Determinants of tax avoidance – evidence on profit tax-paying companies in Romania

Abstract Our objective is to investigate the determinants of tax avoidance in Romanian companies in 2013–2017. Our initial sample comprises 236 privately owned companies that are payers of profit tax and have been found guilty of tax evasion. This was matched with 236 ‘compliant’ companies structured similarly by industry, whereas the final sample comprises 1674-year-observations. We defined ‘compliant’ those companies that have never been prosecuted for tax evasion. Our main finding is that larger companies with lower financial performance and lower leverage ratio are more inclined towards tax avoidance. The geographical region and the industry sector in which companies operate in are also determining their tax avoidant-behaviour. Surprisingly, the fiscal regulations amended starting with 2016 did not lead to an apparent exacerbation of tax avoidance among profit tax payers.


Introduction
Tax avoidance is of great importance because it restricts state's ability to collect money and to put policies into practice as taxpayers find ways to reduce their taxable base. This is the reason why investigating the determinants of tax avoidance has been an important concern in the accounting field for the past two decades (Halioui et al., 2016). Governments conduct fiscal policies to make laws complete and accurate and to fight against tax evasion because of their need for tax revenues (Martinez, 2017). At the same time, when doing business, companies around the world minimise tax liabilities through tax aggressive activities (Lanis & Richardson, 2012;Martinez, 2017).
Different terms are relevant in the context of our study. Hanlon and Heitzman (2010) define tax avoidance very broadly. If tax avoidance represents a continuum of tax planning strategies, at the lower end we have perfectly legal tax reducing strategies, while terms such as 'noncompliance,' ''evasion,' and 'sheltering' would be closer to the other end of the continuum. For our research purposes, we define tax evasion as tax aggressiveness outside the boundaries of the tax law, while tax aggressiveness within the boundaries of the tax law is in our view tax avoidance. More precisely, tax evasion is 'the wilful attempt to defeat or circumvent the tax law in order to illegally reduce one's tax liability' (Gottschalk, 2010). In other words, tax evasion is pecuniary and criminally sanctioned (Constantin, 2014). On the other hand, by definition, tax avoidance is the lawful underreporting of tax obligations (Lipatov, 2012). Tax avoidance is the legally avoiding of stating and paying taxes to the state budget. Tax avoidance consists in escaping taxes by legal means using legislative gaps in taxpayers' favour (the so-called 'loopholes'), and therefore it is only possible when the law is incomplete or has inaccuracies.
The academic literature on tax avoidance in Romania is rather scarce. Romanian authors prefer to focus on tax evasion: its definition, its causes, its forms, its effects, as well as the ways of preventing and fighting it (Comândaru et al., 2018;Constantin, 2016;2018;Corîci &Ilincuīa, 2017;David & Pojar, 2011;Manea, 2011;Perpelea & Beldiman, 2016;Popescu & Cochinţu, 2014;Socoliuc et al., 2018;S , uvelea, 2014;Ungureanu et al., 2016). The reason for this growing concern for tax evasion is the increased frequency of scandals regarding corrupt state officials involved in tax evasion crimes (Manea, 2015) and the inefficiency of the fiscal supervision performed by tax authorities in Romania (Pana, 2019). However, at international level, many recent studies focused on understanding the corporate taxpayers' behaviour and especially companies' tax avoidant behaviour (Chen et al., 2010;Chen & Chu, 2005;Desai & Dharmapala, 2006;2009a;2009b;Frank et al., 2009). An issue that remains unsolved in the academic debate is the adequate measurement construct for tax avoidance. A review of tax research by Hanlon and Heitzman (2010) criticises 12 (twelve) measures used as a proxy of corporate tax avoidance by various authors along the years.
The objective of our research was to investigate the determinants of tax avoidance of companies headquartered in Romania over a five-year period (2013 through 2017). Our sample includes evading companies (companies that were prosecuted and found guilty for engaging in tax evasion in the analysed timeframe) and a matched group of compliant companies (for which no such prosecution existed).
