Sustainability of current accounts: evidence from the quantile unit-root test

ABSTRACT

Regarding the stationarity of current accounts, previous studies used panel unit-root tests to improve the power of augmented Dickey–Fuller (ADF) test. This paper applies a quantile autoregression (QAR) approach to improve ADF test in the presence of outliers, and found that first, the stationarity was present in a QAR framework, rather than ADF test. Second, current accounts exhibited symmetric (e.g. Taiwan) and asymmetric patterns, which showed that positive shocks in larger quantiles induce current accounts to adjust towards a long-term equilibrium for Korea, Thailand, the Philippines and Singapore. Japan exhibited an asymmetric pattern in response to negative shocks in smaller quantiles.

KEYWORDS:

• Current accounts sustainability
• augmented Dickey–Fuller (ADF) test
• quantile autoregression (QAR) test
• unit-root test
• asymmetric patterns

JEL CLASSIFICATION:

• C32
• F32

Notes

¹ Prior studies proposed the importance of the issue, e.g. the intertemporal model of current account indicates that current accounts are stationary (Wu 2000); stationarity of current accounts is consistent with implications of the intertemporal model, thus supporting its validity (Obstfeld and Rogoff 1996, 90).

² Shiller and Perron (1985) observed that the power of ADF tests is low during short time spans. Otto (1992) and Ghosh (1995) used the approach to test the validity of the intertemporal model. They observed that current accounts are nonstationary by using the conventional unit-root test proposed by Dickey and Fuller (1979). These works include Husted (1992) and Fountas and Wu (1999).

³ Equation (3)–(5) estimates α₀(τ) that denotes the coefficient at τth quantiles and captures cumulative magnitude of shocks from period t–q to period t; one-off shocks of a similar magnitude, which are classified as falling into the same quantile, are the shocks that determine the fit of this quantile. α₀(τ) denotes the magnitude of these shocks (Nikolaou 2008). Unlike ADF tests focusing on conditional mean, QAR model estimates α₁(τ) that can identify localized reversion within various quantiles. A stationarity in current accounts at a specific quantile suggests that the persistence of the series is destroyed by shocks within this quantile. By contrast, a unit-root behaviour in a certain quantile reveals a strengthening of the persistence. A series can exhibit localized unit-root behaviours in a certain quantile, followed by reverting patterns in other quantiles, thereby inducing stationarity in the overall process (Koenker and Xiao 2004).

⁴ This calculation method has been used in previous research (Sheffrin and Woo 1990; Wu 2000; Wu, Chen, and Lee 2001; Taylor 2002; Lau and Baharumshah 2005; Lau, Baharumshah, and Haw 2006; Boileau and Normandin 2008; Raybaudi, Sola, and Spagnolo 2004).

⁵ Some researchers (e.g. Gauss, Laplace and Legendre) suggested that a minimization of absolute deviations might be preferable to the least-squares estimator when the reliability of some observations is dubious (Koenker and Bassett 1978, 35). The tests based on regression quantiles in linear models can substantially improve least-squares methods in non-Gaussian situations (Koenker and Bassett 1978, 47). This is because the least-squares methods estimate the estimator of conditional mean of entire samples and cannot capture estimator of outliers, which must be obtained from a few regression quantiles (Koenker and Bassett 1982, 1978, 48). Tukey (1975) indicated that an alternative estimator should be devised by modifying the least-squares estimator and reducing its notorious sensitivity to outliers. This is especially true when error distributions are longer tailed than the Gaussian distribution (Koenker and Bassett 1982, 48)

⁶ By applying resampling method, this paper generated bootstrapped distribution and asymptotic critical values of t_n(τ) (see Table A1)

⁷ For example, half-lives present 8 and 9 quarters in the 0.4 and 0.6 quantiles, whereas they respectively decrease to 2 and 5 quarters in the 0.9 and 0.1 quantiles (Table 1, panel C).

⁸ For example, the half-lives of Korea decreased from infinity (0.1 quantile) to 2.11(0.9 quantile) (see Table 1, panel C).