Bridging Micro- and Macro-Analyses of the EU Sugar Program: Methods and Insights

Appendix. Econometric procedures and estimates

Stationnary tests

The time series used in the econometric estimations were tested for stationarity using the
augmented Dickey-Fuller (ADF) with drift as well as with the Kwiatkowski-Phillips-
Schmidt-Shin (KPPSS) tests for stationarity (Kwiatkowski et al 1992). As it is well known,
none of these tests is fully conclusive, especially when used on small samples. The ADF
tests for the rejection of non stationarity. If the value of this statistic is lower than the critical
value of -3.6, then this means that we can reject at 5% level of confidence the hypothesis of
non-stationarity. KPSS test has stationarity as the null hypothesis and the unit root as the
alternative. Again, if the value of this statistic is lower than the critical value of 0.146, we can
not reject at the 5% level of confidence the hypothesis of stationarity.

Table A1 shows the values of these two tests for all variables estimated (sugar beet acreage
by Member States) as well as for the prices entering the French equation. The number of lag
in the ADF test is determined according to the Durbin Watson test. As it is often the case,
these two tests are not always consistent. For instance, according to the ADF test, we can not
reject the assumption that the UK sugar beet area variable is non stationary but according to
the KPSS test, we can not reject the hypothesis of stationarity. It appears in Table A1 that for
no series did we find the two tests in favor of non stationarity. With this fragile conclusion,
we are only able to be cautious about the inferences on our econometric estimates.

Table A1. Augmented Dickey Fuller and KPSS tests

Econometric estimation

The equation that is estimated corresponds to (7) in the paper. The series are annual data
from 1981. Data sources include Eurostat for acreages and p2 (spot sugar prices, London) is
from the annual report of the Confédération Générale de la Betterave. Data for p1 including
various taxes are taken from van der Linde et al (2002) and from Confédération Générale de
la Betterave. All prices were deflated by the GNP price using Eurostat data. Price and yields
are expressed as an arithmetic average over the past three years. Various specifications with
trends and lagged variables were tested, but the naïve expectation specification fit the data
best.

25

<<   <   22   23   24   25   26