FIRM-LEVEL%20PREDICTORS%20OF%20LABOUR%20TAX%20EVASION.%20Alice%20Mikk.pdf

41

However, as the recall is lower than specifity, the model could be better at classifying compliant firms than non-compliant. The F-score indicates tha t the classifier has archieved moderate precision and TP rate, AUC indicates a relatively g ood performance of the classifier. Figure 3 presents the ROC curve for 2021 on the left-hand si de and for 2022 on the right-hand side. The ROC curve shows the trade-off between sensitivity a nd specificity and the closer the curve is to the top-left corner, the better the performance of the classifier. Therefore, the classifier performs better compared to a random classifier representing the 45-degree line on the graph.

Figure 3. ROC curve Source: Compiled by author Note: This figure displays the ROC curve based on the test set.

Thirdly, the logit model is used to predict the out -of-sample probability of being engaged in tax evasion for all firms for which the labour tax evasion was unknown (based on Eq. 9). For the given threshold of 65%, the model classifies 9534 firms ( 2022: 10 536) as tax evading and 9062 firms (2022: 8918) as tax compliant for 2022. Gavoille & Zasova (2023) find that 37% of the firms are engaged in tax evasion, compared to Benkovskis & Fa dejeva (2022) who report that 75-80% of Latvia’s firms are evading labour taxes. The percentage of companies classified as tax evading is 51% (2022: 54%) which is in between the estimates of previous studies.

To understand the distribution of tax evading firms better, the evasion according to size and sector of the firm is shown for 2022. Table 13 presents th e share of evading firms by size and the share of employees working in evading firms from the size group. The results reveal that tax evasion is more prevalent in smaller enterprises, peaking to 84% in firms with one employee only.

42

Table 13. Evading firms by firm size in 2022 Size class Number of evading firms
Share of evading firms
Share of employees working in evading 1 employee 5765 84.31% 84.31% 2 to 4 employees 4071 62.75% 60.30% 5 to 9 employees 693 23.15% 20.50% 10 to 19 employees 7 0.42% 0.37% 20 or more employees 0 0% 0% Source: Author ’s calculations Table 14 presents the share of evading firms by sec tor and the share of employees working in evading firms from the size group. The share of eva ding firms is highest in construction and transportation and storage. However, the share of employees working in tax evading firms is much smaller for transportation and storage compared to construction, the latter accounting to nearly 36%. The higher share of employees working in tax evading firms, as can be seen in construction sector, is in line with previous findings that highly competitive and cash-intensive industries tend to have higher share of labour tax evasion.
Table 14. Evading firms by sectors in 2022 Industry Number of evading firms
Share of evading firms
Share of employees working in evading Manufacturing 981 28.43% 2.54% Construction 4752 77.03% 35.50% Wholesale and retail trade 3027 41.66% 7.61% Transportation and storage 1776 69.16% 16.50% Source: Author ’s calculations

Therefore, despite the arguably subjective choice o f probability threshold, the results are in line with previous findings by Putni ņš & Sauka (2015), Gavoille & Zasova (2021), Benkovs kis & Fadejeva (2022). 3.2. Robustness checks To ensure the reliability and validity of the regre ssion results, robustness checks were conducted for both the wage regression and logistic regression. 43

3.2.1. Wage regression The robustness checks carried out for wage regression were exclusion of a variable and cross- validation. Firstly, the educational attainment variable was excluded from the model as there were missing values for approximately 25% of the observations. Even though the distribution of individuals with missing educational attainment seemed random and not systematic, it could still cause issues and influence wage regression results. The wage regression without educational attainment showed relatively similar results compared to the final model and is presented in Appendix 4. The signs of the coefficients remain unchanged, however, the explanatory performance of the model decreases. This is expected, as education is an important factor impacting the job and salary prospects (Mincer, 1975).

Secondly, cross-validation was implemented on the final model. The data was randomly partitioned into training (70%) and test (30%) sets, and an OLS regression model was fitted using the training data. Afterwards, the values of the dependent variable were predicted using the fitted model and testing data. The performance was evaluated by calculating Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and R². The results from five random partitions are presented in table 15.
Table 15. Cross-validation results for OLS model Iteration MSE RMSE R² 1 0.170 0.423 0.316 2 0.169 0.421 0.314 3 0.169 0.411 0.316 4 0.170 0.412 0.315 5 0.170 0.412 0.316 Average 0.170 0.416 0.315 Source: Author’s calculations Note: MSE, Mean Squared Error. RMSE, Root Mean Squared Error. R², R-squared. The values suggest that the model has consistent MSE, RMSE and R² values across iterations. The deviation from average is relatively low, and the proportion of variance in the dependent variable that is explained by the independent variables is similar to the final model. Therefore, the cross- validation results suggest that the OLS model is stable.

