Runze Liu

Quantitative Researcher

Profile Picture of Runze Liu

Skills Summary

Programming & Tools

  • Python (Pandas, Polars, PyTorch, HuggingFace)
  • C++, Linux, LeetCode (300+ problems solved)
  • Machine Learning: LightGBM, XGBoost, Random Forest, CNN, LSTM
  • Quantitative Tools: CVXPY, Barra, Factor Analysis

Domain Expertise

  • Factor Mining & Evaluation Systems
  • Multi-Factor Integrated Models
  • Stock Market Prediction & Backtesting
  • Large Language Models & Transformer Architecture

Education

2023.09 - 2027.06

Shanghai University of Finance and Economics

B.S. in Financial Statistics and Risk Management

First Prize in Chinese Chemistry Olympiad

People’s Scholarship

College Math Competition (2nd Prize)

Core Courses:

Probability & Statistics, Financial Mathematics, Machine Learning

Co-founder, SUFE Quantitative Association QITA

2024.07 - 2024.08

Peking University

Summer School Program

Course: Large Language Models - From Fundamentals to Frontiers

Studied Transformer architecture, classic LLMs, and HuggingFace/PyTorch frameworks. Conducted group presentations on cutting-edge research papers.

Internship Experience

2025.07 - 2025.09

Hainan

MY Capital (AUM: ¥30B)

Quantitative Researcher

Automated Factor Mining System

Developed high-performance factor evaluation framework with Polars. Implemented RL-based factor mining system using minute-level OHLC data, achieving out-of-sample IC > 0.09.

Daily Multi-Factor Integrated Model

Constructed multi-GPU parallel MoE/GRU architecture. Achieved top 3 groups with Sharpe > 2, 25% annualized return, and turnover < 6% in out-of-sample backtests.

Intraday K-Line Image Recognition

Converted 5-minute K-lines to grayscale images for CNN-based prediction. Achieved out-of-sample IC > 0.11 through 3-year rolling predictions.

2025.01 - 2025.04

Long Life Investment Co., Ltd. (AUM: ¥10B)

Quantitative Researcher

Cutting-edge DL Models Implementation

Reproduced and improved Attention-LSTM/MASTER models, increasing prediction IC to 0.12. Developed proprietary DL training framework.

Backtesting Performance

Achieved 19% annualized return, max drawdown < 6%, and Sharpe ratio > 2.5 under ¥10M+ capital with Barra neutralization.

Key Projects

2025.06

Machine Learning Strategy Based on Multi-Frequency Factor Synthesis

  • Constructed 100+ feature factors using 1-minute and daily market data
  • Applied LightGBM for nonlinear factor integration, achieving overall IC 0.09 (daily avg IC 0.07)
  • 40% annualized return with 15% max drawdown under ¥10M capital and 0.15% transaction cost
2024.11 - 2024.12

Machine Learning-Based Stock Limit-Up/Down Prediction

  • Processed 330,000+ records of A-share data (fundamentals, limit events, Dragon & Tiger List)
  • Designed 5-class prediction framework using XGBoost/CatBoost/Random Forest
  • Achieved 75% accuracy for T+2 day 6-10% return category prediction