Worst-Of Options And Correlation Skew

A best-of option pays out on the strongest member of a group; a worst-of option pays on the weakest, flipping its risk profile from an investor's friend into an investor's enemy. It's like a bonus that depends not on your own performance, but on the worst-performing member of your whole team — one bad performer ruins it for everyone. That sounds punishing, and it is — which is exactly why banks can pay a much higher coupon on worst-of notes: the investor is implicitly selling protection against the single weakest link in a basket, and gets paid for taking that concentrated risk.

The payoff and why correlation dominates

A worst-of put (the payoff structure embedded in most worst-of autocallable notes) pays based on whichever asset in the group fell the most:

$$\text{Payoff} = \max\!\Big(K - \min(S_1(T), S_2(T), \dots, S_n(T)),\ 0\Big).$$

In words: $\min(S_1(T), \dots, S_n(T))$ picks out whichever underlying ended up lowest, and the payoff is triggered by how far that single weakest performer fell below strike $K$, regardless of the others. Unlike a basket option, correlation here doesn't average risk away — it's the dominant driver of value. Low correlation means each underlying can independently have a bad day, so the chance at least one falls hard is high even if the group looks fine on average, making the put expensive. High correlation means the assets fall together or not at all, so "the worst performer" isn't much worse than "the group," making the same put considerably cheaper.

Worked example 1 — low correlation inflates the risk

Three stocks, each with a 10% chance of falling more than 20% over a year, nearly uncorrelated. The chance that at least one falls more than 20% is roughly $1-(1-0.10)^3=1-0.729=0.271$, about 27% — nearly triple any single stock's own 10% chance. A worst-of put triggered by a >20% drawdown in the weakest member is priced closer to that 27% than to the 10% an investor might naively associate with "a diversified basket."

Worked example 2 — high correlation shrinks the gap

Same three stocks, same individual 10% chance, but now correlation is very high ($\rho \approx 0.95$, all driven by the same macro factor). When one falls 20%, the other two very likely fell close to 20% too — the joint probability of "at least one falls more than 20%" is now much closer to the single-stock 10%, maybe 12-13%, because the three outcomes are barely distinguishable rather than three separate chances to fail. The same put, same strike, same names, prices roughly twice as risky under low correlation as under high — purely from the correlation assumption, with no change to any stock's own volatility.

Drag correlation toward zero and notice how a wide scatter makes it easy to find a point where at least one coordinate is deep in negative territory even when the "average" point looks fine — that's the mechanism inflating a worst-of put at low correlation. Push correlation up and the cloud tightens into a line, where a bad outcome for one asset means a bad outcome for all, not an extra independent chance to fail.

Correlation skew

Just as implied volatility varies by strike (the volatility "smile"), the correlation the market implicitly prices into worst-of structures varies with how deep the trigger is set — deeper, more punishing triggers tend to imply lower correlation than at-the-money ones, mirroring how out-of-the-money puts imply higher volatility. This "correlation skew" exists partly because banks selling many worst-of notes accumulate a large short-correlation position, and the cost of hedging that concentrated exposure gets baked into the correlation implied back out of the notes' prices, not pulled from a flat historical average.

What this means in practice

Worst-of autocallables are among the most heavily sold retail structured products because low correlation between the underlyings lets the issuer offer a large headline coupon while keeping the note attractively priced — the coupon compensates for exactly the concentrated correlation risk described above. Desks that sell many of these actively manage correlation as its own risk factor, often larger in dollar terms than their exposure to any individual name's volatility.