LFU:online

This course gives an introduction to modern probabilistic forecasting methods and applications using flexible semi-parametric modeling techniques, amongst others. By the end of the course, the students should be able to work on complex probabilistic modeling questions themselves and gain a good overview of the specialist literature.

In predictive modeling, state-of-the-art statistical models need to forecast future outcomes as accurate as possible. Moreover, in modern applications it is crucial to provide full probabilistic forecasts, which are able to assign probabilities to each predicted outcome. In the advent of machine learning applications, except in a few fields, e.g., in weather, energy and economic models, probabilistic forecasting has not received too much attention yet, although frameworks and software for probabilistic forecasting exist over about two decades. One reason might be, that classic statistical textbooks mainly focus on modeling the mean instead of the full response distribution.

 - Repetition of the linear regression model.

- Introduction to generalized additive models (GAM).

- Algorithms for model choice and variable selection for GAMs.

- Introduction to distributional regression models.

- Probabilistic calibration checks.

- Bayesian distributional regression models.

- Boosting and LASSO type regularization for distributional regression models.

- Distributional forests.

- Quantile regression and quantile regression averaging.

- Dimension reduction methods.

Each of the sessions starts with a lecture in the morning followed by a discussion. Afterwards, the methods learned are applied to problems using the statistical programming environment R (https://www.R-project.org/) and the obtained results are discussed in detail. After each session, a homework assignment will be given, which must be presented at the beginning of the next applied session. At the end of the course there will be a forecasting project, which covers 60% of the achievable points. All homework cover 40% of the achievable points.

Corresponding references will be mainly presented in the slides. Recommended textbooks are:

* Wood SN (2006). Generalized Additive Models: An Introduction with R. Chapman & Hall/CRC.

* Fahrmeir L, Kneib T, Lang S (2013). Models, Methods and Applications. Berlin: Springer-Verlag.

* Stasinopoulos M, Rigby RA, Heller GZ, Voudouris V, Bastiani F (2017). Flexible Regression and

    Smoothing: Using GAMLSS in R. Chapman and Hall/CRC.

* Koenker R (2005). Quantile Regression (Econometric Society Monographs). Cambridge: Cambridge

    University Press. DOI:10.1017/CBO9780511754098.

Prerequisite(s): Profound knowledge from statistical data analysis, in particular linear regression. Basic knowledge of variable and model selection. Solid basic knowledge in programming with R.

TERMINE finden im Besprechungsraum des Instituts für Statistik, Sowi, 2. Stock Ost, statt.

Mittwoch, 10.4.2019: 8:00 - 16:00 Uhr

Donnerstag, 2.5.2019: 8:00 - 16:00 Uhr

Mittwoch, 15.05.2019: 8:00 - 16:00 Uhr