Anatolia – An International Journal of Tourism and Hospitality Research

1.

Introduction

*Email: lee.chew-ging@nottingham.edu.my
ISSN 1303-2917 print/ISSN 2156-6909 online q 2012 Taylor & Francis http://dx.doi.org/10.1080/13032917.2012.701596 http://www.tandfonline.com

Anatolia – An International Journal of Tourism and Hospitality Research

349

350

C.G. Lee

Anatolia – An International Journal of Tourism and Hospitality Research

351

352

C.G. Lee

Table 1. The results of ADF and KPSS tests. Variable Level EX IM GDP TOU

*,** and *** Statistically significant at the 10%, 5%, and 1% levels, respectively. t-Statistic and LM-statistic are reported for ADF and KPSS tests, respectively. The brackets beside t-statistic indicate the number of lagged first differences of ADF selected based on the Schwarz information criterion. The brackets beside LM-statistic indicate the choice of bandwidth parameter in the Bartlett-kernel-based sum-ofcovariances estimator selected based on Newey –West data-based automatic bandwidth parameter methods.

suggest that TOU is stationary at level but non-stationary at first difference. This type of property is impossible to occur. The results of KPSS test suggest that TOU is stationary at level and first difference. Therefore, it is concluded that TOU is stationary. To investigate for a cointegrating relationship between these variables, the bounds test within the autoregressive distributed lag (ARDL) framework and based on the F-statistic proposed by Pesaran et al. (2001) is used. The bounds test instead of other popular approaches, such as Engle and Granger (1987), Johansen (1988), and Johansen and Juselius (1990) approaches, is used because the bounds test can be applied irrespective of whether the variables are purely I(1), purely I(0), or mutually cointegrated and has better small sample properties. Pesaran and Shin (1999) have shown that the ARDL-based estimators of the long-run coefficients are super-consistent. For this study, it is believed that the bounds test is more suitable because the sample has only 28 observations and the variables are either I(1) or I(0). The bounds test investigates whether a cointegrating relationship exists in the following unrestricted error correction models: q X i¼1 q X i¼0 q X i¼0 q X i¼0

DGDPt ¼ a0 þ

aGi DGDPt2i þ

aTi DTOUt2i þ

aIi DIMt2i

þ

aEi DEXt2i þ a1 GDPt21 þ a2 TOUt21 þ a3 IMt21 þ a4 EXt21 þ 11t ð1Þ

DTOUt ¼ b0 þ

q X i¼1

bTi DTOUt2i þ

q X i¼0

bGi DGDPt2i þ

q X i¼0

bIi DIMt2i

þ

q X i¼0

bEi DEXt2i þ b1 GDPt21 þ b2 TOUt21 þ b3 IMt21 þ b4 EXt21 þ 12t ð2Þ

DIMt ¼ c0 þ

q X i¼1

cIi DIMt2i þ

q X i¼0

cGi DGDPt2i þ

q X i¼0

cTi DTOUt2i þ

q X i¼0

cEi DEXt2i ð3Þ

þ c1 GDPt21 þ c2 TOUt21 þ c3 IMt21 þ c4 EXt21 þ 13t

Anatolia – An International Journal of Tourism and Hospitality Research q X i¼1 q X i¼0 q X i¼0 q X i¼0

353

DEXt ¼ d0 þ

dEi DEXt2i þ

d Gi DGDPt2i þ

d Ti DTOUt2i

þ

d Ii DIMt2i þ d1 GDPt21 þ d2 TOUt21 þ d3 IMt21 þ d4 EXt21 þ 14t

ð4Þ

To minimize the loss of degrees of freedom and to meet no autocorrelation assumption required by the bounds test, the value of q in each equation is the lowest value where the Breusch– Godfrey LM test with lag order of 2 is unable to reject the null hypothesis of no autocorrelation at 5% significance level. The critical values for bounds test are obtained from Table CI(iii) in Pesaran et al. (2001). At k ¼ 3, the critical value bounds are (2.72, 3.77) at the 10% significance level, (3.23, 4.35) at the 5% significance level, and (4.29, 5.61) at the 1% significance level. The null of no cointegration is rejected if the computed F-statistic obtained from each null hypothesis exceeds the respective upper critical value. If the computed F-statistic is lower than the respective lower critical value, the null hypothesis is accepted. If the computed F-statistic falls inside the lower and upper critical values, no conclusive decision can be made unless the order of integration of the variables under consideration is known. The null hypothesis of no cointegration of each equation is stated in the second column of Table 2. The null of no cointegration is rejected at the 1% significance level for Equation (1). The null of no cointegration is accepted even at the 10% significance level for Equations (2) and (4). For Equation (3), the bounds test shows inconclusive result at the 10% significance level and accepts the null of no cointegration at the 5% level. Therefore, it is concluded that there is no cointegration for Equation (3). Since there is evidence of cointegration when GDP is used as the dependent variable, the Granger causality test applied on the case where GDP is the dependent variable will be augmented with an error correction term based on the suggestion of Engle and Granger (1987). To obtain the cointegrating relationship, the following ARDL model is estimated: ð1 2 f1 L 2 · · · 2 fp L p ÞGDPt ¼ b0 þ ð1 2 b1 L 2 · · · 2 bq L q ÞTOUt þ ð1 2 a1 L 2 · · · 2 ar L r ÞIMt þ ð1 2 u1 L 2 · · · 2 us L s ÞEXt þ nt ð5Þ

