WG Risk : “Economic Tracking Forests: Leveraging Tree-based Models for Macroeconomic Forecasts”

The Working Group on Risk - CREAR, with the support of the ESSEC IDO dpt/Ceressec, Institut des actuaires, Labex MME-DII (CY), and the group Risques AEF - SFdS, has the pleasure to invite you to the seminar by:

Prof. Anastasija TETEREVA, Erasmus School of Economics Rotterdam, Netherlands

 Dr. Anastasija Tetereva holds a Ph.D. in Economics and Finance from the University of St. Gallen, Switzerland (after a M.S in Statistics at Humboldt University, Berlin). Prior to joining Erasmus University Rotterdam as an Assistant Professor, she had postdoctoral appointment at the Swiss Institute for Empirical Economic research. Her research interests are mainly in the field of Financial Econometrics with a focus on machine learning methods for modeling high-dimensional and high-frequency time series. Her second area of interest is the incorporation of novel data sources into econometric models.

 “Economic Tracking Forests: Leveraging  Tree-based Models for Macroeconomic Forecasts”

Economic tracking portfolios (ETPs) play a crucial role in economic forecasting, risk management, and in uncovering the links between financial markets and macroeconomic dynamics. This study enhances ETP construction by innovatively adapting random forests (RFs) and local linear forests (LLFs) to better model complex, non-linear dependencies between asset returns and macroeconomic variables. Specifically, RF and LLF-based ETPs are utilized to track inflation, consumption growth, and industrial production growth across various horizons. Our analysis reveals that machine learning-based ETPs consistently outperform traditional linear approaches. For instance, at the 1-year horizon, LLF ETPs significantly enhance the ability to track inflation and consumption growth, as evidenced by increases in R^2 values from 4.4% to 6.9% for inflation and from 3.1% to 7.4% for consumption growth. Furthermore, Shapley value analysis uncovers that the relationships between asset returns and macroeconomic factors are highly sensitive to prevailing economic conditions. Additionally, kernel principal component analysis applied to LLF kernels identifies distinct economic regimes, providing a novel lens for analyzing dynamic economic relationships.

Dual format:

ESSEC Paris La Défense (CNIT), Room TBA, and via Zoom, please click here .