An End-to-End Structure with Novel Position Mechanism and Improved EMD for Stock Forecasting

Paper: arXiv 2404.07969

Abstract

As a branch of time series forecasting, stock movement forecasting is one of the challenging problems for investors and researchers. Since Transformer was introduced to analyze financial data, many researchers have dedicated themselves to forecasting stock movement using Transformer or attention mechanisms. However, existing research mostly focuses on individual stock information but ignores stock market information and high noise in stock data. In this paper, we propose a novel method using the attention mechanism in which both stock market information and individual stock information are considered. Meanwhile, we propose a novel EMD-based algorithm for reducing short-term noise in stock data. Two randomly selected exchange-traded funds (ETFs) spanning over ten years from US stock markets are used to demonstrate the superior performance of the proposed attention-based method. The experimental analysis demonstrates that the proposed attention-based method significantly outperforms other state-of-the-art baselines. Code is available at https://github.com/DurandalLee/ACEFormer.

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

Why this score: The paper employs advanced deep learning architectures like Transformers and a novel signal processing algorithm (ACEEMD) with mathematical derivations, indicating high mathematical complexity. It also provides empirical validation using real financial data (two ETFs over ten years) and compares against state-of-the-art baselines, with code available, showing substantial empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Improve Stock Movement Forecasting"] --> B["Data Preparation<br>Two ETFs (10+ years)"]
  B --> C["Proposed Method<br>ACEFormer"]
  subgraph C_Method ["ACEFormer Components"]
      C1["Novel Position Mechanism"]
      C2["Improved EMD<br>Denoising"]
      C3["Transformer Attention<br>Market + Individual Info"]
  end
  C --> D["Computational Process<br>Model Training & Evaluation"]
  D --> E{"Performance Comparison"}
  E --> F["Outcome<br>Outperforms SOTA Baselines"]
  E --> G["Outcome<br>Superior ETF Forecasting"]

Cited References (23)

1. Exploring the impact of investor’s sentiment tendency in varying input window length for stock price prediction (2023) — link
2. TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis (2022) — link
3. Non-stationary Transformers: Exploring the Stationarity in Time Series Forecasting (2022) — link
4. Coal Thickness Prediction Method Based on VMD and LSTM (2022) — link
5. Stock Selection via Spatiotemporal Hypergraph Attention Network: A Learning to Rank Approach (2021) — link
6. Prediction of Trends in Stock Market using Moving Averages and Machine Learning (2021) — link
7. A Deep Neural Network Based Model For Stock Market Prediction (2021) — link
8. Prediction of Stock Closing Prices Based on Attention Mechanism (2020) — link

0 of 23 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.