Instructors: Marco Bee
Class hours: 36 (6 ECTS)
The course aims at introducing the main time series forecasting tools used in business-oriented organizations. In particular, the program is focused on time series regression models, time series decomposition and exponential smoothing. The approach is data-driven, i.e. we will introduce only the theory necessary for the analysis of real datasets. At the end of the course the student is expected to be able to (i) estimate a multiple time series regression model and interpret the results; (ii) predict a time series by means of an appropriate forecasting method and interpret the results; (iii) master the R software.
My main goal in teaching this class consists in transmitting to the students the belief that apparently complicated statistical techniques are not only mathematical sophistications, but also tools with a tremendous impact in practical applications, and that they are important assets for professional success. In my opinion, a manager who masters complex statistical methods has a better understanding of the processes and projects going on in an organization.
I am convinced that the most effective way of learning quantitative methods is based on problem solving. Accordingly, I will assign regular homeworks that have a threefold advantage:
1. they mimick a situation that the students will often encounter in their career: they will have to solve a problem within a deadline;
2. they encourage group work: even though the students will be individually responsible for the solution, hopefully they will realize that working together can produce a better result in a shorter time;
3. they will give the students a motivation for a thorough understanding of the methods presented in class.
Some preliminary knowledge of basic mathematics and statistics is assumed, at the level of Newbold, P., Carlson, W. and Thorne, B. (2010), Statistics for Business and Economics, 7th edition, Prentice Hall.
The concept of forecast and the forecaster's toolbox: time series graphics, simple forecasting methods, preliminary transformations, evaluation of forecast accuracy. Time series regression models: least-squares estimation, goodness of fit, selection of predictors, nonlinear regression. Time series decomposition: component factors of the classical multiplicative time series model, moving averages, using moving averages to extract the seasonal component. Exponential smoothing: simple exponential smoothing, the Holt-Winters approach, forecast computation and evaluation.
Standard classroom lectures, class laboratories and regular homework assignments. All the techniques shall be implemented in class using real datasets by means of the R software. The exercises are mostly focused on R, whose knowledge is essential for a complete understanding of the topics of this course.
Evaluation of learning
The final mark is computed as follows: M=0.7*FE+0.3*HW, where FE is the mark of the final written exam and HW is the average of the marks obtained in the homework assignments. The written exam consists in solving three/four problems that, broadly speaking, belong to three possible categories:
1. Open questions about any topic treated in class;
2. Comments to the R output obtained as the solution of a real-data problem.
3. Multiple choice questions.
Finally, I may adjust the result taking into account class participation.
Required text: Hyndman, R.J. and Athanasopoulos, G. (2018), "Forecasting: principles and practices", second edition. Available online at OTexts.org/fpp2/. The R codes available online at the same address will be used, commented and extended in class.
Additional datasets, R codes and homeworks will be made available during the course at www.comunitaonline.it.
As said above, the practical application of the quantitative techniques presented in class requires appropriate software in order to be fully effective. Accordingly, both class lectures and homework are firmly based on the R software, which is an open source environment for statistical computing and graphics, freely available for download at www.R-project.org.