Tải bản đầy đủ - 0 (trang)
5 Code: Cliquet with uncertain volatility, in similarity variables

5 Code: Cliquet with uncertain volatility, in similarity variables

Tải bản đầy đủ - 0trang

924



Part Five advanced topics



Dim Vmax, AssetStep, TStep, QStep, Delta, Gamma, Theta, Tim, Vol, qafter, _

frac, V1, V2 As Double

Dim i, j, k, M, iafter, kafter, N, NumSoFar, NumQSteps As Integer

ReDim jtest(1 To NumFixes) As Integer

Vmax = Application.Max(VolMin, VolMax)

AssetStep = 1 / NumAssetSteps

TStep = 0.95 * AssetStep ^ 2 / Vmax ^ 2 / 2 ^ 2 ’ A) This ensures stability _

of the explicit method

M = Int(Expiry / TStep) + 1

TStep = Expiry / M

QStep = AssetStep

NumQSteps = Int(Strike2 / QStep) * NumFixes

ReDim Q(0 To NumQSteps)

ReDim VOld(-NumAssetSteps To NumAssetSteps, 0 To NumQSteps) ’ First dimension _

centered on xi = 1

ReDim VNew(-NumAssetSteps To NumAssetSteps, 0 To NumQSteps)

NumSoFar = 1

For j = 1 To NumFixes - 1

jtest(j) = Int(j * Fixing / TStep) ’ Used in testing whether fixing date _

has been passed

Next j

For k = 0 To NumQSteps

Q(k) = k * QStep

For i = -NumAssetSteps To NumAssetSteps

xi(i) = 1 + AssetStep * i ’ i = 0 corresponds to xi = 0

VOld(i, k) = Application.Max(Strike1, Q(k) + _

Application.Max(0, Application.Min(Strike2, xi(i) - 1))) ’ B) _

Payoff

Next i

Next k

For j = 1 To M ’ C) Time stepping

For k = 0 To NumQSteps

For i = -NumAssetSteps + 1 To NumAssetSteps - 1

Delta = (VOld(i + 1, k) - VOld(i - 1, k)) / 2 / AssetStep ’ Central difference

Gamma = (VOld(i + 1, k) - 2 * VOld(i, k) + VOld(i - 1, k)) / AssetStep / AssetStep

Vol = VolMax

If Gamma > 0 Then Vol = VolMin ’ Volatility depends on gamma in the uncertain _

volatility model

Theta = IntRate * VOld(i, k) - 0.5 * Vol * Vol * xi(i) * xi(i) * Gamma _

- (IntRate - Div) * xi(i) * Delta ’ _

The Black-Scholes equation

VNew(i, k) = VOld(i, k) - TStep * Theta

Next i

VNew(-NumAssetSteps, k) = VOld(-NumAssetSteps, k) * (1 - IntRate * TStep) ’ D) _

Boundary condition at xi = 0

VNew(NumAssetSteps, k) = VNew(NumAssetSteps - 1, k) ’ D) Boundary condition at _

xi = infinity. Delta = 0

For i = -NumAssetSteps To NumAssetSteps



volatility case study: the cliquet option Chapter 56



VOld(i, k) = VNew(i, k) ’ E) Overwriting with new data

Next i

Next k

If jtest(NumSoFar) = j Then ’ F) Test for a fixing date

For i = -NumAssetSteps To NumAssetSteps

For k = 0 To NumQSteps

qafter = Q(k) + Application.Max(0, Application.Min(Strike2, xi(i) - 1)) ’ G) _

The updating rule

kafter = Int(qafter / QStep)

frac = (qafter - QStep * kafter) / QStep

V1 = 0

V2 = 0

If kafter < NumQSteps Then

V1 = VNew(0, kafter)

V2 = VNew(0, kafter + 1)

End If

VOld(i, k) = (1 - frac) * V1 + frac * V2 ’ The jump condition. Linear _

interpolation

Next k

Next i

NumSoFar = NumSoFar + 1

End If

Next j

cliquet = VOld ’ Output the whole array

End Function



This program clearly leaves much to be desired, for example in the discretization, the treatment of the jump condition etc. But it does have the benefit of transparency.

A:



The time step is set so that the explicit finite difference method is stable. If the time step

is any smaller than this, the method will not converge.



B:



Here the payoff is set up; the dependent variable as a function of the independent variables.



C:



The time-stepping engine. Delta and gamma are discretized versions of the first- and

second-order derivatives with respect to S. This part of the code also treats the uncertain

volatility. See how the volatility depends on the sign of gamma.



D:



The boundary conditions, for ξ = 0 and large ξ .



E:



Updating the next step back in the grid.



F:



Here the code tests for a fixing date.



G:



Across fixing dates the updating rule is applied. This is really the only point in the code

that knows we are pricing a cliquet option.



