A0711
Title: Transform and sparsify: Advancing macroeconomic predictions
Authors: Tim Reinicke - ETH Zurich (Switzerland) [presenting]
Maurizio Daniele - ETH Zürich, KOF Swiss Economic Institute (Switzerland)
Philipp Kronenberg - ETH Zurich (Switzerland)
Abstract: Exploring the potential of sparsity and multiple data transformations simultaneously reveals opportunities for enhancing forecast accuracy for macroeconomics. An extensive out-of-sample forecast exercise using the FRED-MD data assesses the performance of various factor models with different sparsity implementations for predicting key macroeconomic indicators. Comparisons involve standard factor models, specifications using variable pre-selection, and sparse factor models with sparsity integrated directly into the estimation process. Performance is also evaluated against machine-learning models such as lasso, elastic-net, and random forests. Findings highlight the enhancement of forecast accuracy through sparsity and diverse data transformations. These results gain particular significance for all models during turbulent periods, underlining the importance of reconsidering macroeconomic data transformations across different regimes.