The values of p, q, r, and s in Equation (5) are selected based on SIC. However, because of a relatively small sample size and annual data used in this study, the maximum possible values of p, q, r, and s are set at 1. SIC selects p ¼ 1, q ¼ 0, r ¼ 0, and s ¼ 1. Equation (5) is reparameterized to obtain the long-run coefficients and the corresponding asymptotic
Table 2. The results of the bounds test for cointegration. Equation (1) (2) (3) (4) H0 a1 ¼ a2 ¼ a3 ¼ a4 ¼ 0 b1 ¼ b2 ¼ b3 ¼ b4 ¼ 0 c1 ¼ c2 ¼ c3 ¼ c4 ¼ 0 d1 ¼ d2 ¼ d3 ¼ d4 ¼ 0 F-Value 31.1500*** 0.4735 3.0144 1.0514 q 2 1 2 2

*,** and *** Statistically significant at the 10%, 5%, and 1% levels, respectively.

354

C.G. Lee

standard errors as reported in the following: ^ GDPt ¼ 227809:2 þ 0:0036TOUt þ 1:7086IMt 2 0:8849EXt þ ht ð9891:6Þ ð0:0060Þ ð0:7328Þ ð0:5076Þ ð6Þ

^ where ht is the error correction term and the asymptotic standard errors are reported in parentheses. IM is statistically significant at the 5% significance level. The estimated coefficient of IM is positive implying that IM has positive effects on GDP in the long run. TOU is statistically insignificant even at the 10% level. The statistically insignificant coefficient of TOU is positive. EX is marginally significant at the 10% level because the p-value of the null hypothesis of the coefficient of EX equal to zero is only 9.6%. But, the sign of the estimated coefficient of EX is negative indicating that EX has negative impacts on GDP. Instead of providing support for export-led growth, the result suggests that an increase in export activities lead to a fall in the income of Singapore. Since EX is only marginally significant and its coefficient has a wrong sign, it is suggested that EX has no effect on GDP in the long run. To implement the Granger causality test, the following models are specified: DGDPt ¼ e0 þ q X i¼1 q X i¼1 q X i¼1 q X i¼1 q X i¼1 q X i¼1 q X i¼1 q X i¼1 q X i¼1

eGi DGDPt2i þ

q X i¼1

eTi DTOUt2i þ

q X i¼1

eIi DIMt2i

þ

eEi DEXt21 þ e1 ECTt21 þ m1t

ð7Þ

DTOUt ¼ f 0 þ

f Ti DTOUt2i þ

f Gi DGDPt2i þ

f Ii DIMt2i

þ

f Ei DEXt2i þ m2t

ð8Þ

DIMt ¼ g0 þ þ

gIi DIMt2i þ

gGi DGDPt2i þ

gTi DTOUt2i ð9Þ

q X i¼1

gEi DEXt2i þ m3t q X i¼1 q X i¼1

DEXt ¼ h0 þ

q X i¼1

hEi DEXt2i þ

hGi DGDPt2i þ

hTi DTOUt2i

þ

q X i¼1

hEi DIMt2i þ m4t

ð10Þ

^ where ECTt21 is the error correction term. ht21 of Equation (6) is used to replace ECTt21 of Equation (7) because ECTt21 is not observable. The value of q in each equation is selected with the same approach as that applied to Equations (1) – (4). The results of shortrun and long-run Granger causality are presented in Table 3. The long-run Granger causality is implemented only on Equation (7). To investigate for the presence of long-run Granger causality, the t-test is used to check whether the coefficient of the error correction

355

2 0.4419*** (0.1298)

# *,** and *** Statistically significant at the 10%, 5%, and 1% levels, respectively. Estimated value of e1 is reported in this column. The corresponding standard error is given in parentheses.

