Black-Scholes Option Pricing Using Three

Volatility Models: Moving Average, GARCH(1, 1),

and Adaptive GARCH

Bachelor Thesis

by
Heru Sataputera Na
260330
Informatics & Economics

Erasmus University
Rotterdam, Netherlands
August 2003

copyright © Heru Sataputera Na 2003

Acknowledgements
I thank God for He has given me wisdom, strength, and time to be able to write this
bachelor thesis.

I am also very grateful to my supervisor, Dr. Ir. Associate Professor Jan van den Berg, for
giving me the opportunity to evaluate this topic and especially for his time, inputs and
very helpful comments during the writing process.

Finally, I am really thankful for my family and friends who always supported and
encouraged me along the way.

2
Table of Contents

Abstract 4
1. Introduction 5
2. Option pricing and volatility models
2.1. Black-Scholes option pricing model 7
2.2. Volatility models
2.2.1. Moving Average model 9
2.2.2. GARCH (1,1) model 10
2.2.3. Adaptive GARCH 12
3. Experimental setup
3.1. Data description 13
3.2. Calculating option price 15
3.3. Performance Measure 16
4. Results 16
5. Discussion 18
6. Conclusion 21

References 23
Appendix A – Tables 24
Appendix B – Figures 31

3
Abstract

This article evaluates the empirical performance of three different models to estimate the
parameter volatility in the Black-Scholes option pricing model. The three models are: (a)
Moving Average model, (b) GARCH(1,1) model, and (c) Adaptive GARCH model. I
examine these models using two datasets: AEX index and ING GROEP CERTS asset.
Performance is measured by minimizing the error between the model-determined price
and the real market price. The Adaptive GARCH model and the Moving Average model
outperform the GARCH(1,1) model. This might be due to the ability of the Adaptive
GARCH model and the Moving Average model to update the information used in the
maximization process daily.

The Adaptive GARCH model outperforms the Moving Average model in the AEX data.
Conversely, the Moving Average model does better when using the ING GROEP CERTS
data. The underperformance of the Adaptive GARCH model via the ING data might be
because of the extra volatility of the ING asset compared to the AEX index.

4
1. Introduction
When Fischer Black and Myron Scholes developed the Black-Scholes model in
the early 1970’s [1], it soon became a major breakthrough. Since then many traders use
the Black-Scholes model as their premier model for pricing and hedging options. An
important property of the Black-Scholes model is that all variables in the equation are not
influenced by the risk preferences of investors. In particular, the analysis is based on a
risk-neutral pricing approach1, which in return simplifies the analysis of derivatives.

In the classic Black-Scholes model, the volatility is assumed to be constant.

However, empirical research shows that the volatility of financial asset prices is
following a stochastic process and varies through time. It means that while other
properties of an option- such as exercise price, time to maturity, current price of
underlying asset- can be observed directly from the market, the return volatility is the
uncertainty factor in the Black-Scholes model.

As volatility increases, the probability that stock price will raise or fall increases,
which in response will also increase the value of both call and put options. Return
volatility thus plays a major role in option pricing. Therefore, accurate measures and
good forecasts of volatility are critical for option pricing theories as well as trading
strategies. At present, there have been many models developed to determine volatility,
and some of them act as alternatives or improvement from earlier models. The family of
GARCH models is an example, starting from the autoregressive conditional
heteroskedasticity (ARCH) model of Engle (1982) [4].

This paper will analyze the performance of option pricing by using three different
models to estimate one-day-ahead volatility in the Black-Scholes model. The estimated
volatility of each model is used as an input in the Black-Scholes option pricing formula to
price 3-months-options daily during the lifetime of the options. The errors between the
model-determined prices and the real price will then be computed. Smaller errors will
denote a better performance. Call options from two underlying assets, namely the AEX

1
Explanation over the risk-neutral valuation is given in Hull [8], chapter 12.7.

5
index and the ING GROEP CERTS asset - one of the assets that construct the AEX
index-, are chosen for this purpose.

The first volatility model is the Moving Average model. In this model, daily
volatility is determined by taking the mean of historical returns of underlying asset within
a certain time horizon. The time horizon is set to be fixed. It is called the Moving
Average volatility model because the average-based determined volatility moves with
time. An assumption lying behind this model is that expected volatility today is given by
the average of the returns in the past period. Estimated volatility is updated daily by
recalculating the average of past returns within the same time horizon.

The second model is chosen from the family of GARCH models, the GARCH(1,1)
model. Although there are many more complicated GARCH models, GARCH(1,1) often
performs as well as others. A study of comparing volatility models done by Hansen and
Lunde [7], using GARCH(1,1) as benchmark, had as result that the best models do not
provide a significantly better forecast than the GARCH(1,1) model. For this reason
GARCH(1,1) is preferred here above all family of GARCH models. The volatility is
computed from the observations of historical daily asset prices, taking both the
conditional and unconditional variance into account in the estimation process.

An alternative to the GARCH(1,1) model is also proposed to improve the

performance of estimating volatility, which serves as our third model. Because the
parameters of the GARCH model are maximized by means of historical data and then
held constant afterward to forecast future volatility, it is possible that the information
contained in the parameters differs from what will occur in the future. To avoid this, the
parameters of the GARCH model are adjusted over time and updated daily through the
option’s lifetime. Using this approach, the parameters are re-maximized on daily basis
and expected to show an improvement in the performance of the model.

The remainder of the paper is organized as follow: Section 2 explains the Black-
Scholes model and the three models to forecast volatility. Section 3 describes the

6
experimental setup. Section 4 gives the results obtained from the experiment and Section
5 discusses the experiment’s outcomes. Finally, Section 6 concludes this thesis.

2. Option pricing and volatility models

2.1. Black-Scholes option pricing model
A basic assumption of the Black-Scholes model is that the stock price is log-
normally distributed. One of the attributes the lognormal distribution has is that stock
price can never fall by more than 100 percent, but there is some small chance that it could
raise by much more than 100 percent [3]. Several assumptions that lie behind the Black
and Scholes model are [1]:
a) Markets are efficient, which implies that people are unable to consistently predict
the direction of the market or an individual asset. Stock prices are supposed to
follow the continuous Itô process. To understand the Itô process, we have to know
what a Markov process is. A Markov process is “a process where the observation
at time t depends only on the previous observation.” An Itô process is simply a
Markov process in continuous time2.
b) Trading on securities is continuous and short selling is allowed.
c) No commissions are charged to buy and sell options (no transaction costs or
taxes). All securities are perfectly divisible.
d) The stock pays no dividends during option’s life. European exercise terms are
used; where the option can only be exercised on the expiration date.
e) Risk free arbitrage opportunities are not available.
f) The risk free interest rate (r) is known and remains constant for all maturities.

Option prices can be determined by a risk neutral pricing approach. This valuation
serves as the most important tool for the analysis of derivatives, because pricing a
derivative that provides a payoff at one exact time in the future simply becomes taking
the discounted value of the expected payoff from the option at its maturity. Using the
risk-neutral valuation, the expected return (µ) from the underlying asset and the discount
value is the risk-free interest rate (r). The Black-Scholes formulas for the prices of a

2
Further details over the Itô process are given in Hull [8], chapter 11.

7
European call option on a non-dividend paying stock and a European put option on a
non-dividend paying stock are given in the following equations [8]:
c = S o N (d1 ) − Ke − rT N (d 2 )
(1)
p = Ke − rT N (− d 2 ) − S o N (−d 1 )

The variables d1 and d2 are calculated by:

ln(S 0 / K ) + (r + σ 2 / 2)T
d1 =
σ T
(2)
ln( S 0 / K ) + (r − σ 2 / 2)T
d2 = = d1 − σ T
σ T

The variables c and p are the European call and European put price, So is the stock
price at time zero, K is the strike price of the option, r is the continuously compounded
risk-free rate, σ is the stock price standard deviation, and T is the time to maturity of the
option. N(x) is the cumulative probability distribution function for a standardized normal
distribution, or the probability that a variable with a standard normal distribution, with
zero mean and standard deviation is one, will be less than x.

