Applying Machine Learning Techniques to Estimate the Size of the Romanian Shadow Economy

Ivan Andreea-Daniela (), Davidescu Adriana Anamaria (), Agafiţei Marina-Diana () and Geambaşu Maria Cristina ()
Additional contact information
Ivan Andreea-Daniela: Bucharest University of Economic Studies, Romania
Davidescu Adriana Anamaria: Bucharest University of Economic Studies/National Scientific Research Institute for Labour and Social Protection, Bucharest, Romania
Agafiţei Marina-Diana: Bucharest University of Economic Studies/National Scientific Research Institute for Labour and Social Protection, Bucharest, Romania
Geambaşu Maria Cristina: Bucharest University of Economic Studies, Bucharest, Romania

Proceedings of the International Conference on Business Excellence, 2025, vol. 19, issue 1, 2525-2541

Abstract: The research investigates the underground economy in Romania, an essential phenomenon that affects tax revenue, competitive conditions in the marketplace, and economic stability. This empirical research estimates the informal economy using the most recent machine learning algorithms, with a focus on support vector regression (SVR) In our application, support vector regression (SVR) provided estimates of the coefficients needed for estimating the currency demand approach (CDA) and measuring the size of unobserved economic activities. The literature has examined different forms of measurement of the informal economy: direct approaches, indirect approaches, and statistical approaches such as the CDA and multiple-indicator multiple-cause (MIMIC) model. The classical methods have a number of issues, such as the numerous assumptions and poor data availability. Artificial intelligence (AI) or Machine learning (ML) can offer even more potentially accurate measures than what the classical methods have offered in the past. This paper integrates measures of economic indicators, such as cash in circulation, tax revenues, inflation and unemployment rates, with measures of digitalisation including numbers of cards, ATMs and POS terminals. In addition to SVR, other machine learning algorithms, such as Random Forests and Gradient Boosting, were estimated in order to evaluate which type of model produces the best predictive models of liquidity demand. Overall, the evidence establishes that SVR is a clear improvement over standard methodologies because it models complex and nonlinear relationships in the data. The research illustrates the acute fluctuations registered by Romania’s underground economy from 2000 to 2023, thus illustrating how economic crises and changes in fiscal policies dictate the amount of informal economic activity. The paper suggests improved calculation methods resulting in better estimates of natural economic activity and provides ongoing information for economic forecasters and policymakers wishing to maintain the level of informal activities at acceptable levels.

Keywords: underground economy; machine learning; Romania; artificial intelligence; currency demand; informal sector (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
https://doi.org/10.2478/picbe-2025-0196 (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:vrs:poicbe:v:19:y:2025:i:1:p:2525-2541:n:1023

DOI: 10.2478/picbe-2025-0196

Access Statistics for this article

Proceedings of the International Conference on Business Excellence is currently edited by Alina Mihaela Dima

More articles in Proceedings of the International Conference on Business Excellence from Sciendo
Bibliographic data for series maintained by Peter Golla ().