Put away the crossword and the sudoku:  it’s the “credit spread puzzle” that’s occupying some leading financial minds.  On May 3, 2012, Prof. Joost Driessen of Tilburg University spoke to a Global Association of Risk Professionals (GARP) webinar audience about recent work done by his research group to solve this puzzle.

The term “credit spread puzzle” refers to the fact that credit spreads are much higher than can be justified by historical default losses.  A typical example Driessen cited was a long-term AA bond that had an expected default loss of 0.06% yet whose average credit spread, calculated using real-life data, was 1.18%.  More about the spread calculation later.

Over the time period Driessen surveyed, from 2004 to 2008, the credit spread profile looks similar to the landscape of Cupecoy, St. Martin (pictured here) with the beach extending from 2004 to mid-2007 and then abruptly the white cliffs of super-high spreads jutting up when the financial crisis hit.  The expected default loss did not change significantly throughout this period but suddenly the expected return on bonds shot up. [Note:  The expected return is calculated as the difference between credit spread and expected loss.]  

Driessen used an asset pricing approach to examine liquidity effects in corporate bond markets.  (The grandaddy of all asset pricing approaches, the CAPM model, is an asset pricing approach.)  The Driessen model is a multi-factor asset-pricing model that includes liquidity effects and risk premia.  I can’t do justice to the details of the work of Driessen et al.; you can read about it in on the www.joostdriessen.com site under the “Working Papers” link.  They concluded that the credit spread puzzle can be explained when liquidity is taken into account as an additional characteristic of pricing.

There was a lot of discussion during and after Driessen’s talk about “slicing” the data to construct different portfolios.  What caught my interest was the nifty way his group teased out information on the actual costs of the bid-ask spreads of the corporate bonds.  They used the TRACE database that covers all trades made in US corporate bonds.  They looked up all “paired” transactions of the same size for the same bond made on the same day.  By inferring that one transaction involved a customer-to-dealer buy and the other involved a dealer-to-customer sale, they were able to collect thousands of paired datapoints for exact bid-ask transactions, and in this way they determined the actual cost of the liquidity problems during the financial crisis.  (Driessen et al. are not the first to calculate the actual liquidity cost using paired trades.  They cite the technique as reported by Dick-Nielsen, Feldhütter and Lando, in a 2011 Journal of Financial Economics paper on corporate bond liquidity.)

It’s terrific to have a lot of data, like the TRACE database does.  But sometimes there’s too much raw data, and it’s tough to see the relevance to the problem you wish to solve.  As data work-ups go, this was beautifully simple.  It was a hunt for points that followed a modest rule.  But it was an insightful rule to search on, and one that has already borne fruit for Joost Driessen and his collaborators.  ª

Click here for the webinar presentation.

Click here for the research paper it’s based on. 

The striking cliffs at Cupecoy are from the Where To Stay website.