A classifier trained to predict a positive next-day return may be solving the wrong investment problem if the strategy enters later, pays substantial costs or exits at a stop before the day ends. Labels specify what success means. Their outcome intervals also determine whether neighbouring observations share the same market moves. This lesson uses the timestamp contract to define targets and the dependence that validation must respect.
Learning objectives. By the end of this note you should be able to:
- define a return target from feasible entry and exit references;
- distinguish fixed-horizon and first-barrier labels;
- calculate concurrency and average uniqueness for overlapping labels;
- explain why label weights are not prediction-time features;
- identify when a sample becomes eligible for model fitting.
In this lesson
- The target is an economic decision
- Fixed horizons and path-dependent exits
- Shared returns create shared information
- Weights, splits and availability do different jobs
- A label audit before model selection
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 Quantitative Research, Data and Backtesting syllabus.