44

3.2.2. Logistic regression To validate and assess the model's generalizability and performance on unseen data, cross- validation was applied. Over all samples, 10-fold cross-validation suggests an accuracy of 92.57% for 2021 (2022: 92.30%). The kappa statistic measure value of 0.45 for 2021 (2022: 0.44) suggests a moderate agreement beyond chance. Therefore, the model has a good overall performance, but the predictive performance could be improved by, for example, adjusting model specifications.

As logistic regression is sensitive to probability threshold selection, the probability threshold of 84% following Benkovskis & Fadejeva (2022) is applied and the results are compared. The probability threshold of 65% presented conservative results, therefore it is expected that higher probability threshold results in a lower TP rate. The confusion matrix in table 16 presents results for the comparative analysis. The model is good at finding compliant companies but performs poorly in finding non-compliant companies. Table 16. Confusion matrix

2021 2022

False True False True 0 102 6 106 1 1 774 157 721 144 Source: Author’s calculations Therefore, increasing the probability threshold to 84% following Benkovskis & Fadejeva (2022) results in lower accuracy and more conservative results (table 17). The TN rate for 2021 in this case would be 94% (2022: 99%), but TP rate is 18% (2022: 17%), which is low. The relatively high AUC shows that the model is able to distinguish between true and false type well, but the chosen threshold might be inaccurate. Table 17. Prediction performance ratios

Accuracy TP rate FP rate TN rate FN rate F-score AUC 2021 24.93 17.86 5.56 94.44 83.14 0.29 0.84 2022 25.72 16.65 0.93 99.07 83.35 0.29 0.88 Source: Author’s calculations As a result, probability threshold of 84% results in 15 466 firms (2022: 16 021) classified as tax compliant and 3130 firms (2022: 3433) classified as tax evading. Therefore, 17% of firms (2022: 18%) firms are classified as evading labour taxes.

45

3.3. Discussion The results of the analysis suggest that larger com panies (in terms of number of employees) are less likely to evade labour taxes and labour tax ev asion is highest among self-employed. The results regarding financial ratios suggest that the probability to evade labour taxes decreases as turnover, debt to assets, or cost of goods sold to assets increases. Conversely, the predicted probability to evade labour taxes increases as shor t-term debt to assets or turnover to assets increases. What is more, labour tax evasion is mor e prevalent in construction sector. The prediction of out-of-sample probability of being en gaged in labour tax evasion suggested that in 2021 51% of companies (54% in 2022) are classified as labour tax evading in the four NACE sectors under investigation.

Even though using administrative data from SA has advantages of being representative, there are also some limitations to be considered in the inter pretation and discussion of the results. Administrative data on wages is only available for years 2021 and 2022, which on the positive side is the most recent data but on the negative si de limits the time frame to be used for analysis purposes. More comprehensive results could be obtai ned if investigating longer time series and implementing a panel data approach.

From the methodological perspective, logistic regression is suitable if the main focus is not solely on prediction power but rather on interpretation of the results. Cecchini et al. (2010) also compare the performance of other fraud detection methods applied in previous papers and find that support vector machines using the financial kernel performs best for recall. Gavoille & Zasova (2023) apply gradient boosting decision trees for better p rediction performance. However, these models offer less opportunities for interpretation and could be used if predictive power is of importance.

It is also important to acknowledge that the wage r egression model and logistic regression model do not capture all the individual or firm-level het erogeneity that could explain the decision to be engaged in tax evasion, but only factors available in the database. There are many other unobserved factors, such as the personal views of the key personnel and employees, overall company culture, understanding the tax system and experience with ta x authorities that can have significant effect on the decision of tax evasion or tax compliance.

