Paper: arXiv 2510.14986
Authors: Yiyao Zhang, Diksha Goel, Hussain Ahmad, Claudia Szabo
Abstract
Financial markets are inherently non-stationary, with shifting volatility regimes that alter asset co-movements and return distributions. Standard portfolio optimization methods, typically built on stationarity or regime-agnostic assumptions, struggle to adapt to such changes. To address these challenges, we propose RegimeFolio, a novel regime-aware and sector-specialized framework that, unlike existing regime-agnostic models such as DeepVol and DRL optimizers, integrates explicit volatility regime segmentation with sector-specific ensemble forecasting and adaptive mean-variance allocation. This modular architecture ensures forecasts and portfolio decisions remain aligned with current market conditions, enhancing robustness and interpretability in dynamic markets. RegimeFolio combines three components: (i) an interpretable VIX-based classifier for market regime detection; (ii) regime and sector-specific ensemble learners (Random Forest, Gradient Boosting) to capture conditional return structures; and (iii) a dynamic mean-variance optimizer with shrinkage-regularized covariance estimates for regime-aware allocation. We evaluate RegimeFolio on 34 large cap U.S. equities from 2020 to 2024. The framework achieves a cumulative return of 137 percent, a Sharpe ratio of 1.17, a 12 percent lower maximum drawdown, and a 15 to 20 percent improvement in forecast accuracy compared to conventional and advanced machine learning benchmarks. These results show that explicitly modeling volatility regimes in predictive learning and portfolio allocation enhances robustness and leads to more dependable decision-making in real markets.
Complexity vs Empirical Score
- Math Complexity: 4.5/10
- Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs only partially modelled; see penalty flags)
- Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 8.0)
Why this score: The paper uses standard ML models (RF, GBDT) and classic portfolio optimization (mean-variance with shrinkage) rather than heavy advanced mathematics; the implementation details are straightforward. However, it presents strong empirical results with specific backtest metrics (137% return, 1.17 Sharpe) over a clear 2020-2024 dataset of 34 equities, demonstrating high backtest readiness.
Research Flowchart
flowchart TD Goal["Research Goal<br>Develop a Regime-Aware ML System<br>for Sectoral Portfolio Optimization"] Data["Data & Inputs<br>34 Large Cap U.S. Equities<br>2020-2024<br>VIX & Sector Data"] Proc1["Methodology: Regime Detection<br>Interpretable VIX-based Classifier<br>Identifies Volatility Regimes"] Proc2["Methodology: Forecasting<br>Regime & Sector-Specific Ensemble<br>RF/GBM for Conditional Returns"] Proc3["Methodology: Optimization<br>Dynamic Mean-Variance<br>Shrinkage-Regularized Covariance"] Out["Key Outcomes & Findings<br>Cumulative Return: 137%<br>Sharpe Ratio: 1.17<br>Max Drawdown: -12%<br>Forecast Accuracy: +15-20%"] Goal --> Data Data --> Proc1 Proc1 --> Proc2 Proc2 --> Proc3 Proc3 --> Out
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes RegimeFolio, a novel regime‑aware and sector‑specialized framework that integrates volatility regime segmentation, sector‑specific ensemble forecasting, and adaptive mean‑variance allocation.
Source sentence
To address these challenges, we propose RegimeFolio, a novel regime-aware and sector-specialized framework that, unlike existing regime-agnostic models (e.g., DeepVol, DRL optimizers), integrates explicit volatility regime segmentation with sector-specific ensemble forecasting and adaptive mean‑variance allocation.
Assumes 5
- Market volatility regimes can be identified using the CBOE VIX and a rolling‑tercile methodology. market
Source sentence
Market volatility regimes are classified using the CBOE VIX as the primary segmentation criterion.
- Within each identified regime, asset return dynamics are locally stationary. statistical
Source sentence
The modeling assumes local stationarity within regimes, which may not hold during abrupt structural breaks.
- The investment universe consists of the historical constituents of the S&P 500, using point‑in‑time CRSP membership to avoid survivorship bias. data
Source sentence
Our investment universe is based on the historical constituents of the S&P 500 index[42] to avoid survivorship bias. We use a point in time database from CRSP that provides the exact list of index members for any given date in our study period (2020 – 2024).
- Transaction costs are represented only by fixed per‑trade fees, ignoring price impact, slippage, and liquidity constraints. execution
Source sentence
Finally, transaction cost modeling includes fixed per trade costs but omits price impact, slippage, and liquidity constraints.
Show 1 more
- Sector‑specific ensemble models (Random Forest, Gradient Boosting) trained separately for each regime capture conditional return structures. model
Source sentence
sector-specific ensemble forecasting models trained separately for each regime
Data
2020 – 2024 · 34 large cap U.S. equities · equities · daily
- CRSP daily stock file market-prices commercial
Source sentence
The dataset comprises 34 carefully selected large cap U.S. equities distributed across 7 major economic sectors.
- CBOE VIX options public
Source sentence
Historical VIX data is obtained from the Chicago Board Options Exchange, providing daily volatility expectations derived from S&P 500 index options.
Costs
Partial
fixed per trade costs but omits price impact, slippage, and liquidity constraints
Source sentence
Finally, transaction cost modeling includes fixed per trade costs but omits price impact, slippage, and liquidity constraints.