The novel character of our approach consists in the fact that we include in our sample tax evasion perpetrators. Moreover, to the best of our knowledge, no other authors investigate the determinants of tax avoidance at company-level in Romania. Another element of novelty is that we do not limit our study to the tax-avoidant behaviour of listed companies. Most of the prior international studies only analysed the behaviour of public entities, which in our opinion is rather restrictive (Buijink et al., 1999;Chennells & Griffith, 1997;Collins & Shackelford, 1995). Moreover, earlier studies (e.g., Siegfried, 1972) used industry-level data to their disadvantage: companies' business typically spans many different industry groups and aggregated industry data obscure differences among individual companies' (Stickney & McGee, 1982). In our study, we use company-level data, which provides more insightful results. The relevance of our sample and the fact that this area of research received little attention in Romania make the contribution of our study particularly significant. Our conclusions could also be useful to those interested in the corporate tax avoidance phenomenon in countries with a similar historic background such as ours.
Our paper has the following roadmap. The first section presents relevant studies published both in Romania and internationally. The next section describes the data collection process and offers arguments in favour of the chosen hypotheses. Following, we present the econometric results, as well as robustness checks. The last section highlights the major conclusions, the limits and the implications of our study.

Literature review
Corporate tax avoidance has been investigated in a variety of studies that aimed to understand the determinants and economic implications of such tax minimising behaviours. Table 1 offers a summary of relevant recent international studies on corporate tax avoidance. The Table is useful in providing an overview of the methodology employed over the years in this line of study and is fundamental for our own choice of the research design. Corporate tax avoidance was subject of integrative literature reviews, such as the one carried out by Whait et al. (2018) on the basis of 143 academic articles. Another recent methodological approach was a between-subjects experiment performed by Antonetti and Anesa (2017) to investigate how stakeholders react to corporate tax strategies that are either aggressive or conservative. Only a few authors chose an event-study methodology to examine the stock price reaction to news referring to corporate tax avoidance (Blaufus et al., 2019;Hanlon & Slemrod, 2009).
Most authors performed a regression analysis where corporate tax avoidance was the analysed as (dependent) variable (Armstrong et al., 2012;Chen et al., 2010;Chyz et al., 2013;Lanis & Richardson, 2012;Mafrolla & D'Amico, 2016;Richardson et al., 2013;2015b;Steijvers & Niskanen, 2014). In other studies, the corporate tax avoidance is treated within the regression model as an independent (explanatory) variable. For instance, Richardson et al. (2014) examine the influence of tax avoidance on corporate debt policy, while Kubick and Lockhart (2017) investigate the association between tax avoidance and corporate debt maturity. We have chosen to embark on a multivariate regression analysis where corporate tax avoidance is the dependent (explained) variable, thus following the majority of other authors in the field.
Regarding the analysed geographical region, we have chosen to focus on one country only -Romaniaas did many researchers before us. Few studies are cross-country analyses which comprise several European countries (such as Buijink et al., 1999;Jaafar & Thornton, 2015). Most studies available in international journals use data on companies established in the United States (such as Armstrong et al., 2012;Chen et al., 2010;Chyz et al., 2013;Gallemore & Labro, 2015;Gupta & Newberry, 1997;Higgins et al., 2015;Jennings et al., 2012 etc.). Some other one-country studies are focusing on Australia and New Zealand (Harris & Feeny, 2003;Lanis & Richardson, 2012;Richardson & Lanis, 2007;Taylor & Richardson, 2012;Wilkinson et al., 2001).  countries (Brazil, China, India and Russian Federation). Herbert and Overesch (2014) and Kraft (2014) 2018) and Onofrei et al. (2018) wrote about the determinants of the effective tax rate in the Eastern European companiesincluding Romania. Our research aims at filling in this gap in the literature and at exploring in depth the determinants of tax avoidance in Romanian companies, irrespective whether listed or not.

