Unpacking Sources of Comparative Advantage: A Quantitative Approach

Transcription

1 Unpacking Sources of Comparative Advantage: A Quantitative Approach Davin Chor (Singapore Management University) 5 Jan 2008 NAWM of the Econometric Society (New Orleans)

2 Background Recent empirical work on sources of comparative advantage: 1. Productivity levels (Ricardian framework): Eaton and Kortum (2002): Multi-country Ricardian model that yields closed-form trade flows Good fit to bilateral manufacturing trade data in the OECD Structural approach facilitates counterfactual exercises: eg computing welfare gains from the reduction of distance barriers

3 Background Recent empirical work on sources of comparative advantage: 1. Productivity levels (Ricardian framework): Eaton and Kortum (2002): Multi-country Ricardian model that yields closed-form trade flows Good fit to bilateral manufacturing trade data in the OECD Structural approach facilitates counterfactual exercises: eg computing welfare gains from the reduction of distance barriers 2. Factor endowments (Heckscher-Ohlin framework): Romalis (2004): Countries export more from industries that use their abundant factors more intensively

4 Background (cont.) 3. Institutional determinants: Financial development (Beck 2003, Manova 2006) Legal/Contracting institutions (Levchenko 2007, Nunn 2007, Costinot 2006) Labor market institutions (Cuñat and Melitz 2007) Empirical strategy: Countries with particular institutional strengths export more in industries that are dependent on these institutional conditions. Focus on identifying effects, rather than welfare implications

5 Background (cont.) 3. Institutional determinants: Financial development (Beck 2003, Manova 2006) Legal/Contracting institutions (Levchenko 2007, Nunn 2007, Costinot 2006) Labor market institutions (Cuñat and Melitz 2007) Empirical strategy: Countries with particular institutional strengths export more in industries that are dependent on these institutional conditions. Focus on identifying effects, rather than welfare implications How much do different sources of comparative advantage matter for patterns of specialization and country welfare?

6 What this paper does: Model Develops a structural framework for quantifying the importance of different sources of comparative advantage: Eaton-Kortum model extended to industry trade flows Comparative advantage driven by the interaction of country and industry characteristics Intuition: Countries specialize in industries whose production needs they can best satisfy with their factor endowment mix, institutional environment, and technological strengths. In general equilibrium, specialization patterns and welfare are functions of underlying country and industry characteristics, and distance barriers

7 What this paper does: Empirics Estimation on a dataset of bilateral industry trade flows (82 countries and 20 manufacturing industries) 1. OLS baseline Corroborates literature on the significance of distance barriers, factor endowments, and institutional conditions in explaining bilateral trade patterns 2. Simulated method of moments (Pakes and Pollard 1989) To account for the bias from omitting zero trade flows (about two-thirds of the data)

8 What this paper does: Counterfactuals 1. Effect of Distance: Large welfare gains from reduction of physical distance barriers (25.6%) Comparable to what EK find for the OECD (16.1%-24.1%) Developing countries benefit more than the OECD (29.8% vs 7.0%)

9 What this paper does: Counterfactuals 1. Effect of Distance: Large welfare gains from reduction of physical distance barriers (25.6%) Comparable to what EK find for the OECD (16.1%-24.1%) Developing countries benefit more than the OECD (29.8% vs 7.0%) 2. Policy experiments: Raising all countries factor endowments or institutional characteristics to the world frontier level... Also... illustrate welfare gains and the shift in industry composition for a single country (Indonesia) when raising its country attributes

10 Related literature Estimation of model of bilateral trade flows / MNC activity: Eaton and Kortum (2002), Alvarez and Lucas (2007), Shikher (2007) Lai and Trefler (2002), Lai and Zhu (2004) Ramondo (2006) Costinot and Komunjer (2006) This paper: Ties pattern of industry trade flows to country characteristics Estimation in a manner consistent with the zero trade flows

11 Plan of Talk 1. Introduction and Motivation 2. to the industry level 3. Estimation OLS baseline Simulated method of moments (SMM) 4. on... Distance barriers Country characteristics 5. Conclusions

12 Modelling industry trade flows: Utility k = 0, 1,..., K industries: Industry 0: non-tradable, domestic numeraire Industry k 1: tradables Utility: ( ) 1 η U n = Qn 0 ( 1 ) β (Qn k (j k )) α dj k α k 1 0 η β, α, β, η (0, 1) (1) ε = 1 > 1: elasticity of substitution between varieties from the same industry 1 α φ = 1 > 1: elasticity of substitution between varieties across industries 1 β Assume: ε > φ > 1.

