Beware the “atypicality” of remote events
Taleb warns that historical scenarios may miss important tail risks. He also distinguishes predicting whether an event occurs from protecting against an open-ended amount of loss.
THE IDEA
4. Beware the “atypicality” of remote events.
There are suckers’ methods called “scenario analysis” and “stress testing”—usually based on the past (or on some “make sense” theory). Yet (I showed earlier how) past shortfalls do not predict subsequent shortfalls, so we do not know what exactly to stress-test for. Likewise, “prediction markets” do not function here, since bets do not protect an open-ended exposure. They might work for a binary election, but not in the Fourth Quadrant.
When to use it
A financial institution evaluating tail risk for investment portfolios must account for the atypicality of remote events beyond historical scenarios.
Government agencies preparing for low-probability high-impact events like pandemics need to avoid over-reliance on stress tests derived from past outbreaks.
Companies using scenario planning for strategic decisions should recognize that plausible scenarios may not cover all remote yet consequential possibilities.
Insurance companies assessing catastrophe risks must consider that historical loss data may not adequately represent the true tail distribution.
FROM THEORY TO EVERYDAY LIFE
In practice
Bank stress-test failure
Suppose a bank tests only scenarios drawn from its recent experience and misses a crisis like the 2008 crash. The example illustrates a limitation of the chosen scenarios, rather than proof that all stress testing is useless.
Prediction market for novel risks
A market can offer a binary contract on whether a pandemic occurs. Such a contract does not automatically cover an institution’s potentially much larger losses from that pandemic. Forecasting occurrence and hedging the size of the exposure are different tasks.
A MOMENT TO REFLECT
Put it to work.
How does relying on past data for risk assessment lead to unexpected failures, especially in complex systems?
What distinguishes a binary prediction from an open-ended exposure, and why is the distinction crucial for risk management?
Can you identify a real-world event that was considered 'atypical' and not captured by standard models, and what were the consequences?
CONTINUE THE CONVERSATION
Source texts
- The Black Swan The Impact of the Highly Improbable With a new section On Robustness and Fragility (Nassim Nicholas Taleb)