Paper: arXiv 2508.18592

Authors: Lin Cai, Zhiyang He, Caiya Zhang

Abstract

This paper proposes a novel stock selection strategy framework based on combined machine learning algorithms. Two types of weighting methods for three representative machine learning algorithms are developed to predict the returns of the stock selection strategy. One is static weighting based on model evaluation metrics, the other is dynamic weighting based on Information Coefficients (IC). Using CSI 300 index data, we empirically evaluate the strategy’ s backtested performance and model predictive accuracy. The main results are as follows: (1) The strategy by combined machine learning algorithms significantly outperforms single-model approaches in backtested returns. (2) IC-based weighting (particularly IC_Mean) demonstrates greater competitiveness than evaluation-metric-based weighting in both backtested returns and predictive performance. (3) Factor screening substantially enhances the performance of combined machine learning strategies.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs only partially modelled; reported Sharpe ratio of 33.06 is implausibly high for daily or lower-frequency data; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)

Why this score: The paper uses formal statistical concepts like IC (Information Coefficient) and involves ensemble weighting logic, placing it in the middle-to-high range for math, while it demonstrates strong empirical rigor through backtesting on CSI 300 data, comparing multiple methods, and discussing feature selection.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Develop Novel Stock Selection Strategy<br>using Combined ML & Dynamic Weighting"] --> B["Data: CSI 300 Index Stocks<br>with Technical & Fundamental Factors"]

  B --> C["Methodology:<br>Two Weighting Schemes"]
  
  C --> D["Static Weighting<br>Based on Model Evaluation Metrics"]
  C --> E["Dynamic Weighting<br>Based on Information Coefficient IC"]

  D & E --> F["Ensemble Execution:<br>Combined ML Algorithms<br>(e.g., Random Forest, XGBoost, LightGBM)"]
  
  F --> G["Performance Comparison<br>via Backtesting & Predictive Accuracy"]
  
  G --> H{"Key Outcomes"}
  H --> I["1. Combined ML > Single Models<br>in Backtested Returns"]
  H --> J["2. IC_Mean Weighting<br>Best Performance (Returns & Accuracy)"]
  H --> K["3. Factor Screening<br>Critically Enhances Strategy"]

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 stock selection strategy framework that combines multiple machine learning algorithms with dynamic weighting methods.

Source sentence
This paper proposes a novel stock selection strategy framework based on combined machine learning algorithms.

Assumes 3

  • Training data for each prediction cycle consists of the previous 12 months of historical data. data
    Source sentence
    During each prediction cycle, the training set is constructed using historical data from the previous 12-month period.
  • Realized returns are adjusted for a flat transaction cost of 0.3%. execution
    Source sentence
    Realized returns are adjusted to account for transaction costs, with the total cost rate set at 0.3%.
  • Information Coefficient (IC) values are computed via Spearman correlation between predicted and actual returns. model
    Source sentence
    we calculate the IC values through Spearman correlation between predicted returns and real returns, which can reflect both the direction and m…

Data

Backtest sample: 2020-01 – 2022-12 · CSI 300 Index · equities · monthly

  • CSI 300 index data market-prices public
    CSI 300 constituents
    Source sentence
    Using CSI 300 index data, we empirically evaluate the strategy' s backtested performance and model predictive accuracy.
  • Publicly accessible databases other public
    Source sentence
    The data employed in this study are obtained from publicly accessible databases[11][12].

Costs

Unclear - sources disagree The platform classifier and the paper-text extraction give different answers, so no value is asserted.

Platform classifier: Partial

Paper text extraction: Modeled - A flat total cost rate of 0.3% is applied to realized returns.

Source sentence (the authors' words)
Realized returns are adjusted to account for transaction costs, with the total cost rate set at 0.3%.

Finds 2

Reported performance (ICMean weighted combined forecast): Sharpe 33.06 (benchmark 0.00) · annual return 39.1% · max drawdown 10.9%

  • IC‑based weighting (especially ICMean) yields the highest backtested returns, far exceeding evaluation‑metric weighting and single‑model approaches. performance
    Source sentence
    the strategy returns based on IC Mean and IC IR stand at 39.09% and 29.89%, respectively.
  • Combined machine learning algorithms significantly outperform single‑model approaches in backtested returns. performance
    Source sentence
    The strategy by combined machine learning algorithms significantly outperforms single-model approaches in backtested returns.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 12 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 (23)

  1. Using support vector machine with a hybrid feature selection method to the stock trend prediction (2009) — link
  2. Intelligent optimization based multi-factor deep learning stock selection model and quantitative trading strategy (2022) — [link](https://doi.org/https: //doi.org/10.3390/math10040566)
  3. Multifactor prediction model for stock market analysis based on deep learning techniques (2025) — link
  4. A comprehensive evaluation of ensemble learning for stock-market prediction (2020) — link
  5. Predicting Stock Prices: Strategies of Ensemble Learning with Transformer, ARIMA, and Linear Regression Models (2024) — link
  6. Stock Type Prediction Based on Multiple Machine Learning Methods (2024) — link
  7. Multi Factor Quantitative Stock and Transaction Timing Selection Model Based on Information Coefficient Mean Value Scoring (2021) — link
  8. Overview of Machine Learning for Stock Selection Based on Multi-Factor Models (2020) — link
and 15 more
  1. Extra-market components of covariance in security returns (1974) — link
  2. Barra China Equity Model (CNE5)(2012)
  3. Tushare Big Data Open Community
  4. Shenwan Industry Classification Standard
  5. Common risk factors in the returns on stocks and bonds (1993) — link
  6. Returns to Buying Winners and Selling Losers: Implications for (1993) — link
  7. Security Analysis (1934)
  8. A five-factor asset pricing model (2015) — link
  9. Quality minus junk (2019) — link
  10. The cost of capital, corporation finance and the theory of investment (1958) — link
  11. Leverage and the beta anomaly (2020) — link
  12. Asset pricing and the bid-ask spread (1986) — link
  13. Liquidity risk after 20 years (2019) — link
  14. Regression shrinkage and selection via the lasso (1996) — link
  15. The fundamental law of active management (1989) — link

0 of 23 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.