New Evidence on Exchange-Rate Volatility and Export Flows in Thailand: Nonlinearity and Asymmetric ARDL Investigation

ABSTRACT

This is the first study of Thailand’s export demand relations utilizing the new nonlinear autoregressive distributed lag (NARDL) technique by Shin, Yu, and Greenwood-Nimmo. The study reports long-run elasticities and exposes the short-run adjustment of real exports to changes in foreign economic activity, relative export price, and exchange-rate risk using long-run and short-run asymmetry as well as nonlinear asymmetric cointegration. Results from nonlinear and asymmetry cointegration analysis and short-run asymmetric error-correction modeling indicate cointegration and negative effects of exchange-rate risks on export volume in both the long run and short run.

KEYWORDS:

• Foreign activity
• relative price
• exchange-rate volatility
• export flows
• asymmetry and nonlinear asymmetric cointegration

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ The inflation rate for Thailand averaged 4.81% over the period of 1973:1 to 2016:4. Inflation rose sharply in the mid-1970s and remained highly volatile until early in the 1980s. Inflation then decreased and remained low until the late 1980s and rose somewhat following the financial crisis of 1997 to 1998.

² Many economists refer to it as the world’s first financial crisis since the Great Depression of the 1930s.

³ Risk-neutral exporters will take advantage of exchange-rate volatility to increase their export returns through more exports.

⁴ The issue of nonlinearities is raised here because De Grauwe (1988) argues that a risk-averse firm would choose to raise supply to compensate for a sufficiently large revenue loss.

⁵ This implies that producers tend to adjust product prices more quickly to increases in cost than they do to decreases in cost.

⁶ Hysteresis has a tendency to cause firms to hestitate in their decision making and to accept increasing costs when the domestic currency appreciates, creating an asymmetric competitive effect and, thus, asymmetric exposure. Therefore, firms can no longer cherry-pick their investments. Moreover, exporters hold foreign-currency prices constant to avoid endangering “sunk costs” or consumer allegiance through current pricing practices.

⁷ Intuitively, a partial sum of negative (positive) changes at time t is the cumulative sum of all $\mathrm{\Delta }x$ prior to time t where positive (negatives) changes have been replaced by zeros.

⁸ The constant term has a t-value of 1.68, which is statistically significant at better than the 10% level, so taking a first difference of the model in levels with a significant constant and no serial correlation will eliminate the constant and introduce misspecification in the first difference equation. Clements and Mizon (1991, 894) point out that treating the exchange rate as I(1) does not mean that it will yield an appropriate model. Moreover, Campbell and Perron (1991) note that it would be econometrically advantageous to treat a stationary variable with a significant drift term as a nonstationary one for forecasting purposes. Furthermore, Nelson (1990a, 1990b) and McKenzie (1997) note that the order of the AR model does not have any significant effect on the estimated GARCH model. Finally, the ADF test rejects the series as I(1) at the 10% level (see Appendix B online).

⁹ The appendix can be found online at www.tandfonline.com/uitj.

¹⁰ This is so because the critical values were simulated based on the assumption that the variables must be I(0) or I(1). Its use under these circumstances could lead to spurious results.

¹¹ Caution is needed for these results because Bahmani-Oskooee and Goswami (2003) have argued that the value of the F-statistic may be sensitive to the number of lags imposed on the differenced variables, whereas Arize et al. (2000b) have argued about the sensitivity of the long-run relationship to optimal lag selection of the model.

¹² Briefly, the Hatanaka process requires that one obtains estimates of the regressand by an instrumental variable approach. The instrument for the lagged dependent variable is the predicted value from a regression of the dependent variable ΔX_t on a constant, four lagged values of Δσ_t, ΔP_t, and ΔY_t (as well as as the current values of all variables except ΔX_t). The predicted value for the dependent variable is used as an instrument for the lagged dependent variable in estimating the model to generate consistent standard errors; the model was re-estimated subject to errors following a first-order autocorrelation process. The final model, thus, accounts for persistence effects and the effects of a shock last period on exports (see Green (1993), for a more detailed explanation of the process).

¹³ Because both the DW statistics (i.e., 2.04 and 1.88) and (4 – DW) are greater DW_0.01 (1.6), there is no statistical evidence that the error terms are positively or negatively autocorrelated.

¹⁴ The appendix can be found online at www.tandfonline.com/uitj.