The Covid-19 pandemic created one of the most substantial negative exogenous shocks in decades, forcing firms to rapidly adapt. This brief examines an adjustment mechanism that played a significant role in Latvia, and potentially in other countries in Eastern and Central Europe. Specifically, we focus on the role of envelope wages as a buffer for absorbing the shock. Our analysis demonstrates that this form of tax evasion indeed acted as a cushion during the Covid-19 pandemic. Our results indicate that, in the short run, tax-evading firms experienced smaller employment losses in response to the Covid-19 shock compared to compliant firms.
The Covid-19 pandemic generated one of the largest negative, exogenous shocks in decades. To absorb this shock, firms had to swiftly adapt. Prior literature has demonstrated that firms responded by reducing employment and investment (Lastauskas, 2022; Fernández-Cerezo et al., 2023; Buchheim et al., 2020). In this brief, we discuss another margin of adjustment – potentially important for many countries in the region. We focus on the role of envelope wages as a buffer for negative shock absorption.
Envelope wages is a widespread form tax evasion, in which, for employees that are formally registered, a portion of their salary (often at the minimum wage level) is reported to tax authorities, while the remaining ‘envelope’ portion is paid unofficially. The prevalence of this phenomenon has been extensively documented in Eastern and Central Europe (see Kukk and Staehr (2014) and Paulus (2015) for Estonia, Gorodnichenko et al. (2009) for Russia, Putniņš and Sauka (2015) for the Baltic States, Tonin (2011) and Bíró et al. (2022) for Hungary).
In addition to the evident objective of reducing tax obligations, a primary incentive for firms to employ this evasion scheme is the extra flexibility it provides. The unreported portion of wages operates outside of the legal framework, offering firms a means of adaptation in the face of production restrictions, supply chain disruptions, and overall substantial uncertainty caused by the Covid-19 pandemic. In this brief, we argue that firms utilizing envelope wages reduced their employment less than compliant firms during the pandemic in Latvia.
Identifying Firms That Pay Envelope Wages
We identify firms that paid (at least partly) their employees in cash before the pandemic using a rich combination of Latvian administrative and survey data and the methodology proposed by Gavoille and Zasova (2021).
The idea is as follows: We use a subsample of firms for which we can assume that we know whether they pay envelope wages and, using this subsample, train an algorithm that is capable of distinguishing compliant and evading firms based on their observed characteristics and reported financials.
Following Gavoille and Zasova (2021), we use firms owned by Nordic investors as a subsample of tax-compliant firms. To obtain a subsample of non-compliant firms, we combine data on administrative (i.e., reported) wages with several rounds of Labor Force Survey data in order to spot employees who are paid suspiciously little given their personal characteristics (education, experience, etc). Firms employing these employees form the subsample of evading firms. Using these samples of compliant and evading firms, we train a Random Forest algorithm to classify firms according to their type. We then use the algorithm to classify the universe of firms used in this study. Table 1 shows the classification results.
Table 1. Classification results: share of tax-evading firms and employees.
We find that almost 40 percent of firms (employing about 20 percent of employees) underreport at least some of their workers’ wages. The cross-sectoral heterogeneity is consistent with survey evidence: the construction and transport sectors are the sectors with the highest prevalence of envelope payments. Comparing the share of tax-evading firms with the share of workers working within these firms also indicates that on average, tax-evading firms are smaller than tax-compliant ones. This is yet again in accordance with survey evidence.
Employment Response During Covid-19
Figure 1. Average firm-level change in employment during the Covid-19 pandemic.
The Covid-19 crisis had a severe impact on Latvia. The government declared a state of emergency as early as March 13, 2020, which entailed significant restrictions on gatherings and on-site work, leading to a six-fold increase in the proportion of remote workers within a matter of months.
During the second wave, in Autumn 2021, Latvia had the highest ranking in the world in terms of new daily positive cases per capita. A substantial number of firms were directly affected by the pandemic (see Figure 1).