46

Combining together survey results, for example evas ion behaviour in company managers (or personal views of employees) with financial and non-financial variables of firms could give a more comprehensible overview of the topic. Finally, using administrative data combined together with surveys such as HBS, offering information on consumption for validating the results of the labour tax evasion, as relatively smaller income compared to consumption can refer to envelope salary. Hajek & Henriques (2017) note that text in annual r eports can be used to detect fraudulent firms. They found that using linguistic variables (frequen cy count of different positive, negative, uncertain or other words) in addition to financial variables improved the performance of logistic regression compared to including only financial variables, resulting in improved accuracy, higher true positive rate and true negative rate.

The findings may also suggest that measures to motivate labour tax compliance for self-employed should be improved. Personal views on taxation largely determine the decision to comply or evade, however, nudging techniques could improve tax compliance among self-employed. Therefore, the effect of tax compliance ratings displayed to taxpa yers by EMTA on wages could be further investigated to see, whether the tool has been effe ctive in increasing labour tax compliance. Additionally, it would be important to measure the revenue lost due to non-compliance, for example, by evaluating the volume of unreported wage, following Benkovskis & Fadejeva (2022). What is more, it could be investigated how the minimum wage hike effects the compliant and non- compliant firms differently, following Gavoille & Zasova (2023). 47

CONCLUSION The aim of this thesis was to analyse the relations hip between firm-level financial and non- financial indicators and the probability of a firm being engaged in labour tax evasion. The analysis aimed to answer two main research questions:

1. Which firm-level predictors contribute to the proba bility of being engaged in labour tax evasion?
2. What is the proportion of labour tax evading companies?

To answer the research questions raised, a database comprising three merged datasets was 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 analysis findings indicate that larger companies, as measured by their number of employees, exhibit lower tendencies to evade labor taxes. Conv ersely, labor tax evasion is most prevalent among the self-employed, reaching 84% in 2022. Cons equently, the results support the need to implement strategies aimed at fostering labor tax c ompliance among the self-employed. While individual attitudes towards taxation significantly influence compliance decisions, employing various cost-effective nudging techniques could enh ance tax compliance among the self- employed. Therefore, further investigation into the impact of tax compliance ratings displayed to taxpayers by EMTA on wages is required to ascertain the tool's effectiveness in increasing labor tax compliance. Additionally, it is crucial to quan tify the revenue loss resulting from non- compliance, such as by assessing the volume of unre ported wages, following the methodology outlined by Benkovskis & Fadejeva (2022).

What is more, labour tax evasion is more prevalent in the construction sector. The results for 2022 suggest that 77% of companies active in the construction sector are engaged in labour tax evasion ; however they employ 36% of the employees in the sec tor. Therefore, more attention should be 48

paid to the construction sector, which is by nature competitive and cash-intensive. The prediction of the out-of-sample probability of being engaged i n labour tax evasion suggested that in 2021 51% of companies (2022: 54%) are classified as labo ur tax evading in the four NACE sectors under investigation.

The results regarding financial ratios suggest that the probability of evading labour taxes decreases as turnover, debt to assets, or cost of goods sold to assets increases. Conversely, the predicted probability of evading labour taxes increases as sh ort-term debt to assets or turnover to assets increases.

There are opportunities for improvements as well as for future research on this topic. The results suggest that a more refined model is required or an other machine learning approach should be applied to improve the prediction performance of th e model. Additionally, the topic could be further developed to examine the dynamics and patterns in employee salaries, changes in minimum wage and financial statements. For example, it could be investigated how the minimum wage hike affects the compliant and non-compliant firms diffe rently, following Gavoille & Zasova (2023). Lastly, besides the extensive margin, the intensive margin of labour tax evasion could be further investigated to understand the features of companies with unreported employees. 49

KOKKUVÕTE TÖÖJÕUMAKSUDEST KÕRVALEHOIDUMIST ENNUSTAVAD TEGURID
ETTEVÕTTE TASANDIL Alice Mikk Käesoleva magistritöö eesmärk oli hinnata erinevate finants- ja mittefinantsnäitajate seost tööjõumaksudest kõrvale hoidumise tõenäosusega ette võtte tasandil. Analüüsi käigus keskenduti põhiliselt kahele uurimisküsimusele:

1. Millised ettevõttespetsiifilised tegurid panustavad tööjõumaksudest kõrvale hoidumise tõenäosuse suurenemisse?
2. Kui suur on tööjõumaksudest kõrvale hoiduvate ettevõtete osakaal?

