To study the correlation among a group of univariate time series, the goal is to understand how they co-move over time — whether their patterns are synchronous, lagged, or independent.
Correlation captures linear co-movement on a common timeline. For misaligned, nonlinear, or lagged dependencies, use DTW or cross-correlation instead.
Steps:
-
Align Time Frames – Resample or interpolate to ensure all series share a common time index.
-
Z-Normalisation or Standardisation – Convert each series to zero mean and unit variance to remove scale effects:
-
Optionally Detrend or Difference – Remove long-term trends or seasonality to focus on short-term co-movement.
-
Compute Correlations
- Use
df.corr()for standard Pearson correlations. - Compute rolling correlations to study how relationships change over time.
- Use
-
Visualize
- Heatmap for pairwise correlation matrix.
- Network graph or dendrogram for clustering similar series.