We study firm-level employment response at a monthly frequency in compliant and tax-evading firms, from January 2020 to December 2021. Our empirical approach is in the spirit of Machin et al. (2003) and Harasztosi and Lindner (2019), who study the effect of minimum wage shocks. In essence, this approach consists of a series of cross-section regressions, where the dependent variable is the percentage change in employment in a firm between a reference period (set to January 2020) and any subsequent month until December 2021. Our key interest is the difference in cumulative employment response between tax-compliant and evading firms, controlling for a set of (pre-pandemic) firm characteristics, such as the firm’s age, average profitability, average export share, and average labor share over the 2017-2019 period.
The Aggregate Effect
Figure 2 shows the estimated coefficients that measure the difference between employment effects in compliant and tax-evading firms, aggregate for all sectors. Period 0 denotes our reference period, i.e., January 2020, while the estimated coefficients in other periods show the cumulated difference between tax compliant and tax-evading firms in the respective period relative to January 2020 (e.g., the estimated coefficient in period 10 shows the cumulated differential employment response in October 2020 vis-à-vis January 2020).
We document a noticeable difference in the employment response between the two types of firms starting in April 2020. The positive coefficient associated with evading firms indicates that the change in employment growth was not as negative in evading firms as in compliant firms (see Figure 2). Labor tax-evading firms exhibit, on average, a less sensitive employment response than tax-compliant firms. In March 2021, the point estimates are about 0.025, implying that compared to March 2020, tax-evading firms contracted, on average, 2.5 percentage points less than compliant ones. This difference however fades over time and turns insignificant (at the 95 percent level) about halfway through 2021.
Figure 2. Evasion and total employment.
Differences by Sector
Figure 3 displays the estimated difference in employment response, disaggregating the sample by sector. We show the results for two sectors: trade and transportation. These two sectors exhibited the most significant differences in employment response between evading and non-evading firms.
For trade, evading firms have been able to maintain employment losses at approximately 5 percentage points less than compliant firms (see Figure 3(a)). This is consistent with the envelope wage margin mechanism. Contrary to the aggregate results, the difference in employment response does not fade over time. This suggests that this margin is not a shock absorber only in the very short run.
The decrease of the evader effect at the aggregate level is caused by negative point estimates of the evasion indicator in the transportation sector, starting in the first quarter of 2021 (see Figure 3(b)). In this sector, evading firms have on average experienced a larger employment decline in 2021 than compliant firms.
Figure 3. Employment effect – by sector.
The outcome in the transportation sector is likely influenced by the taxi market. There were two major changes in 2021 that particularly affected taxi drivers receiving a portion of their remuneration through envelopes. Firstly, amendments to State Revenues Service’s (SRS) regulations made it more difficult to underreport the number of taxi trips, as each ride was now automatically recorded in the SRS system through taxi apps. Secondly, commencing in July, legal amendments mandated a minimum social security tax, which had to be paid based on at least the minimum wage. Given that many taxi drivers work part-time, and that those associated with evading firms tend to underreport their rides, this new requirement was more binding for evading firms. Additionally, there was a significant shift of taxi drivers to the food delivery sector, where demand for driver services surged during the pandemic.
Our results indicate that employment losses in response to the Covid-19 shock were smaller in tax-evading firms than in compliant firms in the short run. We also demonstrate that by the end of 2021, the discrepancy between the two types of firms had disappeared. This can be explained by significant heterogeneity in employment responses across sectors.
These findings contribute to our understanding of the pandemic’s impact on the size of the informal sector. Despite tax-evading firms generally having more restricted access to finance, the added flexibility provided by unreported wages may have increased their resilience to the negative shock.
This brief is based on a forthcoming working paper COVID-19 Crisis, Employment, and the Envelope Wage Margin. The authors gratefully acknowledge funding from EEA and Norway, grant project “Micro-level responses to socio-economic challenges in face of global uncertainties” (Grant No. S-BMT-21-8 (LT08-2-LMT-K-01-073)).
