It has been generally agreed that a structural break exists in time series data. In fact, visually checking the EG (real per capita GDP) plots in Appendixes 5 to 9, India seems to have a break around 1991, whereas the other four countries have prominent breaks over the period 1997-1998. We therefore consider it important to take the element of structural break into our analysis for obtaining more plausible estimates. To this end, the structural break in economic growth dummy (SBGD) is allocated by seeking structural break(s) in each country’s EG series through the test developed by Bai and Perron (hereafter the BP test).
The BP test specifies multiple structural changes in a linear regression model estimated by least squares, treating the dates of structural breaks as unknown and endogenous events. The rationale for performing the BP test is that it allows us to determine break points statistically and objectively not setting the break dates based on a priori information.
To eliminate autocorrelation in estimation, each EG-VAR has already been included: SGD and PCD for India; SGD and SFD for Indonesia; SFD and SFCD for Korea; SGD and SFCD for Malaysia; and SGD, SFD and SFCD for Thailand (Note 11). As reported in Table 2, the sample periods differ across the five countries due to data availability. Subsequently, we check the lag order selection statistics of each EG-VAR and set three lags for Korea, Malaysia and Thailand and four lags for India and Indonesia. Exhaust Emissions
Based on the break dates reported in Table 1, different SBGDs are created. Referring to Thailand’s two-break result, for instance, we produced SBGD as illustrated in Figure 1. Thus actually allocating each of those SBGDs — as the deterministic component outside the cointegrating vector — into each country’s VECM and ARDL estimations, we have detected that for both India and Indonesia, the one break result is the best (1990Q1 for India and 1997Q4 for Indonesia), whereas for Thailand, the two-break result (1997Q2 and 2003Q1). Here, the selection mainly depends on whether the SBGD allocation provides a single cointegration (r = 1) and/or no autocorrelation in estimation. However, SBGDs are not essential for both Korea and Malaysia. In Korea’s case, instead of the BP test, we have performed the Zivot and Andrew (hereafter ZA) test and detected a single structural break in 1997Q4 (Note 13). Based on this single break result, we allocate a zero-one dummy, which is named the ZA dummy (ZAD), in Korea’s estimation. On the other hand, in Malaysia’s case, any dummy allocations, which are specified either by the BP test or by the ZA test, do not provide better estimates, so that no SBGD is contained in Malaysia’s analysis. Finally, Table 2 shows the combinations of dummy variables that are included in the five countries’ assessments.
Table 1: Bai and Perron test results
|Country||Number of Break(s)|
|Korea||1998Q3||1996Q4||1988Q3; 1996Q4||1987Q4; 1992Q4|
Table 2: Sample periods and dummy variables included
|Country||Sample period||Dummy variables|
|India||1982Q1 to 2007Q4||SGD; SBGD (one break); PCD|
|Indonesia||1982Q1 to 2007Q4||SGD; SFD; SBGD(one break)|
|Korea||1983Q1 to 2007Q4||SFD; SFCD; ZAD|
|Malaysia||1982Q1 to 2007Q4||SGD; SFCD|
|Thailand||1986Q1 to 2007Q4||SGD; SFD; SFCD; SBGD (two breaks)|
Figure 1: Thailand’s SBGD (two breaks)