In the above formulas, standard deviation per annum is used instead of daily
standard deviation as an input for the variable σ. Because we forecast the daily volatility,
standard deviation per annum is obtained using:
σ per .annum = σ daily × trading .days. per. year (4)

The normal assumption in equity markets is that there are 252 trading days per year [8].

Nevertheless, recent findings show that volatility tends to vary over time and thus
the assumption of constant volatility is unrealistic. Many models have been developed to
relax the constant volatility assumption. One of them is GARCH option model, which
assumes that the conditional volatility of stock prices depends on the past pricing errors.

8
 2.2. Volatility models
2.2.1. Moving Average model
The first model to estimate volatility3 is based on historical returns from stock
prices. Suppose that the value of the asset at the end of day i is Si. Define ui as the
percentage change of asset price between the end of day i-1 and the end of day i, so that:
S i − S i −1
ui = (5)
S i −1

We can also define ui as the continuously compounded return during day i (between the
end of day i-1 and the end of day i) by taking the logarithm of current asset value divided
by value of the day before:
 S 
u i = ln i  (6)
 S i −1 
The unbiased estimate of one-day volatility, using ui of m days before today, is
1 m
σ n2 = ∑ (u n−i − u ) 2
m − 1 i =1
(7)

In the equation above, ū is the mean of the ui’s:

1 m
u= ∑ u n −i
m i =1
(8)

A simplified approach to estimate volatility is given by [8]:

1 m 2
σ n2 = ∑ u n −i
m i =1
(9)

Variable ui is estimated using the equation (5). Two modifications takes place in this
equation compared to the equation (7): ū is assumed to be zero and m-1 is replaced by m.
The assumption of zero mean of ui could be taken if the actual mean does give a value of
zero or almost zero. We will use the equation (9) to estimate volatility for the Moving
Average model.

3
In this paper, we will use the term ‘volatility’ also for the variance rate.

9
 The amount of historical days used to estimate volatility should be determined
carefully. Some expect that more data would lead to better precision, but data that are too
old might be unrelated to predict the future. Closing prices from daily data over the last
90, 180 or even 252 days are often used. Another option is setting the amount of days,
variable m in the equation (9), to be equal to the number of days to which the volatility is
to be applied [8]. In another word, to value for instance a three month option, daily data
for the last three months are used. We will also use this method in this paper.

2.2.2. GARCH(1,1) model

Many models are developed that correspond to stochastic volatility process
characteristic. One widely known model is the ARCH model, introduced by Engle (1982)
[4]. This model is setting unconditional volatility constant, while allowing the conditional
volatility to change over time. This conditional volatility is subsequently altered by past
returns. The ARCH (q) model is specified as:
q
σ n2 = ω + ∑ α i u n2−i (10)
i =1

The variable ω is:

ω = γ .V L (11)

In this equation, ω > 0, and αi ≥ 0 hold. VL is the long-run average variance rate and γ is
the weight assigned to VL. Daily return (ui) is calculated using the equation (5). The
variable q is the order of dependency to past returns. This model differs with the Moving
Average model in that it assigns more weight to recent data, and also a weight is assigned
for the long-average variance rate (the unconditional volatility).

An assumption underlying this model is that volatility is changing over time and
there is tendency that a large error will be likely followed by a large error and a small
error followed by a small error. The variable q is the period the conditional variance
depends on. The larger the variable q, the longer is the period of volatility clustering. This
characteristic coincides with the findings of Fama [6] and Mandelbrot [9] that the
volatilities of financial series cluster.

10
 A generalized approach of ARCH models was proposed by Bollerslev (1986) [2],
and as well known as GARCH model. The GARCH (p, q) model is specified as:
q p
σ n2 = ω + ∑ α i u n2−i + ∑ β iσ n2−i (12)
i =1 i =1

In this equation, ω > 0, αi ≥ 0, and βi ≥ 0 hold.

As in ARCH models, variables p and q are the order of dependency. The

distinction of GARCH model is that the conditional variance is specified not only as a
linear function of past sample variances, but also including lagged conditional variances
to enter the equation as well. This corresponds to some sort of adaptive learning
mechanism.

The simplest GARCH model is GARCH(1,1) model, which is expressed as:

σ i2 = ω + α ⋅ u i2−1 + β ⋅ σ i2−1 (13)

Although there are many GARCH models developed, GARCH(1,1) often performs as
well as others4. GARCH(1,1) model is also the most popular among GARCH models.

The weights assigned to both conditional and unconditional volatility - γ, α, and β

- must sum to one. For a stable GARCH(1,1) process, α + β < 1 is required, otherwise the
weight applied to the long-term variance is negative. An interesting empirical finding is
that in financial series, particularly in daily series, α + β is often close to one. Engle and
Bollerslev (1986) [5] introduced the IGARCH (Integrated GARCH) model in which the
sum of α and β equals to 1.

The parameters in the equation of the models are estimated using the maximum
likelihood method. This method will choose the values of parameters by maximizing the
probability of a set of observations occurring. Define vi = σ2, which is the estimated
variance for day i. We assume that there are m observations, consisting of u1, u2… um and

4
For further details, see Hansen and Lunde [7].

11
that the probability distribution of ui is normal. The likelihood of the m observations is
[8]:
m  1  − u i2 
∏ 
π
exp  (14)
i =1 
 2 v i  2vi 
The best parameters are the ones that maximize this expression. Fortunately, maximizing
an expression and maximizing the logarithm of the expression is equivalent. Taking
logarithms of the expression in the equation (14), the log-likelihood is given by:
m  1 u i2 
∑ 
i =1 
c + 
2
− ln(v i ) − 
vi 

Ignoring constant multiplicative factors c and ½, we accordingly want to maximize:
m
 u i2 
∑ 

i =1 
− ln( v i ) −
vi

 (15)

To find the maximum, we differentiate this formula to variable vi and set the equation
result to zero. The maximum likelihood estimator of vi is:
1 m 2
vi = ∑ ui
m i =1

2.2.3. Adaptive GARCH model

The original GARCH(1,1) model estimates the parameters using maximum
likelihood estimation given historical data. These parameters are then held constant and
used to forecast future volatility by the equation (13) above. Hence, if historical returns of
the last three months, for example, have a different structure compared to the returns in
the next three months, then the maximized parameters might not be able to perform well
to estimate daily volatility for the next three months.

As an alternative, we can also define the parameters in GARCH as a function of

time. It means that the parameters at day i are estimated by means of maximum
likelihood just like the original GARCH(1,1) model, and subsequently these parameters
are used to forecast volatility at day i. The difference is the parameters’ maximization is
repeated daily during the lifetime of the options. The equation (13) is then adapted to:

12
 σ i2 = ω i + α i ⋅ u i2−1 + β i ⋅ σ i2−1 (17)

Parameters ω, α, β are changing over time. Assumption made is that the contributions of
VL, u2i-1 and σ2i-1 change over time and that they can be estimated over a fixed time
window. Because of daily maximization of the parameters, it is expected to obtain better
maximized parameters which will be used to estimate future volatility on a daily basis.
This model will be labeled ‘the Adaptive GARCH model’.