- Bíró, A., Prinz, D. and Sándor, L. (2022). The minimum wage, informal pay, and tax enforcement. Journal of Public Economics, 215, 104728.
- Buchheim, L., Dovern, J., Krolage, C. and Link, S. (2020). Firm-level Expectations and Behavior in Response to the COVID-19 Crisis. CESifo Working Paper No. 8304
- Fernández-Cerezo, A., González, B., Izquierdo Peinado, M. and Moral-Benito, E. (2023). Firm-level heterogeneity in the impact of the COVID-19 pandemic. Applied Economics 55(42), 4946-4974.
- Gavoille, N. and Zasova, A. (2021). What we pay in the shadows: Labor tax evasion, minimum wage hike and employment. SSE Riga/BICEPS Research paper No.6.
- Gorodnichenko, Y., Martinez-Vazquez, J. and Sabirianova Peter, K. (2009). Myth and reality of flat tax reform: Micro estimates of tax evasion response and welfare effects in Russia. Journal of Political Economy 117 (3), 504-554.
- Harasztosi, P. and Lindner, A. (2019). Who pays for the minimum wage? American Economic Review, 109, 2693–2727.
- Kukk, M. and Staehr, K. (2014). Income underreporting by households with business income: evidence from Estonia. Post-Communist Economies, 26(2), 257-276.
- Lastauskas, P. (2022). Lockdown, employment adjustment, and financial frictions. Small Business Economics 58(2), 919-942.
- Machin, S., Manning, A. and Rahman, L. (2003). Where the minimum wage bites hard: Introduction of minimum wages to a low wage sector. Journal of the European Economic Association, 1, 154–180.
- Paulus, A. (2015). Tax Evasion and Measurement Error: an Econometric Analysis of Survey Data Linked with Tax Records. ISER Working Paper Series 2015-10.
- Putniņš, T. and Sauka, A. (2015). Measuring the shadow economy using company managers. Journal of Comparative Economics, 43(2), 471-490.
- Tonin, M. (2011). Minimum wage and tax evasion: Theory and evidence. Journal of Public Economics, 95(11-12), 1635-1651.
Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.
Despite an increase in women’s representation since the beginning of the 20th century, women remain underrepresented in academia and other high-skilled professions. Academia has been prone to gender disparities both within and across fields as well as across academic ranks. In an endeavour to examine and address the underrepresentation of women in the academic profession, the Centre of Economic Analysis (CenEA), together with the Stockholm Institute of Transition Economics (SITE) and other partners of the Forum for Research on Gender Economics (FROGEE) at the FREE Network, organized the two-day conference “The playing field in academia: Why are women still underrepresented?”, in Warsaw June 21-22, 2023. This brief offers insights from the presentations and panel discussions held at the conference.
To date, there are few, if any, high-skilled professions exhibiting gender balance, and academia is no exception. Consequently, this imbalance has been subject to increased multidisciplinary research attention, exploring its origins and potential remedies. However, attaining a comprehensive understanding of gender disparities remains a challenge. For instance, much remains to be learnt about their long-run dynamics, a subject addressed by Carlo Schwarz, in one of the conference’s keynote lectures.
A Century of Progress
Carlo Schwarz (in joint work with Alessandro Iaria and Fabian Waldinger, 2022) trace the evolution of gender gaps in academia across a variety of domains at the global level throughout the 20th Century. Facilitated by an unprecedentedly large database of nearly 500,000 academics, spanning 130 countries and supplemented by publication and citation data, the authors specifically examine gender imbalances in recruitment, publishing, citation patterns, and promotions.
