It presents a problematic situation as common techniques of panel analysis are incapable of handling both cross sectional dependence and
serial correlation simultaneously.
While
serial correlation is well understood, it has been largely ignored by researchers using DD estimation.
The
serial correlation LM test along with ACF and PACF plot results reveals that
serial correlation in the residuals not exists.
All independent variables in model 4 are consistent with theories and the model have no severe problem of
serial correlation, multicollinearity, unit root and heteroscedasticity which fulfil the BLUE properties.
The result from LM test proves that there is no
serial correlation among the variables as shown in Table-4.
Heteroskedasticity and
serial correlation Heteroskedasticity
Serial correlation Model P-Value [Chi.sup.2] P-Value F value Model 1 0.63 0.23 0.57 0.31 Model 2 0.12 2.33 0.54 0.37 Model 3 0.31 1.00 0.24 1.39 Model 4 0.23 1.40 0.24 1.39 Model 5 0.22 1.47 0.30 1.06 Table 4.
Bounds and cointegration tests in rows (a) and (b) are all highly significant, and the null of no
serial correlation cannot be rejected by any p value in row (c).
Table 5 Diagnostics Tests of the Models (P-Values) UM RWD RWDT Benchmark Models
Serial Correlation [0.45] [0.18] [0.30] Heteroskedasticity [0.40] [0.93] [0.97] Normality [0.35] [0.81] [0.59] Models with Survey Expectations
Serial Correlation [0.46] [0.12] [0.1] Heteroskedasticity [0.40] [0.74] [0.72] Normality [0.36] [0.65] [0.54] AR MA ARIMA Benchmark Models
Serial Correlation [0.96] [0.1] [0.54] Heteroskedasticity [0.55] [0.53] [0.11] Normality [0.38] [0.56] [0.68] Models with Survey Expectations
Serial Correlation [0.96] [0.29] [0.27] Heteroskedasticity [0.67] [0.92] [0.10] Normality [0.38] [0.32] [0.46] The tests for
serial correlation, heteroscedasticity and normality are Breusch-Godfrey
Serial Correlation LM Test, White Test and Jarque Berra, respectively.
In conclusion, for a VEC estimation model, considering a dependent variable, LOG_ASSETS and an independent variable, RETURN, residuals are normally distributed, the model does not have an ARCH effect and there is no
serial correlation, which is desirable in all 3 cases, so we have a proper regression model.
We test for
serial correlation using the method discussed in Wooldridge (2002).
The Arellano-Bond test for first-order
serial correlation in the disturbances--in differences--rejects the null that there is no first-order
serial correlation, but it does not reject the null of no second-order
serial correlation, as shown in Tables 1 and 2.
Tables 7 and 8 report the output of regressions that use the Cochrane-Orcutt technique to correct for
serial correlation.