3. Experimental setup
3.1. Data description
Data used for implementation consists of daily close prices from AEX index and
ING GROEP CERTS from August 27, 2001 to February 15, 2002, which are collected
for the purpose of estimating volatility; while daily close prices of AEX index and ING
GROEP CERTS call options between November 19, 2001 and February 15 2002 are
obtained to determine the performance result.

We use this data based on the following consideration. First, the data is freely
available at Erasmus University Rotterdam, acquired from DataStream. Second, AEX
index is actively traded in the Euronext Amsterdam Exchange. AEX index is made up of
the 25 most active securities in the Netherlands. While many studies have used indexes
such as S&P 500 or Dow Jones as the experiment data because they are actively traded,
we therefore also choose an index, AEX, to test the performance of the models. Third, the
ING GROEP CERTS asset is one of the stocks included in the AEX index, which we
shall use as our second underlying asset. A change in the return of the ING asset will
have a fraction effect to the return of the AEX index. Hence, we are interested to see
whether the performance result of each model is still the same when we use a stock in
place of a market index. We choose for ING GROEP CERTS stock because the evolution
of the price is rather similar with the AEX price during the period we are interested in.

Figure 1A plots the movement of the AEX index levels in our sample, and Figure
1B plots the daily return for the period of 6 months. We can notice from Figure 1A that in

13
the first 30 days, the index prices fall substantially, and afterward rise to in the
neighborhood of € 500, and eventually vary within this region. The price of the index at
the end of the period is 498.1. From Figure 1B, we can see that the volatility of AEX
index changes over time. Meanwhile, Table 1 demonstrates that the returns have an
average of 0.0011.

The evolution of ING GROEP CERTS prices and the daily return are plotted in
Figure 2A respectively Figure 2B. One can notice that the prices also exhibit a decreasing
movement in the first 30 days, just like in the AEX index levels, except that these prices
quickly increase to subsequently vary in the region of € 28. The price of the ING asset at
the end of the period is 28.15. The daily return in Figure 2B shows that the volatility
change is higher over time than the AEX volatility. The mean of return sequences is
0.0008, as we can read from Table 1.

It is interesting to verify whether our implementation using the ING GROEP data
will give the same result as when using the AEX index data. Therefore these two data are
chosen to test the empirical performance of our models above. The options data have
several attributes. In the first place, the time to expiration is 3 months. Second, we
consider 5 options with different exercise prices for each underlying assets, whereas in-
the-money, near at-the-money, and out of-the-money options are all included. For AEX
index, we use options with exercise price 460, 480, 500, 520, and 540; while for ING
asset, options with exercise price 26, 28, 30, 32, 34 are selected. We would like to see
whether the performance of the option-pricing is constant on the options mentioned
above. In Table 1, we give the actual means of the returns during the lifetime of the
options. Because the mean values are always slightly above or below 0 and the average is
approximately zero, we can then safely assume that the mean is zero. It is possible
therefore to use the equation (9) above to estimate volatility.

We also obtain the daily yield rates of Netherlands CBS Government Bond that
matures within 9-10 years, during the life of the options. While there is no such thing as
risk-free interest rate in the real world, we assume that the yield rates of long-term

14
government bond would serve as a good representative to fill in this parameter5. The
interest rates of financial institutions are less appropriate because they are not risk-free
rates. The reason is that financial institutions bear the risk to default. Even the most
established institution bears this risk, though the amount is rather small. In contrast,
government is very unlikely to default. The daily rates are then converted to continuously
compounded rates.

3.2. Calculating option price

In computing the spot volatility on a given day, the daily prices of each asset
within 3-months prior to this day are used. Consequently, 6-months asset prices are
obtained for this purpose, starting from August 27, 2001 until February 15, 2002. For the
Moving Average model, we calculate the average of the last 3 months daily returns and
apply this average of past squared returns of underlying asset as the obtained volatility to
price an option. The average is updated daily and serves as the estimated volatility on the
given day. This process is done through the options’ lifetime.

For GARCH model, we use the daily returns of the first 3 months to estimate the
parameters of the GARCH(1,1) equation, and afterward by means of these parameters to
forecast volatility for the next 3 months. We make use of the maximum likelihood
method to estimate the parameters that maximize the equation (15) above. We choose the
Solver program in Microsoft Excel to implement the maximization process.

Eventually for the Adaptive GARCH model, we estimate parameters and

volatility each day through the lifetime of the options by means of assets’ daily returns of
the last 3 months. Analogous to the Moving Average model, this process is also carried
out on a daily basis starting from the first day the options come out to the market until the
options expire. The next step to be taken is calculating the option price by means of the
obtained volatility above. This would be a straightforward effort, since all inputs needed

5
We thank Mr. Rob Stevense for giving the advice that the long-term Netherlands Government Bond is
usually used as input for the parameter ‘risk-free interest rate’ in the Black-Scholes model.

15
for the parameters in the Black-Scholes formulas are already available within the data
acquired.

3.3. Performance Measure

The pricing error of the model for each option is expressed as:
η t = Ot − qt (18)
Parameter ηt denotes the pricing error for an option at time t, Ot is the real market price of
this option, and qt is the model-determined option price.

We implement this formula to all daily option prices calculated above. Because
the error could be positive or negative, we make use of the square amount of the pricing
errors. The Root Mean Squared Error (RMSE) is then the square root of the average of
squared pricing errors of options in the whole sample, and given by:
m

∑η i
2

RMSE = i =1
(19)
m
Variable m is the lifetime of the options in the sample, or 3 months.

The smaller the RMSE, the smaller pricing errors are, which subsequently means
better price forecasting. Another possible measure is MAE (Mean Absolute Error), which
stands for average of the absolute values of difference between the market option prices
and model determined prices. In contrast with RMSE, MAE returns the mean from
pricing errors that are given in absolute term. We will implement RMSE as our
performance measure in this paper.

4. Results
The parameters estimates of the GARCH(1,1) model appear in Table 2. Parameter
α is 0.0346, β is 0.7664, and γ is 0.1990 for AEX, while the parameters α, β, and γ of the
ING data are respectively 0, 0.8337, and 0.1663. Because the parameter α of ING is 0, it
means that the past returns do not provide any weight to estimate the volatility in this
model.

16
 The parameters of the Adaptive GARCH are given in Table 3. For AEX, the
weights α, β, and γ begin by 0.0382, 0.7653, and 0.1965, respectively; and end by 0,
0.9938, and 0.0062, correspondingly. For ING, the weights α, β, and γ start by 0, 0.8288,
and 0.1712, respectively; and finish by 0.0125, 0.9602, and 0.0273. The sum of α and β
in the first ten days for AEX and the first twenty-one days for ING asset are less than
0.900.

The evolution of the parameters is given in Figure 3A for the AEX index and
Figure 3B for the ING GROEP CERTS asset. The sum of α, β, and γ is always 1. From
Figure 3A we can see that the parameter α is going up first, and afterward decreasing to
zero value at the end of the period. The parameter β declines in the beginning of the
period and afterward rises gradually to almost reach value one, while the parameter γ
rises in the first week and subsequently diminishes to slightly below the start value.