13 Modelling industry trade flows: Goods Prices Market structure: perfect competition Production: constant returns to scale Firms price at average cost: p k ni(j) = ck i d k ni z k i (j) n: importer i: exporter j: varieties (2) Unit production costs: c k i = f (w if ) sk f, s k f (0, 1), f sk f = 1 Distance: d k ni 1, d k ni d k nmd k mi Productivity: z k i (j)

14 Modelling industry trade flows: Goods Prices Specify: ln z k i (j) = λ i + µ k + {l,m} β lm L il M km + β 0ɛ k i (j) (3) Systematic component: λ i + µ k + {l,m} β lml il M km L il : country characteristics M km : industry characteristics Stochastic component: ɛ k i (j) is a draw from the Type I extreme-value (Gumbel) distribution, F (x) = exp ( exp ( x)) ln p k ni(j) = ln c k i d k ni λ i µ k {l,m} β lml il M km β 0ɛ k i (j) Head-to-head competition: Each variety is procured from the lowest-price exporter

15 Modelling industry trade flows Normalized trade flows: ( ) X k ni X k nu = (ck i d k ni) θ ϕ k i (c k u d k nu) θ ϕ k u (4) where: (i) ϕ k i = exp {θ(λ i + µ k + {l,m} β lml il M km )}; (ii) θ = 1 β 0 is an inverse spread parameter. Comparison with EK (2002): K = 1 ( ) Xni = (c id ni ) θ T i (5) (c ud nu) θ T u X nu ϕ k i unpacks productivity as a function of country and industry characteristics

16 Plan of Talk 1. Introduction and Motivation 2. to the industry level 3. Estimation OLS baseline Simulated method of moments (SMM) 4. on... Distance barriers Country characteristics 5. Conclusions

17 Estimation: OLS Baseline Specify: d k ni = exp{β d D ni + δ k + ζ ni + ν k ni} (6) D ni : distance measures from gravity literature ζ ni N(0, σ 2 ζ), and ν k ni N(0, σ 2 ν) From CEPII and Rose (2004): (i) Log distance; (ii) Common language dummy; (iii) Colonial dummy; (iv) Common colonizer dummy; (v) Common border dummy; (vi) GATT dummy; (vii) Regional Trade Agreement dummy

18 Estimation: OLS Baseline Estimating equation: (relative to u: US) ( ) X k ln ni Xnu k = F f =1 ( θβ f ln V if ln V ) uf sf k + θβ lm (L il L ul )M km... V i0 V u0 {l,m} θβ d (D ni D nu) + I i + I nk θζ ni θν k ni (7) Natural decomposition: Normalized trade flows as a function of: 1. Distance 2. Heckscher-Ohlin determinants 3. Institutional determinants 4. Exporter and importer-industry fixed effects 5. Idiosyncratic noise

19 OLS estimates: Caveats OLS provides a baseline for comparison with existing empirical work. But Can only interpret these as effects conditional on observing positive trade flows, even though majority (67.6%) of the data is zeros Coefficients are biased when omitting zeros (Haveman and Hummels 2004, Anderson and van Wincoop 2004, Silva-Santos and Tenreyro 2006, Helpman, Melitz and Rubinstein 2007) 2. Cannot identify θ, which we need to compute welfare counterfactuals

20 Modifying the theory to generate zeros Let the ɛ k i (j) s now be independent draws from a truncated Gumbel distribution with bounded support [x, x]: F (ɛ) = F (ɛ) F (x) F ( x) F (x), where F (ɛ) = exp( exp( ɛ)) Zero trade flows arise when there are large cross-country productivity gaps Lose closed-forms for trade flows, but can simulate them Motivates a simulated method of moments (SMM) estimator; akin to Ramondo (2006)

21 Procedure for simulating trade flows 1. For each variety j, compute the prices presented by all N countries to each importing country n: ln(p k ni )(j) = 1 F θβ d D ni θβ f s k f ln V if θβ lm L il M km + Ĩ i + Ĩ k (ɛ k i θ V )(j) i0 f =1 {l,m} Proxy for exporter fixed effects: Ĩ i = c 1 (ln( Y i L i )) γ 1 ; c 1 < 0, γ 1 > 0 Proxy for industry fixed effects: Ĩ k = c 2 (ln(t k )) γ 2 ; c 2 < 0, γ 2 > 0 2. Identify the exporting country that presents the lowest price, and compute the corresponding quantity. 3. Sum over varieties for the value of bilateral industry trade flows.

