Value At Risk (VAR) is a risk measure used to determine the probability of a certain percentage loss (or gain!) in the value of a security based on historical data. Although widely used by risk managers to determine worst case scenarios and handle the risk a firm can take (in a bid to earn more!), it is deeply flawed.
Not because it uses historical data to calculate future risk, but because it assumes that the historical returns (daily. monthly etc.) fall under a normal distribution or bell curve – the one used to fix employee appraisals (another flawed application!).
It is widely documented that VAR models is one of the reasons for why, and the manner in which the 2008 crash occurred. Watch the movie margin call for a non-technical account or get hold of the book, Alchemists of Loss: How Modern Finance and Government Intervention crashed the financial system!
The title of this post (written by a student of the subject) is a reference to US hedge fund manager, David Einhorn’s quote that VAR is “an airbag that works all the time, except when you have a car accident.”
Technically, it does not work at any time.
I do not wish to get into the details of the VAR model only to later show that it is flawed. Instead, using a simple measure called the SKEW, will demonstrate why using the VAR is incorrect. Alternatives to the VAR are not intuitive but that is no excuse for using a faulty model.
This is a how a textbook handles this issue:
“The most questionable assumption, however, is that of normality because evidence shows that most securities prices are not normally distributed. Despite this, the assumption that continuously compounded returns are normally distributed is, in fact, a standard assumption in finance”.
Terrific! The ‘standards’were set by the alchemists of loss!
Let us start with how the normal distribution looks.
The x-axis will be monthly or daily stock returns and the y-axis the corresponding probability of occurrence. So VAR can be used to calculate the probability associated with a range of returns.
(1) The peak corresponds to the average return. Notice that the data is distributed symmetrically on either side of the peak. That is the SKEW of a normal distribution is zero
(2) Also, notice how smoothly and rapidly the probability falls off from the peak. Extreme events are extremely rare! Meaning, a Harshad Mehta scam which saw the Sensex shoot up by 270% or the 2008 crash would have such a low probability of occurring that you can safely discount them. Yet as B Mandelbrot writes in his book – the misbehaviour of markets – such events are common enough.
These two are the hallmarks of a normal distribution. Stock price returns look nothing like this!
Mandelbrot showed the even over a period of several decades, price movements do not follow a bell curve. That his contributions were largely ignored and he was sidelined by the proponents of the modern financial theory* is another story!
(*) also depends on the normal distribution which makes mutual fund star ratings questionable!
Here is the skew observed in the daily and monthly returns of a few stocks (15 year history). For an ideal normal distribution, the value should be zero. Even small departures from zero can result in bad predictions.
It would be foolhardy to bet that these distributions can be approximated as normal and use the VAR model.
Have a look at the monthly returns distribution of Manali Petro Chem.
Normal distributions do not behave like that!
As Ramesh Mangal once told me, why choose a model that is exactly wrong? In order to assess volatility, we need to use models that do not depend on the normal distribution assumption.
What do you think? Would you buy a car with a faulty airbag because it looks elegant?