| Title |
Graph-theory based identification of company groups for tax evasion risk |
| Authors |
Ruzgas, Tomas ; Armonaite, Karolina |
| DOI |
10.15388/DAMSS.16.2025 |
| ISBN |
9786090712009 |
| Full Text |
|
| Is Part of |
16th conference on data analysis methods for software systems (DAMSS), 27-29 November 2025, Druskininkai, Lithuania.. Vilnius : Vilnius University press, 2025. p. 111-112.. ISBN 9786090712009 |
| Keywords [eng] |
graph theory ; tax evasion ; company groups |
| Abstract [eng] |
The identification of economically related groups of companies is an important element in the assessment of tax evasion risks. This study proposes and applies a graph-theoretic approach to identify such groups using data from tax registers and company ownership networks. Unlike traditional classifications based on legal or administrative criteria, the proposed method defines company groups as structural units derived from interconnected ownership links between legal entities. The methodology is based on the construction of a directed graph, where each node represents a company, and the edges represent ownership relationships with assigned weights corresponding to ownership percentages. Three graph-based grouping algorithms are explored and evaluated. The first identifies weakly connected components (WCCs) in the graph, which ensures maximal inclusion of all indirectly related companies. The second and third approaches apply different graph filtering techniques: one uses eigenvector centrality-based filtering, while the other removes edges below a specified ownership threshold, helping to eliminate insignificant or formal links. The empirical analysis is based on a dataset of legal entities with shareholder information from Lithuania’s tax registry. The study analyses the resulting group structures formed by each method, comparing the number of groups, their sizes, and structural characteristics. The WCC-based approach yields fewer, larger groups with high node connectivity, while threshold-based methods produce more granular clusters with stronger internal ownership links. Groups are further analysed through the centrality of nodes to identify key companies within each group. The proposed methodology enables visual representation of ownership networks and supports the detection of business groups that may operate in a coordinated manner for tax planning or evasion purposes. Visual analysis of selected groups illustrates the effectiveness of centrality filtering in revealing economically meaningful substructures, such as hubs and branching ownership paths. Overall, the study demonstrates the feasibility and interpretability of graph-theoretic techniques for identifying company groups. These approaches provide useful insights for auditors, analysts, and policymakers when assessing the complexity of inter-company relationships and detecting structures that may increase the risk of non-compliance or tax base erosion. |
| Published |
Vilnius : Vilnius University press, 2025 |
| Type |
Conference paper |
| Language |
English |
| Publication date |
2025 |
| CC license |
|