22 SMM Estimator ε = 3.8 (from Bernard et al. 2003); φ = 2 J = 100 where: ˆΘ = arg min (b( ˆΘ) b(θ)) Ψ (b( ˆΘ) b(θ)) (8) Θ = {x, x, θ, η, c 1, γ 1, c 2, γ 2, β d1,..., β d7, β f 1,..., β f 4, β lm1,..., β lm6 } Moments matched, b(θ): Covariances: Cov(X k ni, Log(Distance)),..., Cov(X k ni, SVOL FLEX ) Mean trade flows by industry

23 Dataset 82 countries 20 US SIC-87 industries (2-digit manufacturing) = 132, 840 observations Year: 1990 (same as EK) X k ni: Feenstra et al. (2005), concorded from SITC Rev 2 to SIC-87

24 Dataset Heckscher-Ohlin determinants: ( ) ln V if V i0 ln V uf V u0 sf k Follows Romalis (2004) 1. s h log(h/l) Skill intensity Skill abundance 2. s k log(k/l) Capital intensity Physical capital abundance 3. s m log(forest/l) Materials intensity Forest land s m log(arable/l) Materials intensity Arable land

25 Dataset Institutional determinants: L il M km 1. CAPDEP FINDEV (Beck 2003, Manova 2006) External capital dependence Financial Development 2. HI LEGAL (Levchenko 2007) Input concentration Strength of legal institutions 3. RS LEGAL (Nunn 2007) Input relationship-specificity Strength of legal institutions 4. COMPL LEGAL (Costinot 2006) Job complexity Strength of legal institutions COMPL log(h/l) Job complexity Skill abundance 5. SVOL FLEX (Cuñat and Melitz 2007) Sales volatility Labor market flexibility

26 Estimates (Table 3) 1. Gravity matters: Distance variables have expected sign, though not all significant (1) (3) (4) OLS Probit SMM Distance and Geography: β d1 : Log (Distance) 1.13*** 0.64*** 0.36*** (0.03) (0.03) (0.13) β d2 : Common Language 0.66*** 0.50*** 0.15 (0.06) (0.04) (0.80) β d3 : Colony 0.49*** (0.10) (0.08) (8.20) β d4 : Common colonizer (0.15) (0.11) (9.05) β d5 : Border (0.15) (0.15) (4.08) β d6 : RTA 0.45*** 0.40*** 0.78** (0.10) (0.15) (0.39) β d7 : GATT ** 0.73** (0.25) (0.15) (0.35)

27 2. Factor endowments matter: Heckscher-Ohlin interactions for human capital and physical capital generally positive and significant (1) (3) (4) OLS Probit SMM Heckscher-Ohlin determinants: β f 1 : s h log(h/l) 15.37*** 3.54*** 10.13*** (2.00) (0.98) (0.60) β f 2 : s k log(k/l) ** (0.20) (0.08) (0.08) β f 3 : s m log(forest/l) ** 0.04 (0.10) (0.04) (0.68) β f 4 : s m log(arable/l) 1.00*** 0.26*** 0.29 (0.15) (0.06) (0.70)

28 3. Institutions matter: Even when tested jointly (1) (3) (4) OLS Probit SMM Institutional determinants: β lm1 : CAPDEP FINDEV 1.17*** 0.25*** 1.81*** (0.09) (0.06) (0.59) β lm2 : HI LEGAL *** 1.77 (1.95) (0.78) (3.86) β lm3 : RS LEGAL 4.70*** 4.26*** 1.43** (0.76) (0.32) (0.57) β lm4 : COMPL LEGAL 4.87*** 0.51** 0.36** (0.45) (0.22) (0.16) β lm5 : COMPL log(h/l) 2.16*** 0.36** 1.81* (0.33) (0.15) (1.00) β lm6 : SVOL FLEX 10.57*** (2.07) (1.18) (2.23)

