TALLINN UNIVERSITY OF TECHNOLOGY School of Business and Governance
Alice Mikk FIRM-LEVEL PREDICTORS OF LABOUR TAX EVASION Master’s thesis Program Economic Analysis
Supervisor: Karsten Staehr, PhD
Tallinn 2024
I hereby declare that I have compiled the thesis independently
and all works, important standpoints and data by other authors
have been properly referenced and the same paper
has not been previously presented for grading.
The document length is 14 889 words from the introduction to the end of conclusion.
Alice Mikk 07.05.2024
3
TABLE OF CONTENTS ABSTRACT .................................................................................................................................... 4 INTRODUCTION ........................................................................................................................... 5
- THEORETICAL AND EMPIRICAL BACKGROUND ............................................................ 7
1.1. Institutional background ....................................................................................................... 7
1.2. Shadow economy and tax evasion ...................................................................................... 10
1.3. Dynamics of tax evasion .................................................................................................... 14
1.4. Detecting tax evasion by firms ........................................................................................... 16 - DATA AND METHODOLOGY .............................................................................................. 19
2.1. Data ..................................................................................................................................... 19
2.1.1. Wage and population data ........................................................................................... 20
2.1.2. Annual report data ....................................................................................................... 24
2.2. Methodology ....................................................................................................................... 27
2.2.1. Obtaining subsets of tax evading and tax compliant firms .......................................... 28
2.2.2. Firm-level predictors of tax evasion ............................................................................ 30 - EMPIRICAL ANALYSIS ......................................................................................................... 34
3.1. Main results ........................................................................................................................ 34
3.1.1. Wage regression .......................................................................................................... 34
3.1.2. Logistic regression ....................................................................................................... 37
3.2. Robustness checks .............................................................................................................. 42
3.2.1. Wage regression .......................................................................................................... 43
3.2.2. Logistic regression ....................................................................................................... 44
3.3. Discussion ........................................................................................................................... 45
CONCLUSION ............................................................................................................................. 47
KOKKUVÕTE .............................................................................................................................. 49
LIST OF REFERENCES .............................................................................................................. 51
APPENDICES ............................................................................................................................... 55
Appendix 1. Binary and categorical variables in wage regression ............................................ 55
Appendix 2. List of variables in wage regression ..................................................................... 56
Appendix 3. List of variables in logistic regression .................................................................. 58
Appendix 4. Robustness checks for wage regression ................................................................ 59
Appendix 5. Non-exclusive licence ........................................................................................... 60
4
ABSTRACT The objective of this master’s study is to analyse the relationship between different firm-level financial and non-financial indicators and the prob ability of a firm being engaged in labour tax evasion. Numerous studies have investigated tax eva sion and the shadow economy in Estonia, however, little attention has been paid to the conn ection between labour tax evasion and patterns in financial reports. Considering that labour tax e vasion is a difficult challenge for governments due to which a large share of revenue is lost, it is important to investigate it further.
The thesis aims to answer which firm-level predicto rs contribute to the probability of being engaged in labour tax evasion, as well as estimatin g the proportion of labour tax evading firms. The methodological approach uses Mincer wage regressions to find the firms paying “suspiciously low wages” to employees and logistic regression to analyse the relationship between different firm- level indicators and the probability of the firm be ing a tax evader. Administrative matched employer-employee wage data, population data and firms’ annual reports are used for the purpose of this thesis.
The results show that smaller companies (in terms o f number of employees) are more likely to evade labour taxes. What is more, labour tax evasion is more prevalent in the construction sector. The results regarding financial ratios reveal that the probability to evade labour taxes decreases as turnover, debt to assets, or cost of goods sold to assets increases. However, the predicted probability to evade labour taxes increases as shor t-term debt to assets or turnover to assets increases. The prediction of out-of-sample probabil ity of being engaged in labour tax evasion suggested that 51% of companies in 2021 and 54% in 2022 are classified as labour tax evading.
