Teoria dos leilões nas contratações públicas paraibanas - uma estimação dos custos de transações através do aprendizado de máquina

Abstract

This master’s thesis aims to fill a specific gap in the literature by investigating transaction costs in public procurement, using mathematical models and machine learning techniques. The personal and practical motivation for this research arises from the need to improve efficiency and transparency in public tenders, particularly in the state of Paraíba, where the optimization of public resources is of utmost importance. The main objective of this study is to develop and apply a mathematical model to analyze public procurements and use machine learning techniques to predict the transaction costs that impact public pricing. The research is divided into two main parts. The first part is dedicated to developing a mathematical model (game-theoretic), adapting classic auctions to the most common public procurement modalities, such as Competitive Bidding and Auction. This section mathematically explores the impact of participants’ strategies and behaviors on auction outcomes, focusing on the presence of transaction costs, including entry prices. In the second part, machine learning techniques are applied to predict transaction costs in public procurements, using data such as invoices from public entities in the state of Paraíba, as well as economic, geographical, social, and accounting information. The methods include the use of Random Forest and LASSO to create predictive models, aiming to estimate procurement prices more accurately. The research results indicate that the Random Forest model presented a coefficient of determination (R2) of 0,97, explaining about 97% of the variability in transaction costs, with a root mean squared error (RMSE) of 0,14 standard deviations of normalized prices. The analysis revealed that factors such as Average Payment Time and the timeframe for fulfilling judicial debts (precatórios) are crucial determinants of transaction costs. These results show that it is possible to predict transaction costs in public procurements with high precision using advanced machine learning techniques. In conclusion, the practical implications of the research are highlighted, such as the possibility of implementing predictive models to improve the management of public procurements, promoting greater efficiency and transparency in the use of public resources. The interdisciplinary approach adopted, which combines statistics, economics, mathematics, computer science, and public administration, reflects the complexity and relevance of the topic, offering practical and theoretical tools to enhance public procurement processes.