They find that in 1900 women constituted roughly 1 percent of all hires in academia (226 women, with only 113 hired as full professors). By 1969 the share of female academics had risen to about 6.6 percent, and by the year 2000 it had grown to approximately 17 percent. These rates varied across disciplines, institutions, and countries. For instance, teaching-centric disciplines such as pedagogy and linguistics, exhibited higher representation relative to research-oriented ones.
The research subsequently reveals a hump-shaped evolution of the gender gap in academic output – starting small before peaking at 45 percentage points fewer publications by women in 1969, thereafter declining to 20 percentage points. These publication disparities were also found to share a U-shaped relationship with the share of women in academia, indicating the interconnectedness of gender gaps.
The authors also address gender gaps in citations, identified by the use of a novel machine learning approach, forecasting a paper’s citations had it been written by a man. The results indicate a progressive reduction in the citation gap during the 20th century, decreasing from 27 percentage points (pre-WW1) to 14 percentage points (interwar) and eventually to 8 percentage points (post-WW2) fewer citations of papers by female relative to male academics. These gender gaps in academic output reiterated current evidence from Mexico, presented at the conference by Diana Terrazas-Santamaria, showing that women are associated with lower citation rates. Terrazas-Santamaria attribute the low rates to gender differences in both the number of publications and duration of academic careers.
The work by Iaria, Schwarz and Waldinger (2022) further showcase the gender disparities in career advancement in academia, which similarly decreased over the years. At the point of the greatest gender disparity, women required an approximately 6 percentage points better publication record to have the same promotion probabilities as their male counterparts.
The Leaky, Dry Pipeline
In the conference’s second keynote, Sarah Smith highlighted how academia, much like other professional occupations, exhibits a leaky pipeline. This is a phenomenon characterized by a declining representation of women as they ascend through the academic hierarchy. When examining specific fields, Smith’s results indicate that the gender disparities in economics much more closely align with those observed in STEM fields (science, technology, engineering, and mathematics) than other social science disciplines. Furthermore, the economics’ field illustrate a significant lack of diversity among its new entrants. This phenomenon, referred to as the dry pipeline, generates future cohort implications, as they result in less demographically representative cohorts from which future professors can be recruited (see Stewart et al., 2009).
The cross-disciplinary comparison of the dry pipeline addressed in the keynote, contest the mathematical rigor of economics as a barrier to entry, as mathematics itself demonstrated higher women representation at A-level and undergraduate levels. In a following discussion panel, which focused on ensuring a fair start in academia (comprised of Yaroslava Babych, Alessandra Casarico, Federica Braccioli and Marta Gmurek, and moderated by Maria Perrotta Berlin), the panellists acknowledged that deeply engrained social expectations, gender trained behaviours and a lack of awareness constitute some of the persistent hindrances to the (early) involvement of women in specific fields, and the academic profession in general.
Additional factors influencing the gender balance in recruitment and promotion are gendered references, and the presence or absence of shared research interests between candidates and recruitment panels. These themes were extensively investigated in the work presented by Alessandra Casarico on the conference’s opening day. Specifically, results from collaborative work with Audinga Baltrunaite and Lucia Rizzica, highlight that grindstone words (e.g., “determined”, “hardworking”, etc.) are frequently used in recommendation letters to describe female candidates, while standout words (e.g., “excellent”, “strongest” etc.) typify male candidates’ references. Compared to their male counterparts, women are also shown to be more inclined to accentuate personality traits when serving as referees. This added to a broader literature demonstrating that female candidates’ recommendation letters frequently exhibit brevity, raise doubts, carry a weak tone, and emphasize candidates’ interpersonal skills and personality traits rather than their ability. Moreover, separate results from Casarico’s work (with Piera Bello and Debora Nozza) illustrate that research similarity between the recruiting committee and the candidate predict the likelihood of recruitment. The authors argue that the relationship is indicative of a bias against women if – as shown by the authors – women are less likely to be the candidates with the highest similarity.
