This example shows how local correlation can reveal structure that global correlation misses. The data has positive correlation in the upper region (x > 0.5), negative correlation in the lower region (x < -0.5), and no correlation in the middle. The global correlation would be close to zero, but local correlation shows the true structure. **Try the Bootstrap t-statistic mode** to see which regions have statistically significant correlations. Values beyond ±1.96 indicate significance at p < 0.05.
Local Gaussian Correlation shows how the correlation between two variables varies across their joint distribution. The heatmap displays the local correlation at each (x, y) point, computed using a Gaussian kernel to weight nearby observations.
The green line on the right shows the average correlation at each Y value (integrated over X), weighted by the kernel density. The red line at the bottom shows the average correlation at each X value (integrated over Y).
Blue = positive, Red = negative, White = zero.
Bootstrap t-statistic: Resamples the data 200 times to estimate the standard error of the local correlation. The t-statistic (correlation / SE) indicates statistical significance. Values beyond ±1.96 are significant at p < 0.05.
Data: varying_data.parquet
Examine how the correlation between stock and bond returns varies. During normal markets, stocks and bonds may be uncorrelated or negatively correlated. During crisis periods, correlations can break down or become positive (contagion). Use the filter to compare regimes.
Local Gaussian Correlation shows how the correlation between two variables varies across their joint distribution. The heatmap displays the local correlation at each (x, y) point, computed using a Gaussian kernel to weight nearby observations.
The green line on the right shows the average correlation at each Y value (integrated over X), weighted by the kernel density. The red line at the bottom shows the average correlation at each X value (integrated over Y).
Blue = positive, Red = negative, White = zero.
Bootstrap t-statistic: Resamples the data 200 times to estimate the standard error of the local correlation. The t-statistic (correlation / SE) indicates statistical significance. Values beyond ±1.96 are significant at p < 0.05.
Data: financial_data.parquet
Explore local correlations between different feature pairs. Use the X and Y selectors to choose which features to analyze. Filter by sector to see if local correlation patterns differ across groups. Features A and B should show positive local correlation, A and C negative, while D is independent of all others.
Local Gaussian Correlation shows how the correlation between two variables varies across their joint distribution. The heatmap displays the local correlation at each (x, y) point, computed using a Gaussian kernel to weight nearby observations.
The green line on the right shows the average correlation at each Y value (integrated over X), weighted by the kernel density. The red line at the bottom shows the average correlation at each X value (integrated over Y).
Blue = positive, Red = negative, White = zero.
Bootstrap t-statistic: Resamples the data 200 times to estimate the standard error of the local correlation. The t-statistic (correlation / SE) indicates statistical significance. Values beyond ±1.96 are significant at p < 0.05.
Data: multi_data.parquet
This example demonstrates the difference between choices and filters: - **Sector (choice)**: Single-select dropdown - pick exactly ONE sector at a time Use choices when the user must select exactly one option. Compare this to Example 3 above which uses a multi-select filter for sector instead.
Local Gaussian Correlation shows how the correlation between two variables varies across their joint distribution. The heatmap displays the local correlation at each (x, y) point, computed using a Gaussian kernel to weight nearby observations.
The green line on the right shows the average correlation at each Y value (integrated over X), weighted by the kernel density. The red line at the bottom shows the average correlation at each X value (integrated over Y).
Blue = positive, Red = negative, White = zero.
Bootstrap t-statistic: Resamples the data 200 times to estimate the standard error of the local correlation. The t-statistic (correlation / SE) indicates statistical significance. Values beyond ±1.96 are significant at p < 0.05.
Data: multi_data.parquet
When Y = -0.5X² + X + noise, the global correlation is weak. But local correlation reveals the underlying structure: positive correlation where the parabola is increasing (x < 1) and negative where it's decreasing (x > 1).
Local Gaussian Correlation shows how the correlation between two variables varies across their joint distribution. The heatmap displays the local correlation at each (x, y) point, computed using a Gaussian kernel to weight nearby observations.
The green line on the right shows the average correlation at each Y value (integrated over X), weighted by the kernel density. The red line at the bottom shows the average correlation at each X value (integrated over Y).
Blue = positive, Red = negative, White = zero.
Bootstrap t-statistic: Resamples the data 200 times to estimate the standard error of the local correlation. The t-statistic (correlation / SE) indicates statistical significance. Values beyond ±1.96 are significant at p < 0.05.
Data: nonlinear_data.parquet