Tax Collection in Developing Countries – New Evidence on Semi-Autonomous Revenue Agencies (SARAs)

Abstract

Over the last two decades, semi-autonomous revenue agencies (SARAs) have become a key element of public administration reform. They are supposed to improve revenue mobilisation and stabilise state–taxpayer relations. But do SARAs really outperform conventional tax administrations? This article argues that they do. Presenting the results of a panel analysis of local tax collection in Peru between 1998 and 2011, it shows that municipalities with SARAs collect more revenue than those with conventional tax administrations. The results also indicate that local revenue is more stable in municipalities with SARAs, which is good for budget policy and planning.

Acknowledgements

An earlier version of this article has been presented at the Primeras Jornadas Iberoamericanas de Financiación Local in Toledo, Spain (24–25 November 2011). We would like to thank the participants of the conference for their helpful comments. In addition, the article has benefitted from the detailed comments provided by two anonymous referees. Finally, we gratefully acknowledge the excellent research assistance of Sergio Gemperle and Marco Diamantini.

Notes

1. Like most innovations, this new approach builds on a broad set of older patterns and traditions. Kidd and Crandall (2006) relate the emergence of independent revenue administrations (RAs) to the UK’s experience with semi-autonomous governmental bodies. Other historical narratives might refer to the ‘functional deconcentration’ (Schmidt, 1989) of public administrations in the United States and Latin America from the 1950s onwards, or even to the historical Roman or French experience of privatised tax collection (Levi, 1988).

2. Peru has two levels of local government: 194 provinces and (at present) 1,642 districts. Provincial mayors govern the main district (the ‘capital’) of each province and perform supervisory functions relating to the district municipalities in their province. In addition, provinces have some tax competencies that districts do not have (Rühling, 2005). The present analysis therefore focuses solely on provincial municipalities, even though some district municipalities, primarily in the metropolitan area of Lima, have substantial revenue of their own.

3. In the following sections we use the acronym ‘SAT’ to refer to the specific case of Peru, and ‘SARA’ to address general aspects of the subject matter.

4. This ‘easy phase’ refers to an initial period in the life of SARAs, when high growth rates in tax collection thanks to the activation of open liabilities from previous years (generation of debt) are accompanied by robust tax collection practices.

5. Ica, Cajamarca and Chiclayo in 2003, Huancayo in 2004, Tarapoto in 1 and Huamanga in 2008. This section draws on von Haldenwang et al. (2009).

6. To give an example, the municipality of Tacna has been considered a benchmark for reform-oriented tax administrations that do not follow the SAT model. Tacna is a city in the southern coastal area of Peru that has managed to increase its locally collected revenue by 188.5 per cent between 2003 and 2011, almost tripling its income from that source. Field research conducted in March 2008 showed that the local government attributed this achievement primarily to high levels of investment in information and communication technologies, changes in human resources management and improved customer-orientation (von Haldenwang et al., 2009). However, Tacna’s revenue collection in this period was quite volatile, with growth rates oscillating between –12.4 per cent (2003) and 38.9 per cent (2010), and negative growth figures in three out of nine years. In a period of high and sustained national economic growth rates, the volatile character of Tacna’s revenue collection appears to indicate that this city has failed to fully tap its local revenue potential.

7. The FONCOMUN is financed mainly through the ‘municipal promotion tax’ (Impuesto de Promoción Municipal, IPM), a surcharge of 2 per cent on the national sales tax (Impuesto General a las Ventas, IGV). The Canon allocates a share of the corporate income tax on specific economic activities (50 per cent in the case of mining companies) to the municipalities and regions affected by them. Apart from mining, which is by far the most important sector, revenue is derived from Canon and royalty transfers in the fishery, forestry, gas, oil and hydro-energy sectors (Comisión Multisectorial, 2010; von Haldenwang, 2010).

8. The calculation is based on the mean exchange rate (sell) for the year 2005, as provided by the Peruvian Central Bank (http://www.bcrp.gob.pe/estadisticas.html).

9. Most municipalities in Peru are simply too small to be viable in terms of local self-administration, public service delivery and revenue collection. Following the recommendations of the Peruvian Association of SATs (ASAT), a municipality should be able to collect some 10 million soles (ca. 3 million US dollars) in taxes and fees in order to run a SAT efficiently (see von Haldenwang et al., 2009). Tarapoto, which collected 4.5 million soles in 2005 – two years before the instauration of its SAT – has raised local collection to 8.5 million soles in 2011. By taking half of Tarapoto’s intake shortly before the introduction of the SAT as the threshold value for our basic model, we account for the considerable revenue oscillations and collection potential at the local level in Peru and include all municipalities that could think about introducing a SAT. However, the robustness of findings was tested with smaller and larger sample sizes, as explained in Section 5. See Appendix 1 for the summary statistics and Appendix 3 for a list of the municipalities included in the sample.

