METHODOLOGICAL APPROACHES TO FORECASTING ENTERPRISE PROFITS

Abstract

The article considers the issues of increasing the reliability of assessing the level of net profit at enterprises producing bakery products. The influence of 21 factors on the dependent variable (the amount of net profit of the enterprise) was studied. In this study, we built single-factor and multifactor regression models using the least squares method. The multicollinearity of the factors was tested using the Ferrara- Glauber test. Taking into account the presence of multicollinearity between the factors, the values of the factors were transformed using the principal components method to build multifactor models using the least squares method. To build separate single-factor models, we used transformations based on the decimal and natural logarithm. The internal parameters of the models were tested using the least squares method by calculating their p-values. The models were compared using the following indicators: coefficient of determination, Fisher's criterion, average approximation error. All models built using the least-squares method were excluded from further research for various reasons (checking the significance of factors, statistical significance, approximation accuracy). At the next stage of the study, it was decided to try to build multivariate models for estimating the level of net profit using the methods of k-nearest neighbors and binary decision tree. Before applying these methods, the values of the factors were standardized. Based on the results of comparing the quality of the obtained models, four of the five models were eliminated: one model was eliminated due to the low value of the coefficient of determination, and three models due to the low accuracy of the approximation. Based on the results of comparing the quality of the obtained models, four of the five models were eliminated: one model was eliminated due to the low value of the coefficient of determination, and three models due to the low accuracy of the approximation. The regression model based on a binary decision tree passed all quality checks and can be recommended for predicting net profit at bakery enterprises. The average approximation error of the model is 2.58%.