Financial Engineering

KU Leuven 2024

PART 1: EQUITY DERIVATIVES MODELLING

This part introduces and applies advanced models for the

pricing of equity derivatives.

Students should develop a solid understanding of the these

frameworks for pricing these instruments and the
mathematical and practical background necessary to apply
the various pricing methodologies to the market.

Financial Engineering [G0Q22a] 1

 OUTLINE PART 1

• Exotic Options

• Financial Models

• Fast Pricing of Vanilla Options under Advanced Equity Models

– Advanced pricing formulas
– Application, pros and cons
– Fast Fourier Transform (FFT) techniques versus direct integration
– Modelling stochastic volatility with the Heston and Bates model
– Other examples of advanced models for pricing (jump models)

Financial Engineering [G0Q22a] 2

OUTLINE PART 1

• Calibration on Option Surfaces

– Basics
– Objective functions
– Calibration algorithm

• Monte Carlo Simulation and Pricing of Exotics

– Basics of Monte-Carlo Simulation
– Sampling of Heston paths : Euler and Milstein scheme
– Pricing of exotics under Heston using Monte Carlo Simulation

• Model Risk and Calibration Risk

– Model risk
– Calibration Risk

• Bid and Ask Pricing

– Introduction to bid-ask pricing
– Distorted expectations and the concept of acceptability

• Conclusion
Financial Engineering [G0Q22a] 3
• We start with overviewing a few traditional exotic derivatives

– Modest path-dependent structures

• Asian options
• Cliquets
• …

– Heavily path-dependent structures

• Lookback
• Barriers
• …

Financial Engineering [G0Q22a] 4

• We start with some basic Structured Products.
• A Principal Proctected Note (PPN) for example may be suitable
for those seeking full protection of their original investment and for
investors who have long-term financial obligations.

• PPNs generally offer a return at maturity linked to an underlying

asset.

• Investors typically give up a portion of the potential gains in

exchange for principal protection.

• In its most basic form, a principal protected note typically consists

of a zero-coupon bond and an exotic derivative. At maturity, the
zero-coupon bond is redeemed at par, while the derivative offers
participation in an underlying reference asset.
Financial Engineering [G0Q22a] 5
• ZCB + “Fraction” of ATM Call
• Typical Principal Protection is 100 %
• Most of the time fraction = fixed percentage of participation < 1.
100.00%

80.00%

60.00% Stock
40.00%
PPN
20.00%
Protection 100% Participation 90%
0.00%
13
19
25
31
37
43
49
55
61
67
73
79
85
91
97
1
7

103
109
115
121
127
133
139
145
151
157
163
169
175
181
187
193
199
-20.00%
Price underlying
-40.00%

-60.00%

-80.00%

-100.00%

Financial Engineering [G0Q22a] 6

• EC is ATM
100 + A * max(ST – S0, 0)=
100 + A* S0 * max((ST – S0)/ S0, 0)

Depending on implied Participation rate

volatility
• Example :

– EC = 9; ZCB = 82; S0=40

– Participation rate = 2 * 40 /100 = 80 %
– Buy A = 2 ATM EC

Financial Engineering [G0Q22a] 7

• EC is ATM
100 + A * max(ST – S0, 0)=
100 + A* S0 * max((ST – S0)/ S0, 0)

Participation rate
• If EC is more expensive (high vol):

– EC = 18; ZCB = 82; S0=40

– Participation rate = 1 * 40 /100 = 40 % is NOT ATTRACTIVE
– Buy A = 1 ATM EC

Financial Engineering [G0Q22a] 8

• How to make note more attractive (higher particpation rate)
• SOLUTION 1: Principal is only protected to 90:
90 + A*max((ST – S0), 0)
• Example : Allocate less for ZCB
– EC = 18; ZCB = 74; S0=40 Increase participation rate
– A = 26/18 = 1.4444
– Participation rate = 1.4444 * 40 /100 = 57.78 %

• SOLUTION 2: Make payoff barrier dependent:

100 + ( A*max((ST – S0), 0) if barrier H has never been hit )
• Example : DOBC < EC
– DOBC = 12; ZCB = 82; S0=40
– A = 18/12 = 1.5
– Participation rate = 1.5 * 40 /100 = 60 %
Financial Engineering [G0Q22a] 9
• Some traditional Exotic option structures
– Barriers: Payoff is conditional on whether a low or high barrier is hit.
• Down-out-Barrier Call : DOBC = (ST-K)+ 1(min{St, , 0  t  T }  H)
• Down-in-Barrier Call : DIBC = (ST-K)+ 1(min{St, , 0  t  T }  H)
• Up-out-Barrier Call : UOBC = (ST-K)+ 1(max{St, , 0  t  T }  H)
• Up-in-Barrier Call : UIBC = (ST-K)+ 1(max{St, , 0  t  T }  H)
• Down-out-Barrier Put : DOBP = (K-ST)+ 1(min{St, , 0  t  T }  H)
• Down-in-Barrier Put : DIBP = (K-ST)+ 1(min{St, , 0  t  T }  H)
• Up-out-Barrier Put : UOBP = (K-ST)+ 1(max{St, , 0  t  T }  H)
• Up-in-Barrier Put : UIBP = (K-ST)+ 1(max{St, , 0  t  T }  H)
– Basic Relations:
• DOBC+DIBC = EC
• UOBC+UIBC = EC
• DOBP+DIBP = EP
• UOBP+UIBP = EP

Financial Engineering [G0Q22a] 10

• An investor buys a DIBC on the S&P 500. The call has six months to
expiration, is struck at the money and knocks in when the index falls
10% from its initial level.
• Assume that this is offered for 0.94%. With the index at 950, this puts
the strike at 950, the barrier at 855 and the premium is 8.93 index
points.