In her presentation, Anne Sophie Lassen offered a different factor that may contribute to the attrition in the pipeline: the influence of parenthood on academic careers. Results from her work (with Ria Ivandić) indicate that while parenthood does not significantly influence graduation rates, it extends doctoral studies by an average of 7 months for women. Moreover, Lassen highlighted a declining trend of remaining in academia after becoming a parent, particularly pronounced among women.
More Areas of Imbalance
The remaining conference presentations and panel discussions explored additional domains of gender imbalances within academia. Iga Magda showcased evidence from her joint work with Jacek Bieliński, Marzena Feldy and Anna Knapińska of gender differences in remuneration during the early stages of an academic career, substantiating a gap within a year of graduation. These disparities endure throughout respondents’ careers and are contingent on the field of study – largest among engineering and technology graduates and lowest among those from the humanities and arts fields. Furthermore, it was observed that productivity plays a negligible role in the identified pay gaps, as its impact is similar for both genders.
The panel composed of Eleni Chatzichritou, Marta Łazarowicz-Kowalik, Jesper Roine and Joanna Wolszczak-Derlacz, and moderated by Michał Myck, deliberated on exposed disparities in the application for, and the success rates in attaining research funding in Poland and Europe – as seen in the National Science Centre (NCN) and the European Research Council research grants, respectively. The discussion highlighted how quantitative measures used in the allocation of research funding are riddled with subjective criteria that often benefit male academics. They also recognized how quests to allocate funds to the most successful candidate inadvertently penalize women with career breaks.
Another panel including Lev Lvovskiy, Carlo Schwarz, Sarah Smith, Marieke Bos and Joanna Tyrowicz, and moderated by Pamela Campa, lauded the growing objective data shedding light on gender inequalities in academia. The panellists discussed current challenges in identifying and quantifying aspects of gender disparities. For instance, currently used proxies do not allow to capture more subtle disparities, like microaggressions faced by female academics from students – emphasizing the need for more individual level survey data.
The panels were further enriched by personal anecdotes and filled with retrospective advice shared by both early career and established academics. To contextualize the above, a few cases from the FREE Network countries follow.
Evidence From Within the FREE Network
Yaroslava Babych shared insights concerning women in higher education in Georgia and other countries of the South Caucasus. Preliminary findings of her study confirm the presence of gender inequality in academia, evident in disparities in access to higher education as well as gender segregation across both fields and countries. Notably, women comprise a majority of the graduates in bachelor’s and master’s of art programs, whereas higher research-level programs such as doctors of science, and top echelons of the academic hierarchy remain predominantly male. Moreover, female academic output is found to be lower than that of male counterparts.
Lev Lvovskiy discussed the case of Belarus, highlighting the influence of the Soviet legacy. A significant factor linked to this legacy is exploiting university enrolment to circumvent compulsory conscription of men, allowing male university admissions to serve a secondary purpose beyond acquiring knowledge. This increases the perceived opportunity cost of enrolling a woman. Lvovskiy further documented the academic trajectories of Belarusians, revealing a majority of women at college and doctoral levels, but being underrepresented among doctoral graduates. The results further indicate significant cross-disciplinary gender disparities, with humanities having close to 80 percent women representation and engineering and information and technology (IT) fields having less than 30 percent women representation.
Monika Oczkowska provided evidence of gender disparities in Poland. Findings from the country reveal an overrepresentation of women graduates from bachelor through doctoral levels, and relative parity at post-doctoral level, but lower proportions at habilitation, associate professor, and professor levels. These general results confirm the higher detail findings presented by Karolina Goraus-Tanska on the first day of the conference. Results from Goraus-Tanska’s work (with Jacek Lewkowicz and Krzysztof Szczygielski) suggest that the drop-off among female academics from habilitation levels is not attributed to higher output expectations for women, but rather stems from the impact of parenthood.