Examining Figure 3B, we could notice a similar progress for the three parameters
of the ING asset. The difference lies in that the parameter α is beginning with zero value,
persisting in the first 10 days before rising and then decreasing to a value of 0.0125. As
for the parameter β, it declines further than AEX in the first two weeks of December
2001, and suddenly increases significantly at 18th of December 2001, followed by a
steady increase until the end of the period. The significant increase of the parameter β is
matched by a significant decrease of the parameter γ at the same day, after stepping
forward in the first month, and later on parameter γ evolves in the value range between
0.01 and 0.04, as we can read from Table 3. A worthy of note outcome is for the ING
asset, in the first ten days, the Adaptive GARCH model estimates the parameter α as 0,
which means that return from the preceding day during this period is not taken into
account when calculating the volatility.

The total squared differences between the real market prices and the model-
determined prices and the RMSE (root mean squared error) are reported in Table 4 for
the AEX index and Table 5 for the ING asset. For all the five options of the AEX index

17
considered, we could identify that the RMSE of the GARCH(1,1) model is larger than the
RMSE of the Moving Average model, while the Adaptive GARCH model produces the
smallest RMSE compared to the other two models. The RMSE of the Moving Average
model is also better than the RMSE of the GARCH(1,1) model for the options of the ING
asset; but we obtain a contrast result for the Adaptive GARCH model, where the RMSE
is no longer superior. The Moving Average model generates the smallest RMSE,
followed by the Adaptive GARCH model and the latest is GARCH(1,1) model.

5. Discussion
In the last chapter we have seen that the Adaptive GARCH model gives the best
estimation of the AEX option prices measured up to the Moving Average model, in the
second place, and the GARCH(1,1) model in the third place. We obtain similar result for
the GARCH(1,1) model using the ING options, creating the largest RMSE than the other
two models. A possible explanation of the underperformance of the GARCH(1,1) model
is that the historical returns used to maximize the parameters α, β, and γ differ a lot from
the volatilities throughout the lifetime of the options. These parameters do not act as the
appropriate weights for the equation of volatility estimates because we hold the
parameters constant during the options’ lifetime.

In the Moving Average model, volatility is calculated by the mean of the last 3-
months returns, and this calculation is updated daily. The oldest return is removed and
replaced with the newest return, where new information is taken into account in the
volatility estimates. The Adaptive GARCH model follows this process by removing the
earliest return with the latest return and uses this daily-updated-set of returns to re-
maximize its parameters. This process is also repeated daily. We have obtained contrast
results for the Moving Average model and the Adaptive GARCH model. The Adaptive
GARCH is superior to the Moving Average model in estimating the AEX option prices,
but the other way around in estimating the ING option prices.

There are three outcomes that might explain why the Adaptive GARCH model
does not perform well in estimating the volatility of the ING asset. First, an empirical

18
finding in financial series (as said before), particularly in daily series, α + β is often close
to one. For AEX index, this characteristic happens after the 11th day of the options’
lifetime, where the sum of the parameters α and β rises from 0.8254 to 0.9826 after
formerly generates value only in the range of 0.7663 and 0.8254. The sum then diverges
itself to a value close to one, between 0.9695 and 0.9938. In contrast, the sum of α and β
of the ING asset fluctuates in lower values than the sum of α and β of the AEX index.
The sum alters in the first 21 days between 0.6112 and 0.8288 along with a rising in the
21st day (at December 18, 2001) from 0.6359 to 0.9875 and varies close to one, between
0.9534 and 0.9886. This means that the duration of the sum of α + β not close to one is
longer in the volatility estimates of ING than AEX, which in turns might affect the
performance of the model.

Second, the parameter α of the AEX is estimated as zero for the last 3 days of the
period. On the contrary, the parameter α of the ING is estimated as zero for a longer
period, explicitly the first 10 days. The zero values of the parameter α mean that the
previous-day return is not taken into account when estimating volatility. This condition
might also affect the performance result of the model since the effect of this condition is
heavier for the ING option prices estimates because the period is longer.

Third, we can compare the estimated volatility of the three models with the
implied volatility (volatility acquired from option real prices). These data are given in
Figure 4A for the AEX index and Figure 4B for the ING asset. The implied volatility is
obtained by taking the average of the implied volatilities of the individual call options
each day throughout the lifetime of the options. We can notice from Figure 4A that the
estimated volatility of the GARCH(1,1) model is far from the implied volatility. The
volatility of the Moving Average model evolves progressively while the volatility of the
Adaptive GARCH model responds quickly after 3 weeks and then evolves near by the
implied volatility. In the last 2 weeks the implied volatility rises again and this is the only
period the volatility of the Moving Average model is closer to the implied volatility than
the volatility of the Adaptive GARCH model. Generally, during the 3-months period, the
forecasted volatility of the Adaptive GARCH is closer to the implied volatility than the

19
other two models. It clarifies why the Adaptive GARCH model provides the best
volatility estimates in the AEX index.

In Figure 4B, we can see that the estimated volatility of the GARCH(1,1) model is
also far from the implied volatility. The volatility of the Moving Average model is closer
to the implied volatility than the volatility of the Adaptive GARCH model in the first
month, but in the second month the volatility of the Adaptive GARCH model is closer. In
the third month, the first 20 days the volatilities of both models are more or less near by
the implied volatility. In the following 10 days, the implied volatility rises significantly to
a value of 0.0033 (above the volatility estimates of the GARCH(1,1) model – 0.0030),
and the volatility of the Moving Average is closer to the implied volatility matched up to
the volatility of the Adaptive GARCH.. The volatility estimates of the Moving Average
model is in total amount of days closer to the implied volatility than the volatility
estimates of the Adaptive GARCH model. This condition also clarifies why the Moving
Average model has a better (smaller) RMSE than the Adaptive GARCH model, though
the difference is small.

Identical to what happens in the parameter estimates, the volatility of the

Adaptive GARCH model is suddenly improved at 18th of December 2001. An
explanation of this outcome is that the returns of the first month (August 27th, 2001 to
September 27th, 2001) have left the basket of returns used to maximize the parameters.
These returns do not represent as the proper estimation of the volatility for the rest of the
period, because the movement is quite different with the rest. This leads to an
improvement in the maximized parameters which consecutively improves the estimated
volatility. In the AEX data, the returns of the first month have a similar shape as the
returns of the ING data, except the volatility of the AEX returns is less than the volatility
of the ING returns. As we can see from Figure 4A, the estimated volatility of the AEX
data also improves significantly in the first period, though the improvement is smoother.

20
6. Conclusion
The empirical performance of the Black-Scholes option pricing model on the
AEX options and the ING GROEP CERTS options, using three models to estimate
volatility, has been evaluated in this paper. The three models chosen to forecast volatility
are the Moving Average model, the GARCH(1,1) model and the Adaptive GARCH
model.

The Adaptive GARCH model outperforms the Moving Average model, in the
second place, and the GARCH(1,1) model while pricing the AEX options. We obtain a
different result when examining the ING option prices estimates. The GARCH(1,1) still
underperforms the other two models, but the Moving Average model is giving a better
result compared to the Adaptive GARCH model.

We have found from this experiment that the Adaptive GARCH model and the
Moving Average model have performed better than the GARCH(1,1) model to estimate
daily volatility during the lifetime of the 3-months option. The performance result of the
GARCH(1,1) model depends on the past returns used to maximize the parameters in the
equation of volatility estimates. The poor performance of the GARCH(1,1) model is
caused by the training data (in this case the first 3-months period of returns used to
maximize the parameters) having a different evolution compared to the test data (the next
3-months period – the options’ lifetime). Because we hold the parameter constant during
the volatility estimates, these parameters do not represent as the suitable weights for the
equation and therefore do not able to forecast volatility accurately.