• The first path crosses the barrier at 855 in June. At expiration, the index
is at 876.05. Since the strike is at 950, the option expires worthless.
• The second path knocks in in August. At expiration, the index is at
1030.78, so the option pays out 80.78 = 1030.78 - 950.00 index points.
Financial Engineering [G0Q22a] 11
• An investor buys a UIBP on the S&P 500. The put has six months to
expiration, is struck at the money and knocks in when the index rises
5% from its initial level.
• The DIBP is offered for 2.60%. With the index at 950, this puts the strike
at 950, the barrier at 997.50 and the premium is 24.70 index points.

• The first path never reaches a level greater than 952.53, which is much
less than the knock in barrier at 997.50. Hence, the option expires
worthless even though the index at expiration is below the strike.
• The second path does knock in, crossing above the 977.50 level in mid
June. At expiration, the level of the index is 911.40, so the option pays
out 38.60 = 950.00 - 911.40 index points. Financial Engineering [G0Q22a] 12
 Lookbacks: Payoff is depending on the minimum or maximum of the
underlying over the lifetime:
• Lookback Call : LBC = ( max{St, , 0 <= t <=T } – K)+
• Lookback Put : LBC = ( K - min{St, , 0 <= t <=T })+
• Conditional Lookback Call : ( max{St, , 0 <= t <=T } – K)+ 1(ST>H)
• Conditional Lookback Put : ( K- min{St, , 0 <= t <=T } – K)+ 1(ST>H)

Example:
An investor purchases a six-year lookback on
the FTSE 100, with monthly observation.

At maturity, the investor gets 90% of the

highest performance of the FTSE over the
investment period, with full capital protection

Financial Engineering [G0Q22a] 13

• A Bonus Certificate (BC) pay out a bonus at maturity if certain
conditions are satisfied, e.g. the underlying stock has never hit a
low barrier during the life time of the certificate.
• It gives typically a reduction of the downside risk, a potential
higher bonus in case markets move sideways and an unlimited
upside potential.
Standard example: BC on FTSE-100, T=3y
At expiry a holder recives
A. the final index level if it is above 140% of the initial level, or
B. 140% of the initial level, if the final level is between 75% and
140% of the initial level, unless the index level has fallen below
75% of the initial level during the lifetime of the certificate in
which case…
C. one receives just the final level
Financial Engineering [G0Q22a] 14
• The Bonus Certifcate (BC) is a combination of a
– Zero strike European call,
– Barrier option: down-and-out barrier put (DOBP).

QUESTION : Is the payoff always better than a direct investment ?

ANSWER: No, a BC doesn’t pay dividends. These dividends are
used to finance the DOBP.
This makes high yield stocks/indices appealing in to be used in
BCs.
Financial Engineering [G0Q22a] 15
• Booster :
– Zero strike European call + Fraction of ATM call

200.00%

Participation 150%
150.00%

100.00% Booster

50.00%

Price underlying
0.00%
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
1
6

101
106
111
116
121
126
131
136
141
146
151
156
161
166
171
176
181
186
191
196
-50.00%

-100.00%
Strike

-150.00%

Financial Engineering [G0Q22a] 16

• Variation of a Booster with Barrier clausule
• Zero strike Call + fraction of ATM Call + DOBP
1
– If underlying always above barrier level: 0.8
0.6 Participation 150%
• geared exposure if final level is above initial level 0.4
• nomial if final level is below initial level 0.2
• Call works if in the money 0

11
21
31
41
51
61
71
81
91
1

101
111
121
131
141
151
161
171
181
191
201
-0.2
• Put works if out of the money -0.4
-0.6
-0.8 barrier
-1
200.00%
– If barrier level has been breached:
• geared exposure if final level is above initial level 150.00%
Participation 150%
• underlying if below initial level 100.00%

• Call works if in the money 50.00%

0.00%

11
21
31
41
51
61
71
81
91
1

101
111
121
131
141
151
161
171
181
191
Important : Note is non-principal -50.00%

protected; entire capital is at risk -100.00%

-150.00%
Financial Engineering [G0Q22a] 17
• Zero strike Call + fraction of ATM Call + 2 DOBP
100.00%
– If underlying always above barrier level: 80.00% Participation 150%
• geared exposure if final level is above initial level 60.00%
• positive exposuref to a fall if final below initial level 40.00%

• Call works if in the money 20.00%

• 2 Put works if out of the money 0.00%

11
21
31
41
51
61
71
81
91
1

101
111
121
131
141
151
161
171
181
191
-20.00%
-40.00%
-60.00%
– If barrier level has been breached: barrier
• geared exposure if final level is above initial level 200.00%

• underlying if below initial level 150.00%

• Call works if in the money 100.00%

Participation 150%
50.00%

Question: 0.00%

11
21
31
41
51
61
71
81
91
1

101
111
121
131
141
151
161
171
181
191
Compare with Power (same setting). -50.00%

Is Power-Barrier higher/lower than Twin- -100.00%

-150.00%
Twin-Barrier ?
Financial Engineering [G0Q22a] 18
• Zero strike Call + series of UIBP (with strikes=barrier).
• Investor receives at maturity the greater of the lock-in level if
achieved or the performance of underlying.

Question: Is this note capital protected ?

