Advanced3 min read

Tail Risk

The rare, large losses that dominate long-run outcomes. Standard risk measures are built on a distribution that understates exactly these events.

MadStockAlerts Research · Updated August 28, 2026

What to take away

  • Return distributions have fatter tails than a normal distribution implies.
  • Standard deviation understates the frequency of extreme moves.
  • Value at risk states a threshold and says nothing about what lies beyond it.
  • Correlations converge in the tail, so diversification weakens when it is needed.
  • The practical defence is sizing and structure rather than measurement.

MAD Academy Training Video · 0:46

The Events the Model Says Cannot Happen

Financial returns have fatter tails than a normal distribution, which is why 'once in ten thousand years' events keep arriving every decade.

This lesson is part of a Stock Alerts + Tools plan.

See the library

Why the tails are fat

Under a normal distribution, moves of several standard deviations should be vanishingly rare. In market data they occur regularly: several-sigma daily moves have appeared many times in histories where the normal distribution predicts approximately none.

How often extreme days actually occur
How often extreme days actually occur0200400600Predicted never; observed repeatedly2 sigma3 sigma4 sigma5 sigmaOccurrences over a long history

Scroll the chart sideways to see all of it.

  • Predicted by a normal distribution
  • Observed
The gap between the two bars at the right is the whole of the tail problem. It is not a small discrepancy at the edge; it is the difference between never and repeatedly.

Volatility clustering is part of the explanation. Large moves follow large moves, which violates the independence the normal model assumes and concentrates extreme days into periods.

What value at risk does not say

A value at risk figure states a loss that will not be exceeded on a stated proportion of days. It is silent about the size of the loss on the days when it is exceeded, which is the question that matters.

MeasureThe question it answers
Value at riskWhat is the threshold that is exceeded 1 percent of the time
Expected shortfallGiven that it is exceeded, how large is the loss on average
Maximum drawdownWhat actually happened, historically
Stress testingWhat would happen under a specified scenario

The second row exists precisely because the first is inadequate. Two portfolios can share a value at risk figure and have completely different losses beyond it, and it is the beyond that ends accounts.

Correlations in the tail

The diversification a portfolio relies on is measured in ordinary conditions and consumed in extreme ones. Assets that behaved independently for years move together in a crisis, which is when the independence was being counted on.

This is why a risk model calibrated on a calm period underestimates a crisis twice over: the individual moves are larger than the model predicts, and they arrive together rather than offsetting.

What actually helps

  • Sizing that survives a move several times larger than the model's worst case.
  • Limiting leverage, since leverage converts a survivable loss into a forced one.
  • Avoiding structures with unbounded loss, which is where the tail is unbounded rather than merely fat.
  • Holding some assets whose payoff is largest in exactly those conditions, accepting that they cost money the rest of the time.
  • Accepting that the event will be one nobody modelled, which is what makes it a tail event.

The last item is the honest conclusion. Tail risk is managed structurally rather than measured precisely, because the measurement depends on a distribution that the events themselves demonstrate is wrong.

Stress testing rather than modelling

Because the distribution cannot be relied upon, the practical alternative is to specify scenarios directly and compute what they would do, without any assumption about how likely they are.

  • A move of a stated size across everything held at once, rather than position by position.
  • Correlations set to one, which is the crisis case rather than the estimated one.
  • A gap through every exit, so the realised loss exceeds the planned one.
  • A funding stress: margin requirements raised while positions are losing.

The output is not a probability. It is an answer to whether the account survives a specified event, which is the question that matters and is the one a value at risk figure does not address.

The fourth item is the one most often omitted and the one that has ended the most accounts. Requirements rise as volatility rises, so the capital needed grows exactly as the capital available shrinks.

Educational content only. MadStockAlerts provides market commentary, research, and educational content. It is not personalized investment advice, and nothing here is a recommendation to buy or sell any security. Trading and investing involve substantial risk, including loss of capital. See the Risk Disclosure and Customer Agreement.