Küsimustele vastamiseks kasutati kolme ühendatud an dmestikku, mis hõlmas endas nii palkade, rahvastiku kui ka ettevõtete majandusaasta aruannet e andmeid. Analüüsis kasutati Minceri palgaregressiooni, et leida ettevõtted, kes maksava d töötajatele “kahtlaselt madalat palka” ning logistilist regressiooni, et analüüsida erinevate e ttevõtte finants- ja mittefinantsnäitajate seost palgamaksudest kõrvale hoidumise tõenäosusega.

Analüüsi tulemused viitavad sellele, et suuremad et tevõtted (töötajate arvu mõistes) hoiduvad väiksema tõenäosusega tööjõumaksude tasumisest. Töö jõumaksude maksmisest kõrvale hoidumine on kõrgeim ühe töötajaga ettevõtete hulga s, moodustades 84% kõigist ühe töötajaga ettevõtetest. Tulemused ilmestavad, et tööjõumaksude laekumise parandamiseks tuleb just mõelda ühe töötajaga firmade võimalikult kuluefektiivsele motiveerimisele, näiteks nügimismeetodeid (inglise keeles nudging ) kasutades. Näiteks võiks analüüsida EMTA poolt ma ksumaksjatele kuvatava maksukuulekuse reitingu mõju palkade dekla reerimisele ja tööjõumaksudele. Antud analüüsi laiendusena võiks leida saamata jäänud töö jõumaksude ulatuse kõigi ettevõtete osas järgides Benkovskis & Fadejeva (2022) lähenemist.

50

Tööjõumaksudest hoidumine on kõige prevalentsem ehi tussektoris, moodustades analüüsi tulemuste põhjal 2022. aastal koguni 77% kõigist eh itussektori ettevõtetest ning pakkudes tööd 36% ehitussektori tööjõule. Seetõttu tuleb tööjõumaksudest hoidumise vaates rohkem tähelepanu pöörata ehitussektorile kui kõrge konkurentsiga ja sularahaintensiivsele sektorile. Magistritöö tulemusel hinnati, et uuritud neljal tegevusalal ko kku hoidus 2021. aastal tööjõumaksudest 51% ettevõtetest ning 2022. aastal 54% ettevõtetest.

Finantssuhtarvude ja tööjõumaksudest hoidumise tõenäosuse uurimisel selgus, et tööjõumaksudest kõrvalehoidumise tõenäosus väheneb käibe, võlakordaja ja müüdud toodangu (kaupade, teenuste) kulu suhe varadesse kasvades. Vastupidiselt kasvab tõenäosus tööjõumaksudest kõrvale hoidumiseks kui suureneb lühiajalise võla kordaja või varade käibekordaja.

Antud analüüsi puhul on mitmeid võimalusi täiustust eks ja edaspidiseks uurimiseks. Tulemused viitavad ka sellele, et mudeli ennustusvõime parand amiseks on vaja mudeli spetsifikatsioone täpsustada või rakendada mõnd teist masinõppe lähenemist. Lisaks võiks analüüsi edasiarendusena uurida töötajate palkade, miinimumpalga muutuste ja raamatupidamisaruannete dünaamikat ja mustreid. Näiteks võiks Gavoille & Zasova (2023) lä henemist järgides hinnata, kui palju erineb miinimupalga tõusu mõju maksukuulekate ja maksudest hoiduvate ettevõtete finantsnäitajatele. Täiendavalt võiks analüüsida ka tööjõumaksudest kõr vale hoidumist ettevõtetes registreerimata töötajate vaatest. 51

LIST OF REFERENCES Abdixhiku, L., Krasniqi, B., Pugh & G., Hashi, I. ( 2017). Firm-level determinants of tax evasion in transition economies. Economic Systems, 41 (3), 354-366. https://doi.org/10.1016/j.ecosys.2016.12.004

Allingham, M. G. & Sandmo, A. (1972). Income tax ev asion: a theoretical analysis. Journal of Public Economics, 1(3-4), 323-338.

Alm, J. & Malézieux, A. (2021). 40 years of tax eva sion games: a meta-analysis. Experimental Economics, 24 (3), 699-750.

