Taiwan's trade imbalance and exchange rate revisited

Abstract

This article applies the recently developed econometric method of autoregressive distributed lag (ARDL) model to re-examine the relationship between the exchange rate and bilateral trade imbalance for Taiwan and its several trading partners: China, Hong Kong, the US, Japan, South Korea and Malaysia. The implication of this issue is critical to policy makers, particularly after China and Taiwan joined the World Trade Organization (WTO) in late 2001 and early 2002, respectively. The empirical evidence shows a stable long-run relationship of bilateral trade balance and real exchange rate between Taiwan and its trading partners except Japan. Other findings indicate that except for the US, there is no specific pattern supporting the J-curve phenomenon for these trading partners.

Acknowledgements

We thank Jyh-Lin Wu, Wei-Ming Lee, Song Zan Chiou-Wei, Yung-hsiang Ying, Gerald Baker and participants of the Fifth Annual Conference of Empirical Economics for their valuable comments.

Notes

¹ Earlier studies have employed various approaches to examine the issue for Taiwan; for example, Moreno (1989), Chou and Shih (1991) and Cheung et al. (1997) apply the ordinary least squares (OLS) regression, whereas Kam and Lin (1998) and Chen (2002) use the Johansen cointegration analysis.

² Accordingly, major trading partners and competitors started to closely monitor whether the Taiwanese government's intervention in the exchange rate undermined their interests by any means. The pressure from these trading partners mounted almost immediately.

³ In 2001, Taiwan's economy suffered a negative economic growth rate of −2.3% in 30 years.

⁴ When trade deficit occurs, that is, trade balance is negative, taking logarithmic transformation is meaningless. Therefore, the bilateral trade balance in this study is measured as a ratio of exports to imports so that not only the model could be expressed in logarithmic form but also to make the measure of trade balance unit-free (Kam and Lin, 1998; Arora et al., 2003; Bahmani-Oskooee and Goswami, 2003).

⁵ For an excellent review of the J-curve effect, refer to Bahmani-Oskooee and Ratha (2004).

⁶ In such case, following Kremers et al. (1992) and Banerjee et al. (1998), the error-correction term will be a useful way of establishing cointegration (Bahmani-Oskooee and Goswami, 2003).

⁷ As Peersman and Smets (2002) point out, the use of the industrial production index instead of GDP allows an easier identification of the state of the business cycle, since this series shows a stronger cyclical pattern.

⁸ This study applies the Akaike's Information Criterion (AIC) to determine the optimal number of lags after imposing a maximum of 12 lags on each of the first-differenced variables. Based on the AIC criterion, the optimal ARDL (a, b, c, d) models are ARDL (3, 4, 0, 3), ARDL (6, 3, 1, 0), ARDL (11, 9, 0, 10), ARDL (6, 12, 0, 0), ARDL (11, 2, 0, 0) and ARDL (7, 0, 0, 8) for China, Hong Kong, the US, Japan, Korea and Malaysia, respectively.