Paper: arXiv 2504.20058

Authors: Ambedkar Dukkipati, Kawin Mayilvaghanan, Naveen Kumar Pallekonda, Sai Prakash Hadnoor, Ranga Shaarad Ayyagari

Abstract

Fluctuations in stock prices are influenced by a complex interplay of factors that go beyond mere historical data. These factors, themselves influenced by external forces, encompass inter-stock dynamics, broader economic factors, various government policy decisions, outbreaks of wars, etc. Furthermore, all of these factors are dynamic and exhibit changes over time. In this paper, for the first time, we tackle the forecasting problem under external influence by proposing learning mechanisms that not only learn from historical trends but also incorporate external knowledge from temporal knowledge graphs. Since there are no such datasets or temporal knowledge graphs available, we study this problem with stock market data, and we construct comprehensive temporal knowledge graph datasets. In our proposed approach, we model relations on external temporal knowledge graphs as events of a Hawkes process on graphs. With extensive experiments, we show that learned dynamic representations effectively rank stocks based on returns across multiple holding periods, outperforming related baselines on relevant metrics.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)

Why this score: The paper employs advanced mathematics including heterogeneous Hawkes processes on temporal knowledge graphs and graph convolution operators, but also demonstrates strong empirical rigor through the construction of two comprehensive datasets (NASDAQ/NSE SmKGs) and extensive experiments outperforming baselines.

Research Flowchart

  flowchart TD
  A["Research Goal: Forecast Stock Returns Under External Influence"] --> B["Construct Temporal Knowledge Graph TKG Dataset"]
  B --> C["Model External Relations as Events in Graph Hawkes Process"]
  C --> D["Learn Dynamic Stock Representations"]
  D --> E["Experiments: Rank Stocks by Predicted Returns"]
  E --> F["Key Findings: Outperforms Baselines, Effective Dynamic Ranking"]

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 time‑series forecasting algorithm that incorporates knowledge of external events via a heterogeneous temporal knowledge graph and a temporal point‑process model.

Source sentence
We propose a time‑series forecasting algorithm that incorporates knowledge of external events influencing the evolution of the series.

Assumes

No explicit assumptions found in the text.

Data

  • NASDAQ100 · Yahoo Finance market-prices public
    30 January 2012 to 20 October 2022 · NASDAQ100 constituents
    Source sentence
    Three datasets are created, containing a subset of stocks from NASDAQ and NIFTY, with datasets NASDAQ100 and S&P500 from the NASDAQ market and NIFTY500 from NSE, using Yahoo Finance.
  • S&P500 · Yahoo Finance market-prices public
    30 January 2012 to 20 October 2022 · S&P500 constituents
    Source sentence
    Three datasets are created, containing a subset of stocks from NASDAQ and NIFTY, with datasets NASDAQ100 and S&P500 from the NASDAQ market and NIFTY500 from NSE, using Yahoo Finance.
  • NIFTY500 · Yahoo Finance market-prices public
    30 January 2012 to 20 October 2022 · NIFTY500 constituents
    Source sentence
    Three datasets are created, containing a subset of stocks from NASDAQ and NIFTY, with datasets NASDAQ100 and S&P500 from the NASDAQ market and NIFTY500 from NSE, using Yahoo Finance.

Costs

Not modeled A trading or execution result is reported without costs.

Finds 2

  • TA‑HKGE consistently achieves the strongest or near‑strongest ranking quality (ND‑CG/ACC) and investment returns (IRR/AIRR) across all holding periods on both NASDAQ and NSE markets, outperforming five baseline methods. performance
    Source sentence
    In Tables 2 and Table 4 (Appendix), we compare TA‑HKGE with five baseline methods across all holding periods (∆∈{1, 5, 20}). The proposed TA‑HKGE consistently achieves the strongest or near‑strongest ranking quality (ND‑CG/ACC) and investment returns (IRR/AIRR) across both the NASDAQ and NSE markets.
  • Moderate sliding‑window sizes generally yield stronger forecasting performance, as shown by the effect of window size W on Top‑5 return ratio across different holding periods. performance
    Source sentence
    Our method integrates historical stock prices with temporal events and relationships to forecast future trends. To analyze the effect of the historical context length, we varied the sliding window size W used in the sequential embedding layer. Moderate window sizes generally yield stronger...

