Paper: arXiv 2511.18076
Authors: Fermat Leukam, Rock Stephane Koffi, Prudence Djagba
Abstract
This research proposes an enhancement to the innovative portfolio optimization approach using the G-Learning algorithm, combined with parametric optimization via the GIRL algorithm (G-learning approach to the setting of Inverse Reinforcement Learning) as presented by. The goal is to maximize portfolio value by a target date while minimizing the investor’s periodic contributions. Our model operates in a highly volatile market with a well-diversified portfolio, ensuring a low-risk level for the investor, and leverages reinforcement learning to dynamically adjust portfolio positions over time. Results show that we improved the Sharpe Ratio from 0.42, as suggested by recent studies using the same approach, to a value of 0.483 a notable achievement in highly volatile markets with diversified portfolios. The comparison between G-Learning and GIRL reveals that while GIRL optimizes the reward function parameters (e.g., lambda = 0.0012 compared to 0.002), its impact on portfolio performance remains marginal. This suggests that reinforcement learning methods, like G-Learning, already enable robust optimization. This research contributes to the growing development of reinforcement learning applications in financial decision-making, demonstrating that probabilistic learning algorithms can effectively align portfolio management strategies with investor needs.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 3.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper employs advanced reinforcement learning algorithms (G-Learning, GIRL) and Markov Decision Processes, involving sophisticated probabilistic modeling and mathematical formulations. However, the empirical validation relies on simulated data with no mention of code, specific datasets, or detailed backtesting methodology, resulting in low implementation readiness.
Research Flowchart
flowchart TD
A["Research Goal<br>Maximize portfolio value & minimize contributions<br>within defined time horizon"] --> B["Data & Inputs<br>Historical market data & volatility"]
B --> C["Methodology: Reinforcement Learning<br>G-Learning + GIRL algorithms"]
C --> D{"Computational Process<br>Dynamically adjust portfolio positions<br>based on market volatility"}
D -- G-Learning Core --> E["Model Optimization<br>Robust RL optimization"]
D -- GIRL Refinement --> F["Parametric Tuning<br>Reward function optimization"]
E --> G["Key Findings<br>Sharpe Ratio: 0.42 → 0.483"]
F --> G
G --> H["Outcome<br>Effective alignment of RL strategies<br>with investor financial goals"]
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 an enhanced portfolio optimization approach that combines the G‑Learning algorithm with parametric optimization via the GIRL algorithm to maximize portfolio value by a target date while minimizing periodic contributions.
Source sentence
This research proposes an enhancement to the innovative portfolio optimization approach using the G-Learning algorithm, combined with parametric optimization via the GIRL algorithm (G-learning approach to the setting of Inverse Reinforcement Learning) as presented by.
Assumes 5
- The evolution of assets follows a Markov process, i.e., future prices depend only on the current price. model
Source sentence
To model our market, we start with the assumption that the evolution of assets follows a Markov process.
- Asset price dynamics are modeled by geometric Brownian motion (GBM). model
Source sentence
This assumption accurately reflects the reality of financial markets. In fact, it is characteristic of geometric Brownian motion (GBM), which is widely used in mathematical finance to model stock prices.
- A predetermined target portfolio value ˆPt+1 exists at each time step. other
Source sentence
Furthermore, we assume that at each time step t, a predetermined target value ˆPt+1 exists for the portfolio at the subsequent time t + 1.
- Transaction costs are proportional to the squared change in positions via a weighting matrix Ω. execution
Source sentence
−uT t Ωut: Transaction costs. This term corresponds to the cost of transactions. Each change in the positions in the portfolio ut (buying or selling assets) incurs costs, such as brokerage fees or costs of adjusting the portfolio.
Show 1 more
- The market is highly volatile but the portfolio is well‑diversified, leading to a low‑risk level for the investor. market
Source sentence
Our model operates in a highly volatile market with a well-diversified portfolio, ensuring a low-risk level for the investor.
Data
- Synthetic market data generated from GBM synthetic unknown
Source sentence
To achieve this, we rely on popular financial models, known for their ability to accurately replicate the dynamics of financial markets, to generate our data.
Costs
Modeled
Costs are represented by the term -uᵀΩu in the reward function, where Ω is a weighting matrix controlling magnitude of costs.
Source sentence
−uT t Ωut: Transaction costs. This term corresponds to the cost of transactions. Each change in the positions in the portfolio ut (buying or selling assets) incurs costs, such as brokerage fees or costs of adjusting the portfolio.
Finds 3
- The proposed G‑Learning approach achieved a Sharpe ratio of 0.483, improving over the prior 0.42 benchmark. performance
Source sentence
Results show that we improved the Sharpe Ratio from 0.42, as suggested by recent studies using the same approach, to a value of 0.483 a notable achievement in highly volatile markets with diversified portfolios.
- GIRL parameter optimization (λ = 0.0012 vs. 0.002) yields only a marginal improvement in Sharpe ratio. performance
Source sentence
However, we found that the optimal parameters obtained using the GIRL model (e.g., λ = 0.0012 compared to an initial value of 0.002) result in only a marginal improvement in the Sharpe ratio.
- The goal‑based formulation aligns better with investors’ needs than classical Markowitz approaches. theoretical
Source sentence
Our results also emphasize that the goal-based formulation (focused on specific objectives) is better aligned with investors’ needs than Markowitz’s classical approaches.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 32 related papers). Coverage is partial, so this is neither replication nor refutation.
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 (25)
- Machine learning in finance (2020)
- G-learner and girl: Goal based wealth management with reinforcement learning (2020) — link
- Is deep learning an RG flow? (2019)
- How Machine Learning Is Transforming Portfolio Optimization (2024)
- Portfolio management using reinforcement learning (2016)
- A deep reinforcement learning framework for the financial portfolio management problem (2017) — link
- Dynamic portfolio allocation in goals-based wealth management (2020)
- Market self-learning of signals, impact and optimal trading: Invisible hand inference with free energy (2018) — link
and 17 more
- Statistical Inference for Sharpe Ratio (2010)
- Modèle de Markowitz et détermination d’un portfeuille optimal (2018)
- my Human Brain Project (mHBP) (2021)
- The Markowitz Model (2002)
- Deep Reinforcement Learning Model for Stock Portfolio Management Based on Data Fusion (2024)
- Modern portfolio theory (1952)
- Analysis of new approaches used in portfolio optimization: a systematic literature review (2020)
- Theory of portfolio optimization in markets with frictions (2001)
- A survey of international financial risk management system (2013)
- Finance and diversification (2010)
- Estimating fundamental Sharpe ratios: A Kalman filter approach (2015)
- Teaching the Financial Markets Course (1976)
- Benchmarking performance indices: pitfalls and solutions (2003)
- The failure of risk management: Why it’s broken and how to fix it (2020)
- Benchmarking Reinforcement Learning (RL) Algorithms for Portfolio Optimization (2024)
- Stochastic differential portfolio games (2000)
- Q-learning (1992)
0 of 25 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.