29 SMM estimates (cont.) 1. θ = On the high end of EK s range of estimates (2.44 to 12.86) 2. {x, x} = { 1.61, 7.07 } 3. {η, c 1, γ 1, c 2, γ 2} = {0.24, , 2.34, 0.06, 2.52}

30 Assessing the fit 1. Generated income levels vs actual GDP (WDI) Pin down (Y 1,..., Y n) via trade balance: EXP i = IMP i = ηy i Spearman rank correlation: 0.69Figure 1 Assessing the Goodness of Fit: Predicted Income Levels vs Observed Country GDP Levels Pearson correlation (log incomes): 0.71 (normalized, US=1) Predicted GDP levels JAM PAN DOM URY CRI ECU LKA ZMB KEN CMR PRY CIV NER SLE HTI NIC BOL CAF TGO GHA PNG TUN TCD HND MDG MWI UGA SLV ZAR MLI SEN JOR SYR GTM BDI ZWE CHE NOR NLD JPN DEU USA SWE DNKBEL FIN CAN FRA NZL AUT GBR SGP IRL ISR ITA HUN GRC KOR AUS ESP PRT MYS POL PHL IDN MAREGY THA ARG TUR MEX PER CHL IND CHN BRA ZAF COL NGA PAK IRN DZA Actual country GDP levels Notes: Observed country GDP levels plotted on the horizontal axis are from the World Development Indicators (WDI). The predicted country GDPs on

31 Assessing the fit 2. Simulated vs actual trade flows Slope coefficient: 0.77 Figure 2 Number Assessing of zeros: the Goodness 81,990 of Fit: vs Predicted 89,806 (64,506 Trade Flows of vs these Observed shared) Trade Flows Predicted trade flows 1.0e e e e e e e e e+07 Observed trade flows

32 Plan of Talk 1. Introduction and Motivation 2. to the industry level 3. Estimation OLS baseline Simulated method of moments (SMM) 4. on... Distance barriers Country characteristics 5. Conclusions

33 Computing welfare counterfactuals Indirect utility: W n = (1 η) 1 η η η Y n (p 0 n) 1 η ( k 1 (Pk n ) 1 φ ) η 1 φ (9) Welfare change: W n W n = Yn Y n (1 η) (p0 n ) p 0 n η ( ) 1/(1 φ) k 1 (Pk n ) 1 φ ( ) 1/(1 φ) k 1 (Pk n ) 1 φ (Assumption: Factors are mobile domestically, but not across borders)

34 Distance counterfactuals 1. Large average gains from removing physical distance (25.6% ) Comparable to what Eaton-Kortum find for the OECD (16.1%-24.1%) Decomposition of welfare change: Most of the increase stems from change in country GDP as market access is opened % Welfare Change Decomposition Due to change in: Std. Wtd. Prices Prices Min. Max. Dev. Avg. GDP (k 1) (k = 0) Reducing Distance Barriers: Log (Distance) OECD: ROW: Halving distance mark-up OECD: ROW:

35 Distance counterfactuals (cont.) Countries that are more isolated see greater gains Developing countries gain more than the OECD (29.8% vs 7.0%); similar to Lai and Zhu (2004) Figure 4 Welfare Effect of a Counterfactual Removal of Physical Distance Barriers % welfare change BDI MWI BOL ZWE IRNHTITCD CAF SEN HND MDG SLV DZA JOR PAK TGO CIV GHA IND CRI ZAR BRA ZAF NER MLISLE THA SYR LKA GTM ARG URY CMR UGA KEN PER IDN COL DOM MEX JAM NGA NIC ZMB PHL PAN CHL EGY ECU MYS PNG PRTMAR TUR PRY GRC KOR POL AUS ESP GBR IRL HUN ITA CHN CAN ISR SGP NZL FRA AUT BEL DEU DNK FIN SWE USA NLD JPN NOR CHE TUN Log Average GDP-weighted distance Notes: The vertical axis plots the percent welfare change from a hypothetical removal of physical distance, as described in Section 5.1. This is plotted against Davin the Chor log mean (Singapore GDP-weighted Management distance University) of each country from Unpacking all countries Sources in the ofsample. Comparative The linear Advantage: best fit line Ais Quantitative illustrated Approach

