Models for S&P500 Dynamics: Evidence from Realized

Volatility, Daily Returns, and Option Prices

Peter Christoersen Kris Jacobs Karim Mimouni
Desautels Faculty of Management, McGill University

December 16, 2008

Abstract
Most recent empirical option valuation studies build on the ane square root (SQR) sto-
chastic volatility model. The SQR model is a convenient choice, because it yields closed-form
solutions for option prices. However, relatively little is known about the resulting biases.
We investigate alternatives to the SQR model, by comparing its empirical performance with
that of ve dierent but equally parsimonious stochastic volatility models. We provide em-
pirical evidence from three dierent sources: realized volatilities, S&P500 returns, and an
extensive panel of option data. The three sources of data we employ all point to the same
conclusion: the SQR model is severely misspecied. The best of the alternative volatility
specications is a model with linear rather than square root diusion for variance, which
we refer to as the VAR model. This model captures the stylized facts in realized volatilities,
it performs well in tting various samples of index returns, and it has the lowest option
implied volatility mean squared error in- and out-of-sample. It ts the option data better
than the SQR model in several dimensions: it improves the t of at-the-money options, it
provides a more realistic volatility term structure and implied volatility smirk.
JEL Classication: G12
Keywords: Stochastic volatility; option valuation; particle ltering; skewness; kurtosis;
mean reversion.

Christoersen and Jacobs are also aliated with CIRANO and CIREQ and want to thank FQRSC, IFM
2
and SSHRC for nancial support. Mimouni was supported by IFM
2
and FQRSC. We are grateful for helpful
comments from Yacine At-Sahalia (the Editor), Torben Andersen, Mikhail Chernov, Jerome Detemple, Jin Duan,
Bjorn Eraker, Bob Goldstein, Jeremey Graveline, Mark Kamstra, Nour Meddahi, Marcel Rindisbacher, Stephen
Taylor, and two anonymous referees. Correspondence to: Peter Christoersen, Desautels Faculty of Management,
McGill University, 1001 Sherbrooke Street West, Montreal, Quebec, Canada, H3A 1G5; Tel: (514) 398-2869; Fax:
(514) 398-3876; E-mail: peter.christoersen@mcgill.ca.
1
1 Introduction
Following the nding that Black-Scholes (1973) model prices systematically dier from market
prices, the literature on option valuation has formulated a number of theoretical models designed
to capture these empirical biases. One particularly popular modeling approach has attempted to
correct the Black-Scholes biases by modifying the assumption that volatility is constant across
maturity and moneyness. Estimates from returns data and options data indicate that return
volatility is time-varying, and modeling volatility clustering leads to signicant improvements in
the performance of option pricing models. It has also been demonstrated that it is necessary to
model a leverage eect. The leverage eect captures the negative correlation between returns
and volatility, and thus generates negative skewness in the distribution of the underlying asset
return.
1
The existing literature has almost exclusively modeled volatility clustering and the leverage
eect within an ane or square root structure. In particular, the Heston (1993) model, which
accounts for time-varying volatility and a leverage eect, has been implemented in a large number
of empirical studies. Henceforth, we will refer to this model as the SQR model. The SQR model
is convenient because it leads to (quasi) closed-form solutions for prices of European options. It is
not surprising that the choice of model is partly driven by convenience: the estimation of option
valuation models using large cross-sections of option contracts is computationally burdensome
because of the need to lter the latent stochastic volatility. As the solution to the Heston (1993)
SQR model relies on univariate numerical integration, it is relatively easier to estimate than
many other models that require Monte Carlo simulation to compute option prices.
It is well recognized that the SQR model cannot capture important stylized facts. In order
to address the limitations of the square root structure, the model is often combined with models
of jumps in returns and/or volatility, or multiple volatility components are used.
2
However,
relatively few studies analyze non-ane stochastic volatility models, and therefore much less is
known about the empirical biases that result from imposing the ane square root structure on
the stochastic volatility dynamic in the rst place. Notable exceptions that investigate option
valuation using non-ane stochastic volatility models are At-Sahalia and Kimmel (2007), Jones
(2003) and Benzoni (2002). The non-ane model in Benzoni (2002) does not improve on the
performance of the Heston (1993) model. Jones (2003) analyzes the more general constant
elasticity of variance (CEV) model using a bivariate time series of returns and an at-the-money
short maturity option, and a number of his specication tests favor the non-ane CEV model over
the SQR model. At-Sahalia and Kimmel (2007) also estimate dierent models using joint time
series of the underlying return and a short-dated at-the-money option price or implied volatility.
They also nd that the SQR model is misspecied, but the nature of the misspecication and
the empirical evidence are dierent from Jones (2003).
1
The leverage eect was rst characterized in Black (1976). For empirical studies that emphasize the im-
portance of volatility clustering and the leverage eect for option valuation see among others Benzoni (2002),
Chernov and Ghysels (2000), Eraker (2004), Heston and Nandi (2000), and Pan (2002).
2
For empirical studies that implement the Heston (1993) model by itself or in combination with dierent types
of jump processes, see for example Andersen, Benzoni and Lund (2002), Bakshi, Cao and Chen (1997), Bates
(1996, 2000), Broadie, Chernov and Johannes (2007), Benzoni (2002), Chernov and Ghysels (2000), Huang and
Wu (2004), Pan (2002), Eraker (2004) and Eraker, Johannes and Polson (2003). Bates (2000) also introduces a
square root model with two volatility components.
2
A number of papers estimate the SQR and alternative stochastic volatility specications on
returns data. These studies nd strong evidence against the SQR specication, see for example
Chacko and Viceira (2003) and Chernov, Gallant, Ghysels and Tauchen (2003). Similarly, the
realized volatility literature seems to strongly favor logarithmic, non-ane SV specications. See
for example, Andersen, Bollerslev, Diebold and Ebens (2001). Yet state-of-the-art option pricing
papers, such as for example Broadie, Chernov and Johannes (2007), continue employing the SQR
model as a building block.
This paper further investigates the empirical implications of adopting a square root stochastic
volatility model for option valuation. We compare the empirical performance of the Heston
(1993) square root stochastic volatility model with that of ve alternatives. In contrast with
At-Sahalia and Kimmel (2007) and Jones (2003), who compare the SQR model with more richly
parameterized stochastic volatility models, we deliberately choose alternative specications with
the same number of parameters as the SQR model. This approach also diers from several other
studies that enrich the SQR model by adding for example jumps to the basic model, which also
increases the number of model parameters. We deliberately ask a very dierent question: is the
SQR model the best possible stochastic volatility model to build upon? If not, what are the
empirical deciencies? Part of our motivation for keeping the number of parameters constant is
our desire to make the in-sample comparison between models as meaningful as possible. It is
well known that in-sample model comparisons favor relatively richly parameterized models, but
that these in-sample model rankings can easily be reversed in out-of-sample experiments.
We conduct an extensive empirical investigation on the six stochastic volatility specications
using several dierent data sources. First, we use realized volatilities to assess the properties of
the SQR model and to guide us in the search for alternative specications. Second, we estimate
the model parameters using maximum likelihood on returns only. Third, we employ nonlinear
least squares on a time series of cross-sections of option data with extensive cross-sectional
variation across moneyness and maturity. Fourth, we value options using realized volatility and
VIX data as proxies for spot volatility. Finally, we investigate the robustness of the option
valuation results with respect to the ltering method used. All evidence points to the same
conclusion: the ane SQR model is severely misspecied, and it is signicantly outperformed
by the simple non-ane models we consider. Overall, the best of the alternative volatility
specications is a model we refer to as the VAR model, which is of the GARCH diusion type.
This model captures the stylized facts in realized volatilities, and it provides the best t for two
of the three return samples. Most importantly, when estimating the models using options data,
it has the lowest option price mean squared errors in- and out-of-sample.
Our estimation on option contracts uses a rich panel of data. It therefore critically diers
from At-Sahalia and Kimmel (2007) and Jones (2003), who estimate model parameters using
a bivariate time series. Indeed, to the best of our knowledge our analysis of the SQR model
uses substantially larger cross-sections of option contracts than any other available study that
explicitly solve the ltering problem. Existing studies either solve the ltering problem and use
a small cross-section of option data, or use a large cross-section and do not explicitly solve the
ltering problem. For example, At-Sahalia and Kimmel (2007), Chernov and Ghysels (2000),
Eraker (2004) and Jones (2003) explicitly solve the ltering problem using a relatively limited
cross-section of option data. Studies that investigate a wider cross-section are for example Bakshi,
Cao and Chen (1997), who estimate one cross-section at a time, and Bates (2000) and Huang and
3
Wu (2004), who estimate a separate volatility parameter for each cross-section. Our estimation on
a rich option data set while explicitly solving the ltering problem is a non-trivial contribution:
One can arguably only reliably estimate the parameters that determine the model smirk and
volatility term structure when using such a wide range of option contracts.
We are able to estimate a large number of stochastic volatility models using comprehen-
sive cross-sections of options data thanks to our use of the particle lter. This methodology
is extensively used in the engineering literature, and has also recently been used for nancial
applications.
3
Particle ltering provides a convenient tool for the analysis of latent factor models
such as stochastic volatility models. It is easy to implement, and it can be adapted to provide
the best possible t to any objective function and any model of interest. In our opinion, the
trade-o with other empirical methods is a very favorable one in the context of computationally
intensive option valuation problems.
The paper proceeds as follows. In Section 2 we discuss the benchmark Heston (1993) model in
light of the evidence from realized volatilities. We also use the realized volatilities to help guide
the search for alternative specications. Section 3 discusses particle lter based estimation on
index returns and index options. Section 4 presents the empirical results obtained using return
and options data. Section 5 concludes. Evidence on the nite sample properties of the particle
lter based estimator and on Monte Carlo based option pricing are provided in the appendix.
2 Stochastic Volatility Specications
This section discusses alternatives to the square root volatility specication in Heston (1993),
which has become the standard building block for more elaborate models in the option valuation
literature. Note that it is relatively straightforward to write down more heavily parameterized
models that outperform the square root model. Trivially, a more heavily parameterized model
will always outperform a simpler nested model in-sample. Moreover, Bates (2003) points out
that even in short-horizon out-of-sample experiments, the richer models may outperform nested
models because the smile or smirk patterns in option prices is very persistent. This nding is
more likely if the models are re-estimated frequently, say daily as in the classic study by Bakshi,
Cao and Chen (1997) for example.
Thus, we deliberately conne attention to models that have the same number of parameters
as Hestons SQR model, and we estimate on a long sample of data. The models we consider
can be thought of as alternative building blocks for more elaborate models containing jumps in
returns and volatility, as well as multiple volatility factors. However, such extensions are beyond
the scope of this paper. In our paper, the objective is to establish a well-specied volatility
dynamic, which in our opinion ought to precede the development of more elaborate versions of
a possibly misspecied basic model.
3
See Gordon, Salmond and Smith (1993) and Pitt and Shepherd (1999). See Johannes, Polson and Stroud
(2008) for an application to returns data.
4
2.1 Realized Variance: Implications for the Square Root Model
The Heston (1993) square root (SQR) model assumes that the instantaneous change in variance,
\ , has the following dynamics
SQR: d\ = i(o \ )dt +o
p
\ dn (1)
where i denotes the speed of mean reversion, o is the unconditional variance, and o deter-
mines the variance of variance. The variance innovation dn is a Brownian motion, and we have
co::(d.. dn) = j, where d. is the innovation in the underlying spot price, o, given by
do = jodt +
p
\ od. (2)
where j is the mean of the instantaneous rate of return. We will use the return specication (2)
in each of the volatility models considered below.
In order to explore the implications of the SQR specication, consider the instantaneous
volatility dynamic implied by the model. Using Itos lemma, we can write
SQR: d
p
\ = j(\ )dt +
1
2
odn (3)
where the volatility drift j(\ ) is a function of the variance level. Note that the SQR model
implies that the instantaneous change in volatility should be Gaussian and homoskedastic: \
does not show up in any way in the diusion term for d
p
\ . This is a strong implication, which
can be quite easily evaluated empirically.
As a rst piece of empirical evidence, we construct daily realized variances, 1\
t
, from intraday
returns. We obtain intraday S&P500 quotes for 1996 through 2004 from which we construct a
grid of two-minute returns. From these two-minute returns we construct robust realized variances
using the two-scale estimator from Zhang, Mykland, and At-Sahalia (2005).
4
The two-scale
estimator is dened as
1\
TS
t
= 1\
Avr
t
0.06141\
All
t
where 1\
Avr
t
is the average of all the possible low-frequency RV estimates that use (in our
case) 30-minute squared returns on the two-minute grid, and where 1\
All
t
is the high-frequency
estimator summing over all the available two-minute squared returns.
5
The coecient on 1\
All
t
depends on the frequency of the sparse estimators (30 minutes here) and of the high-frequency
estimator (2 minutes here). Zhang, Mykland, and At-Sahalia (2005) show that even in the
presence of market microstructure noise this estimator converges to the true integrated variance
dened by
_
t
t1
\

