The Default Line: THE INSIDE STORY OF PEOPLE, BANKS AND ENTIRE NATIONS ON THE EDGE (33 page)

In the wake of the crisis, central bankers around the world have responded to a campaign against Gaussian thin tails by the former-trader-turned-probability-philosopher Nassim Taleb. Taleb noted wryly that Friedrich Gauss and the bell curve itself were the images printed on the last German 10-mark note before it was replaced by the non-existent bridges of the inane 5-euro note in 2002. Taleb makes the point that the exchange rate on the 10-mark note fluctuated between 4 per dollar to 4 trillion per dollar in the hyperinflation of the early 1920s, and so was the very illustration ‘that the bell curve is meaningless as a description of the randomness in currency fluctuations’. (Intriguingly, though, the postwar Deutschmark went from 4 per dollar to 2 per dollar in the five decades of its existence – a rather more stable career. For the thirteen-year period that the occasionally maligned Gauss was actually on the DM10 note, the Deutschmark went from 1.8 per dollar to 2 per dollar. Gauss’s image turns out to have been a stabilising influence on that note – or perhaps Taleb himself was fooled by the random data point.) Taleb told me during the crisis that he had been trying to persuade the King of Sweden to rescind the Nobel Prize for Economics, on account of the damage done by its winners to mankind.

To the extent that the world’s central bankers can understand what Taleb is saying, they are listening. Anand Sinha, the deputy governor of the Reserve Bank of India, puts it best when he says that economists have ‘physics envy’. ‘The mistake has been in elevating quantitative finance to the status of physics,’ Sinha mused in Mumbai, in a speech that drew on the work of Taleb’s collaborator Pablo Triana. ‘Physics deals with the laws of nature governing the universe. The objects have unique physical attributes (i.e. position, velocity, temperature, etc.) and the universe evolves according to the immutable laws of nature. Any observation or measurement of physical attributes does not change them, or even if it does, it does so in a predictable way. During measurement interactions the results are not deterministic but follow a probability distribution which, however, is stable. On the other hand, in finance, there is no such law of financial markets.’

Where there is interaction, there is non-normality, argues Haldane – as LTCM spectacularly found out in 1998 when its Nobel laureate-inspired quantitative strategies blew a $4 billion hole in its finances. The model could not account for the influence of the model itself in altering the market. Formulae forged in observation, when deployed as a trading strategy, immediately changed the basis of those formulae. It was a VaR model.

As Sinha argues, ‘The “values” of assets are not inherent attributes of the financial instruments and the economic agents are not outside observers of the financial system. In fact, it is the human mind, its ambitions, drive, competitiveness, caprice and greed which drive the actions of the economic agents, and it is these actions which determine the value of the financial instruments. Thus, unlike in physics, in finance it is the observers who provide value to the financial instruments. There is no unique value: it is determined by the collective psychology of economic agents.’

So the normal distributions were wrong, ignoring the fat tails was wrong. Even if they had been right, there was insufficient historic data upon which to base price patterns. And measures like VaR in any case ignore systemic risk. But it doesn’t end there. In fact, away from the flaky trading books, back on the bank loan books, things are getting rather strange indeed. Abnormal, even.

There is a point to this whistlestop tour through correlation trading, Gaussian copulas, credit derivatives, earthquakes, dodgy trading books, probability philosophers and bell curves. This is a world of risk and uncertainty begging for quantification and false comfort. As Friedrich Hayek said in his Nobel Prize lecture of 1974, entitled
The Pretence of Knowledge
, ‘While in the physical sciences the investigator will be able to measure what, on the basis of a prima facie theory, he thinks important, in the social sciences [such as economics] often that is treated as important which happens to be accessible to measurement.’

None of which is to say that attempts to model and measure risk should be condemned. Some banks launched simple models, and then made them better. They skewed their normal distributions. They used a variety of models. They based their risk judgements on data that stretched back before the boom. Unfortunately, many banks saw these flawed models as a target to hit rather than as a guide to their safety.

