CAPÍTULO IV 4 ANÁLISIS E INTERPRETACIÓN DE RESULTADOS
DESARROLLO DE MECANISMOS DE APRENDIZAJE
Panel cointegration tests investigate the presence of a long-run relationship between the two variables. This is a test of whether there is a long-run relationship (Ghali, 1998; Basu, et al., 2003 and Lund, 2010:30). Cointegration tests were conducted using the Pedroni (1999 and 2004) and Kao (1999) tests. The Pedroni (1999 and 2004) tests heterogeneous panels. Pedroni’s cointegration tests amount to seven in total; and they can be grouped into two: the ‘panel statistics’ or ‘within dimension’, which are equivalent to the unit- root statistic against homogenous alternatives (Breitung and Pesaran, 2008); and the ‘group mean statistics’ or ‘between dimension’, which involve the averaging of the individually estimated AR coefficients for each country, individually. All the seven Pedroni tests test the null hypothesis of no cointegration. The Pedroni (1999 and 2004) statistics are one-sided tests with a critical value of -1.64. This means that a test statistic of less than -1.64 (k < -1.64) implies rejection of the null for all other tests except the v-statistic. For the panel-v, the critical value is 1.64. Thus a test statistic greater than 1.64 (k > 1.64) suggests rejection of the null hypothesis of no cointegration. Lund (2010) also noted that when there is a conflict in the results, the panel ADF and group ADF should be used. Table 9 shows the Pedroni panel and group results for both low-income and middle-income groups.
Table 9: Pedroni Panel Cointegration Tests for FDI and GDP for SADC countries groups
Method
Low Income Group Middle Income Group
Test Statistic (p-value) Test Statistic (p-value) Within Dimension / Panel Statistic Panel v-Statistic (+) -0.709 (0.761) 1.681** (0.046) Panel rho-Statistic 1.584 (0.943) -1.769** (0.038) Panel PP-Statistic 2.020 (0.978) -2.386*** (0.009) Panel ADF-Statistic 1.442 (0.925) -2.362*** (0.009) Between Dimensions / Group Mean Statistic
Group rho-Statistic 1.523 (0.936) -0.851 (0.197) Group PP-Statistic 2.045 (0.98) -2.242** (0.013) Group ADF-Statistic 1.770 (0.962) -2.085** (0.019)
Notes: All statistics are from Pedroni’s procedure (1999) where the adjusted values can be compared to the N(0,1) distribution. The Pedroni (2004) statistics are one-sided tests with a critical value of -1.64 (k < -1.64 implies rejection of the null), except the v- statistic that has a critical value of 1.64 (k > 1.64 suggests rejection of the null). Numbers in parentheses () are the p-values; Selection of lags is based on Schwarz Information Criterion (SIC); Newey-West automatic bandwidth selection and Bartlett kernel; Probabilities are computed assuming asymptotic normality; *** Rejects the null at the 1% level; ** Rejects the null at the 5% level; * Rejects the null at the 1% level. + the first test is a right-tail test while the other tests are left-tail tests; for consistence we used un-weighted test statistics.
For the low-income country group, none of the four within-dimension panel-cointegration tests reject the null hypothesis that there is no cointegration between FDI and GDP. Furthermore, none of the three group tests could reject the null hypothesis. Overall, the seven Pedroni (1999 and 2004) tests show evidence of no cointegration between the two variables for the low-income country group. The Kao (1999) panel cointegration test was also conducted, for robustness; and it did not reject the null hypothesis of no cointegration (See Table 10) for the low-income country group. Thus, both the Pedroni and Kao cointegration tests fail to support the existence of any long-run relationship between FDI and economic growth in low-income SADC countries.
This result was not expected, given that both theory and the empirical literature support the existence of a long-run relationship between these two variables. Most seminal studies found evidence of the existence of a long-run relationship between the FDI and economic growth (Ghali, 1998; Zhang, 2001; Lund, 2010 and Fowowe, 2011). In some studies cointegration was found in certain countries (see Urbain, 1992; Zhang, 2001a and Moudatsou and Kyrkilis, 2011).
On the other hand, the four within-dimension panel cointegration tests for the middle-income country group rejects the null hypothesis that there is no cointegration between the FDI and the GDP at a minimum of 5% level of significance. Only one of the group tests could not reject the null hypothesis (See Table 9).
Overall, six out of the seven tests show evidence of cointegration. The panel ADF and group ADF show consistent results. Thus, the Pedroni tests show evidence of a co-integrated long-run relationship between the FDI and the GDP in middle-income SADC countries. Table 10 presents panel cointegration tests results based on Kao (1999).
Table 10: Kao (1999) Panel cointegration test by SADC country group
Method
Low Income Group Middle Income Group
Test Statistic (p-value)
Test Statistic (p-value)
Kao ADF Test 0.049
(0.480)
-2.138
(0.016)**
Notes: Numbers in parentheses () are the p-values; Selection of lags is based on Schwarz Information Criterion (SIC); Newey- West automatic bandwidth selection and Bartlett kernel; Probabilities are computed assuming asymptotic normality; ** Rejects the null at the 5% level.
The Kao (1999) panel cointegration test (See Table 10) rejects the null of no cointegration at 5% level for the middle income country group. However, for the low income, the Kao ADF Test also agrees with the Pedroni (1999, 2004) tests that the panels are not cointegrated. These cointegration results support evidence for the existence of a long-run relationship between FDI and economic growth in middle-income SADC countries, but not in the low-income group.
Since FDI and GDP panels in low-income SADC countries were found to be I(1), but not cointegrated, the error correction term (ECT) cannot be included in the estimation equations. Granger-causality tests were therefore conducted through the VAR panel framework as proposed by Dumitrescu and Hurlin (2012). Conversely, FDI and economic growth panels for middle-income countries were co-integrated; and therefore, causality tests were conducted, using the ECM framework (see Granger, 1986 and 1988; Urbain, 1992; Ghali, 1998; and Pesaran, et al., 1999).