scientific article; zbMATH DE number 1542250
Time series, auto-correlation, regression, etc. in statistics (GARCH) (62M10) Reasoning under uncertainty in the context of artificial intelligence (68T37) Research exposition (monographs, survey articles) pertaining to game theory, economics, and finance (91-02) Economic time series analysis (91B84)
This book studies forecasting of short time series. A time series is defined to be short if the number of observations is less than the minimum required to achieve a given accuracy of forecasts. This is compensated for by the use of expert information where experts give opinions about future values of the time series in the form of inequalities (these may be between future values at different horizons, such as ``the series will be monotone over the next 5 time units). The models considered are trend plus noise where the trend is linear in the unknown parameters while the time functions may be arbitrary. The basic idea is that the expert information is of higher priority than the fit. The problem is formulated as a minimization problem where the loss function, representing the fit, is minimized over a subset of the parameter space consistent with the expert information. The various problems considered in the book are best seen from its table of contents (abbreviated here): Chapter~1: Short time series and their description; Chapter~2: Forecasting under consistent expert judgements; Chapter~3: Forecasting under inconsistent (contradictory) expert judgements; Chapter~4: Model selection; Chapter~5: Combining models and predictions. In each section, there is an illustrative example and a summary in algorithmic form of the estimation/model selection method considered in it.
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4520148)