Why Are Extreme Outcomes More Common Than We Expect?

How fat tails, power laws, and reinforcing feedback explain why extraordinary events occur more frequently than ordinary statistics predict.
Imagine measuring the heights of 10,000 adults. Most cluster around the average. A few are exceptionally tall or short, but nobody stands 20 feet tall.
Now measure the wealth of 10,000 adults. The distribution looks radically different. A small number possess vastly more than everyone else.
Why?
Some measurements cluster around an average, while others produce enormous extremes. The difference lies partly in their statistical distributions and the processes that generate them.
Why Do Some Extremes Seem Almost Impossible?
The normal distribution, commonly called the bell curve, describes many measurements that cluster around a central value.
Human height is a useful approximation. Most adults fall within a relatively narrow range, and extreme deviations become progressively rarer.
In a perfect normal distribution, an observation at least five standard deviations from the mean occurs about once in 1.7 million observations.
But many real-world systems behave differently.
Financial losses, earthquake energy, city populations, and concentrations of wealth exhibit patterns in which extreme outcomes occur far more frequently than a normal distribution predicts.
These distributions have fat tails, also called heavy tails.
The difference matters because a risk considered practically impossible under one statistical model could be significant under another.
What Is a Power Law?
A power law is one mathematical pattern capable of producing extremely unequal outcomes.
Rather than extreme values becoming negligible as quickly as they do under a normal distribution, their frequency decreases according to a power relationship.
For example, in a hypothetical distribution, an outcome ten times as large might occur one-hundredth as often. Increase the size another tenfold, and its frequency decreases by the same factor.
That consistent scaling relationship defines a power law.
Figure: Illustrative comparison of probabilities of extreme deviations. The heavy-tailed model assigns substantially greater probability to very large deviations. Both curves are theoretical models, not observed data.
Not every fat-tailed distribution follows a power law.
In a 2009 study, Aaron Clauset, Cosma Shalizi, and M. E. J. Newman examined 24 datasets previously associated with power laws. Some were statistically consistent with power-law behavior; others were not.
That distinction matters when predicting risk. Identifying an extreme outcome is not enough to determine its probability. The underlying distribution must be tested rather than assumed.
Why Do Extreme Outcomes Become So Large?
Three mechanisms help explain why some systems produce enormous differences.
1. Multiplicative Growth
Imagine two hypothetical businesses, each starting with $100,000 in annual revenue.
Business A grows 5% annually. Business B grows 20%.
After ten years:
| Business A | Business B | |
|---|---|---|
| Starting revenue | $100,000 | $100,000 |
| Annual growth | 5% | 20% |
| Revenue after 10 years | $162,889 | $619,174 |
Both businesses started at exactly the same point. Their different growth rates produced nearly a fourfold difference in revenue.
When growth rates vary across thousands of businesses and compound over time, outcomes spread dramatically.
This example demonstrates divergence through compounding, not a power law by itself. The eventual distribution depends on how growth rates vary, how long growth continues, and other constraints.
2. Cumulative Advantage
Success creates opportunities for additional success.
A popular product attracts customers. Those customers generate reviews, recommendations, and recognition. Greater visibility attracts still more customers.
The advantage feeds on itself.
The Barabási–Albert model, published in 1999, demonstrates how networks develop highly connected hubs when new connections preferentially attach to already well-connected nodes.
A few nodes accumulate enormous numbers of connections while most remain relatively small.
This explains one route through which modest initial differences develop into highly unequal outcomes.
3. Interdependence and Cascades
Some extreme events emerge when one disturbance triggers a chain reaction.
A financial institution experiences losses, exposing other institutions to trouble. Those institutions sell assets, driving prices lower and forcing additional losses.
The initial shock becomes a cascade.
The damage depends not only on the original event but also on how strongly the system's components influence one another.
Interdependence transforms local problems into systemwide crises.
What Did Mandelbrot Discover About Financial Markets?
In his influential 1963 paper, The Variation of Certain Speculative Prices, mathematician Benoit Mandelbrot investigated cotton-price fluctuations.
His research challenged the assumption that price movements followed a normal distribution.
One particularly important observation was that large price changes tended to cluster. Large movements were followed by other large movements, whether upward or downward, while quieter periods tended to remain quieter.
This meant market volatility was not adequately described by a model of independent, normally distributed daily changes.
Mandelbrot proposed a mathematical alternative that gave much greater weight to extreme fluctuations.
The practical implication remains important: a risk model calibrated to ordinary market behavior can dramatically underestimate the likelihood of exceptional losses.
Why Do We Keep Underestimating Extremes?
Three common assumptions distort our expectations.
| Assumption | What to examine instead |
|---|---|
| Outcomes cluster around the average | Examine the complete distribution, especially its tails |
| Events happen independently | Identify feedback loops, shared exposures, and cascading failures |
| Historical averages describe future risk | Test extreme scenarios and assess how the system responds |
The central problem is not that averages are useless. It's that averages alone conceal concentration, dependence, and vulnerability to rare events.
An organization preparing for disruption should ask what happens when several unfavorable conditions occur together.
What if demand falls while borrowing costs rise? What if a supplier fails precisely when inventory is low? What if a market decline forces multiple investors to sell simultaneously?
These are stress tests, not predictions.
Their purpose is to identify vulnerabilities before those vulnerabilities become emergencies.
Margin of safety matters because extreme outcomes cannot always be forecast precisely. Cash reserves, backup suppliers, diversified exposures, and spare operating capacity create room to absorb surprises.
The appropriate buffer depends on the cost of failure, the range of plausible outcomes, and the resources available.
What Should We Learn From Extreme Outcomes?
Extraordinary outcomes are not necessarily statistical anomalies.
In systems governed by heavy-tailed distributions, extremes occur far more frequently than bell-curve assumptions suggest.
Multiplicative growth creates divergence. Cumulative advantage amplifies differences. Interdependence spreads disturbances.
None of these mechanisms means every system follows a power law or that every catastrophe is predictable.
The lesson is to understand the system before deciding what qualifies as unlikely.
The next time an enormous success, market collapse, or unexpected disruption surprises everyone, ask:
Was the outcome truly extraordinary, or were our expectations built around the wrong distribution?
📚Bookmarked for You
The Black Swan by Nassim Nicholas Taleb: Explores unexpected events and the limits of prediction.
The Misbehavior of Markets by Benoit Mandelbrot and Richard L. Hudson: Examines financial risk, extreme price movements, and the limitations of conventional market models.
🧬 QuestionStrings to Practice: Understanding Extremes
Use this sequence to evaluate a system where unusually large outcomes matter.
What outcomes are unusually large? → How frequently do they occur compared with ordinary outcomes? → What processes amplify initial differences? → How connected are the events producing those outcomes? → What safeguards or decisions should change?
Explore another question every day at QuestionClass.
You just thought more carefully than most people do all day.
One question like this, every morning. Free.
Start My Daily Practice →No password needed · Takes 30 seconds
Discover more from QuestionClass
Subscribe to get the latest posts sent to your email.