Статистика и экономика (Jul 2019)
Neural models in diagnostics of the financial result of housing and utility enterprises
Abstract
The aim of the research is the usage of an artificial neural network as a tool not only for forecasting, but also for operative diagnostics of a financial state through combining deterministic and stochastic factors in one model. This circumstance expands the possibilities of effective influence on the formation of an acceptable level of the company’s financial condition in various activities. The proposed universal model is presented in the article in relation to the company’s characteristics in the housing and utilities sector.The article proposes a method for diagnosing the level of the housing and utility company’s financial condition based on the use of a factor neural model of the financial results of their activities.Materials and methods. The neural network modeling methodology allows you to create models that have several advantages: learning ability (they adapt to various changes); universality (able to solve a wide range of data analysis and processing tasks); speed (process various data in parallel mode); ease of use (easy to operate after training); fault tolerance (resistant to local damage to the neural network structure and external noise).One of the main tasks that neural networks successfully solve is the problem of classification – the assignment of the sample to one or several predefined classes. Most often, the input sample is determined by the input data vector. The components of this vector are the various characteristics of the sample. The classifier in the form of a neural network relates the object to one of the classes in accordance with the partitioning of the N-dimensional input space. The number of components of the vector determines the dimension of this space. In the context of this article, the input sample is the financial condition of the organization at a particular point in time. The input vector that characterizes the sample includes a set of direct and indirect factors of the financial results of a housing and utility company. The neurons of the output layer are a set of different classes. In the course of operation, the neural network assigns to each input vector a neuron in the output layer. The significance of the input data can be adjusted using connections between neurons and changing the neural network architecture. Neural networks can have a complex architecture when different parts of the neural network include different numbers of connections and different neurons.The article develops the ideas laid down by its authors in [7, 8], where a neural network of direct propagation and a way of learning with a teacher have already been used. The model, described below, has been modified due to the authors’ desire to improve it, as well as dictated by the specifics of the housing and utility companies: a list of key indicators has been developed that affect not the financial result, but, consequently, the financial condition of companies in this sector of the Russian economy; the number of input factors characterizing the input sample was increased, each direct factor or group of direct factors was supplemented with an indirect factor; direct and indirect factors explaining the same processes are combined into clusters that influence the corresponding neuron; the number of neurons in the output layer has been expanded, the number of classes has been increased, the data are classified by means of the neural network in more detail; in the course of the program, it is possible to select the period to which the input data (month, quarter, half year, year) belong. The additions made a positive impact on the work of the neural network. The accuracy of attributing the input sample to a specific cluster and the sensitivity of the neural network has increased. The number of clusters has grown up to 50. Innovations have increased the usability of the program. New interface allowed to analyze data monthly. The programmatic way of interpreting the data has changed due to the fact that not all input data changes depending on the period.
Keywords
