Working Paper Series: Special Edition of 2016 to 2018 Interns

Autocorrelation or serial correlation, which is the correlation of errors between different space and time; in a statistical notation it is where ( ≠ 0 for ≠ j. The test states its null hypothesis as no autocorrelation and has not found any traces of this statistical phenomenon. Homoscedasticity or the absence of heteroscedasticity concerns itself with a variance that fails to be constant over time or ( ′ ) = 2 . Employing the White heteroscedasticity (no cross terms) test has indicated that the equation do not suffer from a varying variance over time. It was also necessary to determine if the data set was well-modelled in accordance to a normal distribution which would aid in speaking to the goodness of fit with accuracy . Testing with the null hypothesis as normality, it has been concluded by the joint Jarque-Bera statistic that the model has been structured properly. Inverse roots of the AR characteristic polynomial show covariance stationarity of the variables.

Table 6: Diagnostic Tests

Portmanteau Autocorrelation

Residual Heteroscedasticity

Residual Normality test (Jarque-Bera)

3.798196 (0.7040) 4.527728 (0.6056)

71.24070 (0.0163) 68.48190 (0.0277)

7.803110 (0.0991) 8.680758 (0.0696)

Eq. 2 w/ TO

Eq. 2 w/ TA

Note: figures in parentheses are p-values and all tests taken at 5% level of significance.

The Breusch-Pagan Chi-square cross-section dependence test which is suitable in small samples was utilized. In both versions of versions of equation 2, it was found that disturbances have cross- sectional independence, see Table 7.

116