On the relation between central and sub-national government fiscal balances¹

¹ The views in this article are those of the author and should not be attributed to the International Monetary Fund.

Abstract

Unbalanced panel least squares regression results (with fixed cross-section and period effects) suggest that changes in the fiscal balances of central and sub-national tiers of government in OECD countries tend to be highly positively correlated and that this partly reflects the existence of administrative controls and fiscal rules that govern central- and sub-national government fiscal relations. In contrast, cooperative institutional frameworks and reliance on market forces do not appear to have secured complementary fiscal adjustment at the sub-national tier of government.

Notes

¹ The views in this article are those of the author and should not be attributed to the International Monetary Fund.

² A well-known shortcoming of the GFS data is that the level of sub-national revenues and expenditures is a poor proxy for the degree of autonomy that sub-national governments have in practice. For example, surveys of fiscal relations by Joumard and Kongsrud (2003) and Darby et al. (2003), and OECD (1999) show that limits on the discretion of sub-national governments to determine tax rates and tax bases significantly reduce local fiscal autonomy.

³ The countries and respective sample periods are as follows: Australia 1970–1999; Austria 1970–1999; Belgium 1978–1998; Canada 1974–2001; Denmark 1970–2000; Finland 1972–2001; France 1972–1998; Germany 1970–1998; Iceland 1983–1988; Ireland 1970–1997; Mexico 1972–2000; Netherlands 1973–1997; Norway 1972–1999; Spain 1970–1997; Sweden 1970–1999; United Kingdom 1970–1999; and USA 1972–2001.

⁴ The source for the GDP growth and inflation data is the IMF's International Financial Statistics database.

⁵ The choice of the fixed country effects was also based on the Haussman (1979) specification test (Chi-square) which consistently favoured fixed over random effects for each equation. In fact, the results from simple OLS estimates of the pooled data were not markedly different in terms of the sign and statistical significance of the estimated coefficients.