Abstract Using a panel threshold model, we examine the heterogeneous effects of foreign aid on tax revenue due to government stability in the West African Economic and Monetary Union countries over the period 1986-2010. Panel Smooth Threshold Regressions indicate the existence of strong threshold effects in the aid-tax rela- tionship depending on the level of government stability. They also indicate that the effect of aid on tax revenue is gradual and varies across countries according to the level of government stability. We find that aid directly reduces tax revenues but for higher levels of government stability it enhances tax performance. We provide estimates of country time-varying coefficients of aid effect. We find on average a positive impact of aid. However, the size of this impact is very small suggesting that there is still much to do at the institutional level to improve the effectiveness of aid for tax performance in WAEMU countries. Keywords: Foreign aid, Government Stability, Tax Revenue, PSTR, WAEMU JEL-Classification: F35, 017, H20, C23, O55 2
36
Embed
Keywords: Foreign aid, Government Stability, Tax Revenue ...cerdi.org/uploads/pagePerso/11/Main document JAE accepted version.… · Smooth Threshold Regression (PSTR) recently developed
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Abstract
Using a panel threshold model, we examine the heterogeneous effects of foreign
aid on tax revenue due to government stability in the West African Economic and
Monetary Union countries over the period 1986-2010. Panel Smooth Threshold
Regressions indicate the existence of strong threshold effects in the aid-tax rela-
tionship depending on the level of government stability. They also indicate that
the effect of aid on tax revenue is gradual and varies across countries according to
the level of government stability. We find that aid directly reduces tax revenues
but for higher levels of government stability it enhances tax performance. We
provide estimates of country time-varying coefficients of aid effect. We find on
average a positive impact of aid. However, the size of this impact is very small
suggesting that there is still much to do at the institutional level to improve the
effectiveness of aid for tax performance in WAEMU countries.
Keywords: Foreign aid, Government Stability, Tax Revenue, PSTR, WAEMU
JEL-Classification: F35, 017, H20, C23, O55
2
1 Introduction
Whether or not foreign aid hinders tax mobilization is a question of great importance
for development funding. Despite the recent growing interest of donors and research
into the aid / tax relationship, we still have a limited knowledge of the precise impact of
aid on tax revenues. The literature remains theoretically and empirically inconclusive.
According to aid skeptics, aid reduces the incentive of the recipient countries to reform
and implement good policies to increase revenues (Azam et al., 1999; Remmer, 2004;
Knack, 2009; Brautigam and Knack, 2004). The findings of Gupta et al. (2003) and
Benedek et al. (2012) provide evidence that aid harms tax performances. These results
have been challenged by Morrissey et al. (2006), Clist and Morrissey (2011) and Brun
et al. (2011b) who emphasize the enhancing effect of aid on a population’s tax compli-
ance, and the capacity of tax administrations. However, the importance of governance
quality for a positive impact of aid on tax performance tends to be less controversial
(Gupta et al., 2003; Brun et al., 2011a,b; Benedek et al., 2012).
Governance is a multi-faceted phenomenon, but stability plays a central role in
the ability of governments to adopt and implement good policies, suggesting that the
impact of aid on tax revenues may depend on government stability. Indeed, Alesina
and Perotti (1996), Alesina et al. (1996) and Carmignani (2003) argue that government
instability may generate uncertainty about the sustainability of the current and future
course of economic policies. In particular, when governments have doubts about their
survival in office, they may forego their initial commitments by delaying or reversing
the required structural reforms (Carmignani, 2003). The resulting frequent ignoring
of their commitments can be expected to undermine people’s trust in government to
promote good quality services and growth. This contributes to lower the population’s
tax compliance so that the enhancing effects of aid through this type of action may not
be effective. Moreover, government instability in developing countries often causes an
upheaval in the whole of the administration, including the competent and less corrupt
staff in charge of tax reforms. This may complicate the discussions with the funders
and compromise a good follow-up of the initial reforms. For instance, Fossat and
Bua (2013) point out that the implementation of tax reforms in sub-Saharan African
(SSA) francophone countries is impaired by the inadequate political support and the
insufficient commitment of finance ministers and tax administration managers due to
their frequent turnover.
Although such research sheds light onto our understanding of the role of governance
in the aid/tax relationship, there remain, however, some challenges for econometric
3
specifications as for the whole aid/tax literature (Prichard et al., 2013). In particular,
how the existing empirical studies examine this conditional effect of aid on tax related
to governance quality is somewhat questionable. They often use subsamples, group
dummies, or interactive terms (typically the aid × governance variable). According to
Gonzalez et al. (2005), Fok et al. (2005), Hansen (1999), and Colletaz et al. (2006),
such approaches suffer from several limitations and may pose some important inference
problems. Subsamples and group dummies are set arbitrarily (and according to only
one criteria) and do not allow a country to move from one group to another. The
interactive term assumes that the impact of governance on the aid/tax relationship is
invariant, and is the same for all countries and over time. Yet the impact of governance
may vary simply due to the fact that recipient governments have learned from their
failures and/or successes in policy implementation and aid management year by year, a
so-called effect of learning by doing. These methods may lead to an underestimation of
the role of heterogeneities across countries and over time, and so to misleading policy
implications.
