estimation efficiency by exploiting information in the cross-equation error covariance matrix. This
estimation procedure takes into account the effects of random price shocks in different markets
on the price level in a particular market. The decision to use the SURE model was also based on
the test of contemporaneous correlations of the residuals from the SURE model. If the coefficient
of contemporaneous correlation between the different residuals is not significant it implies that
the random price shock in one market has no effect on the price level in another market. Under
this condition there is no gain in the efficiency of estimating the various coefficients using SURE
in that SURE estimates are not different from that of OLS estimates and OLS can be used (Judge
et al. 1985).
The manner in which variables are specified in the model is sensitive to assumptions about the
degree of spatial market integration. If markets are assumed to be fully integrated, then prices
in location i would not just be affected by local rainfall and food aid but rather by movements in
these variables all over the country, which would motivate for some national-level indicators.
On the other hand, under the assumption of weak market integration, it would be reasonable to
use region-specific rainfall and food aid variables that vary across equations.
Under the assumption that grain markets are integrated, the SURE model used to analyze the
effects of various factors on the levels of the real price of a given grain for a given number (n) of
markets is formulated as follows:
Pt = μ + IID + DDLIB + RFF + AADDD + PPt_ 1 + Vt
(2)
where Pt is a nx1 vector of real prices of a given grain for different markets at time t, μ is a nx1
vector of regression intercepts, ∏ is a nx11 matrix of coefficients on monthly seasonal dummy
variables, D is 11x1 vector of monthly seasonal variables, Φ is a nx1 vector of coefficients on
liberalization dummy variable DLIB, Ω is a nx1 vector of coefficients on rainfall variable RF, λ
is a nx1 vector of coefficients on food aid variable FAID, γ is a nx1 vector of coefficients on nx1
vector of own lagged price values, and Vt is a nx1 vector of the disturbance term at a time t.
Under the assumption that regional markets are only weakly integrated, the SURE model is
specified in the same way as in (2) above except for the rainfall and food aid variables. These
variables, instead of being a scalar, are now vectors with a dimension of nx1 for n different
markets. Since it is generally believed that Ethiopian grain markets are not fully integrated and
suffer from numerous infrastructural and coordination problems, we adopt the latter specification
in which region-specific food aid and rainfall variables are used to estimate equilibrium prices for
markets in their respective regions. Rainfall and food aid in each region affects market prices in
other regions indirectly through the residual terms in the SURE system estimation procedure.
The expected sign for each variable in (2) are discussed briefly:
Seasonality: There are clear seasonal patterns in Ethiopian grain prices due to different volumes
hitting the market in different months, and also due to costs of storing grain over time. The
coefficients on months following harvest seasons are expected to be lowest, gradually rising to
a peak before harvest.
-6-