I wanted to take a look at how the correlation matrix of the S&P 500 has changed over time, but with close to 125k unique pairwise correlations, direct comparison of each matrix is difficult. In the past I showed how to visualize correlation of a large number of variables using Correlation Filtered Graphs, thus I wanted to see what a graph of the S&P 500 correlation matrix would look like as time marched on.
The charts start with the correlation matrix from October 2008 to 2009 and advance forward by one year until the present. One thing that is striking is just how tightly coupled price dynamics were in 2009, 2010, and especially 2011. Its an old trading rule that in a crisis correlations go to one. This translates visually into highly connected graphs during the crisis years immediately following Lehman’s collapse.
After 2012 the graphs become disconnected, reflecting the drop in overall correlation. This was an environment where individual stocks traded less like a herd and where stock picking seemed to matter much more. The last chart is the current year: while stocks are not nearly as coupled as in 2009, the distinct clusters are starting to merge as volatility creeps back into the equity market.
You can reproduce and explore these graphs for yourself using SliceMatrix. Make sure you enter the following settings:
- Threshold: Low = 0.67, High = 1.0
- Layout: Gravity: 1.0, Charge: -2000
- Zoom: 0.25
- Labels: Hidden
- Set the appropriate Time Range and Regenerate the Graph
Categories: Quantitative Trading