The goal of this paper is to thoroughly address the issue of heterogeneity in the
aid/tax relationship by focusing on the role of government stability using a Panel
Smooth Threshold Regression (PSTR) recently developed by Gonzalez et al. (2005).
This approach allows for strict testing for any non-linearity to determine endogenously
the threshold level of government stability where the effect of aid on tax revenue shifts
critically. It then also provides time-variant estimates for each country.
Our sample consists of the countries of the West African Economic and Monetary
Union (WAEMU). The aid/tax relationship in these countries has not been studied in
detail until now; the interest is that they share a number of common fiscal schemes,
but they differ in political stability. Since 1994, they have adopted convergence criteria
targeting inflation, public debt, and deficits. To reach these goals, a common fiscal
transition program has been adopted aimed at increasing tax revenues to over 17%
of GDP, this has recently been increased to 20%. In addition, administrative reforms
have been undertaken to decentralize the fiscal administration, and audit offices and
computer systems have been established to improve public financial management. Tax
administrations have also adopted a code of ethics and a code of practice. The objectives
are to eliminate fraud and corruption, and to improve tax compliance. In spite of these
reforms, the tax to GDP ratio remains below the target, reaching about 15.6% of GDP
on average during the last decade. However, large differences between the 8 countries
remain (World Bank, 2014; Keen and Mansour, 2010). Aid reliance remains high with
the aid-to-GDP ratio at 10% for 2000-2010 against 17.8% for 1987-1999. The degree
4
of political stability also varies across countries. For instance, Benin and Senegal have
successfully engaged in a democratic process since 1990, but only after a series of
military coups in the 1970s and 1980s in Benin. The democratic transition in Cote
d’Ivoire and Togo is not yet complete. These diverse experiences may have affected the
quality of policy decision-making and the effect of aid differently from one country to
another, even if they share common fiscal rules.
This paper is organized as follows. Section 2 reviews the existing literature on
the effect of aid on tax performance. Section 3 outlines the fiscal and institutional
challenges, as well as the reliance on aid in the WAEMU. Section 4 describes the
methodology and data. Section 5 discusses the results, and Section 6 offers conclusions.
2 Literature review
The literature does not enable us for now to reach a conclusion about the precise impact
of aid on tax revenues. Indeed, there are theoretical arguments that aid can promote or
hinder the recipient tax revenues. Harmful effects may come from adverse incentives,
difficulties in public administration, or economic instability. For instance, governments
may use foreign aid to avoid the social costs induced by the tax burden. For instance,
Azam et al. (1999) demonstrate that aid does not encourage the recipient governments
to adopt good policies and develop efficient institutions, that is to say to develop an
efficient domestic tax system. Moore (2001), Martens et al. (2002) and Moss et al.
(2006) and Svensson (2006) underline that aid may lead the recipient government to
privilege donor satisfaction to the detriment of accountability towards citizens. This
may divert public administrations attention towards aid projects to the detriment of
the tax administration. Furthermore, in degrading the quality of provision of public
services, this may hinder domestic tax compliance and so revenues. As well as the level
of aid dependency, the volatility of aid poses serious problems for the definition of the
budget and more generally for the administration of public expenditure. The lack of
coordination between multiple donors may dramatically increase these adverse effects
on public administration (Knack and Rahman, 2007; Kanbur et al., 1999; Brun et al.,
2011a). As well as the direct adverse effects on tax administration, aid can also have
some adverse consequences for macroeconomic stability and economic growth leading
to a reduction in the tax base (Gupta et al., 2003).
The positive effects of aid on tax also rely on incentives, a direct impact on public
administration and more indirectly through its impact on the economy. The various
costs explained above may give incentives to government to reduce their aid dependence,
5
particularly by increasing their tax effort (Brun et al., 2011a). It can also be expected
that aid in the form of technical assistance should strengthen directly the capacities of
the tax and customs administrations, facilitating spending and tax reforms. Finally,
aid should enhance tax performances indirectly by improving the effectiveness of public
expenditure, human development, and tax compliance (Morrissey, 2015). Inconclusive
lessons from the theoretical literature are also reflected in the empirical works, especially
since findings are sensitive to data quality, sample and econometric approach (Moss
et al., 2006)
As regards general tests and findings, in a seminal work Heller (1975) found that aid
had a negative effect on tax revenues for 11 African countries. In contrast Khan and
Hoshino (1992) found a positive impact for 5 South and South-east Asian countries.
Ouattara (2006) does not find a significant relationship using a sample of 46 developing
countries. On larger samples, Remmer (2004) finds that aid lowers tax revenues.
Empirical investigations have also examined the impact of aid composition on the
aid/tax relationship. The basic argument is that loans encourage tax effort since they
must be reimbursed, in contrast to unconditional grants. This argument has been
marked by the recent debate between Gupta et al. (2003), Benedek et al. (2012) on the
one hand and Clist and Morrissey (2011), and Morrissey et al. (2014) on the other hand.