dt.
Consider now the left column of Figure 1. Using daily realized volatilities from 1996 through
2004, the top-left panel of Figure 1 shows a quantile-quantile (QQ) plot of the daily realized
volatility changes compared with the Gaussian distribution. The deviation of the data points
from the straight line indicates that the Gaussian distribution is not a good assumption for daily
changes in volatility. The observed tails (both left and right) are considerably fatter than the
4
We scale up the open-to-close realized variance estimates to match the overall variance in close-to-close daily
returns for the sample period.
5
See At-Sahalia, Mykland and Zhang (2005) for a discussion of the optimal sampling frequency.
5
normal distribution would suggest. Now consider the middle-left panel in Figure 1, which scatter
plots the daily volatility changes against the daily volatility level. According to the SQR model,
this scatter plot should not display any systematic patterns. However, as the volatility level
increases on the horizontal axis, a cone-shaped pattern in the daily volatility changes on the
vertical axis is apparent. The bottom-left panel of Figure 1 conrms this nding: it scatter plots
the absolute daily volatility changes against the daily volatility level. A simple OLS regression
line is shown for reference. Notice the apparent positive relationship between the volatility level
and the magnitude of the volatility changes. This pattern is in conict with the homoskedastic
volatility implication of the SQR model.
Using Itos lemma, the dynamics for log variance in the SQR model can be written as
SQR: d ln(\ ) = j(\ )dt +o
1
p
\
dn (4)
The right column of Figure 1 summarizes the empirics of the logarithms of the realized vari-
ances. The top panel shows that daily changes in the realized log variances follow the Gaussian
distribution quite closely. The result clearly diers from the corresponding top left panel for
the daily volatility changes. Furthermore, the scatter of daily log variance changes against the
log variance level in the middle-right panel of Figure 1 does not reveal a cone-shaped pattern.
The bottom-right panel in Figure 1 conrms this result: it shows a virtually horizontal line
when regressing absolute changes in log variance on log variance levels. The absence of a clear
relationship between changes in the log variances and the log variance level in Figure 1 casts
further doubt on the SQR specication, for which the instantaneous changes in log variances are
heteroskedastic, as is evident from (4).
2.2 Alternative Specications for Variance Dynamics
The evidence in Figure 1 suggests specifying the variance dynamic as follows
VAR: d\ = i(o \ )dt +o\ dn (5)
which has the property that ln(\ ) is homoskedastic.
6
d ln(\ ) = j(\ )dt +odn
Heston (1997) and Lewis (2000) suggest another alternative to the SQR model: the so-called
three halves model dened by
3/2N: d\ = i\ (o \ )dt +o\
3=2
dn (6)
In addition to a dierent power on \ in the diusion term, the 3/2N model has the interesting
implication that the variance drift is nonlinear in the level of the variance. We highlight this
feature by denoting the model 3/2N. This allows for a potentially fast reduction in volatility
6
Although it would seem natural from the realized variance analysis above, we do not consider the homoskedas-
tic log SV model where d ln(V ) = ( ln(V )) +dw, because it typically performs very similar tobut slightly
worse thanthe VAR model in (5).
6
following unusually large volatility spikes, such as October 1987, and for slow increases in volatil-
ity during prolonged low-volatility episodes, such as the mid 1990s. The 3/2N model also leads
to a closed-form solution for European call option prices (see Lewis (2000)), even though to the
best of our knowledge the model has not yet been implemented using this solution.
So far we have considered three dierent specications of the diusion and two dierent
specications for the drift of \ . We can think of the three models considered so far as belonging
to the class
d\ = i\
a
(o \ )dt +o\
b
dn, for c = f0. 1g and / = f1,2. 1. 3,2g (7)
Although we will focus our discussion on the SQR, VAR and 3/2N models, this framework
encompasses a total of six models, which we denote
c / Name
0 1/2 SQR
1 1/2 SQRN
0 1 VAR
1 1 VARN
0 3/2 3/2
1 3/2 3/2N
All of these models will be estimated below.
The literature contains several good discussions of the properties of stochastic dierential
equations. See for instance Karlin and Taylor (1981). At-Sahalia (1996) contains a discussion of
the properties of general interest rate processes and Lewis (2000) contains a treatment of certain
variance processes. Jones (2003) provides an excellent discussion of the CEV stochastic variance
process. He shows that for a model with c = 0 and / 1, zero and plus innity are inaccessible
values for the stochastic variance. He also shows that a solution to the variance SDE exists and
is unique, and that the stationary distribution exists. Thus Jones (2003) covers the model we
refer to as 3/2. Similar arguments can be used to show that these results hold also for the VAR
model, the VARN model and the 3/2N model. For the SQRN model, all but the existence of the
stationary distribution can be shown using similar arguments as well.
2.3 Volatility Risk Premia
In order to compute option prices, which depend on the price of volatility risk, we need to specify
a volatility risk premium relationship for each of the above six models. Heston (1993) assumes
that the volatility risk premium `(o. \. t) is equal to `\ , so that the risk neutral dynamic for
the SQR model is
SQR: do = :odt +
p
\ o d.

(8)
d\ = (i(o \ ) +`\ ) dt +o
p
\ dn

(9)
= (i `)(io,(i `) \ )dt +o
p
\ dn

where d.

and dn

are Brownian motions under the risk-neutral measure, with co::(d.

. dn

) =
j. Notice that the variance dynamic takes the same form under the physical and risk neutral
measures.
7
To provide some intuition for the properties of risk-neutral volatility, Figure 2 plots various
stylized facts for the model-free, option-implied VIX volatility index, provided by the CBOE.
The VIX is measured in annualized percentage standard deviations. Similar to Figure 1, we
report the QQ plot of the VIX in the top left panel. The daily VIX changes are scattered against
the VIX level in the center-left panel, and the daily absolute VIX changes are scattered against
the VIX level in the bottom-left panel. In the right column, we provide the same analysis using
the natural logarithm of the VIX rather than the VIX level.
7
The similarities with the RV-based
plots in Figure 1 are quite striking. The level of the VIX displays strong evidence of nonnormality
and heteroskedasticity. The distribution of the logarithm of the VIX appears to be much closer
to the normal distribution, and shows much less evidence of heteroskedasticity.
A key lesson from Figure 2 is that the (model-free) risk-neutral volatility distribution has
similar features to the (model-free) physical volatility distribution in Figure 1. Therefore, when
deciding on the risk premium specication for the non-ane models, we choose a form that
preserves the functional form of the spot variance process under the two measures. In particular,
we use the following functional form for the variance risk premium
`(o. \. t) = `\
a+1
(10)
This specication ensures that the variance dynamic takes the same form under the two measures,
as is the case in most available empirical studies that estimate the Heston (1993) SQR model.
Under the risk neutral measure, we have
do = :odt +
p
\ o.

(11)
d\ = (i `)\
a
(io,(i `) \ )dt +o\
b
dn

with co::(d.

. dn

) = j.
In all six models we investigate, the risk-neutralization can be obtained using a no-arbitrage
argument. A utility-based argument behind the risk neutralization can be explicitly characterized
for some cases when assuming log-utility. See Lewis (2000) for a thorough discussion of these
issues.
Note that the risk-neutral process in (11) can be rewritten
do = :odt +
p
\ o.

(12)
d\ = i

\
a
(o

\ )dt +o\
b
dn

where i

= (i `), and o

= io,i

. This representation highlights that ` cannot be identied

in any of the models when using only options in estimation. Below we perform two types of
valuation exercises: one with ` as well as the other parameters estimated using option data and
returns jointly, and one with ` equal to zero and the other parameters estimated from returns
only.
2.4 Computing Option Prices
Heston (1993) demonstrates that the SQR model admits a closed form solution for option prices,
which can be written as
C(\
t
) = o
t
1
1;t
Ac
r(Tt)
1
2;t
(13)
7
Note that the analysis for the natural logarithm of the VIX is equivalent to the analysis of the natural
logarithm of the squared VIX.
8
where 1
1;t
and 1
2;t
are computed using numerical integration of the conditional characteristic
function.
Heston (1997) and Lewis (2000) show that for the 3/2N model, a similar closed-form solution
is available, but in this case the characteristic function involves the gamma and conuent hyper-
geometric function with complex arguments. To the best of our knowledge, this solution is very
numerically intensive and has not yet been implemented empirically. Barone-Adesi, Rasmussen
and Ravanelli (2003) and Sabanis (2003) develop approximate analytical solutions for the VAR
model, but as far as we know an exact solution is not available.
In order to ensure that dierences across models are not driven by a particular numerical
technique, we compute all model option prices using Monte Carlo. Thus the call option prices
are computed via the Monte Carlo sample analogue of the discounted expectation
C(\
t
) = c
r(Tt)
1

t
[o
T
A. 0] (14)
where the expectation is calculated using the risk neutral measure.
In order to calculate Monte Carlo prices, we use 1,000 simulated paths and a number of
numerical techniques to increase numerical eciency, namely the empirical martingale method
of Duan and Simonato (1998), stratied random numbers, antithetic variates and a Black-Scholes
control variate technique. In the appendix we use the SQR model to demonstrate that the Monte
Carlo prices match closely the prices computed via numerical integration.
3 Estimation Using the Particle Filter
The SQR model has been investigated empirically in a large number of studies. It is often used
as a building block, together with models of jumps in return and volatility. For our purpose,
it is important to note that when estimating the model on option data, a number of dierent
techniques can be used. First, the models parameters can be estimated using a single cross-
section of option prices treating \
t
as a parameter. This is done for example in Bakshi, Cao and
Chen (1997). A second type of implementation of the SQR model uses multiple cross-sections of
option prices but does not use the information in the underlying asset returns. Instead, for every
cross-section a dierent initial volatility is estimated, leading to a highly parameterized problem.
This approach is taken for instance in Bates (2000) and Huang and Wu (2004). A third group
of papers provide a likelihood-based analysis of the stochastic volatility model. See for example
At-Sahalia and Kimmel (2007), Bates (2004), Jones (2003), and Eraker (2004), who provides a
Markov Chain Monte Carlo analysis. A likelihood-based approach can combine information from
the option data and the underlying returns, and impose consistency between the physical and
risk neutral dynamics in estimation. Both the return data and the option data carry a certain
weight in the objective function. Finally, Chernov and Ghysels (2000) use the ecient method
of moments and Pan (2002) uses a method of moments technique as well. These methods can
also combine information in option data with the information in underlying returns.
The empirical challenge in stochastic volatility models is that the spot volatility \
t
is an
unobserved latent factor. In order to estimate the parameters in stochastic volatility models,
we thus need to apply a ltering technique using observed index returns. The last two groups
of papers discussed above explicitly solve the ltering problem. The ltering problem is also
9
explicitly considered in some papers that use return data to estimate continuous-time stochastic
volatility models, such as Chernov, Gallant, Ghysels and Tauchen (2003).
We provide an alternative implementation of the ltering problem for the six stochastic
volatility models under investigation using the particle lter (PF) algorithm. As shown by
Gordon, Salmond and Smith (1993), the PF oers a convenient lter for nonlinear models such as
the stochastic volatility models we consider here. The PF has recently been applied by Johannes,
Polson and Stround (2008), who use return data to estimate continuous-time stochastic volatility
models with jumps. But because the PF procedure is relatively new in nance, we discuss the
implementation of this method in some detail.
We use the PF both for maximum likelihood estimation on returns and for nonlinear least
squares estimation on option prices. Below, we rst describe the PF, then the maximum likeli-
hood importance sampling (MLIS) estimation methodology we use to estimate the models using
return data, and nally the nonlinear least squares importance resampling (NLSIS) estimation
methodology we use to estimate the models using option data.
3.1 Volatility Discretization
Working with log returns, we can write the generic SV process as
d ln(o) =
_
j
1
2
\
_
dt +
p
\ d. (15)
d\ = i\
a
(o \ )dt +o\
b
dn (16)
Note that the equations in (15) and (16) specify how the unobserved state is linked to observed
stock prices. This relationship allows us to infer the volatility path using the returns data. We
rst need to discretize (15)-(16). There are dierent discretization methods and every scheme
has certain advantages and drawbacks. We use the Euler scheme, which is easy to implement
and has been found to work well for this type of applications.
8
Discretizing (15)-(16) gives
ln(o
t+1
) = ln(o
t
) +
_
j
1
2
\
t
_
+
_
\
t
.
t+1
(17)
\
t+1
= \
t
+i\
a
t
(o \
t
) +o\
b
t
n
t+1
(18)
We implement the discretized model in (17) and (18) using daily returns, but all parameters will
be expressed in annual units below. We now describe the volatility ltering procedure.
3.2 Volatility Filtering
The particle lter algorithm relies on the approximation of the true density of the state \
t+1
by a set of ` discrete points or particles that are updated iteratively through equations (17)
and (18). The lter is implemented using particles or draws,
_
\
j
t+1
_
N
j=1
. from the empirical
distribution of \
t+1
, conditional on particles or draws,
_
\
j
t
_
N
j=1
. from the empirical distribution of
\
t
. Our particular implementation of the particle lter is referred to as the Sampling-Importance-
Resampling (SIR) particle lter and follows Pitt (2002).
We recursively proceed through the following three steps for t = 1. .... 1 1:
8
See for example Eraker (2001).
10
3.2.1 Step 1: Simulating the State Forward: Sampling
We rst simulate the state forward by computing ` raw particles {
~
\
j
t+1
}
N
j=1
from the set of
smooth resampled particles {\
j
t
}
N
j=1
using equation (18), taking the correlation between returns
and variance into account. We x the particles in the rst period as \
j
1
= o, for all ,.
9
We can then get correlated shocks via
.
j
t+1
=
_
ln(o
t+1
) ln(o
t
)
_
j
1
2
\
j
t
__
,
_
\
j
t
(19)
n
j
t+1
= j.
j
t+1
+
_
1 j
2

j
t+1
where
j
t+1
are independent random draws from the standard normal so that co::(.
j
t+1
.
j
t+1
) = 0
and co::(.
j
t+1
. n
j
t+1
) = j. Substituting (19) into (18) we get
~
\
j
t+1
= \
j
t
+i
_
\
j
t
_
a
_
o \
j
t
_
+o
_
\
j
t
_
b
n
j
t+1
(20)
This simulates ` raw particles and thus provides a set of possible values of \
t+1
.
3.2.2 Step 2: Computing and Normalizing the Weights
At this point, we have a vector of ` possible values of \
t+1
and we know according to equation
(17) that given the other available information, \
t+1
is sucient to generate ln(o
t+2
). Therefore,
equation (17) oers a simple way to evaluate the likelihood that the observation o
t+2
has been
generated by \
t+1
. Hence, we are able to compute the weight given to each particle (or the
likelihood or probability that the particle has generated o
t+2
). The weight is computed as
follows
~
\
j
t+1
=
1
_
2:
~
\
j
t+1
exp
_
_
_

1
2
_
ln
_
S
t+2
S
t+1
_

_
j
1
2
~
\
j
t+1
__
2
~
\
j
t+1
_
_
_
(21)
for , = 1. ... `. Finally, we normalize the weights via