Data collection
We identified all entities that were found guilty of tax evasion crimes carried out over the period 2013-2017 (in accordance with Law 241/2005 on the prevention and fighting of tax evasion). The names of these entities were published by the Ministry of Justice in Romania and the National Anticorruption Directorate on their websites (http://portal.just.ro and https://www.pna.ro). Then, we eliminated from our sample the following types of entities: national companies (e.g., companies where the state is the sole shareholder); autonomous administrations; national research and development institutes; associations; non-banking financial institutions; as well as insurance companies and banks.
Consequently, our final sample comprises 236 privately owned limited liability and stock companies that are payers of profit tax and submitted their annual financial statements in at least one of the five years from within the analysed period 2013-2017. For each of these 236 companies, we collected the financial data published by the Romanian Ministry of Public Finance over the period 2013-2017. Additional to data of financial nature, we also obtained the following pieces of information about the sampled companies: the average number of employees, the county where the company is registered in and the main type of activity each company performs as per the categories from the Statistical classification of economic activities in the European Community (NACE). The resulting panel is unbalanced. Tables 2-4 show the structure of our sample in agreement with three criteria: fiscal year, industry and region.
With every passing year, fewer and fewer evading companies submitted their annual financial statements to the fiscal authorities. Thus, the sub-sample of the financial year 2013 includes most observations (namely 190, representing approximately 30% of the total year-observations), while the sub-sample of 2017 includes only 75 observations, meaning that only 75 evading companies submitted their financial statements during that year.
Regarding the region where the selected evading companies are registered in, the analysis of the sample structure reveals that 51 evading companies out of 236 are registered in Bucharest-Ilfov. All others operate in one of the following regions: North-West; South-Muntenia; Center; West; South-East; North-East; and South-West Oltenia.
Concerning the industry sector in which the evading companies operate in, more than a third (90 companies, representing 38% of the sample) belong to the NACE category 'Wholesale and retail trade; repair of motor vehicles and motorcycles'. Additionally, we created a match for the group of evading companies, respectively a sample of 236 companies that were never prosecuted for tax evasion crimes and existed between 2013 and 2017. We called this control samplecompliant   companies. The sample of compliant companies was selected randomly, making sure that it perfectly matches the group of evading companies in accordance with the industry criterion. The data collection framework was the database containing the 2017 financial data submitted by companies across Romania and made publicly available by the Ministry of Public Finance on one of its official websites. To begin with, we eliminated those companies that we knew were accused of tax evasion crimes. Next, we selected solely companies that were payers of profit tax. Only afterwards, we applied random sampling to the remaining list of entities, with the assistance of Stata, so that the sample of compliant companies is the exact mirror of the initial sample of evading companies from the point of view of their industry. Another legislative change relevant to our study is the amendment of the Tax Procedure Code by introducing the principle 'in dubio contra fiscum', applicable starting with January 2016. This principle states that in case of uncertainties in the tax provisions, the law shall be interpreted in favour of the taxpayers, thus protecting them against an abusive application of the law by the tax authorities (Blanch, 2016). Therefore, companies could interpret the 2016 amendments to the Tax Code in their favour in the sense that they could match more expenses with their economic activity, thus classify them as deductible and ultimately lowering their profit tax liability. Following all these legislative amendments, the range of deductible expenses is much wider for payers of profit tax, which in our view encourages them to engage in tax avoidance.

Hypotheses development
Second, prior research finds that long-run tax avoidance is positively associated with company size (Dyreng et al., 2008;Richardson et al., 2015b;Richardson & Lanis, 2007). This could be explained by the fact that larger-sized companies have greater power in comparison to smaller-sized companies. Larger companies are able to influence the political process in their favour to a greater extent than small companies (Stickney & McGee, 1982). In particular, large companies can use their resources and power to negotiate their tax burden or influence legislation in their favour (for instance, through lobbying activities) (Gupta & Newberry, 1997;Siegfried, 1972;Stickney & McGee, 1982). Consequently, we investigate whether company size is associated with tax avoidance. Size is calculated as the natural logarithm of total assets (variable SIZE).
