1

Multi-asset Options 1 In this lecture. . .

G G G G

how to model the behaviour of many assets simultaneously estimating correlation between asset price movements how to value and hedge options on many underlying assets in the BlackScholes framework the pricing formula for European non-path-dependent options on dividend-paying assets

2 Introduction
In this lecture we see the idea of higher dimensionality by examining the BlackScholes theory for options on more than one underlying asset. This theory is perfectly straightforward; the only new idea is that of correlated random walks and the corresponding multifactor version of Its lemma. Although the modeling and mathematics is easy, the final step of the pricing and hedging, the solution, can be extremely hard indeed. We see what makes a problem easy, and what makes it hard, from the numerical analysis point of view.

3 Multi-dimensional lognormal random walks

The basic building block for option pricing with one underlying is the lognormal random walk
G6 { 6 GW  6 G; 

This is readily extended to a world containing many assets via models for each underlying
G6
L

{ L6 L G W  L6 L G ;

     G , and { L and L Here 6 L is the price of the Lth asset, L are the drift and volatility of that asset respectively and G ; L is the increment of a Wiener process.

4 We can still continue to think of G ; L as a random number drawn from a Normal distribution with mean zero and standard deviation   GW so that
( >G ; L@ 

and

( >G ;
M

 L@

GW

but the random numbers G ; L and G ;

( >G ;
L

are correlated:
 LM G W

G;

M@

here  LM is the correlation coefficient between the Lth and M th random walks.
G

The symmetric matrix with  LM as the entry in the Lth row and M th column is called the correlation matrix.

5 For example, if we have seven underlyings G tion matrix will look like this:


and the correla-

             

                                                 

    

   

     

          M L.

         

           

Note that  LL   and  LM definite, so that y7 E y M

The correlation matrix is positive . The covariance matrix is simply ME M

where M is the matrix with the L along the diagonal and zeros everywhere else.

6 To be able to manipulate functions of many random variables we need a multidimensional version of Its lemma. If we have a function of the variables 6  ,. . . ,6 G and W, 9 6      6 G  W, then
G9

 '

9

 

; ;
G G L 

L M  LM 6 L6

# 9
M

M 

# 6 L# 6

  /9  ;
G

#9 #6
L

G 6 L

L 

We can get to this same result by using Taylor series and the rules of thumb:  G ; L  G W and G ; LG ; M   LM G W

4 Measuring correlations
If you have time series data at intervals of s W for all G assets you can calculate the correlation between the returns as follows. First, take the price series for each asset and calculate the return over each period. The return on the Lth asset at the N th data point in the time series is simply
5 LWN  6 LWN  s W > 6 LWN  6 LWN  

The historical volatility of the Lth asset is

; : : 9

 I 9

;


> 

# L9N  > #{ L

N 

where  is the number of data points in the return series and 5{ L is the mean of all the returns in the series.

8 The covariance between the returns on assets L and M is given by

 s W0

;


> 

5 LWN  > 5{ L5

M WN 

> 5{ M 

N 

The correlation is then

 I 9

;
0 M N 

>  P LP

# L9N  > 5{ L5

M WN 

> 5{ M 

9 Correlations measured from financial time series data are notoriously unstable.
1.0 0.8

0.6

0.4

0.2

0.0 0 -0.2 0.5 1 1.5

1.A correlation time series. The other possibility is to back out an implied correlation from the quoted price of an instrument. The idea behind that approach is the same as with implied volatility, it gives an estimate of the markets perception of correlation.

