Robustness in Statistical Forecasting
Heidelberg/Dordrecht/New York/London: Springer, 2013. 356 P.
Traditional procedures in statistical forecasting of time series, which are proved to be optimal under the hypothetical model, are often not robust under relatively small distortions (misspecification, outliers, missing values, etc.), leading to the actual forecast risks (mean square errors of prediction) which are much higher than the theoretical values.
This monograph fills a gap in the literature on robustness in statistical forecasting giving solutions to the following topical problems:
- development of mathematical models and descriptions of typical distortions in applied forecasting problems;
- evaluation of the robustness for traditional forecasting procedures under distortions;
- obtaining of the maximal distortion levels allowing for “safe” use of the traditional forecasting algorithms;
- construction of new robust forecasting procedures to provide the risk that is not much sensitive to definite distortion types.
The book is primarily intended for mathematicians, statisticians and software developers in applied mathematics, computer science, data analysis, operation research, econometrics, financial engineering and medicine. It can also be recommended as a textbook for a one-semester course for advanced undergraduate and postgraduate students of the mentioned disciplines.