Manajemen | Fakultas Ekonomi Universitas Maritim Raja Ali Haji 00074910600873666

Bulletin of Indonesian Economic Studies

ISSN: 0007-4918 (Print) 1472-7234 (Online) Journal homepage: http://www.tandfonline.com/loi/cbie20

Local government taxation: An analysis of
administrative cost inefficiency
Blane D. Lewis
To cite this article: Blane D. Lewis (2006) Local government taxation: An analysis of
administrative cost inefficiency, Bulletin of Indonesian Economic Studies, 42:2, 213-233, DOI:
10.1080/00074910600873666
To link to this article: http://dx.doi.org/10.1080/00074910600873666

Published online: 20 Aug 2006.

Submit your article to this journal

Article views: 182

View related articles

Citing articles: 5 View citing articles


Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=cbie20
Download by: [Universitas Maritim Raja Ali Haji]

Date: 18 January 2016, At: 21:42

Bulletin of Indonesian Economic Studies, Vol. 42, No. 2, 2006: 213–33

LOCAL GOVERNMENT TAXATION:
AN ANALYSIS OF ADMINISTRATIVE COST INEFFICIENCY

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

Blane D. Lewis*
The World Bank, Jakarta
Local governments in Indonesia administer taxes inefficiently. The average cost of
local tax administration as a percentage of revenue generated is estimated to be
over 50%. There is, however, a wide variation in administrative inefficiency across
local governments. The estimation of a stochastic cost frontier model suggests that

administrative cost inefficiency increases significantly as fiscal transfers from the
centre rise; the investigation also demonstrates that local governments with elected
executives are no more administratively cost efficient than those with appointed
heads. The simple and complex measures of cost inefficiency yield broadly similar results concerning the level and variation of inefficiency across local governments, but can offer significantly different estimates of the relative inefficiency of
individual local governments. This poses a dilemma for the central government
in monitoring and evaluating local government tax administration performance.

INTRODUCTION
Tax administration is concerned with identifying taxpayers, assessing tax liability, collecting taxes and enforcing tax payment. Its main objective is to produce
the desired amount of public revenue at minimum cost, all other things remaining the same (Mikesell 1982). Applied analysis of the tax administrative performance of governments typically focuses on realisations and/or costs (Berne and
Schramm 1986). The former concentrates on the extent to which generated revenues approach potential or desired levels, and the latter on the degree to which
costs have been minimised—i.e. the cost efficiency of tax administration.
This paper examines the cost efficiency of tax administration by kabupaten and
kota (districts and municipalities, or local governments). The investigation considers two approaches to estimating administrative cost efficiency. The first is based
on a straightforward indicator—the cost-to-yield ratio (i.e. cost of tax administration divided by tax revenue). The measure is common in applied work, owing to
its simplicity and the general availability of relevant data (McMaster 1991). The
*

The author is Senior Adviser for Fiscal Decentralization at the World Bank in Jakarta under financing from the Dutch Trust Fund (TF050378). The views expressed here are those
of the author and should not be attributed to the World Bank or the government of the

Netherlands. The author would like to thank Indonesia’s Ministry of Finance for access to
data, Guenther Schultz for useful comments on an earlier draft of the paper, and the editor
and two anonymous referees for constructive criticism.

ISSN 0007-4918 print/ISSN 1472-7234 online/06/020213-21
DOI: 10.1080/00074910600873666

© 2006 Indonesia Project ANU

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

214

Blane D. Lewis

second approach derives from the specification and estimation of a stochastic cost
frontier regression model. This method is theoretically more rigorous than the
simple approach, and allows for the simultaneous testing of some hypotheses
about the determinants of administrative inefficiency.
Stochastic frontier models have traditionally focused on exploring questions

related to efficiency of private production (Kumbhaker and Lovell 2000). A limited number of studies have, however, used stochastic frontier methods to analyse public sector phenomena. De Borger et al. (1994), for example, appraise the
general technical productivity of municipalities in Belgium via stochastic frontier
methods. And Sakata (2004) employs a stochastic cost frontier approach to study
the efficiency of local government fire protection services in Japan. The present
examination appears to be the first to apply stochastic frontier techniques to questions related to public tax administrative efficiency.
The particular objectives of this paper are to establish some initial ideas about
local government tax administrative efficiency and to develop preliminary notions
about the viability of different measures of cost inefficiency. The paper proceeds
as follows. First, some background on local taxation in Indonesia is presented
with a view to offering a rationale for the analysis that follows. Second, sources
and yields of local revenue are reviewed and the role of local governments in revenue administration is described. Third, some initial empirical results on the cost
of administering local taxation, using the cost-to-yield measure of inefficiency, are
presented. Fourth, the stochastic cost frontier model is developed and the hypotheses to be tested are fully detailed. Fifth, the econometric model is estimated and
the derived results are discussed. Sixth, the simple (i.e. cost-to-yield ratio) and
complex (i.e. stochastic cost frontier) measures of cost inefficiency are compared
and contrasted. The final section of the paper summarises the main conclusions
of the study and draws attention to some problems in the analysis of local tax
administrative efficiency in Indonesia.