Checked by others

Contradicted by

  • Bellwether Trades: Characteristics of Trades influential in Predicting Future Price Movements in Markets (2024)
    On the question: How does the size of the sliding window affect the predictive performance of trade‑based signals?
    Compare the findings

    This paper: Moderate sliding‑window sizes generally yield stronger forecasting performance, as shown by the effect of window size W on Top‑5 return ratio across different holding periods.

    That paper: Lower window sizes provide superior predictive power, which plateaus as window size increases.

    Finding A reports lower window sizes give superior predictive power that plateaus as size grows, while Finding B reports moderate window sizes generally yield stronger forecasting performance.

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 (89)

  1. Ci-sthpan: Pre-trained attention network for stock selection with channel-independent spatio-temporal hypergraph (2024)
  2. The efficient market hypothesis and its critics (2003)
  3. Individualized Indicator for All: Stock-wise Technical Indicator Optimization with Stock Embedding (2019)
  4. Financial market bubbles and crashes: features, causes, and effects (2018)
  5. Mdgnn: Multi-relational dynamic graph neural network for comprehensive and dynamic stock investment prediction — → 2402.06633
  6. Effect of insider trading on stock characteristics (2020)
  7. Stock manipulation and its impact on market quality (2005)
  8. Anomalies in stock market pricing: problems in return measurements (2006)
and 81 more
  1. Exploring graph neural networks for stock market predictions with rolling window analysis (2019) — link
  2. Applied attention-based LSTM neural networks in stock prediction (2018)
  3. Trade the event: Corporate events detection for news-based event-driven trading (2021) — link
  4. Temporal relational ranking for stock prediction (2019)
  5. Heterogeneous edge-enhanced graph attention network for multi-agent trajectory prediction (2021) — link
  6. Stock price prediction via discovering multi-frequency trading patterns (2017)
  7. Listening to chaotic whispers: A deep learning framework for news-oriented stock trend prediction (2018)
  8. Incorporating transformers and attention networks for stock movement prediction (2022)
  9. Incorporating corporation relationship via graph convolutional neural networks for stock price prediction (2018)
  10. Guided attention multimodal multitask financial forecasting with inter-company relationships and global and local news (2022)
  11. Deep learning for stock prediction using numerical and textual information (2016)
  12. A graph neural network-based stock forecasting method utilizing multi-source heterogeneous data fusion (2022)
  13. Stock selection via spatiotemporal hypergraph attention network: A learning to rank approach (2021)
  14. Spatiotemporal hypergraph convolution network for stock movement forecasting (2020)
  15. FinBERT: Financial Sentiment Analysis with Pre-trained Language Models (2019) — link
  16. When flue meets flang: Benchmarks and large pre-trained language model for financial domain (2022) — link
  17. Bloomberggpt: A large language model for finance (2023) — link
  18. Pixiu: A large language model, instruction data and evaluation benchmark for finance (2023) — link
  19. Llama: Open and efficient foundation language models (2023) — link
  20. Translating embeddings for modeling multi-relational data (2013)
  21. Knowledge graph embedding by translating on hyperplanes (2014)
  22. Rotate: Knowledge graph embedding by relational rotation in complex space (2019) — link
  23. Embedding entities and relations for learning and inference in knowledge bases (2014) — link
  24. Holographic embeddings of knowledge graphs (2016)
  25. Complex embeddings for simple link prediction (2016)
  26. A novel embedding model for knowledge base completion based on convolutional neural network (2017) — link
  27. Knowledge embedding based graph convolutional network (2021)
  28. Adriana Romero Pietro Lio and Yoshua Bengio. Graph attention networks (2018)
  29. node2vec: Scalable feature learning for networks (2016)
  30. Deepwalk: Online learning of social representations (2014)
  31. Inductive representation learning on large graphs (2017)
