Objectifs

This course capitalizes on the knowledge acquired in the Time Series and Forecasting class taught in Spring 2018. The students will learn different forecasting methods but also how to assess and combine them. The accent is made on practical situations and students will face live forecasting experiences. Upon the completion of this course, students will have an overview of the methodologies that can be employed for forecasting as well as a strong sense on how to interactively apply them to different problems while keeping track of the progress by documenting and making reports.

Contenus

This class is intended to present to the students a wide range of forecasting tools. Tentative list of topics that will be discussed in this class are listed below:

• Time series models: A broad range of “classical” models such as autoregressive models, ARMA, ARIMA, GARCH, … and their extensions to add external information (dynamic regression)
• Neural Networks: An introduction to neural networks for time serie data
• Model selection: Different ways to assess the performance of a model using back-testing, cross-validation, bagging, information criterion
• Model combination: Combining different forecast startegies and assess them
• Judgmental forecasting: Making forecast in the absence of historical data
• Web scraping: Automatic extraction of data from websites using SelectorGadget, rvest and quantmod
• Data visualizations: Exploratory data analysis with Base R and ggplot2
• Reporting: Performances reporting

The accent is made on practical aspects of forecasting and case studies will be presented on several occasions. Students will be required to participate in groups to “forecasting competitions”. Familiarity with a programming language is assumed. Within this class, we will use the statistical language R, but the students are welcome to use another programming language as long as it allows them to complete the different tasks of this class.

Références

Most of this class is based on the online textbook:

• Required: Forecasting: Principles and Practice by Rob J Hyndman and George Athanasopoulos

Pré-requis

It is strongly recommended that students follow the Time Series and Forecasting class prior to ours.

Familiarity with the R environment and basic mathematical background is assumed.

Notions of programming, as taught in the Programming tools in data science and Data science in business analytics classes will be a plus.

Evaluation

1ère tentative

Examen:

Sans examen (cf. modalités)  

Evaluation:

The learning outcomes are continuously assessed during the semester with forecasting competitions (group), reports (group), a presentation (group) and the participation (individual) (check this link for more details).

Rattrapage

Examen:

Sans examen (cf. modalités)  

Evaluation:

A second opportunity to pass the forecasting competitions and/or the reports and/or the presentation is proposed to the failing students/groups.