BACKGROUND

Local government own-source revenue is quantitatively unimportant in local
budgets. In 2003, own-source taxes and charges made up less than 8% of total
local revenue. As such, the interest in examining local tax administrative efficiency
may perhaps seem odd at first. However, the significance of local own-source revenue might well change in the future, if the central government decides to award
additional tax authority to local governments, as has been advocated by many
analysts (Lewis 2003a). Apart from possible changes in control over local sources
of revenue, the interest in local government tax administration is motivated by at
least four facts.
First, as is well known, local governments were given the authority to create
their own taxes and charges under the government’s decentralisation program,
whose implementation began in 2001.1 Since that time, local governments have
apparently established new revenue instruments in a rather aggressive manner.
1 See Undang-Undang 34 Tahun 2000 tentang Perubahan atas Undang-Undang 18 Tahun
1997 tentang Pajak dan Retribusi Daerah [Law 34/2000 concerning the Revision to Law
18/1997 on Regional Taxes and Charges].

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

Local government taxation: an analysis of administrative cost inefficiency


215

An early study estimated that just under 1,000 new taxes and charges were created by local governments in the run-up to and during the first year of decentralisation (Lewis 2003b). More recent work indicates that local governments may
have established as many as 6,000 new taxes and charges during the period 2000
through mid-2005 (LPEM–FEUI 2005).2
Second, local governments seem to have created many of their official new
revenue instruments in an extra-legal manner. Lewis (2003b) estimated that only
about 40% of the newly established taxes and charges in 2000–01 were submitted to the central government for review and evaluation, as required by law. The
remainder were presumably implemented directly via local by-laws, without central government appraisal and therefore contrary to applicable legislation.
Third, the conventional wisdom in Indonesia is that many of the newly established local revenue instruments have proven to be economically harmful (Barnes
et al. 2005). The Survey of Regional Investment Attractiveness carried out by
Regional Autonomy Watch (Komite Pemantauan Pelaksanaan Otonomi Daerah,
KPPOD) in 2003 (Regional Autonomy Watch 2003) suggested, for example, that
private businesses find that local tax regimes, in general, constrain investment
and other entrepreneurial activities. An often mentioned specific problem relates
to restrictions on inter-regional domestic trade evidently imposed by many local
government taxes and charges (Ray 2001).
Fourth, corruption related to local taxation has become more than a little apparent under decentralisation. One concern has been the increased formation of
strictly unofficial and illegal local levies, the payment of which does not produce
any revenue at all for local government budgets (Von Luebke 2005). Another issue

relates to the bribery of local officials that occurs in connection with legitimate
taxation (which does generate some local government revenue). Kuncoro (2004)
argues that the proliferation of official local taxes and charges under decentralisation has been accompanied by an increase in the bribing of local officials by businesses. He shows that such bribery significantly increases business compliance
costs, while at the same time reducing tax revenue for local governments.
The assertive and occasionally unsanctioned (by central government) creation
of official new local revenue instruments, many of which are believed to have
had a negative impact on local economies, along with the establishment of illegal
levies and other corrupt practices related at least to some local taxation, naturally
stimulates interest in examining local government tax administration in general.
As a result of the above local tax phenomena, and of other perceived weak outcomes related to decentralisation, interest in evaluating the performance of local
governments has intensified; for example, the president has recently charged the
National Development Planning Agency (Bappenas) with coordinating the development of a system to monitor and evaluate local government performance across

2 A caveat is in order with regard to the figure of 6,000 ‘new taxes and charges’. This actually represents the number of new tax and charge by-laws (peraturan daerah, or perda) issued
during the period 2000 through mid-2005. Some of these new perda undoubtedly were
written in order to change the tariffs and/or bases of existing taxes and charges, as allowed
by Law 34/2000. That is, the new perda did not in all cases authorise new tax and charge
instruments. As such, the quoted figure might best be thought of as an upper bound on the
number of newly established revenue sources.


216

Blane D. Lewis

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

a wide range of indicators. One aspect of performance that the centre has expressed
interest in assessing is the efficiency of own-source revenue administration. An early
proposal of significance in the current context was to use the cost-to-yield ratio to
monitor local government administrative cost efficiency over time, and to evaluate
the relative administrative cost performance of individual local governments.

LOCAL REVENUE AND REVENUE ADMINISTRATION
Local revenue
Local government own-source revenue in Indonesia comprises local taxes, user
charges and fees, and various other types of revenue. The main local taxes include
those on electricity, hotels, restaurants, mining, advertising, entertainment and parking.3 The central government defines tax bases and determines maximum rates; local
governments may set tax rates at any level up to these maxima. The most significant
charges are those for health services provided by local public clinics (puskesmas), for
the issue of building permits, and for public market licences. The central government provides guidelines for establishing user charges and fees; local governments

