Government size, fiscal policy and the level and growth of output: a review of recent evidence

Abstract

Theoretical developments, improved methodologies and more extensive data have helped generate a dramatic increase in the literature testing for the impact of government size and fiscal policy on economic growth in recent years. We review a range of the more recent evidence and examine (1) the consistency or robustness of the results, (2) how these results differ from the earlier literature and (3) their usefulness as a guide to policy reform in practice. We find that the last decade has produced more robust evidence and more plausible orders of magnitude on the impact of fiscal policy on growth. However, the value of this evidence remains limited as a basis for quantifying macro-economic responses to fiscal policy reform in practice.

Keywords:

• fiscal policy
• government expenditure
• taxation
• economic growth

Acknowledgements

The authors are grateful for helpful comments on an earlier draft of this paper from Professor Noel Gaston, an anonymous referee of this journal, and participants at the First Annual Workshop, on Economic and Policy Developments in East Asia, held at Bond University, Gold Coast, Australia (October 2011). The views expressed in this article are those of the authors and do not necessarily represent those of the New Zealand Treasury.

Notes

1. See also Myles (2000).

2. The distinction here between effects on income ‘levels’ or ‘growth rates’ is essentially a distinction between short-run, temporary effects on income growth which lead to permanent changes in income levels, versus persistent effects on income growth that thus have ‘permanent’ (or at least long-lasting) effects on income growth rates.

3. Similar patterns are found if 2007, rather than 2009, is used suggesting that the observed patterns are not specific to the post-2008 global recession period. Furthermore, averaging the data over the 1995–2009 period reveals similar patterns.

4. For the same sample of OECD countries, the T/Y ratio is heavily concentrated within a very small interval, largely falling within 15%–30%.

5. Statistically, weak negative associations are obtained in both cases, not significant at the 5% level.

6. The scatter in Figure 4 does reveal a statistically significant (at 5%) negatively signed relationship between the two variables.

7. For example, persistence within the decadal data for individual countries could yield a positive association between current government size and subsequent growth that essentially mirrors the equivalent contemporaneous association between the two variables.

8. For example, the estimated slope of a linear OLS regression on the data in Figure 5 changes from −0.078 (t-ratio = −5.68) when all countries are included, to −0.035 (t-ratio = −2.28) when just the six observations (7.6% of the sample) on the extreme left of the figure are omitted: 3 for Korea; 1 each for Japan, Greece and Spain. The regression fit (R ²) also falls from 0.30 to 0.07. This is not intended to suggest that such observations should be omitted but merely to demonstrate the sensitivity to inclusion/omission of small numbers of extreme observations.

9. This latter category would include consumption taxes in the absence of labour supply effects.

10. See, for example, the endogenous growth models of Peretto (2003, 2007). Similar ‘transitional’ (as opposed to steady-state) growth effects of fiscal policy are obtained from neoclassical models; see, for example, Turnovsky (2004).

11. Here we treat as given the institutional structures that determine which goods are delivered via market or non-market mechanisms.

12. For income taxes this is clearly built into the tax code where an income definition forms the tax base; however, an indirect relationship with income levels is also often observed for other taxes, such as expenditure taxes, where the tax base (e.g. consumption expenditure) moves in association with income.

13. For example, the familiar multi-step income tax used in many countries usually involves individual's marginal (and associated average) tax rates rising with income levels.

14. As noted above a statutory income tax involving a progressive multi-rate structure, generally leads to effective average and marginal tax rates rising as all income levels increase. In most such tax structures the relationship between effective average and marginal rates would not be strictly proportional as both would approach the top statutory marginal rate as incomes rise.

15. When d is sufficiently negative to generate b(1 + bd) < 0, then Y_t will be lower than Y_{t− 1}.

16. The mis-specification of typical cross-country OLS regressions (in which income levels or growth rates are regressed on a measure of government size) can be seen by re-arranging Equation (1) to give: Y_t =−(a/b)+(1/b)g_t or, taking first differences, ΔY_t =(1/b)Δg_t =(1/b)g_t −(1/b)g _t−1. Hence regression testing for relationships between current government size and current income levels or changes are likely to pick up these endogenous demand-driven relationships between the variables depicted in Equation (1) rather than, or as well as, the hypothesised causal relationships from government size to income levels or growth.

17. The Barro (1990) model, for example, effectively implies a value of d in Equation (3) that is positive at low income levels, becomes zero then turns negative at high income levels due to the non-linear distortionary effects of rising marginal tax rates. Slemrod (1995) and others have argued that this may, in part, explain the failure of empirical studies to identify robust linear cross-country relationships between income levels and government size.

18. Two earlier studies which broadly meet these criteria are Mofidi and Stone (1990) and Miller and Russek (1997) – the former applied to cross-State data for the US. Note that almost all third generation evidence appears after the critical reviews of Slemrod (1995), Agell, Lindh, and Ohlsson (1997) and Myles (2000).

19. See Dungey and Fry (2009), Ilzetzki, Mendoza, and Vegh (2010) and Beetsma and Giuliodori (2011) for more detailed reviews of methods and evidence.

20. Other evidence prior to 2000 includes time-series regression estimates of business cycle models for individual countries which estimate or simulate fiscal policy shocks (e.g. McGrattan, Rogerson, and Wright 1997), and simple Granger-causality-based tests that provide some, but limited, insights (e.g. Saunders 1993).

21. See, for example, Romer and Romer (2010), Ramey (2011) and Barro and Redlick (2011).

22. For example, where studies ignore the GBC and report parameter estimates for, say, an ‘income tax’ effect on output growth, this could not be used to investigate the growth effects of possible tax reform scenarios without knowing which other element(s) also changed (or were implicitly assumed to change) in the empirical exercise.

23. The table also reports similar orders of magnitude for simulations of the output effects of capital tax rate changes for the US by Turnovsky and Chatterjii (2002).

24. We prefer the term ‘persistent’ to ‘permanent’ here since the generally limited periods of analysis used in most studies do not allow ‘persistent’ but nevertheless transitional effects (in the sense used in theoretical models), to be readily distinguished from permanent effects.

25. See, for example, Glaeser, Lopez-de-Silanes, and Schleifer (2004) and Mauro (1995).