Quantitative Research · Deep Learning · On-Chain Data

Venti

Turning messy financial data into testable trading signals.

I build end-to-end quantitative research pipelines — data engineering, factor modeling, deep learning and backtesting — and stress-test them for the only thing that matters: whether the signal survives real-world frictions.

量化研究 · 深度学习 · 链上数据

Venti

把杂乱的金融数据，变成可检验的交易信号。

我搭建端到端的量化研究流程——数据处理、因子建模、深度学习与回测，并始终追问同一个问题：这条信号能否在实际摩擦下存活。

View research查看研究 Toolbox技能栈

Selected research研究精选

Two independent studies, written up as full reports on this site.两项独立研究，均以完整报告形式发布于本站。

Deep LearningEquitiesPyTorch

Mixed-Frequency Deep Learning for Cross-Sectional Stock Prediction

A mixed-frequency model combining CNN-LSTM price-volume features with MLP-encoded fundamentals via self-attention, trained with a pairwise ranking objective on CSI 300 stocks.

+25% peak validation RankIC (0.0268 → 0.0335) from adding fundamentals to a price-only baseline
Top-10 portfolio backtest: 69.3% annualized, Sharpe 3.28, +43.1% annualized excess vs CSI 300 (simulation)

Read the full report →

深度学习股票PyTorch

面向横截面股票预测的混频深度学习模型

将 CNN-LSTM 量价特征与 MLP 编码的财务基本面通过自注意力融合，并以成对排序损失训练，用于沪深 300 股票的横截面排序预测。

引入财务信息使验证集峰值 RankIC 提升 25%（0.0268 → 0.0335）
Top-10 组合回测：年化 69.3%、夏普 3.28，较沪深 300 年化超额 +43.1%（模拟）

阅读完整报告 →

CryptoOn-Chain DataMarket Microstructure

On-Chain MEV Signals for Cryptocurrency Timing

A 5-minute BTC/USDT pipeline combining Binance price-volume data with Dune MEV (sandwich attack) features, de-correlated with Jaccard-similarity spectral clustering.

Directional signal: 52.46% validation accuracy (ROC-AUC 0.5369); MEV features are near-orthogonal to price factors
Strongest on volatility magnitude; zero-friction backtest: 66.4% annualized

Read the full report →

加密资产链上数据市场微观结构

链上 MEV 信号与加密货币择时

结合 Binance 量价数据与 Dune 链上 MEV（三明治攻击）特征的 5 分钟 BTC/USDT 研究管线，使用基于 Jaccard 相似的谱聚类去相关。

方向信号验证准确率 52.46%（ROC-AUC 0.5369），MEV 特征与量价因子接近正交
对波动幅度预测最强；零摩擦回测年化 66.4%

阅读完整报告 →

Toolbox技能栈

AI-Native DevelopmentAI 原生化开发

Codex / CLI agentsPrompt iterationAI-assisted research

Programming & Data编程与数据

Python (NumPy, pandas, scikit-learn, Hugging Face)SQL / SQLite CC#RedisWebSocket WindBinanceDune AnalyticsDashPower BI

Quant & Analytics量化与分析

Factor researchBacktestingStatistical analysis StataExcelFinancial modelingDerivatives pricing

National University of Singapore — Master of Financial Engineering (2026–) · Renmin University of China — FinTech & Accounting (2022–2026). 新加坡国立大学 — 金融工程硕士（2026–）· 中国人民大学 — 金融科技与会计（2022–2026）。