have authority to determine charges and fees based on the guidelines. Other ownsource revenue includes that generated by local government enterprises (especially
water authorities) and interest income on unspent balances.
Table 1 shows local government own-source revenue by category for 2003.4
Each of the three main types of local revenue—taxes, charges and ‘other’—contributes roughly the same amount to total own-source revenue. The table also
shows the relative importance of individual items in each own-source revenue
category. Own-source revenue makes up less than 8% of total local government
revenue; revenue budgets are dominated by intergovernmental transfers, which
account for slightly more than 92% of total local fiscal resources.
The table also shows the relative importance of various categories of transfers.
Revenue sharing (property tax, income tax and natural resource revenues), the
general allocation fund (Dana Alokasi Umum, DAU), the special allocation fund
(Dana Alokasi Khusus, DAK), and other transfers (emergency funds from the centre and grants from provinces) account for about 25%, 65%, 2% and 8% of total
intergovernmental transfers, respectively.5 A transfer of special significance in the
current context is shared property tax revenue. Total property taxes were about
Rp 10.0 trillion in 2003, about Rp 7.5 trillion of which were transferred to local
governments. While property tax transfers are small relative to total transfers,
such revenue is important in the context of local government tax administration,
as will be seen below.

3 These are the so-called ‘positive list’ taxes, that is, those that are explicitly enumerated

in Law 34/2000 as allowable.
4 All data used in this paper are from the Ministry of Finance Regional Financial Information System. The Ministry of Finance has generously provided the author with access to
these data.
5 See Lewis (2002) for detailed descriptions of the various kinds of intergovernmental
transfers, including allocation methods, and Lewis (2005) for an examination of the impact
of transfers on local government spending, own-source revenue generation, and savings.

Local government taxation: an analysis of administrative cost inefficiency

217

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

TABLE 1 Local Government Revenue, 2003
Rp billion

%

8,707.1
3,188.7


7.6
2.8

1,676.9
807.2
206.1
137.2
83.7
13.5
264.1

52.6
25.3
6.5
4.3
2.6
0.4
8.3
100.0

Total user charges
of which:
Health
Building permits
Market fees
Other

2,823.2

2.4

935.4
289.5
213.3
1,385.0

33.1
10.3
7.6
49.1
100.0

Other own-source revenue

2,695.1

2.3

Transfers
of which:
Shared revenues
General allocation fund
Special allocation fund
Other

106,528.5

92.4

26,804.8
69,280.5
2,269.0
8,174.2

25.2
65.0
2.1
7.7
100.0

Total revenue

115,235.6

100.0

Own-source revenue
Total local taxes
of which:
Electricity
Hotels & restaurants
Mining
Advertisements
Entertainment
Parking
Other

Source: Ministry of Finance Regional Financial Information System.

Revenue administration6
The local government agency charged with revenue administration is the Dinas
Pendapatan Daerah (the Regional Revenue Office, DiPenda). The DiPenda typically comprises departments for general affairs (agency administration); planning
(own-source revenue forecasting); data collection (taxpayer identification, tax liability assessment and enforcement of tax payment); and billing and recording (tax
billing, tax collections and tax revenue recording). The head of each department
reports to the director of the agency who, in turn, answers to the executive. The

6 This section of the paper draws heavily on Oosterman (2004).

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

218

Blane D. Lewis

structure and function of the DiPenda (and of other local government offices)
is regulated by the central government, and there appears to be little variation
across local governments.7
The DiPenda administers all local taxes except the electricity tax, which is managed by the National Electricity Company (Perusahaan Listrik Nasional, PLN).
It performs a minor role (mainly revenue planning and recording) in administering local user charges, which for the most part are overseen by local government
technical implementing units (Unit Pelaksana Teknis Dinas, UPTD). The DiPenda
provides no administrative support for ‘other’ types of local government ownsource revenue as defined above. However, it does play a small part (assistance
with property identification, liability assessment and billing) in administering the
property tax, which is otherwise managed by deconcentrated offices of the Ministry of Finance’s Directorate General of Taxation.8
Local government tax administration is, in general, very labour intensive. The
number of DiPenda employees varies quite considerably across local governments.
While small rural local governments may employ only around 50 civil servants,
large cities such as Medan and Surabaya may engage several hundred full-time
staff. Few DiPenda have made use of standard information technologies in the
administration of taxes. A computerised tax administration system that was established by the Ministry of Home Affairs in some of the larger local governments in
the early 1990s no longer appears to function anywhere (Oosterman 2004).

PRELIMINARY INDICATION OF
ADMINISTRATIVE COST INEFFICIENCY
As noted at the outset of the paper, a frequently used measure of local government administrative cost inefficiency is the cost of tax administration divided by
revenue generated—the cost-to-yield ratio. All other things being equal, as the
cost-to-yield ratio rises, local governments are seen as increasingly cost inefficient as regards revenue administration. The measure is standard in applied work
owing to its straightforward character and the usual availability of relevant data
with which to estimate it.
The most recent available data on DiPenda costs and on local government revenue associated with DiPenda operations are for 2003. Cost data include those related
to personnel, materials, operations and maintenance, travel, and other (unspecified)
costs.9 As noted above, not all local government own-source revenue is fully administered by the DiPenda, and at the same time it has some administrative responsibility for the central property tax. Therefore some assumptions need to be made with
regard to revenue associated with DiPenda operations. In this paper, such revenue is
taken to be all local tax revenue except for that from the electricity tax, plus 25% of
local user charge revenue, plus 25% of property tax collections. The latter two percentages are based on recent case study research (Oosterman 2004), the goal of which

7 The DiPenda does not set local tax policy; its functions are purely administrative.
8 Deconcentrated offices are divisions of central government departments; that is, they
are not part of decentralised regional administrations.
9 The data are available for 224 of the 348 local governments that existed in 2003.