Third, companies with higher financial performance generally have a higher taxable profit and therefore a higher tax on profit. For these reasons, we expect them to be more incentivised to minimise their tax liabilities, in other words to be more inclined towards tax avoidance. Eichfelder and Hechtner (2018) argue that profitable companies have more resources which allow them to employ good fiscal consultants who assist them in reducing their due tax. We thus posit a positive association between a company's financial performance and its tax-avoidant behaviour. We use the Return on Assets as indicator of a company's profitability (variable PERF).
Fourth, Mills and Newberry (2005) and Drucker (2006) explain the existence of a negative association between a company's leverage and its tax-avoidant behaviour. They suggest that some companies wish to create off-balance-sheet financing via tax avoidance, thereby removing debt from the financial statements and lowering leverage. Such companies use tax avoidance actions to prevent great sums of money raised from being booked as debt on the company's financial statements. For this reason, we include in modelling tax avoidance a company's leverage ratio measured as debt divided by total assets (variable LEV). We expect the leverage ratio is negatively associated with tax avoidance, as prior studies suggest (Allen et al., 2016;Chyz et al., 2013;Drucker, 2006;Majeed & Yan, 2019;Mills & Newberry, 2005).
Finally, we also include in our analysis various control variables. We investigate the entire sample of both evading and compliant companies; therefore we define a dummy variable (EVZ) that takes the value 0 if the disclosing company was compliant in the analysed timeframe and the value 1 if it was prosecuted and found guilty for tax evasion. Our assumption is that tax evaders are more tax avoidant than 'clean' companies which were never prosecuted for tax evasion crimes. Next, we include in the analysis the region where the selected companies operate in. We define a binary variable that takes the value 1 if the company is based in Bucharest and 0 otherwise (variable REGION). We draw on the fact that studies (e.g., Zeng, 2011) have shown that companies pay higher/lower tax when they are located in a region with more/less developed market and legal institutions. We do not predict any sign for the region dummy. Second, we are aware of the fact that each product market of service market has its own specificities (Bus , u, 2014;Jind richovsk a et al., 2020). Thus, we include a dummy variable for the industry sector that takes the following values: 'A' for Agriculture, forestry and fishing; 'C' for Manufacturing; 'F' for Construction; 'G' for Wholesale and retail trade, repair of motor vehicles and motorcycles (variable INDUSTRY). These categories are compared to the category 'Other' which cumulates all other sectors not mentioned above. The reason for including this control variable is the possibility for tax avoidance to vary across different industry sectors (Rego, 2003;Richardson et al., 2015a). No sign predictions are made for the industry dummies either.
To test the above hypotheses, we employ linear regression with cluster-robust standard errors. The general model is as described below.
The dependent variable Scaled_BTD it is a proxy for tax avoidance. We employ the book-tax differences scaled by total assets. This is calculated as the company's i booktax differences, which equal book profit less taxable profit, scaled by total assets. Book (or accounting) profit is pre-tax (or gross) profit in year t which we extracted from the set of financial data published by the Romanian Ministry of Public Finance for each reporting company. Taxable profit is not among the publicly available financial data, and this is the reason why we calculate it indirectly by dividing the profit tax payable in year t by the statutory flat tax rate of 16%. Last, we scale this difference by total assets and it results Scaled_BTD it (Badertscher et al., 2016;Mills & Newberry, 2001). We assume that large book-tax differences could be a useful indicator of the company's attempts to minimise their tax-related costs. Table 5 displays the descriptive statistics for the total sample, as well as for the two sub-samples of evading and compliant companies. When comparing the two groups, the descriptive statistics suggests that, on average, evading companies have a lower financial performance with data points more spread out (mean ROA 0.0930 and standard deviation 0.1685) than compliant companies (mean ROA 0.1215 and standard deviation 0.1444). However, the leverage measured by Debt-to-Assets Ratio is higher in the group of evading companies compared to the group of conforming companies (mean D_A_Ratio 0.6577, respectively 0.5959) while both groups have a similarly high standard deviation. With respect to size