summarized by the average price trends during the two regimes. A reduced form approach is very
well suited to this task.
Before the various grain price series are subjected to econometric analysis, time series properties
such as stationarity of the different price series are investigated using the standard procedures.
First, the deseasonalized real price series are obtained as a residual from the regression of real
price series on seasonal dummy variables. Residual series are obtained from the following OLS
regression:
11
(1)
Pj,t = a + D № + ⅝∙,t
n = 1
where Pij,t is the price of jth grain for the ith market at time t, a is a regression constant, Dn is a
monthly seasonal dummy variable for the nth month, βn is a coefficient on the seasonal dummy
variable and εij,t is a regression residual which represents the deseasonalized real price series at
time t for the jth grain in ith market. Stationarity tests on deseasonalized real price series and price
spreads were performed using the augmented Dicky-Fuller (ADF) and Philips-Peron (PP) tests.
The stationarity test results on price levels are presented in Appendix 4. The results reject the
hypothesis of non-stationarity in 9 of 16 cases using the PP test and in 8 of 16 cases using the
ADF test at *P<0.10.5 Both tests provide strong evidence of stationarity in the all cereal prices
for the central terminal market in the country, Addis Ababa. Assuming that there is some long-
run relationship between Addis prices and prices in other regional markets, this would then
require an assumption of stationarity among the other market price series as well. Given the
unlikelihood of no long-term relationship between prices in the central market in the country and
other major wholesale markets, even in a country with relatively weak infrastructure such as
Ethiopia, it would appear from the data that the most reasonable assumption to make is one of
stationarity, even though the hypothesis of non-stationarity could not be rejected for all markets.
The results of unit root tests on price spreads are given in Appendices 5 to 7. The results here
clearly support the hypothesis of stationarity. In light of these results, the models for both price
levels and price spreads are specified in levels.
Equilibrium cereal prices in each market were hypothesized to be affected by a set of factors
including seasonality, grain market liberalization, food aid released in the region, and rainfall. For
each commodity, prices were estimated as a system, using a Seemingly Unrelated Regression
(SURE) estimation process. Iterative SURE is a maximum likelihood estimator that increases
5 This finding is in contrast to Dercon (1994) who concluded that at least teff prices in Ethiopia were
non-stationary over the 1985-93 period. A potential cause of the different finding is that we deseasonalized
the price data to account for seasonal effects before testing for unit-roots. Since grain prices typically
exhibit clear seasonal effects, it would seem to us important to purge the data of this effect before testing
for unit-roots. Without doing so, there may be a tendency to accept the hypothesis of non-stationarity when
in fact this may not be so, which would then motivate for different model specifications that would not be
appropriate if in fact the data were stationary.
-5-