\
j
t+1
=
~
\
j
t+1

N
j=1
~
\
j
t+1
. (22)
3.2.3 Step 3: Smooth Resampling
The set
_

\
j
t+1
_
N
j=1
can be viewed as a discrete probability distribution of
~
\
t+1
from which we
can resample. Let the ordered particles be dened by
~
\
(j)
t+1
then the corresponding cumulated
sum of ordered weights,

\
(j)
t+1
. gives the discrete, step-function CDF corresponding to the dis-
tribution of \
t+1
. Pitt (2002) proposes smoothing the

\
(j)
t+1
-based CDF in order to ensure that
the subsequent likelihood function is smooth in the model parameters, which renders numerical
9
In the returns-based MLIS estimation, t = 1 is simply the rst day of observed returns. In the options-based
NLSIS estimation, t = 1 is one year prior to the rst available option quote.
11
optimization feasible. The smooth bootstrap in Efron and Tibshirani (1993) oers a possible
but computationally expensive solution to this problem. Fortunately, Pitt (2002) provides a
very ecient smoothing approach which is crucial for our extremely intensive option valuation
application.
Pitts (2002) approach is simple and intuitive: The discrete, step-function empirical CDF
implied by

\
(j)
t+1
can be converted to a smooth continuous CDF by partitioning the sample space
for \
t+1
into regions delimited by the observed ordered particles. Let region , be dened by
1
j
= [
~
\
(j)
t+1
.
~
\
(j+1)
t+1
]. Now dene the probability of \
t+1
being in region , as
Pr (\
t+1
2 1
j
) =
1
2
_

\
(j)
t+1
+

\
(j+1)
t+1
_
. , = 2. 3. ...` 2 (23)
and dene the conditional density for region i by
q (\
t+1
j\
t+1
2 1
j
) =
1
_
~
\
(j+1)
t+1

~
\
(j)
t+1
_. , = 2. 3. ...` 2 (24)
By appropriately dening the probability distribution in the end-regions, that is for , = 1
and , = ` 1, we ensure that the probabilities sum to one and that the continuous CDF
corresponding to (24) passes through a midpoint of each step in the discrete CDF. As ` becomes
large, the continuous resampled CDF will converge to the discrete CDF, which in turn will
converge to the true CDF. We refer to Pitt (2002) and Smith and Gelfand (1992) for the details.
Note from (24) that we are assuming a uniform distribution on each region of the sample
space. Because the uniform distribution is easy to invert, this is a computationally ecient way
to get a set of smooth particles. We resample the particles a total of ` times in proportion to
their smooth likelihood in (23) and (24) to get the smooth resampled particles,
_
\
j
t+1
_
N
j=1
.
The ltering for period t + 1 is now done. The ltering for period t + 2 starts in step 1
above by computing shocks for period t + 2 in (19) using the smooth resampled particles, \
j
t+1
,
as inputs. Repeating steps 1-3 for all t yields a discrete distribution of ltered spot variances on
each day.
3.3 Maximum Likelihood Estimation
Our rst empirical strategy uses a long sample of daily S&P500 index returns to estimate each of
the SV models by maximizing the likelihood. To this end, we need an estimation methodology
which allows for models with a latent volatility factor.
Pitt (2002) builds on Gordon, Salmond and Smith (1993) to showthat the parameters of latent
factor models in general, and of the SV model in (17) and (18) in particular, can be estimated
by maximizing the Maximum Likelihood Importance Sampling criterion, which is simply dened
by
`11o (j. i. o. j. o) =
T

t=1
ln
_
1
`
N

j=1

\
j
t
_
(25)
As described in the previous section, the particle weights,

\
j
t
, are determined via the condi-
tional likelihood of particle , at time t. These individual likelihoods in turn are given from the
12
model specication in (17) and (18), taking into account that .
t
and n
t
are correlated normal
random variables. The MLIS criterion then simply averages the particle weights across particles,
takes logs, and sums over time to create a log likelihood function.
A key challenge in the use of the MLIS for estimation and inference is that it is not generally
smooth in the underlying parameters. However, as discussed above, Pitts (2002) ingenious
implementation of the particle lter, where the resampling in Step 3 is done in a smooth fashion,
ensures that the MLIS criterion is smooth in the parameters. This smoothing drastically improves
the numerical optimization performance and it enables us to compute reliable parameter standard
errors using conventional rst-order techniques.
In the appendix, we provide a small scale Monte Carlo experiment comparing the `11o
estimator to other estimators available in the literature. We nd that the `11o estimator
performs well in comparison with much more computationally intensive methods.
3.4 Nonlinear Least Squares Estimation
Our second empirical strategy is to take a large panel of options traded on the S&P500 index
and estimate each of the SV models by minimizing the option pricing errors on this sample. For
all the SV models, our implementation uses the nonlinear least squares importance sampling
(NLSIS) estimation technique, which minimizes the following mean squared implied volatility
error
1\ `o1 (j. i. o. j. o. `) =
1
`
T

t;i
_
1\
i;t
1o
1
i
_
C
i
(

\
t
)
__
2
(26)
where 1\
i;t
is the Black-Scholes implied volatility corresponding to the market price of option i
quoted on day t. C
i
_

\
t
_
is the model price evaluated at the ltered variance, and 1o
1
_
C
i
(

\
t
)
_
denotes the Black-Scholes inversion of the model option price.
10
The total number of options
in the sample is denoted by `
T
=
T

t=1
`
t
, where 1 is the total number of days included in the
options sample, and where `
t
is the number of options with various strike prices and maturities
included in the sample at date t. The ltered spot variance

\
t
is simply the average of the smooth
resampled particles

\
t
=
1
`
N

j=1
\
j
t
(27)
Our implementation minimizing (26) is dierent from existing studies that estimate SV mod-
els from option data while solving the ltering problem. Most other studies are likelihood-based.
One advantage of our approach is that it is relatively straightforward and fast, which allows
us to estimate the models using much more extensive cross-sections of options. In our opinion,
estimation using (26) has an additional advantage, because matching the objective function used
in parameter estimation with the function subsequently used to evaluate the models ensures
the best possible performance of the models in- and out-of-sample. This is motivated by the
insights of Granger (1969), who demonstrates that the choice of objective function (also labeled
10
Notice that we could alternatively compute the model price as the weighted average of the option prices
computed for each particle. However, such an approach would be computationally very costly. Note that if the
distribution of particles is centered around the mean, the two approaches will yield very similar results.
13
loss function) is an integral part of model specication. It follows that estimating a model using
one objective function and evaluating it using another one amounts to a suboptimal choice of
objective function. Christoersen and Jacobs (2004) demonstrate that this issue is empirically
relevant for the estimation of the deterministic volatility functions in Dumas, Fleming and Wha-
ley (1998). We thus choose to implement the SV models in a way that is consistent with these
insights. Notice however that our use of the particle ltering algorithm is completely general:
we can apply this technique to any volatility model and using any well-behaved loss function
involving option prices and underlying returns, including a likelihood function.
Our optimization algorithm minimizes (26) using an iterative procedure on the structural
parameters. At each iteration, the volatility is ltered through time using the current parameter
values and the information embedded in observed returns. Using the ltered volatility and the
structural parameters, option prices are computed and the IVMSE is calculated. The procedure
searches in the structural parameter space until an optimum is reached.
4 Empirical Results
This section presents our empirical results obtained using returns and options data. First, we
discuss the return-based MLIS estimation results and dierences in volatility sample paths. Sec-
ond, we introduce the option data set and use the MLIS parameters to compare option valuation
performance across models. Third, we estimate the models using NLSIS estimation and ana-
lyze the patterns in the option valuation errors. Fourth, we check the robustness of our results
to the use of dierent volatility proxies and dierent volatility ltering techniques. In all our
implementations of the PF, we use ve hundred particles.
Out-of-sample experiments, in which the models are evaluated using data not used in estima-
tion, are conducted in two dierent ways. First, we use purely return-based MLIS estimates to
value options. Second, we use NLSIS-based parameters obtained from Wednesday call options
to value Wednesday put options and Thursday call options. We use the same calendar period
for the out-of-sample study so as to avoid the impact of structural breaks in volatility. Because
we are comparing equally parsimonious models estimated on a large sample, the Bates (2003)
critique mentioned in Section 2, that more heavily parameterized models are favored in such
out-of-sample experiments, does not apply.
4.1 MLIS Estimation Results from Index Returns
As mentioned above, our rst empirical strategy uses a long sample of daily S&P500 index returns
and estimates each of the SV models by maximizing the model t for this sample. First, recall
the SV model specications we consider
d\ = i\
a
(o \ )dt +o\
b
dn, for c = f0. 1g and / = f1,2. 1. 3,2g.
We use daily S&P500 returns from CRSP and estimate the physical parameters by maximizing
`11o (i. o. j. o) =
T

t=1
ln
_
1
`
N

j=1

\
j
t
_
14
In the MLIS optimization, the return drift j is xed at the sample average daily return in
all models. The optimal parameters as well as the MLIS optimum values and the volatility
properties for each model are given in Table 1. Consider rst the MLIS objective values. We
report three sets of values. First, we present results for 1996-2004, which matches the sample
period we will use in the subsequent option estimation analysis. Second, we present results for
1989-2004, which is a longer sample period that does not include the 1987 crash, and nally
results for 1985-2004, which includes the crash. Note rst that the ranking of models is very
stable across the various samples. The VAR (c = 0. / = 1) model or the 3/2N (c = 1. / = 3,2)
model is always best or second best. The SQR (c = 0. / = 1,2) model is ranked fourth or fth
and the SQRN specication (c = 1. / = 1,2) is always worst. While we do not have inference
procedures for these MLIS objective values for non-nested models, recall that in standard LR
tests, adding one parameter to a model is signicant at the 5% level if the log-likelihood increases
by approximately two points. The dierences in objective values across these models that have
the same number of parameters therefore appear to be quite large.
Table 1 also contains the parameter estimates for the 1996-2004 sample period and the right-
most four columns of Table 1 provide the rst four sample moments of the ltered volatility. Note
that the mean ltered volatility is similar across models. Note also that the i and o parameters
are not directly comparable between linear (c = 0) and nonlinear (c = 1) drift specications.
Figure 3 therefore plots the drift function for all models (solid lines) as a function of the level
of variance. The linear drift specications are given in the left column and the nonlinear drifts
in the right column. The square root diusions are in the top row, the linear diusions in the
middle row, and the 3/2N diusions in the bottom row. The dierences between linear and
nonlinear drift specications are most evident for large values of the spot variance where the
mean-reversion in the nonlinear models is much stronger.
The diusion parameter o is comparable within a given diusion specication (i.e. for a
given value of /), but not across diusion specications. In order to facilitate comparisons of the
models, Figure 3 shows the diusion functions (dashed lines) plotted against the value of the spot
variance \ . Notice that in comparison with the VAR diusion specication (/ = 1) in the middle
row, and the 3/2 specications (/ = 3,2) in the bottom row, the SQR and SQRN specications
(/ = 1,2) in the top row enable much less diusion in the variance when the variance level is
large.
The nal parameter estimate is that of j, which captures the correlation between the shocks
to return and variance. Ranging from 0.7876 to 0.7411, the estimate of j is stable across
models and relatively large in magnitude. However, recent studies, including Jones (2003) and
At-Sahalia and Kimmel (2007), have obtained similarly large correlation estimates.
4.2 Properties of the Filtered Spot Volatilities
We now present some more empirical evidence on the dierences in model performance over
time. We rst plot the path of ltered volatility over time in Figure 4. Subsequently, we plot
the path of the conditional volatility of variance in Figure 5. Finally, in Figures 6-8 we assess
the distributional characteristics of the ltered spot volatilities for the three models with linear
drift, and compare them to the properties of the realized volatilities in Figure 1.
Figure 4 plots the ltered volatility paths
_

\
t
for each model during the 1996 to 2004 sample
15
period. All volatility paths are shown on the same scale, going from zero to 70 percent volatility
in annual terms. Naturally, the overall pattern in volatility over time is similar across models.
However, when volatility increases, it tends to do so much more sharply in the 3/2 models
and somewhat more sharply in the linear (VAR) diusion models when compared with the
square root diusion models. The VAR and 3/2 diusions thus exhibit more spikes in volatility
when compared with the SQR model. Examples of this include September 1998 (LTCM/Russia
default), September 2001 (9/11), and July 2002.
The path of spot variances in Figure 4 is of course crucial for the models option valuation
performance over time. However, the time paths of the higher moments are equally important.
We can dene the model-based conditional variance of variance as
\ c:
t
(\
t+1
) = o
2

\
2b
t
. (28)
Figure 5 shows the path of the annualized conditional volatility of variance for each model, i.e.
the square root of the expression in (28). We again use the same scale for all the plots so
as to emphasize the dierences across models. Notice how high-volatility episodes such as the
September 1998 LTCM debacle and Russia default, as well as the July 2002 stock market decline,
lead to pronounced dierences between the benchmark models, namely the SQR model in the
top-left panel, the VAR model in the middle-left panel, and the 3/2N model in the bottom-right
panel.
In Figures 6 through 8, we report on the distributional properties of the levels of the spot
volatilities (left column) and the natural logarithms of the variances (right column), for the SQR,
VAR, and 3/2 models with linear drifts. As in Figures 1 and 2, we report the QQ plots in the top
row, the daily changes scattered against the volatility (or log variance) in the middle row, and
the daily absolute changes scattered against volatility (or log variance) in the bottom row. The
dierences across the three models are quite striking. The QQ plots reveal that the SQR model
in Figure 6 has close to normally distributed spot volatilities but nonnormal log spot variances.
The VAR model in Figure 7 has non-normal volatilities but approximately normal log variances,
and the 3/2 model in Figure 8 has strong non-normality in both the spot volatilities and the log
variances. The scatter plots are also quite revealing. Notice that the SQR model in Figure 6 has
close to homoskedastic volatility changes but heteroskedastic log variance changes. The VAR
model in Figure 7 has heteroskedastic volatility changes but close to homoskedastic log variance
changes. The 3/2 model in Figure 8 is characterized by heteroskedastic changes in both volatility
and log variance.
Properties of the distribution of annualized ltered volatilities can also be inferred from the
standard deviation, skewness, and kurtosis of the
_

