Paper: arXiv 2410.01864

Authors: Diego Vallarino

Abstract

This paper introduces a novel approach to optimizing portfolio rebalancing by integrating Graph Neural Networks (GNNs) for predicting transaction costs and Dijkstra’s algorithm for identifying cost-efficient rebalancing paths. Using historical stock data from prominent technology firms, the GNN is trained to forecast future transaction costs, which are then applied as edge weights in a financial asset graph. Dijkstra’s algorithm is used to find the least costly path for reallocating capital between assets. Empirical results show that this hybrid approach significantly reduces transaction costs, offering a powerful tool for portfolio managers, especially in high-frequency trading environments. This methodology demonstrates the potential of combining advanced machine learning techniques with classical optimization algorithms to improve financial decision-making processes. Future research will explore expanding the asset universe and incorporating reinforcement learning for continuous portfolio optimization.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematics including Graph Neural Networks (GNNs) with iterative message-passing equations and Dijkstra’s algorithm for pathfinding, indicating high complexity. It also provides empirical results using historical data from tech stocks and discusses implementation in high-frequency trading contexts, demonstrating solid empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Optimize portfolio rebalancing<br>to minimize transaction costs"] --> B["Data Input:<br>Historical stock data<br>(Tech firms)"]

  B --> C["Methodology:<br>Train GNN model to predict<br>future transaction costs"]
  C --> D["Graph Construction:<br>Create financial asset graph<br>Edge weights = Predicted costs"]
  
  D --> E["Computational Process:<br>Apply Dijkstra's algorithm<br>to find least costly path"]

  E --> F["Key Findings:<br>Significant reduction in<br>transaction costs achieved<br>Effective for high-frequency trading"]
  
  F --> G["Future Work:<br>Expand asset universe<br>Integrate Reinforcement Learning"]

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 hybrid model that combines Graph Neural Networks for predicting transaction costs with Dijkstra's algorithm to find cost‑efficient rebalancing paths.

Source sentence
This paper introduces a novel approach to optimizing portfolio rebalancing by integrating Graph Neural Networks (GNNs) for predicting transaction costs and Dijkstra's algorithm for identifying cost-efficient rebalancing paths.

Assumes 4

  • Larger daily price differences are taken to represent higher transaction costs. data
    Source sentence
    This approach assumes that larger price differences equate to higher transaction costs.
  • Assets are modeled as nodes in a fully connected, undirected graph where edges represent transaction costs. model
    Source sentence
    In this study, financial assets (stocks) are represented as nodes in a fully connected, undirected graph.
  • The selected five technology stocks are sufficiently liquid to reflect realistic rebalancing conditions. market
    Source sentence
    The selection of these companies is driven by their liquidity and relevance in global financial markets.
  • Missing price data can be accurately filled by linear interpolation or forward‑filling. data
    Source sentence
    Any missing data is handled by linear interpolation or forward-filling methods to ensure a continuous dataset.

Data

  • Yahoo Finance historical stock price data · Yahoo Finance market-prices public
    January 2023 to August 2024 · Apple (AAPL), Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), Tesla (TSLA)
    Source sentence
    The historical stock price data for this study was sourced from Yahoo Finance. The dataset includes daily closing prices from January 2023 to August 2024 for five prominent technology companies: Apple (AAPL), Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), and Tesla (TSLA).

Recognised sources: Yahoo Finance

Costs

Modeled Costs are part of the reported headline results.

Transaction costs are predicted by the GNN and used as edge weights in Dijkstra's algorithm; total minimized cost Cmin = Σ w(u,v) over the optimal path.

Source sentence
w(u,v) is the transaction cost between nodes u and v.

Finds 3

  • The GNN predicts transaction costs with high accuracy, achieving an average MSE of 0.0416 on the test set. predictive-accuracy
    Source sentence
    The model achieved an average MSE of 0.0416, indicating a high level of accuracy in predicting transaction costs.
  • Predicted transaction costs correlate strongly with actual costs (R² = 0.9538) over a 30‑day out‑of‑sample period. performance
    Source sentence
    Graphical representation the comparison of real and predicted transaction costs over the last 30 days, with key performance metrics indicating model accuracy (R² = 0.9538).
  • Integrating Dijkstra's algorithm with GNN predictions reduces overall portfolio rebalancing transaction costs compared to traditional approaches. performance
    Source sentence
    Empirical results show that this hybrid approach significantly reduces transaction costs, offering a powerful tool for portfolio managers, especially in high-frequency trading environments.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 16 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.