Ergodicity Economics: Why Career Advice Based on Averages Quietly Ruins People

60-Second Summary

Ergodicity: a system is ergodic if its time-average (one person over long time) equals its ensemble-average (many people at one time). Career outcomes are almost never ergodic.
Traditional advice averages across many careers and reports the mean — but you live one career.
Bets that look good in expectation ('positive EV') can be ruinous for the individual because ruin ends the sequence.
For HR: severance size, emergency savings, insurance, and diversification are the correct response to non-ergodic career outcomes.
The most misleading four words in career advice: 'on average, people…'. On average is not you.

'Startup employees do fine on average — most eventually find another job.' True as an ensemble statement. Useless as personal advice. Because if the one startup you joined lays you off six months before your mortgage renewal, the average across many startups doesn't help you make rent.

Ensemble vs time average

“In economics… the crucial distinction between ensemble averages and time averages has largely been ignored. This has led to systematic and important errors.”
— Ole Peters, Nature Physics (2019)

Peters, working with Murray Gell-Mann, formalised a distinction economics had been fudging. An ensemble average is what happens across many parallel players and averaged. A time average is what happens to one player over time. In ergodic systems these are equal. In non-ergodic systems (multiplicative processes with ruin) they diverge — often by orders of magnitude. Career and financial outcomes are the second kind.

Classic illustration: a coin flip where heads increases wealth by 50% and tails decreases it by 40%. Expected value per flip is positive (+5%). But if you play it repeatedly, your wealth trajectory almost surely goes to zero over time. The ensemble average lies to the individual.

Where non-ergodicity matters

Six domains where averages mislead

1
Startup joining decisions
'Startup employees do well on average' — driven by long-tail winners. Median outcome is far below the mean. Your one draw is more like the median.
2
Equity compensation
Expected value of unvested options is positive. Time-average outcome is dominated by the specific company's fate. Diversification is not paranoia.
3
Layoff exposure
'Most laid-off workers find new jobs within X months' hides the tail that doesn't, and the tail that does but at large permanent income loss.
4
Founder outcomes
'Founders make more on average' — driven by outliers. Median founder has worse lifetime earnings than a senior employee at the same company.
5
Career risk-taking advice
'Take the bold role — most people who take risks are rewarded' — the ensemble may look good; your sequence contains one bad draw that compounds for decades.
6
Retirement planning
Sequence-of-returns risk is the ergodicity problem — average returns are irrelevant; the sequence you actually experience decides whether you run out of money.

Designing around non-ergodicity

Ensemble-thinking vs time-thinking

Ensemble thinking (dangerous)

'Expected value is positive, take the bet'
'Most people do fine'
'The market recovers on average'
'Founders average $X'
'Layoff recovery rates are Y%'

Time thinking (survivable)

'Would this ruin me if it went wrong?'
'What happens to me if I'm on the bad end?'
'Can I survive the sequence I actually experience?'
'Median founder outcome is what matters for me'
'What tail risk do I need to insure against?'

Non-ergodic career design

1
Avoid ruin, always
Any decision where a bad outcome permanently ends the sequence is disproportionately bad, regardless of paper EV.
2
Diversify income sources
One employer + one job function is a concentrated bet. Small side income, contracting capacity, or investment income diversifies sequence risk.
3
Build a 6–12 month runway
Emergency savings aren't fear-based — they're the mathematically correct response to living one career, not many.
4
Design severance and insurance for the tail
Severance policies should not be designed around 'most laid-off workers find jobs quickly'. Design for the tail.
5
Prefer bets with capped downside and long tails
Optionality (skills that compound, small experiments) is ergodicity-friendly: bounded loss, unbounded upside.

The HR leader implication

Every time you frame a policy as 'most people will be fine', you are averaging across an ensemble that doesn't include the specific people affected. The individual with the bad draw is the one your policy actually meets.

FAQ

Frequently asked questions

Isn't this just 'be risk-averse'?

No — more precise. Risk aversion is a preference; ergodicity is a structural fact about which systems produce equal time and ensemble averages. Rejecting positive-EV bets in a non-ergodic system isn't cowardice; it's correct.

Should nobody take startup roles?

They should — with clear eyes about the personal outcome distribution vs the ensemble pitch. Diversification, runway, and downside caps become non-negotiable.

How does this apply to workforce policy?

Severance, mental health support, and job-loss insurance become more defensible as ergodicity fixes — helping individuals survive one draw of a distribution that looks fine in aggregate.

Takeaways

You live one career, not many parallel ones. Averages across people are not averages across your futures.
Non-ergodic systems mean expected-value reasoning can quietly ruin individuals even when the ensemble looks great.
Design career choices (and HR policies) for the person with the bad draw.
Avoiding ruin, diversifying income, building runway, and capping downside are correct responses to non-ergodicity.