Nevertheless, based on this experiment we are not able to conclude whether the
Adaptive GARCH model is better in estimating volatility than the Moving Average
model. We have indeed found that the underperformance of the Adaptive GARCH model
in estimating the volatility of the ING asset might be due to the extra volatility this asset
has contrast to the AEX index. This extra volatility can be identified by comparing the
returns of the AEX index and the ING asset in Figure 1B and Figure 2B.

21
 Consequently, several adjustments could be made to improve the performance of
the Adaptive GARCH model for the ING data. In the first instance, historical data with a
longer period could be used to maximize the parameters of the equation, for example 6
months, 1 year or even later. Due to possible shocks in a short-period history data that in
response might not represent the characteristic of the development of asset prices as a
whole, the use of a longer-term data might result in a better analysis. Other possible
adjustment is by implementing an alternative algorithm to calculate the maximum
likelihood function, for example the Levenberg-Marquardt algorithm. In this paper we
have used Solver algorithm in Microsoft Excel to compute the likelihood function.
Finally, it might be necessary to do this experiment on the other 24 assets that compose
the AEX index in order to be able to take a valid conclusion. We leave this for future
research.

22
References
1. Black, F., and Scholes, M. (1973): “The Pricing of Options and Corporate Liabilities,”
Journal of Political Economy, 81, 637-659.
2. Bollerslev, T. (1986): “Generalized Autoregressive Conditional Heteroscedasticity,”
Journal of Econometrics, 31, 307–327.
3. Brealey, R. A., and Myers, S. C. (2003). Principles of Corporate Finance. Seventh
Edition. Chapter 21.
4. Engle, R. F. (1982): “Autoregressive Conditional Heteroscedasticity with Estimates of
the Variance of UK Inflation,” Econometrica, 50, 987-1008.
5. Engle, R. F. and Bollerslev, T. (1986): “Modelling the Persistence of Conditional
Variances”, Econometrics Reviews, 5, 1-50.
6. Fama, E. (1965): “The behavior of stock market prices,” Journal of Business, 38, 34-
105.
7. Hansen, P.R., and Lunde, A. (2001): “A Forecast Comparison of Volatility Models:
Does Anything Beat a GARCH(1,1)?” Working Paper No. 01-04, Brown University,
Department of Economics.
8. Hull, J.C. (2003). Options, Futures, and Other Derivatives. Fifth Edition. Chapter 5,
11, 12, 17.
9. Mandelbrot, B. (1966): “Forecast of future prices, unbiased markets, and martingale
models,” Journal of Business, 39, 241-255.

23
 Appendix A – Tables
Table 1
Actual Means of Returns (ui) of the AEX Index and the ING GROEP CERTS Asset

The actual means of the returns of the last 3 months, calculated daily during the options’
lifetime are given below.

DATE AEX ING DATE AEX ING

19/11/2001 -0.0009 0.0008 7/1/2002 0.0016 0.0010
20/11/2001 -0.0004 0.0008 8/1/2002 0.0011 0.0010
21/11/2001 -0.0007 0.0008 9/1/2002 0.0007 0.0010
22/11/2001 -0.0006 0.0009 10/1/2002 0.0010 0.0010
23/11/2001 -0.0002 0.0009 11/1/2002 0.0016 0.0010
26/11/2001 0.0000 0.0009 14/1/2002 0.0018 0.0010
27/11/2001 -0.0001 0.0009 15/1/2002 0.0010 0.0010
28/11/2001 -0.0002 0.0009 16/1/2002 0.0016 0.0010
29/11/2001 0.0000 0.0009 17/1/2002 0.0017 0.0010
30/11/2001 0.0005 0.0009 18/1/2002 0.0016 0.0010
3/12/2001 0.0007 0.0009 21/1/2002 0.0009 0.0010
4/12/2001 0.0017 0.0009 22/1/2002 0.0010 0.0010
5/12/2001 0.0017 0.0009 23/1/2002 0.0013 0.0010
6/12/2001 0.0019 0.0009 24/1/2002 0.0008 0.0009
7/12/2001 0.0031 0.0009 25/1/2002 0.0015 0.0009
10/12/2001 0.0025 0.0009 28/1/2002 0.0019 0.0009
11/12/2001 0.0021 0.0009 29/1/2002 0.0017 0.0008
12/12/2001 0.0025 0.0009 30/1/2002 0.0014 0.0008
13/12/2001 0.0031 0.0009 31/1/2002 0.0010 0.0007
14/12/2001 0.0033 0.0009 1/2/2002 0.0007 0.0007
17/12/2001 0.0023 0.0009 4/2/2002 0.0008 0.0006
18/12/2001 0.0024 0.0009 5/2/2002 0.0003 0.0006
19/12/2001 0.0022 0.0009 6/2/2002 -0.0003 0.0005
20/12/2001 0.0022 0.0010 7/2/2002 -0.0002 0.0005
21/12/2001 0.0012 0.0010 8/2/2002 0.0004 0.0005
24/12/2001 0.0023 0.0010 11/2/2002 -0.0004 0.0004
27/12/2001 0.0020 0.0010 12/2/2002 -0.0001 0.0004
28/12/2001 0.0025 0.0010 13/2/2002 -0.0002 0.0003
2/1/2002 0.0018 0.0010 14/2/2002 -0.0003 0.0003
3/1/2002 0.0016 0.0010 15/2/2002 -0.0002 0.0003
4/1/2002 0.0018 0.0010 Average 0.0011 0.0008

24
 Table 2
Maximum Likelihood Estimation of the GARCH(1,1) model for the Whole Sample

The estimated parameters of the GARCH(1,1) model and the long-term variance rates of
the AEX Index and the ING GROEP CERTS asset are reported.

• AEX index
VL ω γ α β α+β
7.7646E-04 1.5450E-04 0.1990 0.0346 0.7664 0.8010

• ING GROEP CERTS asset

VL ω γ α β α+β
3.0308E-03 5.0404E-04 0.1663 0.0000 0.8337 0.8337

25
 Table 3
Maximum Likelihood Estimation of the Adaptive GARCH model during the
lifetime of the options

The estimated parameters of the Adaptive GARCH model and the long-term variance
rates of the AEX Index and the ING GROEP CERTS asset on each day during the
lifetime of the options are displayed.