925



926



Part Five advanced topics



56.6 SUMMARY

The cliquet can be a subtle instrument, with a dependence on volatility that is far from obvious.

In our example there was a high degree of sensitivity to volatility but not to level, nor to current

skew. No, that sensitivity was to a rotation of forward volatility skew about a point 4% (half

way between the floor and the cap) above the previous year’s asset price. And that level just

happens to be at approximately where the risk-neutral asset is expected to be. Pure genius,

albeit an evil genius!

I hope this chapter has shown you, using a concrete example, just how sensitive contracts

can be to volatility, when they might appear to be quite harmless.



FURTHER READING

• This work on cliquet options was first published in Wilmott (2002).



CHAPTER 57



jump diffusion

In this Chapter. . .









the Poisson process for modeling jumps

hedging in the presence of jumps

how to price derivatives when the path of the

underlying can be discontinuous

• jump volatility

57.1



INTRODUCTION



There is plenty of evidence that financial quantities, be they equities, currencies or interest rates,

for example, do not follow the lognormal random walk that has been the foundation of almost

everything in this book, and almost everything in the financial literature. We look at some of

this evidence in a moment. One of the striking features of real financial markets is that every

now and then there is a sudden unexpected fall or crash. These sudden movements occur far

more frequently than would be expected from a Normally distributed return with a reasonable

volatility. On all but the shortest timescales the move looks discontinuous; the prices of assets

have jumped. This is important for the theory and practice of derivatives because it is usually

not possible to hedge through the crash. One certainly cannot delta hedge as the stock market

tumbles around one’s ankles, and to offload all one’s positions will lead to real instead of paper

losses, and may even make the fall worse.

In this chapter I explain classical ways of pricing and hedging when the underlying follows

a jump-diffusion process.



57.2



EVIDENCE FOR JUMPS



Let’s look at some data to see just how far from Normal the

returns really are. There are several ways to visualize the difference between two distributions, in our case the difference

between the empirical distribution and the Normal distribution. One way is to overlay the two probability distributions.

In Figure 57.1 we see the distribution of Xerox returns, from

1986 until 1997, normalized to unit standard deviation. The



Part Five advanced topics



0.7

PDF



928



0.6

0.5

0.4

0.3

0.2

0.1



0

−3 −2.6 −2.2 −1.8 −1.4 −1 −0.6 −0.2 0.2



0.6



1



1.4



1.8



2.2



2.6



3



Scaled returns



Figure 57.1 The probability density functions for Xerox daily returns, scaled to zero mean and unit

standard deviation, and the standardized Normal distribution.



peak of the real distribution is clearly higher than the Normal distribution. Because both of

these distributions have the same standard deviation then the higher peak must be balanced by

fatter tails, it’s just that they would be too small to see on this figure. They may be too small

to see here, but they are still very important.

This difference is typical of all markets, even typical of the changes in interest rates. The

empirical distribution diverges from the Normal distribution quite markedly. The peak being

much higher means that there is a greater likelihood of a small move than we would expect

from the lognormal random walk. More importantly, and concerning the subject of this chapter,

the tails are much fatter. There is a greater chance of a large rise or fall than the Normal

distribution predicts.

In Figure 57.2 is shown the cumulative distribution functions for Xerox daily returns, scaled

to zero mean and unit standard deviation, and that for the standardized Normal distribution.

And in Figure 57.3 is shown the difference between these two. If you look at the latter figure

you will see that there is more weight in the empirical distribution than the Normal distribution

from about two standards deviations away from the mean. If you couple this likelihood of an

extreme movement with the importance of an extreme movement, assuming perhaps that people

are hedged against small movements, you begin to get a very worrying scenario.

Another useful plot is the logarithm of the pdf for the actual distribution and the Normal

distribution. The latter is simply a parabola, but what is the former? Such a plot is shown in

Figure 57.4. The log of the pdf of the actual distribution looks more linear than quadratic in

the tails.

The final picture that I plot, in Figure 57.5, is called a Quantile-Quantile or Q-Q plot.

This is a common way of visualizing the difference between two distributions when you are

particularly interested in the tails of the distribution. This plot is made up as follows.



jump diffusion Chapter 57



Scaled returns



1.0



Xerox

Normal



0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1



−5



−4



−3



−2



−1



0.0

0

CDFs



1



2



3



4



5



Difference between CDFs



Figure 57.2 The cumulative distribution functions for Xerox daily returns, scaled to zero mean and

unit standard deviation, and that for the standardized Normal distribution.



−12



−7



−2



6%



4%



2%



0%

3

−2%



Scaled return



−4%



−6%



Figure 57.3 The difference between the cumulative distribution functions for Xerox daily returns,

scaled to zero mean and unit standard deviation, and the standardized Normal distribution.