Gupta et al. (2003) and Benedek et al. (2012) find that overall aid and grants reduce
tax effort while loans show a positive effect on samples of 107 developing countries
over 1970-2000 and 118 countries over 1980-2009, respectively. These results were
severely challenged by Clist and Morrissey (2011) who show that aid variables become
statistically insignificant when they are lagged. Clist (2014), using the data of Benedek
et al. (2012), and a new data set developed by Prichard et al. (2014), finds a modest
positive effect of aid on tax revenues, whereas the negative effect of aid grants is not
robust. Morrissey et al. (2014) report very similar findings.
Many authors have suggested investigating heterogeneities across countries and over
time. The effect of aid may differ from one country to another according to the level
of development, the region, the level of aid, and the quality of policies and institutions.
The country-level analyses support the idea that heterogeneity matters (Brun et al.,
2011a). However, single country analyses are limited by the lack of data and do not
allow a robust comparison between countries. This certainly explains the preference for
panel data in most studies. 3 major techniques are used to highlight heterogeneities
in panel studies: the use of subsamples, the introduction of regional dummies, and the
use of interaction terms. The subsample technique consists of splitting the sample into
subgroups according to such criteria as the location in a specific region, a time period,
6
or a defined threshold of a specific economic or institutional variable. It is generally
used as a robustness test to check whether results differ between the whole sample and
across subsamples. A regional split was used for instance by Benedek et al. (2012) who
find a negative relationship for total aid and grants in Africa, Asia and the Pacific while
loans show a negative impact only for African countries. On the contrary, Morrissey
et al. (2014) show a positive effect of grants in Sub-Saharan Africa and Latin America
and the Caribbean. Also using regressions over various decades, the latter authors do
not find any evidence of harmful effects of aid.
Less controversially, empirical findings tend to demonstrate a conditional impact
of aid on tax depending on governance quality. By the use of subsamples built on
country corruption levels, Gupta et al. (2003) find that the negative effect of grants
is substantially amplified, while the positive effect of loans vanishes in highly corrupt
countries. Using the same approach Benedek et al. (2012) show that both grants and
loans have a negative effect when corruption is high. Brun et al. (2011a), interacting
aid and governance variables, find that only the quality of bureaucracy conditions the
impact of aid on tax effort. Brun et al. (2011b) using interaction variables also conclude
that IMF programs are less effective in promoting tax effort in SSA countries that are
characterized by weak institutions. On the other hand Alonso and Garcimartın (2011)
show that the aid/tax relationship is not conditional on institutional quality.
However, these previous works present some serious drawbacks that we aim to re-
solve. First, subsamples impose an ad hoc choice on only one conditional variable
(i.e. either the region or the quality of governance) and on threshold values (allow-
ing splitting of the sample between low and high governance quality for instance) and
usually does not allow countries to move from one group to another. Second, as Brun
et al. (2011a) noted, this does not allow testing of the significance of the difference be-
tween subsample estimates. The use of group-dummies in the overall sample estimates
presents the same drawbacks. Finally, taking account of heterogeneities is limited since
these techniques give estimates of an average impact for all countries belonging to each
group. As regards the use of interactive terms, this only tests for a bilinear interaction
impact where the slope between tax revenues and aid changes as a linear function of
institutions. It constrains the nonlinearity to be of a particular shape and completely
ignores the possibility of multiple thresholds. It provides a common conditional aver-
age estimate of the effect of aid on tax revenues for all the countries, ignoring that the
effects may change gradually over time within a country due to the potential effect of
learning by doing and/or depending on the experience of each country.
Carter (2013) provided an interesting discussion about these kinds of effects. But
7
he uses a Pool Mean Group (PMG) approach which only allows for short-term het-
erogeneity, and imposes a common long-term relationship. We propose here to use a
PSTR estimation which is a more flexible approach, to account for heterogeneity across
countries and over time.
3 Stylized facts of the aid/tax relationship and reforms in the
WAEMU
Since the 1980s, the 8 WAEMU members have successively engaged in difficult macroe-
conomic reforms and ’new generation’ reforms. The former reforms were developed in
the framework of the Structural Adjustment Programs (SAPs) from the 1980s to the
middle of the 1990s. They aim to stabilize macroeconomic imbalances and to counter-
act recessions under the auspices of the IMF and the World Bank. They essentially
consist of resizing public services, drastically cutting state interventions in productive
activities, liberalizing trade and stimulating the private sector.
The ’new generation’ reforms have been implemented through the Millennium De-
velopment Goals (MDGs) since the mid-1990s. They translate at the national level into
the Poverty Reduction Strategy Paper (PRSP) in which the government declares its
development goals and targets, and the measures to attain them. These efforts received
the support of the international community, notably through the Heavily Indebted Poor
Countries (HIPC) initiative and through the Multilateral Debt Relief Initiative (MDRI)
in 2006 (African Development Group, 2011). All the member countries of the Union
benefited from these debt relief initiatives. Cote d’Ivoire became the last recipient in
2012 after a decade of military and political instability.