• AEX Index
AEX VL ω γ α β α+β
19/11/2001 7.6572E-04 1.5047E-04 0.1965 0.0382 0.7653 0.8035
20/11/2001 7.5269E-04 1.4504E-04 0.1927 0.0436 0.7637 0.8073
21/11/2001 7.4367E-04 1.4357E-04 0.1931 0.0475 0.7595 0.8069
22/11/2001 7.3544E-04 1.5290E-04 0.2079 0.0487 0.7434 0.7921
23/11/2001 7.3044E-04 1.6132E-04 0.2209 0.0493 0.7298 0.7791
26/11/2001 7.2475E-04 1.6936E-04 0.2337 0.0510 0.7154 0.7663
27/11/2001 7.1768E-04 1.6351E-04 0.2278 0.0596 0.7126 0.7722
28/11/2001 7.1191E-04 1.4873E-04 0.2089 0.0701 0.7210 0.7911
29/11/2001 7.0048E-04 1.2809E-04 0.1829 0.0826 0.7345 0.8171
30/11/2001 7.0390E-04 1.2291E-04 0.1746 0.0949 0.7305 0.8254
3/12/2001 5.2335E-05 9.0877E-07 0.0174 0.1373 0.8453 0.9826
4/12/2001 2.6679E-05 4.0485E-07 0.0152 0.1556 0.8292 0.9848
5/12/2001 2.5461E-05 3.5066E-07 0.0138 0.1708 0.8155 0.9862
6/12/2001 1.5075E-04 4.3045E-06 0.0286 0.0727 0.8988 0.9714
7/12/2001 6.9137E-06 1.4938E-07 0.0216 0.0699 0.9085 0.9784
10/12/2001 6.9137E-06 1.4938E-07 0.0216 0.0699 0.9085 0.9784
11/12/2001 5.0340E-05 1.0642E-06 0.0211 0.0729 0.9059 0.9789
12/12/2001 3.8771E-05 9.7116E-07 0.0250 0.0650 0.9100 0.9750
13/12/2001 3.8771E-05 9.7116E-07 0.0250 0.0650 0.9100 0.9750
14/12/2001 3.0084E-05 6.7542E-07 0.0225 0.0371 0.9404 0.9775
17/12/2001 2.2280E-05 5.4494E-07 0.0245 0.0448 0.9308 0.9755
18/12/2001 1.6783E-04 5.1120E-06 0.0305 0.0394 0.9301 0.9695
19/12/2001 1.3368E-04 3.7917E-06 0.0284 0.0439 0.9277 0.9716
20/12/2001 1.3962E-05 2.8609E-07 0.0205 0.0350 0.9445 0.9795
21/12/2001 6.8300E-05 1.7017E-06 0.0249 0.0325 0.9425 0.9751
24/12/2001 1.1761E-04 2.7991E-06 0.0238 0.0278 0.9484 0.9762
27/12/2001 5.2499E-05 1.0728E-06 0.0204 0.0303 0.9493 0.9796
28/12/2001 1.8172E-05 2.6434E-07 0.0145 0.0181 0.9673 0.9855
2/1/2002 3.2457E-05 6.0595E-07 0.0187 0.0258 0.9555 0.9813
3/1/2002 6.2938E-06 1.0082E-07 0.0160 0.0251 0.9588 0.9840
4/1/2002 7.7324E-05 1.6322E-06 0.0211 0.0294 0.9495 0.9789
7/1/2002 3.9489E-05 7.5187E-07 0.0190 0.0277 0.9533 0.9810
8/1/2002 1.5848E-05 3.0416E-07 0.0192 0.0336 0.9472 0.9808
9/1/2002 1.3592E-05 2.6971E-07 0.0198 0.0433 0.9368 0.9802
10/1/2002 4.0607E-05 9.1357E-07 0.0225 0.0270 0.9505 0.9775

26
11/1/2002 3.6309E-05 8.0311E-07 0.0221 0.0296 0.9483 0.9779
14/1/2002 2.9931E-05 6.3371E-07 0.0212 0.0276 0.9512 0.9788
15/1/2002 1.1140E-04 2.9430E-06 0.0264 0.0236 0.9500 0.9736
16/1/2002 1.3165E-05 1.4317E-07 0.0109 0.0099 0.9792 0.9891
17/1/2002 1.3525E-05 2.3601E-07 0.0175 0.0234 0.9591 0.9825
18/1/2002 1.6733E-05 2.5775E-07 0.0154 0.0156 0.9690 0.9846
21/1/2002 9.9341E-06 1.7549E-07 0.0177 0.0238 0.9586 0.9823
22/1/2002 1.9035E-05 3.8018E-07 0.0200 0.0304 0.9497 0.9800
23/1/2002 6.6753E-06 1.3178E-07 0.0197 0.0256 0.9546 0.9803
24/1/2002 4.0142E-06 8.5878E-08 0.0214 0.0351 0.9435 0.9786
25/1/2002 3.6379E-05 5.7551E-07 0.0158 0.0119 0.9723 0.9842
28/1/2002 8.5827E-06 1.5381E-07 0.0179 0.0216 0.9605 0.9821
29/1/2002 3.4045E-05 6.9711E-07 0.0205 0.0245 0.9551 0.9795
30/1/2002 1.6203E-05 3.1182E-07 0.0192 0.0258 0.9550 0.9808
31/1/2002 2.9254E-05 5.2999E-07 0.0181 0.0169 0.9650 0.9819
1/2/2002 8.5301E-06 1.5142E-07 0.0178 0.0233 0.9590 0.9822
4/2/2002 3.0246E-05 6.1760E-07 0.0204 0.0252 0.9544 0.9796
5/2/2002 2.9284E-05 6.1327E-07 0.0209 0.0268 0.9523 0.9791
6/2/2002 5.6134E-05 9.8579E-07 0.0176 0.0107 0.9718 0.9824
7/2/2002 7.5533E-06 1.1075E-07 0.0147 0.0122 0.9731 0.9853
8/2/2002 3.0494E-05 5.3839E-07 0.0177 0.0123 0.9700 0.9823
11/2/2002 6.8832E-05 6.2688E-07 0.0091 0.0001 0.9908 0.9909
12/2/2002 7.3231E-05 6.5836E-07 0.0090 0.0001 0.9909 0.9910
13/2/2002 1.3375E-05 8.2965E-08 0.0062 0.0000 0.9938 0.9938
14/2/2002 3.3805E-05 2.1983E-07 0.0065 0.0000 0.9935 0.9935
15/2/2002 1.6514E-05 1.0314E-07 0.0062 0.0000 0.9938 0.9938

• ING GROEP CERTS

ING VL ω γ α β α+β
19/11/2001 2.9694E-03 5.0834E-04 0.1712 0.0000 0.8288 0.8288
20/11/2001 2.8944E-03 5.1306E-04 0.1773 0.0000 0.8227 0.8227
21/11/2001 2.8532E-03 5.0667E-04 0.1776 0.0000 0.8224 0.8224
22/11/2001 2.7828E-03 5.4961E-04 0.1975 0.0000 0.8025 0.8025
23/11/2001 2.7349E-03 5.8794E-04 0.2150 0.0000 0.7850 0.7850
26/11/2001 2.7501E-03 6.4055E-04 0.2329 0.0000 0.7671 0.7671
27/11/2001 2.7356E-03 6.7651E-04 0.2473 0.0000 0.7527 0.7527
28/11/2001 2.7389E-03 7.0324E-04 0.2568 0.0000 0.7432 0.7432
29/11/2001 2.7492E-03 7.0048E-04 0.2548 0.0000 0.7452 0.7452
30/11/2001 2.7375E-03 7.3457E-04 0.2683 0.0000 0.7317 0.7317
3/12/2001 2.4642E-03 5.4229E-04 0.2201 0.0135 0.7664 0.7799
4/12/2001 2.4213E-03 5.3604E-04 0.2214 0.0201 0.7585 0.7786
5/12/2001 2.4213E-03 5.3592E-04 0.2213 0.0201 0.7585 0.7787
6/12/2001 2.2784E-03 5.1043E-04 0.2240 0.0293 0.7467 0.7760
7/12/2001 2.2451E-03 5.9016E-04 0.2629 0.0377 0.6995 0.7371
10/12/2001 2.2300E-03 6.7669E-04 0.3035 0.0450 0.6516 0.6965
11/12/2001 2.2274E-03 8.0414E-04 0.3610 0.0518 0.5872 0.6390
12/12/2001 2.2113E-03 8.3735E-04 0.3787 0.0575 0.5638 0.6213

