Nassim Taleb - The #1 Rule to Not Be a Sucker (The Ergodicity Principle)

The central error in misusing averages is mistaking a vertical (ensemble) average for a time average: 100 gamblers each gambling one day produces a valid expected return via the law of large numbers, but a single person gambling for 100 days yields a completely different process because if they go bankrupt on any day the sequence ends — so the two are governed by different mathematics.

“the central problem is that you mistake a vertical average for time average”

The Kelly criterion is a mathematical formulation of the simple concept 'don't risk everything' — bet carefully each time so you never lose the whole kitty — and it helps avoid ruin, making risk-taking calculated across large populations not always rational for an individual.

“The Kelly criterion is a very popularized mathematical formulation of a simple concept. And the simple concept is don't risk everything... Just be very careful how much you bet each time so you don't lose the whole kitty”

Ergodicity is illustrated by Russian roulette: six people each playing once with a billion-dollar prize leaves five billionaires and one dead, but one person playing six times with the same gun ends up worth zero — demonstrating that what is true for many people on average is not the same as one person repeating the gamble.

“The easiest way to see that is by playing Russian roulette. Six people who play Russian roulette once each... versus one person who plays Russian roulette with the same one gun six times is never going to end up a billionaire is going to end up worth zero”

When taking an arithmetic average over an ensemble, a single bankruptcy (a zero) barely affects the average because it is weighted by one-over-n, whereas in a time sequence a single ruin event terminates the entire path.

“number 28 goes bankrupt number 29 is not affected you're taking the arithmetic average So you add a zero, zero doesn't impact average much by one overn. Whereas if I take a sequence... if I on day 28... dies then it doesn't get it”

For an arithmetic average to match continuously compounded returns, you must bet in log size; otherwise a strategy that does not compound returns will lead to ruin even with fair odds because you cannot survive the path.

“for arithmetic average to match the the the continuous compounded thing you must uh bet in log size”

A bet with favorable odds (e.g., 55/45) is not necessarily rational to keep taking: if you repeatedly take such a bet over a lifetime without following a survival strategy like the Kelly criterion, you will eventually go bankrupt even with fair odds, because you are subject to a time average and cannot survive a ruin event.

“if you keep taking that vet all your life unless you follow a specific policy of Kelly criterion for example which has longs incidentally you're going to go bankrupt even if you have fair odds because you cannot survive”

The precautionary principle rests on asymmetry: when there is uncertainty about something where the downside is ruinous (e.g., a pilot of uncertain skill, possibly-poisoned water, climate, or GMOs), the rational response is to avoid the action rather than demand proof of harm, because the burden of the asymmetry means 'no evidence of harm' is not 'evidence of no harm.'

“they tell you they have uncertainty about the skills of the pilot... I do not fly... Life is too important for me”

Concentrating supply chains on a single supplier produces 'pseudo-efficiency' — a better bottom line achieved by being short an option whose cost does not appear in the numbers until a tail event (such as a supplier in Wuhan during COVID) materializes; robustness instead requires redundancy and inventory, which is not a true cost but can be antifragile because in a crisis squeeze the commodity price spikes and stockpiled inventory can be sold.

“a lot of firms concentrated everything on one supplier instead of being diversified... Better bottom line, but what I call pseudo efficiency because they're short that option”

Thaler's concept of 'mental accounting' as a fallacy (treating money won at a casino as different from one's initial endowment) is actually correct behavior for survival: treating house money as distinct and being more aggressive with it than with one's endowment is the only way to survive, so what is labeled a psychological 'disease' is rational under ergodic dynamics.

“he found a disease called mental accounting that say the following if you go to a casino and increase your betting with money won from the casino it's called houses money you are irrational because it's the same dollars”

When you lack skin in the game (as a CEO or fund manager), your incentive is to print good numbers because you collect on the profits but do not pay for the downside — the 'generalized Bob Rubin trade,' exemplified by Robert Rubin collecting ~$100 million from Citi over ~10 years while the bank became near-insolvent in 2008 and required taxpayer bailout, after which he only had to claim it was an unforeseeable black swan.

“If you don't have skin in the game, you're CEO of a company or you're a fund manager... what is your incentive is to print good numbers because you don't pay for the downside”

Measuring inequality statically is wrong; the correct ergodic measure tracks individuals over their lifetimes — 12-15% of Americans will spend at least a year in the top 1% and roughly 60% will spend at least a year in the top 10%, showing the people 'at the top' in 1980 are not the same as in 2010, unlike Europe's flatter, more static structure.

“to do inequality right you got to take every person throughout her or his life rather than take a static inequality”

Static inequality measurement is analogous to a Markov chain with absorbing barriers (unit or zero probabilities); the proper dynamic treatment requires ensuring there are no absorbing states, mirroring how risk management must avoid ruin.

“it's like taking a markoff chain and make sure that there are no absorbent barrier, no absorbent probabilities, no unit one or zero probabilities in a markoff chain”

The physics formulation of Taleb's point is the distinction between time averages and ensemble averages; physicists understand this because of ergodicity theory, and economists avoid the error when modeling stochastic processes (ensuring no absorption) but commit it in the psychology of economics and in topics like inequality.

“the sort of physicalized version of what you're saying has to do with time averages versus ensemble averages”

Over-optimization is self-defeating ('pseudo-optimization'): driving a Ferrari at 500 km/h will not get you there faster than a bicycle because the odds are you will never arrive — extreme optimization without redundancy raises the probability of catastrophic failure.

“if you drive a Ferrari 500 km per hour, you're not going to get there faster than if you ride a bicycle cuz odds are you're never going to get there”

The number one way people get ruined in modern business is not betting too much but cutting corners and doing unethical or illegal things; ending up in prison or with a ruined reputation is equivalent to being wiped to zero, so reputational ruin should be treated as ergodic ruin and avoided absolutely.

“The number one way in which people get ruined in modern business is not by betting too much, but it's by cutting corners and doing unethical things... Ending up in an orange jumpsuit in prison or having a reputation ruined is the same as getting wiped to zero”

Holding cash in a bank makes you more robust and even antifragile because it lets you capture opportunities during a crisis, contrary to the common view that idle cash or excess inventory is merely a cost.

“People think it's silly to have cash in a bank. You're more robust if you have cash in a bank... and actually even antifragile because you can capture opportunities that way”