Identifying and Forecasting Macroeconomic Drivers of Cryptocurrency Returns
Author
Andersen, Martin Husfeldt
Term
4. term
Education
Publication year
2026
Submitted on
2026-06-09
Pages
63
Abstract
This project examines whether macro-financial variables (for example broader market indicators) can help predict returns on Bitcoin, Ethereum, and Litecoin. We compare several variable selection methods designed for data with many potential predictors and high volatility: Lasso (a regularization approach that selects variables), One-Covariate at a Time, Multiple Testing (OCMT), and Boosting with Multiple Testing (BMT). We evaluate performance within the data used to fit the models (in-sample) and on new, unseen data (out-of-sample). In-sample, we find stable associations between cryptocurrency returns and the selected drivers. Out-of-sample, predictive accuracy is modest across all three cryptocurrencies, and simpler, more parsimonious models generally outperform more complex specifications. Overall, macro-financial variables carry some predictive signal, but forecasting cryptocurrency returns reliably remains difficult.
Dette projekt undersøger, om makrofinansielle variable (fx bredere markedsindikatorer) kan hjælpe med at forudsige afkast på Bitcoin, Ethereum og Litecoin. Vi sammenligner flere metoder til variabeludvælgelse, som er udviklet til data med mange mulige forklaringsvariable og høj volatilitet: Lasso (en regulariseringsmetode, der udvælger variable), One-Covariate at a Time, Multiple Testing (OCMT) og Boosting with Multiple Testing (BMT). Vi vurderer resultater både i den del af data, som modellerne er trænet på (in-sample), og på nye, ubrugte data (out-of-sample). In-sample ser vi stabile sammenhænge mellem kryptovalutaafkast og de udvalgte drivere. Out-of-sample er den forudsigende præcision moderat på tværs af alle tre kryptovalutaer, og enklere, mere parsimoniske modeller klarer sig generelt bedre end mere komplekse specifikationer. Samlet set indeholder makrofinansielle variable noget forudsigelsesinformation, men det er stadig vanskeligt at forudsige kryptovalutaafkast på en robust måde.
[This apstract has been rewritten with the help of AI based on the project's original abstract]