\
t
paths provided in Table 1 for all six
models. Not only is the standard deviation of volatility dierent across models, the VAR, 3/2
and 3/2N models are also characterized by much larger skewness and excess kurtosis of volatility
compared to the SQR model. This of course conrms the ndings in Figures 6-8.
Finally, note that when comparing the model-based spot volatilities and log variances in
Figures 6-8 to the model-free realized volatilities and log variances in Figure 1, it appears that
the VAR model seems to best capture the distribution of volatility.
16
4.3 Option Data
We conduct our empirical option-based analysis using S&P500 index option data for the 1996-
2004 period. We only use Wednesday and Thursday options data. For the NLSIS based in-sample
analysis, we use Wednesday call option data. Wednesday is the day of the week least likely to
be a holiday. It is also less likely than other days such as Monday and Friday to be aected by
day-of-the-week eects. The decision to pick one day every week is to some extent motivated
by computational constraints. The optimization problems are fairly time-intensive, and limiting
the number of options reduces the computational burden. Using only Wednesday data allows us
to study a long time-series, which is useful considering the highly persistent volatility processes.
An additional motivation for only using Wednesday data is that following the work of Dumas,
Fleming and Whaley (1998), several studies, including Heston and Nandi (2000), have used this
setup. We use the parameters obtained in the NLSIS estimation exercise together with either
Wednesday put options or Thursday call options in order to assess the out-of-sample performance
of the models.
Panel A of Table 2 presents descriptive statistics for the options data for the 1996-2004
Wednesday closing call option data by moneyness and maturity. The sample includes a total of
16,506 Wednesday call option contracts with an average mid-price of $46.05 and average implied
volatility of 20.26%. The implied volatility is largest for the in-the-money options reecting the
well-known volatility smirk in index options. The average implied volatility term structure is
roughly at during the period. The average bid-ask spread is $1.60 corresponding to about 3.5%
of the average mid-price of $46.05.
Panel B of Table 2 provides the option data characteristics for put options recorded at the
close of each Wednesday. We observe a total of 23,035 puts with an average mid-price of $40.87
and an average implied volatility of 21.80%. The average bid-ask spread is $1.53 or about 3.7%
of the average mid-price. The implied volatility smirk is evident in the put options as well: The
out-of-the-money put options have an average of 24.26% implied volatility compared with around
20% for at-the-money puts. Panel C of Table 2 shows that the Thursday call option sample has
15,390 options with characteristics that are similar to those of the Wednesday call option data.
4.4 Option Valuation with MLIS Estimates
This section presents results from option valuation using the MLIS estimates in Table 1, obtained
using returns only. In this case, we cannot estimate the volatility risk premium ` . We therefore
set ` to zero, implying that the variance dynamics are the same under the two measures. Clearly,
setting ` to zero will worsen the t of all the models. Consequently, the focus in this analysis is
not primarily on the level of the pricing errors, but rather on the relative t across models. In
the subsequent section, we will estimate ` along with the other parameters, using both return
and option data.
Table 3 contains the results for the MLIS-based option valuation exercise. We report the t
of the dierent SV models using our three sets of options data (Wednesday calls, Wednesday
puts, and Thursday calls) and the following three criteria: implied volatility root mean squared
error, 1\ 1`o1; dollar root mean squared error, $1`o1, which uses the option mid-prices;
and out-of-spread root mean squared error, C1`o1, which sets the dollar pricing errors to zero
if the model price falls between the observed bid and ask price for a given option.
17
Table 3 uses the ane SQR model as the benchmark model and it reports the RMSE ratio
between each model and the SQR model. It also reports the Diebold and Mariano (1995) test
of the null hypothesis that the weekly MSE of each model is equal to that of the SQR model.
Boldface font indicates signicance at the 5% level.
Table 3 shows that for each of the 45 pairwise comparisons made, the RMSE ratio with the
SQR model is smaller than one. Thus the SQR model is outperformed by all other models for all
three criterion functions and for all three sets of option data. The ratios tend to be the lowest
for the Co1`o1 loss function, suggesting that the dierences between the SQR and the other
models are economically relevant. From the Diebold-Mariano test statistics we see that a large
majority of the dierences with the SQR model are statistically signicant. A comparison of the
ve non-ane models suggests that none of the models systematically outperforms the others.
A given models performance depends on the criterion and the data used, but in general the
dierences between the ve non-ane models are relatively small.
4.5 NLSIS Estimation on Option Prices and Returns
While the MLIS-based results in Table 3 are of interest, it is likely that a much better t can be
obtained by estimating all the parameters including the volatility risk premium using information
from both underlying returns and option prices. This section presents such results using `1o1o
estimation.
Table 4 contains the parameter values obtained from minimizing the option implied volatility
mean-squared-error, dened above as
1\ `o1 (i. o. j. o. `) =
1
`
T

