2011 3rd International Conference on Information and Financial Engineering

IPEDR vol.12 (2011) © (2011) IACSIT Press, Singapore

A Study on Heston-Nandi GARCH Option Pricing Model

Suk Joon Byun

KAIST Business School, Korea

Abstract. Heston and Nandi (2000) derive an almost closed form GARCH option pricing formula. This
paper considers an implementation of the Heston and Nandi (2000)’s option pricing model. We first estimate
Heston-Nandi’s GARCH parameters using a time series of S&P 500 historical daily index returns from
January 1981 to December 2010. Then we compare Heston and Nandi (2000)’s analytic formula with the
Monte-Carlo simulation results.
Keywords: GARCH, maximum likelihood estimation, Monte-Carlo simulation.

1. Introduction
GARCH (Generalized Auto-Regressive Conditional Heteroskedasticity) models have been used to model
time varying variances of asset prices and are well documented in finance literature. GARCH option pricing
models are developed by Duan (1995) and Heston and Nandi (2000). Heston and Nandi (2000) derive an
almost closed-form option pricing formula. Barone-Adesi, Engle, and Mancini (2008) propose a method for
pricing options based on GARCH models with filtered historical non-normal innovations. Byun and Min
(2010a) refine the GARCH option pricing model in Barone-Adesi, Engle, and Mancini (2008) by using the
theoretical results in Christoffersen, Elkamhi, Feunou, and Jacobs (2010) and letting physical and risk-
neutral one-day ahead GARCH volatilities to be different. An empirical application of Byun and Min (2010a)
shows that by doing so GARCH option fitting improves significantly. Byun and Min (2010b) compare the
empirical performances of several GARCH option pricing models with non-normal innovations using
extensive data on S&P 500 index options.
This paper considers an implementation of the Heston and Nandi (2000)’s GARCH option pricing model.
We first estimate Heston-Nandi’s GARCH parameters using a time series of S&P 500 historical daily index
returns from January 1981 to December 2010 (7,570 daily returns). The parameter estimates are obtained
using maximum likelihood estimation (MLE) procedure. Then we compare Heston and Nandi (2000)’s
analytic formula with the Monte-Carlo simulation results.

2. Heston-Nandi (2000) Model

2.1. Assumptions
The asset dynamics under the physical measure are given by

ln

is the underlying asset price at time . is the log-return of the asset price. is the continuously
compounded risk-free rate. is the standard normal random variable. is the conditional variance of the
log-return between 1 and and is known at time 1. is the equity risk premium. Figure 1 shows
467
the daily logarithmic return on the S&P 500 from January 1981 to December 2010. The dynamics of the
variance is the GARCH(1,1), that is,

0.15

0.1

0.05

-0.05

-0.1

-0.15

-0.2

-0.25
0 1000 2000 3000 4000 5000 6000 7000 8000

Fig. 1: Daily S&P 500 returns (1981-2010)

Heston and Nandi (2000) also assume that the value of a call option with one period to expiration obeys
the Black-Scholes-Rubinstein formula. This assumption is equivalent to Duan’s (1995) locally risk neutral
valuation relationship (LRNVR) assumption. They use the discrete time option pricing framework of
Rubinstein (1976) and Brennan (1979). Rubinstein (1976) and Brennan (1979) developed a discrete time
option pricing framework by making joint conditions on the distributions as well as on the individual’s
preferences.

2.2. The Likelihood Function

The model has five parameters , , , , . We estimate the model parameters using a time series of
S&P 500 historical daily index returns from January 1981 to December 2010 (7,570 daily returns). The
parameter estimates are obtained using maximum likelihood estimation (MLE) procedure. The likelihood
function is
1
exp
2 2
The log-likelihood function is

log 0.5 log 2

2.3. Parameter estimates

Table I shows the maximum likelihood estimates of the Heston-Nandi model under the physical measure.
We estimate the model parameters using a time series of S&P 500 historical daily index returns from January
1981 to December 2010 (7,570 daily returns). We fix the interest rate of 4% and dividend rate of 1.5%. Table
I also reports the parameter estimates from Heston and Nandi (2000) and Christoffersen, Jacobs, and
Ornthanalai (2009). Heston and Nandi (2000) report their parameter estimates using S&P 500 daily index
returns from 1992-1994. Christoffersen, Jacobs, and Ornthanalai (2009) report their parameter estimates
using S&P 500 daily index returns from 1985-2004. The last column in Table I show the parameter estimates
with the constraint of 0.

