub-periods: 2002-2004 and 2005-2007 and investigate how changes in multinational activity between the two periods affect host-country domestic firms. • A firm is considered foreign owned if its global ultimate owner is based in a different country. There are about 36,000 foreign owned manufacturing subsidiaries in the final sample.

21. EmpiricalEvidence: TheSelf-Selection of Multinational Firms • Estimate the following equation: • zM(θ) represents foreign multinationals' binary decision to enter a given host country in 2005-2007, • θ is the ex-ante productivity of multinational firms (estimated based on headquarters activities in 2002-2004) • FEM is a vector of host country-industry dummies, • D bilateral country factors including distance, common border, and common language between headquarters and host countries. • Ex-ante productivity of foreign firms are expected to have an effect on the decision to participate in multinational activity but, unlikely to be directly correlated with the future productivity of host-country firms, Helpman et al. (2004): Exclusion conditions in the second-stage estimations to identify multinational production effects.

22. More productive firms exhibit a greater likelihood of entering foreign countries, consistent with Helpman et al. (2004).

23. Multinational activity exerts, on average, a positive and significant effect on the average productivity of domestic firms. Traditionally, it has been assumed that this productivity effect is associated to knowledge spillovers. But is the gain due to knowledge spillovers, selections, or both?

24. EmpiricalEvidence:TheSelf-Selection ofDomestic Firms • Survival of individual domestic firms by estimating • zD(θ) whether the domestic firm continues production in 2005-2007, • θA is the lagged productivity of the domestic firm, (βZ :cumulative effect of new multinational entry on the survival probability of domestic firms, including the positive knowledge spillover effect and the adverse effects on financing costs and aggregate price). • zM is an indicator for new multinational entry. • Country and industry dummies to control for country and industry factors and country-industry clustering to allow for correlations within each cluster. • To account for the endogeneity of zM, we substitute ^γM obtained.

25. Domestic firms are more likely to exit the market in the presence of new multinational entry.

26. EmpiricalEvidenceTheSelection of DomesticFirms: CutoffProductivity • Higher probability of multinational entry leads to a significant increase 16% of the cutoff.

27. EmpiricalEvidence: Labor MarketReallocation • A higher likelihood of multinational entry leads to a significant decrease in the level of revenue for firms at both the 25th and 50th percentiles (the relatively smaller domestic firms see a bigger contraction in their revenue).

28. EmpiricalEvidence: Capital MarketReallocation • A higher probability of multinational entry leads to a significant increase in the cutoff revenue of domestic firms.

29. EmpiricalEvidence:KnowledgeSpillovers • A higher probability of new multinational firms leads to an increase in the productivity of domestic firms at the 25th and 50th percentiles (3-4, percent upward shift of the productivity in the lower range of the distribution).

30. EstimatedWelfare and TFP Gains • 100-percent increase in theprobability of MNC entry • Aggregatedomesticproductivity (0.87 %): 75% increase in aggregateproductivity. • Knowledgespilloverse (0.6%): 69% of domesticproductivitygain. • Realloction (0.3%): 31% of domesticproductivitygain.

31. Within and BetweenIndustryReallocations

32. FDI PromotionPolicy

33. Country Heterogeneity 100-percent increase in theprobability of MNC entry Tougher domestic market selections and reallocation in developed nations and relatively less knowledge spillovers.

34. Discussion: Measures of Productivty, Correlations, Prices, Markups • Firm productivity based on the output value produced by each firm given its inputs: (we do not observe firm-level physical output quantities and prices). • No bias in cutoff revenue, and revenue distribution, and survival estimates. • Similar estimate of knowledge spillovers using shift in the productivity distribution or exploring domestic survival and changes in revenues distribution moments. • Overestimation of spillovers if not considering both sources even with q, p data. • Estimates of knowledge spillover could be biased downward if the distribution of productivity partly reflects the distribution of markups: More productive, higher markup firms survive, distributions shifts downward (Melitz and Ottaviano, 2008). • Re-estimate spillover equations for industries with homogeneous products (shifts more likely to reflect changes in productivity) using country-industry-specific import demand elasticities > 75th percentile in each country ( Broda, et al., 2006) • Productivity distribution of domestic firms shifts rightward by 3% (25th, 50th percentiles), no significant changes at 75th percentile; productivity distribution becomes more left truncated.

35. Conclusions • The impact of multinational activity on host-country productivity has been a major topic of economic research. • We develop a theoretical and structural empirical framework to distinguish knowledge spillovers and selection, less stressed in the literature: • Knowledge spillovers induce a rightward shift of the productivity distribution; the selection effect, in contrast, causes a weaker, or even leftward, shift of the revenue distribution and an increase in the cutoff productivity and revenue. • Using a large cross-country panel dataset of manufacturing firms, we find significant evidence of selections, market reallocations as well as knowledge spillovers. • We also quantify the productivity gains associated with each effect and explore cross-country heterogeneity.