The core message of the PRSP is to free up additional resources for poverty reduction
and development programs. The program package encompasses significant institutional
and economic reforms. The major reforms in WAEMU have been developed in the
context of fostering the process of regional integration. In 1994, the Union adopted
convergence criteria targeting inflation, public debt, and deficits. The major concern
was then to prevent macroeconomic instabilities due to the CFA Franc devaluation and
to generate sustainable economic growth. Now, after changes in 1999 and 2003, the
Pact breaks down the criteria into key and secondary criteria. The key criteria are those
whose violation leads to the formulation of corrective actions or even sanctions (Bamba,
2004). The secondary criteria are considered as indicative structural benchmarks to
achieve internal and external balance. However the failure to meet them does not
8
result in corrective measures, but only recommendations1.
As concerns fiscal matters, a regional harmonization program in the tax and cus-
toms areas has been developed. VAT and Excise directives were introduced in 1998
with the main objective of gradually substituting domestic taxes for trade duties. In
2000 a customs union with a common external tariff was established. A Fiscal Transi-
tion Program was adopted in 2006 that essentially used the same goals as the VAT and
Excise directives. The aim is to achieve a tax burden of 17% of GDP with 10% derived
from domestic revenues and 7% from import taxes (Mansour and Graziosi, 2013; Fos-
sat and Bua, 2013). External technical support, in particular from the IMF-AFRITAC,
was engaged to improve tax administration capacities through strategy design, organi-
zational reforms, procedural reforms, IT and human capacity building, strengthening
transparency and integrity, fighting corruption and fraud, and increasing tax compli-
ance, etc. In addition, in 2004 the financial support of the WAEMU’s central bank, the
Central Bank of West African States (BCEAO), for national budgets was stopped. In
2007, the BCEAO gained independence and reinforced its mission of fighting inflation.
In short, the convergence Pact aims to ensure greater fiscal discipline in support of
the common monetary policy in order to create favorable conditions for price stability
and strong sustainable growth. However, it leads to a number of challenges. First,
the suppression of the seigniorage means that the member states are able to properly
exploit a disposable fiscal space in support of the objective of the 7% growth required
to substantially reduce poverty in the Union. However, the Union has not been able to
achieve this goal. Between 1994 and 2012, the per capita GDP growth was about 1.4%
against −1.3% for the period 1980-1993 (World Bank, 2014). Second, Table 1 shows
that a number of the 8 members violate the convergence criteria. Only the criterion of
the Total debt to GDP ratio is met by all the members, thanks to HIPC debt relief.
Remarkably, the key criteria of fiscal balance and tax revenue are the most violated
criteria, along with the current account balance, despite tax reforms. This results in
persistent public deficits and high dependency on aid as Table 2 highlights.
Over the period 1987-2010, the fiscal balance is negative in all members, with a few
1The key criteria are: i) The Ratio of fiscal balance to nominal GDP (key criterion) should begreater than or equal to 0% in 2002. Its non-compliance results in sanctions, except in exceptionalcircumstances such as those defined by Regulation No.11/99/CM/UEMOA, Article10; ii) The averageannual rate of inflation should not exceed 3%; ii) The ratio of outstanding debt to nominal GDP shouldnot exceed 70% by the year 2005; iv) Current payment arrears should not be generated. There are also4 secondary criteria: i) The tax burden should reach at least 17% of GDP. ii) The ratio of public wagebill to tax revenue should not exceed 35%; iii) The share of domestically-funded public investmentshould reach at least 20% of tax revenues iv) The ratio of the current account balance excluding grantsto nominal GDP should be greater than or equal to 5%
9
Table 1: WAEMU–Number of countries violating convergence criteria, 2010-2013
WAEMU -8.0 -5.2 -2.5 -3.3 12.2 12.0 12.8 14.7 17.5 17.9 10.42 10.7Source: Authors from Keen and Mansour (2010), various IMF-article IV reports, World Bank (2014)
exceptions, but with a general improvement after 1994. The average ratio decreases
from − 8% over the period 1987-1993 to −5.2% and −3.3% respectively for the periods
1994-1999 and 2007-2010. Aid inflows dramatically decrease while the tax rate increases
slowly from 12% of GDP to 14.7%. These slow improvements can be explained to
some extent by the various reforms undertaken at sub-national level, but at the same
time leads to questions about their effectiveness. The situation varies significantly
between the member states. A closer look at Table 2 reveals that the higher the
income level, the lower is aid dependency, and the higher is the tax revenue ratio. The
apparent sensitivity of tax revenue to aid inflows also varies considerably. Mansour
and Graziosi (2013) argue that the lack of credibility, due to the absence of a clear
mechanism for sanctions in the WAEMU explains the gaps between ’de jure and de facto
coordination’ characterized by flexibility in defining their tax bases and rates which is
induced by various directives, and ongoing tax competition through special tax regimes.
