In a previous article (December 10, Goodhart’s Law and Monetary Policy) we discussed Goodhart’s Law which says that once an observed empirical relationship begins to be relied upon, it will no longer work. In that article we applied the law to monetary policy.

Goodhart’s Law also applies to bank capital regulations.  The Basel Committee on Bank Supervision is an international body that recommends standards for minimum bank capital levels.  According to Basel I, bank assets are weighted according to their perceived riskiness, with cash and government bonds (sovereign debt) receiving a zero risk weight and business loans 100% risk weight.  By overweighting their balance sheets with low risk weight assets, banks can operate with very low equity capital.  For example, Deutche Bank AG (stock symbol DBK) has approximately $3 trillion in assets and equity capital of just over 2 percent of assets.  Yet, DBK’s regulatory capital ratio is reported (September 30, 2011)¹ at 13.8% because risk weighted assets are just 15 percent of total assets.  For the sake of DBK shareholders, one hopes that DBK’s low risk weighted asset calculation is not driven by large investments in sovereign debt of fellow EU countries like Greece, Italy, Portugal and Spain.

Closer to home, another unfortunate selection of risk weight was 20% for highly rated (AAA) mortgage backed securities.  Assuming an 8% Tier 1 capital requirement, this means that Tier 1 capital (mostly equity) must be just 1.6% (=.20*8%) of AAA mortgage securities.  This created an enormous demand for AAA tranches and spawned an explosion in financial innovation.  The basic objective was to produce as many AAA securities out of pools of non-agency loans (securities created from agency loans, those guaranteed by an agency of the U.S. government, were automatically AAA).  The method to accomplish this was subordination and credit tranching.  All losses in the pool of loans were directed to the owners of the equity or “first loss” tranche until that piece was wiped out.  Then, the losses were directed to the “mezzanine” tranches until they were wiped out.  Only then were credit losses applied to the “senior bonds.”  Rating agency models determined for each pool of loans the degree of subordination required so that the senior bonds would be awarded AAA ratings.  Typically, 80% or more of the principal value of a non-agency security would be awarded the AAA rating.  Investors viewed AAA rated bonds as essentially risk free, and there was huge demand for those bonds during the housing boom. 

Paradoxically, the effect of the huge demand for AAA bonds drove underwriting standards lower.  This would not have been disastrous had the ratings agency models taken account of changing underwriting standards and raised subordination levels accordingly.  But they did not.  Perhaps the problem was Goodhart’s Law.  Rising home prices during the boom reduced foreclosure rates and loan losses, even as standards were weakening.  If credit models are trained on historical data, they would tend to progressively generate more benign forecasts during the boom, even as the risk was rising.  As it turned out, the rating agency models underestimated the probability of a severe and widespread drop in housing prices.

Thus, capital regulations that were carefully designed to accurately measure and control risk taking wound up being used to accumulate ultra-risky positions which have failed or are in the process of failing.  This is Goodhart’s Law morphed into Murphy’s Law.

What are solutions to the challenges suggested by Goodhart’s Law?

One answer is simply to be careful about relying on historical relationships.  Not only can they stop working at any time, the likelihood of failure increases as they become more popular.  So, maybe the answer is to only rely on historical relationships that are not generally known.  This means you will have to keep coming up with new ones. 

Another solution is to take great care in developing metrics and setting objectives.  If you can focus on the real objective instead of a proxy, then your reliance on the assumed relationship between target and proxy is lessened. 

A better method of assigning capital in accordance with risk levels is to explicitly measure expected and unexpected losses.  And the best method for doing this is stress testing.  More on this in another blog.

¹Simon Johnson, “Deutsche Bank Could Transfer Contagion,” Bloomberg News, 2011.