Alm, J. (1988). Compliance Costs and the Tax Avoidance-Tax Evasion Decision. Public Finance Quarterly, 16 (1), 31-66. https://doi.org/10.1177/109114218801600102

Arsi ć, M., Arandarenko, M., Radulovi ć, B., Ran đelovi ć, S. & Jankovi ć, I. (2015). Causes of the Shadow Economy. Formalizing the Shadow Economy in S erbia. Contributions to Economics , 21-56. https://doi.org/10.1007/978-3-319-13437-6_4

Benkovskis, K. & Fadejeva, L. (2022). Chasing the Shadow: The Evaluation of Unreported Wage Payments in Latvia. Bank of Latvia Working Paper, 2022/01.

Bobbio, E. (2017). Tax Evasion, Firm Dynamics and G rowth. Bank of Italy Occasional Paper, 357 . http://dx.doi.org/10.2139/ssrn.2910376

Braguinsky, S. & Mityakov, S. (2015). Foreign corpo rations and the culture of transparency: Evidence from Russian administrative data. Journal of Financial Economics 117 (1), 139– 164. https://doi.org/10.1016/j.jfineco.2013.02.016

Brennan, G. & Buchanan, J. (1980). The Power to Tax : Analytical Foundations of a Fiscal Constitution. Cambridge University Press .

DeBacker, J., Heim, B. T. & Tran, A. (2015). Import ing corruption culture from overseas: Evidence from corporate tax evasion in the United S tates. Journal of Financial Economics, 117 (1), 122-138. https://doi.org/10.1016/j.jfineco.2012.11.009.

Dom, R., Davenport, A. & Prichard, S. R. (2022). Innovations in Tax Compliance: Building Trust, Navigating Politics, and Tailoring Reform. World Bank .

Eurobarometer (2020). Undeclared work in the Europe an Union. European Commission, Technical report, 498.

Gavoille, N. & A. Zasova (2021). Foreign ownership and labor tax evasion: Evidence from Latvia. Economics Letters 207 (C), 1–4. https://doi.org/10.1016/j.econlet.2021.110030.
52

Gavoille, N. & Zasova, A. (2023). What we pay in the shadows: Labor tax evasion, minimum wage hike and employment, Journal of Public Economics, 228 . https://doi.org/10.1016/j.jpubeco.2023.105027

Gujarati, D. N. (2003). Basic econometrics (4th ed). McGraw Hill.

Hajek, P. & Henriques, R. (2017). Mining corporate annual reports for intelligent detection of financial statement fraud – a comparative study of machine learning methods. Knowledge-Based Systems, 128 , 139–152. https://doi.org/10.1016/j.knosys.2017.05.001

Harmon, C. & Walker, I. (1995). Estimates of the ec onomic return to schooling for the United Kingdom. American Economic Review, 85 (5), 1278-1286.

Hashimzade, N., Myles, G. D. & Tran-Nam, B. (2013). Applications of behavioural economics to tax evasion. Journal of Economic Surveys, 27 (5). http://dx.doi.org/10.1111/j.1467- 6419.2012.00733.x

Josing, M. (2016). Varimajanduse trendid. Eesti Konjunktuuriinstituut.

Kenny, L. W. (2006). Tax Systems in the World: An Empirical Investigation into the Importance of Tax Bases, Administration Costs, Scale and Polit ical Regime. International Tax and Public Finance, 13, 181-215. https://doi.org/10.1007/s10797-006-3564-7

Kleven, H. J., Knudsen, M. B., Kreiner, C. T., Pedersen, S., & Saez, E. (2011). Unwilling or unable to cheat? Evidence from a tax audit experiment in D enmark. Econometrica, 79 (3), 651- 692. https://doi.org/10.3982/ECTA9113

Kukk, M. & Staehr, K. (2014). Income underreporting by households with business income: evidence from Estonia. Post-Communist Economies, 26 (2), 257-276. https://doi.org/10.1080/14631377.2014.904110

Kukk, M., Paulus, A. & Staehr, K. (2019). Cheating in Europe: underreporting of self-employment income in comparative perspective. International Tax and Public Finance, 27, 363-390. https://doi.org/10.1007/s10797-019-09562-9

Lepassar, A. (2024). Maksukäitumise hinnangud MTA e-teenus ettevõtjale “virtuaalne audiitor”. Maksu- ja Tolliamet.

Levenko, N. & Staehr, K. (2023). Self-reported tax compliance in post-transition Estonia . Economic Systems, 47(3). https://doi.org/10.1016/j.ecosys.2022.101047

Levi, M. (1988). Of Rule and Revenue. University of California Press.

