Paper: arXiv 2412.03038
Authors: Liwei Deng, Tianfu Wang, Yan Zhao, Kai Zheng
Abstract
Portfolio management is an important yet challenging task in AI for FinTech, which aims to allocate investors’ budgets among different assets to balance the risk and return of an investment. In this study, we propose a general Multi-objectIve framework with controLLable rIsk for pOrtfolio maNagement (MILLION), which consists of two main phases, i.e., return-related maximization and risk control. Specifically, in the return-related maximization phase, we introduce two auxiliary objectives, i.e., return rate prediction, and return rate ranking, combined with portfolio optimization to remit the overfitting problem and improve the generalization of the trained model to future markets. Subsequently, in the risk control phase, we propose two methods, i.e., portfolio interpolation and portfolio improvement, to achieve fine-grained risk control and fast risk adaption to a user-specified risk level. For the portfolio interpolation method, we theoretically prove that the risk can be perfectly controlled if the to-be-set risk level is in a proper interval. In addition, we also show that the return rate of the adjusted portfolio after portfolio interpolation is no less than that of the min-variance optimization, as long as the model in the reward maximization phase is effective. Furthermore, the portfolio improvement method can achieve greater return rates while keeping the same risk level compared to portfolio interpolation. Extensive experiments are conducted on three real-world datasets. The results demonstrate the effectiveness and efficiency of the proposed framework.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)
Why this score: The paper presents a novel multi-objective framework with theoretical proofs (e.g., for portfolio interpolation) and extensive experiments on three real-world datasets, accompanied by open-sourced code.
Research Flowchart
flowchart TD A["Research Goal: Portfolio Management with Controllable Risk"] --> B["Phase 1: Return-Related Maximization"] B --> B1["Inputs: Historical Market Data"] B --> B2["Auxiliary Objectives: Return Prediction & Ranking"] B --> B3["Portfolio Optimization to prevent overfitting"] C["Phase 2: Risk Control"] --> C1["Input: User-specified Risk Level"] C --> C2["Method A: Portfolio Interpolation<br>*Theoretical guarantee of risk control*"] C --> C3["Method B: Portfolio Improvement<br>*Higher returns at same risk level*"] B3 --> D["Computational Process"] C2 --> D C3 --> D D --> E["Key Findings & Outcomes"] E --> E1["Effective & Efficient Framework"] E --> E2["Generalization to future markets"] E --> E3["Fine-grained, fast risk adaptation"]
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 MILLION, a general multi‑objective deep‑learning framework with controllable risk for portfolio management, consisting of a return‑related maximization phase and a risk‑control phase.
Source sentence
we propose a general Multi-objectIve framework with controLLable rIsk for pOrtfolio maNagement (MILLION)
Assumes 2
- Assume we have a set of portfolios whose return rates are the same for the interpolation analysis. model
Source sentence
Assume we have a set of portfolios {𝒃1𝑡, 𝒃2𝑡, · · · , 𝒃𝑖𝑡, · · · } whose return rates are the same.
- A holding period is the minimum time unit to invest an asset, and the investment horizon is divided into non‑overlapped periods of fixed length (e.g., one day). market
Source sentence
A holding period is the minimum time unit to invest an asset. We follow previous studies that divide the whole investment period into multiple non‑overlapped holding periods with fixed length, such as one day or one month.
Data
- NAS100 market-prices unknown
Source sentence
NAS100 2009/01/01-2020/02/01 2020/02/01-2021/01/01 78
- DOW30 market-prices unknown
Source sentence
DOW30 2009/01/01-2019/02/01 2019/02/01-2020/01/01 28
- Crypto10 crypto-onchain unknown
Source sentence
Crypto10 2018/08/02-2021/07/01 2021/07/01-2023/10/32 10
Costs
Not modeled
Source sentence
Transaction cost (𝑐𝑡) in Equation 4 is set to 0
Finds 5
- MILLION consistently achieves the best performance in return‑related metrics such as APR, ASR, and ACR across three real‑world datasets. performance
Source sentence
our proposed MILLION framework consistently achieves the best performance in terms of return‑related metrics such as APR, ASR, and ACR.
- The portfolio improvement method outperforms the interpolation method, even when no predicted return rates are used. performance
Source sentence
Improvement consistently outperforms Interpolation, even in scenarios where there are no predicted return rates to guide optimization.
- MILLION shows comparable training times to predict‑then‑optimize algorithms and is faster than RL‑based methods. performance
Source sentence
MILLION demonstrates comparable training times to predict‑then‑optimize algorithms and outperforms RL‑based methods in terms of speed.
- As the specified risk level increases, both realized ASR and AVOL increase, confirming the fine‑grained risk control capability. theoretical
Source sentence
as specified risk increases, both realized ASR and AVOL simultaneously increase.
- Risk‑constrained portfolios can achieve higher ASR than portfolios optimized solely for Sharpe ratio, e.g., LR with risk 5e‑5 yields ASR > 1.5 on DOW30. performance
Source sentence
constraining LR’s risk to 5e-5 results in an ASR exceeding 1.5, surpassing the 1.43 ASR achieved through maximizing the Sharpe ratio alone
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.
