Paper: arXiv 2401.14199

Authors: Junwei Su, Shan Wu, Jinhui Li

Abstract

In this study, we explore the synergy of deep learning and financial market applications, focusing on pair trading. This market-neutral strategy is integral to quantitative finance and is apt for advanced deep-learning techniques. A pivotal challenge in pair trading is discerning temporal correlations among entities, necessitating the integration of diverse data modalities. Addressing this, we introduce a novel framework, Multi-modal Temporal Relation Graph Learning (MTRGL). MTRGL combines time series data and discrete features into a temporal graph and employs a memory-based temporal graph neural network. This approach reframes temporal correlation identification as a temporal graph link prediction task, which has shown empirical success. Our experiments on real-world datasets confirm the superior performance of MTRGL, emphasizing its promise in refining automated pair trading strategies.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces a novel graph neural network architecture with memory modules and temporal graph attention, requiring advanced mathematical concepts; it also reports experiments on real-world datasets and includes an ablation study, indicating strong empirical validation.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Discern Temporal Correlations<br>in Pair Trading"] --> B["Key Methodology: MTRGL Framework<br>Multi-modal Temporal Relation Graph Learning"]
  B --> C["Computational Process:<br>Temporal Graph Link Prediction<br>Memory-based Temporal GNN"]
  C --> D{"Key Data Inputs"}
  D --> D1["Time Series Data<br>Price/Returns"]
  D --> D2["Discrete Features<br>Financial Attributes"]
  D1 & D2 --> E["Key Finding:<br>Superior Performance on<br>Real-world Equities Datasets"]
  E --> F["Outcome:<br>Refined Automated<br>Pair Trading 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 novel Multi-modal Temporal Relation Graph Learning (MTRGL) framework for pair trading.

Source sentence
Addressing this, we introduce a novel framework, Multi-modal Temporal Relation Graph Learning (MTRGL).

Assumes 2

  • Each time interval ∆i has the same size |∆i| = δ = T/K. data
    Source sentence
    We assume that each time interval ∆i has the same size of |∆i| = δ = T/K , i.e., Ω= {[0, δ), [δ, 2δ), ..., [(K −1)δ, Kδ)} for simplification.
  • A correlation measure S(Pi(t), Pj(t)) (e.g., NHD) and a threshold γ are given. model
    Source sentence
    Assuming S(Pi(t), Pj(t)) 7→[0, 1] is a given measure function (for simplicity, we use the NHD) for the correlation of two time series, and γ is a given threshold, we construct the k‑th temporal graph Gk...

Data

Recognised sources: Yahoo Finance

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • MTRGL significantly outperforms existing baselines in predicting future correlations among entities. performance
    Source sentence
    As Table 1 shows, MTRGL significantly outperforms the existing baselines in predicting future correlations among entities.
  • Removing feature information or structural information leads to a notable performance drop in MTRGL. performance
    Source sentence
    The results reveal a significant performance drop when feature information is removed

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 6 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 (92)

