Long-run aggregate import demand function in Taiwan: an ARDL bounds testing approach

Abstract

This article adopts the bounds test, developed by Pesaran et al. (2001), to determine whether there is a level long-run relationship exists between Taiwan's real import demand function and it determinants, namely real domestic income and relative prices. It is found that aggregate import quantities and their determinants do indeed exhibit a level long-run relationship. In addition, the empirical results show that estimated short-run elasticity and long-run income elasticity are both elastic but that short-run income elasticity is considerably greater than that of its long-run counterpart. This indicates that economic growth should have a relatively greater negative impact on trade balance in the short-run than in the long-run.

Notes

¹ As noted by Hong (1999, p. 3), ‘ … import demand in a market economy can be fully modeled by two determinants: income and relative prices. The other factors can all be subsumed within these two factors, at least theoretically.’

² As noted by Pesaran et al. (2001, p. 312), ‘in testing the null hypothesis of the absence of the level long-run relationship in Equation 5(5), namely ψ = δ = 0, it is important that the coefficients of the lagged change remain unrestricted; otherwise, these tests could be subject to a pre-testing problem. However, for the subsequent estimations of the level effects and short-run dynamics of the adjustments, the use of more parsimonious specifications seems advisable.’

³ The UECM model is also estimated with a deterministic trend and the bounds test is conducted. The optimal lag selected by the AIC and SBC are 3 and 4, respectively, values which are the same as those from the UECM model without deterministic trend. The computed F-statistic of the bounds test is less than the lower critical value, suggesting that there is no long-run relationship between the regressors. However, the estimate of the deterministic trend is also not significant at the conventional level, indicating that the UECM model without a deterministic trend is the preferred model.

⁴ The graphs are available from the author upon request.

⁵The robustness of the ARDL model with respective to the Johansen and Juselius (1990) method is also checked and the estimated results from the two methods are quite alike. These results are available from the author upon request.