Multiple Testing Frameworks vs. Penalized Estimators in High-Dimensional Environments
Author
Knudsen, Jens Peter Østergaard
Term
4. term
Education
Publication year
2026
Submitted on
2026-05-27
Pages
68
Abstract
This thesis investigates forecasting commodity prices—particularly crude oil and metals—in high-dimensional settings where numerous candidate predictors, strong multicollinearity, and structural shocks challenge standard approaches. Penalized regression methods such as Lasso and Elastic Net are widely used to induce sparsity, but their assumptions can fail during macroeconomic crises, leading to overfitting and weaker out-of-sample performance. As an alternative, the study evaluates two multiple-testing frameworks—One Covariate at a Time Multiple Testing (OCMT) and Boosting with Multiple Testing (BMT)—designed to handle highly correlated predictors while controlling false discoveries. Using a rich information set of macroeconomic indicators, technical variables, and aggregated uncertainty measures (transformed to achieve stationarity), OCMT and BMT are compared with Lasso and Elastic Net. Performance is assessed by out-of-sample statistical accuracy and by economic value in a portfolio allocation exercise that computes Certainty Equivalent Returns and Sharpe Ratios. The core research question is how OCMT and BMT perform relative to penalized estimators in volatile, high-dimensional environments; the provided text does not report empirical findings, but the thesis aims to determine whether multiple-testing approaches yield more accurate and more robust forecasts.
Denne afhandling undersøger, hvordan man kan forudsige råvarepriser – især råolie og metaller – i højdimensionelle miljøer, hvor mange potentielle forklarende variable, stærk multikollinearitet og strukturelle chok udfordrer traditionelle metoder. Penaliserede regressorer som Lasso og Elastic Net bruges ofte til at skabe sparsitet, men deres antagelser kan brydes under makroøkonomiske kriser, hvilket øger overfitting og forringer out-of-sample præstation. Som alternativ vurderes to multiple-testing-rammer – One Covariate at a Time Multiple Testing (OCMT) og Boosting with Multiple Testing (BMT) – der er udviklet til at håndtere høj korrelation og bedre kontrollere falske fund. Med et omfattende informationssæt af makroøkonomiske indikatorer, tekniske variable og aggregerede usikkerhedsmål (transformeret for stationaritet) sammenlignes OCMT og BMT med Lasso og Elastic Net. Evalueringen omfatter både statistisk nøjagtighed out-of-sample og økonomisk relevans via en porteføljeøvelse, der måler sikkerhedsekvivalent afkast og Sharpe-ratioer. Det centrale forskningsspørgsmål er, hvordan OCMT og BMT klarer sig relativt til penaliserede estimatorer i et volatilt, højdimensionelt miljø; den foreliggende tekst rapporterer ikke konkrete empiriske resultater, men afhandlingen har til formål at afgøre, om multiple-testing-rammer leverer mere præcise og mere robuste prognoser.
[This apstract has been generated with the help of AI directly from the project full text]