Abstract
Using Taiwan data, this empirical study delves into the causal links among four disaggregate real government expenditures, real government revenue and real output. The results substantiate that there is (i) neutrality between real government revenue and real government expenditure on economic development; (ii) unidirectional causality from real government revenue to real government expenditures on national defence, on general administration and on education, science and culture, confirming the tax-and-spend hypothesis; (iii) neutrality between output and the four disaggregate government expenditures; and (iv) unidirectional causality from real output to real government revenue. Several implications emerge from our empirical results.
Notes
Notes
1. ZA (Citation1992) warned that with small sample sizes, the distribution of the test statistic can deviate substantially from those with asymptotic distribution. In general, the critical values for small samples are smaller (more negative) than those with asymptotic distribution (see Narayan and Narayan, Citation2006). Because our test results are larger than the asymptotic critical values, we believe that our conclusions are unchanged if we use the small sample critical values.
2. Like the F-test, the LM test shows no evidence of residual serial correlation. Moreover, Ramsey's RESET test shows no misspecification with respect to its functional form. Furthermore, there is also no evidence of heteroscedasticity, of the non-normality of the residual, or of residual ARCH effect. Aside from this, the preferred specification passes the CUSUM and CUSUMSQ tests of parameter stability. The test results and graphs from the CUSUM and CUSUMSQ tests are available from the author upon request.
3. The results from Chang, Liu and Caudill (Citation2000), which are based on aggregate government expenditures and revenue data from 1951 to 1996, are also in favour of the tax-and-spend hypothesis in Taiwan.
4. As pointed out by an anonymous referee, if the error term in Equations (Equation3–5) is not normally distributed, the Toda–Yamamoto modified Wald test will not perform accurately. To the best of our knowledge, there are many discussions on the power and size performances on Toda and Yamamoto's (Citation1995) MWALD test for causality. In particular, Rambaldi and Doran (Citation1996), Zapata and Rambaldi (Citation1997), and Yamada and Toda (Citation1998) provide simulation evidence that the Toda and Yamamoto (Citation1995) procedure performs well regarding both size and power of the test. Moreover, Zapata and Rambaldi (Citation1997, p. 294) pointed out: ‘The empirical size of the WALD test approaches the nominal size when the model is estimated with the true or the overfitted lag structure. The results at samples of size 50, nonetheless, appear quite accurate’. Therefore, we believe that our empirical evidences are quite reliable irrespective of the error distribution.