1. Sevilla May 2011, WIOD 2 nd Consortium Meeting 1 WP 6 Methodological research related to the database WP 6.4 Multiplier bias from Supply and Use Tables José M. Rueda-Cantuche (JRC-IPTS) Erik Dietzenbacher (RUG, NL) Esteban Fernández (University of Oviedo, ES) Antonio F. de Amores (Pablo de Olavide University, ES)

2. Sevilla May 2011, WIOD 2 nd Consortium Meeting 2 Rationale

3. Sevilla May 2011, WIOD 2 nd Consortium Meeting 3 Following Dietzenbacher (2006): Leontief inverse, L, plays a relevant role in inter- industry economics L captures direct + indirect effects of an exogenous shock on industry/commodity output L can also describes the inter-industry core of a CGE model

4. Sevilla May 2011, WIOD 2 nd Consortium Meeting 4 As far as, typically, L = (I – A) -1, it is subject to the many sources of measurements errors that are very well known for IOT, A and SUT. Hence, it seems plausible assuming A stochastic, which leads to L is biased, with input coefficients: –totally independent (Simonovits, 1975) –biproportionally stochastic (Lahiri, 1983) –moment-associated (Flam and Thorlund-Petersen, 1985).

5. Sevilla May 2011, WIOD 2 nd Consortium Meeting 5 More recently, stochastics was alternatively imposed on the intermediate transactions of an input-output table rather than on its technical coefficients (e.g. Dietzenbacher, 2006). The findings of these experiments turned out that the bias tends to be rather small and needs a large sample size to get significant relevance. Complementary to Roland-Holst (1989).

6. Sevilla May 2011, WIOD 2 nd Consortium Meeting 6 This work however shifts the attention to supply and use tables, which really constitute the basic units of the elements of an input-output table and therefore, of the technical coefficients. Six kind of multipliers discussed in the form of multiplier matrices Supply-use based Monte Carlo experiment

7. Sevilla May 2011, WIOD 2 nd Consortium Meeting 7 Multiplier matrices

8. Sevilla May 2011, WIOD 2 nd Consortium Meeting 8 B = input coefficient per unit of industry output C = share of industry output stemming from producing one commodity D = commodity output proportions (market shares) Domestic supply and use table

9. Sevilla May 2011, WIOD 2 nd Consortium Meeting 9 M1 = Industry technology assumption for product by product tables M6 = Product technology assumption for product by product tables M2 = Fixed product sales structure for industry by industry tables M5 = Fixed industry sales structure for industry by industry tables M3 = D T · M1 (industry by product) M4 = C -1 · M6 (industry by product) M7 = D -T · M2 (product by industry) M8 = C · M5 (product by industry)

10. Sevilla May 2011, WIOD 2 nd Consortium Meeting 10 Monte Carlo experiment

11. Sevilla May 2011, WIOD 2 nd Consortium Meeting 11 Empirical application for Spain (2006): Monte Carlo experiment 1.Data sources: SUT (basic prices), 59 ind. x 59 prod. (Eurostat and INE). 2.Two approaches: Supply-side Use-side

12. Sevilla May 2011, WIOD 2 nd Consortium Meeting 12 Randomization of VRandomization of u, f, h, w Reconciliation h, f, p RAS-based Solution SUPPLY-SIDE APPROACH

13. Sevilla May 2011, WIOD 2 nd Consortium Meeting 13 Randomization of VRandomization of u, f, h, w RAS-based Solution USE-SIDE APPROACH

14. Sevilla May 2011, WIOD 2 nd Consortium Meeting 14

15. Sevilla May 2011, WIOD 2 nd Consortium Meeting 15 ITM C X C FPS I X I ITM I X C PTM I X C FIS I X I PTM C X C

16. Sevilla May 2011, WIOD 2 nd Consortium Meeting 16 CONCLUSIONS: 1.The absolute average values of the bias in all cases are not very significant and may probably be considered negligible as in Roland-Holst (1989) and Dietzenbacher (2006). 2.This is just an almost definite answer that deserves further research for a bigger number of iterations (say, 10,000); more countries and/or years and different levels of variability in the randomized supply and use values.

17. Sevilla May 2011, WIOD 2 nd Consortium Meeting 17 Thank you! http://ipts.jrc.ec.europa.eu