Keywords: tax evasion, wage regression, logistic regression. 5
INTRODUCTION Tax evasion is a key problem for many governments a s taxation is one of the primary sources of government revenue. The importance of efficient and fair tax system has been discussed by Schumpeter (1991), Musgrave (1959), Levi (1988), an d Brennan & Buchanan (1980). Any inefficiencies in the design of tax system and the administration, or unfair treatment of taxpayers can result in a decision to shift to shadow economy . Revenue lost due to tax evasion affects the ability of the government to fund essential public services and infrastructure such as education, healthcare, defence and more. This can lead to budget deficits, increased debt or governments may respond by increasing tax rates to compensate the l osses. However, higher tax burden can lead to deadweight losses as higher tax rates change the be haviour of compliant taxpayers who perceive the tax burden distribution as unfair.
Estonian government receives 80-90% of the revenues from taxation, and labour taxes account for approximately 50% of the tax revenues, being theref ore directly affected by compliance or resistance of the individuals and companies to repo rt wages to the tax authorities (Statistics Estonia, table RR057). Therefore, investigating labour tax evasion is crucial for understanding its economic, social, and ethical implications of the i ssue. Uncovering determinants and patterns of labour tax evasion can inform effective policy measures, which may lead to improved government revenue, more accurate fiscal planning and resource allocation as well as ensure a fair tax burden for participants of the economy. In addition, it may aid to design targeted interventions to promote corporate transparency and ethical financial practices.
Numerous studies have investigated the dynamics of tax evasion and also estimated the extent of the shadow economy (Putni ņš & Sauka, 2015 ; Kukk & Staehr, 2014; Sc hn eider, 2016; Tafe nau et al., 2010). What is more, different approaches to detec t financial statement fraud have been provided and tied together with labour tax evasion as it constitutes a form of financial statement manipulation resulting in specific patterns in the balance sheet and profit and loss statement (Cecchini et al., 2010; Hajek & Henriques, 20 17; Gavoille & Zasova, 2023 ; Be nkovskis & 6
Fadejeva, 2022). Applying the novel approach on Est onian data could therefore provide exciting insights into the labour tax evasion in Estonian firms.
The aim of this thesis is to analyse the relationship between different firm-level financial and non- financial indicators and the probability of a firm being engaged in labour tax evasion. Two research questions covered in thesis are as follows:
- Which firm-level predictors contribute to the proba bility of being engaged in labour tax evasion?
- What is the proportion of labour tax evading companies?
To answer the research questions raised, a database comprising three merged datasets is utilized, including administrative data on wages, population as well as annual reports. The analysis employs Mincer wage regression to find the firms paying “su spiciously low wages” to employees and logistic regression to analyse the relationship bet ween different firm-level indicators and the probability of being a tax evader.
The master’s thesis is structured into three main p arts. The first chapter gives an overview of theoretical and empirical background of tax evasion. This chapter explains the institutional context of tax system, its importance and more precisely, t axation in Estonia. What is more, emphasis is placed on explaining the concept of shadow economy and connections with tax evasion as well as the dynamics of tax evasion. Lastly, an overview of tax evasion prevention and detection opportunities is given. The second chapter presents the data and the methodology employed in this thesis to analyse the different company-level indic ators and the relationship with the probability of being a tax evader. The third chapter gives an overview of the findings, providing an explanation of the results as well as discussion on weaknesses of the analysis and suggestions for future research. 7
- THEORETICAL AND EMPIRICAL BACKGROUND The following chapter presents the theoretical and empirical background of taxation, shadow economy, tax evasion and tax fraud detection. Section 1.1. explains the fundamentals of taxation, gives an overview of the Estonian tax system and mo re precisely, labour taxation. Section 1.2. aims to describe the concept of the shadow economy and its connection with tax evasion, focusing on labour tax evasion. Section 1.3 discusses the motivation behind the decision to be compliant or to evade on the firm or individual level. Section 1 .4. explains the potential approaches and tools to be used for identifying tax fraud on company lev el as well as characterizes the firms engaged in fraudulent activities based on previous research in the field. 1.1. Institutional background The tax system plays a relevant role in shaping the economic landscape of a country. The importance of an effective tax system cannot be ove rstated as it serves as a primary source of government revenue funding essential public service s and infrastructure such as education, healthcare, defence and more. In addition to fundin g the operation of public institutions, the tax system is a tool to redistribute wealth. Taxation directly impacts individuals and businesses in their decisions on establishing business, conducting trade, employment, investments and more.
