La mesure de la performance dans les administrations fiscales et douanières des pays en développement

FERDI, ICTD, CERDIJeudi et vendredi 12 - 13 juin 2014, Clermont-Ferrand, France.

The power to tax in Africa: VAT, LTU, and SARA

Christian Ebeke, Mario Mansour and Gregoire Rota Graziosi

International Monetary FundCERDI and FERDI.

This Working Paper should not be reported as representing the views of the IMF.The views expressed in this Working Paper are those of the author(s) and do not necessarily represent those of the IMF or IMF policy. Working Papers describe research in progress by the author(s) and are published to elicit comments and to further debate.

Introduction• The power to tax is a core function of the state.

– Preponderant role of taxation in the relationship between rulers and citizens (Ardant, 1971, Friedman, 1974, and Tilly, 1990).

– State building => governments’ ability to raise taxes (Brautigan et al., 2008; and Besley and Persson, 2011).

• Three major reforms in tax policy and tax administration in Sub-Saharan Africa (SSA): – Large Taxpayer Units (LTUs): a rationalization of the tax administration through

the segmentation of taxpayers according to their turnover; – Value Added Tax (VAT): a partial delegation of revenue collection to the private

sector (self-enforcing mechanism though the credit-invoice method); – and Semi-Autonomous Revenue Authorities (SARAs): a strategic delegation of

tax collection to an agent.• A common denominator of these reforms: increase revenue through

better enforcement of tax laws.

Introduction• Our analysis: estimating the effectiveness of these three tax reforms on

domestic revenue mobilization. • Our database: tax revenue from non-renewable natural resources to GDP

ratio for 41 SSA countries over 1980-2010 (Mansour, 2014).• Our empirical analysis: a wide range of complementing empirical

approaches to address the issue of self-selection associated with the decision to adopt each reform.– Fixed-effects estimates– Dynamic panel estimates with fixed-effects. – the propensity score matching,– and the synthetic control method.

• Our results: – VAT and SARAs improve significantly tax revenue mobilization; – the impact of LTU is debatable.

LTU: strengths and weaknesses– Initially in Latin America in the late 70s,– Establishing a self-contained office within the tax administration, acting as single

clearance window for large taxpayers for the main domestic taxes: income taxes, VAT…

• Strengths:– Segmentation => the specialization of the personnel of the tax administration,– Centralization of core functions for all taxes,– An opportunity to induce other efficiency-improving reforms such as self-assessment,

adoption of unique taxpayer identification numbers, electronic filling, and new computerized information system (Baer, 2002; and McCarten, 2005)

• Weaknesses:– Concentration of the tax administration efforts and most competent human resources

to a small number of large taxpayers => the emergence of a dual economy– Risk of unbalanced resource allocation between tax collection from existing well

known taxpayers, and extending the tax base to include tax evaders (Terkper, 2003).– Strategic behavior from taxpayers, who may break-up their activities in order to

remain below the LTU threshold= > a missing middle.

VAT: strengths and weaknesses• Introduced in France (1954): a tax on final consumption. • Two properties:

– Neutral on production decisions. – Self-enforcing: collected by firms at the various stages of production and

distribution through the invoice-credit mechanism, which provides an incentive to report purchases (input) in order to claim a credit against VAT on sales (output).

• Strengths:– An incentive to formalize economic activity since the informal sector would be considered as

final consumer and would support all the VAT paid on its inputs; – Devolution of a part of the tax collection effort to the private sector,.

• Weaknesses– Poorly implemented in Africa:

• High VAT rates, which created pressure for exemptions• Inadequate registration threshold

– Exposed to lobby groups for tax expenditure every year.

SARA: strengths and weaknesses– A drastic reform: delegating tax collection to an autonomous agency (Kydland

and Prescott, 1994)• Strengths:

– Autonomy (which may differ significantly across countries) = a signal to a more credible audit policy without any political interference (Taliercio, 2004).

– Signal effect reinforced by the greater flexibility of SARA to manage its human resources than standard public sector agencies (Fjeldstadt and Moore, 2009 and Moore, 2014).

– Merger of tax and customs administrations into a single entity.• Weaknesses

– Credibility of the commitment non sufficient (< an independent Central Bank).– A threat on the consistency of policymaking due to the separation between tax

collection and tax policy (Fjeldstad and Moore, 2009).– The mobility of tax auditors from SARA to the private sector => conflicts of

interests, revolving doors.

Data and stylized facts (Average tax revenue to GDP in SSA over 1980-2010)

12%

14%

16%

18%

20%

1980 1990 2000 2010Year

Non natural resource tax revenue / GDP Fitted valuesTax revenue / GDP Fitted values

The adoption of LTU, VAT and SARA in SSA (1980-2010)

0

5

10

15

20

25

30

35

40

1980 1985 1990 1995 2000 2005 2010

VAT

LTU

SARA

Reform sequencing0

10

20

30

40

Num

ber o

f occ

uren

ces

0 1 2 3 4Order

No reform

SARA

LTU

VAT

LTU and SARA same year

LTU and VAT same year

Source: Authors.

