Системний підхід до моделювання та прогнозування на основі регресійних моделей і фільтра Калмана

• Видавець: The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
• Дата: 2017
• Автори: Shubenkova, Irina A., Petrova, Svitlana K., Bidyuk, Petro I.
• Формат: Стаття
• Мова: Ukrainian
• Опубліковано: The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2017
• Теми: regression model; Kalman filter; short-term forecast; dynamic and statistical estimations of forecasts; probabilistic and statistical methods; регрессионная модель; фильтр Калмана; кратковременный прогноз; динамические и статистические оценки прогнозов; вероятностно-статистические методы; регресійна модель; фільтр Калмана; короткочасний прогноз; динамічні та статистичні оцінки прогнозів; імовірнісно-статистичні методи
• Онлайн доступ: http://journal.iasa.kpi.ua/article/view/108763

A concept for adaptive modeling of financial and economic processes is proposed that is based upon simultaneous application of regression models and optimal Kalman filter for reducing the influence of stochastic disturbances and measurement errors on statistical data. Specialized software has been developed that is necessary for performing computational experiments. Several regression models were constructed for the selected financial and economic processes that were transformed to the state space representation. Testing of the software system developed using various data samples of financial and economic data showed that it was quite possible to reach an acceptable quality of short-term forecasting with the mean absolute percentage error of about 5–8 %. Depending on a specific problem statement, dynamic and static estimates of forecasts were used with an acceptable quality. An application of Kalman filter for preliminary data processing (reduction of the influence of external stochastic disturbances and measurement errors) and short term forecasting provides a possibility for further reduction of forecasting errors by about 1,5–2,0 %. In the future research, it is planned to develop a specialized decision support system for solving the problems of forecasting on the basis of probabilistic and statistical procedures.