t;i
_
1\
i;t
1o
1
i
_
C
i
(

\
t
)
__
2
(29)
In the NLSIS optimization, the return drift, j, is xed at the sample average daily return in all
models, as was the case in `11o estimation.
Consider rst the in-sample root mean squared error (IVRMSE) column in Table 4, which
uses Wednesday call options to compare the models. The SQR model performs the worst with an
IVRMSE of 3.32%, and the VAR model performs the best with an IVRMSE of 2.86%. The RMSE
thus diers by about 14% between the SQR and the VAR model. Note that this improvement
in t is obtained without adding any parameters to the model. The 3/2N model is about 11%
better than the SQR model and the VARN model with nonlinear drift and linear diusion
(c = 1. / = 1) is about 10% better than the SQR model. The VAR, VARN and 3/2N models
signicantly outperform the SQR model when judged by the Diebold-Mariano statistic. The
remaining two models do not signicantly outperform the SQR model.
Consider next the out-of-sample columns in Table 4, which use Wednesday puts and Thursday
calls respectively. For Wednesday put options, the VAR model outperforms the SQR model by
approximately 19%, and the VAR model is the only model that signicantly outperforms the
SQR model according to the Diebold-Mariano test. Using Thursday call options, the VAR model
again performs the best and outperforms the SQR model by a signicant 17%. For Thursday
calls, the 3/2N model improves upon the SQR model by 11%, which is also signicant when
judged by the Diebold-Mariano test. The VARN model has an IVRMSE that is 9% below the
SQR, but this dierence is not statistically signicant. The remaining three models show minor
and statistically insignicant improvements.
18
The overall improvement in IVRMSE provided by the VAR model over the benchmark SQR
model is quite substantial. To judge the magnitude of the improvement in t, the most natural
reference point is the rich literature that uses Poisson jumps in returns and/or volatility in con-
junction with an SQR stochastic volatility model. The evidence on the in-sample and especially
the out-of-sample improvement provided by including jump processes is inconclusive, with some
studies nding moderate improvements, and others concluding that there is no improvement in
t.
11
The improvement in t from adopting the VAR model over the SQR model is roughly
similar in-and out-of-sample. It is also consistent with the evidence from returns and realized
variances discussed earlier.
The improvement in t of the 3/2N model over the SQR model is less impressive but still
substantial. Recall that when judging the models performance using the `11o objective func-
tions in Table 1, the most impressive return-based performance of the 3/2N model obtained when
the 1987 crash was included in the sample. Our evidence on the 3/2N model is interesting in
light of the ndings in Jones (2003) and At-Sahalia and Kimmel (2007). These papers both
estimate a CEV model using a bivariate time series of index returns and options. However, they
obtain very dierent results. Jones (2003) estimates a CEV parameter of 1.33 using a 1986-2000
sample, while At-Sahalia and Kimmels (2007) estimate for the CEV parameter is 0.65. Our
results suggest that one potential reason for these conicting results is the dierent samples used
in these studies: At-Sahalia and Kimmels sample is 1990-2003, and therefore does not include
the 1987 crash. However, it must be noted that Jones (2003) also estimates a CEV parameter of
1.17 using a 1988-2000 sample. It may thus be the case that richer option data sets are needed
in order to reliably estimate the model parameters.
When comparing the option-based NLSIS estimates in Table 4 with the return-based estimates
in Table 1, we nd that the i is generally smaller, indicating a slower mean-reversion of variance
when options are driving the parameters. The o parameter is generally lower in Table 4, with the
SQR model as the notable exception. The volatility of variance parameter o is generally lower
in Table 4, the exception again being the SQR model. The volatility risk premium parameter `
is generally small but positive for all models as expected. Finally, the correlation coecient, j,
is slightly smaller (in absolute value) when estimated using both options and returns.
Not surprisingly, when compared with the NLSIS-based in-sample IVRMSEs in Table 4, the
MLIS-based errors for Wednesday call data in Table 3 are clearly considerably larger, except for
the 3/2 model where the dierence is negligible. In most cases, the IVRMSEs for Wednesday puts
and Thursday calls in Table 3 are also larger than the corresponding out-of-sample NLSIS-based
IVRMSEs in Table 4.
Finally, the footnote to Table 4 reports the IVRMSE from two benchmark models. First,
the simple Black-Scholes model where volatility is kept constant at the average implied volatility
across all options and across the entire sample period. The resulting IVRMSE is 4.95% for
the Wednesday call sample, 5.52% for the Wednesday put sample, and 5.22% for the Thursday
sample. Note that, not surprisingly, all the SV models in Table 4 outperform the simple Black-
Scholes model by a wide margin both in and out of sample. The second benchmark is the
so-called ad-hoc Black-Scholes model where the implied volatility for each option is set to the
average implied volatility across options on the same weekday of the previous week. This ad-
11
See for instance Bates (2000), Pan (2002) and Eraker (2004). In recent work, using a very dierent empirical
setup, Broadie, Chernov and Johannes (2007) nd large improvements in-sample when including Poisson jumps.
19
hoc model has an IVRMSE of 2.99% for the Wednesday call sample, 3.52% for the Wednesday
put sample, and 3.06% for the Thursday sample. Thus, only the VAR model outperforms this
benchmark in all three samples. This result may be surprising but note that the ad-hoc model
allows for the estimation of 9*52 = 468 parameters versus only ve parameters in the SV models.
These results are not easily comparable with other recent studies as the weekly estimated ad-hoc
benchmark is not often compared with SV models estimated on a long sample. Heston and Nandi
(2000) provide a similar benchmark, but their GARCH models are estimated on much shorter
samples.
4.6 Decomposing the Option Valuation Errors
To provide more insight into the models performance, we decompose the in-sample option val-
uation errors along several dimensions. The out-of-sample results yield very similar conclusions
regarding the models relative performance, and are thus omitted from the tables and gures.
Option valuation models can fail for several reasons: they may imply a biased estimate of the
latent volatility variable, which leads to biases in the valuation of at-the-money options. Alter-
natively, one volatility model may outperform another in the maturity dimension if it provides
better estimates of the term structure of volatility, or in the moneyness dimension. We investi-
gate all these model aspects, and we also investigate the models performance as a function of
the level of market volatility, as measured by the VIX.
Figures 9 and 10 plot the performance of the models in valuing at-the-money options over
time. Figure 9 plots the weekly IVRMSE for at-the-money options, dened as options with
moneyness o,A, between 0.975 and 1.025. The 3/2 model and particularly the VAR model
display a relatively low and stable IVRMSE across time. The IVRMSE plots for the SQR model
and the three models with nonlinear drifts exhibit many more spikes.
The MSE-based objective function can be viewed as a sum of error variance and bias squared.
Figure 10 therefore shows the weekly IV bias (average data IV less average model IV) for the
at-the-money options. It is clear that some of the spikes in Figure 9 are driven at least partly
by spikes in the bias. This is particularly the case for the models with nonlinear drift. In
addition to the high-frequency movements evident in Figure 10, it also appears that all models
display a somewhat persistent bias. This would indicate that volatility models with more exible
autocorrelation functions, as provided for example by the multiple volatility component models
in Bates (2000), are needed.
In Table 5, we report the in-sample IVRMSE by moneyness, maturity, and volatility level
as measured by the VIX. Figure 11 depicts the IVRMSE results by moneyness, maturity, and
volatility categories, for the three models with linear variance drift, c = 0, in the left column,
and the three models with nonlinear drift, c = 1, in the right column. The impressive overall
performance of the VAR model (denoted by x) clearly originates from good performance in
virtually all moneyness, maturity and volatility categories. The benchmark SQR model (denoted
by o) on the other hand generally performs worst or near worst in all moneyness, maturity and
volatility categories. The only exception is for the lowest VIX category, where the benchmark
SQR model performs relatively well.
In general, the dierences between models are larger within the class of models with linear
drift (the left column of Figure 11). The right column in Figure 11 indicates that the three models
with nonlinear drift perform quite similarly across the moneyness and maturity dimension, and
20
that in the volatility dimension the models only dier substantially for the highest VIX category.
Comparing the two bottom panels, we see that the nonlinear variance drift improves the t
across the three diusion specications for the lowest VIX category, but it actually hurts the
performance of all three models for the highest VIX category.
The IVRMSE is generally increasing in moneyness for most models, which may be partly
driven by the fact that the average implied volatility is increasing in moneyness (see Table
2). Table 5 indicates that the VAR model performs better than average in all six moneyness
categories. The VAR model performs best at pricing deep-in-the money calls. The 3/2N model
is also better than average for all categories, and is particularly good at valuing out-of-the-
money calls. The SQR model performs below average in all moneyness categories, and it does
particularly poorly for in-the-money options.
Note also that the IVRMSE tends to decrease with maturity for all models. This is only
partially explained by the average implied volatility being larger for short-term options, because
these dierences are small (see Table 2). Panel B of Table 5 indicates that the VAR model
outperforms the other ve models for the most actively traded options (with maturity up to 60
days), and the VAR and 3/2N models are the best performers for maturities between 60 and 180
days. For the longest maturities, the VARN model performs well. The VAR and the 3/2N models
are better than average performers in all maturity categories. The SQR model performs worse
than average for all maturity categories and is the worst model in the four medium-maturity
categories.
The moneyness and maturity patterns clearly indicate that while non-ane volatility dynam-
ics may lead to improved model performance, important biases remain. A potential solution to
remedy these bias patterns is to include return jumps in the model specication. Negative return
jumps are likely to have most eect on the valuation of short-maturity and deep-in-the-money
call options.
For the analysis by VIX level in panel C of Table 5 and the bottom panels in Figure 11, we
split the sample days into six categories sorted by the VIX level on each day of the sample. The
3/2N model is better than average for all VIX categories, and the VAR model is better than
average for all except the lowest VIX levels. The SQR model performs well for the very lowest
VIX level category, but performs poorly otherwise.
Figure 12 and Table 6 report the bias (average market price less average model price) in
percent across moneyness, maturity and volatility level categories, dened as in Table 5. Note
rst and foremost from Table 6 that the overall bias is close to zero in all models, except for the
SQR model where it is 1.06%, and for the 3/2 model, which has a bias of -0.90%.
The bias patterns in the top row of Figure 12 conrm a potentially important role for jumps
across model specications. Panel A of Table 6 indicates that the SQR model underprices all
options, and all models underprice the deep in-the-money calls. All models except for the SQR
model tend to overprice out-of-the money calls.
Panel B of Table 6 reports the bias by maturity. The SQR model underprices all maturity
categories and the 3/2 model overprices all maturity categories. The remaining models tend
to overprice short-term options and underprice long-term options but the biases are generally
small. Interestingly, the middle row of Figure 12 highlights that the bias is generally smaller in
magnitude across maturity in the models with nonlinear drift (right column), suggesting that
some of the IVRMSE patterns across maturity in Figure 11 originate from dierences in error
21
variance across models. The nonlinear drift specication particularly helps in the SQR model
where the bias is large when the drift is linear.
Panel C of Table 6 reports the bias by volatility level. The bias tends to be increasing in VIX
for all models. The dierences in bias patterns between models across VIX categories clearly
explain a signicant part of the IVRMSE VIX patterns in Figure 11. Recall that the bottom-left
panel in Figure 11 showed that the SQR model performs relatively well when VIX was at its
lowest level. The bottom-left panel in Figure 12 reveals that this result is driven partly by a
smaller bias. The bottom row in Figure 11 also showed that specifying a nonlinear variance drift
improves IVRMSE when volatility is low, but worsens IVRMSE when volatility is high. The
bottom row of Figure 12 gives at least a partial explanation: the negative bias observed at low
volatility levels across diusion specications is worst when the drift is linear. The positive bias
observed at high volatility levels across diusion specications on the other hand is worst when
the drift is nonlinear. We conclude that a more exible volatility specication may therefore be
needed.
4.7 Alternative Data and Filtering Techniques
The NLSIS estimation on option data leads to a number of interesting conclusions. The SQR
model displays substantial biases and seems misspecied. While the option valuation errors and
bias patterns for the other ve models suggest a role for alternative model specications, such as
multiple volatility components and return jumps, some of the models substantially outperform
the SQR model in most dimensions. This is particularly the case for the VAR model. These
results conrm our nding, obtained using realized volatility diagnostics and returns-based MLIS
estimation, that the SQR model is misspecied.
We now further investigate the robustness of these ndings with respect to the choice of
ltering technique and the data used in ltering. While the PF used in the MLIS and NLSIS
estimation is computationally ecient and reliable, it is of interest to see if similar results obtain
with alternative ltering techniques. Moreover, while our results are largely consistent across
the MLIS and NLSIS ltering exercises that use daily returns, it is of interest to investigate if
similar results obtain when ltering the spot volatility from alternative data.
Table 7 reports on these robustness exercises. We consider two alternative data sources for
ltering: the daily realized variance data for 1996-2004 considered in Section 2.1, and the daily
VIX data from Section 2.3 for the same period. In Table 7, we rst value the Wednesday call
options using the raw realized variance and (squared) VIX data as proxies for the spot variance.
Subsequently, we lter the latent variance from these two series using the Kalman lter. We use
the simplest possible setup to lter from the VIX and RV series, ignoring returns altogether. We
limit our analysis to the SQR, VAR and 3/2 models, because the measurement equations relating
realized volatility and VIX to the latent volatility are not known for models with nonlinear drift.
Valuation results are obtained using the NLSIS-based parameter estimates from Table 4.
When ltering volatility from realized volatility, a sensible state-space representation for the
SQR, VAR and 3/2 models is
1\
t;t+1
= 1[1\
t;t+1
j\
t
] +n
t
\
t+1
= \
t
+i(o \
t
) +o\
b
t
n
t+1
22
Using the results in At-Sahalia and Kimmel (2007), this can be written as
1\
t;t+1
= o +
_
exp(i,252) 1
(i,252)
_
(\
t
o) +n
t
(30)
\
t+1
= \
t
+i(o \
t
) +o\
b
t
n
t+1
For the VIX, we employ the following state-space representation
\ 1A
t;t+30
= o +
_
exp(30i,252) 1
(30i,252)
_
(\
t
o) +n
t
(31)
\
t+1
= \
t
+i(o \
t
) +o\
b
t
n
t+1
Table 7 presents the IVRMSEs, as well as the IVRMSE ratio between the VAR and 3/2
models and the SQR model. We also report the DM test for the VAR and 3/2 models against
the SQR benchmark. The main conclusion is that the results conrm that the SQR model can
be substantially improved on by either of the two non-ane models. The IVRMSEs for the 3/2
model are smaller than those for the VAR model, but in most cases the dierences are small.
In general, the IVRMSEs using the raw RV data are large compared to the MLIS- and NLSIS-
based IVRMSEs in Tables 3 and 4. This is not surprising because these raw data contain more
outliers than the variance data ltered from returns. When ltering the RV using the Kalman
lter setup in (30), the IVRMSEs are substantially lower than when using raw RV data. The
VIX data yield a lower IVRMSE than the realized volatility data and sometimes lower than the
return-based ltering. This is also not surprising, because this implementation does not enforce
consistency with the daily return on the underlying index. The IVRMSEs for the VAR and 3/2
models are similar, and both are substantially lower than the IVRMSE for the SQR model.
In summary, the results obtained using alternative data sources and ltering techniques con-
rm the misspecication of the SQR model, and indicate that non-ane models can oer sub-
stantial improvements.
5 Summary and Conclusions
This paper provides an empirical comparison of the ane SQR model of Heston (1993) with a
range of non-ane but equally parsimonious option valuation models. The main conclusion is
that non-ane specications can substantially improve upon the ane SQR model. This nding
is robust across a wide range of data sources used for volatility ltering, as well as across dierent
estimation techniques. First, the misspecication of the SQR model is suggested by the stylized
facts characterizing realized volatility data. Second, an extensive estimation exercise on index
returns conrms the presence of misspecication by analyzing measures of t. Third, non-ane
models lead to substantially smaller in-sample errors in option valuation when estimating using
a dataset consisting of long time series of cross sections of call option data. These results are
conrmed when using the estimates out-of-sample by pricing options on dierent days and by
pricing put options. Moreover, the model t obtained by valuing options using estimates from
returns also conrms that the SQR model is misspecied. Fourth, option pricing errors are also
smaller for non-ane models when using realized volatility and the VIX as a proxy for spot
23
volatility, and when using the Kalman lter to obtain spot volatility from realized volatility and
the VIX.
We therefore conclude that while the option valuation literatures focus on ane models is well
motivated, because the resulting closed-form solutions are extremely convenient, this analytical
convenience comes at a price, and non-ane models need to be studied more extensively. Our
empirical results provide some important guidance for the specication of non-ane models. The
VAR model consistently performs very well. It provides the best t for two of the three return
samples, and the second best t using the third sample. When estimating the models using
options data, it is the best performing model in- and out-of-sample. When using the returns-
based parameter estimates to value options, the ranking of the ve non-ane models depends
on the data used.
At the methodological level, this paper uses a method to estimate continuous-time option val-
uation models that has not yet been applied to this particular problem. We use particle ltering,
which is rather exible and straightforward to implement. It can easily be used to investigate
option data and underlying equity returns jointly for a wide range of objective functions as well
as for a wide range of models. In our opinion, because of its simplicity and numerical eciency,
this method is attractive compared to other methods that have been used in this literature.
Our empirical results suggest a number of important extensions. First, while non-ane mod-
els substantially outperform ane models, the biases of the ane model can also be addressed by
alternative model specications, such as multiple volatility components, and/or Poisson or Levy
jumps in returns and volatility.
12
This study analyzes non-ane specications without consider-
ing these alternatives, but ultimately it will be of interest to investigate the relative advantages of
non-ane specications, multiple volatility components, and jump processes. Second, the nding
from the returns data that the 3/2N model is particularly useful in a sample that includes the
1987 crash, indicates that another interesting avenue for future work is to focus more explicitly on
the CEV model, and particularly to investigate its performance when periodically re-estimating
the CEV parameter. Third, the analysis in Section 4.7 provides some interesting insights and
may be worth extending. In particular, model estimates obtained by ltering volatility using
returns and the cross-section of options could be compared with model estimates obtained by
ltering using returns and either the VIX or the realized volatility data or both.
6 Appendix
In this appendix, we rst study the properties of Monte Carlo based option prices. We compare
Monte Carlo prices for the Heston (1993) SQR model with prices computed via Fourier inversion
of the conditional characteristic function. We then study the nite sample properties of the MLIS
estimator, comparing it with other available estimators in the literature.
6.1 Option Price Computation by Simulation
To assess the accuracy of our Monte Carlo setup, we compute a set of analytical option prices
using Hestons (1993) model for various strike prices, moneyness and maturities. We then com-
pute Monte Carlo option prices for the same options. Using the numerical example in Heston
12
See Carr and Wu (2004) and Huang and Wu (2004) for option valuation studies using Levy processes.
24
(1993), the risk-neutral variance process is parameterized as follows
d\ = 2(.01 \ )dt +.2
p
\ dn