Financial Engineering [G0Q22a] 19
• Asian Call : Gives the buyer exposure to the upside in the
underlying stock above the strike, with settlement based on
an average stock price over some period.
AsianCall = (A-K)+, where A is the average of the stock
levels in the averaging period.
• Asian Put : Gives the buyer exposure to the downside in
the underlying stock below the strike, with settlement based
on an average stock price over some period.
AsianPut = (K-A)+, where A is the average of the stock
levels in the averaging period.
• The effect of averaging is to make the observed settlement
level less volatile. Many investors use Asian options in case
the market makes a large move on the day of expiration.
Financial Engineering [G0Q22a] 20
• Asian put on the S&P 500 with 6 months to expiration and struck 5% out of the money.
The strike is fixed on the trade date (not based on averaging) and the terminal level is
taken as the average closing price over the last ten trading days up to expiration.
• Assume this is offered for 4.25% of spot. With the index at 950, this puts the strike at
902.50 and the cost is 40.375 index points.

• The first path begins well below the strike at 888.48, and ends about 15 points higher at
902.79. Although the index closes above the strike, the average over the ten trading days
is 896.39, so the investor receives 6.11 (902.50 - 896.39) index points.
• The second path starts around 901.The average is 897.42, so the investor receives 5.08
index points.
• In the third case, the index begins the averaging period at 918.54.
The average is 907.35, so the investor receives nothing. If the option were ordinary (not
Asian), then the payoff would have been positive: 13.58 (902.50 - 888.92) index points.

Financial Engineering [G0Q22a] 21

• A VARIANCE SWAPS is a forward contract on annualized
variance with payoff at expiration equals to:

where σ²R is the realized variance (in annual terms) over a

period specified by the contract

and Kvar is the agreed fixed variance and N is the notional

amount.

Financial Engineering [G0Q22a] 22

The Kvar for a fair price of the variance swap is given by

where is the current forward price.

All this is related to the VIX calculation:

OTM C(K,T) or OTM P(K,T) Strike near FT

Financial Engineering [G0Q22a] 23
This expression comes from the fact that:

in combination with due to telescoping that

Hence the VS is essentially minus twice a log-contract.

Financial Engineering [G0Q22a] 24

Further the log-contract can be priced by the Breeden-
Litzenberger formula:

where

Note that for

Hence the appearance of the strike-square in the fair strike

formula.

Note that the B-L-formula provides a static hedge.

Financial Engineering [G0Q22a] 25
 Financial Models
About Mean and Variance
n

S i

• Mean is the arithmetic average: S  n * i1

• Variance is the deviation from that average

- S = Stock return for a given period t (in
years)
- S* = mean of stock returns
- n = number of observations

 
n
2
Si  S*
Variance of stock returns:
 
2
s
i 1
Sum of squared
n
deviations
from the mean
s
Annualised volatility:  s 
t
Financial Engineering [G0Q22a] 26
How stock prices move
Proportional return Expected return Stochastic Part.
over t Mean = 0, standard
deviation =  t
S
 t   t
S
Volatility Random number
from a normal distribution
with mean = 0 and standard
deviation = 1

Overall the proportional return of the

stock over a small interval of time
is normally distributed with mean µΔt
and standard deviation  t
Financial Engineering [G0Q22a] 27
 About Volatility
Volatility is the annualized standard deviation of asset returns (or s.d. of
log of asset price).
– Implied volatility:
The volatility that makes the output of an option pricing model equal
to the market value
– Historical volatility:
Annualized standard deviation of observed asset returns
Calculation period and observation frequency affect result; not simple
to measure properly.
QUESTION : If a stock has a volatility of 16% how much is it likely to move in
one day?
If the one year return is normally distributed the stock has 68% probability of moving by
16% or less up or down.
Over one day the standard deviation will be:

Use 256 trading days so daily vol. is 1% ; Really 252 trading days in a year, but 256 is a
good and easy approximation

Financial Engineering [G0Q22a] 28

About Volatility

Financial Engineering [G0Q22a] 29

Black and Scholes
Brownian Motion :
1. starts at zero
2. independent distributed increments
3. stationary distributed increments
4. normally distributed increments :

5. continuous sample paths

SDE :

Geometric BM :

Normal log-returns :

Financial Engineering [G0Q22a] 30

Vanilla Black-Scholes Pricing
OPTION PRICES EQUAL DISCOUNTED EXPECTED PAYOFFS

Probability that option

Delta
will finish in the money c = Call option price
r = Risk free interest rate
q = dividend yield
S = spot
K = strike
T = time to maturity
σ = volatility
N( ) = cumulative normal
probability distribution

Call Delta = N(d1) Put Delta = -N(-d1)

Financial Engineering [G0Q22a] 31
Fast Pricing of Vanillas

• For the Black-Scholes model we have a nice close-form pricing

formula for calls and puts:

• For more advanced models such a closed-form expression is not

readily available.

Financial Engineering [G0Q22a] 32

Fast Pricing of Vanillas

• For many advanced models expressions based on Fourier

transforms are available.

• The Fourier-Stieltjes Transform (or characteristic function) of a

distribution function :

• If the density function f of X is available, one can write :

Financial Engineering [G0Q22a] 33

Fast Pricing of Vanillas

• The characteristic function always exists, is continuous and it

DETERMINES THE DISTRIBUTION FUNCTION UNIQUELY.

