Variable selection in MIDAS models: A comparison of Ridge, Elastic Net, LASSO, sparse-group LASSO, OCMT, and BMT
Translated title
Variabeludvælgelse i MIDAS-modeller: En sammenligning af Ridge, Elastic Net, LASSO, sparse-group LASSO, OCMT og BMT
Author
Jensen, Jonas Vammen
Term
4. term
Education
Publication year
2026
Abstract
This thesis examines which regularization and variable selection approaches most effectively improve nowcasts of U.S. real GDP growth in high-dimensional mixed-frequency MIDAS models. The empirical setup combines macroeconomic, financial, and textual news variables observed at different frequencies to avoid pre-aggregation. Six methods are compared—Ridge, LASSO, Elastic Net, sparse-group LASSO, One Covariate Multiple Testing (OCMT), and Boosting Multiple Testing (BMT). Forecasts are evaluated in a rolling pseudo–out-of-sample framework from 2002Q1 to 2017Q2 at three nowcasting horizons (2-month, 1-month, and end-of-quarter) using RMSE, MSFE, MAE, and the aSPA test. The results indicate that Ridge-U-MIDAS and sparse-group LASSO-MIDAS deliver the strongest forecasting performance, with BMT-U-MIDAS and Elastic Net-U-MIDAS also remaining competitive. In contrast, OCMT-U-MIDAS has the weakest out-of-sample performance despite frequently selecting many predictors. Overall, the findings suggest that shrinkage and structured regularization are effective for GDP nowcasting with high-dimensional mixed-frequency data.
Dette speciale undersøger, hvilke regulariserings- og variabeludvælgelsesmetoder der bedst forbedrer nuværende prognoser (nowcasting) af amerikansk real BNP-vækst i høj-dimensionelle MIDAS-modeller med blandede frekvenser. Analysen kombinerer makroøkonomiske, finansielle og tekstbaserede nyhedsvariabler observeret ved forskellige frekvenser for at undgå forudgående aggregering. Seks metoder sammenlignes: Ridge, LASSO, Elastic Net, sparse-group LASSO, One Covariate Multiple Testing (OCMT) og Boosting Multiple Testing (BMT). Prognoser evalueres i en rullende pseudo out-of-sample-ramme fra 2002K1 til 2017K2 ved tre nowcasting-horisonter (2 måneder, 1 måned og slutningen af kvartalet) med RMSE, MSFE, MAE og aSPA-testen. Resultaterne viser, at Ridge-U-MIDAS og sparse-group LASSO-MIDAS har den bedste prognosepræstation, mens BMT-U-MIDAS og Elastic Net-U-MIDAS også er konkurrencedygtige. Omvendt præsterer OCMT-U-MIDAS svagest, selv om den ofte udvælger mange prædiktorer. Samlet set peger fundene på, at shrinkage og struktureret regularisering er effektive strategier til BNP-nowcasting i høj-dimensionelle datasæt med blandede frekvenser.
[This apstract has been generated with the help of AI directly from the project full text]