So how did this soup of mismeasurement and misspecification come to be hardwired into the rules that seek to ensure the solvency of every major bank in the world? New rules for bank capital requirements, Basel II and Basel III, have emerged over the past decade. At their heart they make use of Vasicek’s work. For the trading book, it is the VaR method, as discussed already. For calculating possible losses on their traditional loan book, banks can volunteer to use the ‘advanced internal ratings basis’ (AIRB) for calculating capital, which was identified in 2005 as a modified version of the Vasicek model. I showed this to the man himself.

Oldrich Vasicek is flabbergasted on a number of levels. ‘This loan loss distribution [the ‘Vasicek distribution’] was developed strictly for corporate loans under some restrictive assumptions,’ he says. ‘It was not intended to address mortgages (which are quite different in structure), credit card loans, or any other types of loans to which Basel II applies it.’ Remember the important role of both the probability of default and the correlation? The Basel formulae blend these two factors. ‘The distribution has two principal parameters [the probability of default and the asset correlation] that are completely independent. An attempt to derive one from the other, as Basel II and III do in their formulae, is completely unjustified and, frankly, outright silly. I am appalled by that arbitrary decree.’ A meaningful estimate of these two parameters for each loan portfolio is necessary, Vasicek says, ‘in order that the portfolio loss distribution has any applicability. It is a lot of calculation, but computer time is cheap. Without it, it is GIGO [garbage in, garbage out].’

He acknowledges that questions remain about the applicability of the shape of the distribution. Andrew Haldane, for example, has written that ‘Standard applications of the Vasicek model assume that underlying risk models, and hence portfolio losses, are normally distributed.’ Crunching an erroneous normal distribution of GDP through the model would lead to the need to hold a 3 per cent capital buffer on a typical loan portfolio. If you use the actual fat-tailed distribution of GDP in the same equation, then the required capital buffer shoots up to 12 per cent.

This creates something of a mystery. It is true that the asset values, the enterprise value of a firm, used in the Vasicek formulae are distributed normally (actually log-normally). But the output of this formula is ‘very very non-normal’, says Vasicek, and always has been from the beginning. The precise shape of the curve was also corroborated by one of the world’s most important regulators, using masses of data in a chart that has never been published. ‘The portfolio loss, derived from this assumption, is actually extremely non-normal,’ says Vasicek. ‘To cover, say, one in a thousand possible cases, the bank needs to hold an amount of capital equal not to the expected loss plus about three times the standard deviation, but to the expected loss plus some ten or twelve times standard deviations.’ Basically, using the original formula would have required banks to hold massively more capital. Somewhere in the process of applying it to the capital requirements of banks, the formula became ‘normalised’.

‘No one in the Basel task force ever consulted me on the applicability and proper use of my formulas,’ Vasicek concludes. ‘In fact, I learned only much later that these equations were incorporated in the regulations. Moreover, Basel makes changes to my formulas that I very much disagree with.’

So the original author of the models that currently determine the safety of the world’s biggest banks says that those models are wrong. This is exacerbated by how banks in practice have responded to Basel II.

One small piece of proof for this comes from the minutes of the HBoS board on 24 June 2003. ‘“Advanced” status was the only credible status for HBoS. “Advanced” banks would have the capacity to undercut competition on chosen tranches of business, with cost of capital being a key strategic weapon,’ said the minutes. In other words, gaining AIRB status, using some version of the Vasicek formula, would enable HBoS to hold much less capital than its competitors. In evidence to the Tyrie Commission, the HBoS leadership team admitted that ‘tens of thousands of hours’ were spent by HBoS staff trying to secure the Basel II waiver. Ex-chief executive Andy Hornby said it was a ‘huge distraction’. HBoS was not granted the waiver in June 2007, but Northern Rock, its key mortgage competitor, did get it, just months before its collapse. HBoS eventually got the waiver in December 2007.

