*4.1. The Test Results, Stationarity and Cointegration*

Before estimating the parameters, stationarity and cointegration tests were performed to show that the nonlinear panel approach ARDL is appropriate for the data. The unit root test is a popular method for stationary tests for both annual time series and panel data. The stationarity test is conducted in both "individual intercept" and "individual intercept and trend" in test equations. There are many types of unit root test for panel data such as Levin, Lin and Chu t (LLC) and Breitung t-stat with common unit root process; I'm, Pesaran and Shin W-stat (IPS), ADF—Fisher Chi-square (ADF), and PP—Fisher Chi-square (PP) with individual unit root process. The panel data in this study are balanced so that both hypotheses can be applied. The LLC test is chosen for the hypothesis "common unit root process" and the hypothesis "individual unit root process" is chosen for the IPS test. The results of panel unit root tests for logarithms of variables are summarized in Table 3.

**Table 3.** Results of stationarity test.


Source: Author's calculation using Eviews. Note: LLC, Levin, Lin & Chu; IPS, I'm, Pesaran and Shin W-stat; ADF, ADF—Fisher Chi-square; PP, PP—Fisher Chi-square; \*\* and \*\*\* for statistically significant at the 0.05 and 0.01 levels, respectively.

According to Table 3, most of the series are non-stationary at level, but stationary at first difference, except for lnVA in LLC test of intercept and trend; lnTC in LLC test of intercept; and lnRF in LLC, IPS and ADF of intercept. Based on the majority of the results, it can be seen that the series are non-stationary at level but stationary at first difference, so a cointegration test should be performed to consider the long-term relationship between variables.

To analyze the cointegration relationship between variables in the panel data model, this study chooses the Pedroni and Kao tests because they are more comprehensive and universal. Cointegration tests are conducted for both "individual intercepts" and "individual intercept and individual trends" in the Pedroni test. By contrast, it is only conducted in the case of individual intercepts in the Kao test. The Pedroni test used seven test statistics (four tests for within-dimension and three tests for between-dimension). The Schwarz Information Criterion (SIC) automatically chooses the lag length with Newey-West automatic bandwidth selection and Bartlett kernel. Table 4 below presents the results of panel cointegration analysis.

**Table 4.** Results of panel cointegration test.


Note: \*\*\* for statistically significant at the 0.01 levels, respectively; deterministic trend specification: Individual intercept for Pedroni test and Kao test; Four tests for within-dimension of Pedroni test are weighted statistics. Source: Author's calculation using Eviews.

According to the results of the Pedroni test in Table 4, 4/7 tests are significant at the 0.01 level for both "individual intercept" and "individual trend and individual intercept". This means that most cointegration tests in the Pedroni test result in the cointegration series. However, the Kao test gives the opposite result, meaning that the Kao test result does not give cointegration series at the level of 0.05, so is not compelling evidence to conclude clearly that series shows cointegration. Because of lnVA, lnTC, lnHR, and lnRF containing both I(0) and I(1), and when the existence of long-run associations is unclear, the ARDL technique is the most appropriate.