Empirical analysis• Several techniques:

– Fixed-effects estimates and dynamic panel with fixed-effects, – Propensity score matching estimates,– Synthetic control method which is a generalization of the difference-in-difference

technique.• Why?

– Due to the issue of the self-selection associated with the decisions to adopt each reform.

– No random assignment of the allocation of LTUs, VAT and SARAs between countries.– Potentially large selection-bias associated with the decision to adopt (and maintain)

a tax reform. For example, countries with poor performance in revenue mobilization are more likely to embark into ambitious reforms aimed at reversing the trend.

• Expectation: – A positive impact on revenues of each reform; – The impact of any combination of these reforms.

Fixed-effects estimates• The baseline specification:

– A linear relationship between the tax revenue-to-GDP ratio and each dummy variable representing each of the three tax reforms. The dummies take the value 1 whenever a country has in place a VAT, LTU, or a SARA, and 0 otherwise.

– where T and R denote the tax revenue-to-GDP ratio and the dummy variable identifying the presence of each of the three tax reforms considered.. X refers to the matrix of control variable discussed above while ui denotes the country fixed-effects.

Fixed effects estimation resultsDependent variable: -1 -2 -3 -4 -5Non-hydrocarbon tax revenues-to-GDP

FE FE FE FE FE

VAT 1.015*** 0.717** 1.489***-0.246 -0.311 -0.329

LTU 0.644** -0.425 0.204-0.25 -0.323 -0.321

SARA 2.158*** 2.016*** 1.509***-0.352 -0.399 -0.415

Fund arrangement dummy, lagged

0.656***

-0.253Trade openness 0.0193** 0.0179* 0.0169* 0.0180* -0.00361

-0.00923 -0.00941 -0.00916 -0.00927 -0.00996Resource rents-to-GDP -2.394*** -2.399*** -2.227*** -2.234*** -1.968***

-0.257 -0.258 -0.256 -0.256 -0.266Real GDP per capita, log 1.378** 1.781*** 1.484*** 1.218** -0.372

-0.553 -0.55 -0.543 -0.556 -0.671Agriculture value added-to-GDP -0.0398** -0.0434** -0.0313 -0.0297 -0.0243

-0.02 -0.0202 -0.0199 -0.02 -0.0232Aid-to-GDP 0.00565 0.00867 0.00277 0.000966 0.0127

-0.0138 -0.0139 -0.0137 -0.0138 -0.0159Constant 9.108** 6.996* 8.136** 9.563** 18.59***

-3.894 -3.902 -3.84 -3.892 -4.69

Country fixed-effects Yes Yes Yes Yes YesObservations 1,080 1,080 1,080 1,080 821R-squared 0.11 0.101 0.127 0.132 0.163Number of countries 39 39 39 39 39Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Table 1. Fixed effects panel data estimates of the effects of Tax reforms in Africa.

Propensity score matching estimates

• A two-step approach– First: the probability of observing either the VAT, LTU or SARA in a

given country at year t, is estimated conditional on observable economic conditions and country characteristics (selection model)

– Second: these probabilities, or propensity scores, are used to match reform countries to non-reformer countries, and thereby construct a statistical control group.

• Test the sequencing of the reforms: for each reform, we investigate whether the initial adoption of a given tax reform determines the probability of adopting a subsequent one.

-1 -2 -3 -4Nearest-neighbor matching

Three nearest-neighbor matching

Radius Kernel

ATT 1.855*** 2.043*** 1.814*** 1.813***-0.629 -0.548 -0.36 -0.36

No. of treated 364 364 364 364No. of controls 475 475 475 475Observations 839 839 839 839

ATT 0.353 0.758 0.517 0.545-0.729 -0.64 -0.408 -0.407

No. of treated 268 268 268 268No. of controls 603 603 603 603Observations 871 871 871 871

ATT 4.529*** 3.585*** 2.634*** 2.805***-0.807 -0.659 -0.535 -0.531

No. of treated 169 169 169 169No. of controls 735 735 735 735Observations 904 904 904 904

Table 2. Matching estimates of treatment effect on the non-natural resource taxrevenue-to-GDP ratio.

Panel A. Treatment effect of VAT adoption on non-hydrocarbon tax revenues-to-GDP ratio

Panel B. Treatment effect of LTU adoption on non-hydrocarbon tax revenues-to-GDP ratio

Panel C. Treatment effect of SARA adoption on non-hydrocarbon tax revenues-to-GDP ratio

Notes: ATT: Average treatment effect on the treated. Observations are matched on the‘common support’ An Epanechnikov kernel is used for kernel matching. Bootstrapped standarderrors are reported in parentheses. They are based on 100 replications of the data. *** p<0.01,** p<0.05, * p<0.1.