468
 Table I Maximum likelihood estimates under the physical measure
1992-1994 1985-2004 1981-2010 1981-2010
HN(2000) CJO(2009) (w=0)
Parameters
0.205 2.015 1.7172 1.7686
w 5.02e-06 -9.038e-07 -1.3277e-06 0
b 0.589 0.9032 0.9138 0.8733
a 1.32e-06 4.439e-06 4.4718e-06 4.3859e-06
c 421.39 119.2 114.196 140.5724

Log-Likelihood 3,503.7 16,449 24,503 24,482

3. Option Pricing in Heston-Nandi Model

3.1. Monte-Carlo simulation
The risk-neutral process of the Heston-Nandi (2000) model can be written as
1
2

is the standard normal random variable in the risk-neutral world and 0.5. The price at
time of a European call option with strike price that expires at time is given by:
max ,0
is the expectation under the risk-neutral distribution. We sample paths to obtain the expected payoff
in a risk neutral world and then discount this payoff at the risk-free rate.

3.2. Analytic valuation

GARCH option pricing models are developed by Duan (1995) and Heston and Nandi (2000). Duan’s
(1995) model is typically solved by Monte-Carlo simulation which can be slow and computationally
intensive for empirical work. In contrast, Heston and Nandi (2000) derive an almost closed-form option
pricing formula. They derive the following option pricing formula for the European call option with strike
price that expires at time :
1 1 1 1
2 2
In this formula, denotes the real part of a complex number. is the conditional
characteristic function of the log asset price using the risk neutral probabilities. is the imaginary number,
√ 1. The put option price can be obtained by put-call parity. Heston and Nandi (2000) show that the
conditional generating function of the asset price under the physical measure takes the following log-linear
form for the GARCH(1,1) model. This is also the moment generating function of the log asset price under
the physical measure:
exp
Heston and Nandi (2000) also derive the recursion formulas for the coefficients and . The two
coefficients can be calculated recursively, by working backward from the maturity date of the option and
using the terminal conditions:
1
log 1 2
2
1
1 2
2 1 2
0

3.3. An Example

469
 This section compares Heston and Nandi (2000)’s analytic formula and Monte-Carlo simulation results.
We consider a European call option with 100 days to expiration and the following parameter values;
$100, $100, 1 0.15 0.15 /252, 0.04/252, 0.015/252. The GARCH
parameter values are from the last column in Table I: 1.7686, 0, 0.8733, 4.3859e 06,
140.5724. The analytic formula gives the call option value of $4.602. Figure 2 shows 95% confidence
intervals for the Monte-Carlo option prices. We increase the number of simulations from 10,000 to 100,000
in Monte-Carlo simulations. The straight line in Figure 2 represents the option price from Heston and Nandi
(2000)’s analytic formula.

4.75

4.7

4.65

4.6

4.55

4.5

4.45
1 2 3 4 5 6 7 8 9 10
4
x 10

Fig. 2: 95% confidence intervals of Monte-Carlo simulations

4. Conclusion
GARCH (Generalized Auto-Regressive Conditional Heteroskedasticity) models have been used to model
time varying variances of asset prices and are well documented in finance literature. Heston and Nandi (2000)
derive an almost closed-form GARCH option pricing formula. This paper first estimates Heston-Nandi’s
GARCH parameters using a time series of S&P 500 historical daily index returns from January 1981 to
December 2010 (7,570 daily returns). The parameter estimates are obtained using maximum likelihood
estimation (MLE) procedure. Then we compare Heston and Nandi (2000)’s analytic formula with the Monte-
Carlo simulation results.

5. References
[1] G. Barone-Adesi, R. Engle, and L. Mancini. A GARCH option pricing model with filtered historical simulation.
Review of Financial Studies. 2008, 21: 1223-1258.
[2] M. Brennan. The pricing of contingent claims in discrete time models. Journal of Finance. 1979, 34: 53-68.
[3] S. Byun and B. Min. Conditional volatility and the GARCH option pricing model with non-normal innovations.
Working paper. KAIST. 2010a.
[4] S. Byun and B. Min. An empirical comparison of GARCH option pricing models with non-normal innovations.
Working paper. KAIST. 2010b.
[5] P. Christoffersen, R. Elkamhi, B. Feunou, and K. Jacobs. Option valuation with conditional heteroskedasticity and
nonnormality. Review of Financial Studies. 2010, 23: 2139-2183.
[6] P. Christoffersen, K. Jacobs, and C. Ornthanalai. Exploring time-varying jump intensities: Evidence from S&P
500 returns and options. Working Paper. 2009.
[7] J. Duan. The GARCH option pricing model. Mathematical Finance. 1995, 5: 13-32.

470
[8] S. Heston and S. Nandi. A closed-form GARCH option pricing model. Review of Financial Studies. 2000, 13: 585-
626.
[9] M. Rubinstein. The valuation of uncertain income streams and the pricing of options. Bell Journal of Economics
and Management Science. 1976, 7: 407-425.

471