Activate this model when making decisions under significant uncertainty, when evaluating risks, or when the downside of being wrong is large.
Most risk intuition is calibrated to normal distributions — where extreme events are rare and the average case is representative. Many real-world risks are not normally distributed. In fat-tailed distributions, extreme events are far more frequent and consequential than Gaussian intuition suggests. Risk management built for normal distributions fails catastrophically in fat-tailed environments.
| Distribution | Characteristic | Planning Implication |
| --- | --- | --- |
| **Normal (thin tail)** | Extremes are rare and bounded | Planning for average + 2σ is sufficient |
| **Fat tail (power law)** | Extremes are common and can be unbounded | The rare extreme event may be more consequential than all average events combined |
Domains with fat tails: financial markets, earthquakes, wars, pandemics, social media virality, reputational events.
**Rule:** In fat-tailed domains, never plan only for the average case. The extreme event is not an anomaly — it is the expected outcome over a long enough horizon.
| Type | Response to Volatility | Example |
| --- | --- | --- |
| **Fragile** | Harmed by stress and volatility | Glass, over-leveraged institutions |
| **Robust** | Neutral to stress | Rocks, well-capitalized businesses |
| **Antifragile** | Gains from stress and volatility | Immune system, optionality portfolios |
**The design principle:** Prefer antifragile structures where possible. Eliminate fragility before adding complexity. Never be in a position where a single bad event is fatal.
The future is not a point but a distribution. Build three explicit scenarios:
| Scenario | Description | Use |
| --- | --- | --- |
| **Base case** | Most likely outcome given current trends | Primary planning assumption |
| **Optimistic** | Favorable variance from base case | Tests whether upside is worth pursuing |
| **Pessimistic** | Adverse variance from base case | Tests whether downside is survivable |
**Rule:** Prefer decisions that are acceptable under all three scenarios over decisions that are optimal in one and catastrophic in another.
Expected value (probability × magnitude) is necessary but not sufficient. Two bets with the same EV can have radically different variance profiles.
| Bet | EV | Variance | Preferred? |
| --- | --- | --- | --- |
| 50% chance of +$200, 50% of -$100 | +$50 | High | Depends on ability to absorb losses |
| 95% chance of +$60, 5% of -$10 | +$55 | Low | Usually yes, if EV is similar |
**Rule:** Always specify both EV and variance. Consider your ability to sustain the variance, not just the expected outcome.
> No bet is worth taking if a bad outcome ends the game entirely. Ruin is not a bad outcome — it is the end of all future outcomes.
**Before any significant commitment:** What is the worst realistic outcome? If it is ruin or near-ruin, the bet is disqualified regardless of expected value.
**Kelly Criterion:** The mathematically optimal bet size is a fraction of your capital, never a size that risks ruin. Kelly = edge / odds. In practice, use half-Kelly to account for model error.
Under uncertainty, structures that preserve future options are worth paying a premium for. Irreversible commitments should be deferred until information improves.
**The asymmetric value of optionality:** The downside is the option premium (small, defined). The upside is preservation of all future choices (large, open-ended).