Heterogeneities among countries are also explained by the different changes in the GDP
composition among members.
Inconsistency depends on the quality of institutions, in particular the ability of
10
governments to lead successful reforms despite pressure from interest groups, their
ability to promote a safe environment, and to gain popular support. When the risk
of instability is low, a fiscal contract between government and citizens is more likely,
aimed at decreasing the reliance on foreign assistance. But the fiscal contract might
also enforce a mutual consensus to not tax and so maintain aid reliance.
The graphs below show the changes in average government stability, foreign aid
and tax ratios in WAEMU for 1985-2010. The 1990s were marked by a low level of
government stability when the SAPs were implemented. In contrast, the reforms of
1994-1999, and the development of PSRP documents involving a large participation of
populations, benefited from a continuous improvement in government stability. After
the peak in 1999, the index of government stability shows a decreasing trend until 2005
before a slight increase. After 1996 when a marked improvement of government stability
is noted, the tax to GDP ratio moved above the foreign aid ratio. 2
Figure 1: Aid, tax revenue and government stability in WAEMU
46
810
scor
e
510
1520
perc
ent o
f GD
P
1985 1990 1995 2000 2005 2010years
Tax collection Foreign aidGovernment stability index
Source: Authors from WDI, Keen and Mansour (2010) database, various IMF-article IV reports andWorld Bank (2014) and PRSP database
In short, the main characteristics of the tax system in WAEMU do not change
significantly because of a lack of credibility in enforcing the reforms, maintaining tax
exemptions, fraud, and the presence of heterogeneities despite the support of inter-
2The peak in foreign aid in 1994 is mainly due to the devaluation of local currency which inflatesthe domestic currency nominal value of aid.
11
national partners. The purpose of this paper is to test econometrically if there is a
threshold level for the government stability index beyond which aid inflows promote
tax collection.
4 Methodology and data
In this section, we present the Panel Smooth Transition Regression (PSTR) model and
show how it contributes to accounting better for the role of institutional heterogeneity
in the aid/tax relationship in WAEMU countries. We assume that this institutional
heterogeneity is related to government stability as the latter determines the probability
of occurrence of the necessary reforms and the quality of their implementation, which
in turn affect the impact of aid on tax revenues.
4.1 The Panel Smooth Transition Regression
The prior assumption of this paper is that the effect of aid on tax revenues varies
gradually year by year and across countries depending on the ability of government to
implement good policies. It requires a certain minimum threshold of quality of govern-
ment to make aid effective. The PSTR model, recently developed by Gonzalez et al.
(2005) and Fok et al. (2005), is suitable for describing such a double heterogeneity.
It provides impact coefficients of aid on tax that are function of the level of gover-
nance that varies across countries and over time. It is a regime-switching model that
allows for a small number of extreme regimes. It can be seen as a generalization of
the Panel Threshold Regression (PTR hereafter) model proposed by Hansen (1999) in
which coefficients of some explanatory variables are a function of the value of another
variable called the transition variable. In the PTR model, the modeled regime shift is
sharp, while in the PSTR the shift is modeled through a smooth transition function.
The estimated coefficients are continuous functions of an observable variable through a
bounded function of this variable (Gonzalez et al., 2005). The PSTR model addresses
both heterogeneity and time variability by allowing coefficients to vary smoothly with
respect to country and time.
As stated above, our transition variable of interest is government stability. It is
defined as the smallest risk that government does not deviate from its declared programs
and is not removed from office. This gathers three conditions. First, it assumes that
there is unity of the executive team and the cabinet around the governments general
policy goals. Second, that the government can lean hard on the legislative to implement
12
its program. As a combination of these two conditions, the third condition states
that the government has popular support. Obviously, this assumption is questionable
at certain times. In a context of fragile institutions, having legislative and popular
support or sticking to declared projects is not a guarantee against military coups or
of good reforms. One might simply admit that keeping strictly to its program and
respecting the three conditions is enough to provide a guarantee against instability,
all things being equal. A stable government should be seen as a good reformer able to
convince its citizens of the relevance of its reforms. In a highly risky environment a weak
government is constantly under threat of being removed from office. So it is tempted
to develop some suboptimal strategies to remain in office. One way that could be tried
would be to increase public spending. It has two possibilities of funding: increasing tax
or substituting aid for tax. As raising tax is costly compared to aid, it will deliberately
delay the tax reforms. Several authors have demonstrated other implications about
the uncertainty of government survival. Alesina and Tabellini (1990), Edwards and
Tabellini (1991) and Cukierman et al. (1992) emphasize that this kind of government
renounces the implementation of good policies to weaken the state for their successors.
Moreover, Murphy et al. (1991) show that an uncertain and weak government is
more prone to pleading from lobbyists and pressure groups, thus leading to a more
direct effect of rent-seeking activities on policy decisions (Alesina et al., 1996). Another
aspect of government instability concerns the staff in tax administrations. In general,
changes in cabinet also lead to changes in the assignments of, or the positions of, the tax
managers in charge of tax reforms. On one hand the risk of being ousted may make them
sensitive to rent-seeking in order to compensate the loss of their job-associated perks.