Finds 5
Sharpe 1.17 (benchmark 0.66) · annual return 18.9% · total return 137.0% · max drawdown 29.3%
- RegimeFolio achieved a cumulative return of 137.0% versus 73.8% for the S&P 500 benchmark. performance
Source sentence
the regime-aware portfolio achieved a 137.0% cumulative return versus 73.8% for the S&P 500 benchmark (Table VII)
- The strategy delivered a Sharpe ratio of 1.17 and a 12% lower maximum drawdown compared to the benchmark. performance
Source sentence
The framework achieves a cumulative return of 137 percent, a Sharpe ratio of 1.17, a 12 percent lower maximum drawdown
- Forecast accuracy improved by 15–20% (MAE reduction) relative to linear baselines, especially in cyclical sectors. predictive-accuracy
Source sentence
Ensemble models (Random Forest, Gradient Boosting) outperform linear baselines by 15–20% in MAE reduction, with larger gains in cyclical sectors such as Energy and Financials
- The strategy outperformed the S&P 500 benchmark by 63.2% over the 5‑year period, generating $63.2 million on a $100 million mandate. performance
Source sentence
The 63.2% outperformance over the S&P 500 benchmark translates to substantial economic value creation: on a $100 million institutional mandate, the strategy generates $63.2 million in additional returns over the 5 year period
- Maximum drawdown was reduced by 12.0 percentage points (‑29.3% vs ‑41.2%). performance
Source sentence
The 12.0 percentage point reduction in maximum drawdown (-29.3% vs -41.2%) provides substantial downside protection during market stress periods.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 61 related papers). Coverage is partial, so this is neither replication nor refutation.
2 papers in the archive
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (63)
- Network centrality and the cross section of stock returns (2017)
- What Skills Do Cyber Security Professionals Need? (2025) — link
- " I think this is the most disruptive technology": Exploring Sentiments of ChatGPT Early Adopters using Twitter Data (2022) — link
- Towards deep learning enabled cybersecurity risk assessment for microservice architectures (2025)
- The future of ai: Exploring the potential of large concept models (2025) — link
- Resilient Auto-Scaling of Microservice Architectures with Efficient Resource Management (2025) — link
- Towards resource-efficient reactive and proactive auto-scaling for microservice architectures (2025)
- Chatnvd: Advancing cybersecurity vulnerability assessment with large language models (2024) — link
and 55 more
- Machine learning driven smishing detection framework for mobile security (2024) — link
- A survey on immersive cyber situational awareness systems (2025)
- Smart HPA: A resource-efficient horizontal pod auto-scaler for microservice architectures (2024)
- The Cross-Section of Volatility and Expected Returns (2006)
- Leveraging LLMs for financial news analysis and macroeconomic indicator nowcasting (2024) — link
- Random forests (2001)
- Strategic asset allocation (1997)
- VIX Index Historical Data (2024)
- DRL-PF: A Double-Scales Deep Reinforcement Learning-Based Portfolio Management Framework (2024) — link
- A hybrid GARCH , convolutional neural network and long short term memory methods for volatility prediction in stock market (2023)
- Federal Reserve Economic Data (2024)
- Greedy function approximation: a gradient boosting machine (2001)
- The fundamental law of active management (1989)
- Asset allocation under multivariate regime switching (2007)
- Size and value anomalies under regime shifts (2008)
- A new approach to the economic analysis of nonstationary time series and the business cycle (1989)
- A deep reinforcement learning framework for dynamic portfolio optimization: Evidence from China’s stock market — → 2412.18563
- Performance hypothesis testing with the Sharpe and Treynor measures (1981)
- Dynamic graph neural networks for enhanced volatility prediction in financial markets — → 2410.16858
- A well-conditioned estimator for large-dimensional covariance matrices (2004)
- GNN-Forecast : A Graph Neural Network-Based Stock Movement Prediction with Pre-training (2024) — link
- Automated volatility forecasting (2024) — link
- Ensemble learning algorithms based on easyensemble sampling for financial distress prediction (2024)
- Identifying bull and bear markets in stock returns (2000)
- Portfolio selection (1952)
- Portfolio optimization using a hybrid machine learning stock selection framework (2024)
- Machine learning, economic regimes and portfolio optimisation (2018)
- Interpretable VIX forecasting using Kolmogorov–Arnold networks (2025)
- DeepVol : Volatility forecasting from high-frequency data with dilated causal convolutions (2024)
- Anatomy of the VIX spike in February 2018 (2018)
- Mutual Fund Performance (1966)
- S&P U.S. Indices Methodology (2025)
- Network-based prediction of financial cross-sector risk spillover in China : A deep learning approach (2024)
- Forecasting intraday volatility and return densities with a new GARCH-f-GAN model (2023)
- Deep Graph-Reinforcement Learning for Portfolio Management (2024)
- Historical Stock Price Data (2024)
- Calmar ratio: A new measure of portfolio performance (1991)
- Volatility forecasting with machine learning and intraday commonality (2024)
- Econometric measures of connectedness and systemic risk in the finance and insurance sectors (2012)
- Fear and greed: A tale of two sentiments (2015)
- Markov-switching dynamic factor models in real time (2015)
- Asset allocation under multivariate regime switching (2008)
- Are stocks really so risky? A long-run perspective (2012)
- Empirical asset pricing via machine learning (2020)
- A Hidden Markov Model approach to detecting regimes in financial time series (2022)
- International stock return comovements (2009)
- The macroeconomic effects of investment-specific technology shocks (2014)
- Scikit-learn: Machine learning in Python (2011)
- A systematic review of reproducibility research in machine learning (2018)
- S&P 500® (2025) — link
- Optimal Versus Naive Diversification: How Inefficient is the 1/N Portfolio Strategy? (2009)
- Capital Asset Prices: A Theory of Market Equilibrium Under Conditions of Risk (1964)
- International asset allocation with regime shifts (2002)
- Hierarchical graph attention network for stock movement prediction (2021)
- Improved estimation of the covariance matrix of stock returns with an application to portfolio selection (2003) — link
2 of 63 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.