Description“All models are wrong, but some are useful”. In practice, all models suffer from built-in imperfections, an inevitable consequence of imperfect understanding of the underlying system that generated the observations. Given a “best” available model and historical observations, how should we go about improving future forecasts? Forecast correction is an important part of a forecasting framework. In forecasting mode, the model imperfections often result in systematic errors, which can be detected and, potentially, corrected. In this project the students will learn how to construct a forecasting system with a set of procedures designed to detect and correct systematic errors in its forecasts. The corresponding empirical experiments will be conducted based upon a nonlinear time series and a nonlinear model. Linear statistical models are popular tool for time series analysis and forecasting. In reality, however, the intrinsic dynamics of the system are often governed by the nonlinear paradigm, where linear approaches are hampered by their linear assumptions. Unlike linear models, when one put a Gaussian uncertainty through the nonlinear model, one will get non-Gaussian forecast error. Nonlinear models will certainly introduce extra challenge to this project. But the students will gain extra experience of analyzing and modeling nonlinear time series.
PrerequisitesCalculus and Probability I, Statistical Concepts IIResources
|
email: Hailiang Du