-1 -2 -3 -4Nearest-neighbor matching

Three nearest-neighbor matching

Radius Kernel

ATT -1.912** -1.533** -1.645*** -1.616***-0.831 -0.737 -0.465 -0.469

No. of treated 159 159 159 159No. of controls 745 745 745 745Observations 904 904 904 904

ATT 5.702 4.608 6.410*** 6.538***-3.499 -3.003 -2.243 -2.209

No. of treated 19 19 19 19No. of controls 881 881 881 881Observations 900 900 900 900

ATT 1.434 1.154 -0.88 -0.934-2.2 -1.914 -1.312 -1.317

No. of treated 6 6 6 6No. of controls 898 898 898 898Observations 904 904 904 904

ATT 0.329 1.202* 2.394*** 2.403***-0.927 -0.711 -0.412 -0.41

No. of treated 94 94 94 94No. of controls 797 797 797 797Observations 891 891 891 891

Panel A. Treatment effect of VAT and LTU adoptions

Panel B. Treatment effect of VAT and SARA adoptions

Panel C. Treatment effect of LTU and SARA adoptions

Panel D. Treatment effect of all reform adoptions

Notes: ATT: Average treatment effect on the treated. Observations are matched on the ‘commonsupport’ An Epanechnikov kernel is used for kernel matching. Bootstrapped standard errors arereported in parentheses. They are based on 100 replications of the data. *** p<0.01, ** p<0.05, *p<0.1.

Synthetic control method (SCM)• Approach developed by Abadie and Gardeazabal (2003) and

extended in Abadie, Diamond, and Hainmueller (2010).• Question: Does the adoption of the VAT, the LTU or the SARA in

year T lead to higher tax-to-GDP ratio in the years T+i (i=1,…,10) compared to similar countries that have not implemented such reforms?

• Advantages of SCM:– A transparent estimation of the counterfactual outcome of the

treated country —the synthetic control. – SCM can deal with endogeneity from omitted variable bias by

accounting for the presence of time-varying unobservable confounders.

Effect of the adoption of LTU11

1213

1415

16N

on-h

ydro

carb

on ta

x re

venu

e-to

-GD

P

-10 -5 0 5 10Time

Actual Counterfactual

Note: Average taken across countries without missing data.

0.1

.2.3

.4.5

.6.7

.8.9

1S

tatis

tical

sig

nific

ance

(Pro

babi

lity)

0 1 2 3 4 5 6 7 8 9 10

Lead specific significance levels (P-values)

Time

Effect of the adoption of VAT12

1314

1516

17N

on-h

ydro

carb

on re

venu

e-to

-GD

P

-10 -5 0 5 10time

Actual Counterfactual

Note: Average taken across countries without missing data

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7S

tatis

tical

sig

nific

ance

(Pro

babi

lity)

0 1 2 3 4 5 6 7 8 9 10

Lead Specific Significance Level (P-Values)

Time horizon

Effect of the adoption of SARA10

1214

16N

on-h

ydro

carb

on re

venu

e-to

-GD

P

-10 -5 0 5 10Time

Actual Counterfactual

Note: Average taken across countries without missing data.

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

8S

tatis

tical

sig

nific

ance

(P-v

alue

)

0 1 2 3 4 5 6 7

Lead specific significance levels (P-values)

Time

Conclusion• Both VAT and SARA have significant positive effects on tax revenue mobilization. • The LTU reform is not found to exert any particularly significant effect on

revenues.– Tax administrations focused on large taxpayers even before LTUs were implemented; – LTUs centralized compliance services to large firms, which may have had an impact on

their compliance costs but limited impact on revenues. – Another factor is the use of information in risk management and audit, areas where

LTUs bring very little. – The lack of impact of the LTU may be due to missing accompanying measures, such as

better access to information, both internal and external to government.• Practical policy implications of our results:

– Fundamental tax reform such as VAT and SARA has a good chance of mobilizing addition revenues in SSA

– Marginal reforms, such as introduction of LTU, especially if unaccompanied by other reforms, have little impact (if any) on revenue.

Caveats• First, the benefits of tax reforms should be weighed against

their costs, which differ across the various reforms– LTU is certainly the cheaper reform. – VAT involves a significant compliance cost for the private sector.– SARAs in SSA countries have been largely supported by donors and

their operational costs are estimated to two percent of tax revenue in the Eastern African Community.

• Second, our analysis covers non-resource revenue only, but as we noted earlier, much of the increase in revenue in SSA since the mid-1990s came from resource revenues.

• Third, a number of SSA countries recently revised their GDP figures upward, and by a significant margin.

Thank you for your attention

The issue of statistical interferences• SUTVA:

– All previous estimation methods are based on the assumption of no interferences between units (countries) = the stable-unit treatment value assumption (SUTVA).

• Potential positive tax spillovers effects among neighboring countries => A potentially downward bias. – The absence of interference among countries = the absence of any tax competition

and more broadly any tax (policies and reforms) spillovers between countries. – If one of the studied tax reform is effective in improving revenue mobilization

Þ increase in the effective tax rate, whatever is the tax base: income or consumption.Þ some tax base effect in neighboring countries: the increase in the effective tax rate in one country after

a tax reform may induce an outflow of mobile factors (capital, skilled workers…) or some cross-border shopping, which means a broadening of the tax base in neighboring countries

Þ tax revenue increase in the synthetic group.Þ A downward bias.