Page 13 - Azerbaijan State University of Economics

M.R. Jamilov, R.M. Jamilov: Factor-Augmented J-Curve

variables are stationary in first-differences. In general, this is not existential for the
purposes of ARDL modeling, as explained earlier in the paper. Nevertheless, this would
have produced a spurious regression problem had we resorted to simple OLS
techniques. We now run 9 ARDL regressions using the factorial balance of trade
parameters as our dependents and the exchange rate as the main covariate.
Table 6 presents the preliminaries for the bounds-testing procedure. The F-
statistic is statistically significant for 6 of the 9 cases, suggesting that in for those
regressions long-run cointegration is achieved. For the case of the final factor – Mineral
and Quarrying Goods – the F-stat is insignificant but the error correction term is both
negative and significant. We can therefore conclude that cointegration is established for
th
this 7 factor as well. The average value of the error correction term is 0.7, suggesting
that, on average, an exchange rate shock gets transmitted to the exports-imports
relationship within 8 months. In other words, less than one calendar year is required for
a devaluation to affect the balance of trade dynamic. The high, considering our small
sample size and one single covariate, coefficients of determination – R-squared – simply
imply that exchange rate works well as an explanatory variably of the total variation in
egression outcomes. The lag order has been chosen according to the Schwarz-Bayesian
criterion (SBC). In rare cases when the SBC suggested zero lags for the exchange rate,
we forced one lag for reasons of short-run analysis.
Table 7 reports the main results of this paper. Three of the nine factors suggest
the fulfillment of the Marshall-Lerner condition: the long-run elasticity of the
exports-imports ratio in response to an exchange rate shock is positive and
significant. For other factors, the impact is not significant at acceptable levels.
Estimates of the lagged exchange rate variable are negative for almost all cases, and
statistically significant for the 3 factors for which the Marshall-Lerner condition
holds. This points at the presence of the J-curve effect: the collapse of the trade
balance ratio in the short run due to the price effect and the volume-driven
expansion in the long-run. All in all, negative values for the lagged exchange rate
estimate and positive long-run elasticity estimates for factors 1, 6, and 8 suggest that
both the M-L condition and the J-curve effect hold true for “All Industries”, “Heavy
Metals and Inorganic Chemicals”, “Agriculture and Organic Chemicals” factors.
Note that the M-L and the J-curve hypotheses are supported for the trade
sector as a whole since the factor “All Industries”, representing all 59 industries in
the sample, improves following a dollar depreciation. As far as disaggregated factors

13