  1. Article Title (2003)
  2. Convolutional networks on graphs for learning molecular fingerprints (2015) — link
  3. Simple and Deep Graph Convolutional Networks (2020) — link
  4. Gated Graph Sequence Neural Networks (2017) — link
  5. Gaan: Gated attention networks for learning on large and spatiotemporal graphs (2018) — link
  6. Simplifying graph convolutional networks (2019)
  7. Semi-Supervised Classification with Graph Convolutional Networks (2017) — link
  8. Graph attention networks (2017) — link
and 84 more
  1. Inductive Representation Learning on Large Graphs (2018) — link
  2. Diffusion convolutional recurrent neural network: Data-driven traffic forecasting (2017) — link
  3. Predict then propagate: Graph neural networks meet personalized pagerank (2018) — link
  4. Fastgcn: fast learning with graph convolutional networks via importance sampling (2018) — link
  5. Stochastic training of graph convolutional networks with variance reduction (2017) — link
  6. Graph Convolutional Neural Networks for Web-Scale Recommender Systems (2018) — link
  7. Adaptive sampling towards fast graph representation learning (2018) — link
  8. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks (2019)
  9. Understanding graph sampling algorithms for social network analysis (2011)
  10. Relational inductive biases, deep learning, and graph networks (2018) — link
  11. Neural Message Passing for Quantum Chemistry (2017) — link
  12. Evaluation of pairs-trading strategy at the Brazilian financial market (2009)
  13. Profitability of pairs trading strategy in an illiquid market with multiple share classes (2012)
  14. Pairs trading: Performance of a relative-value arbitrage rule (2006)
  15. Deep learning for financial applications: A survey (2020)
  16. Foundations and modeling of dynamic networks using dynamic graph neural networks: A survey (2021)
  17. Predicting dynamic embedding trajectory in temporal interaction networks (2019)
  18. TIGER: Temporal Interaction Graph Embedding with Restarts (2023) — link
  19. Temporal graph networks for deep learning on dynamic graphs (2020) — link
  20. Towards Better Evaluation for Dynamic Link Prediction (2022)
  21. Inductive representation learning on temporal graphs (2020) — link
  22. Learning phrase representations using RNN encoder-decoder for statistical machine translation (2014) — link
  23. Enhancing a pairs trading strategy with the application of machine learning (2020)
  24. Convolutional neural networks for time series classification (2017)
  25. ChatGPT Informed Graph Neural Network for Stock Movement Prediction — → 2306.03763
  26. Temporal Graph Networks for Deep Learning on Dynamic Graphs (2021)
  27. Inductive representation learning on temporal graphs (2020) — link
  28. Apan: Asynchronous propagation attention network for real-time temporal graph embedding (2021)
  29. Dyrep: Learning representations over dynamic graphs (2019)
  30. Predicting dynamic embedding trajectory in temporal interaction networks (2019)
  31. ReFresh: Reducing Memory Access from Exploiting Stable Historical Embeddings for Graph Neural Network Training (2023) — link
  32. Model-Architecture Co-Design for High Performance Temporal GNN Inference on FPGA (2022)
  33. PipeGCN : Efficient Full-Graph Training of Graph Convolutional Networks with Pipelined Feature Communication (2022)
  34. Gnnautoscale: Scalable and expressive graph neural networks via historical embeddings (2021)
  35. Sancus: sta le n ess-aware c omm u nication-avoiding full-graph decentralized training in large-scale graph neural networks (2022)
  36. Tgl: A general framework for temporal gnn training on billion-scale graphs (2022) — link
  37. Representation learning for dynamic graphs: A survey (2020)
  38. TGOpt: Redundancy-Aware Optimizations for Temporal Graph Attention Networks (2023)
  39. Stochastic training of graph convolutional networks with variance reduction (2017) — link
  40. Accelerating training and inference of graph neural networks with fast sampling and pipelining (2022)
  41. Do We Really Need Complicated Model Architectures For Temporal Networks? (2023) — link
  42. BNS-GCN: Efficient full-graph training of graph convolutional networks with partition-parallelism and random boundary node sampling (2022)
  43. Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks (2019)
  44. Dysat: Deep neural representation learning on dynamic graphs via self-attention networks (2020)
  45. Fastgcn: fast learning with graph convolutional networks via importance sampling (2018) — link
  46. Spatial temporal graph convolutional networks for skeleton-based action recognition (2018)
  47. ByteGNN: efficient graph neural network training at large scale (2022)
  48. P3: Distributed Deep Graph Learning at Scale. (2021)
  49. Inductive representation learning on large graphs (2017)
  50. MariusGNN: Resource-Efficient Out-of-Core Training of Graph Neural Networks (2023)
  51. GNNlab: a factored system for sample-based GNN training over GPUs (2022)
  52. Graph convolutional neural networks for web-scale recommender systems (2018)
  53. Pipe-SGD: A decentralized pipelined SGD framework for distributed deep net training (2018)
  54. SAPipe: Staleness-Aware Pipeline for Data Parallel DNN Training (2022)
  55. Hogwild!: A lock-free approach to parallelizing stochastic gradient descent (2011)
  56. More effective distributed ml via a stale synchronous parallel parameter server (2013)
  57. Mxnet: A flexible and efficient machine learning library for heterogeneous distributed systems (2015) — link
  58. Pytorch: An imperative style, high-performance deep learning library (2019)
  59. Deep graph library: Towards efficient and scalable deep learning on graphs (2019)
  60. Minimal variance sampling with provable guarantees for fast training of graph neural networks (2020)
  61. On the importance of sampling in learning graph convolutional networks (2021) — link
  62. Mining of massive data sets (2020)
  63. IGE+: A Framework for Learning Node Embeddings in Interaction Graphs (2019)
  64. Learning temporal interaction graph embedding via coupled memory networks (2020)
  65. Continuous-time dynamic network embeddings (2018)
  66. Megatron-lm: Training multi-billion parameter language models using model parallelism (2019) — link
  67. Deepspeed: System optimizations enable training deep learning models with over 100 billion parameters (2020)
  68. Implementation and experimentation of producer-consumer synchronization problem (2011)
  69. Transfer learning-based multi-scale denoising convolutional neural network for prostate cancer detection (2022)
  70. Combination of UAV Images and DSM for Land Cover Classification using Convolutional Neural Network (2022)
  71. Evaluation of the performance of traditional machine learning algorithms, convolutional neural network and AutoML Vision in ultrasound breast lesions classification: A comparative study (2021)
  72. Privacy-preserving machine learning with fully homomorphic encryption for deep neural network (2022)
  73. Can chatgpt forecast stock price movements? return predictability and large language models — → 2304.07619
  74. Are ChatGPT and large language models “the answer” to bringing us closer to systematic review automation? (2023)
  75. ChatGPT and a new academic reality: Artificial Intelligence-written research papers and the ethics of the large language models in scholarly publishing (2023)
  76. Machine learning for image based species identification (2018)
  77. A survey on the state-of-the-art machine learning models in the context of NLP (2016)
  78. The language of proteins: NLP, machine learning & protein sequences (2021)
  79. Graph learning: A survey (2021)
  80. Runtime performance prediction for deep learning models with graph neural network (2023)
  81. STRUCTURAL PROPERTIES OF CORE–PERIPHERY COMMUNITIES (2023)
  82. Towards robust graph incremental learning on evolving graphs (2023)
  83. Structure of Core-Periphery Communities (2022)
  84. PRES : Toward Scalable Memory-Based Dynamic Graph Neural Networks (2023) — link

2 of 92 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.