term is different from zero. Further, the sign of the coefficient of the error correction term must be negative. The rejection of this null suggests that long-run Granger causality is present. The short-run Granger causality can be investigated with the f-test. If the lagged first differences of an explanatory variable are jointly insignificant, there will be no shortrun Granger causality from that explanatory variable to the dependent variable. In Equation (7), only the lagged first differences of EX are jointly significant. These results suggest that there is only short-run Granger causality running from EX to GDP. The coefficient of error correction term is statistically significant with the correct sign. Granger (1988) points out that there will be causality among the variables at least in one direction, if there is a cointegrating relationship among them. However, based on the results of Equation (6), it is concluded that there is a long-run Granger causality running from IM to GDP because only IM is statistically significant. In Equation (8), the lagged first differences of GDP are jointly significant at the 1% level. This result suggests that there is a short-run Granger causality running from GDP to TOU. The lagged first differences of EX are also jointly statistically significant at the 10% level. This result suggests EX Granger causes TOU in the short-run. In Equation (9), the lagged first differences of GDP and EX are jointly significant at the 1% level. This implies that there is a short-run Granger causality running from GDP to IM and from EX to IM. The lagged differences of TOU are

term is different from zero. Further, the sign of the coefficient of the error correction term must be negative. The rejection of this null suggests that long-run Granger causality is present. The short-run Granger causality can be investigated with the f-test. If the lagged first differences of an explanatory variable are jointly insignificant, there will be no shortrun Granger causality from that explanatory variable to the dependent variable. In Equation (7), only the lagged first differences of EX are jointly significant. These results suggest that there is only short-run Granger causality running from EX to GDP. The coefficient of error correction term is statistically significant with the correct sign. Granger (1988) points out that there will be causality among the variables at least in one direction, if there is a cointegrating relationship among them. However, based on the results of Equation (6), it is concluded that there is a long-run Granger causality running from IM to GDP because only IM is statistically significant. In Equation (8), the lagged first differences of GDP are jointly significant at the 1% level. This result suggests that there is a short-run Granger causality running from GDP to TOU. The lagged first differences of EX are also jointly statistically significant at the 10% level. This result suggests EX Granger causes TOU in the short-run. In Equation (9), the lagged first differences of GDP and EX are jointly significant at the 1% level. This implies that there is a short-run Granger causality running from GDP to IM and from EX to IM. The lagged differences of TOU are

term is different from zero. Further, the sign of the coefficient of the error correction term must be negative. The rejection of this null suggests that long-run Granger causality is present. The short-run Granger causality can be investigated with the f-test. If the lagged first differences of an explanatory variable are jointly insignificant, there will be no shortrun Granger causality from that explanatory variable to the dependent variable. In Equation (7), only the lagged first differences of EX are jointly significant. These results suggest that there is only short-run Granger causality running from EX to GDP. The coefficient of error correction term is statistically significant with the correct sign. Granger (1988) points out that there will be causality among the variables at least in one direction, if there is a cointegrating relationship among them. However, based on the results of Equation (6), it is concluded that there is a long-run Granger causality running from IM to GDP because only IM is statistically significant. In Equation (8), the lagged first differences of GDP are jointly significant at the 1% level. This result suggests that there is a short-run Granger causality running from GDP to TOU. The lagged first differences of EX are also jointly statistically significant at the 10% level. This result suggests EX Granger causes TOU in the short-run. In Equation (9), the lagged first differences of GDP and EX are jointly significant at the 1% level. This implies that there is a short-run Granger causality running from GDP to IM and from EX to IM. The lagged differences of TOU are

TOU

GDP

EX

IM

X X

Y indicates X Granger causes Y in the short-run Y indicates X Granger causes Y in the long-run

Figure 1. Summary of short-run and long-run Granger causalities.

356

C.G. Lee

also jointly significant only at the 10% level. This suggests that there is a short-run Granger causality running from TOU to IM. In Equation (10), only the lagged first differences of GDP are statistically significant. This indicates that GDP Granger causes EX in the short-run. Figure 1 summarizes the results from the f- and t-tests.

References
Abu-Qarn, A., & Abu-Bader, S. (2004). The validity of the ELG hypothesis in MENA region: Cointegration and error correction model analysis. Applied Economics, 36(15), 1685– 1695. Awokuse, T.O. (2006). Export-led growth and the Japanese economy: Evidence from VAR and directed acyclic graphs. Applied Economics, 38(5), 593– 602.

Anatolia – An International Journal of Tourism and Hospitality Research

357

358

C.G. Lee

Copyright of Anatolia: An International Journal of Tourism & Hospitality Research is the property of Anatolia and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.