Macroeconometrics and Machine Learning

Objectif

Large and alternative data sets are widely used for macroeconomic forecasting/nowcasting together with machine learning-based tools. This course presents state-of-the-arts methods for macroeconomic forecasting and nowcasting in presence of large data sets. In particular, the methods illustrated during the course are useful to deal with large-dimensional data sets (big data), to deal with data sets with missing observations, and to estimate forecasting models that are potentially complex and nonlinear. Choice of tuning parameters and how this choice affects the accuracy of the forecasting will be discussed.

At the end of the course students will be able to select the most appropriate techniques for macroeconomic forecasting estimation depending on the data available and on the assumptions made about the forecasting model.

The course evaluation will be based on a project. It is an applied project that has to be done in groups of 2 or 3 students. The project has to be returned by January 22nd on the platform https://app.compilatio.net/v5/document-submission/D37-AFF-754. During the class, we will analyse the FRED-MD data set and we will see how to implement some of the methods seen in class. Students are then asked in their project to develop, on this dataset, another Machine Learning-method among the ones seen in class. They have to : (1) understand the methodology, (2) compare the results with the ones seen in class, (3) interpret them from a macroeconomic point of view. Each group has to return a report (no longer than 5 − 7 pages with reasonable size and font) containing an explanation of the methodology used, the empirical results and an interpretation of the empirical results. In addition of the page limit you are required to provide the code in the R Markdown format. Hint : a good project is a project that explains the model and the results in a concise way by pointing out the important message and information.

Plan

Part I : Factor Models

1. Static and Dynamic Factor Models (FMs)

2. Estimation of the Factors and Parameters

3. Determination of the number of factors

4. Dynamic FMs for Macroeconomic Monitoring and Forecasting

5. Implementation in R of some of the methods seen in class.

Part II : Machine Learning (ML) methods in Macroeconometrics

1. Introduction : data-poor versus data-rich environments, features of ML.

2. Penalized linear models, dense versus sparse models.

3. Tree-based methods and Random forests.

4. Neural Networks.

5. Explainable machine learning

Références

This class has an applied part that will interact with the two theoretical parts. For this applied part you are required to have installed on your laptop RStudio. RStudio can be download from https://posit.co/download/rstudio-desktop/. Prior to install RStudio you have to install R and RTools. You can nd guidance on this on the following webpages : https://larmarange.github.io/analyse-R/installation-de-R-et-RStudio.html and https://ohdsi.github.io/Hades/rSetup.html.

This course we will not follow a textbook. References to several academic articles will be provided during the course.