Basket and Rainbow Options
Every exotic so far has had exactly one underlying. A
multi-asset option pays off on several underlyings at once — a
currency basket, a handful of tech stocks, a commodity index's components. Two payoff shapes
dominate in practice. A basket option pays on a weighted
average of the underlyings, as if they had been blended into one synthetic asset. A
rainbow option pays on the best or worst performer among
them — a genuinely different bet, since it singles one asset out rather than blending them.
\text{basket call: } \Big(\textstyle\sum_i w_i S_i(T) - K\Big)^+ \qquad
\text{best-of call: } \Big(\max_i S_i(T) - K\Big)^+
A worst-of structure is the mirror image — often written as a put,
(K - \min_i S_i(T))^+, since desks mostly sell worst-of
protection, not worst-of upside. The single fact that dominates pricing all of these,
far more than it dominated anything earlier in this module, is how the underlyings
move together.
Why correlation is everything here
A basket built from n assets with weights
w_i, volatilities \sigma_i and
pairwise correlations \rho_{ij} behaves — to a standard
moment-matching approximation — like a single lognormal asset with its own effective
volatility:
\sigma_{\text{basket}}^2 = \sum_i \sum_j w_i w_j\, \rho_{ij}\, \sigma_i \sigma_j.
Push every \rho_{ij} \to 1 (the underlyings move in lockstep) and
this collapses to \sigma_{\text{basket}} = \sum_i w_i \sigma_i,
the simple weighted average of the individual volatilities — no diversification benefit
survives. Push correlation down toward 0 or negative, and the
assets' wiggles start to cancel inside the sum, shrinking
\sigma_{\text{basket}} well below that weighted average. Since an
option's value rises with the volatility of the random variable in its payoff, a
basket call gets cheaper as correlation falls — lower correlation means more
netting, less spread in the basket's own outcome.
Best-of and worst-of rainbows move the opposite way for a subtler reason: they
don't care about netting, they care about the chance that at least one name does
something extreme. With many weakly-correlated assets, it becomes far more likely that
some asset in the basket has a great year (helping a best-of call) — and, just as
easily, that some asset has a terrible one (hurting a worst-of put's seller). Low
correlation therefore makes a best-of call more valuable and a worst-of put
more valuable too, even though it makes a basket call less valuable.
Correlation isn't a single dial that moves every multi-asset option the same way — it moves
basket and rainbow structures in opposite directions.
Seeing the correlation effect
Take a basket that is a straight 50/50 blend of two stocks, each starting at
100, with volatilities 25\% and
30\%, struck at-the-money with a year to run. The chart below
prices an at-the-money call on that basket, using the moment-matching approximation above,
as correlation sweeps from -1 to +1.
The value climbs steadily as the two legs become more correlated — exactly the pattern the
formula predicts, and a striking reminder that the same two options, on the
same two underlyings, can be worth very different amounts depending on a number
(correlation) that has nothing to do with either stock's own volatility.
A worst-of put on the same two names would trace the opposite shape — most
expensive when correlation is low (plenty of ways for at least one leg to crater
independently) and cheapest when correlation is high (the two legs live or die together, so
there's little extra "one of them might fail alone" risk to insure against).
- Basket call/put: (\sum_i w_i S_i(T) - K)^+ / (K - \sum_i w_i S_i(T))^+ — a weighted blend treated as one synthetic underlying.
- Best-of/worst-of rainbow: (\max_i S_i(T) - K)^+ or (K - \min_i S_i(T))^+ — singles out one extreme performer, not a blend.
- Correlation enters directly through \sigma_{\text{basket}}^2 = \sum_i\sum_j w_i w_j \rho_{ij}\sigma_i\sigma_j; two assets have a closed-form (Margrabe-related) rainbow price, three or more generally need Monte Carlo over correlated paths.
In the years before the 2008 crisis, retail-facing "worst-of autocallable" notes — paying
an attractive coupon unless the worst performer among, say, three unrelated blue
chips breached a barrier — were sold in huge volume, because a worst-of structure lets an
issuer offer a headline coupon that looks generous for what seems like modest risk. The
catch is exactly the correlation effect above: with three imperfectly correlated stocks,
the chance that at least one of them has a bad year is much higher than the chance
any single stock does. Many buyers priced the note as if it carried single-stock risk, when
it actually carried something closer to "worst of three independent coin flips" risk —
understanding correlation wasn't a technicality, it was the entire difference between a
fair price and a nasty surprise.
It is tempting to price a basket call by just averaging the prices of individual
vanilla calls on each underlying (with strikes chosen so their weighted sum equals
K). This is a real and costly mistake, and not merely
"approximately fine": a portfolio of options
\sum_i w_i (S_i(T) - K_i)^+ floors each asset
separately at zero before summing, so a bad leg can never be offset by a good one.
A true basket option floors the sum at zero,
(\sum_i w_i S_i(T) - K)^+, which lets a strong leg's gains net
against a weak leg's losses before the floor is applied. Netting can only reduce
the payoff, so a genuine basket option is always worth less than or equal to
the corresponding portfolio of individual vanilla options — pricing it as that portfolio
systematically overprices it, and ignores correlation entirely.