ACOMB-Correlation in Practice
ACOMB-Correlation in Practice
1. Quanto Option
+
S JPY
T
NUSD 0 −1
S JPY
Imagine that the share price is 100JPY and the delta is 1. In this case fund of
the hedge is done in JPY. If FX rates move then the JPY of the contract
moves, but the value of the hedge does not. Weak exposure to the correlation
between FX and Equity
Like an option on a ADR. Every time we trade delta we fund in USD. Volatility
is the volatility of S measured in USD. Strong exposure to FX and Equity
correlation.
1. Spread Option
+
ST1 ST2
1 − 2 − K
S0 S0
3. Basket Options
+
i
∑ wi STi − K
i S
0
4. Himalayan Options
+
Si
w Max ti −K
∑ i
remaining i S i
i 0
5. Rainbow Options
+
i Si
w1 Max STi + w2 Min Ti − K
i S
0 i
S0
• General idea is that when the correlation is low (negative) there is a more diverse
range of outcomes, hence there will always be one asset which has outperformed
and one that has under performed. (Consider the case of perfect negative
correlation.)
• Following exposure of product to a rise in correlation
Call Put
Best-Of - +
Worst-Of + -
All leave the seller short correlation. Difficult to manufacture a product which has
correlation exposure in the other direction.
First attempt would be to use the time series of two underlyings and measure the correlation of
the time series. Rolling window of 6 months daily data Nikkei 225 and S&P 500
Note enormous range. For ρ=0.2 and 6 months daily data – statistic confidence interval is
[0.02, 0.36]. Are we picking up sampling error, or uncertainty in correlation
Negative Definite matrix implies that there are portfolios with negative variance.
(In portfolio theory can we get away with this if we insist that all assets have
positive weight ?)
Q R = QQT Z j = ∑ q jkWk
Find pseudo-square root so that and k
One would imagine that if we use time series over the same period and estimated
from this the correlation matrix would be positive definite by definition.
For pricing purposes would prefer to use correlations implied from option contracts
that I can see. For example baskets and dispersions.
Many people want to know what happens when correlation goes up by 10%.
Useful transform
ρ → ρ + α (1 − ρ )
If covariance matrix is positive semi-definite then this transform will maintain this
property
If volatility is a function of (percentage) strike then ρ must also be a function of this strike.
σ I2 ( K ) = ∑ wi w j ρij ( K )σ i ( K )σ j ( K )
i, j
We know that index volatility smiles are steeper than single stock volatility smiles so not surprisingly we
find that correlation is higher for low strikes than for higher strikes
1. We know that distributions of assets are not lognormal. To make more sense
of the information we should be using the implied distributions of each stock.
2. Given the distributions of each asset does not determine the joint distribution of
two assets. There is a extensive theory of this called copula.
3. If we used local volatility as a process we would get very different results.
If trading correlation the most objective way of doing so will be with variance swaps.
• α = 10%
ρ = 90% → 91%
ρ = 20% → 28%
On initiation product has little individual vega, but has exposure to correlation.
• When share prices move this will no longer be the case.
• Just as the vega of a vanilla option dissipates when the option move away from ATM, so the
correlation exposure of a basket option dissipates as you move away from ATM.
Variance Swaps provide a simple way of trading variance without an explicit strike dependency.
• Possible on the indices with a small number of assets, but can be approximated on indices
such as the S&P with tracking portfolios
• As these are strike-less the correlation exposure does not dissipate as the underlying assets
move.