10

5 Options on many underlyings

Options with many underlyings are called basket options, options on baskets or rainbow options. The theoretical side of pricing and hedging is straightforward, following the BlackScholes arguments but now in higher dimensions. Set up a portfolio consisting of one basket option and short a number d L of each of the assets 6 L:
h 9 6      6 G  W >

;
G

@ L6 L

L 

11 The change in this portfolio is given by

GD  
 

 '

; ;
G G L  M 

9

P LP M  LM 6 L6

# 9
M

# 6 L# 6

 w ' ;  / 9
G L 

N
> @
L L

$

G 6 L

If we choose
@ #9
L

#6

for each L, then the portfolio is hedged, is risk-free. Setting the return equal to the risk-free rate we arrive at
#9 #W 
 

; ;
G G L 

P LP M  LM 6 L6

# 9
M

M 

# 6 L# 6

 U
M

;
G

#9
L

L 

#6

> U9  
L

(1)

This is the multidimensional version of the BlackScholes equation.

12 The modifications that need to be made for dividends are obvious. When there is a dividend yield of ' L on the Lth asset we have
#9 #W 
 

; ;
/ G L 

P LP M  LM 6 L6

# 9
M

M 

# 6 L# 6


M

;
G

7 >  L$

'
L

L 

$

> 7'
L

13

6 The pricing formula for European non-path-dependent options on dividend-paying assets

Because there is a Greens function for this problem we can write down the value of a European non-path-dependent option with payoff of Payoff6      6 /  at time 7 :

9


> 7 % > 9

 ~ 7 > W

> /

???

Payoff$  ??? $ / 

DetE 

> 

$ 

??? $

M M$ N M
OR J
L

H[ S

>

 

> 

N N

??? P / 
/$


> 

F
 L

??? / $ / 

L7 > 9



7 > 

>

N (2)
% > 9 
L

This has included a constant continuous dividend yield of each asset.

on

14

7 Exchanging one asset for another: a similarity solution

An exchange option gives the holder the right to exchange one asset for another, in some ratio. The payoff for this contract at expiry is
P D [ T  6


> T  6   

where T  and T  are constants. The partial differential equation satisfied by this option in a Black Scholes world is
#9 #W 
 

; ;

L 

P LP M  LM 6 L6

# 9
M

M 

# 6 L# 6


M

7 >  L$

'
L

L 

$

> 7'
L

A dividend yield has been included for both assets. Since there are only two underlyings the summations in these only go up to two.

15 This contract is special in that there is a similarity reduction. Lets postulate that the solution takes the form
9 6   6   W T  6  + }  W

where the new variable is

} 6 6
 

If this is the case, then instead of finding a function 9 of three variables, we only need find a function + of two variables, a much easier task. Changing variables from 6   6  to } we must use the following for the derivatives.
# #6 #
      

 # 6#} } 6
  




# #6


>

} # 6#} #


 } 6
 

#6

 6
 

#}

 

#6

#}

 

} # 6
 

#}

# 6 # 6

 >


#}

> 

 6
 

# #}

The time derivative is unchanged.

16 The partial differential equation now becomes

#+ #W  #    } 


+


#}

 

> 

 }

#+ #}

> '

+



where

 

>   

  

Y ou will recognise this equation as being the BlackScholes equation for a single stock with '  in place of U , '  in place of the dividend yield on the single stock and with a volatility of .

17 From this it follows, retracing our steps and writing the result in the original variables, that
9 6   6   W T 6  0
> 
 %

> 9

 /   > 6  $  0
  P % 

> 

 %

> 9

 /  

where
/

OR J 6  $   6  $     P

> 

> 9

% > 9

and

/

/ >

7 > W

18 The next example is of basket equity swap. This rather complex, high-dimensional contract, is for a swap of interest payments based on three-month LIBOR and the level of an index. The index is made up of the weighted average of 20 pharmaceutical stocks. To make matters even more complex, the index uses a time averaging of the stock prices.