Local government taxation: an analysis of administrative cost inefficiency

219

FIGURE 1 Histogram of Cost-to-Yield Ratios of Local Governments, 2003
No. of local
governments
70
60

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

50
40
30
20
10
0
14

30

47

64

81

97

114

131 147 164

181

198

214

231 248

264

Cost-to-yield ratio (%)
Source: As for table 1.

was to determine local responsibilities over various taxes and associated costs, and to
ascertain methods to improve administrative performance. Revenue as defined just
above is termed ‘local revenue’ throughout the remainder of the paper.
The data described above can be used to generate a preliminary measure of local
government tax administrative cost inefficiency. The appropriate calculation shows
that the overall cost-to-yield ratio of DiPenda operations across all local governments is 53%. The ratio ranges from a low of 14% to a high of 264%.10 About 10%
of the local governments in the sample have cost-to-yield ratios that exceed 100%,
indicating that the costs of administration are greater than the revenue collected.
The strikingly large cost-to-yield ratios are, in part, a function of assumptions about
what constitutes ‘local revenue’. But even under more generous assumptions, the
overall cost-to-yield ratio is still relatively large. Suppose, for example, local revenue
were defined as total local taxes (minus the electricity tax) plus total user charges
plus 25% of the property tax; under these assumptions, the average cost-to-yield
ratio would be about 32%. And if local revenue were taken to be total local taxes
(less electricity tax) plus total local charges plus total property tax, the average costto-yield ratio would be approximately 20%. Figure 1 presents a histogram showing
the size distribution of the cost-to-yield ratios of local governments.
10 By comparison, cost-to-yield estimates from the United States range from less than
1% for most local taxes to around 1.5% for the property tax (Mikesell 1982). The US costto-yield ratio is defined as administrative cost divided by revenue net of cost, however.
Using this definition to make the figures comparable, the overall cost-to-yield ratio for
local governments in Indonesia becomes as high as 110.5%.

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

220

Blane D. Lewis

One problem with the cost-to-yield ratio as a measure of local government
administrative inefficiency is that it does not account for differences in prices or
other conditions faced by local governments. Input prices are typically higher in
Eastern Indonesia than they are in Java, for example; all other things (i.e. ‘cost efficiency’) being equal, the cost of local government tax administration in Eastern
Indonesia relative to revenue generated would exceed that for local governments
in Java because of those higher input prices. Differences in tax rates and economies
of scale might also influence the cost-to-yield ratio independently of considerations related to pure cost efficiency.11 All else being the same, local governments
that employ higher tax rates will have lower cost-to-yield ratios, and if economies
of scale in tax administrations obtain, then local governments with more populous jurisdictions will also have lower cost-to-yield ratios. Beyond incorporating
such factors into a measure of inefficiency, one would also want to explain variation in relative inefficiencies across local governments. The next section of the
paper develops a method for addressing both these questions simultaneously.

MODELLING COST INEFFICIENCY IN TAXATION
This section of the paper employs a stochastic cost frontier model to develop a
preferred measure of administrative cost inefficiency and to explain variation
in inefficiency across local governments. The model is based on the underlying
assumption that the objective of local governments is to minimise administrative costs associated with producing a given level of output (which is taken to
be local revenue—see below). The minimum stochastic cost frontier comprises a
systematic component, c(x, β), and a local government-specific idiosyncratic (i.e.
random) element, v. The model posits that deviations from the minimum stochastic cost frontier, given by c(x, β) + v, are a function of local government inefficiency,
u.
More formally, the model is:
ci = β 'x i + vi + ui ,

(1)

where c is the total administrative cost of revenue generation, x is a vector of
explanatory variables, β is a vector of parameters to be estimated in the cost function, v and u together form the disturbance term (discussed below), and i is a subscript denoting the particular local government. All variables other than dummy
variables are measured in natural logarithms.
Explanatory variables in x include local revenue—or output, in the traditional
cost function framework. It would be preferable, of course, to define output as a
strict function of local government (DiPenda) administrative tasks, such as iden-

11 Any variation in legal authorities and processes across local governments might also, in
theory, influence cost-to-yield ratios (independently of changes to efficiency). In Indonesia,
however, most local governments operate within the same legal and regulatory environment. The exceptions are those in Aceh and Papua, which have special fiscal arrangements
with the central government. These exceptional provisions are concerned mainly with the
amount of intergovernmental transfers received from the centre, however, and are not likely to affect directly the cost of local tax administration or the local revenue generated.

