Outsourcing and public expenditure: an aggregate perspective with regional data

ABSTRACT

Although the theoretical impact of outsourcing on public spending is ambiguous ex-ante, previous cross-country comparisons point towards a positive impact on aggregate. We build a novel database that allows us to explore the association between outsourcing and public spending among Spanish regions between 2002 and 2018. This approach greatly reduces unobserved heterogeneity and limits concerns about bias in the estimates. Our baseline results point towards a positive association between outsourcing and expenditure. However, our analysis also unmasks important divergences among expenditure categories and economic conditions. The results are robust to the inclusion of efficiency indicators and several controls.

KEYWORDS:

• regional public spending
• outsourcing
• fiscal federalism
• public services

JEL:

• C53
• E62
• H6

ACKNOWLEDGEMENTS

This paper benefited from useful suggestions made by two anonymous referees and the editor, as well as Angel de la Fuente, Santiago Lago-Peñas, Guillem López-Casasnovas, Diego Martínez-López, Raúl García-Gámez, Antonio Jesús Sánchez-Fuentes, Jorge Onrubia, José Ramón Martínez Resano and the participants of the Banco de España seminar. The views expressed in this paper are the authors’ alone and do not necessarily reflect those of the Banco de España or the Eurosystem.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors

Notes

1. For a review of Spanish regional public spending, see Pérez García and Cucarella i Tormo (2016).

2. These data are published by the General Comptroller of the State Administration (IGAE).

3. See Figure A2 in Appendix A in the supplemental data online.

4. See Figure A3 in Appendix A in the supplemental data online. The standard deviation has decreased from 5.6 in 2002 to 4.6 in 2018.

5. The ‘hold-up’ problem arises when the government needs to renegotiate outsourcing contracts if unexpected events occur. While this mechanism plays a more prominent role in infrastructure investments not analysed in this paper, it may also operate in outsourcing services with multi-period contracts.

6. The movement of public employees into the lobbying industry.

7. This observation suggests that some form of economies of scale may play a role. This is so because Catalonia, the only region showing a negative relation between outsourcing and spending, is the AC with the highest outsourcing level during the whole sample period.

8. Fiscal targets are only available from 2003 onwards. For 2002, we consider a 0% deficit target. From 2003 to 2007, targets are in national GDP instead of regional GDP. We adjust the target to regional GDP following Delgado-Téllez et al.’s (2017) strategy.

9. It is calculated as the sum of the squared weight of each party's MPs over the total number of MPs.

10. Further robustness checks show that our results remain practically unaltered when controlling for (1) the alignment between the executive and legislative branches of the regional government; (2) the electoral cycle; and (3) the existence of a left- or right-wing ruling party (in the sample under consideration the political sign of the ruling regional party coincides most of the time with the majority in the regional parliament).

11. Drop-out rates are from the Labor Force Survey (EPA in Spanish). Although PISA scores are a good candidate for quality control, its results are not available at the regional level for the initial years of our sample.

12. Barómetro sanitario, published by the Health Ministry (see https://www.mscbs.gob.es/estadEstudios/estadisticas/BarometroSanitario/homėBS.htm). It measures the degree of satisfaction of the public health service (on a scale of 1–10).

13. Partially endogenous regressors are also known as predetermined or weakly exogenous regressors.

14. Another alternative is the system-GMM estimator introduced by Arellano and Bover (1995) and Blundell and Bond (1998). It also exploits first differences of the variables as instruments for the equation in levels, but requires further assuming mean stationarity of the variables, which is not realistic in our sample.

15. The methodology included in this paper is similar to that of Hernández de Cos and Moral-Benito (2013).

16. Intuitively, with $N=17$ and $T=17$, some reduced-form equations would be linear projections with 17 observations and $T-1=16$ regressors.

17. For the results for all main specifications, see in the supplemental data online.

18. Table A2 in Appendix A in the supplemental data online shows the results for all specifications under panel OLS.

19. See also Table A2, columns 1 and 2, in Appendix A in the supplemental data online.

20. Because $0.149/\left(1-0.554\right)=0.334$.

21. This is in line with the country level results shown in Alonso et al. (2017).

22. Table A3 in Appendix A in the supplemental data online shows the results in detail.

23. To rule out doubts regarding the excludability assumption of the instrument used, we have conducted further robustness exercises using as blunt instruments the share of private employment in the sectors of education and health for the whole sample. The time series used are from a regional database on public and private employment described by Fernández-Huertas et al. (2016). When using as instrument the share of private employment in the total employment of health and education sectors, estimates point towards a slightly higher positive effect of outsourcing in expenditure in comparison with our preferred baseline specification. This effect seems to be reinforced when instrumenting only with the share of private employment in the health sector, which accounts for roughly 60% of the sum of education and health sectors in terms of both employment and expenditure. The latter is not true when using as an instrument the share of private employment in the education sector, where the coefficient signals a lower, but still positive, effect of outsourcing in expenditure. However, in the last case the results are not statistically significant. Overall, these robustness checks confirm the findings in the preferred baseline specification.

24. The low Sargan p-values reported in columns (1) to (4) from Table A3 in Appendix A in the supplemental data online are not a source of concern since Sargan-type tests of overidentifying restrictions present extremely low power in the diff-GMM setting in which neither $N$ is large nor $T$ is short (Bowsher, 2002).

25. Table A4 in Appendix A in the supplemental data online reports the results for all specifications.

26. This variable has been deflated with the consumer price index (CPI). The results for all specifications are included in Table A6 in Appendix A in the supplemental data online.

27. Average spending per capita across all regions in 2018 was €3928. The long-run effect is calculated from $0.569/\left(1-0.568\right)=1.32$.

28. Results are not shown due to space considerations, but they can be provided by the authors upon request.

29. We undertake the analysis with 2007 as the base year because is the last year of economic expansion before the crisis.

30. Besides the results shown, two additional proxies for education were used: the share of youth unemployment in total unemployment, and students per professor in publicly funded schools. These controls do not change the baseline results.

31. If monitoring arrangements are not properly put in place, public authorities can partially lose control of the service. This, in turn, can be detrimental to its quality. In case this mechanism is also at work, the negative consequences of outsourcing in terms of public sector efficiency would be even larger than those of spending levels per se.