37. Theoretical Framework: AggregateOutcomes • ND : equilibrium mass of incumbent domestic firms in each country. • Given the country symmetry assumption, the number of foreign owned firms in each country: • The total mass of varieties available to consumers in each country and the total mass of firms competing in each country: N=ND +NM.

38. Theoretical Framework:AggregateOutcomes • Theweightedaverageproductivitylevels of domestic and foreignfirms, are respectivleygivenby. • Theproductivityaveragesummarizestheeffects of thedistribution of productivitylevelsonaggregateoutcomes, (Melitz, 2003)

39. EquilibriumConditions • Steadystateequilibria: anenteringfirmwithproductivityθ wouldimmediatelyexitifitsprofitlevelwerenegativeorwould produce and earnπ(θ) in everyperioduntilitis hit withthebad shock and isforcedtoexit • Zerocutoffprofitconditions r(D) = cfD; r(M) = cfM • Free entrycondition (the expected value of future profits must, in equilibrium, equal the fixed entry cost.) • whereD 1-G(D) istheexanteprobablity of survivalafterentry

40. EquilibriumConditions: Factor MarketClearingConditions • Labor • Total demand for labor in the domestic market equals the total supply of labor L: • Which yields the number of domestic firms, ND, the number of foreign firms NF, and the total number of firms in the domestic market N. • Capital. • For simplicity, we assume firms finance a constant share of their fixed foreign investment cost in home countries and the rest abroad (empirical evidence). • which yields the unit financing cost c.

41. Model: TheImpact of MultinationalProductionAggregateProductivity and Welfare • Producivity: Multinational entry leads to a decrease in the number of domestic firms ND, and an increase in the aggregate productivity. • Reallocations in factor markets and tougher selection of domestic firms. • Surviving domestic firms benefit from the positive externalities from foreign firms and witness an increase in their productivity levels. • Welfare: whenthereisanincrease in the total productvariety, N, thiseffect, togetherwiththeincrease in aggregateproductivity, leads toanincrease in welfare

42. Model • ProductivityDistributionbefore and AfterMultinationalEntry

43. Model • TheRevenueDistributionbefore and AfterMultinationalEntry

44. Data • We employ a cross-country firm-level panel dataset, drawn from Orbis, that contains comprehensive financial, operation, and ownership information for over 1 million manufacturing firms in 60 countries. • Orbis provides several distinct advantages: • Ownership information (over 30 million shareholder/subsidiary links); time-series financial information; broad country coverage. • Four categories of information: • Industry information including the 4-digit NAICS (primary industry); • Ownership information including domestic and global parents and domestic and foreign subsidiaries; • Location information; • Financial information including revenue, employment, asset, and investment.

45. Data • We estimate total factor productivity using the semiparametric estimator developed by Olley and Pakes (1996). • The production function is estimated for each country and each NAICS 4-digit industry; • We consider two sub-periods: 2002-2004 and 2005-2007 and investigate how changes in multinational activity between the two periods affect host-country domestic firms. • A firm is considered foreign owned if its global ultimate owner is based in a different country. There are about 36,000 foreign owned manufacturing subsidiaries in the final sample.

46. List of Countries

47. Data: Distribution of DomesticFirms • Productivity distribution of domestic firms remained largely similar in the two periods for countries and industries where there was no multinational entry. Only the top range firms experienced a slight rightward shift in their productivity levels in 2005-2007. • For countries and industries with positive multinational entry, the productivity distribution of domestic firms shifted slightly to the right, even for the small- and medium-productivity firms, while the distribution became more left truncated.

48. Estimating EquationsThe Selection of Multinational Firms A foreignfirminvests in a host country ifM(),  >M. • Weconsiderthefollowingestimation: • Estimate the probability of investing abroad, conditional on being active in the home country market zM()=1, as a function of firm productivity θ, host country demand conditions E and P, and fixed investment cost cfM. • Host-country factors: country-industry fixed effect FEM. • Bilateral factors: distance between host and headquarters countries, common land border and language. • Based on estimates, we obtain: • Predicted probability of entry for each multinational, expected productivity of multinationals in each host country, expected probability of new multinational activity in each host country,

49. Estimating EquationsThe Selection of DomesticFirms A domesticfirmssurvivesD(),  >D. • Weconsiderthefollowingempiricalespecification Pr[zD() = 1]= Pr[ > D] • Dependent variable zD(θ): whether the domestic firm survives. • Based on the estimates, we obtain: • Predicted probability of survival for each domestic firm, expected productivity of surviving domestic firms, expected survival rate, • Cutoff productivity of domestic firms (θA), capital price (cA), aggregate price (PA) to multinational entry.

50. Estimating Equations MarketReallocations: Labor and Capital • LarborMarketReallocation • Foreignmultinational activity shifts the revenue distribution of domestic firms either rightward or leftward depending on whether PτθzM > PA, • qA: qth (25th, 50th and 75th) percentile of the ex-ante revenue distribution. • Capital MarketReallocation • c/cA is expected to be greater than 1 and rD(θA) is the cutoff revenue prior to the opennesstomultinationalactivity.