(32)
The correlation j is set to 0.5, and the risk-free rate is set to zero.
Figure A.1 reports the results. The Monte Carlo prices, denoted by +, oer a very good
approximation to the analytical prices, denoted by solid lines, in all cases. The gure contains two
maturities: one month (left column) and three months (right column). The gure considers three
spot variance levels: \
0
= .005, which is half the unconditional variance (top row), \
0
= .01 which
is equal to the unconditional variance (middle row), and \
0
= .02 which is twice the unconditional
variance (bottom row).
6.2 Finite Sample Performance of MLIS
We assess the nite sample performance of the relatively new MLIS estimation procedure based
on the particle lter. Following Bates (2006), who introduces a new approximate ML estima-
tion procedure, we add the MLIS estimator to the large-scale Monte Carlo study in Andersen,
Chung and Srensen (1999), henceforth ACS. ACS use the following simple SV model as a data
generating process
ln(o
t+1
) = ln(o
t
) +
_
\
t
.
t+1
(33)
ln (\
t+1
) = . +cln (\
t
) +on
t+1
(34)
where .
t+1
and n
t+1
are uncorrelated. They consider a large number of estimators including the
QML method from Harvey, Ruiz and Shephard (1994), the GMM from Andersen and Srensen
(1996), MCMC from Jacquier, Polson and Rossi (1994), as well as the EMM implementation
suggested by ACS themselves. Table A.1 reproduces the results from ACS and includes the
approximate maximum likelihood (AML) technique from Bates (2006). We of course also include
the MLIS estimation procedure from Pitt (2002) which is used in this paper. The various
estimators are compared to the benchmark (but of course unrealistic) case where the spot variance
is observed and standard ML estimation is straightforward. We report results for the sample size
1 = 2. 000 which is relevant for our subsequent empirical study.
Table A.1 reports parameter bias and root mean squared error (RMSE) for each estimator
using 500 Monte Carlo replications. The MCMC estimator has the lowest absolute bias for the
. parameter followed by the AML and the MLIS. The MLIS, MCMC and AML estimators have
the lowest absolute biases for c. The AML, EMM and MCMC estimators have the lowest biases
for the o parameter followed by the MLIS. The MCMC and MLIS have the lowest RMSE for
.. AML, MCMC and MLIS have the lowest RMSE for c, and MCMC has the lowest RMSE
for the o parameter followed by AML and MLIS. Overall, the MLIS estimator seems to perform
well. The fact that it is also easily implementable across a wide range of models and objective
functions makes it a very suitable estimation method in our large-scale empirical study.
25
References
[1] At-Sahalia, Y. (1996), Testing Continuous-Time Models of the Spot Interest Rate, Review
of Financial Studies, 9, 385-426.
[2] At-Sahalia, Y. and R. Kimmel (2007), Maximum Likelihood Estimation of Stochastic
Volatility Models, Journal of Financial Economics, 83, 413-452.
[3] At-Sahalia, Y., P. Mykland, and L. Zhang (2005), How Often to Sample a Continuous-Time
Process in the Presence of Market Microstructure Noise, Review of Financial Studies, 18,
351-416.
[4] Andersen, T.G., L. Benzoni, and J. Lund (2002), Estimating Jump-Diusions for Equity
Returns, Journal of Finance, 57, 1239-1284.
[5] Andersen, T.G., T. Bollerslev, F.X. Diebold, and H. Ebens (2001), The Distribution of
Realized Stock Return Volatility, Journal of Financial Economics, 61, 43-76.
[6] Andersen, T.G., H. Chung, and B.E. Srensen (1999), Ecient Method of Moments Esti-
mation of a Stochastic Volatility Model: A Monte Carlo Study, Journal of Econometrics,
91, 61-87.
[7] Andersen, T.G. and B.E. Srensen (1996), GMM Estimation of a Stochastic Volatility
Model: A Monte Carlo Study, Journal of Business and Economic Statistics, 14, 328-352.
[8] Bakshi, G., C. Cao, and Z. Chen (1997), Empirical Performance of Alternative Option
Pricing Models, Journal of Finance, 52, 2003-2049.
[9] Barone-Adesi, G., H. Rasmussen, and C. Ravanelli (2003), An Option Pricing Formula for
the GARCH Diusion Model, Manuscript, University of Southern Switzerland.
[10] Bates, D. (1996), Jumps and Stochastic Volatility: Exchange Rate Processes Implicit in
Deutsche Mark Options, Review of Financial Studies, 9, 69-107.
[11] Bates, D. (2000), Post-87 Crash Fears in the S&P 500 Futures Option Market, Journal of
Econometrics, 94, 181-238.
[12] Bates, D. (2003), Empirical Option Pricing: A Retrospection, Journal of Econometrics, 116,
387-404.
[13] Bates, D. (2006), Maximum Likelihood Estimation of Latent Ane Processes, Review of
Financial Studies, 19, 909-965.
[14] Benzoni, L. (2002), Pricing Options Under Stochastic Volatility: An Empirical Investigation,
Manuscript, University of Minnesota.
[15] Black, F. (1976), Studies of Stock Price Volatility Changes, in: Proceedings of the 1976
Meetings of the Business and Economic Statistics Section, American Statistical Association,
177-181.
26
[16] Black, F., and M. Scholes (1973), The Pricing of Options and Corporate Liabilities, Journal
of Political Economy, 81, 637-659.
[17] Broadie, M., Chernov, M. and M. Johannes (2007), Model Specication and Risk Premiums:
Evidence from Futures Options, Journal of Finance, 62, 1453-1490.
[18] Carr, P. and L. Wu (2004), Time-Changed Levy Processes and Option Pricing, Journal of
Financial Economics, 17, 113141.
[19] Chacko, G. and L. Viceira (2003), Spectral GMM Estimation of Continuous-Time Processes,
Journal of Econometrics, 116, 259-292.
[20] Chernov, M., A.R. Gallant, E. Ghysels, and G. Tauchen (2003), Alternative Models for
Stock Price Dynamics, Journal of Econometrics, 116, 225-257.
[21] Chernov, M. and E. Ghysels (2000), A Study Towards a Unied Approach to the Joint
Estimation of Objective and Risk Neutral Measures for the Purpose of Option Valuation,
Journal of Financial Economics, 56, 407-458.
[22] Christoersen, P. and K. Jacobs (2004), The Importance of the Loss Function in Option
Valuation, Journal of Financial Economics, 72, 291-318.
[23] Diebold, F.X. and R.S. Mariano (1995), Comparing Predictive Accuracy, Journal of Business
and Economic Statistics, 13, 253-265.
[24] Duan, J.-C. and J.-G. Simonato (1998), Empirical Martingale Simulation for Asset Prices,
Management Science, 44, 1218-1233.
[25] Dumas, B., J. Fleming, and R. Whaley (1998), Implied Volatility Functions: Empirical
Tests, Journal of Finance, 53, 2059-2106.
[26] Efron, B. and R. Tibshirani (1993), An Introduction to the Bootstrap. Chapman and Hill.
New York.
[27] Eraker, B. (2001), MCMC Analysis of Diusion Models with Application to Finance, Journal
of Business and Economic Statistics, 19, 177-191.
[28] Eraker, B. (2004), Do Stock Prices and Volatility Jump? Reconciling Evidence from Spot
and Option Prices, Journal of Finance, 59, 1367-1403.
[29] Eraker, B., M. Johannes, and N. Polson (2003) The Role of Jumps in Returns and Volatility,
Journal of Finance, 58, 1269-1300.
[30] Gordon, N.J., D.J. Salmond, and A.F.M. Smith (1993), A Novel Approach to nonlinear/non-
Gaussian Bayesian state estimation, IEE-Proceedings F, 140, 107-133.
[31] Granger, C. (1969), Prediction with a generalized cost of error function, Operations Research
Quarterly, 20, 199-207.
27
[32] Harvey, A., E. Ruiz, and N. Shephard (1994), Multivariate Stochastic Variance Models,
Review of Economic Studies, 61, 247-264.
[33] Heston, S. (1993), A Closed-Form Solution for Options with Stochastic Volatility with Ap-
plications to Bond and Currency Options, Review of Financial Studies, 6, 327-343.
[34] Heston, S. (1997), A Simple New Formula for Options with Stochastic Volatility, Manuscript,
John M. Olin School of Business, Washington University.
[35] Heston, S. and S. Nandi (2000), A Closed-Form GARCH Option Pricing Model, Review of
Financial Studies, 13, 585-626.
[36] Huang, J.-Z. and L. Wu (2004), Specication Analysis of Option Pricing Models Based on
Time-Changed Levy Processes, Journal of Finance, 59, 14051439.
[37] Jacquier, E., N.G. Polson, and P.E. Rossi (1994), Bayesian Analysis of Stochastic Volatility
Models, Journal of Business and Economic Statistics, 12, 1-19.
[38] Johannes, M., N. Polson, N. and J. Stroud (2008), Learning about Jumps and Stochas-
tic Volatility: Filtering Stochastic Dierential Equations with Jumps, Review of Financial
Studies, forthcoming.
[39] Jones, C. (2003), The Dynamics of Stochastic Volatility: Evidence from Underlying and
Options Markets, Journal of Econometrics, 116, 181-224.
[40] Karlin, S. and H.M. Taylor (1981), A Second Course in Stochastic Processes, Academic
Press, San Diego, CA.
[41] Lewis, A. (2000), Option Valuation Under Stochastic Volatility, Finance Press, Newport
Beach, CA.
[42] Pan, J. (2002), The Jump-Risk Premia Implicit in Options: Evidence from an Integrated
Time-Series Study, Journal of Financial Economics, 63, 3-50.
[43] Pitt, M. (2002), Smooth Particle Filters for Likelihood Evaluation and Maximization, Man-
uscript, University of Warwick.
[44] Pitt, M., and N. Shephard (1999), Filtering via Simulation: Auxiliary Particle Filters,
Journal of the American Statistical Association, 94, 590-599.
[45] Sabanis, S. (2003), Stochastic Volatility and the Mean Reverting Process, The Journal of
Futures Markets, 23, 3347.
[46] Smith, A. and A. Gelfand (1992), Bayesian Statistics Without Tears: A Sampling-
Resampling Perspective, American Statistician, 46, 84-88.
[47] Zhang, L., P.A. Mykland, and Y. At-Sahalia (2005), A Tale of Two Time Scales: De-
termining Integrated Volatility with Noisy High-Frequency Data, Journal of the American
Statistical Association, 100, 1394-1411.
28
Figure 1: Diagnostics of Realized Volatility and Log Realized Variance. 1996-2004.
-4 -2 0 2 4
-4
-2
0
2
4
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Realized Volatility
0 2 4 6
-6
-4
-2
0
2
4
6
Volatility Level
V
o
l
a
t
i
l
i
t
y

C
h
a
n
g
e
s
0 2 4 6
0
1
2
3
4
Volatility Level
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
-4 -2 0 2 4
-4
-2
0
2
4
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Log Realized Variance
-2 0 2 4
-4
-2
0
2
4
Log Variance
L
o
g

V
a
r
i
a
n
c
e

C
h
a
n
g
e
s
-2 0 2 4
0
1
2
3
4
Log Variance
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
Notes to gure: Using realized volatilities (left column) and log realized variances (right column)
from 1996 through 2004, the top panels plot the quantiles of the daily changes against quantiles
from the normal distribution. The middle panels scatter plot the daily changes against the daily
levels. The bottom panels scatter plot the absolute daily changes against the daily levels and
show an OLS regression line for reference.
29
Figure 2: Diagnostics of Option Implied Volatility (VIX) and Log VIX. 1996-2004.
-5 0 5
-5
0
5
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
VIX
10 20 30 40 50
-10
-5
0
5
10
VIX Level
V
I
X

C
h
a
n
g
e
s
10 20 30 40 50
0
2
4
6
8
10
VIX Level
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
-5 0 5
-5
0
5
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Log VIX
2 2.5 3 3.5 4
-0.2
-0.1
0
0.1
0.2
0.3
Log VIX
L
o
g

V
I
X

C
h
a
n
g
e
s
2 2.5 3 3.5 4
0
0.1
0.2
0.3
0.4
Log VIX
A
b
s
o
l
u
t
e

V
I
X

C
h
a
n
g
e
s
Notes to gure: Using VIX (left column) and log VIX (right column) from 1996 through 2004, the
top panels plot the quantiles of the daily changes against quantiles from the normal distribution.
The middle panels scatter plot the daily changes against the daily levels. The bottom panels
scatter plot the absolute daily changes against the daily levels and show an OLS regression line
for reference.
30
Figure 3: Drift and Diusion Functions for Various SV Models.
0 0.02 0.04 0.06 0.08 0.1 0.12
-0.5
0
0.5
SQR Model: a=0, b=1/2.
Drift
Diffusion
0 0.02 0.04 0.06 0.08 0.1 0.12
-0.5
0
0.5
SQRN Model: a=1, b=1/2.
Drift
Diffusion
0 0.02 0.04 0.06 0.08 0.1 0.12
-0.5
0
0.5
VAR Model: a=0, b=1.
Drift
Diffusion
0 0.02 0.04 0.06 0.08 0.1 0.12
-0.5
0
0.5
VARN Model: a=1, b=1.
Drift
Diffusion
0 0.02 0.04 0.06 0.08 0.1 0.12
-0.5
0
0.5
3/2 Model: a=0, b=3/2.
Spot Variance, V
Drift
Diffusion
0 0.02 0.04 0.06 0.08 0.1 0.12
-0.5
0
0.5
3/2N Model: a=1, b=3/2.
Spot Variance, V
Drift
Diffusion
Notes to gure: The solid lines denote the drift function i\
a
(o \ ) plotted against the level of
spot variance, \ . The dashed lines denote the diusion function o\
b
plotted against the level of
spot variance. The parameter estimates are from Table 1 using daily returns from 1996 through
2004.
31
Figure 4: Spot Volatility Paths for Various SV Models. 1996-2004.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
SQR Model: a=0, b=1/2.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
SQRN Model: a=1, b=1/2.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
VAR Model: a=0, b=1.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
VARN Model: a=1, b=1.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
3/2 Model: a=0, b=3/2.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
3/2N Model: a=1, b=3/2.
Notes to gure: For each SV model of the form d\ = i\
a
(o \ )dt + o\
b
dn, we plot the
annualized daily ltered spot volatility path,
_

\
t
during 1996-2004. The parameters are from
the MLIS estimation on daily S&P500 returns from 1996 through 2004 as reported in Table 1.
32
Figure 5: Conditional Volatility of Variance Paths for Various SV Models. 1996-2004.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
SQR Model: a=0, b=1/2.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
SQRN Model: a=1, b=1/2.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
VAR Model: a=0, b=1.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
VARN Model: a=1, b=1.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
3/2 Model: a=0, b=3/2.
1996 1998 2000 2002 2004
0
0.1
0.2
0.3
0.4
0.5
0.6
3/2N Model: a=1, b=3/2.
Notes to gure: For each SV model of the form d\ = i\
a
(o \ )dt +o\
b
dn, we plot the annu-
alized daily conditional volatility of variance path dened as the square root of the conditional
variance of the variance of returns, o

\
b
t
. The parameters are from the MLIS estimation on daily
S&P500 returns from 1996 through 2004 as reported in Table 1. The vertical axes have been
truncated for the 3/2 diusions in order to facilitate comparisons with the other models.
33
Figure 6: Diagnostics of Spot Volatility from SQR Model. 1996-2004.
-6 -4 -2 0 2 4 6
-5
0
5
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Spot Volatility
0 0.1 0.2 0.3 0.4 0.5
-0.05
0
0.05
0.1
Volatility Level
V
o
l
a
t
i
l
i
t
y

C
h
a
n
g
e
s
0 0.1 0.2 0.3 0.4 0.5
0
0.02
0.04
0.06
0.08
0.1
Volatility Level
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
-6 -4 -2 0 2 4 6
-5
0
5
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Log Spot Variance
-6 -5 -4 -3 -2 -1
-1
-0.5
0
0.5
1
Log Variance
L
o
g

V
a
r
i
a
n
c
e

C
h
a
n
g
e
s
-6 -5 -4 -3 -2 -1
0
0.2
0.4
0.6
0.8
1
Log Variance
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
Notes to gure: Using the ltered

\
1=2
t
(left column) and ln
_

\
t
_
(right column) from 1996 through
2004, the top panels plot the quantiles of the daily changes against quantiles from the normal
distribution. The middle panels scatter plot the daily changes against the daily levels. The
bottom panels scatter plot the absolute daily changes against the daily levels and show an OLS
regression line for reference.
34
Figure 7: Diagnostics of Spot Volatility from VAR Model. 1996-2004.
-6 -4 -2 0 2 4 6
-5
0
5
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Spot Volatility
0 0.1 0.2 0.3 0.4 0.5
-0.05
0
0.05
0.1
Volatility Level
V
o
l
a
t
i
l
i
t
y

C
h
a
n
g
e
s
0 0.1 0.2 0.3 0.4 0.5
0
0.02
0.04
0.06
0.08
0.1
Volatility Level
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
-6 -4 -2 0 2 4 6
-5
0
5
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Log Spot Variance
-6 -5 -4 -3 -2 -1
-1
-0.5
0
0.5
1
Log Variance
L
o
g

V
a
r
i
a
n
c
e

C
h
a
n
g
e
s
-6 -5 -4 -3 -2 -1
0
0.2
0.4
0.6
0.8
1
Log Variance
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
Notes to gure: Using the ltered

\
1=2
t
(left column) and ln
_

\
t
_
(right column) from 1996 through
2004, the top panels plot the quantiles of the daily changes against quantiles from the normal
distribution. The middle panels scatter plot the daily changes against the daily levels. The
bottom panels scatter plot the absolute daily changes against the daily levels and show an OLS
regression line for reference.
35
Figure 8: Diagnostics of Spot Volatility from 3/2 Model. 1996-2004.
-6 -4 -2 0 2 4 6
-5
0
5
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Spot Volatility
0 0.1 0.2 0.3 0.4 0.5
-0.1
-0.05
0
0.05
0.1
Volatility Level
V
o
l
a
t
i
l
i
t
y