  32. Semi-supervised classification with graph convolutional networks (2016) — link
  33. Graph attention networks (2017)
  34. metapath2vec: Scalable representation learning for heterogeneous networks (2017)
  35. Hin2vec: Explore meta-paths in heterogeneous information networks for representation learning (2017)
  36. Magnn: Metapath aggregated graph neural network for heterogeneous graph embedding (2020)
  37. Heterogeneous embedding propagation for large-scale e-commerce user alignment (2018)
  38. Representation learning for attributed multiplex heterogeneous network (2019)
  39. Heterogeneous graph transformer (2020)
  40. Heterogeneous graph attention network (2019)
  41. Heterogeneous graph structure learning for graph neural networks (2021)
  42. A comprehensive survey of graph embedding: Problems, techniques, and applications (2018)
  43. A survey on heterogeneous graph embedding: methods, techniques, applications and sources (2022)
  44. Continuous-time dynamic network embeddings (2018)
  45. Inductive representation learning on temporal graphs (2020) — link
  46. Dyngem: Deep embedding method for dynamic graphs (2018) — link
  47. Dynamic graph convolutional networks (2020)
  48. Evolvegcn: Evolving graph convolutional networks for dynamic graphs (2020)
  49. Temporal heterogeneous interaction graph embedding for next-item recommendation (2021)
  50. Modeling dynamic heterogeneous network for link prediction using hierarchical attention with temporal rnn (2021)
  51. Dynamic heterogeneous graph embedding using hierarchical attentions (2020)
  52. Temporal network embedding with micro-and macro-dynamics (2019)
  53. Embedding temporal network via neighborhood formation (2018)
  54. Attention is all you need (2017)
  55. Revisiting approximate metric optimization in the age of deep neural networks (2019)
  56. Dynamic heterogeneous graph embedding via heterogeneous hawkes process (2021)
  57. TRNN: An efficient time-series recurrent neural network for stock price prediction (2024)
  58. Neural machine translation by jointly learning to align and translate (2014) — link
  59. Inductive Representation Learning on Temporal Graphs (2020) — link
  60. Neural latent space model for dynamic networks and temporal knowledge graphs (2021)
  61. Neural temporal point process for forecasting higher order and directional interactions (2025)
  62. Dynamic heterogeneous graph embedding via heterogeneous hawkes process (2021)
  63. Fusing external knowledge resources for natural language understanding techniques: A survey (2023)
  64. The neural H awkes process: A neurally self-modulating multivariate point process (2017)
  65. Dyrep: Learning representations over dynamic graphs (2019)
  66. Recurrent Neural Networks for Time Series Forecasting: Current status and future directions (2021) — link
  67. Time Series Forecasting Using LSTM Networks: A Symbolic Approach (2020) — link
  68. Unlocking the Power of LSTM for Long Term Time Series Forecasting (2025) — link
  69. Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting (2021) — link
  70. iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (2024) — link
  71. From News to Forecast: Integrating Event Analysis in LLM-Based Time Series Forecasting with Reflection (2024) — link
  72. Using Twitter Attribute Information to Predict Stock Prices (2021) — link
  73. D oc2 EDAG : An End-to-End Document-level Framework for C hinese Financial Event Extraction (2019)
  74. ForeSeer: Product Aspect Forecasting Using Temporal Graph Embedding (2023)
  75. Beyond Trend and Periodicity: Guiding Time Series Forecasting with Textual Cues (2024) — link
  76. Context Matters: Leveraging Contextual Features for Time Series Forecasting (2025) — link
  77. MM-Forecast: A Multimodal Approach to Temporal Event Forecasting with Large Language Models (2024)
  78. TimeXer: Empowering Transformers for Time Series Forecasting with Exogenous Variables (2024) — link
  79. ChronosX: Adapting Pretrained Time Series Models with Exogenous Variables (2025) — link
  80. Graph Deep Learning for Time Series Forecasting (2023) — link
  81. Knowledge Graph Representations to enhance Intensive Care Time-Series Predictions (2023) — link

1 of 89 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.