Paper: arXiv 2305.16364

Authors: Zikai Wei, Bo Dai, Dahua Lin

Abstract

Active investing aims to construct a portfolio of assets that are believed to be relatively profitable in the markets, with one popular method being to construct a portfolio via factor-based strategies. In recent years, there have been increasing efforts to apply deep learning to pursue “deep factors’’ with more active returns or promising pipelines for asset trends prediction. However, the question of how to construct an active investment portfolio via an end-to-end deep learning framework (E2E) is still open and rarely addressed in existing works. In this paper, we are the first to propose an E2E that covers almost the entire process of factor investing through factor selection, factor combination, stock selection, and portfolio construction. Extensive experiments on real stock market data demonstrate the effectiveness of our end-to-end deep leaning framework in active investing.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 5.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces novel, advanced mathematical architectures (gated attention blocks, directional linear estimators, graph neural networks) and a custom loss function, indicating high mathematical complexity. It is backed by extensive experiments on real-world stock market data and claims outperformance over baselines, providing sufficient empirical rigor for a research paper, though it lacks the raw data/code links or exhaustive statistical validation typical of immediate backtest readiness.

Research Flowchart

  flowchart TD
  A["Research Goal: Construct Portfolio via End-to-End Deep Learning"] --> B["Data Input: Historical Stock Market Data & Factors"]
  B --> C["Methodology: E2E Deep Learning Framework"]
  C --> D["Core Process: Factor Selection & Combination"]
  D --> E["Core Process: Stock Selection & Prediction"]
  E --> F["Core Process: Portfolio Construction"]
  F --> G["Outcome: Active Investment Portfolio"]
  G --> H["Key Finding: Demonstrated Effectiveness on Real Market Data"]

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 an end-to-end deep learning framework (E2EAI) for active investing that integrates factor selection, stock selection, and portfolio construction.

Source sentence
we propose a novel E2E Active Investing framework (E2EAI) that encourages all modules, i.e., factor selection, stock selection, and portfolio construction, to fulfill the same learning objective

Assumes

No explicit assumptions found in the text.

Data

  • China stock market dataset (CSI 300, CSI 500, CSI 1000 constituents) · tushare.pro (publicly available, also Refinitiv or WIND) market-prices public
    2010 to mid-2022 · constituents of CSI 300, CSI 500 and CSI 1000
    Source sentence
    of the most representative stocks of China stock markets, i.e., the constituents of three major China Stock Indices (CSIs): CSI 300, CSI 500 and CSI 1000, which represent more than 80% of total market capitalization of the entire stock market in China

Recognised sources: Refinitiv / LSEG (Datastream, Eikon, Tick History) Tushare Wind

Costs

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

Finds 2

  • The proposed E2EAI framework outperforms stepwise baselines in portfolio performance metrics such as excess return (α) and information ratio (IR). performance
    Source sentence
    our E2E method outperforms the other stepwise baselines. Overall, our E2EAI can outperform different baselines under different portfolio construction paradigms w.r.t metrics such as α and IR.
  • Using a directional buffer, the model can identify both positive and negative contributions of each factor and reveal dynamic allocation across factors. other
    Source sentence
    Using a directional buffer, we can identify both the positive and negative contribution of each factor and also determine the dynamic allocation of each orignal factor.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 14 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 4 papers in the archive (4 by independent authors)

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.