Credit Crisis: The Movie
A short take on the current financial turmoil from the network physics/econophysics perspective

There are two correlation networks, one based on the arithmetic return of stocks over time from August 1, 2007 and the correlation network based on the daily log return of stocks.

Arithmetic return since 8/1/2007

To download the 17MB AVI for full-screen viewing and storage: Click here

For a shorter 1MB preview that increments in months instead of days Click here

Daily log return

To download the 24MB AVI for full-screen viewing and storage Click here

For a shorter 2MB preview that increments in months instead of days Click here

Published! Journal of the Korean Physical Society, Vol. 54, No. 6 p. 2460-2463 Click here

I have copyright to these but feel free to distribute them with credit for non-profit purposes.

Financial correlation networks and the credit crisis.

Note this is for informational and educational purposes only, so due to questions I have received, please see the FAQ below the images: 

Arithmetic return from 8/1/07 network / Daily log return network


August 1, 2007

August 10, 2007 (the beginning)

December 20, 2007 

January 15, 2008

May 15, 2008

July 10, 2008

September 15, 2008 (post-Lehman)

October 10, 2008 (biggest DJIA point drop in history)


What is this?

Basic explanation: This is a network showing the correlation (similar rise and fall movements) among the stocks in the S&ampP 500 and NASDAQ-100 using the latest stocks in the index (as of 10/10/2008). The abbreviations are the ticker symbols. In a network, you have nodes (the dots) and edges (the links) which I created by connecting stocks (nodes) based on their correlations. I then wrote a program and used the software pydot to color the nodes based on their return (excluding dividends) from August 1, 2007. If they had a return greater than or equal to -10% the nodes are green, less than -10% but greater than -25% yellow, and less than or equal to -25% red. I made several charts of this for each trading day and connected them into an AVI animation.

Technical Explanation: This is a network of undirected edges which is based off of an adjacency matrix derived from the correlation matrix of the closing prices of stocks throughout the S&ampP 500 and NASDAQ-100 indices from August 1, 2007 to October 10, 2008. The network is based on a minimum spanning tree which was created from using an adjacency matrix of weighted edges between the stocks. The weight is a distance metric based on the Pearson correlation coefficient between two stocks, i and j, where d = sqrt(2(1-cij)) and cij is the correlation of i and j. This is taken from the idea of correlation distance for stocks by JC Gower (J. C. Gower, Biometrika 53, 325 (1966)) and later the idea of correlation networks by Rosario Mantegna(R. N. Mantegna, Eur. Phys. J. B 11, 193 (1999)).

The correlation is the correlation over the entire time series and is not recalculated over time so the graph remains stable. Correlation for the two series of animations and images was calculated in two ways. For the first (the more narrow branched structure) correlation was based off of the daily arithmetic return of stock closing prices versus the close at August 1, 2007. The second was based on daily log returns. If the stock share price return had a return (minus dividends) greater than or equal to -10% the nodes are green, less than -10% but greater than -25% yellow, and less than or equal to -25% red. I made several charts of this for each trading day and connected them into AVI animations. Stock splits are accounted for.

Can I use this to trade stocks?

Above, I said this was for educational purposes only. Please feel free to email me with any questions about the plot but I do not give financial advice.

You used the correlation for such a long time but doesn’t correlation increase with volatility?

Yes, it does. This has been proven in many works. However, this is just a simple illustration (though there is a paper which gives more detail) so I do not update the correlation for increased volatility. If I were to create this using more recent market fluctuations than it would definitely show a much different graph due to the stronger correlations among a wider variety of stocks.

Is this an example of financial contagion?

I will preface this with the statement I have been using primarily my knowledge of physics and finance (from business school and the work world) and not high level economics or portfolio theory to create this. Financial contagion is officially defined as the spread of asset price collapses across national borders (like the 97 Asian crisis and the current de-decoupling) and since these are US stocks it really can’t be an example of this. I am working on some international stock index comparisons as well though that may be posted in the near future.

Is this a model like infection disease epidemiology?

Well, it looks like it but looks can be deceiving. The spread of stock price collapses is not due to stocks spreading a contagion among themselves like in human or computer epidemiology models, but rather because the systemic instability and risk have fallen off of a cliff and have allowed this type of cascading collapse to happen. Also derivatives and all sorts of other factors affecting asset prices aren’t in here so it is not a complete (though interesting) picture. 

Do you work in finance?

No, I actually have found my calling in supply chain in manufacturing. Things like this are a side hobby (though supply chain and logistics network analysis is increasingly important).