The tax system being one of the cornerstones of all political regimes has already been recognized by Schumpeter (1991), Musgrave (1959), Levi (1988), Brennan & Buchanan (1980) and others. Schumpeter (1991) viewed taxation as a tool that co uld, in addition to generating revenue for the government, either support or hinder economic devel opment and emphasized the importance of the government as a designer of tax policies. Musgr ave (1959) proposed three main functions of taxation to be revenue collection, redistribution and macroeconomic stabilization. He also stressed the importance of equity, efficiency and feasibilit y of the administrative side of policies. Levi (1988) contributed by showing that taxation was not only a means of generating revenue but also a mechanism through which states assert their power, authority and legitimacy over a territory or a population. He also discussed the factors that in fluenced tax compliance and resistance among 8
taxpayers. Brennan & Buchanan (1980) presented taxa tion from a public choice perspective, viewing tax policies as a social contract between citizens and the state as well as elected officials making decisions about taxation as an extension of the hand of the public. Just as Musgrave (1959), Brennan & Buchanan (1980) examined the trade-off be tween equity and efficiency in tax policy as the desire is to raise revenue for public goods and services but keep the burden fair for taxpayers.
A study published by the World Bank found that coun tries with simpler and more efficient tax systems had higher rates of economic growth and businesses were more eager to make investments and create new jobs, compared to countries with more complicated tax systems (Dom et al ., 2022). Kenny & Winer (2006) also discussed that democracy is followed by higher cooperation with tax authorities, thus the government receives higher ta x revenues in case substantial degree of voluntary tax compliance is required (such as self-reporting).
Estonia has ranked first in OECD countries in the International Tax Competitiveness Index Ratings published by Tax Foundation for the last 10 years (Mendgen, 2023). The four main considerations have been 20% tax rate on corporate income applying only to distributed profits, flat 20% tax on individual income, property tax covering only the v alue of the land, rather than the value of property or capital and territorial tax system that exempts foreign profits earned by domestic firms from domestic taxation in full (with few exemptions) (Ibid.).
The Estonian tax system consists of national taxes established by tax laws and local taxes established by the city or municipality council in its administrative territory. State taxes include income tax, social tax, land tax, gambling tax, sales tax, customs duty, excise duty, heavy truck tax and business income tax. Local taxes include advert ising tax, road and street closure tax, motor vehicle tax, livestock tax, entertainment tax and parking fees. (Rahandusministeerium, 2024)
According to Statistics Estonia (Statistikaamet, he reinafter SA) data on quarterly consolidated
revenue and expenditure of general government, the three main contibutors to the revenues of the
Estonian state budget are taxes on production and i mports (mainly Value Added Tax, hereinafter
V AT), social contributions (mainly employers’ actual social contributions) and current taxes on
income, wealth etc. (mainly personal income tax, he reinafter PIT) (Statistics Estonia, table
RR057). Figure 1 presents the distribution of government revenue from abovementioned sources,
illustrating the importance of revenues from different taxes for the last decade.
9
Figure 1. Distribution of government revenue from taxes and other sources Source: Compiled by author based on Table RR057 (Statistics Estonia, table RR057). Approximately 35% of the revenues are collected in the form of taxes on production and imports, social contributions fluctuate between 25-35% and current taxes on income, wealth etc. contribute approximately 20% of the total revenue. Other sources of revenue fluctuate between 10-20%. The distribution of government revenue from different s ources has been rather stable in the last 10 years. In addition to V AT, social contributions and PIT being main sources of income, the latter two are associated with labour taxation. Individual s are obliged to pay PIT, employees’ unemployment insurance premium (and if applicable, insurance premiums of mandatory funded pension) on their income and additionally, their em ployer is obliged to pay social tax and employers’ unemployment insurance premiums on the i ncome from employment. Therefore, government revenue arising from social contributions and current taxes on income, which in 2013- 2022 fluctuated around 50% of total government reve nue, is directly affected by compliance or resistance of the individuals and companies to report wages to the tax authorities.
The design of tax system can therefore significantl y affect the tax revenues of a government as
well as distribution of income, social equity and e conomic efficiency (Arsi
ć et al ., 2015).
