Conclusions

Key findings, limitations and next steps

Raw weather-CCI evidence is weak

Pearson, Spearman, lag and seasonal t-tests do not support a direct linear CCI story.

Retail is more predictable

Random Forest reaches strong retail performance once weather and macro controls are combined.

Macro context dominates

Inflation and unemployment carry major signal; weather contributes but is not the sole driver.

Claims remain associative

The project does not claim causality; it builds a transparent predictive and statistical analysis.

Future Work

Time-series validation

Use rolling-origin or blocked time splits instead of shuffled K-Fold.

Sub-national data

Add city or regional retail proxies to reduce country aggregation bias.

Causal design

Use event-study or instrumental-variable logic before making causal claims.

Thank you for exploring the project

Weather may not move confidence in a simple straight line, but the full pipeline shows how data science can separate weak direct evidence from stronger predictive structure.

View on GitHub