Oczkowska further demonstrated that female academics in Poland are characterized by fewer international collaborations and lower levels of international output. Polish female academics were also showcased to engage in more international mobility during their doctoral studies relative to men, with the converse holding true after obtaining a doctoral degree. A potential explanation for this mobility decline among female academics, could be the increased burden of familial responsibilities at the post-doctoral and higher levels. Moreover, fewer women were reported to have applied for NCN grants and were underrepresented among the beneficiaries of these calls. Lastly, female academics in Poland record significantly lower total project costs relative to their male counterparts.
‘Plugging’ the Leak
In light of the aforementioned, what measures can be taken to address the gender imbalances in academia? As summarized by Sarah Smith, early initiatives have involved tracking women representation (e.g., in admissions, progression, hiring, etc.) within departments and/or institutions to identify where in the pipeline their progress is impeded. Attempted initiatives include formulation of seminar guidelines to overcome unfair experiences, as well as using gender-blind recruiting and objective hiring criteria to equalize hiring opportunities. Some other efforts, such as diverse recruitment panels have been unsuccessfully adopted, as they seem to embolden hostile male recruiters and load female panellists with unrewarded administration tasks. Conversely, mentoring has helped women build networks, publish more, and advance professionally. Awareness raising campaigns have reduced disparities in teaching evaluations and remain vital in addressing the dry pipeline and both transparent workload allocation and rewarding of administrative tasks have been shown to reduce promotion gaps in academia. In addition to the above, initiatives such as fostering gender-neutral networking opportunities, collaborations and a more diverse faculty were also deliberated during the conference.
The conference advanced dialogue on societal and structural constraints to gender equality in academia and provided a platform to exchange ideas on how the shared objective of a more inclusive and equitable academic environment can be achieved. While the challenges remain abundant, and the costs associated not always negligible, it remains crucial to assess achievements, such as those resulting from mentoring and awareness intervention initiatives and recognize that further opportunities to enhance equity within the profession exist.
Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.
Labor tax evasion is a major policy issue that is especially salient in transition and post-transition countries. In this brief, we use firm-level administrative data, tax authorities’ audit data and machine learning techniques to detect firms likely to be involved in labor tax evasion in Latvia. First, we show that this approach could complement tax authorities’ regular practices, increasing audit success rate by up to 35%. Second, we estimate that about 30% of firms operating in Latvia between 2013 and 2020 are likely to underreport the wage of (some of) their employees, with a slightly negative trend.
Tax evasion is a major policy issue that is especially salient in transition and post-transition countries. In particular, “envelop wage”, i.e., an unofficial part of the wage paid in cash, is a widespread phenomenon in Eastern Europe (European Commission, 2020). Putnins and Sauka (2021) estimate that the share of unreported wages in Latvia amounts to more than 20%. Fighting labor tax evasion is a key objective of tax authorities, which face two main challenges. The first is to make the best use of their resources. Audits are costly, so the choice of firms to audit is crucial. The second challenge is to track the evolution of the prevalence of labor tax evasion. For this purpose, most of the existing literature relies on survey data.
In our forthcoming paper (Gavoille and Zasova, 2022), we propose a novel methodology aiming at detecting tax-evading firms, using administrative firm-level data, tax authorities’ audit data and machine learning techniques.
This study provides two main contributions. First, this approach can help tax authorities to decide which firms to audit. Our results indicate that the audit success rate could increase by up to 20 percentage points, resulting in a 35% increase. Second, our methodology allows us to estimate the share of firms likely to be involved in labor tax evasion. To our knowledge, this paper is the first to provide such estimates, which are however of primary importance in guiding anti-tax evasion policy. We estimate that over the 2013-2020 period, about 30% of firms operating in Latvia are underreporting (at least some of) their workers’ wages.
