Wednesday, 13 October 2010

What Does Fitch's Study on CDS Spreads as Predictors of Default Tell Us About Modern Finance Theory?

Fitch released a report "CDS Spreads and Default Risk Interpreting the Signals" which has received a good deal of coverage in the press. One example from the Financial Times.

Analyzing five property-sensitive sectors in USA market in the wake the recent housing crisis, Fitch found that all five of these sectors had wild swings in their derived default risk. For example, USA REITs went from a CDS-derived probability of default ("PD") of 0.7% at June 2007, to 10.1% in October 2008, to 18.0% in March 2009, and then to 4.4% in August 2010. Details on the other four sectors are in the Table on Page 3. Despite these highly elevated PD's, actual defaults did not increase as predicted. The derived PDs were inaccurate. Fitch cites the volatility of CDS markets and the tendency for directional momentum (that is, an imbalance of demand and supply) as the cause of these false positives.

There are two key conclusions.  

  1. CDS-derived PDs can give potentially erroneous and costly portfolio management signals. 
  2. More importantly, as shown in the Table on Page 9, if the derived PDs are used in Basel II IRB models, financial institutions would be forced to dramatically and needlessly increase provisions at precisely the wrong time in a crisis. A pro-cyclical move which would depress the probability and strength of a recovery.

Fitch's study confirms what various analysts (including yours truly) have been saying about CDS spreads. The market is too thin to give reliable information. These are at best directional indications of PD. Fitch's detailed empirical analysis on this topic is therefore highly welcome and useful.

But there's more here.  Fitch's report raises fundamental questions about current finance theory:  both the construction of models and the use of market data – whether direct or derived –  as inputs to those models during periods of crisis. And so challenges some of the fundamental assumptions of corporate finance orthodoxy about market prices – both at the macroeconomic and microeconomic levels.

The simple empirical fact is that during periods of stress or exuberance, markets are dysfunctional. Prices no longer reflect, if they ever did, intrinsic values. During a crisis, there is a dramatic increase in liquidity preference motivated usually by fear. We see this most clearly in the breakdown of "normal" correlations among markets and asset classes. During a boom, a dramatic decrease in liquidity preference motivated by greed and irrational exuberance. 

This happens not only in illiquid markets like those for CDS but in the most liquid markets. In the period after 9/11, the NYSE plunged dramatically. While the attack was horrific, our way of life was not under serious threat. Our economy was not in danger of being destroyed, particularly by a relatively small band of cave dwellers in Afghanistan and Pakistan. For that task it would need and subsequently got some timely domestic help.

In such circumstances as in booms, the usual assumption about what market equilibrium means has to be thrown out the window. When markets are not rational, in no sense do their prices reflect intrinsic or fair values. Using the values they give – potential inputs into our "sophisticated" models – makes little sense. Market equilibria are much more unstable than during more "normal" times.

But, even if we assume that markets continue to function in such periods, we are misled by another myth: the imaginary no profit equilibrium derived from microeconomics. As this theory goes, intense competition leads sellers to lower their price until goods are sold at cost. Now, I recognize this is the one sacred doctrine on which all or nearly all the various economic cults agree. They may dispute vociferously with one another over which is the sole efficacious economic sacrament – the gold standard, the quantity theory of money, deficit spending, tongue of Newt, tax cuts etc  But on this issue there is by and large doctrinal unanimity. 

However, back in the real world, I don't know many businesses that price at cost. Or that stay in business if they do. Yet, we derive our financial models based on this illusion.   Often we use it to derive inputs for those models.  As you'll notice from Fitch's report, the standard equation for deriving PD from a CDS is to take the CDS spread in bps and divide it by the presumed loss on default (LGD) in percentage terms. The result is the PD expressed in percentage terms.  So a 100 bps CDS spread with a 50% LGD turns into a 2% PD. 

As you'll notice, this calculation assumes that the seller of risk protection is content to receive as his compensation exactly the amount of his expected loss. This is far removed from the standard microeconomic theory of equilibrium, which would have the seller's price result in an overall break even position. Here the seller does not recover his operating costs – salaries and other expenses which I believe it would be safe to say are not small in most investment banks. One might, I suppose, argue that in normal non crisis or non boom markets we can ignore these costs because they are spread over large volumes of business. Perhaps.

But in a crisis or in a boom where there is excess demand for a product, I'd expect any rational business man or trader to take advantage of supply/demand dynamics and increase his profit margin. which as I've argued above was not at "zero" from "normal" times.  More importantly, in a crisis where an institution and a trader are assuming an ongoing risk (like a CDS), I'd expect there to be a strong incentive - both  for the trader and his firm - to price up to cover that risk taking. The personal and institutional consequences of a wrong bet can be rather serious – just ask AIG.  So we should rationally assume that during a crisis CDS spreads include  not just the protection seller's  objective best estimate of the PD and LGD but also a fear/caution based adjustment of those factors plus a rather hefty profit margin.  It's not hard to imagine a seller demanding profit margin well in excess of 50%.  
As before, in line with best doctrinal thinking, we're ignoring costs. Clearly, the profit margin and the additional fear induced safety margin in pricing are going to be a major component of the pricing.  All of which of course explain why often CDS spread-derived PDs are greater than 100%. And why using them in models makes scant sense because the resulting PDs are inflated.  Probably by significant factors. Not percentages that would be considered "normal" tolerances.

We like to think that we are more advanced than our predecessors. We have elegantly constructed apparently "sophisticated" and "scientific" models. Finance theories are expressed with imaginary mathematical precision. 

Yet at the very heart of these models are assumptions that would make a medieval scholastic blush. 

Assume a market with perfect information and no transaction costs and you will discover, perhaps - but hopefully not - to your surprise, that it turns out to be an efficient market. Something I believe has to do more with logic than economics: the principle of tautology.    

Assume that the market price reflects intrinsic value and you are highly likely to input the silliest numbers as variables into your model. 

What's needed at the core of this discipline like any other is a healthy does of skepticism, a constant challenging of revealed assumptions and a cold hard eye on results. 

No comments: