Course teached as: B031816 - STATISTICAL METHODS FOR FORECASTING AND QUANTITATIVE MARKETING Second Cycle Degree in STATISTICS AND DATA SCIENCE Curriculum GENERALE
Teaching Language
English
Course Content
Module A (Magrini): Statistical models to analyse and forecast multivariate time series, with applications in economics and business.
Module B (Nikiforova): Optimal designs to build choice experiments; random utility models for the analysis of choice experiments. Particular attention will be devoted to applied and practical aspects, by considering real case-studies and the last developments in the literature
Module A (Magrini):
• R. J. Hyndman, G. Athanasopoulos (2021). Forecasting: principles and practice. OTexts, 3rd edition, Melbourne, AU. https://OTexts.com/fpp3
• J. H. Stock, M. W. Watson (2019) Introduction to econometrics. Pearson, 4th edition, London, UK.
• B. H. Baltagi (2011) Econometrics. Springer-Verlag, 5th edition, Berlin, DE. [Chapters 1-8, 14]
Module B (Nikiforova):
• K. Train (2009). Discrete choice methods with simulation. Cambridge University Press, 2nd ed., Cambridge, UK. [Chapters 2, 3, 4, 6, 10]. https://eml.berkeley.edu/books/choice2.html
• D. Hensher, J. Rose, W. Greene (2005). Applied Choice Analysis. Cambridge University Press, Cambridge, UK. [Chapters 5, 6, 10, 14, 15, 16].
Slides and further teaching materials will be supplied by the lecturers.
Learning Objectives
The teaching is divided into two modules. Module A (Magrini) has the objective to provide the methodological basis of multivariate time series analysis in economics and business. Module B (Nikiforiva) aims at providing a deep knowledge on the theory and application of choice experiments for the study of the consumers’ behaviour in the field of quantitative marketing.
Prerequisites
Module A (Magrini): Basic knowledge of the multiple linear regression model, and of R programming.
Module B (Nikiforova): Statistical inference; Statistical models (logit model).
Teaching Methods
Lectures and practical laboratory sessions on real case studies.
Type of Assessment
The exam consists, separately for each module, in the presentation through slides of a project carried out at home individually, followed by an oral examination on the topics covered during the teaching. The final grade will be given by the average of the grades obtained in each of the two modules. Particular attention will be devoted to the student's constructive and critic skills, and to the ability to make use of the methods presented during the teaching.
Course program
Module A (Magrini)
• Brush up on the linear regression model and on OLS estimation.
• Stationarity, trends, unit root tests, methods for achieving stationarity.
• The linear regression model with time series data: OLS estimation, inference, prediction.
• Autoregressive (AR), distributed lag (DL), and autoregressive distributed lag (ARDL) models.
• Model diagnostics: residual analysis, cross-validation to assess forecast accuracy.
• Estimation with correlated errors: Newey-West estimator, GLS, maximum likelihood.
• The vector autoregressive model (VAR), cointegration, error correction model (ECM).
• Management of seasonality, outliers and structural breaks.
• Hints to models for panel data.
Module B (Nikiforova):
• Introduction to multi-attribute evaluation methods, with a focus on choice experiments.
• Introduction on random utility theory and modelling, conditional logit model for the analysis of choice experiments and its limits.
• Optimal design theory for choice experiments (exact and approximated optimal choice designs) and design optimality criteria.
• Optimal choice designs for the conditional logit model.
• Basics on Bayesian and Semi-Bayesian optimal choice designs to determine a priori the parameters values.
• Nested logit model: estimation and interpretation of the results.
• Respondents’ heterogeneity: mixed logit model (optimal choice design, estimation and interpretation of the results).
• Heteroscedasticity of the alternatives: the Hetroscedastic Extreme Value (HEV) model (estimation and interpretation of the results).
• Notes on further statistical models for choice experiments (latent class models).