1. pagnese@iese.edu Japan and her dealings with offshoring: An empirical analysis with aggregate data Pablo Agnese 5 th PhD presentation, RES Meeting City University London January 16 th - 17 th 2010

2. pagnese@iese.edu Motivation and goal Is Japan a particular case study? Relatively scarce literature Slump in the 1990s (bubble, East Asian crisis, etc.) Goal: understand the employment and productivity effects of offshoring at the aggregate (industry) level

3. pagnese@iese.edu Main findings 1.Offshoring of services (as opposed to production) still of less importance 2.Offshoring has negligible effects on aggregate employment 3.Offshoring can improve productivity

4. The offshoring literature (Japan) –Head and Ries (2002), Tomiura (2005), Hijzen et al. (2006), Ito et al. (2007) –Results so far: Disaggregate (firm), manufacturing sector only, materials offshoring only, FBTC (Head and Ries, 2002), positive productivity effects only (Hijzen et al., 2006) –This contribution: Aggregate (industry), whole economy, both materials and services offshoring, both employment and productivity effects pagnese@iese.edu

5. Measuring offshoring (I) Direct indicators not easy to come by Indirectly (Feenstra and Hanson): share of imported intermediate inputs in total inputs used, output, exports. Formally: pagnese@iese.edu

6. Measuring offshoring (II) Common drawbacks: –offshoring does not necessarily imply an increase of imports, and vice versa –import penetration of inputs is usually taken as equal for every unit Rationale: intermediate trade stands for an important amount of intra and inter firm trade nowadays, from which offshoring could be proxied. pagnese@iese.edu

7. Estimation methodology: Employment (I) The Cobb-Douglas production function: Accepting that the industry behaves as profit-max firm: First order conditions yield the conditional demand for inputs. For labor we have: pagnese@iese.edu

8. Estimation methodology: Employment (II) Among input prices we can identify the price of foreign labor services (Amiti and Wei, 2005, 2006): Since data on p OS are difficult to get, we use the offshoring intensity index as inverse proxy: Three channels through which offshoring affects employment: substitution effect (-), SR productivity effect (-), LR productivity or scale effect (+) pagnese@iese.edu

9. Estimation methodology: Employment (III) pagnese@iese.edu Our alternative estimating equations thus become:

10. Estimation methodology: Productivity(I) Total factor productivity (TFP) as a multiple-factor measure Different measures of output and inputs and, thus, of productivity, reflect different representations of the same production process We only consider two of these measures: value-added- based and output-based We follow the two-stage estimation methodology, but adjusting for non-CRS Once this is done, we can see the direct effects of offshoring on productivity pagnese@iese.edu

11. Estimation methodology: Productivity(II) pagnese@iese.edu Differentiating with respect to t, we get: Under CRS and P.C. we have:

12. Estimation methodology: Productivity(III) However, under non-CRS, Euler’s eq. can be rewritten: The reduced-form estimating equation is: pagnese@iese.edu

13. Data JIP (2006, 2008) database: –1980 to 2005 –108 industries, 83 after data cleaning –Similar to ISIC rev. 3 Our offshoring index: we adapt (1) to account for both services and materials offshoring. Our measures stand for the import content in all services and material inputs, respectively. pagnese@iese.edu

14. Descriptive statistics (I) pagnese@iese.edu Note: Japan’s manufacturing and services offshoring indices (OSM, OSS) according to formula (1). Broad measures, weighted by industry output (JIP database). Materials and services offshoring (%)

15. pagnese@iese.edu Descriptive statistics (II) Note: mean values across 83 industries. Total factor productivity, growth rate (%)

16. But how good are the offshoring measures? –Decomposition analysis (1980-2005) –Following the “within” and “between” exercise (Hummels et al., 2001, Strauss-Kahn, 2004, and Horgos, 2008) –To see what proportion of the change in the index is due either to a change in offshoring or a change in the industries’ relative weights within the economy pagnese@iese.edu Descriptive statistics (III)

17. pagnese@iese.edu Descriptive statistics (IV) Whole economyWithinBetweenTotal (w+b)Within/Total OSM 1980-1990 2.77-0.052.72101.8% 1990-2005 5.13-0.194.94103.8% 1980-2005 7.87-0.207.67102.6% OSS 1980-1990 1.290.011.3099.1% 1990-2005 -0.44-0.03-0.4793.5% 1980-2005 0.830.010.8399.4% Manufacturing OSM 1980-1990 4.03-0.173.86104.5% 1990-2005 3.94-0.763.18124.0% 1980-2005 7.91-0.887.04112.5% OSS 1980-1990 1.230.001.2399.7% 1990-2005 -0.440.02-0.42105.5% 1980-2005 0.760.050.8193.9% Services OSM 1980-1990 2.34-0.062.27102.7% 1990-2005 5.300.165.4697.0% 1980-2005 7.610.127.7398.5% OSS 1980-1990 1.340.011.3599.2% 1990-2005 -0.43-0.06-0.4988.4% 1980-2005 0.87-0.010.86101.7%

