An Approach to the Determinants of the Agricultural Output Dynamics in the USA - The Agricultural Economics of the 21st Century - page 15

Table 2.2 Results obtained with time series econometric techniques, based on the function of
production model (linear model obtained with logarithms), for agricultural output in the period
1961-2012
Model Prais-Winsten

Constant 9.626*

(5.570)
[0.000]

Agriculture value added per worker (constant 2005 US$) 0.870*

( 5.770)
[0.000]

Employment in agriculture (% of total employment)
Augmented Dickey-Fuller test for unit root

6.311*
[0.000]

EG-ADF test for co-integration 1.809

[0.376]

Portmanteau test for white noise for autocorrelation 224.764*

[0.000]

Durbin's alternative test for autocorrelation 0.342

[0.558]

Breusch-Godfrey LM test for autocorrelation 0.388

[0.533]

Breusch-Pagan/Cook-Weisberg test for heteroskedasticity 0.710

[0.398]

Ramsey RESET test using powers of the fitted values 3.720*

[0.024]

LM test for autoregressive conditional heteroskedasticity (ARCH) 1.362

[0.243]

Note : *Statistically significant at 5 %

agricultural value added in percentage of the GDP), farming productivity (Agricul-
ture value added per worker at constant 2005 prices), the population in urban
agglomeration, and the GDP per capita. On the other hand, there is a strong, positive
relationship between the dependent variable and, namely, the agricultural land
percentage and the weight of the rural population.
The results obtained in Table 2.2 with the econometric time series estimations
show that there is, indeed, a negative and strong, statistically significant, relation-
ship between agricultural output and farming productivity. Considering the form as
the values of the variables presented (the output in the percentage relative to others
sectors) and the productivity in absolute values, these results only mean that the
improvements in productivity were not enough to reduce the decrease in the weight
of the agricultural GDP in the whole US economy. The results for the several tests
considered to evaluate the autocorrelation, the co-integration of the variables, and
the heteroskedasticity confirm the absence of these statistic infractions. The Ram-
sey RESET test, using powers of the fitted values, shows a lack of variables and
because of this finding the model was again estimated with other variables,