by Mark Kirkland, VP Treasury, Bombardier Transportation
The causes of the credit crisis of 2009 will be discussed by many for numerous years to come, although probably for fewer years than we now think. People have a unique ability to forget, perhaps black out, the worst episodes. I have sat down on a number of occasions and tried to think, what were the possible causes of the crisis? An inherent weakness in accounting of results, large numbers of over the counter derivatives with large fair values, weak governance by regulatory bodies or even that bankers were paid too much? In the end, I believe that none of the above was a key contributor to the crisis. In my mind there are two unrelated causes.
It is now clear that very few shareholders of banks understood the risks thtat some banks were in fact taking.
The first is the mode of compensation in the financial industry. Not the amounts. Most bankers receive a kind of option pay out. If the firm makes a large profit (based on the mark to market of future uncertain cash flows), the employees receive large cash bonuses. If the firm makes a loss, in the worst case, staff may receive no bonus. Clearly, for a betting man, this gives carte blanche to load up the company with significant risk. Since most bonuses are not discussed with the owners of the company (the shareholders) but set by a compensation committee, often chaired by senior employees, there is a tendency to overpay since this justifies the compensation of the very people making the decisions. I will not dwell on this cause much longer – except to stress that the whole model encourages large risk taking.
The second is the point of this article. Risk was and still is, very badly understood, managed and reported. It is now clear that very few shareholders of banks understood the risks that some banks were in fact taking. In part, this is because disclosure of risk is unclear. A more fundamental issue, however, is that it appears that some of the banks did not fully comprehend the risk and actually outsourced much of their risk assessment to the rating agencies and then used flawed measures such as Value at Risk (VaR) not only to manage risk but also to report to management and shareholders alike.
A recipe for disaster?
Consider first the structured products themselves. Collateralised loan obligations (CLO), collateralised debt obligations (CDO) and even collateralised mortgage obligations (CMO) were all highly structured to maximise yield while ensuring that the most senior tranches would be rated AAA/ Aaa by the rating agencies. Bankers followed the formulae given by the rating agencies, which, coincidentally, were paid to help structure the products.[[[PAGE]]]
I think the creation of this AAA/Aaa junk was similar to creating ‘fresh orange juice’ in the UK – add the minimum required orange concentrate + maximum amount of water and sugar that is allowed. Add the maximum colours and flavours and hey presto a health drink – ‘fresh orange juice’. Both products can be created at the edge of the rules. This of course, was a recipe for disaster.
The danger increased when some banks themselves believed that these structured instruments actually were AAA/Aaa. The fact that these AAA/Aaa investments were yielding Libor +25 while AAA sovereign risk/ supranational was sub Libor should have warned many buyers. (I have, painfully, learnt that there are few free lunches to be had in the markets. Not for long anyway. There are too many canny investors out there to be tricked indefinitely.) When the market woke up, the instruments of course became illiquid. The auctions of securities that could be auctioned every three months at close to par, failed. There were simply no buyers. The correlations to other highly rated bonds became negative, plunging the very assumptions used by banks for risk management into the abyss.
So before discussing some of the issues with VaR as a risk measure, let us review what VaR actually is. Consider a portfolio or financial instruments and derivatives. The value at time zero is V. Now measure the value after three months. Let’s say this is W. (W-V) is the profit and if negative the loss. Now build a distribution of the possible losses. The probability is 95% that the actual loss of a portfolio over a three-month period is less than the value at risk v(0.95, 3 months) (at the 95% confidence level) for a three- month move.[[[PAGE]]]
VaR can show that by adding some negatively correlated instruments to a portfolio, the VaR of the portfolio and hence the risk of the portfolio reduces.
The VaR statistic does has some clear advantages:
- A clear one-figure summary which is easily understood by non-risk experts – often used in board meetings of banks and corporates alike. The IRFS allows companies to use this measure to disclose risk to the shareholders
- You can see the effect of single instruments on your portfolio
- Combines effects across many asset classes
- Can be extended to ‘cash flow at risk’ concepts
There are, however, some drawbacks:
1. One of the assumptions, that market returns follow a normal distribution, is nonsense. The mathematician Benoit Mandelbrot noticed 40 years ago that if stocks really followed a bell curve then a swing of more than 7% in the Dow Jones would happen every 300,000 years. In fact there were 48 such days during the 20th century! It is clear that returns follow a fat-tailed distribution and so the assumption of normality is not correct.
2. Correlations and volatilities are not stationary. Figure 1 shows the euro/Swedish kroner cross for the last 10 years. The time series clearly displays at least three different behaviours. Between 2001 and 2008 the SEK was highly correlated to the EUR. The rapid devaluation in 2009 of the SEK versus the EUR shows a sharp change in that behaviour.
3. Even simple models require enormous numbers of correlations and volatilities to be estimated. Consider an interest rate and currency portfolio with CHF, EUR, USD, JPY, CAD, CNY and SEK. The interest rate in each currency is measured by two variables – short-and long-term:
- There are 6 + (7 x 2) = 20 variables
- There are 20 volatilities and 190 correlations= 210 covariances to estimate!
Assuming you have five years of data or less, your estimates will be rather rough.
Understatement of risk
So why did none of the banks quote their VaR in 2008 to be in excess of 1bn euro? The answer is clear. The combination of treating the structured products as AAA/Aaa bonds, which were assumed to behave like other AAA/Aaa instruments, in models which do not allow for correlations to switch from positive to negative, led to a massive understatement of risk.
The rest of the story is now of course history. Many banks suddenly realised that these AAA/Aaa products were not as liquid as they thought. As a result they started to rein in commercial lending, which was a shock to industry and potential homeowners alike, who were both now addicted to cheap and easy credit.
To sum up, an academic once told me that VaR is a useful measure of risk except during risky periods. Since I want to know my downside when everything goes wrong, VaR is not a particularly useful measure.
I read recently in the Financial Times that the VaR of a large bank had exceeded 200m USD. It is a shame that standard setters and banks themselves still do not recognise that VaR is not a useful measure!
This article represents the personal opinions of the author.