Fig 1.
Pseudo code for the described topological filtration methods.
(a) Minimum Spanning Tree, (b) Planar Maximally Filtered Graph, (c) Triangular Maximally Filtered Graph.
Fig 2.
Degree agreement between the full and filtered correlation matrices with different values of p for the windows with maximum volatility.
(a) US, (b) UK, (c) DE, (d) IN, (e) CH.
Fig 3.
Degree agreement between the full and filtered correlation matrices with different values of p for the windows with minimum volatility.
(a) US, (b) UK, (c) DE, (d) IN, (e) CH.
Fig 4.
Degree agreement between the full and filtered correlation matrices with different values of p for the windows with median volatility.
(a) US, (b) UK, (c) DE, (d) IN, (e) CH.
Fig 5.
Agreement between edges between the full and filtered correlation matrices with different values of p for the window with maximum volatility.
(a) US, (b) UK, (c) DE, (d) IN, (e) CH.
Fig 6.
Agreement between edges between the full and filtered correlation matrices with different values of p for the window with lowest volatility.
(a) US, (b) UK, (c) DE, (d) IN, (e) CH.
Fig 7.
Agreement between edges between the full and filtered correlation matrices with different values of p for the window with median volatility.
(a) US, (b) UK, (c) DE, (d) IN, (e) CH.
Fig 8.
Correlation between degree centrality in the full and filtered networks over the dataset.
(a) US, (b) UK, (c) DE, (d) IN, (e) CH.
Fig 9.
Edge agreement between the full and filtered networks over the time period of the dataset.
(a) US, (b) UK, (c) DE, (d) IN, (e) CH.
Table 1.
Spearman correlation between the degree and edge agreement measures and the volatility of the market.
Correlations insignificant at p < 0.05 are in italic.
Fig 10.
Accuracy when distinguishing between correlation matrices created from the stock windows with the maximum and minimum volatility using the SF method and a random forest classifier with US stock returns.
(a) US p = 50, (b) US p = 100, (c) US p = 150, (d) US p = 200.
Fig 11.
Accuracy when distinguishing between correlation matrices created from the stock windows with the maximum and minimum volatility using the SF method and a random forest classifier for UK stock returns.
(a) UK p = 20, (b) UK p = 40, (c) UK p = 60, (d) UK p = 70.
Fig 12.
Accuracy when distinguishing between correlation matrices created from the stock windows with the maximum and minimum volatility using the SF method and a random forest classifier for Germany.
(a) DE p = 5, (b) DE p = 10, (c) DE p = 15, (d) DE p = 20.
Fig 13.
Accuracy when distinguishing between correlation matrices created from the stock windows with the maximum and minimum volatility using the SF method and a random forest classifier for India.
(a) IN p = 10, (b) IN p = 20, (c) IN p = 30, (d) IN p = 40.
Fig 14.
Accuracy when distinguishing between correlation matrices created from the stock windows with the maximum and minimum volatility using the SF method and a random forest classifier for China.
(a) CH p = 10, (b) CH p = 15, (c) CH p = 20, (d) CH p = 25.
Fig 15.
Accuracy when distinguishing between correlation matrices inferred from adjacent time windows based on the median volatility using the SF method and a random forest classifier for US stock returns.
(a) US p = 50, (b) US p = 100, (c) US p = 150, (d) US p = 200.
Fig 16.
Accuracy when distinguishing between correlation matrices inferred from adjacent time windows based on the median volatility using the SF method and a random forest classifier for UK stock returns.
(a) UK p = 20, (b) UK p = 40, (c) UK p = 60, (d) UK p = 70.
Fig 17.
Accuracy when distinguishing between correlation matrices inferred from adjacent time windows based on the median volatility using the SF method and a random forest classifier for German stock returns.
(a) DE p = 5, (b) DE p = 10, (c) DE p = 15, (d) DE p = 20.
Fig 18.
Accuracy when distinguishing between correlation matrices inferred from adjacent time windows based on the median volatility using the SF method and a random forest classifier for Indian stock returns.
(a) IN p = 10, (b) IN p = 20, (c) IN p = 30, (d) IN p = 40.
Fig 19.
Accuracy when distinguishing between correlation matrices inferred from adjacent time windows based on the median volatility using the SF method and a random forest classifier for Chinese stock returns.
(a) CH p = 10, (b) CH p = 15, (c) CH p = 20, (d) CH p = 25.