The event classifier can keep ranking matured outcomes while the book loses money, or the book can make money while the score is no longer the thing that was validated. This note writes the monitoring contract for that classifier from drift, decay and retirement: a reference population, one smoothed diagnostic, and a response that does not pretend the alarm found the cause.
Learning objectives. By the end of this note you should be able to:
- score an event only after its outcome has matured;
- compute the source's exponentially weighted diagnostic on a short cost path;
- choose investigate, cut risk, or retest without treating the threshold as a significance level;
- state why an automatic refit is a new model version.
In this lesson
- Score and P&L are different series
- One smoothed diagnostic
- The contract for this classifier
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.