An overemphasis on risk models was the key flaw in the regulatory architecture

The future is not unknown, but unknowable.
John Maynard Keynes

No amount of modelling will reveal the future to us.

Risk models in banking may be useful thought experiments, but they have proved to be lousy predictive tools. There are very good reasons for this. Most economists and risk professionals would generally accept that risk outcomes are both dynamic and conditional, that is that the sum of the risks taken by a bank are dependent on the correlation between the various risks taken by the bank and that these correlations vary depending, primarily, on where we are in an economic cycle. Risk models that actually recognise dynamic conditional correlation are rare, not least because the maths can become very intractable, unless the path of the correlations is described in relatively simple terms (by, say, the use of a Gaussian Copula).

The more fundamental issue is whether the correlations are likely to be stable at all.

Andrew Sheng, the chief adviser to the China Banking Regulatory Commission, said in an interview that appeared in the November 2010 issue of Central Banking that financial stability policy failed because it did not quantify the right risks. "Are we dealing with a puzzle or are we dealing with a mystery? A puzzle is something where if you find the right information, then you can find the right solution. A mystery is something with unknown unknowns. We may never have the knowns because markets are about interactive behaviour," he said. In such a situation where there are unknown unknowns, it follows that there will be no stable correlations.

It does seem very likely that given the differences in market structure and in the role of institutions (such as banks) between developed and emerging markets that correlations will not be the same across asset classes for these different economies. It also seems likely that given the way society, economies and institutions change over time, correlations will also change over time. As an example of this, we only have to see the way that the development of new high risk-traded asset classes resulted in the need for trading institutions to acquire funding by selling low-risk assets during the recent crisis. This led to some assumed negative correlations turning positive. It also produced some unexpected second-order effects as very low-risk government bonds became infected as some governments absorbed failed banks' asset risk.

When it gets to asset correlations, certainly in highly stressed circumstances (which is when they really matter), we thus are probably dealing with a mystery.

This is not to say that modelling is useless. Banks will wish to understand their risks and capitalise against them all of the time, so improving risk models is an endless quest for efficient capital allocation. In such circumstances, risk models have their uses.

For regulators, however, the focus of their work should be fundamentally different.

Regulators must be primarily interested in ensuring banks are as secure as possible in highly stressed circumstances and that, if they do fail, they do so in a manner that causes the least disruption to financial stability. And, in such circumstances, it is likely that we are dealing with unknown unknowns.

Yet regulators, too, were seduced by risk modelling.

Basel II was, until recently, the high watermark of risk-based regulation, supported by a massive new risk management "industry" based largely on statistical method and econometric models and with strong roots in some quite controversial economic theory such as the efficient markets hypothesis.

Certainly the regulatory authorities, as well as the banks, developed in Basel II a comprehensive approach to bank capital adequacy, formed around the modelling of banking risks where possible (Pillar I and models), the assessment of other risks and a comprehensive supervision process (Pillar II) and disclosure requirements sufficient to populate the modelling and supervision processes (Pillar III). Comprehensive as this was intended to be, liquidity plays almost no role in the system.

Basel III simply adds more risk management and more capital (plus a balance sheet gearing measure and a liquidity regime for good measure, both of which look, at the current time, very much like afterthoughts).

The failure to appreciate the difference between the needs of risk managers in banks and their regulators is fundamental to the future of both risk management and regulation, not because it points them in the same direction but because it points them in different directions. Banks' shareholders should be interested in an economic capital but regulators should be more interested in banks' liquidity and in the existence of an efficient resolution regime that ensures that the failure of any individual bank does not threaten the financial stability of the banking system as a whole. It is only these changes to regulation and not those encompassed by Basel III changes to risk-based regulation that will most improve the safety and soundness of the banking system.

Brandon Davies is a member of the Financial Markets Group at the London School of Economics.