• Out of the characteristic function, one can easily derive the

moments of X. If , then

Financial Engineering [G0Q22a] 34

Fast Pricing of Vanillas

• Examples :

Financial Engineering [G0Q22a] 35

Fast Pricing of Vanillas

• Examples :

Financial Engineering [G0Q22a] 36

Heston Stochastic Volatility Model
• Detailing the HESTON Stochastic Volatility Model.
– Stock price process (negatively)
correlated
Brownian
motions
– (squared) volatility process

Mean reverting
property of ”vol-of-vol”
volatility
Financial Engineering [G0Q22a] 37
Heston Stochastic Volatility Model
• Heston parameters:
– Spead of mean reversion: κ > 0
– Level of mean reversion: η > 0
– Vol-of vol : θ > 0 (BETTER: vol of var)
– Correlation vol-stock : -1 < ρ < 1
– Initial vol : v0 > 0
• Model dates back to 1993
• Heston is popular among practitioners: it incorporates
stochastic volatility in an tractable and intuitive way.
• Exotic option pricing can be done using MC; also PDE
methods are doable.
• Pure Heston does not include jumps.
Financial Engineering [G0Q22a] 38
Heston Stochastic Volatility Model

– The correlation ρ governs joint movements in the stock and its

variance.
– ρ is typically negative for equity: if stock drops, vol rises.
– Typically, the more negative ρ the steeper the skew.
– The price of an option C under Heston satisfies the PDE:

Vega Vega convexity Vanna

Additional Terms due to variance being stochastic

Financial Engineering [G0Q22a] 39

Heston Stochastic Volatility Model

– One says that the Feller condition is satisfied if :

– In that case, theoretically the process never hits 0.

– Doing MC simulation however, one must take care that, due to

discretization errors, negative variance is avoided (even under the
Feller condition)

Financial Engineering [G0Q22a] 40

 Equity Exotic Options
Heston Stochastic Volatility Model

– The fair strike Kvar for a variance swap under Heston equals.

– Note the fair variance swap strike is independent of vol-of-var and the
correlation between stock and variance.
– Further :

Financial Engineering [G0Q22a] 41

Characteristic Function HESTON:
– The characteristic function of Heston is available closed-form.
– Be aware of the Heston trap and the axis of evil !

where

Financial Engineering [G0Q22a] 42

BATES Model (Heston with Jumps)

Financial Engineering [G0Q22a] 43

BATES Model (Heston with Jumps)

Characteristic function:

Financial Engineering [G0Q22a] 44

BATES Model (Heston with Jumps)

Source: Numerix

Financial Engineering [G0Q22a] 45

VG model

BASIC IDEA: replace in BS the Normal distribution by a more flexible distribution

Characteristic function of VG law:

Financial Engineering [G0Q22a] 46

Fast Pricing of Vanillas

• Vanilla pricing under advanced models (Heston, Lévy, Jump

diffusions, Sato, Savy, …) can be done using Carr-Madan formula

Characteristic function of
the logarithm of the stock
price in the point (v-(α+1)i)
Financial Engineering [G0Q22a] 47
Fast Pricing of Vanillas

• The proof of the Carr-Madan formula is not so-difficult.

• We write k= log(K) for log-strike.
• Let q(x,t) be the density of the log stock price at maturity.
• We assume we have the characteristic function of the log stock
price at T available:

• We have :

Financial Engineering [G0Q22a] 48

Fast Pricing of Vanillas

• However, the call function C(k, T) converges to a non-zero

constant if (the zero strike call price). Hence this
function is not square integrable and Fourier theory would not
apply.

• To obtain a square integrable function, we consider the modified

call price for some α>0.

• Next we take the Fourier-Stieltjes transform of .

Financial Engineering [G0Q22a] 49

Fast Pricing of Vanillas

Financial Engineering [G0Q22a] 50

Fast Pricing of Vanillas

• The inverse Fourier transform recovers the original function out of

the Fourier transform:

Inverse Fourier Transform

Financial Engineering [G0Q22a] 51

Fast Pricing of Vanillas

• Characteristic function of the logarithm of the stock price is

available in closed form for many models.

• The integral is typically calculated using a Fast Fourier Transform.

• Using the FFT one can actually calculate the prices for a whole
range of strikes in one run.

• The algorithm chooses these strikes in a cleaver way.

• The price in your strike is obtained via interpolation (the grid of

strikes is quite dense).
Financial Engineering [G0Q22a] 52
Fast Pricing of Vanillas

Financial Engineering [G0Q22a] 53

Fast Pricing of Vanillas

• FFT is an efficient algorithm for computing the following transform

of a vector into a vector :

• Typically N is a power of 2.

• The number of calculations of the FFT is of order N log(N) and this

in contrast to the straightforward evaluation of the above sums,
which give rise to order N² of calculations.

Financial Engineering [G0Q22a] 54

Fast Pricing of Vanillas

• Let us apply the FFT for the calculations of the Call prices via the
Carr-Madan formula:

• Make a grid for the range of the integral :

0 η 2η 3η (N-2)η (N-1)η

N grid points (step size η) ignore this part (≈0)

Financial Engineering [G0Q22a] 55

Fast Pricing of Vanillas

• We are going to calculate the values for a whole range of strikes:

• This gives

• This is almost in the form of the FFT.

FFT :

Financial Engineering [G0Q22a] 56

Fast Pricing of Vanillas

• We just need to take to obtain :

• Hence the sum is just the FFT of the vector

FFT :

Financial Engineering [G0Q22a] 57

Fast Pricing of Vanillas

• There is a whole theory to choose optimally α>0

• Carr-Madan (1995) report that the following values give

satisfactory results :

• These values lead to :

Financial Engineering [G0Q22a] 58

Fast Pricing of Vanillas

• An more refined weighting (Simpson’s rule) for the integral in the

Carr-Madan formula on the N points-grid (0, η, 2η, 3η, ..., (N − 1)η)
leads to the following approximation:

where δj−1 = 1 if j = 1 and zero otherwise.

• This approximation gives a much more accurate integration.

