FinUties 101 · Risk · Advanced
Tail risk: when the bell curve fails
Tail risk is the chance of a very large loss that normal models treat as almost impossible. Fat tails, value at risk and expected shortfall show what the usual measures miss and how risk control adapts.
In this lesson
- Explain what a fat tail is and why a normal distribution understates extreme losses
- Calculate value at risk and expected shortfall from a list of returns
- Name three ways risk control limits the damage of a tail event
Read first: Correlation and its limits, Position sizing basics
Most risk measures describe ordinary days well. Tail risk is about the rest: rare events with very large losses, in the "tails" at the ends of the distribution of returns. They matter because one such event can outweigh years of ordinary results.
The bell curve and its tails
A normal distribution is the bell curve. Under it, most results sit near the average and extreme ones are very rare. A move of 5 standard deviations is expected roughly once in several million observations. At about 250 trading days a year, that is once in many thousands of years.
Market history disagrees. Moves of that size appear within a single career; see Black Monday, 1987. On 19 October 1987 the S&P 500 fell about 20% in a day, when a typical day moved around 1%. Benoit Mandelbrot argued that price changes follow distributions with fat tails: extreme moves are far more common than the bell curve allows. The bell curve is a fair description of the middle and a poor one of the ends.
Why tails occur
- Volatility clusters. Stress days come together, and the clustering produces many more extremes than independent days would.
- Correlations jump. In a crisis, diversification shrinks as assets fall together.
- Liquidity vanishes. Sellers outnumber buyers, so prices jump past the levels where orders sit.
- Leverage forces selling. Borrowers sell because they must, which pushes prices lower and forces more sales.
The 2008 and 2020 case files show each of these. Nassim Taleb popularised the label "black swan" for rare events with large effects that models did not foresee, though the point he stresses is practical: judge a position by what it can lose, not by how unlikely the loss seems.
Measuring a tail: VaR and expected shortfall
Take 100 daily returns and sort the five worst: -2.0%, -2.4%, -2.6%, -3.1%, -5.5%.
Value at risk (VaR) at 95% is the loss that was exceeded on 5% of days. Here it is the fifth-worst day, a loss of 2.0%. The statement is "on 95 days in 100 the loss was no worse than 2.0%".
Expected shortfall is the average loss on those worst days: (2.0 + 2.4 + 2.6 + 3.1 + 5.5) / 5 = 3.12%. It describes the part VaR ignores.
Notice what VaR hides: the day of -5.5% and the day of -2.0% look the same to it. Both measures also rest on the past, and the past sample may contain no true crisis. They are useful gauges, and they are not limits on what can happen.
Living with tail risk
Risk control cannot predict tails. It limits the damage:
- Size for survival. Keep each position, and the sum of related positions, small enough that a gap of several times the usual move is survivable. See position sizing basics.
- Avoid forced selling. Leverage turns a tail event into a permanent loss, because the lender decides when you sell. Unborrowed positions can wait.
- Stress test, don't average. Ask what happens in a repeat of 1987, 2008 or 2020 with today's holdings. A fixed scenario is more honest than a statistical comfort figure.
- Do not rely on one hedge. Hedges, such as bonds against shares, work for some causes of a fall and not others.
Options that pay in a crash exist, but they cost a premium every period, which is the price of the insurance.
Common mistakes
- Treating VaR as a maximum loss.
- Calibrating risk to the last calm years.
- Treating "very unlikely" as "will not happen in my holding period".
- Relying on stops to cap a tail loss.
- Selling insurance-like positions that earn small, steady amounts and carry a rare, large loss.
In practice
Read any risk figure with the question "what is outside it?". For a daily 95% VaR, the answer is the worst 1 day in 20. History is informative about size: Reinhart and Rogoff document how repeatedly large crises have occurred across countries and centuries.
Check yourself
What does a fat-tailed distribution mean for the chance of a 5-standard-deviation day?
Under a normal distribution such a day is expected about once in millions of days. With fat tails it happens far more often, which is why such days appear in market history within a lifetime.
Value at risk at 95% is 2% for a day. What does it tell you, and what does it leave out?
On 95 days out of 100 the loss is expected to be no larger than 2%. It says nothing about how bad the other 5 days are.
Why does a hard stop not remove tail risk?
In a tail event prices gap and liquidity vanishes, so the exit happens far from the stop. Sizing and exposure set the loss; the stop only states the intent.
Key terms
Sources
- The (Mis)behavior of Markets: A Fractal View of Financial Turbulence. Basic Books (Benoit Mandelbrot, Richard L. Hudson), 2004
- The Black Swan: The Impact of the Highly Improbable. Random House (Nassim Nicholas Taleb), 2007
- This Time Is Different: Eight Centuries of Financial Folly. Princeton University Press (Carmen M. Reinhart, Kenneth S. Rogoff), 2009