Modelo preditivo para despesas discricionárias sob restrição fiscal no ensino Superior: evidências a partir da UFPB

Abstract

Public budgeting is a central instrument of government planning; however, in Brazilian federal universities, its effectiveness is constrained by recurring fiscal restrictions, the authorizing nature of budget execution, and the persistence of incremental forecasting practices. In this context, the use of historical averages adjusted for inflation proves insufficient to capture the dynamics of discretionary expenditures. In light of this, this study analyzes the dynamics between institutional demand and budgetary constraints at the Federal University of Paraíba (UFPB) and develops a predictive model to estimate the monthly execution of these expenditures. The research is characterized as applied, with a quantitative approach and documentary research procedures, developed within the scope of a case study on UFPB. An integrated monthly database was built for the period from January 2015 to December 2025, bringing together budgetary, macroeconomic, institutional, and academic information, with committed expenditure as the dependent variable. Time series and machine learning models were compared, with emphasis on ARIMAX, SARIMAX, Random Forest, and XGBoost. The comparison among the algorithms revealed that the SARIMAX model presented the best predictive performance, recording a Mean Absolute Error (MAE) of R$ 7.91 million, a Root Mean Squared Error (RMSE) of R$ 9.97 million, and a Mean Absolute Scaled Error (MASE) of 0.55, outperforming the ARIMAX, Random Forest, and XGBoost models. Additionally, the feature importance analysis revealed that expenditure dynamics are strongly influenced by macroeconomic and academic variables, especially the exchange rate (14.4%), the National Construction Cost Index (INCC) (10.3%), and the number of student assistance beneficiaries (9.7%). It is concluded that budget planning should not rely solely on incremental parameters, but rather requires predictive tools capable of modeling the structural tension between growing institutional needs and severe external fiscal constraints.