Financial Engineering [G0Q22a] 59

Fast Pricing of Vanillas

• Examples : Black-Scholes setting

Financial Engineering [G0Q22a] 60

 Fast Pricing of Vanillas
function y =bsfftcf2(u,p,r,t,x)
sig=x;
y=exp(i.*u*(log(p)+r.*t-(1/2).*sig.^2.*t)).*exp(-(1/2).*sig.^2.*u.^2*t);
return

N=4096; alpha=1.5; eta=0.25;

p=100; strike=90; sig=0.2; r=0.03; q=0; t=1;
lambda=2*pi/N/eta;
b=lambda*N/2;
k=[-b:lambda:b-lambda];
KK=exp(k);
v=[0:eta:(N-1)*eta];
sw=(3+(-1).^(1:1:N));
sw(1)=1;
sw=sw/3;
rho=exp(-r*t)*bsfftcf2(v-(alpha+1)*i,p,r,t,sig)./(alpha^2+alpha-v.^2+i*(2*alpha+1)*v);
A=rho.*exp(i*v*b)*eta.*sw;
Z=real(fft(A));
CallPricesBS=exp(-alpha*k)/pi.*Z;
CallPriceBSFFT=spline(KK,CallPricesBS,strike);

Financial Engineering [G0Q22a] 61

 Calibration
Calibration

Financial Engineering [G0Q22a] 62

Basics of Calibration

- Calibration in a nutshell

- Finding minima and maxima of a function

- The problem with local minima and maxima

- Some basic search algorithms

- Examples

Financial Engineering [G0Q22a] 63

 Basics of Calibration
• Calibration in a nutshell : Pricing model Model
parameters

Derivatives Derivatives
Find:
market prices model parameters model prices
that match
model prices
as best as possible
Optimization problem: with market prices
Finding minimum “distance”

Financial Engineering [G0Q22a] 64

Basics of Calibration

• A function can have global and local maxima/minima

Financial Engineering [G0Q22a] 65

Basics of Calibration
in the neighbourhood of x∗

• A function f defined is said to have a local maximum point at the

point x∗, if there exists some ε > 0 such that
f(x∗) ≥ f(x) when |x − x∗| < ε.
• Similarly, f has a local minimum point at x∗, if there exists some
ε > 0 such that
f(x∗) ≤ f(x) when |x − x∗| < ε.
• A function has a global maximum point at x∗ if
f(x∗) ≥ f(x) for all x.
• Similarly, a function has a global minimum point at x∗ if
f(x∗) ≤ f(x) for all x.
• The global maximum and global minimum points are also known
as the arg max and arg min: the argument (input) at which the
maximum (respectively, minimum) occurs.
Financial Engineering [G0Q22a] 66
Basics of Calibration

• Finding global maxima and minima is the goal of optimization.

• One does this by systematically choosing the values from within

an allowed set.

• The first optimization technique, which is known as steepest

descent and goes back to Gauss.

• Gradient descent is a 1st order optimization algorithm.

Financial Engineering [G0Q22a] 67

Basics of Calibration

• To find a local minimum, one takes steps proportional to the

negative of the gradient of the function at the current point.

• If instead one takes steps proportional to the gradient, one

approaches a local maximum (gradient ascent).

• The gradient of f is defined to be the vector whose components

are the partial derivatives of f. That is:

• So for a univariate function f = f’ and can be approximated by

for h small
Financial Engineering [G0Q22a] 68
 Step size γ>0 and is
Basics of Calibration allowed to change in
each iteration
• We let the algorithm look for a local minumum.
• We start with an x1 and calculate x2 (which is closer to the
minimum) by the formula:
x2 = x1 - γ1 f(x1) , with γ1 small.

• Next, we reiterate this procedure :

x3 = x2 - γ2 f(x2) , with γ2 small.

• In general :
xn= xn-1 – γn-1 f(xn-1) , with γn-1 small.

• We stop if we don’t see much improvement anymore (say less

than a preset tolerance).
Financial Engineering [G0Q22a] 69
Basics of Calibration

• Let f(x)= x². We know f’(x)=2x and f has a minimum in zero.

• We start with an x1 and calculate x2 (which is closer to the

minimum).

• We start with x1 =1; take h=0.01 and γ1 =0.1.

• f(1)=1²=1.
• f’(1) ≈ (1+0.01)² - 1²) / 0.01 = 2.01.
• Note that f’(1)=2.
• x2 = 1 – 0.1 * 2.01 = 0.7990

Financial Engineering [G0Q22a] 70

 Basics of Calibration
• After 20 iterations (with a constant γ )
h 0,01
gamma 0,1

n x f(x) f'(x) true f'(x) approx

1 1,0000 1,0000 2,0000 2,0100
2 0,7990 0,6384 1,5980 1,6080
3 0,6382 0,4073 1,2764 1,2864
4 0,5096 0,2597 1,0191 1,0291
5 0,4066 0,1654 0,8133 0,8233
6 0,3243 0,1052 0,6486 0,6586
7 0,2585 0,0668 0,5169 0,5269
8 0,2058 0,0423 0,4115 0,4215
9 0,1636 0,0268 0,3272 0,3372
10 0,1299 0,0169 0,2598 0,2698
11 0,1029 0,0106 0,2058 0,2158
12 0,0813 0,0066 0,1627 0,1727
13 0,0641 0,0041 0,1281 0,1381
14 0,0503 0,0025 0,1005 0,1105
15 0,0392 0,0015 0,0784 0,0884
16 0,0304 0,0009 0,0607 0,0707
17 0,0233 0,0005 0,0466 0,0566
18 0,0176 0,0003 0,0353 0,0453
19 0,0131 0,0002 0,0262 0,0362
20 0,0095 0,0001 0,0190 0,0290 Financial Engineering [G0Q22a] 71
Basics of Calibration
First order
approximation
Why does this work, i.e. f(x1)> f(x2) ?