27
13/12/2001 2.2122E-03 8.6015E-04 0.3888 0.0592 0.5520 0.6112
14/12/2001 2.1658E-03 7.8191E-04 0.3610 0.0666 0.5724 0.6390
17/12/2001 2.1694E-03 7.8996E-04 0.3641 0.0757 0.5602 0.6359
18/12/2001 2.4978E-05 3.1262E-07 0.0125 0.1004 0.8870 0.9875
19/12/2001 8.1162E-06 9.2822E-08 0.0114 0.0996 0.8889 0.9886
20/12/2001 2.9026E-05 3.3778E-07 0.0116 0.1051 0.8833 0.9884
21/12/2001 3.6505E-04 6.0762E-06 0.0166 0.1139 0.8694 0.9834
24/12/2001 5.3296E-05 9.7397E-07 0.0183 0.0860 0.8957 0.9817
27/12/2001 3.7635E-05 7.0973E-07 0.0189 0.0893 0.8919 0.9811
28/12/2001 0.0000E+00 0.0000E+00 0.0153 0.0860 0.8986 0.9847
2/1/2002 4.5896E-05 9.8053E-07 0.0214 0.0734 0.9052 0.9786
3/1/2002 6.1976E-05 1.0685E-06 0.0172 0.0543 0.9285 0.9828
4/1/2002 1.9456E-04 5.2955E-06 0.0272 0.0520 0.9208 0.9728
7/1/2002 1.5544E-04 3.9693E-06 0.0255 0.0445 0.9300 0.9745
8/1/2002 4.0130E-04 1.0916E-05 0.0272 0.0390 0.9338 0.9728
9/1/2002 3.1170E-04 7.8780E-06 0.0253 0.0429 0.9319 0.9747
10/1/2002 1.9608E-04 4.6480E-06 0.0237 0.0440 0.9323 0.9763
11/1/2002 5.0940E-05 1.0758E-06 0.0211 0.0428 0.9360 0.9789
14/1/2002 1.9096E-04 4.2955E-06 0.0225 0.0395 0.9380 0.9775
15/1/2002 2.9618E-04 7.4029E-06 0.0250 0.0437 0.9313 0.9750
16/1/2002 6.0571E-04 2.0039E-05 0.0331 0.0415 0.9254 0.9669
17/1/2002 5.4920E-04 1.6532E-05 0.0301 0.0448 0.9251 0.9699
18/1/2002 3.7026E-04 9.4940E-06 0.0256 0.0422 0.9322 0.9744
21/1/2002 5.1295E-04 1.3085E-05 0.0255 0.0524 0.9221 0.9745
22/1/2002 4.2211E-04 1.0128E-05 0.0240 0.0514 0.9246 0.9760
23/1/2002 3.1262E-04 6.5893E-06 0.0211 0.0527 0.9262 0.9789
24/1/2002 3.1262E-04 6.5893E-06 0.0211 0.0527 0.9262 0.9789
25/1/2002 8.6753E-05 1.5001E-06 0.0173 0.0135 0.9692 0.9827
28/1/2002 3.5962E-04 1.3778E-05 0.0383 0.0193 0.9424 0.9617
29/1/2002 2.0949E-04 5.2335E-06 0.0250 0.0156 0.9594 0.9750
30/1/2002 7.2620E-05 1.0919E-06 0.0150 0.0129 0.9721 0.9850
31/1/2002 5.0917E-05 7.6162E-07 0.0150 0.0131 0.9719 0.9850
1/2/2002 5.0917E-05 7.6162E-07 0.0150 0.0131 0.9719 0.9850
4/2/2002 4.8315E-04 2.2533E-05 0.0466 0.0135 0.9399 0.9534
5/2/2002 4.4565E-04 2.0458E-05 0.0459 0.0173 0.9367 0.9541
6/2/2002 3.9532E-04 1.4888E-05 0.0377 0.0152 0.9472 0.9623
7/2/2002 4.8607E-05 6.5621E-07 0.0135 0.0100 0.9765 0.9865
8/2/2002 4.5094E-06 5.9178E-08 0.0131 0.0108 0.9761 0.9869
11/2/2002 3.7603E-04 6.8210E-06 0.0181 0.0068 0.9750 0.9819
12/2/2002 4.4337E-04 1.6935E-05 0.0382 0.0114 0.9504 0.9618
13/2/2002 3.8674E-04 1.2234E-05 0.0316 0.0114 0.9569 0.9684
14/2/2002 3.4448E-04 1.0072E-05 0.0292 0.0135 0.9573 0.9708
15/2/2002 3.2759E-04 8.9389E-06 0.0273 0.0125 0.9602 0.9727

28
 Table 4
Aggregate Squared Loss and Pricing Errors of the Black-Scholes Option Pricing for
the AEX Index

The sum of squared differences between the market option prices and the model-
determined prices for each option is given under ‘Total Squared Loss’. The RMSE (root
mean squared error) of the three models is also reported.

Exercise
Price Moving Average GARCH(1,1) Adaptive GARCH
Total Squared Total Squared Total Squared
Loss 1122.7235 Loss 2684.8085 Loss 1107.5599

460 RMSE 4.3257 RMSE 6.6893 RMSE 4.2964

Total Squared Total Squared Total Squared
Loss 2239.8875 Loss 5576.7700 Loss 1470.9104

480 RMSE 6.1099 RMSE 9.6409 RMSE 4.9513

Total Squared Total Squared Total Squared
Loss 3369.8176 Loss 8120.6981 Loss 1648.7049

500 RMSE 7.4942 RMSE 11.6338 RMSE 5.2420

Total Squared Total Squared Total Squared
Loss 5114.0224 Loss 8478.1884 Loss 2979.5121

520 RMSE 9.2322 RMSE 11.8871 RMSE 7.0469

Total Squared Total Squared Total Squared
Loss 3915.4690 Loss 6992.3757 Loss 1865.2816

540 RMSE 8.0782 RMSE 10.7954 RMSE 5.5757

29
 Table 5
Aggregate Squared Loss and Pricing Errors of the Black-Scholes Option Pricing for
the ING GROEP CERTS asset

The sum of squared differences between the market option prices and the model-
determined prices for each option is given under ‘Total Squared Loss’. The RMSE (root
mean squared error) of the three models is also reported.

Exercise
Price Moving Average GARCH(1,1) Adaptive GARCH
Total Squared Total Squared Total Squared
Loss 30.5696 Loss 141.0968 Loss 47.9347

26 RMSE 0.7138 RMSE 1.5335 RMSE 0.8938

Total Squared Total Squared Total Squared
Loss 54.8849 Loss 214.2465 Loss 78.3715

28 RMSE 0.9564 RMSE 1.8896 RMSE 1.1429

Total Squared Total Squared Total Squared
Loss 72.2765 Loss 243.4650 Loss 101.0783

30 RMSE 1.0975 RMSE 2.0144 RMSE 1.2979

Total Squared Total Squared Total Squared
Loss 73.9135 Loss 216.4026 Loss 107.7862

32 RMSE 1.1099 RMSE 1.8991 RMSE 1.3403

Total Squared Total Squared Total Squared
Loss 61.3883 Loss 166.3522 Loss 94.3840

34 RMSE 1.0115 RMSE 1.6651 RMSE 1.2542

30
 26 26
/8 /8
/ /

-0.16
-0.14
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
380
400
420
440
460
480
500
520
540
560
1 / 20 0 1/ 2 0 0
9/ 1 9/ 1
7 / 2 00
9 7/ 200
9
1 3 /2 0 1 13 /20 1
/9 0 1 /9 0 1
1 9 / 20 1 9 /2 0
/9 0 /9 0

