CMStatistics 2015: Start Registration
View Submission - CMStatistics
B0998
Title: Semiparametric modeling of intraday temperature data Authors:  Iryna Okhrin - Dresden University for Technology (Germany) [presenting]
Yarema Okhrin - Universitaet Augsburg (Germany)
Wolfgang Schmid - European University Viadrina (Germany)
Abstract: The modeling of weather data differs from the modeling of other type of data like financial, biological, medical data. Temperature have property of daily and annual seasonality. Parametric and semiparametric procedures are derived for modeling this type of data. The parametric approach is based on the truncated Fourier series. The extraction of both daily and annual seasonal factors results stationary time series. This approach provides a pretty but not perfect forecast for temperature at any time of a year and of a day. The curve of temperature behaves similarly for all days. This property motivates the Shape Invariant Modeling. The parameters of the model, which are estimated for each day separately, are by theirself a multivariate time series. This procedure is quite flexible and allows a forecasting. The application of Dynamic Semiparametric Factor Models provides good results, too. This method is parametric in time and nonparametric in space and can be adapted to our type of data. The application of these three approaches to real data shows an advantage of semiparametric modeling.