The experience of some OECD economies on tax smoothing

Abstract

Observed random walk behaviour of a tax rate does not necessarily support the tax smoothing hypothesis though the latter implies the former. This article presents a direct test of tax smoothing by showing that if the tax smoothing hypothesis holds then the future tax rate should cointegrate with the current permanent government expenditure rate even though the tax rate is a random walk. This test is a direct and robust test of a number of ‘random walk models’ available in the literature. This procedure also enables us to differentiate among ‘strong tax smoothing’, ‘weak tax smoothing’ and ‘no-tax smoothing’, all of which are consistent with the random walk behaviour of a tax rate. Application of this test to Australia, Canada, Italy, Japan, the Netherlands, New Zealand, the UK and the US show evidence in support of weak forms of tax smoothing.

Keywords:

• random walk
• cointegration
• strong tax smoothing
• weak tax smoothing
• BN decomposition

JEL Classification::

• E62
• H21
• C32

Acknowledgements

The authors would like to thank two anonymous referees for their valuable comments on the article. The authors also benefitted from comments by Robert Alexander, John Creedy, Chia Ngee Choon, Aditya Goenka, Alfred Haug, Basant Kapur, and Jeffrey Sheen on earlier versions of the exercise and Peter C.B. Phillips for his insightful discussions on some econometric aspects.

Notes

¹ Kydland and Prescott (1977, 1980) and Barro (1979) have shown that the Ramsey tax rule could also be applied to derive a criterion for optimal taxation if many goods being taxed are interpreted as a single aggregate consumption good. If taxable goods are represented by the level of income (Gross National Product (GNP)), the optimal fiscal policy implies that the average tax rate should be uniform over time (see also, Lucas and Stockey, 1983). There are discussions on optimal tax policies under different conditions. Kydland and Prescott (1977), Chari and Kehoe (1999), and Golosov et al. (2003) examined optimal taxation in an economy with capital. Abel (2005) discussed optimal tax policies when consumers have an endogenous benchmark level of consumption. Fleurbaey and Maniquet (2006), Albanesi and Sleet (2006), and Hungerbühler et al. (2006) examined optimal taxation under different private and government information.

² As it would be clear from the subsequent derivations , where it is assumed . Note that dynamic efficiency of an economy requires .

³ In the case of Equation 13, the underlying dynamic model in levels may be written (without the constant and x terms) as . After obtaining the ECM transformation, and can be replaced with and respectively. The estimators of linearly transformed models share similar statistical properties (Sims et al., 1990).

⁴ We obtained the data for the central government revenue, expenditure, Gross Domestic Product (GDP), M1 money component, the unemployment rate and the Consumer Price Index (CPI) from the International Statistical Yearbook and the International Financial Statistics of the International Monetary Fund (IMF). US unemployment data are from the web site of the US Department of Labour. Australian unemployment data for the period 1957–1967 are from the web site of the Australian Bureau of Statistics. Data for New Zealand are from the website of the New Zealand Bureau of Statistics. The tax rate and the government expenditure rate are given as a percent of GDP.

⁵ The average marginal tax rate should be computed using changing weights. Even if the marginal tax rates remain constant, due to shifting proportions of households in each tax bracket the effective tax rate changes over time. The average tax rate would be a better proxy for the effective tax rate than a fixed-weighted average marginal tax rate. If we use, for example, changing income shares (w_i ) to average the marginal tax rates then the two tax rates should be roughly the same. If represents the marginal tax rate for the ith income category, then  = Total tax assessed/Total assessed income .

⁶ Beveridge and Nelson (1981) originally used this method to examine the business cycle pattern in macroeconomic variables. Later, Watson (1986) employed this method to decompose permanent and temporary components of the US GNP, disposable income and nondurable consumption expenditure. We fit an Autoregressive Integrated Moving Average(p,1,q) (ARIMA(p,1,q)) model to the data series and derive the unit root component first and then obtain the transitory component by subtracting the unit root component and the initial conditions from the original series.

⁷ We use the built in solution in the STAMP package for the KF. The transitory component is derived as the difference between the actual series and the smooth series.

⁸ To address any concerns that may not be exogenous we conducted a causality test by estimating , and testing . The zero restrictions were not rejected by the data (see Rajaguru and Abeysinghe, 2008, for methodological issues). These results are not included here for brevity.

⁹ For Japan, if we ignore a jump in the tax rate in 1980, we obtain a statistically significant estimate of . For New Zealand, estimating a bivariate ECM shows substantial endogeneity of resulting in a close co-movement with . However, seems to have less predictive power on . For the US, statistically significant indicates that either both the tax rate and the government expenditure rate are I(1) or I(0). Since evidence in favour of I(1) behaviour of the government expenditure rate is strong, the tax rate must be sharing the same I(1) behaviour.

¹⁰ See Afonso (2005) and references therein for a large body of literature on fiscal sustainability that show evidence both in favour and against fiscal sustainability.