This action might not be possible to undo. Are you sure you want to continue?
The Market for Volatility Trading; VIX Futures
Menachem Brenner
Stern School of Business
New York University
New York, NY 10012, U.S.A.
Email: mbrenner@stern.nyu.edu
Tel: (212) 998 0323, Fax: (212) 995 4731
Jinghong Shu
School of International Trade and Economics
University of International Business and Economics
Huixindongjie, Beijing, P. R. China
Email: shujinghong@uibe.edu.cn
Jin E. Zhang
School of Economics and Finance
and School of Business
The University of Hong Kong
Pokfulam Road, Hong Kong
Email: jinzhang@hku.hk
First Version: August 2006
This Version: May 2007
Key words: Volatility Trading; VIX; VIX Futures
JEL Classification Code: G13
________________________________
* We would like to thank David Hait for his helpful comments. Jin E. Zhang has been supported by a
grant from the Research Grants Council of the Hong Kong Special Administrative Region, China
(Project No. HKU 7427/06H).
The Market for Volatility Trading; VIX Futures
Abstract
This paper analyses the new market for trading volatility; VIX futures. We first use market
data to establish the relationship between VIX futures prices and the index itself. We
observe that VIX futures and VIX are highly correlated; the term structure of VIX futures
price is upward sloping while the term structure of VIX futures volatility is downward
sloping. To establish a theoretical relationship between VIX futures and VIX, we model the
instantaneous variance using a simple square root meanreverting process. Using daily
calibrated variance parameters and VIX, the model gives good predictions of VIX futures
prices. These parameter estimates could be used to price VIX options.
2
1. Introduction
Stochastic volatility was ignored for many years by academics and practitioners.
Changes in volatility were usually assumed to be deterministic (e.g. Merton (1973)). The
importance of stochastic volatility and its potential effect on asset prices and
hedging/investment decisions has been recognized after the crash of ’87. The industry and
academia have started to examine it in the late 80s, empirically as well as theoretically. The
need to hedge potential volatility changes which would require a reference index has been
first presented by Brenner and Galai (1989).
In 1993 the Chicago Board Options Exchange (CBOE) has introduced a volatility index
based on the prices of index options. This was an implied volatility index based on option
prices of the S&P100 and it was traced back to 1986. Until about 1995 the index was not a
good predictor of realized volatility. Since then its forecasting ability has improve markedly
(see Corrado and Miller (2005)) though it is biased upwards. Although many market
participants considered the index to be a good predictor of short term volatility, daily or
even intraday, it took many years for the market to introduce volatility products, starting
with over the counter products like variance swaps. The first exchange traded product, VIX
futures, was introduced in March 2004 followed by VIX options in February 2006. These
volatility derivatives use the VIX index as their underlying. The current VIX is based on a
different methodology than the previous VIX, renamed VXO, and uses the S&P500
European style options rather than the S&P100 American style options. Despite these two
major differences the correlation between the levels of the two indices is about 98%. (see
Carr and Wu (2006)).
3
VIX is computed from the option quotes of all available calls and puts on the S&P500
(SPX) with a nonzero bid price (see the CBOE white paper
1
) using following formula
2
0
2
2
1
1
) (
2


.

\

− −
∆
=
∑
K
F
T
K Q e
K
K
T
i
RT
i i
i
σ , (1)
where the volatility σ times 100 gives the value of the VIX index level. T is the 30 day
volatility estimate. In practice options with 30day maturity might not exist. Thus, the
variances of the two nearterm options, with at least 8 days left to expiration, are combined
to obtain the 30day variance. F is the implied forward index level derived from the nearest
to the money index option prices by using putcall parity.
i
K is the strike price of ith outof
money options,
i
K ∆ is the interval between two strikes,
0
K is the first strike that is below
the forward index level. R is the riskfree rate to expiration. ) (
i
K Q is the midpoint of the
bidask spread of each option with strike
i
K .
Carr and Madan (1998), and Demeterfi et al (1999) developed the original idea of
replicating realized variance by a portfolio of European options. In September 2003, the
CBOE used their theory to design a new methodology to compute VIX.
We now briefly review the theory behind equation (1). If we assume that the strike price
is distributed continuously from 0 to ∞ + and neglect the discretizing error, equation (1)
becomes
(
¸
(
¸


.

