Skewness and Asymmetry: Why Win Rate Tells You Nothing

Most people evaluate a strategy by asking: does it work more often than it fails? Win rate, hit rate, batting average — the frequency of positive outcomes. If you're right 70% of the time, the strategy seems good. If you're right 30% of the time, it seems bad.

Nassim Taleb's second major argument in Fooled by Randomness is that this evaluation is a category error. Win rate is only informative if the magnitude of wins and losses are roughly symmetric. In skewed distributions — which describe most of the interesting cases — the frequency of outcomes tells you almost nothing about whether a strategy is profitable.

The Trader Who Wins 999 Days Out of 1,000

Here's the setup. A trader runs a strategy with: - 999/1,000 chance of making $1 - 1/1,000 chance of losing $10,000

The expected value per bet: (999/1,000 × $1) + (1/1,000 × −$10,000) = $0.999 − $10 = −$9.001

This is a money-losing strategy with a negative expected value of roughly $9 per bet. And yet: any given day, week, or month is almost certain to be profitable. The win rate over a year approaches 99.9%. The Sharpe ratio — which measures return per unit of volatility — looks excellent because the volatility is small until the loss event.

The strategy looks like genius right up until the 1/1,000 outcome arrives. At that point, all the accumulated $1 wins are insufficient to offset the single $10,000 loss. The strategy is a money-loser, mathematically, even though it felt like a money-maker for a very long time.

Why This Matters for Real Strategies

The 999/1,000 example is stylized to make the math visible. But the structure it illustrates is present in real strategies, constantly.

Short-volatility funds sell options and collect premiums in exchange for being exposed to large moves. In normal markets, the premiums arrive steadily and the moves don't. Win rate is high, drawdowns are rare, Sharpe ratios look clean. Then a volatile period arrives — August 2007, March 2020, or the vol event of February 2018 — and decades of premium income evaporate in weeks.

Yield-enhancement strategies — mortgage-backed securities before 2008, structured notes, carry trades — offer higher-than-market yields by implicitly taking on tail risk. Every coupon payment is a positive outcome. The win rate is very high. The strategy works until the tail event the yield was compensating for arrives, at which point the loss swamps the accumulated income.

Insurance-like positions generally are the category. You collect a small premium or advantage in the normal case, while being exposed to a rare but catastrophic loss. Your win rate is whatever percentage of periods are normal. Your expected value depends on the magnitude of the catastrophic loss and its probability — and if those are misweighted, the high win rate is actively misleading.

The Dentist Who Counts Nickel Wins

Taleb's vivid framing: picking up nickels in front of a steamroller. You walk in front of traffic and pick up $0.05. Every pass is a "win." Your win rate is perfect. The strategy continues to work until the one pass where the steamroller doesn't stop in time, and the single event eliminates all previous gains and a great deal more.

The nickel/steamroller structure appears wherever: - A strategy has many small positive events - And a rare, very large negative event - And the strategy is implicitly "short" of that negative event (exposed to it without compensation)

The key diagnostic: if a strategy's returns look suspiciously smooth over a long period, ask what it is implicitly short of. Smooth, consistent returns in a volatile world almost always mean someone is selling tail insurance, intentionally or not. The premiums are the smooth returns. The steamroller is the thing they haven't met yet.

The Asymmetry Works Both Ways

Taleb's insight also applies in the other direction, and this is where it becomes useful rather than just cautionary.

A strategy with: - 1/10 chance of making $1,000 - 9/10 chance of losing $1

Expected value: (1/10 × $1,000) + (9/10 × −$1) = $100 − $0.90 = $99.10 per bet.

This strategy loses money on nine out of ten bets. Win rate is 10%. By frequency alone, it looks terrible. But the rare good outcome is so large that the expected value is strongly positive. The insurance buyer, the long-option holder, the concentrated venture investor all live here. Most positions fail; the one that hits covers all the others.

This is the structure Taleb's surviving traders use. They accept a high rate of small losses in exchange for rare, disproportionately large wins. Their Sharpe ratios look bad by conventional standards. Their win rates look dismal. And they're running a mathematically sound strategy that is robust to the rare events their short-volatility competitors are exposed to.

What to Count Instead of Win Rate

The corrective is expected value: probability weighted by magnitude.

For any strategy or decision, ask: 1. What are the realistic outcome states? (Not just the likely ones — all of them, including the tails.) 2. What is the probability of each? 3. What is the magnitude of each? 4. What is probability × magnitude summed across all states?

If the expected value is positive, the strategy is sound regardless of win rate. If the expected value is negative, the strategy is unsound regardless of win rate.

The practical challenge: tail probabilities and tail magnitudes are the hardest to estimate. Tail events are rare by definition, which means the historical data underrepresents them. This is exactly the regime in which the 999/1,000 trader's strategy looks most convincing — there's little data on the event the strategy is short of, which is why the risk is invisible until it arrives.

For the full framework, read Fooled by Randomness: How Luck Masquerades as Skill.