Preliminary and Indicative For Discussion Purposes Only

International Pharmaceutical Basket Equity Swap

Indicative terms Trade Date Initial Valuation Date Effective Date Final Valuation Date Averaging Dates [] [] [] 26th September 2004 The monthly anniversaries of the Initial Valuation Date commencing 26th March 2004 and up to and including the Expiration Date US$25,000,000

Notional Amount
Counterparty floating amounts (US$ LIBOR) Floating Rate Payer Floating Rate Index Designated Matrurity Spread Day Count Fraction Floating Rate Payment Dates Initial Floating Rate Index The Bank Fixed and Floating Amounts (Fee, Equity Option) Fixed Amount Payer Fixed Amount Fixed Amount Payment Date Basket

[] USD-LIBOR Three months Minus 0.25% Actual/360 Each quarterly anniversary of the Effective Date []

XXXX 1.30% of Notional Amount Effective Date A basket comprising 20 stocks and constructed as described in attached Appendix Will be set at 100 on the Initial Valuation Date XXXX

Initial Basket Level Floating Equity Amount Payer

Floating Equity Amount

Will be calculated according to the performance of the basket of stocks in the following way:
BASKET average 100 Notional Amount * max 0, 100

where
BASKET average = 100 * Paverage Weight * P 120 stocks initial

Floating Equity Amount Payment Date

And for each stock the Weight is given in the Appendix P_initial is the local currency price of each stock on the Initial Valuation Date P_average is the arithmetic average of the local currency price of each stock on each of the Averaging Dates Termination Date

Appendix Each of the following stocks are equally weighted (5%): Astra (Sweden), Glaxo Wellcome (UK), Smithkline Beecham (UK), Zeneca Group (UK), Novartis (Switzerland), Roche Holding Genus (Switzerland), Sanofi (France), Synthelabo (France), Bayer (Germany), Abbott Labs (US), Bristol Myers Squibb (US), American Home Products (US), Amgen (US), Eli Lilly (US), Medtronic (US), Merck (US), Pfizer (US), Schering-Plough (US), Sankyo (Japan), Takeda Chemical (Japan).
This indicative termsheet is neither an offer to buy or sell securities or an OTC derivative product which includes options, swaps, forwards and structured notes having similar features to OTC derivative transactions, nor a solicitation to buy or sell securities or an OTC derivative product. The proposal contained in the foregoing is not a complete description of the terms of a particular transaction and is subject to change without limitation.

21

8 Realities of pricing basket options

The factors that determine the ease or difficulty of pricing and hedging multi-asset options are
G G G G

existence of a closed-form solution number of underlying assets, the dimensionality path dependency early exercise We have seen most of these in a single-asset setting. The solution technique that we use will generally be one of finite-difference solution of a partial differential equation numerical integration Monte Carlo simulation

G G G

22 8.1 Easy problems If we have a closed-form solution then our work is done; we can easily find values and hedge ratios. This is provided that the solution is in terms of sufficiently simple functions for which there are spreadsheet functions or other libraries. If the contract is European with no path-dependency then there may be a solution in the formof a multiple integral. If this is the case, then we often have to do the integration numerically. 8.2 Medium problems If we have low dimensionality, less than three or four, say, the finitedifference methods are the obvious choice. They cope well with early exercise and many path-dependent features can be incorporated, though usually at the cost of an extra dimension. For higher dimensions, Monte Carlo simulations are good. They cope with all path-dependent features. Unfortunately, they are not very efficient for American-style early exercise.

23 8.3 Hard problems The hardest problems to solve are those with both high dimensionality, for which we would like to use Monte Carlo simulation, and with early exercise, for which we would like to use finite-difference methods. There is currently no numerical method that copes well with such a problem.

24

9 Realities of hedging basket options

Even if we can find option values and the greeks, they are often very sensitive to the level of the correlation. But the correlation is a very difficult quantity to measure. So the hedge ratios are very likely to be inaccurate. If we are delta hedging then we need accurate estimates of the deltas. This makes basket options very difficult to delta hedge successfully. When we have a contract that is difficult to delta hedge we can try to reduce sensitivity to parameters, and the model, by hedging with other derivatives. This is the basis of vega hedging. We could try to use the same idea to reduce sensitivity to the correlation. Unfortunately, that is also difficult because there just arent enough contracts traded that depend on the right correlations.