\

− − +
(
(
¸
(
¸
+ =
∫ ∫
∞ +
1 ln
2
) (
1
) (
1 2
0 0 0
2 2
2
0
0
K
F
K
F
T
dK K c
K
e dK K p
K
e
T
K
K
RT RT
σ . (2)
By construction,
0
K is very close to F, hence 1
0
−
K
F
is very small but always positive. With
a Taylor series expansion we obtain
1
The CBOE white paper can be retrieved from http://www.cboe.com/micro/vix/vixwhite.pdf
4
3
0
2
0 0 0 0
1 1
2
1
1 1 1 ln ln


.

\

− +


.

\

− −


.

\

− =
(
¸
(
¸


.

\

− + =
K
F
O
K
F
K
F
K
F
K
F
.
By omitting the third order terms,
3
0
1


.

\

−
K
F
O , the last term of equation (2) becomes that of
equation (1). Carr and Madan (1998) and Demeterfi et al (1999) show that due to the
following mathematical identity,
∫ ∫
+∞
− − − −


.

\

− =
0
0
) 0 , max(
1
) 0 , max(
1
1 ln
2
0
2
0 0 K
T
K
T
T T
dK K S
K
dK S K
K K
S
K
S
,
the riskneutral expectation of the log of the terminal stock price over strike
0
K is
∫ ∫
+∞
− −


.

\

− =
(
¸
(
¸
0
0
) (
1
) (
1
1 ln
2
0
2
0 0
0
K
RT
K
RT T Q
dK K c
K
e dK K p
K
e
K
F
K
S
E .
Hence equation (2) can be written as
,
1
) (ln
2
ln ln
2
ln ln
2
0
2
0
0
0
0
0
0 0
0
0
2
∫ ∫
=
(
¸
(
¸
− =
(
¸
(
¸


.

\

− =
(
¸
(
¸


.

