Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DeepAries: Adaptive Rebalancing Interval Selection for Enhanced Portfolio Selection

Jinkyu Kim, Hyunjung Yi, Mogan Gim, Donghee Choi, Jaewoo Kang

TL;DR

DeepAries tackles dynamic portfolio optimization by jointly learning when to rebalance and how to allocate assets, using a Transformer-based state encoder and Proximal Policy Optimization to handle a mixed discrete-continuous action space. By adaptively selecting rebalancing intervals and incorporating transaction-cost awareness, it reduces unnecessary turnover while capturing regime shifts, outperforming fixed-interval baselines across four international markets. The approach introduces an adaptive horizon mechanism, multiple Transformer backbones (notably iTransformer), and horizon-specific value estimation within a unified RL objective, demonstrating robust gains in risk-adjusted returns and drawdown management. This work offers a practical paradigm shift for portfolio management, enabling market-adaptive decisions with interpretable rebalancing and allocation strategies, as evidenced by extensive ablations and real-data experiments.

Abstract

We propose DeepAries , a novel deep reinforcement learning framework for dynamic portfolio management that jointly optimizes the timing and allocation of rebalancing decisions. Unlike prior reinforcement learning methods that employ fixed rebalancing intervals regardless of market conditions, DeepAries adaptively selects optimal rebalancing intervals along with portfolio weights to reduce unnecessary transaction costs and maximize risk-adjusted returns. Our framework integrates a Transformer-based state encoder, which effectively captures complex long-term market dependencies, with Proximal Policy Optimization (PPO) to generate simultaneous discrete (rebalancing intervals) and continuous (asset allocations) actions. Extensive experiments on multiple real-world financial markets demonstrate that DeepAries significantly outperforms traditional fixed-frequency and full-rebalancing strategies in terms of risk-adjusted returns, transaction costs, and drawdowns. Additionally, we provide a live demo of DeepAries at https://deep-aries.github.io/, along with the source code and dataset at https://github.com/dmis-lab/DeepAries, illustrating DeepAries' capability to produce interpretable rebalancing and allocation decisions aligned with shifting market regimes. Overall, DeepAries introduces an innovative paradigm for adaptive and practical portfolio management by integrating both timing and allocation into a unified decision-making process.

DeepAries: Adaptive Rebalancing Interval Selection for Enhanced Portfolio Selection

TL;DR

DeepAries tackles dynamic portfolio optimization by jointly learning when to rebalance and how to allocate assets, using a Transformer-based state encoder and Proximal Policy Optimization to handle a mixed discrete-continuous action space. By adaptively selecting rebalancing intervals and incorporating transaction-cost awareness, it reduces unnecessary turnover while capturing regime shifts, outperforming fixed-interval baselines across four international markets. The approach introduces an adaptive horizon mechanism, multiple Transformer backbones (notably iTransformer), and horizon-specific value estimation within a unified RL objective, demonstrating robust gains in risk-adjusted returns and drawdown management. This work offers a practical paradigm shift for portfolio management, enabling market-adaptive decisions with interpretable rebalancing and allocation strategies, as evidenced by extensive ablations and real-data experiments.

Abstract

We propose DeepAries , a novel deep reinforcement learning framework for dynamic portfolio management that jointly optimizes the timing and allocation of rebalancing decisions. Unlike prior reinforcement learning methods that employ fixed rebalancing intervals regardless of market conditions, DeepAries adaptively selects optimal rebalancing intervals along with portfolio weights to reduce unnecessary transaction costs and maximize risk-adjusted returns. Our framework integrates a Transformer-based state encoder, which effectively captures complex long-term market dependencies, with Proximal Policy Optimization (PPO) to generate simultaneous discrete (rebalancing intervals) and continuous (asset allocations) actions. Extensive experiments on multiple real-world financial markets demonstrate that DeepAries significantly outperforms traditional fixed-frequency and full-rebalancing strategies in terms of risk-adjusted returns, transaction costs, and drawdowns. Additionally, we provide a live demo of DeepAries at https://deep-aries.github.io/, along with the source code and dataset at https://github.com/dmis-lab/DeepAries, illustrating DeepAries' capability to produce interpretable rebalancing and allocation decisions aligned with shifting market regimes. Overall, DeepAries introduces an innovative paradigm for adaptive and practical portfolio management by integrating both timing and allocation into a unified decision-making process.
Paper Structure (29 sections, 19 equations, 1 figure, 3 tables, 1 algorithm)

This paper contains 29 sections, 19 equations, 1 figure, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Classical and Static Portfolio Optimization
  4. Reinforcement Learning for Portfolio Management
  5. Transformer-Based Time Series Models
  6. Limitations of Existing RL Approaches and Adaptive Rebalancing Interval Selection
  7. Problem Formulation
  8. Price Relative Vector.
  9. Portfolio Return and Reward Signal.
  10. Transaction Costs.
  11. Portfolio Value Update.
  12. Objective.
  13. Proposed Method
  14. Challenges in Portfolio Selection
  15. Key Ideas in DeepAries
  16. ...and 14 more sections

Figures (1)

  • Figure 1: Effect of transaction costs on portfolio performance. The six panels are arranged in three columns and two rows. In each column, the transaction cost is fixed: the left column shows results with the baseline fee of 0.01%, the middle column uses fees increased 5-fold, and the right column uses fees increased 10-fold. The top row displays results for the FTSE 100 (reflecting bullish market conditions), while the bottom row shows results for the KOSPI (reflecting volatile market conditions).