C
h
a
n
g
e
s
0 0.1 0.2 0.3 0.4 0.5
0
0.02
0.04
0.06
0.08
0.1
Volatility Level
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
-6 -4 -2 0 2 4 6
-5
0
5
Standard Normal Quantiles
D
a
t
a

Q
u
a
n
t
i
l
e
s
Log Spot Variance
-6 -5 -4 -3 -2 -1
-1
-0.5
0
0.5
1
Log Variance
L
o
g

V
a
r
i
a
n
c
e

C
h
a
n
g
e
s
-6 -5 -4 -3 -2 -1
0
0.2
0.4
0.6
0.8
1
Log Variance
A
b
s
o
l
u
t
e

C
h
a
n
g
e
s
Notes to gure: Using the ltered

\
1=2
t
(left column) and ln
_

\
t
_
(right column) from 1996 through
2004, the top panels plot the quantiles of the daily changes against quantiles from the normal
distribution. The middle panels scatter plot the daily changes against the daily levels. The
bottom panels scatter plot the absolute daily changes against the daily levels and show an OLS
regression line for reference.
36
Figure 9: Weekly IVRMSE for At-The-Money Options.
1996 1998 2000 2002 2004
0
5
10
SQR Model: a=0, b=1/2.
I
V
R
M
S
E
1996 1998 2000 2002 2004
0
5
10
SQRN Model: a=1, b=1/2.
I
V
R
M
S
E
1996 1998 2000 2002 2004
0
5
10
VAR Model: a=0, b=1.
I
V
R
M
S
E
1996 1998 2000 2002 2004
0
5
10
VARN Model: a=1, b=1.
I
V
R
M
S
E
1996 1998 2000 2002 2004
0
5
10
3/2 Model: a=0, b=3/2.
I
V
R
M
S
E
1996 1998 2000 2002 2004
0
5
10
3/2N Model: a=1, b=3/2.
I
V
R
M
S
E
Notes to gure: For each model we plot the weekly IVRMSE for at-the-money options, dened
as options with moneyness, o,A, between 0.975 and 1.025. The parameter estimates are from
Table 4.
37
Figure 10: Weekly IV Bias for At-The-Money Options.
1996 1998 2000 2002 2004
-5
0
5
10
SQR Model: a=0, b=1/2.
I
V

B
i
a
s
1996 1998 2000 2002 2004
-5
0
5
10
SQRN Model: a=1, b=1/2.
I
V

B
i
a
s
1996 1998 2000 2002 2004
-5
0
5
10
VAR Model: a=0, b=1.
I
V

B
i
a
s
1996 1998 2000 2002 2004
-5
0
5
10
VARN Model: a=1, b=1.
I
V

B
i
a
s
1996 1998 2000 2002 2004
-5
0
5
10
3/2 Model: a=0, b=3/2.
I
V

B
i
a
s
1996 1998 2000 2002 2004
-5
0
5
10
3/2N Model: a=1, b=3/2.
I
V

B
i
a
s
Notes to gure: For each model we plot the weekly IV bias for at-the-money options, dened as
those with moneyness, o,A, between 0.975 and 1.025. The parameter estimates are from Table
4.
38
Figure 11: IVRMSE by Moneyness, Maturity and VIX Level.
.94 .97 1 1.03 1.06
2
3
4
5
Linear Drift, a=0
I
V
R
M
S
E
Moneyness, S/X
SQR
VAR
3/2
.94 .97 1 1.03 1.06
2
3
4
5
Nonlinear Drift, a=1
I
V
R
M
S
E
Moneyness, S/X
SQRN
VARN
3/2N
30 60 90 120 180
2
3
4
5
I
V
R
M
S
E
Days to Maturity
30 60 90 120 180
2
3
4
5
I
V
R
M
S
E
Days to Maturity
15 20 25 30 35
2
3
4
5
6
I
V
R
M
S
E
VIX Level
15 20 25 30 35
2
3
4
5
6
I
V
R
M
S
E
VIX Level
Notes to gure: For each model we plot the in-sample IVRMSE by moneyness (top row), maturity
(middle row), and volatility level as measured by the VIX (bottom row). The left column shows
the models with linear drift and the right column shows the models with nonlinear drift in spot
variance. The parameters estimates are from Table 4.
39
Figure 12: IV Bias by Moneyness, Maturity and VIX Level.
.94 .97 1 1.03 1.06
-4
-2
0
2
4
Linear Drift, a=0
I
V

B
i
a
s
Moneyness, S/X
SQR
VAR
3/2
.94 .97 1 1.03 1.06
-4
-2
0
2
4
Nonlinear Drift, a=1
I
V

B
i
a
s
Moneyness, S/X
SQRN
VARN
3/2N
30 60 90 120 180
-4
-2
0
2
4
I
V

B
i
a
s
Days to Maturity
30 60 90 120 180
-4
-2
0
2
4
I
V

B
i
a
s
Days to Maturity
15 20 25 30 35
-4
-2
0
2
4
I
V

B
i
a
s
VIX Level
15 20 25 30 35
-4
-2
0
2
4
I
V

B
i
a
s
VIX Level
Notes to gure: For each model we plot the in-sample IV bias by moneyness (top row), maturity
(middle row), and volatility level as measured by the VIX (bottom row). The left column shows
the models with linear drift and the right column shows the models with nonlinear drifts in spot
variance. The parameter estimates are from Table 4.
40
Figure A.1: Analytical (Fourier Inversion) and Monte Carlo Prices for the Heston (1993) Model.
0.94 0.96 0.98 1 1.02 1.04 1.06
0
2
4
6
L
o
w

S
p
o
t

V
a
r
i
a
n
c
e
One-Month Maturity
Fourier Inversion
Monte Carlo
0.94 0.96 0.98 1 1.02 1.04 1.06
0
2
4
6
Three-Month Maturity
L
o
w

S
p
o
t

V
a
r
i
a
n
c
e
0.94 0.96 0.98 1 1.02 1.04 1.06
0
2
4
6
M
e
d
i
u
m

S
p
o
t

V
a
r
i
a
n
c
e
One-Month Maturity
0.94 0.96 0.98 1 1.02 1.04 1.06
0
2
4
6
M
e
d
i
u
m

S
p
o
t

V
a
r
i
a
n
c
e
Three-Month Maturity
0.94 0.96 0.98 1 1.02 1.04 1.06
0
2
4
6
H
i
g
h

S
p
o
t

V
a
r
i
a
n
c
e
One-Month Maturity
Moneyness
0.94 0.96 0.98 1 1.02 1.04 1.06
0
2
4
6
H
i
g
h