On the other hand, the frequent turnover of tax managers and staff may complicate
and make inefficient the discussions between the country and its international partners,
especially in terms of consistency and follow-up of the required reforms. Because of
these obvious economic inefficiencies, we assume that aid is more efficient in promoting
good tax collection performance at higher levels of government stability and less efficient
at lower levels.
The model we use is drawn from the standard empirical analysis on the aid-tax
issue Gupta et al. (2003), Teera and Hudson (2004), Tanzi (1992), Brun et al. (2011a)
and Clist and Morrissey (2011), and others. We augment this reduced-form model
by introducing a nonlinearity of the effects of the usual factors of tax-to-GDP ratio
depending on government stability. The basic PSTR model with 2 extreme regimes
and a single transition function is defined as:
13
taxit = µi + β′0xit + β′
1xitg(stabit; γ; stab) + uit (1)
where ı = 1, . . . , n represent countries and t = 1, . . . , T years. The variable tax is
the ratio of tax revenue to GDP, the vector xit includes the ratio of foreign aid to GDP
and the other traditional factors of tax-to-GDP ratio.
β0 and β1 are 2 vectors of parameters to be estimated. µi and uit denote the fixed
individual effects and the errors, respectively. The errors are assumed to be ı.ı.d. stabit
is the government stability index in country ı and year t. The transition function
g(stabit; γ; stab) is a continuous function of the threshold variable, so that the value of
stabit determines the value of g(stabit; γ; stab). Thus the effects of x on tax revenue for
country ı at time t is given by:
∂taxit
∂xit
= β0 + β1g(stabit; γ; stab) (2)
The transition function value is bounded between 0 and 1 defining the 2 extreme
regimes: when it equals 0, the effects of x on tax revenue equals β0 and when it equals
1, the effects of x on (the x-elasticity of) tax revenue equals β0 + β1. Granger and
Terasvirta (1993) and Gonzalez et al. (2005) specify g as the following logistic function:
g(stabit; γ; stab) =
1 + exp(−γm∏ȷ=1
(stabit − stabȷ))
−1
(3)
where stab (stab = stab1, . . . , stabm)′ is an m-dimensional vector of location (threshold)
parameters and the estimated term γ measures the slope of the transition function.
As we briefly noted, in comparison with the use of sub-sample and the use of a
simple interaction term, the PSTR provides some advantages. It allows coefficients (in
particular aid coefficient) to vary between countries and over time. Moreover, in the
PSTR model, there are as many values of the (aid) impact coefficient(s), lying between
β0 and β0 +β1 , as country-year observations. This is why in the following we interpret
only the signs of β0 and β1 rather than their values.
Regarding the value of the slope γ, when it equals 0 the transition function reduces to
a constant and the model is the standard linear model with individual effects, i.e. con-
stant and homogeneous coefficients; if it tends towards infinity, the transition function
becomes an indicator function and the PSTR model in (1) reduces to the two-regime
PTR model of Hansen (1999) in the case where m = 1, for instance. When m ≻ 1 and
γ tends to infinity, the number of regimes remains 2 but the function switches between
0 and 1 (Colletaz et al., 2006).
14
The control variables reflect the sector composition of the economy, the initial level
of economic development, trade openness, macroeconomic policies, and the quality of
the institutional environment. Institutional quality improves tax collection (Brun et al.,
2011b; Benedek et al., 2012). Agricultural and industrial value added as a percentage
of GDP is used as proxy of economic structure. A large share of agriculture in total
output and employment, being largely a subsistence activity, lowers the possibility of
a modern tax system based on personal income taxes and value added taxes (Tanzi
and Zee, 2000). Agriculture could also be considered as a proxy for the informal sector
(Mahdavi, 2008), which is administratively difficult to tax (Fox and Gurley, 2005). In
addition, being a highly labor-intensive sector, it often employs children at the expense
of school enrollment. It is then less demanding of public services and activities (Tanzi,
1992) and is associated with low tax ratios. On the contrary, the industrial sector is
easier to tax (Clist and Morrissey, 2011).
The correlation between tax revenues and real GDP per capita, is expected to
be positive as economic development increases both the demand for public services
and the tax base. Trade openness is proxied by the sum of exports and imports as
a percentage of GDP. Trade taxes are likely to be easier to collect. According to
Rodrik (1998) and Gupta (2007), trade openness calls for a greater role for the public
sector in providing social insurance in more open economies subject to outside risks.
