Title: Large panel time series forecasting using functional model
Authors: Mohammad Reza Yeganegi - Islamic Azad University, Central Tehran Branch (Iran) [presenting]
Abstract: Analysing Large panel data (panels with large number of corss-section units, $N$, and time periods, $T$) arises certain issues. Main challenges in large panel data analysis are Heterogeneity (i.e. fitting models with different parameters to each unit), Dynamics (i.e. using more complicated dynamic model since large amount of data is available for each unit, during time), Crosse-Section Dependency (i.e. modeling dependency between large number of units) and High Dimensionality (i.e. analysing the panel when $N$ is considerably larger than $T$). There are different approaches to address these issues in large panel data, the proposed models usually does not address all the issues in the same time. For instance, whilst multivariate time series models and multivariate filters (e.g. State-Space models and Kalman filter) are powerful tools to address first three issues, they are not originally designed for high dimensional data. The potentials of functional time series models, as an approach to address above issues in forecasting large panel data, are investigated.