Productivity growth and convergence in US agriculture: new cointegration panel data results

Abstract

Dynamic effects of health and inter-state and inter-industry knowledge spillovers, Total Factor Productivity (TFP) growth and convergence in US agriculture are examined using recently developed procedures for panel data and a growth accounting model. Strong evidence is found to support the hypothesis that TFP converges to a steady state. Health care supply in rural areas and research spillovers from other states and from nonagricultural sectors are found to have significant impacts on the productivity growth rate in both the short and the long run. These results suggest a richer set of opportunities for policy makers to enhance productivity growth than previously considered.

Acknowledgements

This project was supported by the Washington Agricultural Research Center, the USDA Cooperative State Research, Education and Extension Service Hatch grant WPN000275, and the USDA Economic Research Service. The views expressed are those of the authors and do not necessarily correspond to the views or policies of ERS or the US Department of Agriculture.

Notes

¹ Innovation is only partly a public good. It is largely nonrival in consumption, but because of patents, it may not satisfy the nonexcludability condition for a public good. Innovation from public research is generally regarded as more of a public good, but changes in patent law over the last 25 years have permitted public entities to also patent their discoveries.

² In addition to human capital and R&D spillovers, this article also examines the impact on agricultural productivity growth of average farm size and selected policy variables, including investments within the state on public and private R&D and public extension.

³A dummy variable is also included in the estimation equation to account for the impact of the 1983, 1 year, US Department of Agriculture Payment-In Kind (PIK) programme.

⁴ Before estimating the ECM, we first conduct stationarity and cointegration tests for all variables in the equation.

⁵ The critical values of these test statistics are tabulated in Pedroni (1999, Table 2) for up to seven explanatory variables. Because we use nine explanatory variables, the adjustment terms for the panel cointegration tests were obtained as suggested by Pedroni (1999) using Monte Carlo simulation. Following his approach, we took 10 000 draws of nine independent random walks (i.e. the number of regressors) so T=10 000. Under the alternative hypothesis, all the statistics diverge to negative infinity (one to positive infinity). Therefore, each is a one-sided test for which a large positive value for the ‘panel v-statistic’ or a large negative value for the other tests results in rejection of the null hypothesis of no cointegrated relation among the variables.

⁶ There are various other ways to measure farm size, e.g. acreage or gross value of sales or product. Although these traditional ways are easy to quantify, acreage does not account for differences in the productive capacity of the land input (Yee and Ahearn, 2005), and gross value of sales is an output measure rather than an input capacity measure.

⁷ Applying a nonparametric approach, Chavas and Cox (1992) found that the effects of private research on the US agricultural production increased slowly in the first 7 years, increased rapidly in the next 8 years, then decreased, with no effects beyond 23 years.

⁸ These lag lengths are similar to those found by Makki et al. (1999), Thirtle et al. (2002) and Liu and Shumway (2006, forthcoming).

⁹ For I(2) variables, third differences were included.

¹⁰ These estimates were computed using a GAUSS programme by Pesaran et al. (1999).

¹¹ Although the average convergence speed exceeds 40%, it exhibits high volatility across the 48 states. For example, the state-specific convergence speed (not reported here) estimated from ECM II ranges from a low of 10% for Connecticut to a high of 105% for South Carolina. Intuitively, Connecticut is the state relatively closest to its steady state while South Carolina is furthest away from its steady state.

¹² Our study advances previous studies in three ways. First, our model accounts for data nonstationarity and cointegration relationships using more reliable test procedures. Second, the effects of the policy variables are assessed within a dynamic panel data framework. Third, we consider the impacts of health care supply and inter-state and inter-industry innovation spillovers.

¹³ However, the long-run impact of an increase in public extension investments is insignificant in the presence of dynamic impacts from exogenous shocks (ECM II).

¹⁴ It is also possible that the procedure used to allocate private patents may be driving some of these unexpected results with regard to private research.

¹⁵ This procedure was used since it results in least bias among the estimators (Nair-Reichert and Weinhold, 2001). In the mixed fixed and random coefficients model, the coefficient on the lagged dependent variable is specific to the group and the coefficients on the exogenous explanatory variables are treated as being randomly distributed.