But the trade liberalization reforms engaged in the region like in the other developing
countries make this enhancing effect uncertain (Agbeyegbe et al., 2006; Benedek et al.,
2012). Indeed, quantitative barriers have been replaced with import duties. This could
result in higher trade tax revenues depending on the level of duties and on the change
in import volumes. The reforms have also involved reductions in tariffs. But these
changes have also often been compensated by increases in tax pressure from VAT and
other taxes (Cnossen, 2015; Bird and Gendron, 2007) as well as a strengthening of
the tax administrations’ capacities by donors3. The expected net enhancing effect of
these reforms may vary over time and country depending on the depth of involvement
of each country and lessons from earlier implementations. Inflation, measured by the
percent change in average consumer prices, is assumed to harm tax revenues since
it negatively affects their real value following the so-called Oliveira-Tanzi effect. In
addition, we construct a dummy variable to capture the major reforms engaged in the
WAEMU area since the devaluation in 1994. It takes the value of 1 in 1994, 1998-2000,
2002, and 2006-2007, and 0 otherwise, in line with the above overview of public and
macroeconomic reforms. To further robustness, we include the level of foreign debt to
3We are very grateful to the referee for this suggestion.
15
proxy the need to generate revenue to service the debt, and population growth to proxy
the potential increase in the tax base. It is expected that the effects of the control
variables on tax improve at higher levels of government stability.
4.2 Estimation and tests of the specification
The estimation procedure consists of eliminating the individual effects µi by removing
country-specific means and applying non-linear least squares to the transformed model.
Gonzalez et al. (2005) propose a testing procedure which proceeds as follows i) testing
the linearity against the PSTR model (or testing homogeneity against the PSTR alter-
native), ii) determining the number, r, of transition functions, that means the number
of extreme regimes which is equal to r+1. The test of homogeneity in the PSTR model
can be done by testing: H0 : γ = 0 or H0 : β1 = 0. However under the null hypothesis,
the tests are non-standard as the PSTR model contains unidentified nuisance param-
eters. This identification problem is circumvented by replacing g(stabit; γ; stab) by its
first order Taylor expansion around γ = 0 and to test with an equivalent hypothesis
based on the auxiliary regression:
taxit = µi + β′∗0 xit + β
′∗1 xitstabit + ... + β
′∗mxitstab
mit + um
it (4)
Hence testing the linearity of aid-tax model against PSTR is equivalent to testing
H∗o : β∗
1 = ... = β∗m = 0 in equation(4). SSR0 being the panel sum of squared residuals
under H0, and SSR1, the panel sum of squared residuals with regimes, the correspond-
ing F-statistic is then defined by: LMF = (SSR0−SSR1)/mkSSR0/(TN−N−mk)
∼ F (mk, TN −N −m(k +
1)); where k is the number of explanatory variables in the aid/tax function, T is the
number of years, and N the number of countries. The test of homogeneity is also a tool
for determining sequentially the number of transitions in the model. Given a PSTR
model, we test the null hypothesis that the model is linear at a predetermined signifi-
cance level α. If it is rejected, a two-regime PSTR model is estimated. If the two-regime
is in turn rejected a three-regime is estimated. The testing procedure continues until
the first acceptance of the null hypothesis of no remaining heterogeneity. At each step
of the sequential procedure, the significance level must be reduced by a constant factor
0 ≺ τ ≺ 1 in order to avoid excessively large models.
To address the issue of a potential endogeneity bias, it is common to employ various
instrumental methods. But the results are very sensitive to the instruments used and
suffer from a lack of transparency (Clist and Morrissey, 2011). As such methods have
not yet been developed in a PSTR context; we simply lag the aid variable and the
16
variables for institutions, real per capita GDP which are potentially endogenous. Ac-
cording to Clist and Morrissey (2011) and Carter (2013), this is the best way to account
for the expected lagged effect of aid on tax. The reference paper by Gonzalez et al.
(2005) used a similar approach. Moreover, after multiple regressions controlling for
endogeneity, Fouquau et al. (2008), Bereau et al. (2012), and Jude and Levieuge (2013)
conclude that PSTR reduces the problem of endogeneity because it provides a specific
value of the parameter for each level of the threshold variable. On the theoretical side,
Yu (2013) and Yu and Phillips (2014) demonstrate that in threshold regressions models,
both the threshold point and the threshold effect do not need instrumentation to be
identified. In particular, Yu (2013) shows that 2SLS estimators in threshold models
with endogeneity are inconsistent.
4.3 Data
We use a panel data set that covers 6 out of the 8 countries of the WAEMU over the
period 1986-2010 due to data availability (Cote d’Ivoire, Burkina Faso, Mali, Niger,
Senegal, and Togo; excluding Benin and Guinea-Bissau). ICRG data on institutional
quality are not available for Benin. The main reason for removing Guinea-Bissau is data
accuracy and missing data. Apart from the data on quality of institutions, inflation,
and tax revenues, all the other data are taken from the World Development Indicators
of the World Bank.
Tax revenue is the ratio of total tax excluding social contributions to GDP. The
database is from the Tax Policy Division of the Fiscal Affairs Department of the In-
ternational Monetary Fund provided by Keen and Mansour (2010) and Benedek et al.
(2012). It is an extended and improved version of the dataset used in the paper by Keen
and Mansour (2010). Foreign aid is measured as the total net official development as-
sistance as a percent of GDP like in most of the related empirical studies. Even though
we focus on this variable, we also break it down into non-technical grants, technical
grants, and concessional loans to test robustness.
