Econometric analysis of VAT GAP determinants

AuthorAdam Smietanka - Mikhail Bonch-Osmolovskiy - Grzegorz Poniatowski
VAT Gap in the EU-28 Member States
5. Econometric Analysis of VAT Gap Determinants
a. Introduction
The examination of tax non-compliance determinants is not new to the economic literature.
Most of the literature dealing with such factors focuses on personal income taxes, voluntary
tax compliance, and deterrence effects. This focus is clearly related to data availability. The
empirical studies are based mostly on micro-data gathered in surveys and audit statistics.
Thus, they concentrate on the impact of individuals characteristics (see e.g. Feinstein
[1991]). Similarly, studies scrutinising the determinants of compliance in corporate and
consumption taxation usually look at micro-level revenue figures from fiscal registers or
audit data (see e.g. Casey and Castro [2015]). The studies based on fiscal registers and
audit and survey data face an important limitation, i.e. the inability to observe the variability
of determinants across tax systems and economies. A rather limited number of studies
looking at such cross-country variations focus on the variation of dynamics in tax revenue
(see e.g. Aizenman and Jinjarak, [2018]) or have a qualitative nature (see e.g. Keen and
Smith [2007]).
The European Commission’s VAT Gap Study made available a large set of standardised
data on tax compliance from a group of countries with varying economic and institutional
characteristics. The series are available across a time period long enough to cover
economic upturns and downturns. As a result, the Study provides an opportunity to conduct
econometric analyses looking at the determinants of tax non-compliance from a new
perspective. The panel data derived from the VAT Gap Study have already been used by a
number of researchers such as Barbone et al. (2013), Zídková (2017), Lešnik et al. (2018),
Poniatowski et al. (2018 and 2019), Szczypiska (2019), and Carfora et al. (2020).
The econometric analysis outlined in this Study extends the above-mentioned studies
several-fold. Concerning the data preparation procedure, we eliminate potential bias in the
data by correcting the VAT Gap series for each country for revisions in subsequent vintages
of the Study. Moreover, we account for measurement errors, i.e. changes in the VAT Gap
not related to change in compliance but rather to specific one-off factors. To deal with the
scarcity of observations of exogenous variables, we perform a dummy variable adjustment.
Although this operation rises the number of explanatory variables, overall it increases the
degrees of freedom due to higher number of observations included in the estimation. In
regard to the specification of the models, we extend the list of covariates relating to tax
policy characteristics, macroeconomic variables, variables describing the structure of the
economy, and proxies of tax fraud.
b. Data and Variables
Our endogenous variable is the VAT Gap of country i in year t taken from each of the
European Commission’s VAT Gap Studies (i.e. the 2013, 2014, 2015, 2016, 2017, 2018,
and 2019 Studies). To ensure the comparability of vintages across time, the data was
transformed using the methodology described in the following section.
VAT Gap in the EU-28 Member States
page 56 of 99
The wide set of covariates included in the analysis originates from the 2019 Study but
includes around 16 new variables8. The covariates could be grouped as those describing
tax policies, indicators of the macroeconomic situation, variables describing the exogenous
factors to the tax administration economic characteristics of a country, and proxies of VAT
The inclusion of tax policy characteristics is expected to show how the various efforts of
tax administrations relate to the VAT Gap in each country. It could be expected that the
greater the efforts of the administration are, the higher the level of tax compliance, both
voluntary and involuntary. Expenditure on tax administration in relation to GDP alone might
not be enough to capture how effectively the funds are used the “IT expenditure” variable
is expected to pick up the effect of innovative processes introduced into administrative
processes. Similarly, the Administrative effectiveness variable, meaning the
independence of the tax administration from political pressures as well as the quality of
policy formulation and implementation, should account for general proficiency in collecting
taxes and the credibility of government.
The set of macroeconomic variables aims to explain the cyclical conditions that affect
taxpayer behaviour. For example, the Unemployment variable should be able to capture
situations when taxpayers face stronger incentives to evade tax liabilities due to the
increased number of bankruptcies and liquidity constraints. Similarly, GDP per capita is
expected to capture periods of economic stress as well as decreasing with wealth incentives
not to comply. We also expect that the level of government debt could complement the list
of core determinants by accounting for the economic constraints and prudence of public
We suspect that certain economic characteristics which show large variation across
countries and rather low variation in time are also related to VAT compliance. Thus, we
include variables describing the sectoral and company structure of the economy. In
particular, we distinguish the retail sector, which could be the key sector, along with other
labour-intensive sectors, as well as real estate, construction, industry, telecommunications,
and art. The model also takes into consideration the structure of companies by size of
employment and the relative size of the shadow economy. One of the newly introduced
variables is the value of credit transfer payments involving non-MFIs this variable should
help to explain how advanced the financial system is in terms of cashless transactions,
which are more secure and easier to control by the tax administration.
Since the variability of tax fraud, a significant component of the VAT Gap, may be related
to very specific factors not included in the covariates list, we proxy the scale of fraud using
8 See Table 5.1, EC (2019).

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT