Разработка и исследование моделей временных данных со смешанными частотами на примере анализа урожайности зерновых (Кристина Пивоварова, ISPRASOPEN-2018)
Материал из 0x1.tv
- Кристина Пивоварова
The problem of econometric models specifications, as well as the problem of analyzing mixed-frequency data, arises in many issues related to the modeling of processes and real-world phenomena, in particular — the modeling of yields. The main problem is that despite the variety of ready- model specifications, it is necessary to rework factors of influence in order to apply them to a specific region, therefore, each time it’s necessary to re-verify and refine the model specification. In addition, it is often difficult to make a choice between models on panel data and an aggregate set of time series analysis models.
The research methodology proposed allows us to determine the specification of models faster and more qualitatively; the developed software allows conducting research on various sets of initial data, obtaining qualitatively high results with an automatically selected specification of models.
In the framework of this paper, we considered the singular spectral analysis algorithm as an indicator of seasonal components and the trend in time series of yield regressors, as well as we considered models on mixed-frequency data using MIDAS models and construction of models that take into account panel structure of the original data.