Basel II’s formulae should give the answer to the question: ‘When I have lent this much, how much capital do I need?’ The input should be the portfolio of loans, and the output should be the size of the capital buffer required to keep that bank safe. When I discussed this with Professor Hyun-Song Shin of Princeton University, he made the following point: ‘In practice, the banks ask “If I have this much capital, how much can I lend?” The same Vasicek formula gives the answer to both questions, but the second question is a recipe for a credit boom when the financial market becomes more tranquil. What’s worse, the greater lending by the banks leads to further compression of spreads and other measures of risk, which induces the banks to lend even more. So, Basel II made banks and the financial system much more procyclical and prone to booms and crashes.’

In other words, this mis-specified simplification of a formula was reverse-engineered by regulators and bankers, and used inevitably as a target, rather than as one of a series of constraints to ensure bank safety.

Many questions remain – above all, why?

The chief executive of one of the world’s biggest banks took me through how his team calculates its capital under the formula, by sketching out various expected loss, curves on a piece of paper. ‘Conventional risk modelling is very bad at calculating the risk on mortgages,’ he says. ‘The reason it’s so dangerous is that if you have a very low expected loss, the models generate a very low-risk weighted asset. You’re sitting there with the thing that is fine most of the time, but when it blows up there’s no way you’ll have enough capital for it.’

Vasicek himself confirmed that this reading of his actual formula was correct – unlike the Basel derivation. ‘A portfolio with low probabilities of default and high correlations may need as much or more capital to support it as a portfolio with high individual probabilities and low correlations,’ he told me. ‘That is what my formulas are supposed to quantify.’

It is clear there was repeated interference with the Basel formula from bankers and governments to help channel lending into mortgages. A study by Manchester Business School showed that the advanced internal ratings basis (AIRB) formula ultimately results in banks putting insufficient capital aside to cover a systemic mortgage and property crisis, while at the same time putting too much aside to cover the risks in lending to businesses. American researchers have blamed German lobbying. Europeans have blamed the US Federal Reserve. George Christodoulakis, author of the Manchester study, says, ‘The Vasicek model was adopted because of analytical tractability – [it is the] only structural model that produces a formula. Vasicek’s work is great – [it] contributed very important knowledge, but it was the starting point. But Basel adopted it, made very strong assumptions, misused it and misunderstood it.’

One of the world’s most lauded and respected investment bankers and the inventor of many derivative technologies says that governments skewed finance towards mortgages on purpose to keep house prices high and voters happy. ‘That’s always been the case,’ he told me. ‘I’d go as far to say it was an objective of all regimes. Because governments like people to own houses. That’s why most credit crises comes out of the mortgage market.’

In effect the formulae, as applied by banks, bias lending towards mortgages and away from business. To some degree this would always be the case. The original 1988 Basel formula, Basel I, attached an arbitrarily low-risk value to mortgages and sovereign debt, but a high risk to business lending. Bankers either want exhaustive amounts of information about your ability to repay a loan, or they want collateral. Property is the strongest form of collateral. Yet what countless studies of the Basel model show is that its formulae go further than this. Property has low specific or idiosyncratic risk, but very high systemic risk. Mortgages rarely go wrong, except when they really go wrong, which can cause an epic existential disaster for entire banking systems.

The way these formulae have been applied, therefore, amounts to a form of interference in the banking system, helping divert lending from productive to unproductive purposes. They show that at the very heart of the banking system there is a continuing problem: bankers deliberately deploy complexity to confound and bedazzle, with the aim of extracting hefty bonuses from profits that turn out to be illusory. This involves the concentration and multiplication of credit risk, rather than the more sensible strategy of spreading risk around the banking system. Most people know that already. But there is also a high-level political interference in the banking system, dressed up as science. In fact it is no more than a dangerous extrapolation of past losses using faulty models. If interference like this has to exist, why not tweak the core of the banking system towards useful production rather than useless credit bubbles?

Mis-specified credit derivative models have been a significant factor in the default of households, banks and whole nations. Even assuming the bankers
had
got everything right, the crisis would still have occurred – for the simple reason that regulators miscalculated risks. At the same time they created regulations that motivated risk-taking by banks. But it was the pseudo-science behind credit derivatives that created the shadow banking system.