Establishing the right balance in taxation is cruci al – understanding and optimizing tax systems,
especially in the context of labour taxes is important for implementing a fair and sustainable path
for generating revenue to fund public services prov ided by government. Any inefficiencies in the
design of tax system and the administration, or unfair treatment of taxpayers can result in a decision
to shift to a shadow economy.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
2013 2014 2015 2016 2017 2018 2019 2020 2021 2022
Other revenues
Social contributions
Current taxes on income, wealth
etc.
Taxes on production and
imports
10
1.2. Shadow economy and tax evasion The shadow economy, also known as the informal, underground, cash or black economy, refers to economic activities that are not regulated, monitor ed, or taxed by the government. It operates outside the official channels of the economy and in cludes various forms of unreported or underreported transactions. The shadow economy is c haracterized by a lack of formal oversight, taxation, and compliance with labor laws. (Lippert & Walker, 1997)
In economics, three characteristics are used to classify the acitivities in shadow economy (Ibid.):
- Are the activities market-based (monetary) or non-market based (non-monetary)?
- Are the activities legal or illegal?
- Are the activities carried out for tax evasion or tax aversion?
Table 1 illustrates non-exhaustive list of differen t activities included in the classifications based
on the three characteristics listed above that indi viduals, households and firms can engage in.
Firstly, the activities are allocated based on the existence of a financial aspect. Drug trafficking,
prostitution and undeclared work expect a transfer of money from the service recipient to the
service provider, which does not occur in barter trade, household work or production of drugs for
personal use. Secondly, the legality of the activities is considered. Drug trafficking and prostitution
or drug handling are considered illegal, working or household work are considered legal activities.
Finally, the intentions behind the legal activities can be either to evade taxes or avoid taxes. Tax
avoidance entails finding loopholes in tax systems such as deductions and exemptions to reduce
the tax obligation and is seen as legitimate act (Alm, 1988). Tax evasion on the other hand involves
purposeful deception of tax authorities to evade tax obligations (Ibid.).
Table 1. Parts of shadow economy
Type of activity Monetary transactions Non -monetary transactions
Illegal activities Drug trafficking, prostitution,
smuggling, scams, etc
Barter trade of illegal goods and
services, production of drugs for
personal use , etc
Tax evasion Tax avoidance Tax evasion Tax avoidance
Legal activities Undeclared
income and work
Fringe benefits Barter trade of
legal goods and
services
Household work
Source: Lippert & Walker (1997); Müürsepp (2015)
11
When comparing monetary and non-monetary transactio ns, it is more difficult to estimate the extent of non-monetary transactions. For example, individuals’ earnings reported to tax authorities in the form of tax returns can be compared to indiv iduals’ answers in the Labour Force Survey (hereinafter LFS) regarding income or the Household Budget Survey (hereinafter HBS) regarding expenditure. Large discrepancies between administra tive data and survey data can indicate unreported income. On the other hand, it is much harder to analyse the prices of the non-monetary transactions taking place in the form of barter trade of goods and services and to estimate the size of the non-market based activities in the shadow economy.
Even though shadow economy can be defined as a part of the total economy that is unobserved due to the households and businesses keeping their activities undetected, it is important to note that there are alternative definitions that may sig nificantly affect the estimates of the size of the shadow economy (Lippert & Walker, 1997). There are also several approaches to measure the extent of the shadow economy in different countries , using either questionnaires, administrative data, surveys or modelling. For example, the size o f shadow economy differs drastically if comparing the figures estimated by SA, Putni ņš & Sauka (2015), Schneider (2016), Tafenau et al. (2010), or Pissarides & Weber (1989). Also, Turu-uu ringute AS (on behalf of Estonian Tax and Customs Board or Maksu- ja Tolliamet, hereinafter E MTA) and Estonian Institute of Economic Research (Eesti Konjunktuuriinstituut, hereinafter EKI) have investigated the extent of shadow economy in Estonia using surveys.
