Trading Concept
Tail Risk
The risk of extreme events that fall far outside the range of normal experience — events that backtests rarely capture because they haven't happened in the sample.
Benoit Mandelbrot demonstrated in 1963 that financial returns have "fat tails" — extreme moves occur far more frequently than Gaussian (bell curve) models predict. A move that a normal distribution says should happen once every 10,000 years happens every few years in financial markets. The 1987 crash, the 1998 LTCM collapse, the 2008 financial crisis, the March 2020 COVID crash — each was a "tail event" that models said was virtually impossible.
A strategy backtested on 5 years of data has never seen the event that destroys it — because that event did not happen in those 5 years. The worst day in the backtest is not the worst day that can happen. It is the worst day that did happen in the sample. The actual worst case may be significantly worse.
No gate, no tolerance band, no position sizing rule fully protects against tail risk. A configured maximum position size limits exposure. A HALT condition on execution mismatch stops the bot when something unexpected occurs. But a platform outage during a flash crash, a stablecoin depeg, or a regulatory ban on the traded asset — these are events the strategy cannot model because they are outside the distribution the strategy was calibrated on. The user is responsible for understanding that structured discipline reduces risk. It does not eliminate it.