THE MODELING OF DYNAMICS OF THE FINANCIAL INDEXES BASED ON THE “SHORT” SAMPLING
Oksana I. Snytiuk, Lesya Berezhna · Proceedings of Scientific Works of Cherkasy State Technological University Series Economic Sciences · 2018
Forecasting of future financial results of economic entities activity in the conditions of the small number of observation points because of the last years statistics inadequacy for modern conditions is an important and actual problem. That’s why the purpose of this work is the analysis of the existing approaches and methods of forecasting of the basic financial indexes (on the example of the National Bank of Ukraine (NBU)) on the short sampling, the comparison of the opportunities of Ordinary Least Squares (OLS), Group method of data handling (GMDH) and Artificial neural networks (ANN) and the offering of the suggestions of their using specifics.The comparative analysis of using results of OLS, GMDH and ANN testifies about the number of advantages and disadvantages of each of them. So, the OLS is the most studied, the estimates of significance of the model and coefficients were elaborated for it, but its using for solving the real problems has many obstacles due to the non-fulfillment of some preconditions. Besides this method has a low statistical significance for short sampling. So, in the studied model OLS testified its inadequacy and made impossible forecasting of the NBU income in the future.GMDH is an analytical method and it is able to make more accurate forecasting. The calculations made during the investigation confirm this. The use of regularity criterion for short-term forecasting and a small number of observation points allowed to get more accurate results than with OLS.The synthetic approach connected with using of ANN means the calculation of predictive values of the productive function without finding its analytical expression. The good selection of ANN kind, its activation function, the number of layers, the number of neurons in the hidden layer, the algorithm of functioning and enough number of the initial data provide the accurate study of ANN. At the same time its using gives the integrated results that aren’t limited by accounting of given number of factors and sometimes are the most accurate among the received results. The disadvantage of ANN and GMDH is the absence of the determination methods of the adequacy of model to the initial data and its statistical significance.So, three procedures of forecasting on the short sampling using three different methods and special preparatory procedures of Data Processing were examined and researched. The received results analysis points to the advantages and disadvantages of each method.