过拟合检测

Lesson Overview

In "Quant Strategy Development" → "策略 validation", this lesson "过拟合检测" builds knowledge you can apply immediately.

过拟合检测不能证明策略永远有效，但可发现选择依赖。看样本外衰减、参数敏感性、成本压力和置换基线，比一个指标更可靠。 冻结候选→加倍成本→邻域扰动→时间外检验→与简单基线比较→记录所有结果。 本课作业：训练净收益20%、未来−3%，解释为什么不能直接把未来区间删掉再重新报告20%。

Core Concepts

These points align with the video walkthrough—take notes as you watch.

• ▸过拟合检测不能证明策略永远有效，但可发现选择依赖。看样本外衰减、参数敏感性、成本压力和置换基线，比一个指标更可靠。
• ▸冻结候选→加倍成本→邻域扰动→时间外检验→与简单基线比较→记录所有结果。
• ▸本课作业：训练净收益20%、未来−3%，解释为什么不能直接把未来区间删掉再重新报告20%。

Practical Application

After studying, complete the companion practice lesson to move from understanding to doing.

• ▸Follow along in Jupyter or VS Code
• ▸Log how parameter changes affect backtest results
• ▸Submit a reproducible analysis notebook

FAQ

• ▸Q: How does this lesson fit the course? A: Part of "策略 validation"—follow the curriculum order.
• ▸Q: What do I need? A: Demo account and notes; quant lessons need a Python environment.
• ▸Q: How do I know I've mastered it? A: Pass the quiz and explain three key points of 过拟合检测 on your own.