Date posted: November 11, 2007

Risk Management Does Not Eliminate Risk

In the wake of the subprime crisis a lot is being said and written by people who understand nothing of finance and by people who should know better. I thought I’d add some of my own points of view.

Many commentators write that there is something inherently wrong with the way banks and other financial institutions manage risk. Echoing this politicians and other policy makers call for “better risk management”. The misconception is that risk management eliminates risk, which it doesn't. The hidden assumption is that banks are not allowed to incur (trading) losses and go bankrupt.

Much of the criticism focuses on the use of Value at Risk models to measure risk. I don’t know how or where this started, but I’m now reading it in various newspapers. Value at Risk is only one risk measure used by banks. The Basel II Accord also requires banks to perform stress tests using different crisis scenarios. A typical stress test might measure the value of a portfolio of financial assets under the same market conditions as Black Monday in 1987, the Asian crisis of 1997 or the Russia crisis of 1998.

Value at Risk measures the maximum possible loss within a known confidence interval over a given holding period. If we set the holding period at one day and the confidence interval at 99%, what this definition says, is that a loss greater than the Value at Risk happens only once every 100 days. What this does NOT take into account is the value of the maximum loss within the period of, in this case, one day. Thus intraday losses may exceed the Value at Risk calculated the night before. It doesn’t say anything about the amount that can be lost with a probability of 1% either nor does it take into account the fact that losses can accumulate over a period of days. Actual losses can therefore exceed the reported Value at Risk figures.

There are various ways of calculating Value at Risk, all of which rely on assumptions about the behavior of financial assets. When calculating Value at Risk financial institutions therefore face a trade-off between accuracy, consistency and ease of implementation.

A popular calculation method uses a database of historical price movements to calculate the possible price trajectories of the various portfolio components, which are then translated into a return distribution relative to today’s portfolio value. A history of 100 trading days will give 100 possible scenarios. The Value at Risk is then determined by tabulating the profit and loss distribution and taking the desired percentile.

The reason that this method has become popular is that, once you’ve got a history of price movements, it is relatively straightforward to implement.

However, there are various problems with historical simulation, as this method is called. Companies merge and new companies are listed in which case there are no historical data and calculating the right volatility to value an option can be a challenge.

A major problem with historical simulation is that it underestimates the Value at Risk if the historical window happens to be a period of low volatility, as was the case over the past three years. Any sound risk manager should be aware of this and so should any analyst reading the VaR figures financial institutions report every quarter. In the end Value at Risk is just a number. As always, it’s not the figures themselves, but what you do with them, that matters.

I must say that reading the following statement by UBS in its 2007 Third Quarter report: “In third quarter we suffered our first backtesting exceptions - 16 in total - since 1998” does make me wonder about risk management at UBS. All of these 16 “backtesting exceptions”, that is, actual losses that exceeded Value at Risk, occurred in one month, whereas according to the theory one would expect such events to happen at most three times per year.

Instead of using historical data Value at Risk can also be calculated using Monte Carlo simulation, that is, randomly generated numbers. Some years ago, when banks were preparing for the Basel 2 accord, implementing a Monte Carlo simulation based Value at Risk system was intractable for anything but the simplest portfolios. With today’s technology and computing power this may have changed.

It has been suggested that VaR induces banks and hedge funds to simultaenously dispose of the same assets so as to reduce VaR, thus increasing volatility and giving rise to a form of systemic risk. In a paper from 2004 “Bank Trading Risk and Systemic Risk” Philippe Jorion from the University of California at Irvine did not find any evidence for this hypothesis. It would be interesting to study the day-to-day data from various financial institutions over the past few months to see whether this still holds.

I don’t think that VaR itself is the problem. I believe that VaR and Basel 2 have contributed to making the financial world a little more transparent.

Banks will always try to make do with a minimum amount of capital, there’s nothing new about that. It is true that under Basel 2 banks can calculate their Tier 1 capital ratio using risk weighted assets as the denominator. But again, good analysts don't (shouldn't) look at any one ratio, but at a range of metrics and at how they are calculated.

The REAL problem right now is that over the past few years some (?), several (?), many (?) banks have created off-balance investment vehicles with an embedded liquidity put, a contractual agreement to buy back the commercial paper with which the investments were financed. When banks are forced to take these portfolios back onto their balance sheets they have to “re-balance” their assets and liabilities so as to conform to the capital adequacy requirements. The other problem is that nobody knows which banks are exposed to the subprime crisis and to what extent.