CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
Time series exponential smoothing optimal parameters finding for forecasting of currency exchange rate
 |
Read | Pilyugina A.V., Boiko A.A. Time series exponential smoothing optimal parameters finding for forecasting of currency exchange rate // Caspian journal : management and high technologies. — 2016. — №1. — pp. 114-127. |
Pilyugina A.V. - Ph.D. in Economics, Associate Professor, Bauman Moscow State Technical University, 5/1, 2-ya Baumanskaya St., Moscow, 105005, Russian Federation, pilyuginaanna@mail.ru
Boiko A.A. - post-graduate student, Bauman Moscow State Technical University, 5/1, 2-ya Baumanskaya St., Moscow, 105005, Russian Federation, boiko_andrew@mail.ru
Time series exponential smoothing using R.G. Brown and J.S. Hunter models is analyzed. Optimization problems are formulated for following objects: type of exponential smoothing model; smoothing parameter ; number of first time series cases for counting initial smoothed value. Time series of daily exchange rate of the US dollar and Euro to the Russian Federation rouble from 1 January 2009 to 31 December 2015 are used as bench-mark data. In addition to daily course time series three modified time series are considered. They are monthly average exchange rate, time series with missing values in which missing values are filled up with previous value and time series with missing values in which missing values are filled up with interpolated values using previous and following cases. As optimization criterion following objects are used: mean square error (MSE), root mean square error (RMSE), mean average percentage error (MAPE). It is shown that J.S. Hunter model provides smaller value of MAPE. For that model optimal smoothing parameter value is equal to 1 in most cases and initial smoothed value is equal to average of first ten cases of original time series. Grid search is considered as an optimum procedure for exponential smoothing optimal parameters finding.
Key words: валютный курс, прогнозирование, временной ряд, экспоненциальное сглаживание, модель Брауна, модель Хантера, параметр сглаживания, поиск на сетке, квадрат средней квадратичной ошибки (MSE), средняя квадратичная ошибка (RMSE), средняя относительная ошибка п