The event classifier's probabilities were sized under a reference population. A later market can change the features, the mapping from features to outcomes, or the costs that turn a correct side into a loss. This note decides which of those changes the researcher is allowed to call a regime feature, using only the distinctions already in drift, decay and retirement and the drawdown discussion in portfolio risk and the quant quake. It does not introduce a structural-break test.
Learning objectives. By the end of this note you should be able to:
- separate feature drift, relationship change and economic decay for the event book;
- refuse a drawdown print as proof that the classifier's conditional model broke;
- name one feature-distribution check and one economics check that the promotion notes already justify;
- write a regime ticket that freezes what would falsify continued sizing.
In this lesson
- Three changes, three responses
- A drawdown is an operating rule
- The regime ticket
Before you start
The full lesson, with worked examples and exercise solutions, is for members.
Read the full lessonNew to SelfMadeQuant.com? See membership options or browse the Financial Machine Learning syllabus.