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

Local government taxation: an analysis of administrative cost inefficiency

221

tification of taxpayers, assessment of tax liability, and tax collection, rather than
revenue generated. There are no comprehensive data, however, on how well local
governments execute administrative tasks related to revenue generation.
The rationale for using local revenue as a proxy for output is as follows. Local
revenue is a function of the local government’s performance of administrative
functions, the tax rate, and the tax base, across the mix of taxes it employs.12 As
noted above, the central government fixes maximum rates for local taxes, and
local governments have discretion in setting tariffs at levels up to these ceilings.
In practice, most local governments appear to set tariffs at or near maximum
rates.13 The centre defines local tax bases in a uniform manner across all local
governments, although the size of individual tax bases varies across local governments, of course. Perhaps the most influential factor in determining the size
of the local tax base is the local economic base (Bahl and Linn 1992). The model
developed here controls for the size of the economic base as well as for the mix of
taxes employed, as will be made clear later. Because tax rates vary insignificantly
and tax bases are defined uniformly across local governments, and because the
econometric model employs variables to control for variation in the size of local
economic bases and the mix of taxes used, it would seem reasonable to use local
revenue as a proxy for administrative output. In any case, this is the best that can
be done given the data available .
It might be useful to note that the simple cost-to-yield measure of administrative cost inefficiency also implicitly assumes that revenue is a reasonable proxy
for output. The ratio does not, however, account for variation in relative prices
across local governments, nor does it control for differences in the size of tax bases
or the mix of taxes employed among districts and municipalities, as the stochastic
frontier model developed here does. The latter at the very least would seem to
represent an advance over the simple measure, therefore.14
Local revenue is defined in the same manner as earlier described in the context
of calculating the cost-to-yield ratio. It is expected that the cost of tax administration grows as local revenue generated increases.
Other variables in x include wages and the prices of other inputs. Wages are
measured by average civil servant wages, which are estimated for each local government by dividing the total annual local government wage bill by the number
of civil servants employed by the local government. The price of other inputs is
proxied by an index developed by Ministry of Finance officials, who constructed
the measure with a view to capturing cost differentials faced by local governments.

12 Note that corruption related to official taxation may influence local government performance of any of the administrative functions—tax coverage, tax liability assessment,
tax collection—and the amount of (official) revenue generated. Owing to lack of data, the
impact of corruption on administrative performance and revenue generation cannot be
incorporated into this analysis.
13 Similarly, there appears to be little variation in levels of major charges across local
governments (Barnes et al. 2005). In addition, property tax rates are fixed at uniform levels
across districts and municipalities (Lewis 2003a).
14 Finding good measures of local public sector output for use in empirical studies like
the present one is a common and particularly vexing problem (Grosskopf and Yaisawarng
1990; Sakata 2004).

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

222

Blane D. Lewis

Price homogeneity in the cost function is imposed by dividing administrative cost
and civil servant wages by the regional (i.e. district or municipality) price index. It is
expected that administrative costs rise along with (relative) civil servant wages.
In addition, x controls for the size of the economic base in local jurisdictions.
In general, the local economic base is a function of local population and incomes
or output (Bahl and Linn 1992). Thus population15 and gross regional domestic
product (GRDP) are incorporated in x. Finally, x includes a variable denoting the
level of urbanisation in the local jurisdiction, and a location dummy variable (onor off-Sumatra) with a view to controlling for the mix of taxes used by local governments. The number and kinds of taxes local governments employ obviously
depend on the structure of their economies; the latter is associated with the level
of urbanisation, among other things. And it has been shown elsewhere that local
governments on Sumatra have been more aggressive in creating new revenue
instruments than those elsewhere; as such it might be expected that the former
employ a different mix of taxes than do other districts and municipalities (Lewis
2003b). There are no particular expectations about the directions of influence of
the control variables on administrative cost. Table 2 provides a precise definition
of the explanatory variables discussed above, and appendix table A gives basic
summary statistics for the variables enumerated and defined in table 2.
As already mentioned, the disturbance term is made up of two (independent)
components: v and u. The former is a normally distributed error term, representing positive or negative local government-specific idiosyncratic cost effects (i.e.
random noise). The second component of the disturbance term, u, comprises positive deviations from minimum cost, and can be taken as a measure of local government administrative cost inefficiency.
The two disturbance terms can be defined more precisely:
vi ~ N[0, σ 2v ]

(2)

ui ~ N[α' zi , σ u2 ]

(3)

and
ui = α 'zi + wi

(4)

where z is a vector of independent variables that explain cost inefficiency across
local governments, α is a vector of parameters to be estimated, and w is truncated
normal so that u ≥ 0. The model as laid out above is based on a long line of research,
including that of Aigner et al. (1977); Stevenson (1980); Reifschneider and Stevenson
(1991); Huang and Liu (1994); and Battese and Coelli (1995). Kumbhaker and Lovell
(2000) provide a thorough review of the research on stochastic frontier models.
Explanatory variables in z include per capita transfers (net of property tax
allocations) from the central government (and its square) and a dummy variable
indicating whether the head of the local government was elected by the local parliament or appointed by the central government (see table 2 for precise defini-

