The J-curve and bilateral trade balances of Indonesia with its major partners: are there asymmetric effects?

ABSTRACT

With advances in econometric methodologies, old concepts are now receiving a renewed attention and the J-curve is no exception. Previous research that tested the phenomenon assumed that the effects of exchange rate changes on the trade balance are symmetric. Asymmetry analysis and asymmetry cointegration is the new direction that is currently being considered. We add to this later literature by considering the asymmetric effects of exchange rate changes on the bilateral trade balances of Indonesia with each of her major trading partners after accounting for Asian Financial Crisis of 1997 and Global Financial Crisis of 2008. We find support for short-run asymmetric effects in almost all the cases, short-run cumulative or impact asymmetric effects in the trade with five partners and long-run significant asymmetric effects in five bilateral models.

KEYWORDS:

• Indonesia
• trading partners
• J-curve
• nonlinear ARDL
• asymmetry

Acknowledgment

Valuable comments of an anonymous referee and the editor are greatly appreciated. Remaining errors, however, are our own.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Most common reasons for the J-curve pattern are said to be the adjustment lags. See Bahmani-Oskooee (1985) for details.

2. Bahmani-Oskooee and Alse (1994), Lal and Lowinger (2002) and Onafowora (2003) are other studies that have included Indonesia in their sample and again, have used a linear model.

3. This section closed follows Bahmani-Oskooee and Fariditavana (2016) who used such methods to estimate bilateral trade balance models between the U.S. and six of her major trading partners.

4. Therefore, TB_i = M_i /X_i where M_i is Indonesian imports and X_i is her exports.

5. We notice that they are the same if we solve Equation (1(1) $LnTBi,t=a+bLn{Y}_{\text{INDO},t},+cLn{Y}_{i,t}+dLnREX{}_{\text{i,t}}+{ϵ}_t$(1) ) for error term and lag the solution by one period.

6. For precise normalization procedure, see Bahmani-Oskooee and Fariditavana (2015).

7. Since critical values account for degree of integration of variables in a given model, there is no need for pre-unit root testing under this method and as they argue, variables could be combination of I(0) and I(1) which are the properties of almost all macro variables. Indeed, this is said to be another advantage of this method.

8. For some other application of partial sum concept see Apergis and Miller (2006) on the effects of U.S. stock market on consumption; Verheyen (2013) on interest rate pass-through mechanism to deposit rates; and Bahmani-Oskooee and Friditavana (2014) on testing the S-curve. For application of the linear model see De Vita and Kyaw (2008), Halicioglu (2007), Narayan et al. (2007), Dell'Anno and Halicioglu (2010), Tayebi and Yazdani, (2014) and Hajilee et al. (2014).

9. Note that as argued by Bahmani-Oskooee and Fariditavana (2016), expected sign of normalized coefficient estimates of POS and NEG variables in model (4(4) $\begin{array}{l}\Delta LnT{B}_{i,t}=a\text{'}+{\displaystyle \sum _{k=1}^{n1}b{\text{'}}_k\Delta LnT{B}_{i,t-k}}+{\displaystyle \sum _{k=0}^{n2}c{\text{'}}_k\Delta Ln{Y}_{\text{INDO},t-k}^{}}+{\displaystyle \sum _{k=0}^{n3}d{\text{'}}_k}\Delta Ln{Y}_{i,t-k}^{}\\ +{\displaystyle \sum _{k=0}^{n4}e{\text{'}}_k}\Delta PO{S}_{t-k}+{\displaystyle \sum _{k=0}^{n5}f{\text{'}}_k\Delta NE{G}_{t-k}}+{\theta }_{0}LnT{B}_{i,t-1}+{\theta }_{1}Ln{Y}_{\text{INDO},t-1}^{}\\ +{\theta }_{2}Ln{Y}_{i,t-1}^{}+{\theta }_{3}PO{S}_{t-1}+{\theta }_{4}NE{G}_{t-1}+{\xi }_t\end{array}$(4) ) are the same as that of REX in model (2(2) $\begin{array}{l}\Delta LnT{B}_{i,t}=\alpha +{\displaystyle \sum _{k=1}^n{\beta }_{t-k}}\Delta LnT{B}_{i,t-k}+{\displaystyle \sum _{k=0}^n{\delta }_{t-k}}\Delta Ln{Y}_{\text{INDO},t-k}+{\displaystyle \sum _{k=0}^n{\gamma }_{t-k}}\Delta Ln{Y}_{i,t-k}\\ +{\displaystyle \sum _{k=0}^n{\pi }_{t-k}}\Delta LnRE{X}_{i,t-k}+{\lambda }_{1}LnT{B}_{i,t-1}+{\lambda }_{2}Ln{Y}_{\text{INDO},t-1}+{\lambda }_{3}Ln{Y}_{i,t-1}\\ +{\lambda }_{4}LnRE{X}_{i,t-1}+{\mu }_t\end{array}$(2) ).

10. Note that Pesaran et al. (2001, p. 303) also tabulate an upper and a lower bound critical value for this test. However, since their values are for large sample, we use critical values provided by Banerjee et al. (1998, p. 276) for 100 observations.

11. Since part of the trade by some partners of Indonesia go through Singapore and Hong Kong, it is possible for that part to be counted twice. This may introduce some bias to our results. However, the issue cannot be too serious if double counting takes place on both imports and exports side since the trade balance is measured as the ratio of imports over exports.

12. As we have indicated in the table by # and %, Asian Financial Crisis dummy was significant in the linear models of Australia, Japan, Malaysia, Korea, U.K. and the U.S. and in nonlinear models of Australia, Malaysia, U.K. and the U.S. The 2008 Global Financial Crisis dummy was significant in the linear models of Australia, Hong Kong, Malaysia, the Philippines and the U.S., and in the nonlinear models of Australia, Hong Kong, Malaysia and the U.S. Clearly, both crises have affected more Asian partners than non-Asian partners except the U.S., which is one of the largest partners. However, trade with the largest partner, China, is not affected and this could be due to an inelastic demand for Chinese goods.

13. As robustness check we also used Schwarz Bayesian Criterion (SBC) to select optimum models. While there was no change in the long-run asymmetric effects in the cases of Australia, Japan, Singapore, Korea and the U.K., the short-run impact asymmetry was reduced from five to three countries, i.e. to Australia, China and Korea.