The general idea of our approach is to train an algorithm to classify firms as either compliant or tax-evading based on observed firm characteristics. Tax evasion, like any financial manipulation, results in artifacts in the balance sheet. These artifacts may be invisible to the human eye, but machine learning algorithms can detect these systematic patterns. Such methods have been applied to corporate fraud detection (see for instance Cecchini et al. 2010, Ravisankar et al. 2011, West and Bhattacharya 2016).
The machine learning approach requires a subsample of firms for which we know the “true” firm behavior (i.e., tax-evading or compliant) in order to train the algorithm. For this purpose, we propose to use a dataset on tax audits provided by the Latvian State Revenue Service (SRS), which contains information about all personal income tax (PIT) and social security contributions (SSC) audits carried out by SRS during the period 2013-2020, including the outcome of the audit. The dataset also contains a set of firm characteristics and financial indicators, covering both audited and non-audited firms operating in Latvia (e.g., turnover, assets, profit). Assuming that auditors are highly likely to detect misconduct (e.g., wage underreporting) if present, audit outcomes provide information about a firm’s tax compliance. Firms sanctioned with a penalty for, say, personal income tax fraud are involved in tax evasion, whereas audited-but-not-sanctioned firms can be assumed compliant. The algorithm learns how to disentangle the two types of firms based on the information contained in their balance sheets. Practically, we randomly split the sample of audited firms into two parts, the training and the testing subsamples. In short, we use the former to train the algorithm, and then evaluate its performance on the latter, i.e., on data that has not been used during the training stage. If showing satisfying performance on the training sample, we can then apply it to the whole universe of firms and obtain an estimate of the share of tax-evading firms.
In this study, we successively implement four algorithms that differ in the way they learn from the data: (1) Random Forest, (2) Gradient Boosting, (3) Neural Networks, and (4) Logit (for a review of machine learning methods, see Athey and Imbens, 2019). These four data mining techniques have previously been used in the literature on corporate fraud detection (see Ravisankar et al. 2011 for a survey). Each of these four algorithms has specific strengths and weaknesses, motivating the implementation and comparison of several approaches.
Table 1 provides the out-of-sample performance of the four different algorithms. In other words, it shows how precise the algorithm is at classifying firms based on data that has not been included during the training stage. Accuracy is the percentage of firms correctly classified (i.e., the model prediction is consistent with the observed audit’s outcome). In our sample, about 44% of audited firms are required to pay extra personal income tax and social security contributions. This implies that a naive approach predicting all firms to be evading would be 44% accurate. Similarly, a classification predicting all firms to be tax compliant would be correct in 56% of the cases. This latter number can be used as a benchmark to evaluate the performance of the algorithms. ROC-AUC (standing for Area Under the Curve – Receiver Operating Characteristics) is another widespread classification performance measure. It provides a measure of separability, i.e., how well is the model able to distinguish between the two types. This measure is bounded between 0 and 1, the closer to 1 the better the performance. A score above 0.8 can be considered largely satisfying.
Table 1. Performance measures
Random Forest is the algorithm providing the best out-of-sample performance, with more than 75% of the observations in the testing set correctly classified. Random Forest is also the best performing model according to the ROC-AUC measure, with performance slightly better than Gradient Boosting.
Our results imply that a naive benchmark prediction is outperformed by almost 20 percentage points by Random Forest and Gradient Boosting in terms of accuracy. It is important to emphasize that this improvement in performance is achieved using a relatively limited set of firm-level observable characteristics that we obtained from SRS (which is limited compared to what SRS has access to), and that mainly come from firms’ balance sheets. This highlights the potential gain of using data-driven approaches for the selection of firms to audit in addition to the regular practices used by the fiscal authorities. It also suggests a promising path for further improvements, as in addition to this set of readily available information the SRS is likely to possess more detailed limited-access firm-level data.