measured as natural logarithm of total assets, the two groups do not differ widely and, on average, are comparable in size. The dependent variable (the scaled book-tax difference) is negative on average in the full sample analysed over the five years, but positive and tending towards 0 in case of evading companies. Table 6 reports the Pearson and Spearman pairwise correlation coefficients between the variables. The size indicators (number of employees and net sales) are negatively correlated with the book-tax differences. Likewise, there is a negative correlation between the indicators of profitability (Return on Assets -ROA; Return on Equity -ROE) and scaled BTD. Moreover, the correlation matrix suggests correlations between the geographical region, the legislative milestone and the type of company on one hand and the book-tax differences on the other hand. Most importantly, no significant correlations exist between the independent variables employed later on in various models, which demonstrates that multicollinearity is not an issue for our regression analysis. Table 7 shows the results of examining the scaled book-tax differences for the sampled privately owned companies based in Romania and paying profit tax. We performed a linear regression analysis with cluster-robust standard errors. The sign ' Ã ' reads significant at 10%, ' ÃÃ 'significant at 5% and ' ÃÃÃ 'significant at 1%. The table includes the coefficients, the t-scores (in parenthesis), as well as root mean squared error (MSE) and the value of R 2 for each of the alternative linear regression models tested. Model A is the main model applied to the full sample of 1674 company-years. With regards to the link between size and book-tax differences, model A reveals a positive and significant association between total assets and scaled BTD, suggesting that the larger the company, the larger the scaled BTD, namely the more inclined towards tax avoidance that disclosing company is. The link between financial performance and scaled BTD consists in a significant but negative association. In other words, the more profitable a company is (namely the higher its ROA), the less inclined towards tax avoidance that company is (the lower its scaled BTD).

Main findings
The leverage ratio seems to have a significant and positive impact upon the booktax differences, thus showing that, as expected, it is negatively associated with tax avoidance. As hypothesised, it matters in which industry the company operates in. Significant is also the association between the dependent variable and the geographical region: companies located outside Bucharest have lower BTDs, suggesting that entities based around Romania's capital city have a more prominent tendency towards tax avoidance. Our hypothesis that Romanian companies are more tax-avoidant in the years following the legislative changes in Romania in 2016 does not hold.
What the full-sample model shows is that evading companies display larger booktax differences compared to companies that were not prosecuted nor found guilty of  tax evasion crimes. In other words, our hypothesis concerning the increased tax avoidance of evading companies compared to compliant ones was supported by the regression results.

Robustness checks
We performed several robustness checks to the main specification in the regression Model A. First, we extracted two random samples of 95% (1590 observations) and respectively 75% of the initial sample (1256 observations) and re-ran the regression (see Models B and C in Table 7). The results of these regressions are the same as those in the initial regression analysis (Model A). The values of R-Squared indicate that the variation in the variable 'scaled book-tax difference' can be attributed to the variation of the explanatory variables (total assets, ROA, and leverage) in a percentage that remains around 23% across these models (B and C) having controlled for region, industry, company type and legislative changes. Another robustness check was to modify the model specifications by replacing the size indicator (the natural logarithm of total assets) by the natural logarithm of the number of employees (see Model D in Table 7) and by the natural logarithm of net sales (see Model E in Table 7). Last, we have replaced the indicator of financial performance (Return on Assets -ROA) by another similar indicator of profitabilitythe Return on Equity (ROE)see Model F in Table 7. All alternative regression models (B through F) were found valid and their parameters were significant (with a p-value below 0.01). The R-Squared indicates that on average approximately 22% of the variation in the scaled book-tax differences is explained by the dependent variables (size, financial performance and leverage), having controlled for geographical region, industry, company type (evading or compliant) and the legislative turnaround in 2016.