SA assesses the shadow economy following exhaustiveness principle as one of the components of the national accounts when calculating the gross domestic product (GDP). Based on official data, the financial transactions in the shadow economy ar e captured, the main elements of which are unreported employees, illicit trade, envelope salar y and tax fraud (Müürsepp, 2015). However, the size of the shadow economy estimated by SA is r elatively low compared to estimates using other estimation approaches, varying between 3-4% of GDP during 2009-2015 ( Ibid.). As GDP is explicitly a measure of market-based output, severa l transactions are not a part of domestic production definition and are therefore not accounted for (Lippert & Walker, 1997).
Putni ņš & Sauka (2015) used surveys of company managers t o measure the extent of the shadow economy. They argue that due to the unique position , the managers, viewed as experts, ought to know how much of the business income and wages go u nreported in the company. Their method presents estimates of misreported business income, unregistered employees and unreported wages 12
as well as an estimate of the size of a shadow economy as a percentage of GDP. Putni ņš & Sauka (2023) have estimated that the three main components of the shadow economy in Estonia in 2022 are underreporting of salaries (44.5%), underreport ing of employees (28.0%) and underreporting of income (27.5%). What is more, they estimate, that approximately 16.8% of salaries paid by the employers are concealed from the government. The pe rcentage of envelope salaries received has fluctuated between 11.5-18.1% in the last decade (2013-2022) according to their estimations.
Schneider (2016) utilizes the multiple-indicators-m ultiple-causes (MIMIC) procedure as a latent estimator to measure the extent of shadow economy in 25 EU countries. The MIMIC procedure is based on the statistical methodology of unobserved variables, which allows to investigate the relationships between observed variables (indicator s) and unobserved variables (latents). The results suggest that Estonia belongs among the coun tries with largest shadow economies, accounting to 25.4% in 2016. Schneider (2016) obser ves that the size of shadow economy increases from Northern Europe to Southern Europe and from Western Europe to Eastern Europe. Comparing the results from 2016 to 2015, the size o f the shadow economy decreased in most countries. Tafenau et al. (2010) have also estimated the extent of the shado w economy in the regions of the European Union using MIMIC approach on the NUTS 2 level regions and have found that the extent of the shadow economy varies a lot within several European countries. However, Estonia is viewed as a single region and the national average of the shadow economy in 2004 as a percentage of the reported GDP is estimated to be 16.3-16.6%.
Pissarides & Weber (1989) employ an expenditure-based approach to estimate the size of the black economy in Britain. The method involves analysing d iscrepancies between reported income and expenditure. The methodology has been also applied Kukk & Staehr (2014) who found that Estonian households with business income underrepor t 62% of their actual total income, while Kukk et al. (2019) find income underreporting to be more than 40% of self-employed household income on average.
Turu-uuringute AS (2023) estimates the shadow econo my for 2022 by conducting a population survey and the main areas of interest are additiona l income sources, envelope salaries and consumption of illegal tobacco and alcohol. The mai n findings of the survey regarding envelope salaries is that nearly 25% of the population knows someone who earns envelope salary, however, 5% of individuals have received an envelope wage on a regular basis or from time to time in 2022, making up 34% of the individuals’ total salary. The construction sector stood out the most with the 13
share of envelope wages. The same approach was used by EKI (Josing, 2016) and according to their report, 10% of the respondents received envel ope salary in 2015, showing a decline in envelope salaries compared to previous years. Indiv iduals active in the construction sector made up 33% of the individuals earning envelope wage in 2015, supporting even more the existence of unreported income in constrution.
The Eurobarometer (2020) questionnaire on Estonia h as concluded that 27% of the respondents have said that they personally know someone who wor ks without declaring all or part of their income to tax or social security authorities while 6% of the respondents state that they have carried out undeclared paid acitivites themselves. In total , 20% of respondents claimed to be open to the idea of receiving payment from their employer that they knew would not be declared to the tax authorities. Considering the importance of revenue from labour taxation for Estonian government and the estimations for shadow economy and underrep orting of income, analysing labour tax evasion is of great importance.
