Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

E2EAI: End-to-End Deep Learning Framework for Active Investing

Zikai Wei, Bo Dai, Dahua Lin

TL;DR

This work tackles suboptimal active-investing portfolios arising from modular, nonjoint optimization by introducing E2EAI, an end-to-end framework that jointly handles factor selection, deep factor learning on stock graphs, and automatic portfolio construction. It employs gated attention for factor selection, a relational neutralization block on intra- and cross-sector stock graphs to learn multi-horizon deep factors, and a directional buffer with directional attention to interpret factor contributions. A cross-sectional optimizer complements a global optimizer to drive horizon-specific factor returns while the loss function balances portfolio performance, stability, factor returns, and attention accuracy. Experiments on CSI300, CSI500, and CSI1000 with over 2800 stocks demonstrate superior alpha and information ratio relative to baselines, and reveal interpretable, directionally consistent factor contributions that vary with market breadth and liquidity. Overall, the paper offers a practical, interpretable, end-to-end solution for active investing with potential impact for portfolio managers and quantitative analysts.

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.

E2EAI: End-to-End Deep Learning Framework for Active Investing

TL;DR

This work tackles suboptimal active-investing portfolios arising from modular, nonjoint optimization by introducing E2EAI, an end-to-end framework that jointly handles factor selection, deep factor learning on stock graphs, and automatic portfolio construction. It employs gated attention for factor selection, a relational neutralization block on intra- and cross-sector stock graphs to learn multi-horizon deep factors, and a directional buffer with directional attention to interpret factor contributions. A cross-sectional optimizer complements a global optimizer to drive horizon-specific factor returns while the loss function balances portfolio performance, stability, factor returns, and attention accuracy. Experiments on CSI300, CSI500, and CSI1000 with over 2800 stocks demonstrate superior alpha and information ratio relative to baselines, and reveal interpretable, directionally consistent factor contributions that vary with market breadth and liquidity. Overall, the paper offers a practical, interpretable, end-to-end solution for active investing with potential impact for portfolio managers and quantitative analysts.

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.
Paper Structure (11 sections, 9 equations, 3 figures, 1 table, 1 algorithm)

This paper contains 11 sections, 9 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related works
  3. Preliminaries
  4. Method
  5. Factor Selection
  6. Deep Multifactor Model
  7. Portfolio Construction on Automatic Stock Selection
  8. Loss Design
  9. Optimization
  10. Experiment Results
  11. Conclusion

Figures (3)

  • Figure 1: The pipeline of E2E active investing framework (E2EAI).
  • Figure 2: The heat map shows the dynamic contribution of the original factors to the deep factors: 1) the first row shows the dynamic attention allocation to the original factor groups; 2) the second row shows the average attention weights for each style factor on a semi-annual basis.
  • Figure 3: The monotonicity analysis of the deep factor and its attention approximation (AA) on different stock universes. 1) The average returns increase monotonically in stratified groups based on factor exposure, showing good stability of their predictive power; 2) the AA of deep factor can linearly explain most cases in CSI300 and CSI500 from the perspective of portfolio performance in different groups.

Theorems & Definitions (2)

  • definition thmcounterdefinition
  • definition thmcounterdefinition