The Elasticity of Taxable Income: The Case of South Africa

• DOI: http://doi.org/10.1111/saje.12232
• Author: Johannes Hermanus Kemp
• Date: 01 December 2019
• Published date: 01 December 2019

© 2019 Economic Society of Sout h Africa

South African Journal of Economics Vol. 87:4 December 2019

doi : 10.1111/ saje .122 32

417

THE ELASTICITY OF TAXABLE INCOME: THE CASE OF

SOUTH AFRICA

JOHANN ES HERMA NUS KEMP*

Abstract

A key tax policy parameter that has received much attention in the international literature, but

about which there is substantial uncertainty, is the overall elasticity of taxable income. The size of

this parameter is central to the formulation of tax and transfer policy, as well as for the study of

the welfare implications of tax decisions. This paper uses a panel of individual tax returns for the

period 2009–2013 and the phenomenon of “bracket creep” to construct instrumental variable

estimates of the sensitivity of income to changes in tax rates. Estimates suggest that the overall

elasticity of taxable income is approximately 0.3, while that of broad income is significantly lower.

The overall response is primarily driven by the elastic response of taxable income for high-income

earners, who have an elasticity of closer to 0.4. Using the elasticity estimates within an optimal tax

framework, it is determined that the optimal marginal tax rate for the top 10% of income earners

is broadly in line with the current income tax schedule. However, results also suggest that there is

little scope for raising marginal rates on high-income earners further without inducing a negative

revenue response.

JEL Classification: H21, H31, J22

Keywords: Fiscal polic y, elasticity, taxable income, optimal tax pol icy

1. INTRODUCTION

The behavioural response of ta xpayers to changes in (marginal) tax rates ha s long been of

interest to economists. The size of this para meter is central to the formulation of tax and

transfer policy, as well as for the study of the welfa re implications of tax decisions. Public

finance practitioners are often asked to predict responses to alternative policy changes

and to provide empirical estimates of the effec ts to decision-makers (Thoresen a nd Vattø,

2013). To this end, it is important to quantify the behavioural response of taxpayers to

changing policy parameters.

In this regard, the ela sticity of taxable income, or ETI, is a key concept. The ETI aims

to capture all possible behavioural responses to changes in income taxation, including

tax evasion and/or avoidance, in a single measu re, without the need to specify the natu re

of the specific adjustment processes involved (Creedy, 2009; Thoresen and Vattø, 2013).1

This allows policymakers to evaluate the likely impact of policy proposals on both

1 These adjustments could include la bour supply changes (at both the extensive and/or intensive

margins), income shifting between sources which are taxed at different rates and ta x evasion

through non-declar ation of income.

* Corresponding author: Bureau f or Economic Resea rch, Stellenbosch University, Stellenbosch,

South Africa. E-ma il: jhkemp@sun.ac.za

South African Journal

of Economics

418 South African Journal of Economics Vol. 87:4 December 2019

© 2019 Economic Society of Sout h Africa

taxpayer behaviour and policy outcomes (such as the impact on tax revenue of a change

in marginal tax rates).

Modelling the response to tax rates in this reduced form way, as opposed to formu-

lating a structura l model of the behaviour involved, has proven to be very attractive. As

mentioned in Saez et al. (2012), a large body of literature has sought to estimate the ETI

and elasticities for related income measures. Most of this literature has focused on indi-

vidual taxable income in the United States, Ca nada and Western Europe, while a limited

number of studies have focused on countries outside these regions.

Very little work has been done in the South African context. Given the lack of access

to micro-level panel data on personal income taxes (PIT ), studies that have attempted to

estimate the ETI for South Africa focus primarily on publicly available aggregated tax

data. Using a new dataset comprising conf idential tax return data made available for

research purposes by the South A frican Revenue Service (SARS) and the National

Treasury, this paper attempts to estimate the ETI for South Africa. Given the impor-

tance of PIT in overall tax collection in South Africa (personal income tax is the single

largest contributor to tax revenue, accounting for 35% of total gross tax collected be-

tween 2000 and 2018), the study is restricted to estimating the ETI as it applies to per-

sonal income taxes and individual marginal tax rates.2

Much of the recent literature uses tax reforms to identify t he parameters of interest when

studying the sensitivity of income with respect to marginal tax rates. Notwithstanding

the fact that the ETI va ries between countries depending on the nature of the tax s ystem,

the empirical literature has failed to generate a consensus of the size of the elasticity –

even within specific countries.

