A single end-to-end classifier that outputs long, short or flat mixes two jobs: which side and how large. Meta-labeling, as stated in the bet-sizing lesson, separates them. This note adapts only that section: primary side versus secondary size for the event classifier. It does not expand into architectures or loss functions beyond what that note establishes.
Learning objectives. By the end of this note you should be able to:
- define primary and secondary (meta) stages;
- construct binary meta-labels given a frozen primary side;
- explain why size-model overfitting does not invent directional bias the way a joint side model can;
- write the meta-label contract for the running fixture.
In this lesson
- Primary side, secondary size
- The decision for the long–short event book
- Deliverable: meta-label contract
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.