返回项目
Data Science Competition2023.02 – 2023.04
Kaggle 竞赛：JPX 股票预测与 Optiver 波动率预测
JPX 股票预测前 10%，Optiver 波动率预测银牌级别。
PythonLightGBMXGBoostTabNetpandasNumPy
Developed competition solutions during internship at a Kaggle competition R&D company.
JPX Tokyo Stock Exchange Prediction
Built 20+ technical indicators from Tokyo Stock Exchange data. Trained LightGBM with rolling cross-validation to predict T+2 cross-sectional returns for stock ranking. Online Sharpe ratio of 0.119, ranking in top 10%.
Optiver Realized Volatility Prediction
Predicted 10-minute option realized volatility from 20-minute Level 2 LOB data. Constructed slope, depth, and spread order book features with multi-window statistics. Used K-means for stock ID aggregation. Ensembled 3 gradient boosting trees and TabNet via Voting, achieving online RMSPE of 0.21524 (silver medal level).
核心成果
- JPX top 10% (Sharpe 0.119)
- Optiver silver level (RMSPE 0.21524)
- LightGBM + TabNet ensemble
- LOB feature engineering