Revisiting Ricardo: Can productivity differences explain the pattern of trade between EU countries? Wilfried Altzinger (WU) Jo e P. Damijan (UL, IER, Licos) 3 rd FIW Research Conference Vienna, December 11, 2009 This paper was produced in the framework of MICRO-DYN, funded by the EU Sixth Framework Programme 1
Motivation Ricardian model of comparative advantage is probably one of the most exploited theoretical frameworks in trade literature However, surprisingly little empirical tests of the Ricardian model and scarce evidence in favor of it Leamer and Levinsohn (1995): The Ricardian link between trade patterns and relative labor costs is much too sharp to be found in any real data set Deterministic theory with 2 countries and complete specialization based on Dornbusch, Fischer and Samuelson (DFS, 1977) 2
Motivation (2) Eaton and Kortum (2002) demonstrate that random productivity shocks are sufficient to make the Ricardian model empirically relevant explain trade volumes with gravity-like equation with a continuum of goods, transport costs, and more than two countries Costinot and Komunjer (2009) stochastic theory of CA relative productivity performance for any pair of exporters fully determines their relative export performance across industries which is not fully in line with the original Ricardian approach requiring correspondence between bilateral productivity ratios and import-export ratios across industries 3
Aims we revise the empirical tests of the Ricardian model by testing (more) properly the Ricardian hypotheses on bilateral trade flows test whether two measures of productivity unit labor cost and labor productivity can effectively explain the pattern of bilateral trade flows between 21 EU member states period 1994-2004; NACE 2-digit industry aggregation compare the matching between bilateral sectoral exports-to-imports ratio rankings and relative bilateral sectoral productivity rankings for each of 21 x 20 country pairs static and dynamic analysis 4
Outline Motivation Previous empirical studies Theoretical setup Assumptions Continuum of goods Modifications Empirical approach Two hypotheses Methodology and data Results 5
Previous studies MacDougall (1951) 6
MacDougall (1951) 7
MacDougall (1951) 8
Balassa (1963) 9
Other studies Bhagwati (1964) critical comments on earlier emprical studies by MacDougall, Stern and Balassa. states that these studies have only a remote overlap with true Ricardian hypotheses: looking at relative export shares with third countries is not in line with the Ricardian model. Kreinin (1969) Sailors&Branson (1970) The results overhelmingly reject Ricardian hypothesis 10
McGilvray&Simpson (1973) Draw on many of the Bhagwati suggestions. Data for bilateral trade between UK nad Ireland for 34 sectors for 1963-64 Empirical approach: Ranking of sectors according to their export propensity (X/GDP) Ranking of sectors according to their import propensity (M/(GDP+M)) Ranking of sectors according to their relative sectoral productivity (IRL/UK) Computing Spearman rank correlation coefficients To confirm the type III hypothesis rank correlations should be: positive: between labor productivity and export propensity negative: between labor productivity and import propensity 11
McGilvray&Simpson (1973) Results reject the Ricardian hypothesis as none of the rank correlations is significant and all but two are opposite in sign wrt the expected 12
Costinot and Komunjer (2009 Linear (gravity-like) regression model Simple rationale behind it is that if country i is more productive in good k, then it should export more of good k to any importing country j Nothing is said about bilateral (relative) productivities and imports-to -exports ratios. 13
Costinot and Komunjer (2009 Data: 15 exporters (EU-14 + US), 50 importers (OECD, large non-oecd countries), 19 manufacturing industries, period 1988-2001 the elasticity of exports wrt the average unit labor requirement is negative, close to 1 and constant across importers 14
Assumptions Classic Ricardian setup Labor as the sole factor of production Continuum of goods (DFS, 1977) Many countries Differences in productivity (across countries and across goods) However, Product differentiation (Krugman, 1981) Stochastic productivity shocks (Costinot and Komunjer, 2009) Incomplete specialization Intra-industry trade possible 15
Continuum of differentiated goods (DFS, 1977) 16
Modification With the assumptions of Product differentiation (Krugman, 1981) Stochastic productivity shocks (Costinot and Komunjer, 2009)...we arrive at Incomplete specialization Intra-industry trade becomes possible 17
Modification (2) Hence, Instead of the deterministic ordering a weaker ordering version applies 18
Two types of hypotheses wrt wages Type I hypothesis Wages differ both across countries and industries Test of Ricardo model with respect to comparison of unit labor costs (and not productivities) between countries Type II hypothesis Relative industry wages across countries are equalized Test of Ricardo model in its original form: comparison of labor productivities between countries 19
Type I hypothesis 20
Type II hypothesis 21
Data and empirical approach Testing hypotheses Type-I and Type-II Bilateral trade between 21 EU member states Period: 1994-2004, Trade data: COMEXT (Eurostat) Productivity data: EU KLEMS data, University of Groningen Aggregation level: 14 NACE 2-letter product groups Tests compare the matching between: rankings of bilateral relative productivity across industries (LP, ULC), and rankings of bilateral exports-to-imports ratios across industries for all 21 x 20 country pairs for 11 years 22
Methodology Decreasingly ordered set T kzt (exports-to-imports ratio) for any country pair k (i, h) and product z in period t: if T kzt [0, ] > 1: country i is a net exporter to h, Decreasingly ordered set of productivity ratios (A kzt ) and unit labor cost (U kzt ) for any country pair k and good z in period t: A kzt ali U kzt [0, ] > 1: država i ima stroškovno prednost pred 23
Methodology (2) To validate the hypothesis Type-I, rankings of bilateral export/imports ratios should match the rankings of bilateral relative sector unit labor cost: To validate the hypothesis Type-II, rankings of bilateral export/imports ratios should match the rankings of bilateral relative sector productivities: 24
Methodology (3) Regression models: Type-I hypothesis Type-II hypothesis Test on bilateral trade among 21 EU states period1994-2004, 21 NACE 2-product groups Pooled regressions country-by-country 25
Correlation matrix 26
Test of hypothesis Type-I (ULC, levels) - 14/21 27
Test of hypothesis Type-II (LP, levels) - 13/21 28
Dynamics of comparative advantage Regression models: Type-I hypothesis Type-II hypothesis Long differences in ranks of variables over 1994-2004 Pooled regressions country-by-country 29
Test of hypothesis Type-I (ULC, long differences) - 2/21 (neg.) 30
Test of hypothesis Type-II (LP, long differences) 2/21 (neg.) 31
Conclusions No conclusive results which would validate the Ricardian hypotheses generally bilateral trade patterns are surprisingly well explained by the relative sectoral productivities in particular, the relative unit labor cost (i.e. Ricardian Type-I hypothesis) is doing a good job positive and significant correspondence in 67 per cent of country pairs However, mobility of comparative advantages over time is hardly to be explained by the changes in relative productivities among countries 32