1 paper in the archive
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (69)
- Robust Median Reversion Strategy for Online Portfolio Selection (2016)
- Portfolio Selection (1952)
- AlphaStock: A Buying-Winners-and-Selling-Losers Investment Strategy using Interpretable Deep Reinforcement Attention Networks (2019)
- FinRL: deep reinforcement learning framework to automate trading in quantitative finance (2021)
- FinRL-Meta: Market Environments and Benchmarks for Data-Driven Financial Reinforcement Learning (2022)
- Portfolio management system in equity market neutral using reinforcement learning (2021)
- Reinforcement-Learning based Portfolio Management with Augmented Asset Movement Prediction States (2020)
- Financial Trading as a Game: A Deep Reinforcement Learning Approach (2018)
and 61 more
- Deep Learning for Portfolio Optimization (2020)
- Introduction to Computational Finance and Financial Econometrics (2017)
- Rethinking and Benchmarking Predict-then-Optimize Paradigm for Combinatorial Optimization Problems (2023)
- A General Framework for Portfolio Construction Based on Generative Models of Asset Returns — → 2312.03294
- Moving average reversion strategy for on-line portfolio selection (2015)
- The Sharpe Ratio (1994)
- Stock Portfolio Management by Using Fuzzy Ensemble Deep Reinforcement Learning Algorithm (2023)
- A Universal End-to-End Approach to Portfolio Optimization via Deep Learning (2021)
- Mercury: A Deep Reinforcement Learning-Based Investment Portfolio Strategy for Risk-Return Balance (2023)
- Relation-Aware Transformer for Portfolio Policy Learning (2020)
- Stock Movement Prediction Based on Bi-Typed Hybrid-Relational Market Knowledge Graph via Dual Attention Networks (2022)
- Multi-Relational Graph Convolution Network for Stock Movement Prediction (2022)
- Modeling the Stock Relation with Graph Network for Overnight Stock Movement Prediction (2020)
- Multi-task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics (2017)
- Mean-variance portfolio optimization using machine learning-based stock price prediction (2021)
- Constrained Max Drawdown: a Fast and Robust Portfolio Optimization Approach — → 2401.02601
- Confidence Weighted Mean Reversion Strategy for Online Portfolio Selection (2011)
- Heuristics for cardinality constrained portfolio optimisation (2000)
- Optimization of conditional value-at risk (2000)
- Multitrend Conditional Value at Risk for Portfolio Optimization.
- Asynchronous Methods for Deep Reinforcement Learning (2016)
- Proximal Policy Optimization Algorithms (2017)
- Continuous control with deep reinforcement learning (2015)
- A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem (2017)
- Points of Significance: Classification and regression trees (2017)
- Random Forests (2001)
- Probabilistic Outputs for Support vector Machines and Comparisons to Regularized Likelihood Methods (1999)
- Long Short-Term Memory (1997)
- Efficient Continuous Space Policy Optimization for High-frequency Trading (2023)
- Mastering Stock Markets with Efficient Mixture of Diversified Trading Experts (2023)
- Investment Behaviors Can Tell What Inside: Exploring Stock Intrinsic Properties for Stock Trend Prediction (2019)
- WISE: Wavelet based Interpretable Stock Embedding for Risk-Averse Portfolio Management (2022)
- Exploring the Scale-Free Nature of Stock Markets: Hyperbolic Graph Learning for Algorithmic Trading (2021)
- Lower Risks, Better Choices: Stock Correlation Based Portfolio Selection in Stock Markets (2023)
- DoubleAdapt: A Meta-learning Approach to Incremental Learning for Stock Trend Forecasting — → 2306.09862
- Accurate Multivariate Stock Movement Prediction via Data-Axis Transformer with Multi-Level Contexts (2021)
- REST: Relational Event-driven Stock Trend Forecasting (2021)
- Domain Adaptive Multi-Modality Neural Attention Network for Financial Forecasting (2020)
- Knowledge-Driven Stock Trend Prediction and Explanation via Temporal Convolutional Network (2019)
- A Multimodal Event-Driven LSTM Model for Stock Prediction Using Online News (2021)
- Multi-source aggregated classification for stock price movement prediction (2022)
- Deep reinforcement learning for stock portfolio optimization by connecting with modern portfolio theory (2023)
- Deep Learning for Event-Driven Stock Prediction (2015)
- Temporal Relational Ranking for Stock Prediction (2018)
- Graph-Based Stock Recommendation by Time-Aware Relational Attention Network (2021)
- Relational Temporal Graph Convolutional Networks for Ranking-Based Stock Prediction (2023)
- StockFormer: Learning Hybrid Trading Machines with Predictive Coding (2023)
- Adaptive Quantitative Trading: An Imitative Deep Reinforcement Learning Approach (2020)
- Commission Fee is not Enough: A Hierarchical Reinforced Framework for Portfolio Management (2020)
- DeepTrader: A Deep Reinforcement Learning Approach for Risk-Return Balanced Portfolio Management with Market Conditions Embedding (2021)
- A Smart Trader for Portfolio Management based on Normalizing Flows (2022)
- DeepScalper: A Risk-Aware Reinforcement Learning Framework to Capture Fleeting Intraday Trading Opportunities (2021)
- Contrastive Learning and Reward Smoothing for Deep Portfolio Management (2023)
- Adaptive Long-Short Pattern Transformer for Stock Investment Selection (2022)
- Enhancing Stock Movement Prediction with Adversarial Training (2018)
- Joint Admission Control and Resource Allocation of Virtual Network Embedding Via Hierarchical Deep Reinforcement Learning (2024)
- COMET: NFT Price Prediction with Wallet Profiling (2024)
- Learning to Hash for Trajectory Similarity Computation and Search (2024)
- iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (2024)
- Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting (2021)
- Informer: Beyond efficient transformer for long sequence time-series forecasting (2021)
3 of 69 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.