10. According to Beck (2001), it does not seem reasonable to apply common panel methods to our data because these methods are known to be inefficient when the number of units is small. In contrast, the approach developed by Beck and Katz (1995) is justified by asymptotics in T rather than asymptotics in N, and provides robust estimates for T > 10. Our data cover a longer time period (T = 14). For detailed discussions on estimation methods for time-series cross-sectional (TSCS) data see Beck (2001) and Wilson and Butler (2007).

11. The approach is similar to that adopted by Stevenson and Wolfers (2006).

12. Local elections took place in Peru in 2002, 2006 and 2010.

13. Please note that the SATs deduct their commission directly upon collection, meaning that actual collection figures could be even higher than reported. In contrast, the costs of conventional tax administrations appear only on the expenditure side of the budget. For more details, see von Haldenwang (2010).

14. As economic growth data are not available at the local level, population growth is taken as a proxy, considering that this variable is correlated with the economic attractiveness of a municipality in terms of employment opportunities and, therefore, economic growth.

15. At the same time, the distribution of the Canon is highly unequal. Ten per cent of the municipalities obtain roughly 70 per cent of the funds, whereas some of the poorest regions in Peru (Amazonas, Apurimac, Ayacucho, Huánuco, Huancavelica, for instance) receive hardly any Canon transfers at all (Comisión Multisectorial, 2010). While Canon transfers oscillate considerably in some municipalities, they show hardly any variation in many others, making it difficult for the Canon to reach high levels of statistical significance in our model.

16. To take account of the possible effects of the Canon from one budget cycle to the other, we also estimated our model with a lagged variable, but general results did not vary and the results for the Canon variable remained insignificant. For a detailed discussion of the substitution effect of transfers on local tax collection in Peru, see Rühling (2005).

17. The coefficient of the ‘beyond eight years’ dummy should be read with some caution, however, as only three SATs (Lima, Trujillo and Piura) have reached that age.

18. Dummy variables for all one-year SAT periods after establishment were included in the estimation. Table 1 presents only the first eight years due to the small number of older SATs. Full results can be provided by the authors upon request.

19. The negative coefficient for the fourth year of operation may be influenced by a series of specific exogenous factors which affected the SATs more than conventional tax administrations (see von Haldenwang, 2010).

20. We reran the analysis excluding 11 observations where the annual growth of other transfers per capita was higher than 100 per cent. Without these outliers the sign of the coefficient changes and becomes statistically non-significant. All other coefficients experience no major changes, especially those of our main variables of interest. We also estimated the model excluding extreme growth values of Canon transfers, but in this case results did not change significantly.

21. Including the lagged dependent variable in the model actually produced stronger results, supporting the idea of different phases of SARA performance. In this specification, the third, fourth and fifth years of operation were not significant. Yet, given the rather low number of observations and the limitations of our dataset, we find it more appropriate to keep the basic model as parsimonious as possible.

22. For every municipality i in every year t, the variation coefficient V_i,t measures the residual from the actual locally collected revenue (y_i,t) to the individual revenue collection trend line over the period 2002 to 2010 in per cent of the predicted value for revenue collection (ŷ_i,t).

(1)

The locally collected revenue trend line is specified for every municipality individually by the trend equation:

(2)

23. As robustness check we reran the analysis using linear and cubic trend equations. Results do not change substantially with the cubic trend. As for the linear trend, the signs remain but the coefficient for the first two operation years is highly significant, while the coefficient for operation years three and beyond remains negative without reaching statistical significance. This mirrors the fact that the linear trend ignores the exponential growth of revenues and overestimates expected revenue in the middle years of the period analysed. Since assuming a linear trend is far more restrictive than the quadratic and cubic one, we do not consider this as evidence against our main finding.

24. To test for a possible interaction between SATs and local elections, we rerun the specifications presented in Table 2 including an interaction term (‘SAT: Three and more years of operation’ * ‘Election year dummy’). Results suggest that elections affect the variation of revenue less strongly in municipalities with SATs. The coefficient for the interaction term is –1.156 and the F-test for joint significance is highly significant. However, these results should be interpreted with caution as our dataset includes only 19 observations of SATs operating during election years.