It should be noted that labour tax evasion can take place both at the extensive margin and at the
intensive margin. The extensive margin is considere d to encompass undeclared employees –
individuals who work for the firm but are not regis tered in the working registry and therefore do
not receive income reported to the tax authorities at all. The intensive margin is considered to be
underreporting of labour income, meaning that the individual is registered in working registry, but
only part of the income is reported to the tax auth orities and rest is received in cash or in kind, in
other words as an envelope wage. (Alm & Malezieux, 2021)
Mineva & Stefanov (2018) have observed that non-dec lared cash payments in addition to the reported income are one of the most complex tax fra ud issues. Compared to undeclared employment, where the person conducts lawful but undeclared activities which can be rather easily traced, intensive margin could be executed either i n form of under-declaring the actual working time or under-declaring the full-time salary, making it difficult to assess if the administrative wage is accurate or not. 14
1.3. Dynamics of tax evasion Tax evasion by firms and employees is complex and m ultifaceted, involving various personal, economic, psychological, and institutional factors. The attitude towards tax evasion or the decision to engage in illicit practices is a combination of individual and corporate views and incentives. Privitera et al. (2021) have pointed out that psychological motives are more important than the economic incentives. Pickhardt & Prinz (2014) find that personal traits and attitudes toward tax evasion and compliance are rather unchangeable but can be influenced by interactions with other individuals. Levenko & Staehr (2023) also find that personal norms and perceived norms of the peers are key predictors of tax compliance, which is supported by Hashimzade et al . (2013), who find that the tax compliance decision is based on t he context of the social environment of the taxpayer.
Different psychological factors involved include perception of fairness and risk aversion. If the tax system is perceived as unfair or the tax burden dee med disproportionately high, the individuals and companies may incline towards tax evasion. What is more, trust in government and courts is negatively related to tax evasion. From an enterpri se point of view, corporate governance and culture play a huge role as companies understanding the social responsibility and ethics are less likely to engage in tax evasion. Also, internal con trols and governance impact the possibilities to engage in misconduct. (Abdixhiku et al., 2017)
Nevertheless, firms and individuals are rational th inkers and engage in tax evasion to gain economic benefit, either by minimizing tax liabilities or maximizing after-tax profits. Tax evasion engagement is seen as a cost-benefit analysis, wher e perceived benefits from tax evasion should outweigh the risk of being caught and applied costs , fines and legal consequences (Allingham & Sand mo, 1972; Hashimzade et al., 2010). Putni ņš & Sauka (2023) find that greater probability of being caught and more serious consequences for not paying taxes discourage entrepreneurs to get involved in such practices. On the other hand, compliance costs also matter and too much time and money spent on reporting and adhering to regulation s is burdensome to companies. Slemrod (2007) also discusses that tax evasion itself impos es efficiency costs as non-compliance must be camouflaged, however if they do not exceed complian ce costs, tax evasion is perceived more attractive.
15
Another important factor to explain the decision is the legal and regulatory environment of an economy. Complex tax regulations create opportuniti es for exploitation and evasion and weak enforcement mechanisms or inadequate penalties may encourage non-compliance (Musgrave 1959; Bre nnan & Buchanan, 1980). Business legislation, tax p olicy and performance of tax authorities are seen as important determinants of i nvolvement in shadow economy with dissatisfaction encouraging shadow acitivity (Putni ņš & Sauka, 2023). Abdixhiku et al. (2017) find a positive relationship between perceived tax burden and tax evasion. Globalisation and increasing cross-border trade have also enabled international tax planning by profit shifting, using tax havens and regulatory arbitrage by using differ ences in different tax regulations across jurisdictions. Even though international tax planning is rather seen as tax avoidance, it might also encompass tax evasion.
More importantly, a firm’s characteristics are related to the extent of tax evasion. Abdixhiku et al. (2017) have found that firm’s size matters and the larger the firm (in terms of the number of employees), the smaller the extent of tax evasion, also pointing out that firms applying international accounting standards are more likely audited and therefore less likely to engage in tax evasion. Kukk & Staehr (2014) and Kukk et al. (2019) find that self-employed households in Estonia are greatly underreporting their earnings, supported by Slemrod (2007) who points out that absence of third-party reporting of wages and salaries facilitates underreporting of income. Putni ņš & Sauka (2023) have also found that smaller firms engage in tax evasion more often than larger firms. They have observed that younger firms engage in more shadow activity than older firms. Industry differences may also explain the de cision to be engage in tax evasion. Highly competitive industries may experience greater press ure to minimize costs, for example construction sector firms participating in tenders. Generally, the results also support higher tax evasion in sectors that involve higher cash transactions, such as hotels and restaurants, construction and wholesale and retail (Abdixhiku et al., 2017).