929



Part Five advanced topics



Scaled returns

−4



0



−2



0



2



4



−1



Ln(Actual)

Ln(Normal)



−2



ln(PDF)



−3

−4

−5

−6

−7



Figure 57.4 Logarithms of the pdfs for Xerox daily returns and the standardized Normal distribution.

6

Quantile Xerox



930



4



2



0

−4



−3



−2



−1



0

−2



1



2



3



4



Quantile Normal



−4



−6



−8



Figure 57.5 Q-Q plot for Xerox daily returns and the standardized Normal distribution.



First, rank the empirical returns in order from smallest to largest, call these yi with an index

i going from 1 to n. Second, for the Normal distribution find the returns xi such that the

cumulative distribution function at xi has value i/n. Finally, plot each point (xi , yi ).

The better the fit between the two distributions, the closer the line is to straight. In the present

case the line is far from straight, due to the extra weight in the tails.



jump diffusion Chapter 57



Several theories have been put forward for the non Normality of the empirical distribution.

Three of these are

• Volatility is stochastic

• Returns are drawn from another distribution, a Pareto-Levy distribution, for example

• Assets can jump in value

There is truth in all of these. The first was the subject of Chapter 51. The second is a can

of worms; moving away from the Normal distribution means throwing away 99% of current

theory and is not done lightly. But some of the issues this raises have to be addressed in jump

diffusion (such as the impossibility of hedging). The third is the present subject.



57.3



POISSON PROCESSES



The basic building block for the random walks we have considered so far is continuous Brownian motion based on the Normally distributed increment. We can think of this as adding to

the return from one day to the next a Normally distributed random variable with variance proportional to time step. The extra building block we need for the jump-diffusion model for an

asset price is the Poisson process. We saw this process in another context in Chapter 40. A

Poisson process dq is defined by

dq =



0

1



with probability 1 − λ dt

with probability λ dt.



There is therefore a probability λ dt of a jump in q in the time step dt. The parameter λ is

called the intensity of the Poisson process. The scaling of the probability of a jump with the

size of the time step is important in making the resulting process ‘sensible,’ i.e. there being a

finite chance of a jump occurring in a finite time, with q not becoming infinite.

This Poisson process can be incorporated into a model for an asset in the following way:

dS = µS dt + σ S dX + (J − 1)S dq.



(57.1)



We assume that there is no correlation between the Brownian motion and the Poisson process.

If there is a jump (dq = 1) then S immediately goes to the value J S. We can model a sudden

10% fall in the asset price by J = 0.9.

We can generalize further by allowing J to also be a random quantity. We assume that it

is drawn from a distribution with probability density function P (J ), again independent of the

Brownian motion and Poisson process.

The random walk in log S follows from (57.1):

d(log S) = µ − 12 σ 2 dt + σ dX + (log J ) dq.

This is just a jump-diffusion version of Itˆo.

Figure 57.6 is a spreadsheet showing how to simulate the random walk for S. In this simple

example the stock jumps by 20% at random times given by a Poisson process.



931



932



Part Five advanced topics



1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34



A

Asset

Drift

Volatility

Intensity

Jump

Timestep



B



C



D

Time



E

F

G

H

Asset

=B1

0

100

0.01 101.5164

0.02 103.5782

0.03 105.9509

0.04 105.559

0.05 104.8723

0.06 104.9846

0.07 104.237

0.08 104.9858

=E5*(1+$B$2*$B$6+$B$3*SQRT($B$6)*(RAND()+RAND()+RAND()+RAND()+RAND()+R

0.09 106.7324

AND()+RAND()+RAND()+RAND()+RAND()+RAND()+RAND()-6)0.1 110.0078

$B$5*IF(RAND()<$B$4*$B$6,1,0))

0.11 111.1208

0.12 108.3814

0.13 106.6251

150

0.14 101.5317

0.15 79.8246

130

0.16 80.71573

0.17 82.47191

110

0.18 82.37806

0.19 81.69127

90

0.2 83.43508

0.21 83.62845

70

0.22 82.21449

0.23 82.34637

50

0.24 82.93521

0.25 84.00041

30

0.26 69.41815

0.27 69.12042

10

0.28 66.48793

0.29

65.40964

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

−10 0

0.3 66.8851

0.31 67.62094

0.32 66.87121

100

20%

20%

0.1

20%

0.01



Figure 57.6 Spreadsheet simulation of a jump-diffusion process.



57.4 HEDGING WHEN THERE ARE JUMPS

Now let us build up a theory of derivatives in the presence of jumps. Begin by holding a

portfolio of the option and − of the underlying:

= V (S, t) −



S.



The change in the value of this portfolio is

d



=



∂ 2V

∂V

+ 12 σ 2 S 2 2

∂t

∂S



dt +



+ (V (J S, t) − V (S, t) −

Again, this is a jump-diffusion version of Itˆo.



