ANALYSIS OF MATHEMATICAL MODELS USED FOR ECONOMETRICAL TIME SERIES FORECASTING

• Main Author: D. A. Petrusevich
• Format: Article
• Language: Russian
• Published: MIREA - Russian Technological University 2019-05-01
• Series: Российский технологический журнал
• Subjects: russian longitudinal monitoring survey; rlms; quantiles; real salary index; arima; stationarity; time series; prediction; forecast; akaike informational criterion
• Online Access: https://www.rtj-mirea.ru/jour/article/view/149

In the paper changes of the Russian citizens’ welfare are explored. The time lapse of the data is: 2000-2018. In the first part of the paper the representative individual samples of “The Russian Longitudinal Monitoring Survey - Higher School of Economics (RLMS-HSE)” data of the 2008-2017 time period are analyzed. The 10%-95% quantiles of the salary have been constructed with the regard to year inflation, and their behavior has been analyzed. In the second part the monthly wage index based on the dynamic series of macroeconomic statistics of the Russian Federation data (2000-2018) has been explored. The mathematical models of the wage of this time lapse have been presented. They are based on the ARIMA (p, d, q) models with d≤5, p≤5. Forecasts of these models have been compared to predictions of the models with parameters p = 6 or q = 6. The constructed models have made better forecast than the automatically fitted ARIMA model with d≤5, p≤5. They have been compared using two metrics, and also the Akaike information criterion (AIC) has been considered. The seasonal factors of the wage index have been taken into account. It has been shown that the lags of 6 and 12 months are connected to the today wage index; there are maxima of this value situated at the end of the year or in summer. It’s explained with the vacations which traditionally take place in summer, and also officially held vacations in January. The further research is going to target the trend - seasonal - noise decomposition of time series. Statistical packages which are often in use have got different methods to compute the ARIMA coefficients. That fact is also going to be under research.