As discussed above, numerous studies have analysed the attitudes towards tax evasion, mostly relying on survey data and investigating individual attitudes towards tax evasion. However, it is also relevant to understand how the theory of tax evasion regarding individual decision makers is associated with firm compliance and which are the i ndicators to be analysed on firm-level to identify labour tax evasion. 16
1.4. Detecting tax evasion by firms There are different ways to uncover tax evasion at the firm level. Audit data from tax authorities may represent the most accurate way to find a list of companies who have been engaged in illicit tax practices (Benkovskis & Fadejeva, 2022). Howeve r, the data on tax disputes solved in-house is not public and up-to-date tax debt information on a particular firm can be obtained by submitting a debt inquiry on the webpage of EMTA. The audit data from the tax authorities can also be biased towards larger companies or greater benefit as cond ucting a tax audit is costly and the potential increase in tax revenue should cover the cost of conducting an audit.
This is supported by Bobbio (2017), who has shown using Italian firm-level data that smaller firms also tend to spend less on innovation to remain und er the radar and enjoy the cost advantage of evading taxes, resulting in unfair competition. Braguinsky & Mityakov (2015) also present in their results that small enterprises are more eager to en gage in tax evasion. Kukk & Staehr (2014) and Kukk et al . (2019) find that self-employed household income i s greatly unreported. Therefore, it could be assumed that self-employed individuals hav e the possibility and interest to evade taxes. Kleven et al. (2011) also suggest that self-reported income is i ncreasing tax evasion and third- party reporting on the other hand is an effective enforcement device to decrease evasion.
EMTA also monitors the average salary and requests additional checks by the companies as significantly lower salary paid to the employee com pared to average salary for the same position in Estonia may indicate the payment of an envelope salary and therefore labour tax evasion (Lepassar, 2024). However, the ratings are not publ ic if not shared or given access to by the company itself. The more fine-tuned approach, takin g into account individual characteristics would be estimating a wage regression (Mincer, 1975 ). The Mincer wage regression is a widely used model to analyse the relationship between an i ndividual’s earnings and factors such as education, work experience, and other demographic c haracteristics. The model is commonly employed in labour economics to understand how huma n capital influences wages. Gavoille & Zasova (2023) and Benkovskis & Fadejeva (2022) appl y wage regression to spot firms with “suspiciously low wages”.
What is more, Gavoille & Zasova (2023) suggest ther e is a link between labour tax evasion and financial fraud. Labour tax evasion is considered o ne form of financial manipulation and results in particular balance sheet and profit and loss sta tement patterns such as understatement of 17
revenue, assets, expenditure, or liabilities. The a pproach to analyse relationship between the probability of financial manipulation and financial statement variables relies on previous accounting research by Massod Beneish. Beneish (199 9) presents a financial model designed to detect the manipulation of financial statements by assessing the likelihood of earnings manipulation or financial fraud by a firm. The M-sc ore consists of eight financial ratios that are combined to form a single score. These ratios focus on various aspects of firms’ financial statements, such as profitability, cash flows, and accounting quality.
The approach is further developed by Dechow et al . (2009), Cecchini et al . (2010), Hajek & Henriques (2017), showing that different variables from firms’ annual financial reports provide a good indication of whether a company is engaged in financial fraud. Dechow et al . (2009) further assure that financial statement information is usef ul for identifying misreporting and earnings manipulation. Cecchini et al . (2010) provide a methdology for detecting management fraud based on support vector machines, which is a machine lear ning algorithm using supervised learning models to solve complex classification problems. Th eir approach correctly labels 80% of the fraudulent cases and 90.6% non-fraudulent cases, ex ceeding the performance of probit method applied by Beneish (1999) detecting 56% of fraudule nt cases and logistic regression applied by Dechow et al. (2009) detecting 64.5% fraudulent firms.