According to Saez (2003), two reasons might explain this lack of consensus. First,

most tax reforms introduce numerous changes to the definition of taxable income be-

sides tax rate changes. This makes it difficult to compare reported income before and

after a tax reform. Second, most studies rely on a comparison between high- and low- or

middle-income taxpayers. The former often experience large tax rate changes after any

given tax reform, while the latter experience little or no change in marginal tax rates.

This suggests t hat the research design for estimating behavioura l responses to changes

in marginal tax rates should meet two conditions. First, the tax change should affect

only marginal ta x rates without introducing changes in tax rules. Sec ond, the tax cha nge

should affect dif ferently groups of taxpayers that a re comparable in terms of incomes and

other economic characteristics (Saez, 2003).

Taking these factors into account and considering the fact that there was no large

legislated tax reforms in South Africa over our sample period, we follow Saez (2003) in

utilising the phenomenon of “bracket creep” to estimate the behavioural response of ta x-

payers with respect to changing marginal tax rates. Because of inf lation, a taxpayer near

the top end of an income tax bracket might ’creep’ to the next bracket even if their real

income does not change. This is especially true if nominal income tax brackets are not

2 While the la rger share of the international literature ha s also focused on personal income ta xes

(PIT) and the response of individuals to changes in marginal tax rates, most of the conceptual

and empirical issue s discussed remain releva nt for other tax bases, such as corporate i ncome taxes

(CIT ).

419South African Journal of Economics Vol. 87:4 December 2019

© 2019 Economic Society of Sout h Africa

fully adjusted for inflation (so-called fiscal drag). In contrast, taxpayers at the bottom

end of the relevant tax bracket are unlikely to experience an increase in marginal rates.

In practice, the method compares changes in the incomes of taxpayers near the top

end of a bracket to those taxpayers at the bottom end of the bracket using tax return

data. Importantly, “bracket creep” does not affect the definition of reported income.

Therefore, incomes can be easily compared across years. Additionally, this strategy com-

pares groups of taxpayers that are closely related in terms of income levels and economic

characteristics.

Using this approach, the elasticity for ta xable income is estimated at around 0.3, while

that for broad income is estimated at closer to 0.2. Additionally, it was found that be-

havioural responses are concentrated in higher income groups as suggested by the higher

elasticity estimate (0.37 for taxable income) for the top 10% of income earners.

Two limitations of the approach should be mentioned. First, changes in tax rates due

to “bracket creep” are relatively small compared to other ta x reforms across the decades.

Second, because “bracket creep” is not a legislated tax change, it might be harder for

taxpayers to understand the ef fect of this change on their after-tax income, and therefore

the behavioural response to “bracket cre ep” might be muted. This sugges ts that ETI esti-

mates obtained using “bracket creep” ref lect a lower bound for the behavioural response

to changing tax rates.

An important application of ETI estimates is in the determination of optimal mar-

ginal tax rates. Most studies on optimal tax ation focus on the top 1% of income earners.

However, due to data limitations, this paper focuses on the top 10% of income earners.

Embedding the elasticity estimate of 0.37 for the top 10% of income earners in an opti-

mal tax fra mework, it is determined that the optimal marginal tax rate for this group is

around 48% (consistent with a legislated marginal PIT rate of around 40%).

The rest of the paper proceeds as follow. Section 2 discusses the basic conceptual

framework and unpacks the various identification issues that arise when attempting to

estimate the ETI in practice. Section 2 also provides a short review of the international

empirical literature. Section 3 estimates the ETI for South Africa using micro-level tax

return data and Section 4 discusses optimal ta xation results. Section 5 presents some

caveats a nd concludes.

2. CONCEPTUAL FR AMEWORK AND LI TERATURE R EVIEW

In the basic ETI model3, individuals maximise a utility function u(c, l), where c is the

consumption (equal to disposable income in a one-period model) and l is the labour sup-

ply (measured by hours worked). Earnings are given by w·l, where w is the exogenous

wage rate. The (linearised) budget constraint is c = w·l·(1−τ) + R, where τ is the marginal

tax rate and R is the virtual income (i.e. non-wage income equal to the intercept of the

linear section of the budget constraint extended to the c axis, where l = 0).

The ETI literature generalises this model by noting that hours worked is just one of

the possible behavioural responses to income taxation. As shown in Feldstein (1999), a

simple way to jointly model these responses is to assume that utility depends positively

on disposable income (equal to consumption) c and negatively on reported or taxable

3 Following Saez (2001), Gruber and Saez (2002) and Saez et al. (2012)