What you will learn
- Rolling association
- Lagged relationships
- Common drivers
- Structural breaks
Rolling association
Rolling correlations show how measured association varies with the window. They are descriptive and can be unstable. Match returns and report window choice rather than selecting the period with the strongest relationship.
Lagged relationships
Lagged relationships require careful alignment: a feature used to predict a later return must be known beforehand. Apparent leads can arise from asynchronous closes or publication conventions rather than useful forecasting information.
Common drivers
Common drivers can move both series without one causing the other. Conditioning on additional variables may help investigate a hypothesis but introduces modeling choices and does not automatically solve causal identification.
Structural breaks
Structural breaks can make a single full-sample estimate misleading. Compare predefined subperiods or use justified diagnostics, while avoiding endless segmentation until a desired result appears.
Worked example
A market closes hours before another. Its same-date return can appear to lead the later market simply because the time windows overlap differently. The apparent predictive relation may be a timestamp artifact.
Try it yourself
Design a timestamp-aligned comparison that tests this explanation before claiming a forecasting signal.
Show the worked solution
Construct returns over matched intervals or explicitly model nonoverlapping availability times. Preserve the original close conventions and compare the result. Any remaining association still requires out-of-sample and cost-aware evaluation.
Apply this to your course project
Write a cross-asset brief using aligned data and competing explanations.
Keep the calculation inputs, assumptions and decisions with your work. Practical exercises are self-reviewed; the scored knowledge checks assess the questions shown, not an independent certification of practical competence.