Discussions and conclusions
Our research offers a relevant and up-to-date picture of Romanian companies that were payers of profit tax in recent years. Our sample contains 236 evading and 236 compliant companies, adding up to 1674 year-observations from the period 2013-2017. We chose to compare the set of 236 evading companies with another set of 236 compliant ones selected randomly so that the structure by industry of the two groups remains the same. For the purposes of our research, we defined compliant companies as companies that have never been prosecuted for tax evasion crimes (according to the Law 241/2005 on the prevention and fighting of tax evasion) and existed between 2013 and 2017. We analysed the data contained in the financial statements submitted by the selected companies to the Romanian Ministry of Public Finance. Our results show that the disclosure discipline of evading companies fades with every year that passes. On the contrary, compliant companies submit their financial statements regularly within the five-year period. In the fiscal year 2013, 190 evading companies made their financial statements public, and their number drops to only 75 evading companies in 2017. Only about one fifth of our sample of evading companies has been consistent in submitting their financial statements in each of the five years analysed. The reason is that such companies have a lower motivation in providing financial data to the Romanian state and to their other stakeholders.
Our main findings are in agreement with the political power theory (Belz et al., 2019;Gupta & Newberry, 1997;Siegfried, 1972;Stickney & McGee, 1982). This theory has been first described by Siegfried (1972), who argues that larger companies are able to influence the political process in their favor to a greater extent than small companies (Stickney & McGee, 1982). In particular, large companies can use their resources and power to negotiate their tax burden or influence legislation in their favour (for instance, through lobbying activities), resulting in lower taxes payable compared to those of smaller companies (Gupta & Newberry, 1997;Siegfried, 1972;Stickney & McGee, 1982). Indeed, the results of the regression analyses reveal a positive and significant association between book-tax differences scaled by total assets and several indicators of size (total assets, number of employees, net sales). The larger the company, the more inclined towards tax avoidance it is.
We also find a significantly negative association between financial performance and book-tax differences scaled by total assets, which actually suggests the positive impact of performance upon the tax-avoidant behaviour. The higher the profitability of a company, the less inclined towards tax avoidance that company is. Our results are in line with prior studies such as Higgins et al. (2015), Allen et al. (2016) and Majeed and Yan (2019). We also note a negative relationship of leverage with tax avoidance which was highlighted by previous research (Allen et al., 2016;Chyz et al., 2013;Drucker, 2006;Majeed & Yan, 2019;Mills & Newberry, 2005). Surprisingly, the fiscal regulations amended starting with 2016 did not lead to an apparent exacerbation of tax avoidance among Romanian profit tax payers. These amendments increased the range of deductible expenses that could diminish the profit tax and, in our opinion, could have encouraged tax avoidance. Consequently, we can only conclude that, irrespective of the legal environment, companies in Romania are consistent in their attempts to minimise their profit tax liabilities.
Our investigation was subject to several limitations. First, our sample comprises only privately owned companies which are payers of profit tax and does not include entities from the public sector or payers of tax on the income of micro-entities, as defined by the Romanian legislation. Second, we only use publicly available data, put at the disposal by the Ministry of Public Finance. The publicly available financial data offers the following level of detail: assets are split into non-current assets, current assets, and prepaid expenses, while just the sub-category 'current assets' is further detailed into inventory, receivables, and cash. Total debt plus equity comprises liabilities, provisions, equity (and separately owner's equity), as well as prepaid income. Net sales, revenues, expenses, gross profit/loss, and net profit/loss are the figures made available from the profit and loss statement. This limited data also limited our choice of the dependent variable as a proxy of tax avoidance. Despite this, a major plus of our study is the fact that it does not focus exclusively on listed companies, as the majority of studies in the field, but offers a wider view on privately owned companies of all sizes.
Our study is relevant not only for legislators and tax authorities, but also for companies operating in Romania and for potential and actual investors interested in the status-quo of the business environment over the past few years in Romania. Moreover, the results of the present study could be relevant for other countries from South-Eastern Europe and similar studies applied to this geographical area may very well lead to comparable conclusions. In our view, the future in this field of research lies in the combination of both economic and noneconomic factors in order to gain a deeper understanding of tax avoidance in the corporate world.