36 Country policy experiments: Raising all countries Caveat: Welfare changes are due to the underlying shift in industrial composition from introducing the shock to the interaction term. 1. Gains from physical capital accumulation through s k log(k/l) term: 42.0% 2. Combined effect from human capital accumulation through s h log(h/l) and COMPL log(h/l) terms: 37.6% % Welfare Change Decomposition Due to change in: Std. Wtd. Prices Prices Min. Max. Dev. Avg. GDP (k 1) (k = 0) Raising Factor endowments: s h max(log(h/l)) s k max(log(k/l)) Total effect: max(log(h/l))

37 Country policy experiments (cont.) 3. Combined effect of improving legal institutions: 17.7% About two-thirds of this from increased specialization in industries that use a large share of relationship-specific inputs % Welfare Change Decomposition Due to change in: Std. Wtd. Prices Prices Min. Max. Dev. Avg. GDP (k 1) (k = 0) Raising Institutional attributes: HI max(legal) RS max(legal) COMPL max(legal) Total effect: max(legal)

38 Country policy experiments (cont.) 4. Welfare gain negatively correlated with initial level of the country Figure 9 characteristic Welfare Effect of a Counterfactual Improvement in Legal Institutions JOR % welfare change IRN GTM ZWE ZAR HND SYR SLV BOL DZA NIC GHA PAN HTI TUN PERPHL LKA COL PAK MLI NGA UGA JAM SEN MWI MDG BDI ZAF CAF KOR TCD ARG CRI CHL IDN NER PRY TGO MAR POL IND URYBRA ZMB ECU SLE MEX DOM GRC EGYPNG KEN CIV TUR THA CMRPRT MYS ESPIRL GBR ISR CHN HUN ITA JPN SGP FRA SWE DEU NZL AUS CAN AUT BEL DNK FINNOR NLD USA CHE LEGAL Notes: The vertical axis plots the percent welfare change from improving all countries legal institutions to the world frontier (the highest index value observed in the sample), as described in Section 5.2. This is plotted against the initial index value of LEGAL for each country. The linear best fit line is illustrated; the Pearson linear correlation is -0.90, significant at the 1% level.

39 Country policy experiments: Raising one country Raise attributes of one large developing country, Indonesia 1. Largest gains from physical capital accumulation: 43.8% Any beggar-thy-neighbor effects are small 2. Similar gains from human capital accumulation: 38.7% 3. Slightly more modest, but still strong gains from improving LEGAL: 20.2% % Welfare Change for IDN % Welfare Change for sample Due to change in: Prices Prices Std. Wtd. Total GDP (k 1) (k = 0) Min. Max. Dev. Avg. Raising: max(log(h/l)) max(log(k/l)) max(findev ) max(legal)

40 Country policy experiments (cont.) 4. Shift in industry composition: Raising a country characteristic promotes expansion in industries that are Figure intensive/dependent 10 on that characteristic Effect of Raising Indonesia s Human Capital Endowment on Domestic Industrial Composition % change in industry share of IDN exports Skill intensity Notes: The vertical axis plots the percent change in each industry s share of Indonesia s total exports (including domestic absorption) following an increase in Indonesia s human capital endowment to the world frontier (the highest value observed in the sample), as described in Section 5.2. This is plotted against the skill intensity of each industry. The linear best fit line is illustrated; the Pearson linear correlation is 0.59, significant Davin at the 1% Chor level. (Singapore Management University) Unpacking Sources of Comparative Advantage: A Quantitative Approach

41 Conclusion Developed a methodology that facilitates quantification of the welfare impact of different sources of comparative advantage Model: EK model with comparative advantage determined by the interaction of country and industry characteristics Estimation: (i) OLS baseline (ii) SMM procedure: To account for the zeroes Welfare counterfactuals: Reasonable magnitudes for the welfare gains from reducing physical distance, factor accumulation, or improvements in institutions.

42 Ongoing work SMM Estimation with a full set of fixed effects Further counterfactual applications: How much can be gained from regional trade integration? Thinking about dynamics: What does this framework imply about trade and growth?