\

− =
T
t
Q
T
t
t
t Q
T Q T Q
dt E
T
S d
S
dS
E
T
S
S
E
S
F
T K
S
E
K
F
T
σ
σ
where the last equal sign is due to Ito’s Lemma dt
S
dS
S d
t
t
t
t
2
2
1
) (ln σ − = , under the
assumption that the SPX index follows a diffusion process,
t t t t t
dB S dt S dS σ µ + = with a
general stochastic volatility process,
t
σ . So
2
VIX represents 30day S&P 500 variance swap
rate
2
.
2
In practice, the variance swap rate is quoted as volatility instead of variance. It should be noted that the
realized variance can be replicated by a portfolio of all outofmoney calls and puts but the VIX index itself
cannot be replicated by a portfolio of options because the computation of the VIX involves a square root
operation against the price of a portfolio of options and the square root function is nonlinear.
5
On March 26, 2004, the newly created CBOE Futures Exchange (CFE) started to trade
an exchange listed volatility product; VIX futures, a futures contract written on the VIX
index. It is cash settled with the VIX. Since VIX is not a traded asset, one cannot replicate a
VIX futures contract using the VIX and a risk free asset. Thus a costofcarry relationship
between VIX futures and VIX cannot be established.
Our objective is two fold; First, to use market data to analyse empirically the
relationship between VIX futures prices and VIX, the term structure of VIX futures prices
and to estimate the volatility of VIX futures prices. Second, to find parameter estimates,
using a simple stochastic volatility model, that best describe the empirical relationships and
could possibly be used to price VIX futures and options.
2. Data
In this paper, we use the daily VIX index and VIX futures data provided by the CBOE.
The VIX index data, including open, high, low and close levels, are available from January
2 1990 to the present. The VIX futures data, including open, high, low, close and settle
prices, trading volume and open interest, are available from March 26 2004 to the present.
For each day we have four futures contracts: two near term and two additional months
on the February quarterly cycle. For example, on the first day of the listing, 26 March 2004,
four contracts K4, M4, Q4 and X4 were traded which stand for the following futures
expiration: May, June, August and November 2004 respectively. The first letter indicates
the expiration month followed by the expiration year. The underlying value of the VIX
futures contract is VIX times 10 under the symbol “VXB”
VXB = 10 × VIX. (3)
6
The contract size is $100 times VXB. For example, with a VIX value of 17.33 on 26 March
2004, the VXB would be 173.3 and the contract size would be $17,330. The settlement date
is usually the Wednesday prior to the third Friday of the expiration month.
Our empirical study covers the period of two years and eight months from March 16
2004 to November 21 2006, within which there were 34 contract months traded all together.
Table 1 provides a summary statistics of all of them. The average open interest for each
contract is 2784, which corresponds to a market value of 38 million dollars
3
. The average
daily trading volume for each contract is 181, which corresponds to 2.5 million dollars. The
shortest contract lasted 35 days, while the longest 188 days. The average futures price for
each contract changed from 185.6 for contracts that matured in May 2004 to 164.1 for
contracts that will mature in August 2007 (the average was taken for samples up to the
maturity date or November 21 2006, whichever is earlier), while the VIX level ranged from
17.33 on March 26 2004 to 9.90 on November 21 2006. In general, the market expected
future volatility decreased during this period.
3. Empirical evidence
3.1. The relation between VIX futures and VIX (VXB)
Because the underlying variable of VIX futures, i.e. VXB, is not a traded asset, we are not
able to obtain a simple costofcarry relationship, arbitrage free, between the futures price,
T
t
F , and its underlying,
t
VXB . That is,
) ( t T r
t
T
t
e VXB F
−
≠ ,
3
Using the average VIX futures prices 135.5, we compute the market value as 135.5 × 100 × 2784 =
37,723,200.
7
where r is the interest rate, and T is the maturity. Thus, we have gone to the data to see what
we can learn about the relationship between VIX futures prices and VXB. We use this
relationship to estimate the parameters, in a stochastic volatility model, that could be used to
price volatility derivatives.
There are four futures contracts available on a typical day. For example, on 26 March
2004, we have four kind of VIX futures with maturities, in May, June, August and
November, which corresponds to times to maturity of 53, 80, 142 and 231 days. We
construct 30, 60 and 90day futures prices by a linear interpolation technique. For example,
the 30day futures price is computed by using the market data of VXB and May futures on
26 March 2004. The 60day futures price is computed by using the market data of May and
June futures. The 90day futures price is computed with June and August futures. We
calculate these fixed timetomaturity futures price on each day and obtain three time series
of 30, 60, and 90day futures prices. Figure 1 shows the time series of VXB and VIX
futures for three fixed timetomaturities. Intuitively the four time series are highly
correlated. Table 2 presents the correlation matrix between the returns of S&P 500 index,
VXB and VIX futures. The return is computed as the logarithm of the price relative on two
consecutive ends of day prices. All of the four series are negatively correlated with the S&P
500 index. VXB and VIX futures with three different maturities are almost perfectly
correlated. Figure 1 also shows that the trading volume of VIX futures has been gradually
increasing.
Figure 2 shows the relationship between 30day VIX futures and VXB for the market
data from 26 March 2004 to 21 November 2006. In general, the VIX futures price is an
increasing function of VXB. The higher the VIX (VXB), the higher is the price of VIX
futures with a given maturity.
8
3.2. The term structure of VIX futures price
Over the period of March 26 2004 to 21 November 2006, the average VXB was 135.5.
The average VIX futures prices were 144.8, 152.8 and 157.8 for 30, 60 and 90day
maturities respectively. The term structure of the average VIX futures price is upward
sloping, which is demonstrated graphically in Figure 3.
The upward sloping VIX futures term structure indicates that the current level of
volatility is relatively low compared with the longterm mean level and that the volatility is
increasing to the longterm high level.
3.3. The volatility of VXB and VIX futures
With the time series of VXB and fixed maturity VIX futures price, we compute the
standard deviation of daily log price (index) relatives to obtain estimates of the volatility of
these four series, assuming that these series follow a lognormal process. During the two year
period of our study we estimated the volatility of VXB to be 84.4%, while the volatilities of
VIX futures price are 35.7%, 30.0% and 25.8% for 30, 60 and 90 day maturities
respectively. The longer the maturity, the lower is the volatility of volatility. Figure 4 shows
the volatility of VXB and fixed maturity VIX futures price. The term structure of VIX
futures volatility is downward sloping.
The phenomenon of downward sloping VIX futures volatility is consistent with the
meanreverting feature of the volatility. Since the longterm volatility approaches to a fixed
level, longtenor VIX futures should be less volatile than shorttenor ones.
9
4. A Theoretical Model of VIX Futures price
4.1. VIX futures price
We now use a simple theoretical model to price the futures contracts using parameter
estimates obtained from market data. We then test the extent to which model prices can
explain market prices.
In the physical measure, the SPX index,
t
S , is assumed to follow Heston (1993)
stochastic variance model
P
t t t t t
dB S V dt S dS
1
+ = µ ,
P
t t V t
P P
t
dB V dt V dV
2
) ( σ θ κ + − = ,
where µ is the expected return,
P
θ is longterm mean level of the instantaneous variance,
P
κ is the meanreverting speed of the variance,
V
σ measures the volatility of variance,
P
t
dB
1
and
P
t
dB
2
are increments of two Brownian motions that describe the random noises in
SPX index return and variance. They are assumed to be correlated with a constant
coefficient, ρ .
By changing probability measure from P to Q as follows,
dt
V
r
dB dB
t
Q
t
P
t
−
− =
µ
1 1
, dt V dB dB
t
V
Q
t
P
t
σ
λ
− =
2 2
,
where r is the riskfree rate, and λ is the market price of variance risk, we obtain the
dynamics of the SPX index in the riskneutral measure
Q
t t t t t
dB S V dt rS dS
1
+ = ,
Q
t t V t
P P P
t
dB V dt V dV
2
] ) ( [ σ λ κ θ κ + + − = ,
10
where
Q
t
dB
1
and
Q
t
dB
2
are increments of two Q Brownian motions with the correlation ρ .
We define riskneutral longterm mean level θ and meanreverting speed κ as
λ κ
θ κ
θ
+
=
P
P P
, λ κ κ + =
P
.
Then the riskneutral dynamics of the instantaneous variance can be written as
Q
t t V t t
dB V dt V dV
2
) ( σ θ κ + − = , (4)
The transition probability density, given by Cox, Ingersoll and Ross (1985), is
( ) uv I
u
v
ce V V f
q
q
v u
t T
2 )  (
2 /

