Instructor: Bruce E. Hansen, Professor in the Department of Economics, University of Wisconsin-Madison, USA.

Course Objective

The course will cover a rigorous introduction to the theory and practice of econometric model selection, model averaging, and related methods. The goal is to introduce students to a range of tools which can be used in applied econometrics, to provide the theoretical foundation to understand how to use these tools, and to introduce the theory for students interested in pursuing methodological research.


Li Yijie , ASSEE 2017

The whole program reaches an excellent balance!

I learnt a lot about duration analysis, including both theories and how to do empirical analysis with it. The lab sessions are very helpful in improving my understandings of the contents of the lectures. The course is great, and the dinners and excursions are also great.

Day 1: Model Selection

An important practical issue for applied econometrics is how to select regressors. Our first lecture will cover a range of rigorous methods for regressor selection. Methods covered will include Kullback-Leibler information, the Akaike information criterion, the Bayes information criterion, the Takeuchi information criterion, the Mallows information criterion, hold-out evaluation, and the cross-validation criterion.

Day 2: Model Averaging

Model selection is a special case of model averaging. Rather than selecting one specific set of regressors (or model), a weighted average can be used. The critical issue is then how to select the weights. Methods covered will include Bayes weighting, Smoothed AIC, Mallows weighting, and cross-validation weighting.

Day 3: Forecast Combination

One of the classic applications of model averaging methods is for forecast combination. This lecture will review how model averaging methods can be constructive used to improve point forecasting.

Day 4: Stein Shrinkage

A classic method for model combination is due to James Stein and his shrinkage estimator. This estimator has been modernized for contemporary econometrics and can be employed for improved estimation precision. We will also review the related Ridge Regression estimator.

Day 5: Lasso & Ensemble Methods

The term “Machine learning” is a popular buzzword among economists. The most popular methods are the Lasso and Ensemble. The Lasso is a combination selection/shrinkage method and a close cousin of Ridge regression and Stein shrinkage. Ensemble methods are model averaging methods applied to machine learning algorithms. We will review these methods and put them in relative context.

Reading List:

“Lecture Notes: Model Selection” link

“Lecture Notes: Model Averaging” link

“Lecture Note: Shrinkage” link

"Least Squares Model Averaging," Econometrica, (2007) link

"Jackknife Model Averaging," with Jeffrey Racine, Journal of Econometrics, (2012) link

"Nonparametric Sieve Regression: Least Squares, Averaging Least Squares, and Cross-Validation,"  Oxford Handbook of Applied Nonparametric and Semiparametric Econometrics (2014) link

"Minimum Mean Squared Error Model Averaging in Likelihood models,” with Gerda Claeskens and Ali Charkhi, Statistica Sinica, (2016) link

"Model Averaging, Asymptotic Risk, and Regressor Groups," Quantitative Economics, (2014) link

"Efficient Shrinkage in Parametric Models," Journal of Econometrics, (2016) link

"The Risk of James-Stein and Lasso Shrinkage," Econometric Reviews, (2016) link

"A Stein-Like 2SLS Estimator," Econometric Reviews, (2017) link

“Stein Combination Shrinkage for Vector Autoregressions (2016) link

“Multi-step Forecast Model Selection” (2010) link

"Forecasting with Factor-Augmented Regression: A Frequentist Model Averaging Approach," with Xu Cheng, Journal of Econometrics, (2015) link

The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Trevor Hastie, Robert Tibshirani and Jerome Friedman. link

Econometrics, Bruce E. Hansen (2018) link