Agricultural productivity and convergence: Europe and the United States

Abstract

This article applies the Window Malmquist Index (WMI) approach to measure changes in agricultural Total Factor Productivity (TFP) for the United States and a sample of nine European countries for the period 1973 to 1993. The dataset used in this article is obtained from Ball et al. (2001). The WMI is constructed by combining Data Envelopment Analysis, window analysis with the Malmquist index approach. Furthermore, the ‘Kruskal and Wallis rank test’ is used for testing frontier shifts among observed periods. The article also explores the question of convergence in TFP across the countries under consideration, by testing for β- and σ-convergence, as well as for stochastic or long-run convergence. The results show wide variation in the rate of TFP growth across countries with an average trend growth rate of 1.62%. The results indicate the presence of β-convergence but the absence of σ-convergence for the full period under consideration but the presence of both β- and σ-convergence for the sub-period 1983 to 1993. Finally, a wide spectrum of panel unit root test results support the presence of long-run convergence among the sample countries.

Notes

¹ The output and input data are obtained from the Appendix A (Ball et al., 2001). In particular, the output data are obtained from Table A.2 (p. 23), the capital input data from Table A.4 (p. 24), the land input data from Table A.6 (p. 25), the labour input data from Table A.8 (p. 26) and the intermediate input data from Table A.10 (p. 27).

² This reform has been the basis of all subsequent CAP reforms. In particular a subsequent major CAP reform constitute the Agenda 2000 reform agreed in March 1999 to cover the period 2000 to 2006 but mandated a mid-term review in 2003. Note that the MTR sets out the CAP framework until 2013, but it is unlikely to be the last CAP reform because the level of direct payments will be subject to the decision on the EU's Financial Perspective (or medium-term budget) over the period 2007 to 2013 which will determine the resources available for CAP expenditures.

³ The study by Sueyoshi and Aoki (2001) provides the DEA models to calculate IEIs and TSEs measures.

⁴ The study by Sueyoshi and Aoki (2001) provides the DEA models to measure and TSE ^{t − 1∪t} .

⁵ In estimating TFP trend growth rates presented in Table 2 the following regression model is used: In(TFP) = intercept + GROWTHRATE × time

⁶ The first test (LL_1) sets Z_it = 0, i.e. without intercept and time trend; the second (LL_2) sets Z_it = δ ₀, i.e. with intercept and no time trend; the third (LL_3) sets z _it  = δ₀ + δ _i t, i.e. with intercept and time trend; the fourth (LL_4) sets Z_it = ν _t , i.e. without intercept and time trend but with time-specific effect; the fifth (LL_5) sets Z_it = a_i , i.e. without intercept and time trend, but with individual specific effect; the sixth (LL_6) sets z _it  = α _i  + η _i t, i.e. with individual specific effect and individual time trend; and the seventh (LL_7) sets Z_it = 0, i.e. without intercept and time trend, but with serial correlation across time period.

⁷ The first test (LL_8) sets Z_it = 0, i.e. without individual specific effect and individual time trend; the second (LL_9) sets Z_it = a_i , i.e. with individual specific effect, but without time trend; and the third (LL_10) sets z _it  = α _i  + η _i t, i.e. with individual specific effect and individual time trend.

⁸ The first test (HT_1) sets Z_it = 0, i.e. without intercept and time trend, and corresponds to LL_1 test; the second (HT_2) sets Z_it = a_i , i.e. without intercept and time trend but with individual specific effect, and corresponds to LL_5 test; and the third (HT_3) sets z _it  = α _i  + δ _i t, i.e. with individual specific effect and individual time trend and corresponds to LL_6 test.

⁹ The first test (IPS97_1) sets Z_it = α _i , i.e. with individual specific effect but without time trend; and the second (IPS97_2) sets z _it  = α _i  + η _i t, i.e. with individual specific effect and individual time trend.

¹⁰ The first test (IPSLM_1) sets Z_it = α _i , i.e. with individual specific effect but without time trend; and the second (IPSLM_2) sets z _it  = α _i  + η _i t, i.e. with individual specific effect and individual time trend.

¹¹ The panel unit root tests presented in Table 7 were estimated using the GAUSS econometric package and the sub-routines from Chiang and Kao (2002).