Odeychuk A. Methods and information technology for nonstationary time series forecasting with risk assessment in information systems

This thesis is devoted to solving a pressing research and practice task of developing methods and information technology for nonstationary time series forecasting with risk assessment that enable the forecast error to be reduced in information systems. The target of research is processes of nonstationary time series forecasting in information systems. The thesis employs simulation modeling methods, multi-dimensional statistical analysis methods and methods of optimization theory. A simulation model for generating heteroscedastic time series has been improved. The model enables to set a value of changing conditional variance of time series and to analyze the GARCH models. A nonstationary time series forecast risk assessment method has been improved that enables to perform a forecast risk assessment through the development of nonstationary time series volatility model, forecasting volatility and normalization. A method for synthesis of nonstationary time series forecasting models by concatenating basic forecasting methods that employs a procedure of optimization for concatenation of basic forecasting methods has been further developed. A generalized criterion for efficiency of forecasting models has been further developed that enables to lower levels of uncertainty in selecting the efficient forecasting model. The proposed methods and information technology have been implemented in software and their efficiency has been proven through practical application at the NSC KIPT, Ukrsotsbank PJSC, and in the KNURE learning process. The thesis results can be used during development of forecasting services for information systems.