Summary and Conclusions
This study uses the modern time series techniques to study the relationship between government expenditure and economic growth in Malaysia. First, we study the unit root properties of the variables. We find that logarithms of government expenditure and GDP in their various formulations (like total, per capita) are non-stationary in their levels. We also find that all variables except the log of government expenditure as a percentage of GDP are stationary in their first difference form. Next, we conduct Johansen cointegration tests for pairs of variables which do not involve log of the government expenditure as a percentage of GDP. The results indicate all three pairs of variables are cointegrated and the number of cointegrating vector is equal to one in each case. The cointegrating vectors show that all three pairs of variables have a long run positive relationship as expected. Finally, we conduct the augmented Granger causality tests between pairs of variables. The results show that no matter what lag is used, causality does not flow in any direction in any of the cases. Thus, the results indicate that the growth of GDP does not cause growth of government expenditure. This result is possible if non-economic factors are more important in explaining the growth of government expenditure than economic factors. Some writers like Chee (1990) opine that this is the cause in Malaysia. The results for causality tests also indicate that the growth of government expenditure does not cause the growth of GDP. Again, in the context of Malaysia, there seems to be some evidence that government expenditure did not lead to the growth of GDP as we have discussed earlier. The policy implication is that the present structure of government expenditure is not very conducive to economic growth. However, it is quite possible that a different structure of government expenditure can contribute more effectively to economic growth. Thus, while the presence of a long run relationship between GDP and government expenditure (in their various forms) support the Wagner’s Law, causality tests tell another story. However, it should be born in mind that while cointegration tests have been performed on the levels of the variables, the causality tests are performed on the first differences (which give us growth rates of the variables because all variables are expressed in their logarithmic forms). In other words, causality tests indicate the absence of short run relationship whereas the presence of cointegration indicates long run relationship.