18. pagnese@iese.edu Dependent variable: ln L it 83 ind., 1980-2005(1)(2)(3)(4) GMM-DIF GMM-OD ln L it-1 0.95† (0.001)(0.01)(0.001)(0.005) ln w it -0.03†-0.02†-0.03† (0.001)(0.009)(0.001)(0.002) Δ ln w it -0.07†-0.06†-0.10† (0.001)(0.01)(0.002)(0.01) OSS it / 1001.02†0.921.03†0.53‡ (0.07)(0.71)(0.06)(0.25) Δ OSS it / 100-0.50†*-0.59†* (0.03)(0.13) OSM it / 100-0.26†-0.14-0.33†-0.23‡† (0.01)(0.14)(0.01)(0.13) Δ OSM it /100-0.46†*-0.47†* (0.03) ln K it **** Δ ln K it 0.27†0.20†0.21†0.17† (0.004)(0.06)(0.01)(0.06) Sargan test:χ 2 (76) = 90.19χ 2 (53) = 68.65χ 2 (76) = 82.49χ 2 (53) = 66.24 p-value = 0.18p-value = 0.07p-value = 0.28p-value = 0.10 m2 test:z = -16.83z = -1.63z = 0.38z = 1.68 p-value = 0.00p-value = 0.10p-value = 0.70p-value = 0.10 Period dummiesnoyesnoyes s.e.0.05 0.04 Adj. r 2 0.050.110.96 observations1,9922,0751,9922,075 *: strongly non-significant, individually or jointly (variable removed). Note: all specifications estimated with Eviews and based on equation (8). GMM-DIF is the Arellano-Bond (1991) estimator in first differences and GMM-OD the Arellano-Bover (1995) estimator in orthogonal deviations. Both are estimated using the 2- step method by Arellano and Bond (1991), so the standard errors may not be reliable. Results from 1-step estimations with the GMM-DIF were rather similar. The offshoring indices (%) are divided by 100 so as to interpret the semi-elasticities directly. Standard errors in parentheses and †, ‡, and ‡† the usual levels of significance: 1%, 5%, and 10%; Δ is the dif. operator.

19. Services offshoring affects employment positively in all estimations Materials offshoring affects employment negatively Results are robust to the dependent variable and the various specifications of the LD equation Our LD equation minimizes the endogeneity problem by using GMM and ruling out “output” variables Effects are negligible. Quantifying the results: pagnese@iese.edu Employment effects Average employment effects of offshoring, 1980-2005 (3) (4) Δ workers% % OSS9510.9 4890.5 OSM-2,872-2.8-2,001-2.0 Total-1,921-1.9-1,512-1.5 Note: avg. employment increase was 101,425 workers (83 industries).

20. pagnese@iese.edu Dependent variable: Δ τ’ Oit 83 ind., 1980-2005(1’)(2’)(3’)(4’)(5’)(6’) GMM-OD Δ τ’ Oit-1 0.0090.008 (0.01) OSS it / 1000.320.66‡†0.73‡†0.84‡†0.93‡†0.86‡† (0.36)(0.40)(0.42)(0.46)(0.51) OSM it / 1000.020.08‡0.07‡†0.08‡0.07‡†0.06 (0.03)(0.04) HKHK 0.090.080.070.05 (0.08) (0.09) Δ H K -1.42†-1.39†-1.36†-1.31†-1.32† (0.27)(0.26)(0.24)(0.23)(0.24) R&D0.090.100.07 (0.15)(0.16) Δ R&D-0.22 (0.20) Period dummiesyes Sargan (p-value):0.510.320.330.320.350.32 Adj. r 2 0.100.09 0.08 0.09 observations2,0751,660 Note: all specifications estimated with Eviews and based on equation (18). GMM-OD is the Arellano-Bover (1995) estimator in orthogonal deviations (Arellano-Bond 2-step). Results from 1-step estimations with the GMM-DIF or using τ VAit and τ JIPit as dependent variables were ambiguous. Results without period dummies were also not significant. The offshoring indices (%) are divided by 100 so as to interpret the semi-elasticities directly. Standard errors in parentheses and †, ‡, and ‡† the usual levels of significance: 1%, 5%, and 10%; Δ is the difference operator, H K the share of high-skill workers, and R&D the share of software investment in GDP.

21. pagnese@iese.edu Productivity effects Only the output-based measure produces unambiguous results Control variables not significant (R&D and H K ) Both services and materials affect productivity growth positively, with larger effects from services Quantifying: Average productivity effects of offshoring, 1980-2005 (2')(3')(4')(5')(6')Hijzen et al. (2006) OSS1.401.551.791.981.83 1.80 OSM0.640.560.640.560.48 in percentage points Note: means of OSS and OSM are 0.02 and 0.08 respectively (83 industries).

22. Concluding remarks As more services become tradable, more jobs will be at risk of being relocated Materials offshoring still of greater importance Services offshoring does not seem to “take-off” Total loss of jobs is negligible Positive employment effects of services offshoring Negative employment effects of materials offshoring Productivity improvement might be expected Policy-makers to hinder offshoring and profit-seeking Future: spillover effects (general equilibrium) pagnese@iese.edu