∂V



∂S



(µS dt + σ S dX)



(J − 1)S) dq.



(57.2)



jump diffusion Chapter 57



If there is no jump at time t so that dq = 0, then we could have chosen

= ∂V /∂S to

eliminate the risk. If there is a jump and dq = 1 then the portfolio changes in value by an O(1)

amount, that cannot be hedged away. In that case perhaps we should choose

to minimize

the variance of d . This presents us with a dilemma. We don’t know whether to hedge the

small(ish) diffusive changes in the underlying which are always present, or the large moves

which happen rarely. Let us pursue both of these possibilities.



57.5



HEDGING THE DIFFUSION



If we choose

=



∂V

∂S



we are following a Black–Scholes type of strategy, hedging the diffusive movements. The

change in the portfolio value is then

d



=



∂V

∂ 2V

+ 12 σ 2 S 2 2

∂t

∂S



dt + V (J S, t) − V (S, t) − (J − 1)S



∂V

∂S



dq.



The portfolio now evolves in a deterministic fashion, except that every so often there is a nondeterministic jump in its value. It can be argued (see Merton, 1976) that if the jump component

of the asset price process is uncorrelated with the market as a whole, then the risk in the

discontinuity should not be priced into the option. Diversifiable risk should not be rewarded.

In other words, we can take expectations of this expression and set that value equal to the riskfree return from the portfolio. This is not completely satisfactory, but is a common assumption

whenever there is a risk that cannot be fully hedged; default risk is another example of this. If

we take such an approach then we arrive at

∂V

∂V

∂V

∂ 2V

+ 12 σ 2 S 2 2 + rS

− rV + λE [V (J S, t) − V (S, t)] − λ SE [(J − 1)] = 0.

∂t

∂S

∂S

∂S

(57.3)

E[·] is the expectation taken over the jump size J , which can also be written

E[x] =



xP (J ) dJ,



where P (J ) is the probability density function for the jump size.

This is a pricing equation for an option when there are jumps in the underlying. The important

point to note about this equation that makes it different from others we have derived is its nonlocal nature. That is, the equation links together option values at distant S values, instead

of just containing local derivatives. Naturally, the value of an option here and now depends

on the prices to which it can instantaneously jump. When λ = 0 the equation reduces to the

Black–Scholes equation.

There is a simple solution of this equation in the special case that the logarithm of J is

Normally distributed. If the logarithm of J is Normally distributed with standard deviation σ

and if we write

k = E[J − 1]



933



934



Part Five advanced topics



then the price of a European non-path-dependent option can be written as



n=0



1 −λ (T −t)

(λ (T − t))n VBS (S, t; σ n , rn ).

e

n!



In the above

λ = λ(1 + k), σ n2 = σ 2 +



nσ 2

T −t



and rn = r − λk +



n log(1 + k)

,

T −t



and VBS is the Black–Scholes formula for the option value in the absence of jumps. This formula

can be interpreted as the sum of individual Black–Scholes values each of which assumes that

there have been n jumps, and they are weighted according to the probability that there will

have been n jumps before expiry.

If one does not make the assumption that jumps should not be priced in, then one has to

play around with concepts such as the market price of risk.



57.6 HEDGING THE JUMPS

In the above we hedged the diffusive element of the random walk for the underlying. Another

possibility is to hedge both the diffusion and jumps as much as we can. For example, we could

choose

to minimize the variance of the hedged portfolio, after all, this is ultimately what

hedging is about.

The change in the value of the portfolio with an arbitrary is, to leading order,

d



=



∂V



∂S



dS + (− (J − 1)S + V (J S, t) − V (S, t)) dq + · · · .



The variance in this change, which is a measure of the risk in the portfolio, is

var[d ] =



2



∂V



∂S



σ 2 S 2 dt + λE (− (J − 1)S + V (J S, t) − V (S, t))2 dt + · · · .

(57.4)



This is minimized by the choice

=



λE [(J − 1)(V (J S, t) − V (S, t))] + σ 2 S

λSE (J − 1)2 + σ 2 S



∂V

∂S .



(To see this, differentiate (57.4) with respect to

and set the resulting expression equal to

zero.)

If we value the options as a pure discounted real expectation under this best-hedge strategy

then we find that

∂V

∂V

∂ 2V

+ 12 σ 2 S 2 2 + S

∂t

∂s

∂S



µ−



σ2

(µ + λk − r) − rV

d



+ λE (V (J S, t) − V (S, t)) 1 −



J −1

(µ + λk − r)

d



= 0,



Tài liệu bạn tìm kiếm đã sẵn sàng tải về

5 Code: Cliquet with uncertain volatility, in similarity variables

Tải bản đầy đủ ngay(0 tr)

×