Is There a J-Curve Effect in the Korean Service Trade?

ABSTRACT

In this article, we test the asymmetric J-curve hypothesis using data from 10 Korean service industries. When we tested for a symmetric J-curve using the linear autoregressive distributed lag (ARDL) approach, we found support for the J-curve in one industry and the inverse J-curve in two industries. However, when we tested the asymmetric J-curve using the nonlinear ARDL approach, we found support for the asymmetric J-curve in four industries and the asymmetric inverse J-curve in four more industries. The increase from three to eight cases is attributed to the nonlinear adjustment of the real effective exchange rate of the won.

KEYWORDS:

• 10 service industries
• asymmetric J-curve
• asymmetry cointegration
• Korea

Acknowledgments

The valuable comments of the two anonymous reviewers are greatly appreciated. Any remaining error, however, is our own.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/08853908.2024.2379350

Notes

¹ A devaluation or depreciation makes exports cheaper in terms of foreign currency and imports expensive in terms of domestic currency, expecting to improve the trade balance. Symmetric effect means if an x% depreciation improves the trade balance by y%, an X% appreciation will worsen it by the same y%. Asymmetric effect means if an x% depreciation improves the trade balance by y%, an X% appreciation will worsen it by an amount that is different than y%. As Bahmani-Oskooee and Baek (2016) argued, asymmetric effects could be due to different expectations when a currency depreciates versus when it appreciates. Additionally, when a currency depreciates, producers may react faster to satisfy the demand for exports. However, this may not be the case for currency appreciation since they have to add to inventories rather than adjust production.

² Indeed, due to a lack of other studies testing the J-curve effect in Korean service trade, we are unable to include a literature review section.

³ For proof of this test, see Bahmani-Oskooee (2020). For long-run convergence, ${\hat{\lambda }}_0$ must be negative and its estimated value measures the speed with which all variables adjust toward long-run equilibrium.

⁴ Indeed, we verified this by applying the augmented Dickey-Fuller (ADF) test to the level and first-differenced variables. Table A1 in the online Appendix reveals that we have no I(2) variable. They are a combination of I(0) and I(1).

⁵ For other applications of these models, see Nusair (2017), Aftab, Syed, and Katper (2017), Arize, Malindretos, and Igwe (2017), Hajilee and Niroomand (2019, 2021), Baek (2020), and Saha (2022).

⁶ By meaningful, we mean cointegration is supported at least by either the F-test or t-test.

⁷ Note that four additional diagnostics statistics are reported in Panel C. The Lagrange multiplier statistic is reported as LM and is used to identify industries in which the residuals suffer from first-order serial correlation. Clearly, they do not since the LM statistic is insignificant in almost all models. Ramsey’s RESET test is also insignificant, rejecting misspecified models. We also applied the CUSUM and CUSUMSQ tests to establish the stability of all estimates. These tests are reported as CS and CS², and stable estimates are identified by “S” and unstable estimates by “US.” Clearly, all estimates are stable by both tests.

⁸ The Wald test, reported as Wald-L, is only significant in the case of the government services industry. Note also that the other diagnostics are similar to those of the linear models and need no further explanations. The 2008 financial crisis dummy is significant in the construction, financial, and other business services categories. When adding the transportation sector from the linear models to the list, a total of four industries seem to be affected by the global financial crisis of 2008.

⁹ Our findings for Korea are similar to those of Bahmani-Oskooee and Karamelikli (2023) for the U.S. in that nonlinear models yield more support for the J-curve effect compared to linear models. Future research could provide additional support by using data from other countries.