How does the intorduction of leniency influence cartel activity? : evaluation of legal texts using Natural Language Processing

Baumann, Rebecca

doi:10.21256/zhaw-29119

Please use this identifier to cite or link to this item: https://doi.org/10.21256/zhaw-29119

Publication type:	Master thesis
Title:	How does the intorduction of leniency influence cartel activity? : evaluation of legal texts using Natural Language Processing
Authors:	Baumann, Rebecca
Advisors / Reviewers:	Bellert, Nicole Pelli, Maria
DOI:	10.21256/zhaw-29119
Extent:	92
Issue Date:	2023
Publisher / Ed. Institution:	ZHAW Zürcher Hochschule für Angewandte Wissenschaften
Publisher / Ed. Institution:	Winterthur
Language:	English
Subject (DDC):	330: Economics 340: Law 410.285: Computational linguistics
Abstract:	This master's thesis looks at the impact of the European leniency program on detected cartel activity. The starting point is a study by Harrington and Chang, in which the authors conclude that while the introduction of a leniency program aims to reduce cartel activity by increasing the benefit of self-reporting, it can either reduce or increase the cartel rate depending on whether the program diverts resources from non-leniency enforcement to leniency cases, leading to a crowding-out effect. Derived from this, the following research question is explored: “How did the introduction of the leniency program of the European Commission in 1996 affect the cartel activity in the European Union?” Thereby, the change regarding non-leniency enforcement is also investigated. The research methodology of design science research is applied, following the three-cycle concept of design science proposed by Hevner, which consists of the cycles of relevance, rigour, and design. Primary data for the study is collected through web-scraping of publicly available resources from the European Commission and secondary data is accessed from an existing manually compiled dataset. Concerning data extraction directly from the scraped prohibition decisions, different Natural Language Processing techniques are employed, and their accuracy rates analysed. Applying an empirical, quantitative crosssectional survey approach including longitudinal observations, the data collected allows for a comprehensive analysis of temporal trends and patterns across the different cartel entities. Statistical data analysis including panel and multiple linear regression is used to test the hypotheses derived from the research question and its sub-questions. The results show a nuanced impact of the European Commission’s leniency program on cartel activity. After the introduction of the leniency program in 1996, there was a modest but statistically significant decline in the detection of cartel activity. This decline was also observed in cases detected through non-leniency enforcement after the program's intro-duction. Nevertheless, detection rates after 1996 through leniency enforcement were not significantly lower than detection rates for non-leniency enforcement prior to 1996, suggesting a likely reallocation of resources from non-leniency enforcement to prosecution of cases based on leniency applications. However, the introduction of the leniency pro-gram did not have a significant impact on the duration of detected cartels. Some data even suggest a potential increase in cartel duration, but this finding is not statistically sup-ported. The relationship between the average amount of the imposed fines and the duration of cartels is also not found to be significant. Overall, the results paint a complex IV picture of the impact of the European Commission’s leniency program on cartel detection and duration. Analysing various Natural Language Processing techniques showed variable effective-ness based on the type of data extracted. Regular expressions demonstrated a strong performance in identifying case numbers and decision dates, with an impressive 97.87 % and 100 % accuracy rate respectively, though this was heavily dependent on customizing patterns to match the distinct document formats presented by the prohibition decisions. Key-word matching proved to be efficient in detecting instances in which there was an application for leniency, achieving a 90.43 % accuracy rate. Combining Named Entity Recognition, keyword matching and regular expressions delivered mixed results, especially in pinpointing the start and end dates of cartels, which constituted the most difficult data extraction task. Overall, careful selection and combination of Natural Language Processing techniques is vital to meet specific data extraction needs. Future research should expand the scope of the research conducted in this thesis to include cases published in other languages than English, which would help to mitigate selection bias and offer a more comprehensive view of the impact of the leniency program on cartel activity. Further work should also explore different methods or additional data sources to address the limitations of this study. To enhance analysis, Natural Language Processing techniques could be refined and advanced models, such as BERT or Transformers, could be evaluated. This could improve data extraction, especially for important information like formal decision of the European Commission, the names of cartel members and sec-tor information, which would lead to a deeper understanding of the impact of the Euro-pean Commission’s leniency program regarding possible differential impacts in dissimilar industries and could provide insights on repeat offenders.
URI:	https://digitalcollection.zhaw.ch/handle/11475/29119
License (according to publishing contract):	CC BY-NC-ND 4.0: Attribution - Non commercial - No derivatives 4.0 International
Departement:	School of Management and Law
Organisational Unit:	Institute of Business Information Technology (IWI)
Appears in collections:	MSc Wirtschaftsinformatik

Files in This Item:

File	Description	Size	Format
2023_Rebecca-Baumann_MSc_IWI.pdf		1.67 MB	Adobe PDF	View/Open

Show full item record

Baumann, R. (2023). How does the intorduction of leniency influence cartel activity? : evaluation of legal texts using Natural Language Processing [Master’s thesis, ZHAW Zürcher Hochschule für Angewandte Wissenschaften]. https://doi.org/10.21256/zhaw-29119

Baumann, R. (2023) How does the intorduction of leniency influence cartel activity? : evaluation of legal texts using Natural Language Processing. Master’s thesis. ZHAW Zürcher Hochschule für Angewandte Wissenschaften. Available at: https://doi.org/10.21256/zhaw-29119.

R. Baumann, “How does the intorduction of leniency influence cartel activity? : evaluation of legal texts using Natural Language Processing,” Master’s thesis, ZHAW Zürcher Hochschule für Angewandte Wissenschaften, Winterthur, 2023. doi: 10.21256/zhaw-29119.

BAUMANN, Rebecca, 2023. How does the intorduction of leniency influence cartel activity? : evaluation of legal texts using Natural Language Processing. Master’s thesis. Winterthur: ZHAW Zürcher Hochschule für Angewandte Wissenschaften

Baumann, Rebecca. 2023. “How Does the Intorduction of Leniency Influence Cartel Activity? : Evaluation of Legal Texts Using Natural Language Processing.” Master’s thesis, Winterthur: ZHAW Zürcher Hochschule für Angewandte Wissenschaften. https://doi.org/10.21256/zhaw-29119.

Baumann, Rebecca. How Does the Intorduction of Leniency Influence Cartel Activity? : Evaluation of Legal Texts Using Natural Language Processing. ZHAW Zürcher Hochschule für Angewandte Wissenschaften, 2023, https://doi.org/10.21256/zhaw-29119.