Lippert, O. & Walker, M. (1997). The Underground Economy: Global Evidence of Its Size and Impact. The Fraser Institute, Vancouver.

Mendgen, A. (2023). International Tax Competitivene ss Index 2023. Tax Foundation Center for Global Tax Policy, 10.

53

Meriküll, J. & Tverdostup, M. (2023). The gap that survived the transition: The gender wage gap in Estonia over three decades. Economic Systems, 47 (3). https://doi.org/10.1016/j.ecosys.2023.101127

Mincer, J. (1975). Education, experience and the di stribution of earnings and employment: An overview. National Bureau of Economic Research , 71–93.

Musgrave, R.A. (1959) The Theory of Public Finance. McGraw Hill, New York.

Müürsepp, R. (2015, October 12). Kui suur on Eesti varimajandus? Used 21.02.2024. https://www.stat.ee/et/uudised/2015/10/12/kui-suur-on-eesti-varimajandus

Pickhardt, M. & Prinz, A. (2014). Behavioural dynam ics of tax evasion – A survey. Journal of Economic Psychology, 40, 1-19. https://doi.org/10.1016/j.joep.2013.08.006

Pissarides, C. A. & Weber, G. (1989). An expenditure-based estimate of Britain's black economy, Journal of Public Economics, 39 (1), 17-32.

Privitera, A., Enachescu, J., Kirchler, E. & Hartma nn, A. J. (2021). Emotions in Tax Related Situations Shape Compliance Intentions: A Compariso n between Austria and Italy. Journal of Behavioural and Experimental Economics, 92. https://doi.org/10.1016/j.socec.2021.101698

Putni ņš, T. J. & Sauka, A. (2015). Measuring the shadow economy using company managers. Journal of Comparative Economics, 43 (2), 471-490. https://doi.org/10.1016/j.jce.2014.04.001

Putni ņš, T. J. & Sauka, A. (2023). Shadow Economy Index f or the Baltic Countries 2009-2022. 13 th Annual Shadow Economy Conference.

Rahandusministeerium (2024, April 3). Maksud . Used 06.03.2024. https://www.fin.ee/riigi- rahandus-ja-maksud/maksu-ja-tollipoliitika/maksud

Schneider, F. (2016). Estimating the Size of Shadow Economies of Highly-developed Countries: Selected New Results. Leibniz Institute for Economic Research at the Univ ersity of Munich, 14 (4), 44-53.

Schumpeter, J. A. (1991). The Economics and Sociolo gy of Capitalism. Princeton University Press. https://doi.org/10.2307/j.ctv173f01t

Slemrod, J. (2007). Cheating Ourselves: The Economi cs of Tax Evasion. Journal of Economic Perspectives, 21 (1), 25-48.

Smeaton, D. & McKay, S. (2003). Working after State Pension Age: Quantitative Analysis. Department for Work and Pensions, Research Report 182 .

Statistics Estonia. (2023). PA117: Average monthly gross wages (salaries), median and number of employees by Indicator, County and Reference period . Used 23.03.2024. https://andmed.stat.ee/en/stat/majandus__palk-ja- toojeukulu__palk__luhiajastatistika/PA117
54

Statistics Estonia. (2023). PA5335: Gender pay gap by economic activity (EMTAK 2008), October. Used 29.03.2024. https://andmed.stat.ee/en/stat/majandus__rahandus__valitsemissektori- rahandus__valitsemissektori-tulud-kulud/RR057

Statistics Estonia. (2023). RR057: Consolidated revenue and expenditure of general government (quarters) (ESA 2010). Used 21.02.2024. https://andmed.stat.ee/en/stat/majandus__rahandus__valitsemissektori- rahandus__valitsemissektori-tulud-kulud/RR057

Tafenau, E., Herwartz, H. & Schneider, F. (2010). Regional Estimates of the Shadow Economy in Europe. International Economic Journal, 24(4), 629-636, https://doi.org/10.1080/10168737.2010.526010

Tonin, M. (2011). Minimum wage and tax evasion: Theory and evidence. Journal of Public Economics, 95(11-12), 1635-1651. https://doi.org/10.1016/j.jpubeco.2011.04.005

Turu-uuringute AS (2023, October 5). Varimajanduse uuring 2023. Used 20.03.2024. https://www.emta.ee/sites/default/files/documents/2023- 10/varimajanduse_uuring_2023.pdf