Institutional quality is taken from the International Country Risk Guide (ICRG)
database. It provides information on various risk indicators grouped into 3 major
categories of risk: political, financial, and economic risks. It is compiled by Political
Risk Services (PRSP) Group. The ICRG indicators are widely used in empirical studies
to measure political risk and institutional quality. However, in contrast with studies
which focus on a specific variable, we consider here the composite political risk index,
which is the sum of all the 12 major components in this category ranging from 0
to 100 points. This index contains: government stability, socioeconomic conditions,
17
investment profile, internal conflict, and external conflict, rated from 0 to 12, with
corruption, military in politics, religious tensions, law and order, ethnic tensions, and
democratic accountability, rated from 0 to 6, and bureaucracy quality, rated from 0 to
4. The higher the value of the component, the lower the associated risk perceived. This
variable is lagged 1 period to avoid potential endogeneity bias. Data on inflation are
obtained from the IMF World Economic Outlook Database.
Since the individual dimension of our panel is smaller than the time dimension, we
check for the stationarity of variables. We apply Im-Pesaran-Shin (IPS) and Fisher
tests as our panel is not completely balanced, and because the panel is heterogeneous,
in particular in terms of the stability of government, tax, and aid variables (see main
descriptive statistics in appendix)4. The tests reveal that the alternative hypothesis of
stationarity cannot be rejected for all the variables. Table 10 in the appendix displays
pairwise correlations between the variables. Total aid and its components are negatively
associated with tax revenues, while institutional quality and government stability are
positively correlated with tax. Before running regressions, we check whether there
is a concern of multicollinearity by applying a variance inflation factors test (VIF).
The results indicate that the mean of VIF is not high, with no VIF greater than 10
irrespective of whether we account for aid and government stability interactions or not.
The largest VIF is 5.66. Thus there is no concern for collinearity, because according to
Hamilton (2012) collinearity is a problem only when VIFs are greater than 30.
5 Results
Simple panel interaction regressions
For comparison purposes, we begin by panel regressions including a simple interac-
tive term between government stability and aid indicator following Brun et al. (2011a)
among others. We use the Feasible Generalized Least Squares (FGLS) method with
common AR(1) and panel AR(1) and AREG(1)5 to correct the problem of heteroskedas-
ticity and auto-correlation. Table 3 presents all the corresponding results. When we
consider the overall level of aid, the results indicate that aid exerts a negative effect
on tax revenue but this effect is not statistically significant in all specifications. In
contrast, the coefficient of the interactive term between aid and political stability is
positive and statistically significant at 5% only in FGLS common AR(1) specifications.
4More details on within and between statistics can be provided on request.5AREG was used because some data is missing for foreign debt stock as a percentage of GDP
18
This suggests that aid tends to be effective when government stability improves.
Table 3: Interaction Government stability/aid and tax revenue, panel specifications
Model 1 2 Aid compositionFGLSCOM-MONAR(1)
FGLSPANELAR1
FGLSCOM-MONAR (1)
FGLSPANELAR1
AREG(1) FE
FGLSCOM-MONAR(1)
FGLSPANELAR1
AREG(1)
Aid -0.044 -0.009 -0.044 -0.011 -0.053(0.046) (0.044) (0.046) (0.044) (0.061)
AIC criterion 0.913Schwarz criterion 1.756Number of observations 149Notes: *significant at 10%, **significant at 5%, ***significant at 1%. Standard errors in (.)
The results are less robust than those found for the overall level of aid depending on
the type of aid and regime. However, while statistical significance may be marginally
reduced in these regressions, the results always make economic sense. Technical and
non-technical grants are statistically significant only in the regime 3. In regime 3 while
Reforms For. debt Pop growthReforms 1.000For. debt -0.009 1.000Pop growth -0.127 0.190 1.000Notes: Stability=Government stability, Institutions=institutional quality, N-tech grants=Non-tech grants,RPCGDP=Real per capita GDP, For.debt= Foreign debt stock, Pop growth= Population growth.
36
Table 11: Panel Unit root tests
Variables IPS FISHERTax -2.304*** 28.140***Aid -2.434*** 35.775***Non-tech grants -4.696*** 39.437***Tech grants -2.297** -1.843**Loans -4.441*** 44.883***Gov stability -1.985*** 28.760***Institutional quality -1.312* 30.191***Agriculture -2.097** 33.655***Industry -2.627*** 28.751**Real per capita GDP (log) -2.402*** 38.445***Trade openness -2.711*** 28.145***Exports -2.396** 30.571***Imports -3.560*** 24.628**Foreign debt stock -1.460* 23.184**Population growth -6.157*** 44.270***Notes:*,**,***significant at 10%, 5% and 1%. Robust Standards errors in(.).t-bar are reported for IPS assuming that N and time are finite. Fisher testscorrespond to ADF tests.The reported statistics are those of the inverse chi-squared except for Non-tech grants whose statistic is the inverse normal statistic.