LOCAL BUDGET REVENUE ASSESSMENT

Abstract

The article is devoted to the urgent problem of increasing the reliability of assessing the level of income of local budgets. The work analyzed the nature of the existing dependence of the level of income of local budgets of Ukraine on the sources of their receipt. As factors affecting the amount of income of the local budget, the following were considered: tax revenues, non-tax revenues, income from capital transactions, trust funds and official transfers. The construction of a correlation matrix made it possible to identify a high level of dependence between local budget income and non-tax revenues together with targeted funds. The remaining sources of funding have a strong or medium impact on the income level of the local budget. A complex of linear and nonlinear one-factor regression models was built. A group of linear two-factor models were also built. The analysis of p-values made it possible to exclude models from further consideration in which internal parameters turned out to be insignificant. As a result of the study of the coefficient of determination, Fisher's test and the average absolute percent of error, a model was selected. The comparison showed that the best was the linear two-factor model for factors: tax revenues and official transfers. For previously selected combinations of factors, the machine learning method was used - a random forest. Machine learning-derived models were expected to have higher characteristics than the early regression models obtained. The best machine learning model obtained was a model with the same combination of factors as the best linear two-factor model. However, when comparing the values of the coefficient of determination and the average absolute percent of error, the model obtained as a result of applying the random forest algorithm turned out to be worse than the linear two-factor regression model. The developed linear two-factor model can be used to assess the level of income of local budgets of Ukraine and predict the impact on its size of various sources of financing. It is proposed to use other methods of machine learning with a teacher when building future models.