How the Greeks would have hedged correlation

risk of foreign exchange options

Uwe Wystup
Commerzbank Treasury and Financial Products
Neue Mainzer Strasse 32–36
60261 Frankfurt am Main
GERMANY
[email protected]
http://www.mathfinance.de

Abstract

The author shows how to compute correlation coefficients in an n-

dimensional geometric Brownian motion model for foreign exchange rates,
interprets the result geometrically and applies it to eliminate correlation
risk when trading multi-asset options

1 Introduction
A multi-asset option is a derivative security whose payoff depends on the fu-
ture values of several possibly correlated underlying assets. Typical examples
are quanto, basket, spread, outside barrier options and options on the mini-
mum/maximum of several assets. The greatest risk in pricing such options lies
in the choice of the input parameter correlation, which is not directly observ-
able in market prices unless many rainbow options were traded actively on an
exchange. The other market input parameters interest rates and volatilities can
be obtained from and hedged in the money market and vanilla options mar-
ket. In a foreign exchange market the correlation structure can be computed
explicitly in terms of known volatilities - using the interdependence of exchange
rates. This implies in particular that the correlation risk of multi exchange rate
options can be hedged simply by trading FX volatility.
2 Uwe Wystup

2 Foreign exchange market model

We assume a constant coefficient geometric Brownian motion in n dimensions
(i) (i) (i)
dSt = St [µi dt + σi dWt ], i = 1, . . . , n, (1)

where µi is the rate of appreciation and σi is the volatility of FX rate i, and

(i) (j)
the correlation matrix ρ̂ = (ρij )i,j=1,...,n is defined by Cov(ln St , ln St ) =
σi σj ρij t. To illustrate the main idea we consider the example of a triangular
(1) (2) (3)
FX market St (GBP/USD), St (USD/JPY) and St (GBP/JPY) satisfy-
(1) (2) (3)
ing St St = St . Computing the variances of the logarithms on both sides
yields
σ 2 − σ12 − σ22
ρ12 = 3 . (2)
2σ1 σ2
This calculation can be repeated and thus visualized in elementary geometry.
Labeling the corners of a triangle GBP, USD, JPY, the vectors of the edges
σ~1 from GBP to USD, σ~2 from USD to JPY, σ~3 from GBP to JPY, such that
|σ~i | = σi , the law of cosine states that cos φ12 = ρ12 , where π − φ12 is the
angle of the triangle in the USD-corner. This way edge lengths can be viewed
as volatilities and (cosines of) angles as correlations. The correlation structure
turns out to be fully determined by the volatilities. Consequently we do not need
to estimate correlation coefficients and we can hedge correlation risk merely by
trading volatility. Therefore we expect a fairly tight multi exchange rate options
market to develop, because the major uncertainty correlation can be completely
erased. This may not work in the equity, commodity or fixed income market,
although we can imagine admitting one non-currency as long as both domestic
and foreign prices and volatilities are available.

3 The extension beyond triangular markets

Now we solve the slightly harder question how to obtain a correlation coefficient
of two currency pairs which do not have a common currency. For instance,
to compute the correlation between GBP/JPY and EUR/USD, we inflate the
market of these two currency pairs to the following market of six currency
(1) (2) (3) (4)
pairs St (GBP/USD), St (USD/JPY), St (GBP/JPY), St (EUR/USD),
(5) (6)
St (EUR/GBP) and St (EUR/JPY), denoting by σi the volatility of the
(i)
exchange rate St . Geometrically we introduce a tetrahedron with triangular
sides whose corners are the four currencies GBP, USD, JPY, EUR. The six
edge vectors will be labeled σ~i with lengths σi and the direction is always from
FX1
 to FX2 if the FX rate is named FX1/FX2. We need to determine the

6
= 15 correlation coefficients ρij , i = 1, . . . , 5, j = i + 1, . . . , 6, out of
2
which 12 = 3 × 4 are available as described in the above triangular market. The
remaining correlation coefficients ρ34 , ρ25 and ρ16 , can be obtained by using one
How the Greeks would have Hedged Correlation Risk 3

of the obvious relations

(3) (4) (3) (6) (3) (2)
Cov(ln St , ln St ) = Cov(ln St , ln St ) − Cov(ln St , ln St ) (3)

which results in
σ12 + σ62 − σ22 − σ52
ρ34 = , (4)
2σ3 σ4
The hardest part here is to remember how the signs go, but this will, of course,
depend on how one sets up the directions of the volatility vectors or on the
style FX rates are quoted in the market. But in principle, we can solve any
correlation problem in the FX market by the same method. Also note that given
a term structure of volatility one can create a corresponding term structure of
correlation.