Hajek & Henriques (2017) examine whether a financial fraud detection model could be developed, combining financial information and linguistic anal ysis of managerial comments (positive, negative, neutral or other words) from financial re ports. They applied different machine learning methods and found that ensemble methods are best at classifying fraudulent companies as fraudulent and Bayesian belief networks work well f or classyfying non-fraudulent firms as non- fraudulent.
Following the assumption of tax evasion being associated with financial fraud, Gavoille & Zasova (2023) use a set of balance sheet and income statem ent items as predictors to detect tax evading firms and implement gradient boosting decision tree s for classification purposes. However, they point out the black box nature of this method as on e of the main drawbacks. Benkovskis & Fadejeva (2022) also use different explanatory vari ables derived from financial reports and estimate a probit model for classification purposes.
18
Analyzing the relationship between firm-level finan cial and non-financial variables and the probability of engaging in labour tax evasion provi des relevant insights for tax authorities to determine whether the allocation of additional reso urces for tax audits or more profound investigations is necessary. Identifying firms possibly engaged in labour tax evasion by conducting a wage regression is a novel methodology, but as wa ge data is usually confidential and not available for the wider audience, detecting pattern s in publicly available financial reports suggesting tax evasion could be a widely used approach. 19
- DATA AND METHODOLOGY This chapter presents an overview of the data used for thesis purposes as well as the methodological approach applied to analyse firm-lev el predictors related to labour tax evasion. Section 2.1. provides overview of data source and discussion of variables included in the analysis. In section 2.2. explanations regarding the choice of methods as well as the limitations are provided. 2.1. Data The paper relies on a matched employer-employee wag e dataset with a monthly frequency, population data and annual report data. This dataset is collected by SA, the main data competence centre in Estonia. Data on wages are collected from the employment register (TÖR) and Annexes 1 and 2 of the tax declaration form TSD (declaratio n of income and social tax, unemployment insurance premiums, and contributions to mandatory funded pension). Data on wages are administrative data, therefore representing the general population and not a sample. Wage data are anonymised 1, meaning that the personal identification code of an individual is replaced by a SA personal identification number to make it impossible to identify individuals from the dataset. Data regarding annual reports originates from the e-Busi ness Register. Data generated and analysed during the thesis are not publicly available due to confidentiality concerns, and sharing the data is not permitted. 2
Figure 2 illustrates the connections between the th ree datasets. The anonymised identification numbers of the employees allow combining population and wage datasets into a subsequently detailed dataset of individual characteristics of t he employees. The dataset provides information on gross wages, personal income tax payments, social security payments, working time, as well as gender, date of birth, education, date of employmen t, economic activity of the employer,
1 The author did not perform anonymisation but recei ved the dataset already anonymised. Therefore, the author had no possibility to identify anyone personally. 2 Access to various datasets owned by SA for researc h purposes can be obtained by submitting a correspo nding request, if not published on the webpage of SA. 20
occupation, location of workplace and more. Additionally, the company registry code connects the wage data and the annual report data.
Figure 2. Data map Source: Composed by author Notes:
- Wage and population datasets can be merged by anonymised identification numbers of the employees.
- Wage and annual report datasets can be merged by company registry code.
The analysis focuses on four Statistical Classifica tion of Economic Activities (NACE) sectors only: manufacturing (NACE 1-digit level code C), co nstruction (NACE 1-digit level code F), wholesale and retail trade; repair of motor vehicles and motorcycles (NACE 1-digit level code G), transportation and storage (NACE 1-digit level code H), following Gavoille & Zasova (2023). Regarding occupations, 1-digit level code of International Standard Classification of Occupations (ISCO) was included, omitting employees in occupati on code zero, referring to armed forces occupations. For location of workplace, the Nomencl ature of Territorial Units for Statistics (hereinafter NUTS) is derived from Estonian Adminis trative and Settlement Classificator (hereinafter EHAK), dividing the country into five different regions, i.e. Northern Estonia, Central Estonia, North-Eastern Estonia, Western Estonia and Southern Estonia. 2.1.1. Wage and population data Several steps were performed in the preprocessing o f data. The following calculations were performed in order to obtain variables for analysis purposes and to implement the necessary trimming of data:
- monthly average wage was calculated by adding up al l monthly payments and diving the sum by the count of months engaged in employment by employer ;