The intertemporal relation between government revenue and expenditure in the United Kingdom, 1750 to 2004

Abstract

We examine the intertemporal relation between government revenue and expenditure in the UK during 1750 to 2004. We pay particular attention to long run trends by applying a battery of unit root and cointegration techniques to the data, and we use a modified Granger causality test on data spans organized around structural breaks in the series. The results suggest that, allowing for structural breaks, UK real revenue and spending are I(1) series and cointegrated and that Granger causality runs from government spending to revenue. As such, the ‘spend-tax’ hypothesis appears to best characterize the long run intertemporal relation between government revenue and spending in the UK.

Keywords:

• revenue-expenditure nexus
• unit roots
• cointegration
• structural breaks
• causality

JEL Classification::

• E62
• H61
• H62

Notes

¹ Payne (2003) provides a survey of recent studies on the issue.

² In particular, as shown by Quintos (1995), given that government revenues and expenditures are nonstationary, sustainability requires these variables to be cointegrated with a unit slope on expenditures.

³ Even though we report Granger-type causality results only for the sub-samples that are free of structural breaks, our results cover a longer time span and are based on more observations than the other empirical studies of the UK experience. For example, our shortest sub-sample is 1951 to 2004, which still has more observations that the samples analysed in the other studies with the exception of Hasan and Lincoln (1997) who use quarterly data for 1961 to 1993.

⁴ The series are from Mitchell (1988) and the Annual Abstract of Statistics published by the UK National Statistics Office (NSO), and were deflated using the consumer price index compiled by O’Donoghue et al. (2004). The data are calendar year; the fiscal data are for the central government and are total revenue and expenditure (including transfers). While the information on the details about the data used in other similar studies is not readily available, some studies have used the Government Finance Statistics or the International Financial Statistics of the IMF, or the Organization for Economic Co-operation and Development (OECD) Economic Indicators as the sources for the data on central government expenditure and revenue.

⁵ Clark and Dilnot (2002) note that between 1964 and 1970, the increase in UK government revenue was the largest significant and sustained increases in the tax burden in the twentieth century that was not associated with a war or pre-war military build-up.

⁶ In addition, Zivot–Andrews and other similar ADF-type endogenous break tests tend to select the break point incorrectly (one time period before the true break), where the bias and spurious rejections are the greatest. In contrast, the break point(s) tend to be determined correctly at the true break when using the LM test, with the accuracy increasing with the magnitude of the break(s). But even when the size of the break(s) is small and the break point cannot be accurately estimated, the LM test does not suffer a significant loss of power in this case as this is similar to having no break (Lee and Strazicich, 2003, 2004).

⁷ Lee and Strazicich (2003, 2004) do not examine Perron's (1989) ‘changing growth’ Model B as it is commonly held that most economic time series can be adequately described by model A or C.

⁸ Unit root tests with more than two breaks are often the extensions of the tests that do not allow for break(s) under the null hypothesis of unit root, while such extensions for the Lee–Strazicich type tests are, to our knowledge, not available. Hence, note that some loss of power can be expected from ignoring more than two breaks in the one- or two-break test.

⁹ This is in line with the findings of Ahmed and Rogers (1995), who use only the Phillips and Perron (1988) and Perron (1989) tests, the latter of which is based on the assumption of a known single structural break (including for the year 1939, for which, however, Ahmed and Rogers reject the unit root null for the tax revenue series).

¹⁰ We also carried unit root tests of the revenue and expenditure series over the different sub-samples but do not report them here because our main focus is the causal relation between revenue and expenditure, and because the validity of the results from the modified Granger-causality test suggested by Dolado and Lütkepohl (1996) does not depend on knowing the order of integration of the variables and the test does not assume that the cointegration structure of the system under investigation is known. The unit root test results for the sub-samples suggest a likely I(1) behaviour for both revenue and expenditure, though there is more evidence of trend stationarity in the post-World War II period when allowing for the presence of structural breaks. These results are available on request.

¹¹ The results from the Johansen (1988, 1991, 1995) trace tests with dummy variables to correct for large outliers are available on request.

¹² This is likely to reflect a transformation of the VAR system into a more suitable model for the Johansen-type tests, i.e. closer to a Gaussian model.

¹³ It is well known that in a bivariate cointegrated system there must be Granger-causality in at least one direction (Engle and Granger, 1987).