1 2 3 4 5 6
vol 7.50% 13.45% 14.50% 13.00% 11.65% 16.85%
GBP/USD USD/JPY GBP/JPY EUR/USD EUR/GBP EUR/JPY
GBP/USD 1.0000 -0.1333 0.3936 0.4591 -0.1315 0.2478
USD/JPY -0.1333 1.0000 0.8586 -0.1887 -0.1247 0.6527
GBP/JPY 0.3936 0.8586 1.0000 0.0625 -0.1837 0.7336
EUR/USD 0.4591 -0.1887 0.0625 1.0000 0.8203 0.6209
EUR/GBP -0.1315 -0.1247 -0.1837 0.8203 1.0000 0.5333
EUR/JPY 0.2478 0.6527 0.7336 0.6209 0.5333 1.0000

Table 1: matrix of correlation coefficients for a four currency market

4 Geometric interpretation
If we introduce angles φ34 , φ25 and φ16 via cos φij = ρij , we can interpret the
three opposite edge correlations as the angles between the three pairs of skew
lines generated by the vectors of the tetrahedron. Note that equation (4) is also
worth noting as a purely geometric result about tetrahedra, which one might
want to name the law of cosine in a tetrahedron.
As a result we find that the correlation structure uniquely corresponds to visible
angles of a tetrahedron, where - as often is the case - orthogonality in geometry
corresponds to independence in stochastics. We also observe that the FX market
of six currency pairs we considered is actually only three-dimensional. Let us
point out that there may be a generalization to n dimensions in mathematics,
but for practical use a three-dimensional market suffices.
4 Uwe Wystup

Figure 1: four currency market volatilities represented as a tetrahedron

5 Hedging correlation risk

Computing correlation coefficients based on volatilities as above has a striking

implication. Any FX correlation risk can be transformed into a volatility risk.
More precisely, suppose we are given an option value function R(~σ , ρ̂), then,
as we found out, the correlation matrix ρ̂ is a redundant parameter which can
∆
be written as ρ̂ = ρ̂(~σ ). We can therefore write the value function as H(~σ ) =
R(~σ , ρ̂(~σ )). The correlation risk can now be hedged by replacing simple vegas
∂R
∂σi by adjusted vegas of the value function

5 6
∂H ∂R X X ∂R ∂ρjk
= + . (5)
∂σi ∂σi j=1 ∂ρjk ∂σi
k=j+1

This also means that the correlation risk of an option on the currency pairs
EUR/USD and GBP/JPY can be transferred to volatility risk in all six partici-
pating currency pairs, not just in the two EUR/USD and GBP/JPY. However,
hedging six vegas may be still easier than hedging two vegas and one correlation
risk. This is illustrated in the example of an outside barrier option in table 2.
The computation is based on [1].
An alternative approach to hedging correlation risk can be found in [2], where
an explicit relation between correlation risk and cross gamma is presented.
How the Greeks would have Hedged Correlation Risk 5

EUR/USD spot 0.9200 value (in USD) 0.0111 simple hedge adjusted hedge
GBP/JPY spot 167.50 correlation risk -0.0708 0.0000
EUR/USD strike 0.9200
GBP/JPY barrier 155.00 currency pairs vols simple vega adjusted vega
time (days) 180 GBP/USD 7.50% 0.0000 -0.2816
correlation 6.25% USD/JPY 13.45% 0.0000 0.5050
USD rate 5.50% GBP/JPY 14.50% -0.0495 -0.0190
EUR rate 3.00% EUR/USD 13.00% 0.0531 0.0871
JPY rate 0.50% EUR/GBP 11.65% 0.0000 0.4374
GBP rate 6.00% EUR/JPY 16.85% 0.0000 -0.6327

Table 2: correlation hedge for an outside down and out put option

References
[1] HEYNEN, R. and KAT, H. (1994). Crossing Barriers. Risk. 7 (6), pp. 46-51.
[2] REISS, O. and WYSTUP, U. (2000). Efficient Computation of
Option Price Sensitivities Using Homogeneity and other Tricks.
Preprint No. 584 Weierstrass-Insitite Berlin (http://www.wias-
berlin.de/publications/preprints/584).