Macro-Based Regime-Switching Models for Portfolio Allocation
Translated title
Makrobaserede Regimeskiftmodeller i Porteføljeallokering
Authors
Elvstrøm, Daniel Vinther ; Custovic, Emil
Term
4. semester
Publication year
2026
Submitted on
2026-06-01
Pages
56
Abstract
This study examines whether macro-based regime-switching models can improve portfolio allocation compared with simple benchmarks and a machine-learning approach. We use a Hidden Markov Model (HMM), a statistical model, to identify three “hidden” economic regimes—recession, neutral, and expansion—based on lagged macro indicators such as inflation, industrial production, interest rates, the term spread (the gap between long- and short-term rates), and economic policy uncertainty. To keep the approach practical, we rely on the model’s real-time (filtered) probabilities for each regime, apply simple confirmation rules to avoid excessive switching, use fixed regime-dependent portfolio weights, include transaction costs, and evaluate performance on out-of-sample data not used for training. The analysis has four parts: (1) descriptive statistics, (2) regime identification, (3) portfolio performance evaluation, and (4) robustness tests that check alternative model specifications. The descriptive results show that the chosen variables are suitable for modeling and that stock returns are not normally distributed. We assess whether the HMM captures economically meaningful market environments and compare the regime-based strategy with the S&P 500, an equal-weighted benchmark, sector portfolios, and a Random Forest (a common machine-learning method). Overall, the findings indicate that regime-based allocation can improve risk management relative to the market, mainly through lower drawdowns and more dynamic risk exposure. However, the HMM strategy does not clearly outperform the equal-weighted benchmark or the Random Forest approach in cumulative and risk-adjusted returns.
Dette studie undersøger, om makroøkonomiske regimeskiftmodeller kan forbedre, hvordan man fordeler investeringer i en portefølje, sammenlignet med simple benchmarks og en maskinlæringsmetode. Vi anvender en Hidden Markov Model (HMM), som er en statistisk model, til at identificere tre “skjulte” økonomiske regimer—recession, neutral og ekspansion—ud fra forsinkede makroindikatorer som inflation, industriproduktion, renter, term spread (forskellen mellem lange og korte renter) og økonomisk‑politisk usikkerhed. For at gøre metoden realistisk bruger vi modellens løbende (filtrerede) sandsynligheder for hvert regime, simple regler for at bekræfte regimeskift og undgå for hyppige skift, faste porteføljevægte, der afhænger af regime, medtager transaktionsomkostninger og tester på data, modellen ikke er trænet på (out‑of‑sample). Analysen omfatter: (1) deskriptiv statistik, (2) regimeidentifikation, (3) evaluering af porteføljeafkast og risiko, og (4) robusthedstests, der afprøver alternative modelspecifikationer. Den deskriptive del viser, at de valgte variable egner sig til modellering, og at aktieafkast ikke følger en normalfordeling. Vi vurderer, om HMM’en fanger økonomisk meningsfulde markedsmiljøer, og sammenligner den regimebaserede strategi med S&P 500, et ligevægtet benchmark, sektorporteføljer og en Random Forest (en udbredt maskinlæringsmetode). Resultaterne peger på, at en regimebaseret allokering kan forbedre risikostyringen relativt til markedet, især ved lavere drawdowns og en mere dynamisk risikoeksponering. HMM‑strategien overgår dog ikke tydeligt det ligevægtede benchmark eller Random Forest målt på samlet (kumulativt) og risikojusteret afkast.
[This apstract has been rewritten with the help of AI based on the project's original abstract]