Page 82 - Azerbaijan State University of Economics

THE JOURNAL OF ECONOMIC SCIENCES: THEORY AND PRACTICE

Based on the unit root tests’ results, it is clear that each of the two

demand equations has at least two I(1) variables. Therefore, the Johansen
approach to estimation is applicable.

Vector Auto Regressions
When constructing the VAR models, one lag was initially selected

as the starting point of analysis. Following the lag structure test,

however, the export demand VAR is continued with three lags, and the
import demand VAR – with two. An autocorrelation test shows no

problems with error autocorrelation. Similarly, the normality test affirms
that all residuals are normally distributed. Generally, problems of

autocorrelation and normality can be solved by adding more lags to the

system. However, our analysis proceeds without any required
interventions. Results of the autocorrelation and normality tests are not

presented here for brevity.
Vector Error Correction Models

The Johansen co integration test is performed for both demand
functions to reveal the number of long-run co integrating equations.

Each of the two functions has one co integrating equation, which

significantly eases the economic interpretation of the model. Further, the
VEC model is estimated with the N-1 lags, where N is the number of

lags in the respective preliminary VAR. Selection of deterministic
elements is the following: intercept but no trend for the co integrating

equation, and no intercept for the VAR. The long-run equations are
presented below:

ln(X t) = 101.57 + 1.465(RFX t) + 11.467(lnY eur) + ε t (5)

ln(IM t) = -.610 – 0.113(RFX t) – 0.09(lnY az) + ε t (6)

82