Value At Risk (VaR)

In my wage slave financial programmer job, I spend a lot of time making VaR calculations (VaR means "Value At Risk").

VaR is an easy-to-calculate risk measure that provides an illusion of risk management, while being completely meaningless. Almost everyone in the financial industry uses VaR.

For liquid securities, like stocks, VaR is defined and calculated via a historic back-test. For example, VaR usually is the 99th percentile loss, over a year historic period, assuming it takes 1 day to liquidate the position. In that case, you would look at 1 year of historic price data, ~250 trading days. The 99th percentile would be the average of the 3rd and 4th largest loss. This makes VaR very easy to calculate.

Almost everyone in the financial industry uses VaR. Do you see the fallacy?

SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE
SPOILER SPACE

VaR is the 99th percentile loss. However, that remaining 1% can be pretty bad!

Suppose your trading system makes $1M 99.1% of the time, and loses $1T 0.9% of the time. The VaR risk is zero. VaR explicitly encourages large banks to abuse their "too big to fail" status, by ignoring the small chance of a huge loss.

Due to the "fat tails" problem, that remaining 1% will occur more often than your model predicts. For example, a 10 year historic back-test of housing prices would not have predicted the crash. In that 10 year period, housing prices always went up.

Further, "a big bank failed today" automatically is going to be a day in the 1% tail. The VaR distribution is based on a "typical" day. If you actually need it, you're already in the 1% tail.

VaR is easy to calculate. It provides the illusion of risk management, while actually doing nothing. Almost everyone in the financial industry uses VaR. Therefore, the risk manager keeps his job if he also uses VaR. VaR underestimates "tail risk". VaR explicitly encourages banks to abuse their "too big to fail" status. A small chance of a huge loss won't be included in a VaR calculation. The "fat tails problem" means that extreme big losses occur more often than your model predicts. A historic back-test probably won't include the extreme event that happens in the next crash.