February 15, 2002

February 15, 2002
2 5 / 20 1 2 5 /2 0 1
/9 0 /9 0
1 / / 20 1 1 / /2 0 1
10 0 10 0
7 / / 20 1
1 7 / /2 0 1
1 3 0 /2 0 1 1
/1 0 0 1 3 0 /2 0 1
19 0/2 1 /1 0 0
/1 0 1 9 0 /2 1
25 0/2 01 /1 0
/1 0 2 5 0 /2 0 1
31 0/2 01 /1 0
/1 0 0 3 1 0 /2 0 1
0 /1 0 0
6/ /20 1
11 0 0
6 / /2 0 1
1 2 /2 1
/1 0 1
1 2 1 /2 0 1
18 1/2 01 /1 0
/1 0 1 8 1 /2 0 1
24 1/2 01 /1 0
/1 0 2 4 1 /2 0 1
30 1/2 01 /1 0
/1 0 0
1 3 0 1 /2 0 1
6/ /20 1 /1 0 0
12 01 1
1 2 /2 6 / /2 0 1
/1 0 12 0
18 2/2 01 1 2 /2 1
Appendix B - Figures

/1 0 /1 0
24 2/2 01 1 8 2 /2 0 1
/1 0 /1 0
30 2/2 01 2 4 2 /2 0 1
/1 0 0 /1 0 0
2/ 1 3 0 2 /2 1
5/ 200 /1 0 0
1 2/ 1
1 1 /2 0 1 5/ 2 00
/1 0 2 1
1 7 / 20 11 /20 1
/1 0 /1 0 2
2 3 / 20 2 1 7 /2 0
/1 0 2 /1 0
2 9 / 20 2 3 /2 0 2
/1 02 /1 0
/ 2 9 /2 0 2
4 / 20 0
2/ 2 /1 0 2
10 2 0 /
/2 0 2 4/ 2 0 0
Figure 1A: This figure shows the daily AEX Index levels from August 27, 2001 to

/2 2/ 2
00 10 20
2 /2 0 2
/2
00
2

Figure 1B: This figure shows the daily return of the AEX Index from August 27, 2001 to

31
 26 26
/8 /8
/2 /

-0.16
-0.14
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
20
22
24
26
28
30
32
34
36
38

1/ 0 0 1 / 20 0
9/ 1 9/ 1
7/ 2 00 7 / 20 0
9 9 1
1 3 /2 0 1 13 / 20
/9 0 1 /9 0 1
19 /20 19 / 20
/9 0 /9 0
25 /20 1 25 / 20 1
/9 0 /9 0

to February 15, 2002

1/ /20 1 1/ / 20 1
10 0 10 0
7/ /20 1 7/ / 20 1
1 1
13 0/2 01 1 3 0/ 2 01
/1 0 0 /1 0 0
1 9 0 /2 1 1 9 0 /2 1
/1 0 /1 0
2 5 0 /2 0 1 2 5 0 /2 0 1
/1 0 / 1 00
3 1 0 /2 0 1 3 1 0 /2 1
/1 0 0 / 1 00
0 0 1

August 27, 2001 to February 15, 2002

6 / /2 0 1 6 / /2 0
1 11 0
12 1/2 01 12 /2 1
/1 0 /1 0
1 8 1 /2 0 1 1 8 1 /2 0 1
/1 0 /1 0
2 4 1 /2 0 1 2 4 1 /2 0 1
/1 0 /1 0
3 0 1 /2 0 1 3 0 1 /2 0 1
/1 0 0 / 1 00
1 1
6 / /2 0 1 6 / /2 0 1
12 0 12 0
12 /2 1 12 /2 1
/1 0 /1 0 0
1 8 2 /2 0 1 1 8 2 /2 1
/1 0 /1 0
2 4 2 /2 0 1 2 4 2 /2 0 1
/1 0 0 / 1 00
3 0 2 /2 1 3 0 2 /2 1
/1 0 0 / 1 00
2/ 1 2/ 1
5/ 2 0 0 5 / 2 00
1 1
1 1 /2 0 1 11 / 20 1
/1 0 2 /1 0 2
17 /20 17 / 20
/1 0 /1 0 2
23 /20 2 23 / 20
/1 0 /1 0 2
29 /20 2 29 / 20
/1 02 /1 0 2
/ /
4/ 20 0 4 / 20 0
2/ 2 2/ 2
10 2 0 10 20

Figure 2B: This figure shows the daily return of the ING GROEP CERTS asset from
/2 0 2 /2 0 2
/2 /2
00 00
2 2
Figure 2A: This figure shows the daily ING GROEP CERTS asset from August 27, 2001

32
 1

0.75

Alpha
0.5 Beta
Gamma

0.25

0
1

02

02

2
01

01

01

01

01

01

01
00

00

00

00

00

00
20

20
20

20

20

20

20

20

20
/2

/2

/2

/2

/2

/2
1/

2/
1/

1/

1/

2/

2/

2/

2/
12

/1

/1

/1

/1

/2
5/

4/
/1

/1

/1

/1

/1

/1

/1

11

17

23

29

10
6/
18

24

30

12

18

24

30

Figure 3A: This figure shows the evolution of the estimated parameters α, β, and γ for
the AEX Index during the options’ lifetime.

0.75

Alpha
0.5 Beta
Gamma

0.25

0
1

02

02

2
01

01

01

01

01

01

01
00

00

00

00

00

00
20

20
20

20

20

20

20

20

20
/2

/2

/2

/2

/2

/2
1/

2/
1/

1/

1/

2/

2/

2/

2/
12

/1

/1

/1

/1

/2
5/

4/
/1

/1

/1

/1

/1

/1

/1

11

17

23

29

10
6/
18

24

30

12

18

24

30

Figure 3B: This figure shows the evolution of the estimated parameters α, β, and γ for
the ING GROEP CERTS asset during the options’ lifetime.

33
 0.0008

0.0007

0.0006
Moving
Average
0.0005
GARCH
(1,1)
0.0004
Adaptive
0.0003 GARCH

Implied
0.0002

0.0001

0
1

02

02

2
01

01

01

01

01

01

01
00

00

00

00

00

00
20

20
20

20

20

20

20

20

20
/2

/2

/2

/2

/2

/2
1/

2/
1/

1/

1/

2/

2/

2/

2/
12

/1

/1

/1

/1

/2
5/

4/
/1

/1

/1

/1

/1

/1

/1

11

17

23

29

10
6/
18

24

30

12

18

24

30

Figure 4A: This figure shows the estimated volatility of the AEX Index obtained from
the three models and the implied volatility during the lifetime of the options.

0.0035

0.003

0.0025 Moving
Average

0.002 GARCH
(1,1)

0.0015 Adaptive
GARCH

0.001
Implied

0.0005

0
1

02

02

2
01

01

01

01

01

01

01
00

00

00

00

00

00
20

20
20

20

20

20

20

20

20
/2

/2

/2

/2

/2

/2
1/

2/
1/

1/

1/

2/

2/

2/

2/
12

/1

/1

/1

/1

/2
5/

4/
/1

/1

/1

/1

/1

/1

/1

11

17

23

29

10
6/
18

24

30

12

18

24

30

Figure 4B: This figure shows the estimated volatility of the ING GROEP CERTS asset
obtained from the three models and the implied volatility during the lifetime of the
options.

34