15 The use of population also helps control for possible economies of scale in tax administration.

Local government taxation: an analysis of administrative cost inefficiency

223

TABLE 2 Variable Names and Definitions for the Cost Frontier Model
Variable Name

Description

COST

Log of total administrative cost of DiPenda divided by
Ministry of Finance regional price index

Explanators of cost
OUTPUT

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

WAGE
GRDP
POP
URBAN
SUMA

Log of local government own-source revenue administered
by DiPenda
Log of local government average wage divided by Ministry
of Finance regional price index
Log of gross regional domestic product for local government
jurisdiction
Log of population
Log of percentage of the population that is urban
Dummy variable for local government location (= 1 if in
Sumatra, otherwise 0)

Explanators of inefficiency
XREVPC
Log of per capita central transfers (net of property tax)
XREVPCSQ
Log of per capita central transfers (net of property tax)
squared
DEMO
Dummy variable for head of local government (= 1 if elected
by local parliament, otherwise 0)

tions). The variables that form z have been chosen with a view to testing two main
hypotheses.
The first hypothesis concerns the relationship between administrative efficiency
and fiscal transfers from the centre. As noted at the outset of the paper, local government revenue is dominated by transfers from the central government. The
relative importance of transfers in budgets varies significantly across local governments, however; intergovernmental transfers range from about 50% to 99% of
total revenue budgets across all districts and municipalities in Indonesia.16
It is posited here that local governments view transfers as a ‘free good’, and
are therefore relatively unconcerned with their efficient use. More specifically, it
is hypothesised that increasing per capita transfers are associated with growing
administrative cost inefficiency, all other things being equal. If this were to be
the case, then one would expect per capita transfers to be positively related to
the inefficiency variable. The square of per capita transfers is also included in the
specification in order to capture any (quadratic) non-linearity in the relationship
between per capita transfers and inefficiency.

16 Local revenue in kabupaten Badung (Bali), much of which is derived from tax revenue
from tourist resorts, comprises 50% of total revenue. If Badung is excluded, local revenue
ranges from 1% to 30% of total revenue across local governments. Badung has been dropped
from the present analysis, owing to lack of data on other variables. For local governments
in the sample, local revenue varies from about 1% to 25% of total revenue.

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

224

Blane D. Lewis

The second hypothesis has to do with Indonesia’s fledgling local democracy. As
of 2003, about 60% of local governments were managed by heads elected by local
parliaments, and about 40% by heads appointed by the centre.17 It is commonly
assumed in Indonesia and elsewhere that local governments with democratically
elected heads are better governed than those whose executive has been appointed
in some manner. Accordingly it is hypothesised here that local governments with
elected heads are more cost efficient as regards revenue administration than those
with appointed heads. The relevant dummy variable takes on a value of one if the
local government head was elected, and zero if he or she was appointed by the
central government. If local governments are better managed by elected heads,
the dummy variable should be negatively related to the inefficiency variable.18

COST MODEL ESTIMATION
AND DISCUSSION OF EMPIRICAL RESULTS
The cost frontier model as represented in equations 1–4 is estimated by maximum
likelihood techniques.19 Table 3 presents the regression results.20
The table shows that both local government output (revenue) and civil servant wages are significant and positive determinants of administrative cost, as
expected. The elasticities of cost with respect to output (revenue) and wages are
0.44 and 0.38 respectively. As noted above, the other right-hand side variables in
the cost equation (GRDP, population, urbanisation and location) are included to
control for the economic base and the mix of revenue instruments employed by
local governments. The evidence shows that GRDP and population are both significantly and positively associated with cost. The examination also demonstrates
that local governments on Sumatra have lower administrative costs than those
located elsewhere. Finally, the analysis indicates that the urbanisation level of the
local jurisdiction is not significantly related to the cost of tax administration at
generally accepted levels of significance.21
A key concern of the regression analysis is to explain cost inefficiency across
local governments. Table 3 shows that inefficiency is significantly and positively
related to per capita intergovernmental transfers (XREVPC in table 3); the rela-

17 Of the 224 local governments in the sample used in this analysis, 139 (62%) were managed
by elected heads in 2003, while 85 (38%) were still governed by appointed heads.
18 The appendix tests the model specification detailed here for a variety of underlying
assumptions. The analysis shows that the specification used in the present examination is
a reasonable one.
19 Maximum likelihood estimations were performed using LIMDEP econometric software;
see Greene (2002) for a detailed explanation of methods. In addition to the truncated normal
model employed here, LIMDEP allows for the use of half-normal, gamma and exponential
stochastic frontier model specifications. These other models were also tried. Estimation
results were very similar under all four models. The conclusions reached here are robust
with regard to choice of model.
20 λ and σ are estimated directly via the maximum likelihood procedures. The standard
deviations of the two error terms are derived from the estimated values of λ and σ.
21 The insignificance of the urbanisation variable may at least in part be due to its strong
collinearity with other variables, especially local revenue and GRDP.