We have

f(x2) ≈ f(x1)+f’(x1)(x2 - x1) and x2 = x1 - γ f’(x1)

Hence

f(x2) - f(x1) ≈ f’(x1)(x2 - x1) = f’(x1)(-γ f’(x1)) = -γ (f’(x1))² < 0

Financial Engineering [G0Q22a] 72

Basics of Calibration

In practice, we do this in higher dimensions.

• Assume f has a bowl shape.
• The blue curves are the contour lines, that
is, the regions on which the value of f is
constant.
• A red arrow originating at a point shows
the direction of the negative gradient at
that point.
• Note that the (negative) gradient at a point
is orthogonal to the contour line going
through that point.
• We see that gradient descent leads us to
the bottom of the bowl, the minimum.
Financial Engineering [G0Q22a] 73
Basics of Calibration

• Recall:
x2= x1 – γ f’(x1), for some γ
• Rewrite:
x2 - x1 = -γ f’(x1).
• Let’s find γ such that f(x2), is as small as possible (we are looking
for a minimum).
• We use : f(x2) ≈ f(x1)+f’ (x1)(x2 - x1) + ½ f” (x1)(x2 - x1)² .
• This is a quadratic function in z=(x2 - x1): f(x1)+f’ (x1)z + ½ f” (x1)z² ,
which has a min or max if its derivative is zero: f’ (x1) + f” (x1)z =0
• Hence it reaches its extrema if: (x2 - x1) = z = - f’ (x1) / f” (x1).
• Combining gives γ=1/ f” (x1).
Make sure you go
• This is Newton’s method. downhill !
Financial Engineering [G0Q22a] 74
Basics of Calibration

• We try Newton out on our quadratic function f(x)= x².

• f’(x)=2x and f”(x)=2. Recall the minimum is in zero.
• Now γ =0.5.
• x2 = 1 – 0.5 * 2 = 0
• We find the minimum just after one iteration !
• One can prove that if f is quadratic the minimum is found after one
step.
• How does one calculate second derivative numerically:

Financial Engineering [G0Q22a] 75

 Equity Exotic Options
Basics of Calibration
• For f(x)=x4 ; f’(x)=4x3; f”(x)=12x2
Gradient Descent Newton’s Method
h 0,01
gamma 0,1

n x f(x) f' true f' approx n x f(x) f' true f' approx f" true f" approx
1 1,0000 1,0000 4,0000 4,0604 1 1,0000 1,0000 4,0000 4,0604 12 12,0002
2 0,59396 0,1245 0,8382 0,8596 2 0,66164 0,1916 1,1586 1,1851 5,253192 5,253392
3 0,50800 0,0666 0,5244 0,5401 3 0,43605 0,0362 0,3316 0,3432 2,281682 2,281882
4 0,45399 0,0425 0,3743 0,3868 4 0,28564 0,0067 0,0932 0,0982 0,97906 0,97926
5 0,41531 0,0298 0,2865 0,2971 5 0,18533 0,0012 0,0255 0,0276 0,412154 0,412354
6 0,38561 0,0221 0,2293 0,2384 6 0,11840 0,0002 0,0066 0,0075 0,168228 0,168428
7 0,36176 0,0171 0,1894 0,1974 7 0,07370 0,0000 0,0016 0,0020 0,065181 0,065381
8 0,34203 0,0137 0,1600 0,1672 8 0,04376 0,0000 0,0003 0,0005 0,022977 0,023177
9 0,32531 0,0112 0,1377 0,1442 9 0,02354 0,0000 0,0001 0,0001 0,006651 0,006851
10 0,31089 0,0093 0,1202 0,1261 10 0,00955 0,0000 0,0000 0,0000 0,001094 0,001294
11 0,29828 0,0079 0,1061 0,1116
12 0,28712 0,0068 0,0947 0,0997
13 0,27714 0,0059 0,0851 0,0899
14 0,26815 0,0052 0,0771 0,0816
15 0,26000 0,0046 0,0703 0,0745
16 0,25255 0,0041 0,0644 0,0684
17 0,24572 0,0036 0,0593 0,0631
18 0,23941 0,0033 0,0549 0,0584
19 0,23357 0,0030 0,0510 0,0543
20 0,22813 0,0027 0,0475 0,0507

Financial Engineering [G0Q22a] 76

Basics of Calibration

• Newton’s method takes curvature into account to

take a more direct route.
• Is therefore faster.
• Can lead to zig-zagging.
• Both methods cannot be started at points where
derivative is zero.
• If the initial value is too far from the true zero,
Newton's method may fail to converge.

Financial Engineering [G0Q22a] 77

Basics of Calibration

• The Nelder–Mead method or downhill simplex

method or amoeba method is a commonly used nonlinear
optimization technique.
• The Nelder–Mead technique was proposed by John Nelder &
Roger Mead (1965) and is a technique for minimizing an objective
function in a many-dimensional space.
• Nelder–Mead technique is only a heuristic, since it can converge to
non-stationary points.
• The method uses the concept of a simplex, which is a special
geometric object of N + 1 vertices in N dimensions. Examples of
simplices include a line segment on a line, a triangle on a plane,
a tetrahedron in 3D etc.

