Gillian Tett: The Perverse Effects of Value-at-Risk Models

In an interesting bit of synchronicity, the role of value-at-risk models has come into focus in the last couple of days.

By way of background, VAR is a widely used risk management technique. It defines the level of risk by assessing the most one might lose in a set time period (typically one day) with a given confidence level based on the historical performance of the assets in the portfolio.

There has been a bit of discussion of VAR since Morgan Stanley had to explain its August trading losses (see Felix Salmon and Alea).

One thing that these discussions often fail to acknowledge is the shortcomings of VAR, which are well understood in the industry. The biggie is that it assumes a normal distribution of outcomes, when financial assets are subject to both fat tails (kurtosis), which is the market equivalent of “things far away are bigger than they appear to be” and skewness (the distribution isn’t symmetrical around the mean).

The Financial Times’ Gillian Tett today takes another tack on VAR. First, she notes that it can suddenly give signals to sell down very large amounts of risky assets, at of course precisely the worst time to be selling. Second, even though everyone knows that VAR isn’t a great tool, and consequently risk managers tend to give more weight to other analytical techniques, both investors and regulators have become accustomed to VAR. That creates pressure to manage to VAR even when it might be a bad idea.

An aside: Tett mentions managing to a 99% confidence level. That was certainly the level commonly used in the 1990s, and may still be employed in some contexts. It appears from the discussions around Morgan Stanley that 95% is now a typical confidence level for many Wall Street firms.

From the Financial Times:

even months ago, the Bank of England issued a strikingly prescient warning about “value at risk” (VAR) models.

While these models have become endemic in the financial world in recent years, they have a nasty habit of being self-reinforcing, or so the Bank observes.

When volatility is low in the markets – as it has been during most of this decade, when VAR models have flourished – these tools typically offer a very flattering picture of risk-taking.

That prompts banks to take more risk, which reduces market volatility further as more cash chases assets.

However, if markets ever turned volatile, this dynamic could quickly unravel, the Bank warned back in April.

What VAR essentially does is measure the money a bank is likely to lose, with 99 per cent probability, in a specified time, based on historical data.

If prices are swinging around, implied potential losses tend to rise.

In April, the Bank estimated that a typical bank’s VAR might theoretically double, with the same assets, if volatility increased.

Seven months on, that once-theoretical piece of analysis is coming back to bite the banks. In recent weeks, I am told that many investment banks have been furtively trying to reduce their published VAR levels in response to this summer’s tumultuous events.

In the current environment, no bank chief executive who hopes to hang on to that job can afford to give regulators or shareholders the impression that they are being cavalier about risk. And since VAR is often used to define what level of margins – or financial buffers – are set against trades, some banks are doubly keen to cut VAR, to reduce pressure on their own balance sheets.

But as the banks embark on this task, some are finding themselves caught in an unpleasant trap. The easiest way to reduce a risk exposure is to sell risky assets, such as risky loans. In recent weeks, many banks have been trying to do precisely that.

But these sales have been occurring on such a large scale that they have pushed up market volatility. Thus, measured VAR has risen, exactly as the Bank warned all those months ago.

One big investment bank has recently analysed the impact of its own recent asset sales. These suggest that while these sales should have cut VAR by half in recent weeks on constant volatility levels, in practice this gain was more than wiped out by ensuring market price swings.

By scurrying to reduce risk, in other words, the banks may end up simply running to stand still.

This problem will undoubtedly leave many observers to conclude that there are flaws in the VAR concept. Behind the scenes, that is precisely what many risk experts now privately say.

Earlier this week I attended a conference of risk managers in New York, where one expert cheerfully declared that “no one in the industry really has much faith in VAR any more”.

Or as another official at a big bank noted, “VAR has a role as a tool, but it is just one tool. There is no point in using it unless you also engage your brain.”

But the problem for the banks is that VAR has become so central to the financial world this decade, and so closely watched by shareholders and regulators, that it is difficult to ignore.

It is the financial equivalent of the modern examination system in the educational world: though all the practitioners can see endless flaws with the existing rules, nobody is quite willing to flaunt them in public.

It is a fair bet that during the rest of the year the banks will keep trying to lower VAR, while also knowing that their attempts to do this, by making asset sales, could be making their problems worse.

It is a painful vice – or a timely reminder of the golden rule that you rarely get a seemingly virtuous circle in finance that does not turn into a equally vicious cycle later on.

Print Friendly, PDF & Email


  1. a

    Why do we have Var? It’s because the regulators are not very knowledgeable about the businesses they are regulating, and so they prefer the magic bullet of a universal method of risk evaluation to admitting they are completely clueless. But good risk assessment is an activity with a lot of added value, and there is a big difference between those banks with a strong culture of risk evaluation and those which think Var suffices. Such risk assessment (stress tests – what kinds and with what limits, when to allow overshoots and when not to) is very proprietary and only passes from one bank to the next via slow diffusion (new hires and regulators making hints).

  2. Anonymous

    The Gaussian curve fit them all?
    In case one would be interested to dig further none of these markets (equities,CMO,CLO,CDO)were and are meeting with the criteria of normal statitical distribution (Gauss or Poisson)it is hard to understand the VAR in these cases.
    Most of the models I have been through are skewed on the upside,do not read inflexion point untill too late.

Comments are closed.