Examining the contrast between Nassim Taleb's insights into prediction and Nate Silver's based on their respective books Antifragile (2012) and The Signal and the Noise (2012)
2. Can we predict the future?
• Silver suggests we can’t in a large number of
domains after testing a range of past predictions.
• The accuracy of weather forecasting has
significantly improved over the last 30 years.
• Earthquake predictors have generally failed.
• Disease epidemic predictors have generally failed.
• Most stock market predictors generally fail in the
long run.
3. The Signal and the Noise
• US elections
• Baseball performance
• Poker
• Chess
4. The Signal and the Noise
• Weather forecasting is a (limited) success story.
• The science of earthquake forecasting seems barely
to have evolved since the ninth century.
• Economic forecasts have often failed to “predict”
recessions even once they were under way.
5. Prediction, forecasting and modelling
• A prediction is a definitive and specific statement about when
and where an event (e.g. earthquake) will strike (Silver 2012)
• A forecast is a probabilistic statement usually over longer time
scale e.g. 60% chance of earthquake over next 30 years.
(Silver 2012)
• A model is a simplification of a real world system which
attempts to represent relationships between variables.
6. Criticisms of Silver
• Perhaps understandably he does not highlight the
uncertainty around his perfect predictions for
2012 US election. He was lucky!
• He talks about the importance of probabilistic
statements yet describes prediction “failures” in
forecasting election results, financial crisis.
• He has too much confidence in the ability of
forecasting models. They cannot incorporate
typically unique, non repeatable extreme events
which are best studied separately.
7. The Black Swan
• The world is hideously more complex than a
game of poker, baseball or political elections
(ludic fallacy)
• A black swan event is an unexpected outlier that
carries an extreme impact.
10. Did the turkey make a statistical
miscalculation?
• According to Bayes’ theorem, every episode of the turkey
being fed by a generous owner made it more confident of its
safety.
• Think stock market crashes (Lehman Brothers, Enron etc)
where incorporating past data makes you more confident of
the validity of a positive trend and more likely to be harmed
by the eventual crash.
11. Anti-Fragility
• Build in redundancies (Fukushima, flood defences)
• Do not suppress randomness with tight controls.
• Make sure participants have “skin in the game”
• Expose yourself to positive black swans
• Take risks with a known limited downside and unbounded
upside.
12. Extremistan and Mediocristan
• In Mediocristan, large samples contain no single
instance that will significantly change the
aggregate or total. The largest observation is
insignificant to the sum (weights, heights, calorie
intake, mortality rates)
• In Extremistan, inequalities are such that one
single observation can disproportionately impact
the aggregate or total (wealth, income, book
sales, deaths in war, inflation rates, economic
data)
14. Why can’t I predict in Extremistan?
• Imperfect information
• Feedback loops
• Chain reactions (positive/negative spirals)
• Past data is taken from the world in a very
different state to the present (or future) and
extrapolated
• Models are simplifications of the real world.
• Models typically rely on tractable statistical
distributions (e.g. normal distribution) with thin
tails.
15. What can we predict?
• The relative fragility, robustness and anti-
fragility between two systems/objects.
• Mediocristan (sports, poker, elections)
• Short term trends in Extremistan but not the
extreme event (MUCH more important).
• Do not fall for the ludic fallacy of perceiving
the world as a simple, well organised game
subject to consistent rules.
16. “First, do no harm”
• Predicting the weather, making small bets on
sports or political elections is not going to
harm or threaten society (minimal downside)
• Predicting budget deficits, interest rates,
inflation rates, stock prices many years into
the future can threaten the fabric of society if
they are taken seriously and no contingencies
are prepared (potentially large downside)
17. Conclusion
• We generally cannot predict the timing of black
swan events without insider information.
• We cannot predict the severity of black swan
events.
• We can assess the comparative fragility,
robustness or anti-fragility of systems rather than
trying to predict the exact timing, severity of the
rare event.
• Make yourself robust or anti-fragile (if possible)
to uncertain future events.