Page 5 - Working Paper (The Myths and Realities of Tax Performance Under Semi-Autonomous Revenue Authorities)
P. 5
DDTC Working Paper 0213
5
3.1. Data second phase, we use a panel regression using
pooled OLS, fixed effect and random effect. All
In this study, we will use panel data (cross- three approaches are used to find the relationship
countries and over time). The panel data covers between tax institutions and all ex-ante related
49 countries, over the period of 2000-2011 (12 factors.
years). We choose the analysis period based on
several considerations. At the end of the 1980s For multivariate model where there are at least
to the 1990s, many countries were changing the 2 variables, we initially conduct diagnostic test to
institutional model of the tax authority into SARA. describe data normality, check multicollinearity,
As assumed in the post-transition from a model of heteroscedasticity, and autocorrleation test in
non-SARA, SARA in some countries becomes stable order to select best linear unbiased estimators. We
so that more reliable for comparison. Therefore, use Breusch-Pagan and White-Hetero tests to detect
it would be valid if the period of analysis begins heteroscedasticity problems. Furthermore, we also
in 2000, 49 countries that are chosen represent did error plotting to identify autocorrelation. After
countries that apply SARA (28 countries) and conducting diagnostic test, we come up with model
non-SARA (21 countries), regional coverage (Asia, specification below:
Pacific, Africa, America, and Europe), income level
(low-middle income groups, medium-high, and Tax ratio it
income high) classification according to United
Nations or the World Bank classifications. = α + β instmodel + β independent variables + ε it
1it
i
it
2it
it
The data used for this analysis is extrated
from various reliable sources. For economic We choose tax ratio as dependent variable;
and social indicators, data was extracted from whereas, institutional model and other control
Government Finance Statistics (IMF) and the variables as independent variable. We, then, apply
World Development Indicators (World Bank). multivariate analysis to find robust results, which
The institutional variables extracted from OECD are: pooled ordinary least square (pooled OLS) and
(2013), ADB (2012), Mann (2004), and Taliercio linear panels consist of fixed and random effect.
(2004). For the data covering taxation conditions
were taken from the World Bank Enterprise Please note, regression with panel data contains
Survey, while the governance indicators were possibility that the unobservable variable (u_it)
taken from the Worldwide Governance Indicators will be correlated with the explanatory variables
(World Bank). In addition, there are several other (independent). If there is a conviction that
data taken from a variety of sources, e.g. tax morale unobservable variables are not correlated with the
and shadow economy. explanatory variables, it is better to use random
effect (RE). Whereas, if there is a possibility that
3.2. Estimation Strategy there is correlation among the two variables, it
would be better to use fixed effect (FE).
The analysis will be divided into 3 stages. The
first phase provides some descriptive statistics of FE estimator is unbiased under the assumption
the characteristics of taxation in the non-SARA and of strict exogeneity for explanatory variable. FE
SARA countries, for example: tax compliance, tax allows to have correlation between unobservable
bribery indications, to the structure of tax revenues. variable and independent variable in each time
We will testing the so-called myths concerning that period. However, the FE approach has implications
tax conditions under SARA were much better. for the possibility of a variable that is not
constant over time, such as gender or distance.
The second stage, or the core of the research,
Generally, choices over FE or RE will be decided
will try to measure the magnitude of the impact
through several methods, such as hausman test.
of institutional tax authorities to tax revenues by
Nevertheles, we will not do hausman test for this
multivariate model. It is true that, in the literature,
analysis. 16
SARA is not necessarily statistically linked to
tax revenues or the ratio of tax revenue to Gross Lastly, we will also show factors that influence
Domestic Product (GDP), but more on the efficiency the decision to adopt the SARA system using binary
and effectiveness of the agency in collecting taxes. choice model. In this model, the dependent variable
From this perspective, aspects of efficiency and is in the form of binary variables, which in this case
effectiveness will indirectly increase tax revenue. is the institutional model of tax administration, i.e.
At this stage, the tax revenue is assumed to be SARA (1) and non-SARA (0). We utilize a panel logit
influenced not only by factors of institutional
models, but also from an economic level, economic
16 For more detail explanations, please refer to Jeffry M. Wooldridge,
structure, demographic and social factors. In the Econometric Analysis of Cross Section and Panel Data. (Massachusetts:
The MIT Press, 2002).