S
p
o
t

V
a
r
i
a
n
c
e
Three-Month Maturity
Moneyness
Notes to gure: We rst compute option prices from Hestons (1993) model using the Fourier
inversion technique (solid) and then using Monte Carlo simulation (+). We consider dierent
moneyness (o,A) on the horizontal axis. We consider two maturities: one month (left column)
and three months (right column). We consider three spot variance levels: half the unconditional
variance (top row), equal to the unconditional variance (middle row) and twice the unconditional
variance (bottom row).
41
Name a b 1996-2004 1989-2004 1985-2004 Mean StdDev Skewness Kurtosis
SQR 0 1/2 6.5200 0.0352 0.4601 -0.7710 7,064.7 13,359.3 16,595.0 17.698 5.166 0.487 0.086
1.1096 0.0026 0.0309 0.0375
SQRN 1 1/2 100.0291 0.0457 0.3425 -0.7527 7,045.1 13,328.9 16,532.2 17.495 5.172 0.043 -0.411
14.4534 0.0038 0.0193 0.0234
VAR 0 1 3.9248 0.0408 2.7790 -0.7876 7,074.5 13,372.5 16,631.7 17.673 6.146 1.302 2.120
1.1392 0.0067 0.1949 0.0345
Model Filtered Volatility, 1996-2004
Standard Errors:
Standard Errors:
Table 1: Parameter Estimates from Maximum Likelihood Importance Sampling on Returns.
Parameter Estimates for 1996-2004
Standard Errors:
MLIS Objective Values
VARN 1 1 133.9347 0.0560 2.4188 -0.7559 7,066.1 13,362.5 16,615.4 17.708 5.569 0.581 0.316
25.3311 0.0053 0.1654 0.0417
3/2 0 3/2 1.0852 0.0633 11.9534 -0.7411 7,064.9 13,352.9 16,625.8 17.294 5.863 1.989 6.341
0.8260 0.0351 0.9125 0.0432
3/2N 1 3/2 60.1040 0.0837 12.4989 -0.7591 7,068.8 13,362.9 16,638.4 17.683 5.893 1.616 4.659
23.9651 0.0247 0.8575 0.0426 Standard Errors:
Notes: For each model, we estimate the parameters using Maximum Likelihood Importance Sampling (MLIS) on daily returns. The reported
parameter estimates use data from J anuary 4, 1996 to December 31, 2004. The return drift parameter is fixed at the sample average return of
0.091 for all models. The parameters are reported in annual units. Standard errors are computed using the outer product of the gradient at the
optimal parameter values. The MLIS objective value is reported for estimations on longer samples as well. The first four moments of the
distribution of the annualized filtered volatility are reported in the last four columns of the table. Kurtosis is reported in excess of three.
Standard Errors:
Standard Errors:
S/X<0.975 0.975<S/X<1 1<S/X<1.025 S/X>1.025 All
Number of Contracts 5,761 3,859 3,077 3,809 16,506
Average Call Price 29.73 37.22 47.05 78.85 46.05
Average Implied Volatility 19.35 19.36 20.13 22.66 20.26
Average Bid-Ask Spread 1.43 1.54 1.74 1.83 1.60
DTM<30 30<DTM<90 90<DTM<180 DTM>180 All
Number of Contracts 2,552 7,829 2,992 3,133 16,506
Average Price 34.82 40.09 50.83 65.51 46.05
Average Implied Volatility 20.48 20.27 20.18 20.13 20.26
Average Bid-Ask Spread 1.39 1.55 1.67 1.85 1.60
S/X<0.975 0.975<S/X<1 1<S/X<1.025 S/X>1.025 All
Number of Contracts 5,707 4,639 4,111 8,578 23,035
Average Price 80.14 38.41 28.88 21.81 40.87
Average Implied Volatility 20.90 19.44 20.59 24.26 21.80
Table 2: S&P500 Index Option Data. 1996-2004.
Panel A. Option Data Characteristics by Moneyness and Maturity. Wednesday Call Data.
Panel B. Option Data Characteristics by Moneyness and Maturity. Wednesday Put Data.
g p y
Average Bid-Ask Spread 1.95 1.66 1.42 1.24 1.53
DTM<30 30<DTM<90 90<DTM<180 DTM>180 All
Number of Contracts 3,679 9,979 4,565 4,812 23,035
Average Price 30.80 33.94 44.43 59.55 40.87
Average Implied Volatility 21.82 21.52 22.15 22.05 21.80
Average Bid-Ask Spread 1.32 1.46 1.59 1.79 1.53
S/X<0.975 0.975<S/X<1 1<S/X<1.025 S/X>1.025 All
Number of Contracts 4,695 3,775 3,072 3,848 15,390
Average Price 30.29 37.29 46.59 78.52 47.32
Average Implied Volatility 19.26 19.33 20.04 22.77 20.31
Average Bid-Ask Spread 1.40 1.53 1.74 1.87 1.62
DTM<30 30<DTM<90 90<DTM<180 DTM>180 All
Number of Contracts 2,407 6,541 2,717 2,350 15,390
Average Price 34.11 40.32 53.99 77.52 47.32
Average Implied Volatility 20.74 20.14 20.31 20.49 20.31
Average Bid-Ask Spread 1.39 1.56 1.74 1.95 1.62
Notes: We use Wednesday closing call option data, Wednesday closing put option data, and Thursday closing call
option data from OptionMetrics from J anuary 1, 1996 through December 31, 2004.
Panel C. Option Data Characteristics by Moneyness and Maturity. Thursday Call Data.
Name a b IVRMSE $RMSE OSRMSE IVRMSE $RMSE OSRMSE IVRMSE $RMSE OSRMSE
SQR 0 1/2 4.113 8.923 8.321 5.248 9.478 8.893 4.306 8.839 8.243
SQRN 1 1/2 3.487 6.485 5.873 4.062 6.684 6.101 3.746 6.719 6.105
Ratio 0.848 0.727 0.706 0.774 0.705 0.686 0.870 0.760 0.741
DM Test -2.363 -4.617 -4.644 -4.305 -5.153 -5.157 -1.849 -3.825 -3.888
VAR 0 1 3.486 6.904 6.287 3.568 6.705 6.105 3.630 6.980 6.356
Ratio 0.848 0.774 0.756 0.680 0.707 0.687 0.843 0.790 0.771
DM Test -1.921 -2.997 -3.033 -4.325 -3.814 -3.859 -1.730 -2.733 -2.802
VARN 1 1 3629 6869 6238 4093 6788 6177 3877 7062 6424
Wednesday Put Data
Table 3: Option Valuation Root Mean Squared Errors Using MLIS Estimates. 1996-2004.
Model Wednesday Call Data Thursday Call Data
VARN 1 1 3.629 6.869 6.238 4.093 6.788 6.177 3.877 7.062 6.424
Ratio 0.882 0.770 0.750 0.780 0.716 0.695 0.900 0.799 0.779
DM Test -1.531 -2.993 -3.056 -3.454 -3.609 -3.681 -1.209 -2.538 -2.630
3/2 0 3/2 3.275 6.514 5.918 3.994 7.140 6.575 3.517 6.726 6.127
Ratio 0.796 0.730 0.711 0.761 0.753 0.739 0.817 0.761 0.743
DM Test -3.819 -5.191 -5.175 -5.558 -5.817 -5.801 -3.326 -4.780 -4.808
3/2N 1 3/2 3.521 6.875 6.260 3.904 6.558 5.958 3.767 7.023 6.399
Ratio 0.856 0.770 0.752 0.744 0.692 0.670 0.875 0.794 0.776
DM Test -2.050 -3.093 -3.135 -4.576 -4.635 -4.669 -1.700 -2.775 -2.850
Notes: We use the return-based parameters from Table 1 to compute option prices. This table reports the implied volatility root
mean squared error (IVRMSE), the dollar root mean squared error ($RMSE) and the out-of-spread root mean squared error
(OSRMSE). The DM Test refers to the test value in the Diebold-Mariano test that the model's average weekly mean squared error
equals that of the benchmark (SQR) model. The boldface type indicates significance at the 5% level based on the asymptotic
normal distribution.
Name a b Name a b In Sample
Calls Puts Calls
SQR 0 1/2 3.1146 0.0523 0.5826 9.00E-05 -0.6520 SQR 0 1/2 3.317 4.204 3.567
1.50E-03 3.13E-05 1.27E-04 2.16E-03 2.41E-04 Ratio 1.000 1.000 1.000
SQRN 1 1/2 67.8964 0.0340 0.1747 1.14E-03 -0.6069 SQRN 1 1/2 3.174 4.235 3.418
3.94E-02 6.26E-06 7.42E-05 1.47E-03 3.16E-04 Ratio 0.957 1.007 0.958
DM Test -0.609 0.148 -0.441
VAR 0 1 2.4730 0.0272 1.1884 4.37E-03 -0.7116 VAR 0 1 2.855 3.399 2.974
1.12E-03 6.84E-06 3.20E-04 1.17E-03 2.06E-04 Ratio 0.861 0.808 0.834
DM Test -3.178 -4.560 -3.067
VARN 1 1 644378 00367 11214 193E-03 -06749 VARN 1 1 3000 3992 3243
Standard Errors
Table 4: NLSIS Estimates and Implied Volatility Root Mean Squared Error (IVRMSE). 1996-2004.
Model
Standard Errors
NLSIS Estimation on Option Implied Volatilities
Standard Errors
Out-of-Sample
Option IVRMSE Model
VARN 1 1 64.4378 0.0367 1.1214 1.93E-03 -0.6749 VARN 1 1 3.000 3.992 3.243
3.95E-02 1.91E-05 4.36E-04 4.94E-02 2.50E-04 Ratio 0.905 0.950 0.909
DM Test -2.045 -1.380 -1.725
3/2 0 3/2 1.5284 0.0336 7.9501 7.91E-04 -0.7169 3/2 0 3/2 3.273 3.884 3.419
2.22E-03 3.65E-05 2.94E-03 2.15E-03 2.10E-04 Ratio 0.987 0.924 0.958
DM Test -0.147 -1.580 -0.376
3/2N 1 3/2 50.9140 0.0388 6.2593 3.36E-04 -0.6854 3/2N 1 3/2 2.961 4.064 3.187
4.08E-02 2.68E-05 2.41E-03 5.13E-02 2.35E-04 Ratio 0.893 0.967 0.893
DM Test -2.689 -1.193 -2.328
Notes: We estimate the model parameters using NLSIS on the 16,506 Wednesday closing call option quotes observed from J anuary 4, 1996
to December 31, 2004. The return drift parameter is fixed at the sample average return of 0.091 for all models. Standard errors are
computed using the outer product of the gradient at the optimal parameter values. Out-of-Sample refers to the 15,390 Thursday closing call
option prices and the 23,035 Wednesday closing put option prices observed during the same period. The Black-Scholes benchmark yields
an RMSE of 4.95% for the Wednesday call sample, 5.52% for the Wednesday put sample, and 5.22% for the Thursday call sample. The ad-
hoc Black-Scholes benchmark uses the average IV for the same weekday of the previous week and has an RMSE of 2.99% for the
Wednesday call sample, 3.52% for the Wednesday put sample, and 3.06% for the Thursday call sample. The Diebold-Mariano test is
computed on the weekly mean squared errors. Bold type indicates significance at the 5% level.
Standard Errors
Standard Errors
Standard Errors
Name a b S/X<0.94 .94<S/X<.97 .97<S/X<1 1<S/X<1.03 1.03<S/X<1.06 S/X>1.06
SQR 0 1/2 2.7708 3.0377 3.0653 3.2044 3.9611 4.5433
SQRN 1 1/2 2.6416 2.9388 2.9554 2.9341 3.6137 4.6939
VAR 0 1 2.5372 2.7769 2.8047 2.7866 3.1183 3.4327
VARN 1 1 2.5164 2.7456 2.7541 2.7791 3.4342 4.4987
3/2 0 3/2 3.0398 3.1661 3.1407 3.0916 3.5087 4.2378
3/2N 1 3/2 2.5070 2.7100 2.7004 2.7449 3.3742 4.4609
2.6688 2.8958 2.9034 2.9234 3.5017 4.3112
Name a b DTM<30 30<DTM<60 60<DTM<90 90<DTM<120 120<DTM<180 DTM>180
SQR 0 1/2 3.9279 3.5115 3.2255 2.7994 2.9805 2.8364
SQRN 1 1/2 4.0059 3.2621 3.0529 2.6060 2.8723 2.6717
VAR 0 1 3.3337 2.9317 2.7181 2.4162 2.6549 2.6628
VARN 1 1 3.7828 3.1018 2.8216 2.4561 2.6586 2.5909
3/2 0 3/2 3.9129 3.3526 3.1816 2.6327 2.8756 3.0640
Table 5: In-Sample IVRMSE (%) by Moneyness, Maturity, and VIX Level. 1996-2004.
Model
Average
Model
Panel A. IVRMSE by Moneyness
Panel B. IVRMSE by Maturity
3/2N 1 3/2 3.6968 3.0469 2.7192 2.4180 2.6546 2.6679
3.7767 3.2011 2.9532 2.5547 2.7827 2.7490
Name a b VIX<15 15<VIX<20 20<VIX<25 25<VIX<30 30<VIX<35 VIX>35
SQR 0 1/2 2.6406 2.8111 3.1920 3.9811 4.2656 3.9348
SQRN 1 1/2 2.3528 2.8311 3.1742 2.9745 3.4200 6.0664
VAR 0 1 3.3287 2.3612 2.8041 3.1968 3.3560 3.2053
VARN 1 1 2.3876 2.5885 3.0512 3.0804 3.2171 5.0327
3/2 0 3/2 4.1800 2.8601 3.0049 3.6906 3.8623 3.7617
3/2N 1 3/2 2.7186 2.4158 2.9612 3.3071 3.4182 4.3596
2.9347 2.6446 3.0313 3.3717 3.5899 4.3934 Average
Notes: We use the NLSIS estimates from Table 4 to compute the option implied volatility root mean squared error
(IVRMSE) in percent for various moneyness, maturity, and volatility (VIX Level) bins for each model. The contracts
used in the table are for the 1996-2004 in-sample period, which consists of Wednesday closing call option quotes.
Model Panel C. IVRMSE by VIX Level
Average
Name a b S/X<0.94 .94<S/X<.97 .97<S/X<1 1<S/X<1.03 1.03<S/X<1.06 S/X>1.06 All
SQR 0 1/2 0.6125 0.7772 0.7944 1.0763 1.7144 2.3271 1.0642
SQRN 1 1/2 -0.6428 -0.7095 -0.6259 0.1095 1.3751 2.8209 0.0515
VAR 0 1 -0.4979 -0.7031 -0.6997 -0.0777 0.9097 1.8435 -0.1235
VARN 1 1 -0.6356 -0.7195 -0.5927 0.0829 1.2058 2.5779 0.0133
3/2 0 3/2 -1.5830 -1.5403 -1.4433 -0.7849 0.2487 1.3028 -0.9018
3/2N 1 3/2 -0.6449 -0.7313 -0.6219 0.0425 1.1612 2.5196 -0.0168
-0.5653 -0.6044 -0.5315 0.0748 1.1025 2.2320 0.0145
Name a b DTM<30 30<DTM<60 60<DTM<90 90<DTM<120 120<DTM<180 DTM>180 All
SQR 0 1/2 1.2362 1.3286 1.1755 1.2171 0.8340 0.4781 1.0642
SQRN 1 1/2 -0.1901 -0.1099 -0.4071 0.0451 0.6307 0.5470 0.0515
VAR 0 1 -0.5755 -0.2925 -0.5326 0.1422 0.5978 0.3342 -0.1235
VARN 1 1 -0.2453 -0.0841 -0.1653 0.1799 0.1746 0.3788 0.0133
3/2 0 3/2 -0.8248 -0.7800 -1.3773 -0.8494 -0.3953 -1.1008 -0.9018
3/2N 1 3/2 -0.3405 -0.1174 -0.2028 0.2133 0.2085 0.3523 -0.0168
-0.1566 -0.0092 -0.2516 0.1581 0.3417 0.1649 0.0145
Model
Average
Average
Table 6: In-Sample IV Bias (%) by Moneyness, Maturity, and VIX Level. 1996-2004.
Panel B. IV Bias by Maturity
Model Panel A. IV Bias by Moneyness
Name a b VIX<15 15<VIX<20 20<VIX<25 25<VIX<30 30<VIX<35 VIX>35 All
SQR 0 1/2 -1.5081 0.5427 1.3154 1.7536 2.1189 2.1644 1.0642
SQRN 1 1/2 -0.8763 -0.0311 -0.6589 0.3480 1.8700 4.9442 0.0515
VAR 0 1 -2.7125 -0.6450 0.1029 0.6093 0.8517 1.3390 -0.1235
VARN 1 1 -1.2863 -0.1473 -0.4089 0.3412 1.4722 3.7983 0.0133
3/2 0 3/2 -3.6122 -1.4175 -0.5767 -0.1821 0.0651 -0.2396 -0.9018
3/2N 1 3/2 -1.8251 -0.3372 -0.1884 0.4764 1.3507 2.8118 -0.0168
-1.9701 -0.3392 -0.0691 0.5577 1.2881 2.4697 0.0145
Notes: We use the NLSIS estimates from Table 4 to compute the option implied volatility bias (average market prices less average
model price) in percent for various moneyness, maturity, and volatility (VIX Level) bins for each model. The contracts used in the
table are for the 1996-2004 in-sample period, which consists of Wednesday closing call option quotes.
Average
Panel C. IV Bias by VIX Level Model
Name a b RV-Based Ratio VIX-Based Ratio RV-Based Ratio VIX-Based Ratio
SQR 0 1/2 9.089 1.000 6.261 1.000 7.036 1.000 6.196 1.000
VAR 0 1 5.542 0.610 2.552 0.407 3.520 0.500 2.506 0.404
-5.193 -24.249 -7.928 -24.173
3/2 0 3/2 5.150 0.567 2.304 0.368 3.486 0.495 2.265 0.366
-4.749 -30.797 -9.492 -27.645
Name a b RV-Based Ratio VIX-Based Ratio RV-Based Ratio VIX-Based Ratio
SQR 0 1/2 8.613 1.000 6.097 1.000 6.756 1.000 6.074 1.000
VAR 0 1 6.134 0.712 3.784 0.621 4.683 0.693 3.749 0.617
-4.257 -14.371 -5.776 -15.544
DM Test
DM Test
Wednesday Puts: Raw Data Wednesday Puts: Filtered Data
Table 7: IVRMSEs Using Raw and Kalman Filtered VIX and RV Data. 1996-2004.
Model
Model
DM Test
Wednesday Calls: Raw Data Wednesday Calls: Filtered Data
-4.257 -14.371 -5.776 -15.544
3/2 0 3/2 5.520 0.641 3.171 0.520 4.215 0.624 3.146 0.518
-4.216 -21.205 -7.513 -21.173
Name a b RV-Based Ratio VIX-Based Ratio RV-Based Ratio VIX-Based Ratio
SQR 0 1/2 7.783 1.000 6.357 1.000 7.112 1.000 6.269 1.000
VAR 0 1 4.865 0.625 2.720 0.428 3.731 0.525 2.673 0.426
-6.194 -20.052 -7.421 -19.046
3/2 0 3/2 4.735 0.608 2.460 0.387 3.628 0.510 2.415 0.385
-6.672 -25.523 -8.837 -21.918
Notes: We use the NLSIS-based parameters from Table 4 to compute option prices and IVRMSEs for the three option
data sets. RV-based refers to using daily realized volatilities to obtain the volatility state variable used in pricing, and
VIX-based refers to using the daily VIX. The left-side columns use the raw RV and VIX data, and the right-side panels
use RV and VIX filtered using the Kalman filter. The Diebold-Mariano test is computed on the weekly mean squared
errors. Bold type indicates significance at the 5% level.
DM Test
DM Test
Thursday Calls: Raw Data
DM Test
DM Test
Model Thursday Calls: Filtered Data
Estimators
ML|V ML conditional on observing the true spot variance
QML Quasi ML as in Harvey, Ruiz, and Shephard (1994)
GMM Andersen and Srensen (1996)
EMM Efficient method of moments as in Andersen, Chung and Srensen (1999)
AML Approximate ML of Bates (2006)
MCMC Markov Chain Monte Carlo as in J acquier, Polson and Rossi (1994)
MLIS Method used in this paper. See Pitt (2002)
Parameter
True values -0.736 0.9 0.363
Estimators Bias
ML|V -0.015 -0.002 0.000
QML -0.117 -0.020 0.020
GMM 0.150 0.020 -0.080
Table A.1: Monte Carlo Study of MLIS and Alternative Estimation Methods.
EMM -0.057 -0.007 -0.004
AML -0.039 0.005 0.005
MCMC -0.026 -0.004 -0.004
MLIS 0.049 0.004 -0.017
Estimators RMSE
ML|V 0.076 0.010 0.006
QML 0.460 0.060 0.110
GMM 0.310 0.040 0.120
EMM 0.224 0.030 0.049
AML 0.173 0.023 0.043
MCMC 0.150 0.020 0.034
MLIS 0.157 0.020 0.046
Notes: Following Bates (2006), we compare the MLIS estimator to those considered in the
large scale Monte Carlo study by Andersen, Chung and Srensen (1999). We generate 500
Monte Carlo samples of 2,000 returns with zero drift and stochastic volatility following the
logaritmic SV model in (23) and (24) with constant term , persistence , and diffusion
parameter . We then estimate the model using MLIS and compare the bias and RMSE
from MLIS to the bias and RMSE reported in Bates (2006) and Andersen, Chung and
Srensen (1999).