Local government taxation: an analysis of administrative cost inefficiency

225

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

TABLE 3 Cost Frontier Model Regression Output a

Cost frontier
Constant
OUTPUT
WAGE
GRDP
POP
URBAN
SUMA
Inefficiency parameter
Constant
XREVPC
XREVPCSQ
DEMO
λ = (σ u / σ v )
σ = (σ u2 + σ 2v )1/2

Coefficient

t-value

–2.34
0.44
0.38
0.21
0.14
0.03
–0.14

–1.01
9.73*
2.44*
4.29*
2.50*
1.24
–2.43*

–29.57
5.00
–0.20
–0.14
1.70

–0.29
2.10*
–1.71
–0.93
2.84*

0.42

5.02*

σv

0.21

σu

0.36

Number of observations
Log likelihood
a

224
–34.81

See table 2 for an explanation of the variables.

* Coefficient is statistically significant at the 5% level.

tionship between inefficiency and transfers squared (XREVPCSQ) is not statistically significant at conventional levels. The results support the hypothesis that
local government tax administration is increasingly cost inefficient as transfers
from the centre rise.
The Indonesian system of intergovernmental transfers has not been designed
to encourage improvements in local government performance of any kind. On the
contrary, some rather clear performance disincentives are apparent in the transfer system. Transfers have traditionally been used, for example, to fund at least
part of the local government wage bill, without firm restrictions being placed on
hiring. This design encourages over-staffing in general, and arguably helps to
explain the positive association between transfers and cost inefficiency found in
the analysis here.22
22 Lewis (2002) discusses performance disincentives in the Indonesian transfer system.
Starting in 2002, 50% of DAU funds were allocated to cover local civil servant salaries.
From 2006, the central government has decided to cover 100% of the local public sector
wage bill from the DAU, thus further weakening incentives to rationalise local public
sector employment.

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

226

Blane D. Lewis

Finally, the examination shows that the method by which local government
executives were chosen (DEMO) had no effect on administrative cost efficiency.
Apparently, the structures of accountability that result from local parliaments’
election of local government executives in Indonesia are insufficiently robust
to engender more efficient practices in tax administration. In mid-2005, citizens
began to elect heads of local government directly. It is expected that by the end
of 2006 all local governments will have popularly elected executives. There is not
yet, of course, any evidence about the extent to which this change has begun to
lead to more accountable and efficient local governments.
The above econometric results for the cost frontier model can be used to produce estimates of the inefficiency parameter, u. The inefficiency parameter can be
approximated by the following expression due to Jondrow et al. (1982), as cited
and described in Greene (2000) and Greene (2002).

⎡ ⎛ ελ μ ⎞

⎢ ⎜ + ⎟


ελ μ ⎥
σλ ⎢ φ ⎝ σ σλ ⎠
+
E[u | ε] =

⎟⎥ ,


1 + λ 2 ⎢1 − φ ⎛ ελ + μ ⎞ ⎝ σ σλ ⎠⎥


⎥⎦
⎢⎣
⎝ σ σλ ⎠

(5)

where ε = v + u, σ = (σ u2 + σ 2v )1/2 , λ = (σ u / σ v ) , μ = α’z, and ϕ and φ are the standard
normal probability density and cumulative distribution functions, respectively.
Ε[u⎪ε] is calculated by using estimated values of the underlying parameters on
the right-hand side of equation 5, as derived in the frontier regression model and
shown in table 3.
The estimates of u can be employed to provide an index of administrative cost
inefficiency (CI) of individual local governments (Coelli et al. 1998; Sakata 2004).
The cost inefficiency index is given by:
CI i =

E( ci | xi , ui )
E( ci | xi , ui = 0)

,

(6)

where the numerator is the expected actual cost of tax administration and the
denominator is the expected minimum or ideal administrative cost.23 Given the
functional form of the cost function estimated here, it holds that:
CI i = exp(ui ) .

(7)

CI assumes a value between one and infinity and indicates the degree to which
local government administrative costs exceed the minimum cost frontier, given
revenue, prices and control variables. A histogram of the inefficiency index is
shown in figure 2. Note the skewed shape of the distribution of the index, which

23 Other analysts prefer to use an efficiency index (Kumbhaker and Lovell 2000). The
efficiency index is the inverse of the ratio shown on the right-hand side of equation 6. It
ranges in value from zero to one and indicates the extent to which actual cost approaches
the ideal minimum cost.

Local government taxation: an analysis of administrative cost inefficiency

227

FIGURE 2 Histogram of Inefficiency Index of Local Governments, 2003
No. of local
governments
70
60

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

50
40
30
20
10
0
1.1

1.2

1.3

1.5

1.6

1.7

1.9

2.0

2.1

2.3

2.4

2.5

2.7

2.8

2.9

3.1

Inefficiency index (%)
Source: As for table 1

is similar to that of the simple cost-to-yield ratio shown in figure 1. The mean
value of the inefficiency index is 1.40 (median 1.34); the index values range from
1.10 to 3.06 (standard deviation 0.29). These figures indicate a generally high level
of cost inefficiency in local government tax administration and, at the same time, a
significant variation in administrative performance across local governments.24

SIMPLE AND COMPLEX COST INEFFICIENCY MEASURES COMPARED
A question that naturally arises concerns the extent to which the cost-to-yield
measure of inefficiency differs from the inefficiency index derived from the stochastic cost frontier model. This may be especially important in the context of
the central government’s monitoring and evaluation effort, as described at the
outset of the paper. That is, it would be useful for government officials to know
the degree to which they might realistically rely on the simple measure of administrative cost inefficiency, and whether instead they might need to develop other,
perhaps more sophisticated, measures. 25

24 By comparison, in an examination of the costs of providing fire protection services in
Japanese local governments, Sakata (2004), using an empirical model similar to the one
employed here, found an average inefficiency index of 1.09 and a range of index values
from 1.02 to 1.44 (with a standard deviation of 0.07).
25 In the analysis that follows, costs in the cost-to-yield ratio measure of inefficiency have
been divided by the Ministry of Finance price index, as was done in the cost frontier model.