43 Estimation: SMM (extra slide) Simulating trade flows: 1. For each variety, compute the prices presented by all N countries to each importing country n: ln(p k ni )(j) = 1 F θβ d D ni θβ f s k f ln V if θβ lm L il M km + θ V Ĩi + Ĩk (ɛ k i )(j) i0 f =1 {l,m} Requires N K J independent draws from the truncated Gumbel distribution with support [x, x] for the (ɛ k i ) (j) s. 2. For each variety j and each importing country, n, identify the exporting country, i(j), that presents the lowest price, (p k n,i(j)) (j).

44 Estimation: SMM (extra slide) Simulating trade flows: [cont.] 3. Calculate quantities: (Qn,i(j) k )(j) = ηyn (Pk n )ε φ k 1 (Pk n ) 1 φ ((pk n,i(j) )(j) ) ε, (Pn k )1 ε 1 J J ((pn,i(j) k )(j) ) 1 ε j=1 4. Sum over the relevant exporter subscripts to obtain bilateral trade flows: (Xni) k sim = 1 (p k J n,i(j)) (j) (Qn,i(j)) k (j) {j: i(j)=i}

45 SMM estimates: Assessing the fit (Extra slide) 1. Generated income levels vs actual GDP (WDI) Close the model to pin down (Y 1,..., Y n) via trade balance: EXP i 1 J = 1 J N K n=1 k=1 {j: i(j)=i} K N k=1 n=1 {j: i(j)=i} (p k n,i(j)) (j) (Q k n,i(j)) (j) ηy n (P k n ) ε φ k 1 (Pk n ) 1 φ ((pk n,i(j)) (j) ) 1 ε = IMP i = ηy i A linear homogenous system in Y 1,..., Y N. Solve up to a constant of proportionality (set Y US = 1).

46 (Extra slide) Yn Y n and ( k 1 (Pk n )1 φ ) 1/(1 φ) ( k 1 (Pk n )1 φ ) 1/(1 φ) are straightforward to compute. (p0 n ) p 0 n is more tricky: (p 0 n) p 0 n F = sf 0 f =0 (w nf ) w nf Change in w nf computed as the change in total payments to that factor minus any change in endowments: (w nf ) = (s0 f (1 η)yn + J N k 1 sk f j=1 s=1 (pk sn) (j) (Qsn) k (j) ) w nf (sf 0(1 η)yn + J N (V nf ) k 1 sk f j=1 s=1 (pk sn) (j) (Qsn) k (j) ) V nf Assumed factor shares: s 0 h = 0.07, s 0 l = 0.13, s 0 m = 0.6, s 0 k = 0.2. (Calibrated with shares in US agriculture)

47 Notes: The horizontal Davin axis Chor plots (Singapore country price Management indices for University) consumption goods Unpacking from the Sources Penn World of Comparative Tables (PWT), Advantage: adjusted Afor Quantitative purchasing power Approach parity. Extra slide: Assessing the fit 3. Generated ideal price index vs PPP consumption price index (PWT) Spearman rank correlation: 0.60Figure 3 Pearson Assessing correlation: the Goodness of 0.68 Fit: Generated Ideal Price Indices vs Observed Price Indices (normalized, US=1) Generated Ideal Price Indices CHE NLD FRA NOR USA GBRBEL JPN DNK SWE CAN IRLAUT FIN ITA ESP ISR HUN GRCNZL PRT SGP POL AUS KOR EGY TUR TUN MAR MYS CRI CHL MEX CIV ARG NIC PHLDOM THA URY IDN COL CMR JAM BRA SLE LKA IND BDI TGO BOL GHA NER ZAF PER PAK NGA CAF KEN TCDZAR SEN MDG HTI CHNMWI UGA ZMB ZWE ECU PAN JOR DZA SYR HND PNG GTM PRY SLV IRN MLI Price Level: Consumption goods (PWT)

48 Distance counterfactuals (cont.) 2. Gains from a global GATT (6.9% ) Main beneficiaries are: (i) Non-GATT members (ii) Developing countries % Welfare Change Decomposition Correlation Due to change in: Std. Wtd. Prices Prices Min. Max. Dev. Avg. GDP (k 1) (k = 0) Global GATT *** OECD: ROW: