Another nugget from Nassim Taleb’s Incerto series of books. In his book The Black Swan, Taleb discusses what he called the ludic fallacy (ludic from the Latin ludus – games). In essence the ludic fallacy is to believe that real-life events are statistically as simple as games of chance in a casino, and can be predicted as accurately.
In a casino (an honest one, that is) the rules of the games are fixed and clearly stated, and so probabilities can be computed accurately. (Honest) dice always have six sides and are not loaded. (Honest) coins have equal odds of heads or trails. (Honest) roulette wheels have no magnets, so the chances of the ball entering any given bucket are equal, and so on.
In real life, metaphorically, the dice may be loaded, the coin may well be biased, the rules may change unpredictably, and the roulette wheel may well have magnets. Classical statistics is based on the properties of the Gaussian distribution (the Bell curve) with some strong assumptions (observations are independent, and path independent, etc). In games we know the characteristics of the generator (eg – the dice, or the coin, or the roulette wheel). In real-life events the generator is usually very complex, often largely invisible (often including the actions, agendas, decisions, and complex motivations of many people), and often (or perhaps usually) includes random elements. Most games of chance are in the domain of “Mediocristan” where there are few outliers more than 3 standard deviations from the mean; many real-life events are in the domain of “Extremistan” where the distribution may be “fat tailed”, and where there are outliers with very significant impacts, represented by decidedly non-Gaussian distributions.
The ludic fallacy is to assume one can apply the simplified games models to massively complex real-world events. Lots of professional risk managers apparently try to do this, providing ample opportunities for wildly expensive black swans (like the 2008 financial crisis).
In the Incerto series Taleb gives us the non-mathematical logical argument for this. In his massive 455-page technical supplement to the Incerto series, Statistical Consequences of Fat Tails: Real World Preasymtotics, Epistemology, and Applications (which you can buy from Amazon or download as a pdf free from his website here) one can find the rigorous mathematical exposition of his claims.
Posts