Financial Engineering [G0Q22a] 78

Basics of Calibration

• Many variations exist depending on the actual nature of the

problem being solved.
• Nelder–Mead generates a in each iteration a new test position by
extrapolating the behavior of the objective function measured at
each test point arranged as a simplex.
• The algorithm then chooses to replace one of these test points with
the new test point and so the technique progresses.
• It is based on reflection, expansion, contraction and shrink
operations of the simplex.

Financial Engineering [G0Q22a] 79

Basics of Calibration

STOPPING CRITIRIA : When does the algorithm needs to stop ?

• Criterion 1: reach the number of iteration specified by the user;

i>N

• Criterion 2: when the change of x value is smaller than a user

specified threshold;
| xi − xi-1 | < ε1

• Criterion 3: when the change of function value is smaller than a

user specified threshold;
| f(xi)− f(xi-1) | < ε2

Financial Engineering [G0Q22a] 80

Basics of Calibration

STARTING VALUES:
• Many of the optimization techniques depend (heavily) on the
starting values.

• Good intuition for the involved parameters is utmost important:

e.g.: volatility is a number around 20% for equity, not a number of
around 10 Bn.

• Many adhoc method exist to choose good starting values :

– Pick a whole grid of them and do a very quick optimization (just a few steps)
and then take best result.
– Take previous day results or parameters of similar instruments.
– Etc.
Financial Engineering [G0Q22a] 81
Basics of Calibration

CALIBRATION INSTRUMENTS:
• In financial application the instruments (derivatives) on which we
calibrate the model are the available (vanilla) instruments in the
market, for which we have fast model pricers available
• In equity world these are the European Calls and Puts available in
the market.
• There are here two possibilities :
– Calibrate on the prices themselves
– Calibrate on implied volatilities
• Sometimes other derivative (exotic) instruments are also traded,
however the pricing of them under the model may take too long…
(recall a calibration will try-out a huge amount of parameter
combinations).
Financial Engineering [G0Q22a] 82
Basics of Calibration

OBJECTIVE FUNCTIONS: Market

Price
- Root Mean Square Error (RMSE) :

- Average Percentage Error (APE) :

- Average Relative Price Error (ARPE) Model

Price
- Etc.

Financial Engineering [G0Q22a] 83

Basics of Calibration

WEIGHTINGS:

- Equal weighting.

- Weighting according to open interest.

- Weighting inverse to bid-ask spread: If the spread is great, we

have a wider range of prices that the model can imply. In other
words the model is allowed to imply a wider range of prices around
the mid-price. This means that less weight should be placed on
such a price, and vice-versa.

- implied volatilities as weights.

Financial Engineering [G0Q22a] 84
Calibration

Financial Engineering [G0Q22a] 85

Calibration

Financial Engineering [G0Q22a] 86

 Note : Better to calibrate on
OTM Call and Puts than
just on Calls
Calibration

Financial Engineering [G0Q22a] 87

 Monte Carlo Simulation
BASIC IDEA:

– Each simulation consists of a series of steps and is one possible

sequence of asset prices.
– No decisions can be made between start and end of time period
(will not work for American style options)
– Calculation time depends upon the product, the number of steps
and the number of simulations.
– Typical number of simulations: 10.000-100.000
– Therefore fairly slow to calculate.
– Calculating greeks or American styles by MC very slow.

Financial Engineering [G0Q22a] 88

•Simple Example
– Divide whole time interval (T) into a number of periods (n)
so Δt = T/n :

–Calculate asset values at the end of each period by picking

at random from a N(0,1) distribution and using:

–Find the value of the asset or derivative at T and discount it

back to today.

–Repeat many times, take the average and discount.

Financial Engineering [G0Q22a] 89

 Standard
Normal
•Simulation of a standard Brownian motion at the random
number
time points {nΔt , n=0,1,2, …}
•Euler Scheme:

Financial Engineering [G0Q22a] 90

• Simulation of a geometric Brownian motion at
the time points {nΔt , n=0,1,2, …}

Financial Engineering [G0Q22a] 91

• Brownian Bridge:
– important for path dependent options (barriers)
– S(T) generated first (largest variance)
– intermediate values dependent on past (smaller variance)

Conditional
Conditional
Mean standard
deviation

Financial Engineering [G0Q22a] 92

• Heston Stock :

• Variance:

• Euler Scheme:

Correlated Standard
Normal random
numbers

Financial Engineering [G0Q22a] 93

• Stock :

• Variance:

• Milstein Scheme:

Financial Engineering [G0Q22a] 94

 Correlated N(0,1) Numbers

• The Brownian motion driving the vol and the stock in

the Heston model are correlated.

• For the simulation we thus need correlated standard

Normals (correlation ρ).
Independent
Standard Normal
random numbers
Correlated
Standard Normal
random numbers

Financial Engineering [G0Q22a] 95

Avoiding negative variance

• because of the discritization (under Euler as well as

under Milstein scheme), the variance can become
negative under a simulation.

Possible Solutions (all wrong, but that’s life):

- absorption: if negative set equal to zero
- reflection: if negative, take absolute value
- trunctation: Partial trunctation

Full trunctation

Financial Engineering [G0Q22a] 96

 Model and Calibartion Risk
Model Risk

• We show that several advanced equity option models

incorporating stochastic volatility can be calibrated very nicely
to a realistic option surface.
• All these models are hence capable of accurately describing
the marginal distribution of stock prices or indices and hence
lead to almost identical European vanilla option prices.
• As such, we can hardly discriminate between the different
processes on the basis of their smile-conform pricing
characteristics.
• However, due to the different structure in path-behaviour
between these models, the exotics prices can vary significantly.

