- How to Prevent Overfitting in Quant Backtesting: CPCV + Deflated Sharpe Ratio
A high Sharpe in backtest tells you almost nothing. This post explains the three-part anti-overfitting toolkit for serious quant research: CPCV (Combinatorial Purged Cross-Validation), DSR (Deflated Sharpe Ratio multiple-testing correction), and the transaction-cost gate.
- Factor Collinearity: Diagnosis & Deduplication
You think you have 50 factors — the correlation matrix shows only 5-10 truly independent dimensions. Collinearity destabilizes model weights and inflates backtest. Diagnosis triad and dedup methods.
- Factor Mining for China A-Shares: Seven Factor Types & Anti-Overfitting Discipline
A practitioner overview of quantitative factor mining for China A-shares: the seven major factor categories (momentum, reversal, microstructure, money-flow, fundamentals, ML-synthesized) and the OOS + DSR + cost-gate discipline that separates real alpha from noise.
- LLM-Assisted Factor Mining: Tool, Not Product
"Use GPT to mine factors" sounds trendy, but the real value of LLMs in quant is accelerating the research workflow — not generating alpha. Statistical discipline cannot be replaced by LLMs.
- API vs MCP: Two Paradigms for Quantitative Data Access
Two ways to wire quant research to data: traditional HTTP REST API (programmatic, predictable) versus MCP (Model Context Protocol — let LLMs discover and call tools directly). When to use which, and why they coexist rather than compete.
- Portfolio Optimization: Industry Neutrality & Risk Attribution
Turning factor scores into executable position weights — the portfolio optimization layer decides how to allocate, how to control risk, and how to neutralize industry exposure. Without this layer, alpha gets confounded with industry beta.
- Real-time Risk & Algorithmic Order Execution: The Last Line
A great strategy with broken execution wrecks live PnL fast. The execution layer is the last defense between strategy output and real accounts: order slicing, intraday stops, circuit breakers.