Paper: arXiv 2505.03760
Authors: Arishi Orra, Aryan Bhambu, Himanshu Choudhary, Manoj Thakur, Selvaraju Natarajan
Abstract
Portfolio optimization requires dynamic allocation of funds by balancing the risk and return tradeoff under dynamic market conditions. With the recent advancements in AI, Deep Reinforcement Learning (DRL) has gained prominence in providing adaptive and scalable strategies for portfolio optimization. However, the success of these strategies depends not only on their ability to adapt to market dynamics but also on the careful pre-selection of assets that influence overall portfolio performance. Incorporating the investor’s preference in pre-selecting assets for a portfolio is essential in refining their investment strategies. This study proposes a volatility-guided DRL-based portfolio optimization framework that dynamically constructs portfolios based on investors’ risk profiles. The Generalized Autoregressive Conditional Heteroscedasticity (GARCH) model is utilized for volatility forecasting of stocks and categorizes them based on their volatility as aggressive, moderate, and conservative. The DRL agent is then employed to learn an optimal investment policy by interacting with the historical market data. The efficacy of the proposed methodology is established using stocks from the Dow \(30\) index. The proposed investor-specific DRL-based portfolios outperformed the baseline strategies by generating consistent risk-adjusted returns.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematical frameworks like GARCH models and Markov Decision Processes with complex LaTeX formulas, while also demonstrating strong empirical rigor through extensive backtesting on real market data with risk-adjusted metrics and benchmark comparisons.
Research Flowchart
flowchart TD A["Research Goal<br>Investor-Specific Portfolio Optimization"] --> B["Data Acquisition<br>Dow 30 Stocks"] B --> C["Volatility Modeling<br>GARCH Forecasting"] C --> D["Asset Categorization<br>Aggressive, Moderate, Conservative"] D --> E["DRL Agent Training<br>Deep Reinforcement Learning"] E --> F["Simulation & Evaluation<br>Risk-Adjusted Returns"] F --> G["Outcome<br>Outperforms Baseline Strategies"]
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 a volatility‑guided deep reinforcement learning framework that builds investor‑specific portfolios based on risk appetite.
Source sentence
This study proposes a volatility-guided DRL-based portfolio optimization framework that dynamically constructs portfolios based on investors' risk profiles.
Assumes 6
- The GARCH(1,1) model is stationary with 0 < α + β < 1. model
Source sentence
with the stationarity ensured by 0 < α + β < 1.
- The mean equation of returns is assumed to be zero. model
Source sentence
The mean equation of the return in Equation (1) is assumed to be zero.
- An initial capital of $1 million is provided to the agent for investing. execution
Source sentence
To simulate real‑world trading conditions, an initial capital of $1 million has been provided to the agent for investing in selected stocks.
- A flat transaction cost of 0.05% is applied to all trades. execution
Source sentence
Motivated from Orra et al. (2024), a transaction cost of 0.05% has been applied to all trades to reflect market conditions accurately.
Show 2 more
- Daily price data from Yahoo Finance accurately represent market prices without slippage. market
Source sentence
Daily price data for all 30 stocks of the DJI index was obtained from Yahoo Finance over a period from January 1, 2010, to December 31, 2024.
- The top 10 most volatile and bottom 10 least volatile stocks can be identified reliably from the GARCH forecasts. data
Source sentence
The top 10 volatile stocks of the DJI index are used to build an aggressive portfolio, the least volatile 10 are used to build a conservative portfolio, and the remaining are used to develop a moderate portfolio.
Data
2023-01 – 2024-12 · Dow 30 stocks · equities · daily
- Dow Jones Index (DJI) daily price data market-prices public
Source sentence
Daily price data for all 30 stocks of the DJI index was obtained from Yahoo Finance over a period from January 1, 2010, to December 31, 2024.
Costs
Modeled
0.05% flat fee applied to all trades.
Source sentence
a transaction cost of 0.05% has been applied to all trades to reflect market conditions accurately.
Finds 5
Sharpe 1.81 (benchmark 1.21) · annual return 24.0% · total return 53.4% · max drawdown 10.3%
- Aggressive‑DRL achieved the highest annual return (26.714%) and cumulative return (60.629%) among all models. performance
Source sentence
The Aggressive-DRL portfolio, consisting of highly volatile stocks, achieved the highest annual (26.71%) and cumulative (60.63%) returns among all the models.
- Moderate‑DRL attained the highest Sharpe ratio (1.811), indicating the best risk‑adjusted performance. performance
Source sentence
Also, it attains the highest Sharpe ratio (1.81), indicating a better risk‑adjusted performance among all the other models.
- Conservative‑DRL showed the lowest volatility (10.08%) and drawdown (9.97%), confirming its risk‑optimizing capability. performance
Source sentence
The Conservative-DRL achieves the least volatility (10.08%) and a lower drawdown (9.97%), showing its effectiveness in optimizing portfolio risk.
- All DRL‑based portfolios significantly outperformed benchmark strategies (MVO, DJI index, Equal‑Weighted). performance
Source sentence
All the DRL‑based portfolio models significantly outperformed the benchmark models, which considered the entire pool of stocks for constructing the portfolios.
- The proposed volatility‑guided investor‑specific portfolios consistently delivered superior risk‑adjusted returns across the test period. performance
Source sentence
The proposed three investor-specific DRL-based portfolios outperformed the baseline strategies by generating consistent risk‑adjusted returns.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 72 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper 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 (16)
- Foundations of portfolio theory (1991)
- Performance functions and reinforcement learning for trading systems and portfolios (1998)
- A deep reinforcement learning framework for the financial portfolio management problem (2017) — link
- Deep reinforcement learning for stock portfolio optimization by connecting with modern portfolio theory (2023)
- Continuous control with stacked deep dynamic recurrent reinforcement learning for portfolio optimization (2020)
- Multi-period portfolio optimization using a deep reinforcement learning hyper-heuristic approach (2024)
- DeepTrader: a deep reinforcement learning approach for risk-return balanced portfolio management with market conditions Embedding (2021)
- Recurrent ensemble random vector functional link neural network for financial time series forecasting (2024)
and 8 more
- Generalized autoregressive conditional heteroskedasticity (1986)
- Portfolio formation with preselection using deep learning from long-term financial data (2020)
- A multiobjective multiperiod portfolio selection approach with different investor attitudes under an uncertain environment (2024)
- Reinforcement learning: An introduction (2018)
- Proximal policy optimization algorithms (2017) — link
- FinRL: A deep reinforcement learning library for automated stock trading in quantitative finance (2020) — link
- Dynamic Reinforced Ensemble using Bayesian Optimization for Stock Trading (2024)
- Forecasting volatility in the financial markets (2011)
0 of 16 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.