Financial Engineering [G0Q22a] 97

Model Risk
• 7 models :
– Heston Stochastic Volatility Model (HEST)
– Heston Stochastic Volatility Model with Jumps (HESJ)
– The Barndorff-Nielsen-Shephard Model (BNS)
– Lévy Jump Models with Stochastic Volatility
• VG with Integrated Gamma-OU volatility
• NIG with Integrated Gamma-OU volatility
• VG with Integrated CIR volatility
• NIG with Integrated CIR volatility

• We can include many more …

Financial Engineering [G0Q22a] 98

Other Models
• The other models in a nutshell:
– Heston with Jumps (Bates)
• Heston + jumps arriving at a constant intensity
• Jump sizes : log-normally distributed
• 3 new parameters : jump intensity, mean
jump size and variance of jump size

– Barndorff-Nielsen-Model
• Similar to Heston, but different variance process
• In total 5 parameters
• Vol process has only up-jumps and then decays

Financial Engineering [G0Q22a] 99

Other Models
• The other models in a nutshell:
– Lévy Models
• Generalizations of the Brownian motion for other return distributions
• More fat - tail behavior
• Turn out to be pure jump models
• NIG and VG have 3 parameters
• Used in combination with CIR or OU vol model
• CIR vol model is used in Heston
• Gamma-OU vol model is used in BNS
• Vol is incorporated via technique of time change at stochastic business time.

100
Financial Engineering [G0Q22a] 100
Perfect Calibration
• Calibration is seemingly almost perfect (see also next
part).

Financial Engineering [G0Q22a] 101

Perfect Calibration
• Calibration is seemingly almost perfect (see also next
part).

Market Model
Price Price Financial Engineering [G0Q22a] 102
Optimal Parameters
• Parameters seem to make more or less sense:

Financial Engineering [G0Q22a] 103

Exotic Option Pricing
• Next, we are tempted to price (via Monte Carlo) a battery
of exotic options.
– Barriers :

– Lookbacks :

– Cliquets :

• These exotics play a prominent role in Structured

Products and their pricing is key.

Financial Engineering [G0Q22a] 104

Exotic Option Pricing
• We see huge difference for the exotics over the models.
• Note the vanillas are priced almost exactly the same.

Financial Engineering [G0Q22a] 105

Exotic Option Pricing
• Lookback prices :

• Cliquets prices:

Principal Protection

Financial Engineering [G0Q22a] 106

Model Risk Conclusions

• We have look at several models all reflecting stochastic vol and non-
Gaussian returns, properties that are generally supported by empirical
research.
• Although all models are almost perfectly calibrated and hence vanillas
have the same prices under all models, exotic prices can differ
dramatically.
• Vanillas determine the marginal distributions not the process.
• The underlying fine-grain properties of the process have an important
impact on the path-dependent option prices.
• The impact of exotic price ranges is important for price setting of
Structured Products.

Financial Engineering [G0Q22a] 107

Calibration Risk

• Note there are also choices and associated risks in the

calibration exercise.
• Sometimes certain parameters can be fixed upfront on the
basis of historical or other market data, or estimated via
different statistical techniques.
• We can minimizing different objective functions.
• Again this can have an impact on exotic options.

Financial Engineering [G0Q22a] 108

Calibration Risk
• We perform our study under the HESTON model.

• We make use of historical data that tries to identify the

long run (historically implied) variance η > 0.
• Similarly, we try to identify using options on VIX the long
run (market implied) variance η > 0.
• We also make us of market data (VIX) to fix the initial
variance v0 > 0.
Financial Engineering [G0Q22a] 109
Calibration Risk
• Impact of different models on Exotic Prices

Financial Engineering [G0Q22a] 110

 Bid Ask Pricing
Bid-Ask Pricing
• We will make use of the minmaxvar distortion function:

• We use non linear expectation to calculate (bid and ask) prices.

• The distorted expectation of a random variable with distribution function F(x)
is defined

Financial Engineering [G0Q22a] 111

Conic Finance

• The ask price of payoff X is determined as

• The bid price of payoff X is determined as

• Hence for the BID price we have put more weight on the down-side. For the
ASK the upside has been receiving more weighted.

Financial Engineering [G0Q22a] 112

Conic Finance
• These formulas are derived by noting that the cash-flow of selling X at its ask
price and buying X at its bid price is acceptable in the relevant market :

• We say that a risk X is acceptable if

Actually we test whether for many test-measures our cash-flow has a positive
expectation.

• Operational cones were defined by Cherney and Madan and depend solely
on the distribution function G(x) of X and a distortion function Ψ. One can
show that we need to have that the distorted expectation is positive:

Financial Engineering [G0Q22a] 113

BID-ASK PRICING under Monte Carlo Simulation
MC:

Equally weighted

DISTORTED UNIFORM (1/N) WEIGHTS:

Distorted weights
BID PRICE:
Financial Engineering [G0Q22a] 114
How does everything relate to each other ?

• Model Risk : different models – different Q measures – different exotics

• Calibration Risk : same model – different parameters/Q – different exotics

• Acceptability :

• Selling X at its ask price is acceptable :

• Buying X at its bid price is acceptable :

• Ask price is supremum of test-measure prices :

• Bid price is infimum of test-measure prices:

Financial Engineering [G0Q22a] 115

 Conclusion Part I
• We have reviewed some exotic options and structured products.

• We have review some basic and more advanced models.

• We have discussed implementation details of FFT pricing.

• We have discussed calibration of a model on market data.

• We have reviewed some basics of Monte Carlo pricing.

• Model risk is omnipresent. Calibration risk within a model is also

significant. Impact on exotic prices can be quite severe.

• All the above is related to the conic finance theory.

Financial Engineering [G0Q22a] 116