228

Blane D. Lewis

FIGURE 3 Local Government Rank by Inefficiency Index and Cost-to-Yield Ratio, 2003
Inefficiency index rank
225

F
F
F
F
F
F
F
FFF
F
F
F
F
F F F F FF
FFF
F
FF
F
F
F
F
FFF F
FF
F
F
F
F F F
180
F
F
F
F
FF
F
F
F
F
F
F
F
F
F
F
F
FF
F
FF
FF
F
F
F
F
F
F
F
F
F FF
F F
F
F
FFF
F
F
F
F
135 F
F FF
F
F
F
FF
F
F
F
F
F F
F
F
FF
F
F FF
F
F
F
F
F
F
F
F
F
F
F
F F
F
F
F
F F
F F
F
90
F F
F
F
FF
FF F
F FF
F F
F F
F FFFF
F
F F
F
FF F F
F
F
F
F
F
F
F
FF F
FF
FF
F
F
45
FF
F F
F
F
F
F
FF F F
FF
F
F
F FF
F
F
FF F
F
F
F F F
F
FF
F
F
F
F FF F F
F
F F FF
F

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

F

0

0

45

90

135

180

225

Cost-to-yield ratio rank

As can be seen from figures 1 and 2, the shapes of the distributions of the simple and complex measures of inefficiency are reasonably similar. Pearson’s correlation coefficient for the two measures is 0.726, which is significant at the 1%
level. This suggests a reasonable degree of association between the simple and
complex indices of inefficiency. Whether or not the two measures are sufficiently
comparable depends at least in part on how they might be used.
The results here suggest that the simple measure might perhaps serve the general purpose of monitoring average local government capacity to administer taxes
efficiently, as well as the variation in administrative capacity across local governments.
But would the simple measure offer a good indication of a particular local government’s administrative capacity? Since both parameters are best thought of as
indicators of relative inefficiency, an initial answer to the question can be gleaned
by comparing local government rankings according to the two measures. The relevant calculation reveals a rank order correlation of 0.685, which is also statistically significant at the 1% level.
Figure 3 illustrates the relationship between rankings according to the inefficiency
index and rankings according to the cost-to-yield ratio across all local governments
in the sample. The 45-degree line in the figure highlights where local governments
would fall if they had equivalent rankings by both measures. The general statistical

Local government taxation: an analysis of administrative cost inefficiency

229

TABLE 4 Distribution of Local Governments by Performance Measures
(no. of local governments)

Downloaded by [Universitas Maritim Raja Ali Haji] at 21:42 18 January 2016

Cost-to-Yield
Ratio

Cost Frontier
Inefficiency
Index

Top 20%
Middle 60%
Bottom 20%
Total

Agreement
Rate

Top
20%

Middle
60%

Bottom
20%

Total

%

24
20
1
45

21
98
16
135

0
17
27
44

45
135
44
224

53.3
72.6
61.4

association between the two ranked indices notwithstanding, figure 3 demonstrates
quite significant differences in the rankings across the two measures.
Perhaps the rankings might best be employed to place local governments in
broad categories of performance. With a view to examining the possibilities of this
approach, I use the rankings of local governments by the two measures to categorise them into groups representing the top 20% (the most cost efficient), the middle
60%, and the bottom 20% (the least cost efficient). Table 4 shows the distribution
of local governments by performance class according to the two measures.
The far right column of the table shows the extent to which the two standards
agree on group membership. As can be seen, only 53% of the top 20% of local governments by one measure are placed in the top 20% of performers by the other
measure. The agreement rate, at 73%, is highest for the middle 60% of performers,
as might be expected. About 61% of local governments in the lowest 20% of performers by one measure are placed in the same category by the other measure. In
general, these results do not inspire confidence in the use of the simple cost-to-yield
ratio as a measure of individual local government administrative performance.26

SUMMARY AND CONCLUSIONS
Local governments are, in general, very inefficient in matters related to tax administration. This paper estimates that the cost of administering local taxes and charges
is over 50% of local tax and charge receipts. Results derived from the specification and estimation of a stochastic cost frontier model support the conclusion that
local government tax administration is, on average, inefficiently carried out.
Weak average performance notwithstanding, there seems to be considerable
variation in the capacity of local governments to administer taxes efficiently. Each
of the measures of inefficiency employed in this paper—the cost-to-yield ratio
and a cost frontier inefficiency parameter—demonstrates the existenc