Share of Tax-Evading Firms Over Time and Across NACE Sectors
We can now apply these algorithms to the whole universe of firms (i.e., to classify non-audited firms). Figure 1 shows the share of firms classified as tax-evading over the years 2014 to 2019 for our two preferred algorithms – Gradient Boosting and Random Forest. Random Forest (the best performing algorithm) predicts that 30-35% of firms are involved in tax evasion, Gradient Boosting predicts a slightly higher share (around 40%). Both algorithms, especially Random Forest, suggest a slight reduction in the share of tax-evading firms since 2014.
Figure 1. Share of tax-evading firms over time
The identified reduction, however, does not necessarily imply that the overall share of unreported wages has declined. In fact, existing survey-based evidence (Putnins and Sauka, 2021) indicate that the size of the shadow economy as a share of GDP remained roughly constant over the 2013-2019 period, and that there was no reduction in the contribution of the “envelope wages”. With our method, we are estimating the share of firms likely to be involved in labor tax evasion. Unlike the survey approach, our methodology does not allow the measurement of tax-evasion intensity. In other words, the share of non-tax compliant firms may have decreased, but the size of the envelope may have increased in firms involved in this scheme.
Next, we disaggregate the share of tax-evading firms by the NACE sector. Figure 2 displays the results obtained with Random Forest, our best performing algorithm.
Figure 2. Share of tax-evading firms by NACE, based on Random Forest
First, the sector where tax evasion is the most prevalent is the accommodation/food industry, where the predicted share of tax-evading firms is 70-80%. Second, our results indicate that the overall decrease in the share of firms likely to evade is not uniform. It is mostly driven by the accommodation/food and manufacturing sectors. Other sectors remain nearly flat. This highlights the fact that labor tax evasion varies both in levels and in changes across sectors.
We show that machine learning techniques can be successfully applied to administrative firm-level data to detect firms that are likely to be involved in (labor) tax evasion. Machine learning techniques can be used to improve the selection of firms to audit in order to maximize the probability to detect tax-evading firms, in addition to the regular practices already used by SRS. Our preferred algorithms – Random Forest and Gradient Boosting – outperform the naive benchmark classification by almost 20 percentage points, which is a substantial improvement. Once implemented, the use of these tools can improve the audit effectiveness at virtually no extra cost.
Our findings also suggest a promising path for further improvements in the application of such methods. The improvement in predictive power achieved by our proposed algorithm is attained by using a limited set of variables readily available from the firms’ balance sheets. Given that SRS is likely to have access to more detailed firm-level information that cannot be provided to third parties, there is clear room for improving the performance of the algorithms by using such limited-access data.
Acknowledgement: The authors gratefully acknowledge funding from the Latvian State Research Programme “Reducing the Shadow Economy to Ensure Sustainable Development of the Latvian State”, Project “Researching the Shadow Economy in Latvia (RE:SHADE)”; project No VPP-FM-2020/1-0005.
- Athey, Susan, and Guido Imbens. 2019. “Machine Learning Methods That Economists Should Know About.” Annual Review of Economics 11: 685–725.
- Cecchini, Mark, and Haldun Aytug, and Gary J. Koehler, and Praveen Pathak, 2010. “Detecting management fraud in public companies“. Management Science 56, 1146-1160.
- European Commission, 2020. “Undeclared Work in the European Union. Special Eurobarometer 498” (Report)
- Gavoille, Nicolas and Anna Zasova, 2022. “Estimating labor tax evasion using tax audits and machine learning”, SSE Riga/BICEPS Research papers, forthcoming.
- Putnins, Talis, and Arnis Sauka, 2021. “Shadow Economy Index for the Baltic Countries 2009–2020” (Report), SSE Riga
- Ravisankar, Pediredla, and Vadlamani Ravi, and Gundumalla Raghava Rao, and Indranil Bose, 2011. “Detection of financial statement fraud and feature selection using data mining techniques“. Decision Support Systems, 50(2), 491-500.
- West, Jarrod, and Maumita Bhattacharya, 2016. “Intelligent financial fraud detection: a comprehensive review“. Computers & security, 57, 47-66
Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.