Fiscal stability during the Great Recession: putting decentralization design to the test

ABSTRACT

This paper provides an empirical analysis aimed at disentangling the roles played by decentralization level and design as well as the extended decentralized framework provided by subnational borrowing rules and fiscal responsibility laws on a country’s fiscal stability. Using Organisation for Economic Co-operation and Development (OECD) countries’ data from the period 1995–2014, strong regularities are found regarding the effects of decentralization, even during the recent Great Recession. Higher levels of fiscal decentralization have a beneficiary effect on fiscal performance, but the positive impact erodes rapidly with the level of vertical fiscal imbalance. Other fiscal institutions shaping decentralization design, such as borrowing and other fiscal rules, can also contribute to foster fiscal stability.

KEYWORDS:

• subnational governments
• political decentralization
• fiscal stability
• public deficit

JEL:

• H72
• H77

ACKNOWLEDGEMENTS

The authors are grateful to Pablo Beramendi, Miriam Hortas, Andreas Kyriacou, Marta Morano, Oriol Roca-Sagalés and two anonymous reviewers for very helpful comments, and also to Fernanda Martinez and Alejandro Domínguez for superb research assistance.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Notes

1. A possible third fiscal institution that should be mentioned is the recent introduction of independent fiscal councils, with a strong foothold in EU countries. Their main aim is to monitor the fiscal behaviour of all levels of government, with much more attention given to the central government over subnational governments (Wyplosz, 2005; Debrun, Hauner, & Kumar, 2009; Calmfors & Wren-Lewis, 2011).

2. Our empirical analysis is based on OECD countries over the period 1995–2014. The reason why this section is based only on European countries is that the selected countries offer the largest contrasts on how they were affected by the crisis and on how they responded to it.

3. The analysis covers the following 28 countries: Austria, Belgium, Canada, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Latvia, Luxembourg, the Netherlands, Norway, Poland, Portugal, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, UK and the USA. The remaining members of the OECD are set aside owing to lack of data on the OECD fiscal decentralization database at: http://www.oecd.org/ctp/federalism/table4_gov_exp-gdp.xls.

4. Also, the lagged endogenous variable already captures the effect of the past GDP growth rates on the general government primary balance at time t.

5. This index contains information on, for example, the legal base of the rule, the room for revising objectives, the mechanisms for monitoring compliance and enforcement of the rules, and the media visibility of the rule. Ultimately, these scores are aggregated into the composite index following the methodology proposed by Deroose, Moulin, and Wierts (2006).

6. A scheme of different weights is used when more rules apply to the same general government subsector. This weighting captures decreasing marginal benefits of multiple rules applied to the same subsector of general government.

7. In particular, variables from the RAI database are available up to 2010, political variables until 2012, and BBRSUBNATIONAL and FRINDEX only for EU countries.

8. The test is implemented in two steps. First, GDPV_ANNPCT is regressed on its own first lag and lagged values of the remaining variables in the model. Second, the resulting residuals are included in the original equation. The null hypothesis is that they are not statistically significant and then endogeneity is not a serious concern. Computed p-values are 0.51 in column (2) and 0.63 in column (5) of Table 3. Moreover, a Granger causality test clearly supports causation from GDPV_ANNPCT to NLGXQ, but not the opposite one. Using two lags, the p-value of the test on the null hypothesis that NLGXQ does not Granger cause GDPV_ANNPCT is 0.90.

9. More recently, Allison, Williams, and Moral-Benito (2017) and Moral-Benito, Allison, and Williams (2017) show the poor finite properties of panel generalized method of moments (GMM) estimators (in particular, the Arellano–Bond estimator) when N is small, as in the present case. Hence, they propose a new maximum likelihood estimator, but recognize that this estimator tends to work best when panels are strongly balanced, T is relatively small (e.g., < 10), and there are no missing data. In fact, we re-estimated our specification, but both computation and convergence problems arose. Hence, we choose to discard it and rely upon the POLS estimator.

10. Including VFI also in levels in the regressions led to multicollinearity with the interaction, but at any rate our interest in VFI lies on the moderating role it can play on the effect of EXPENDEC on a country’s fiscal performance.

11. As before, we do not include variable VFI in levels to avoid multicollinearity: regressing BORROWAUTO*VFI on both BORROWAUTO and VFI at the same time yielded R²= 0.91.

12. The number of observations increases from 462 (1) to 518 (2), mostly because the time span extends to 2014. The lack of statistical significance of ELECTIONS holds when it was coded 1 in pre-election years, and 0 otherwise.

13. For instance, the coefficient for GDPV_ANNPCT would be β for non-recession years and $\text{β}+{\text{β}}^{\mathrm{\prime }}$, where ${\text{β}}^{\mathrm{\prime }}$ would be the coefficient on the interaction between RECESSION and GDPV_ANNPCT. We do not include the dummy RECESSION alone in the regression because of its perfect multicollinearity with period fixed effects.

14. A higher coefficient on VFI_IMF is compensated by lower means of the variable, as reported in Table 2.

Additional information

Funding

This research was financially supported by the Spanish Ministry of Science and Innovation [grant number CSO2017-85024-C2-2-P].