.

\

=
− −
, (5)
where
,
] 1 [
2
) ( 2 t T
V
e
c
− −
−
=
κ
σ
κ
) ( t T
t
e cV u
− −
=
κ
,
T
cV v = , 1
2
2
− =
V
q
σ
κθ
,
and ) (⋅
q
I is the modified Bessel function of the first kind of order q. The distribution
function is the noncentral chisquare, ) 2 , 2 2 ; 2 (
2
u q v + χ , with 2 2 + q degrees of freedom and
parameter of noncentrality 2u proportional to the current variance,
t
V .
With the riskneutral dynamics of the variance, we can evaluate the first three
conditional moments of the future variance,
s
V , s t < < 0 , as follows
( ) ( )
) ( t s
t s
Q
t
e V V E
− −
− + =
κ
θ θ ,
( )  
( )
κ
θ σ
κ
σ
κ κ
κ
2
1 1
) (
2
) (
2
) (
) ( 2
2
t s
V
t s
t s
t V s
Q
t s
Q
t
e e
e V V E V E
− − − −
− −
−
+
−
= − ,
( )  
( ) ( )
2
3
) (
4
2
2
) (
) ( 4
3 1
2
1 1
2
3
) (
κ
θ σ
κ
σ
κ κ
κ
t s
V
t s
t s
t V s
Q
t s
Q
t
e e
e V V E V E
− − − −
− −
−
+
−
= − ,
where
Q
t
E stands for the conditional expectation in the riskneutral measure.
11
The VIX index at current time t is defined as the variance swap rate over the next 30
calendar days. It is equal to the riskneutral expectation of the future variance over the
period of 30 days from t to
0
τ + t with 365 / 30
0
= τ ,
∫ ∫
+ +
=
(
(
¸
(
¸
= 
.

\

0 0
) (
1 1
100
0 0
2 τ τ
τ τ
t
t
s
Q
t
t
t
s
Q
t
t
ds V E ds V E
VIX
( )  
t
t
t
t s
t
BV B ds e V + − = − + =
∫
+
− −
θ θ θ
τ
τ
κ
) 1 (
1
0
) (
0
, (6)
where
0
0
1
κτ
κτ −
−
=
e
B is a number between 0 and 1. Hence
2
100

.

\

t
VIX
is the weighted average
between longterm mean level θ and instantaneous variance
t
V with B as the weight. Notice
that the correlation, ρ , does not enter into the VIX formula, hence the VIX values do not
capture the skewness of stock return.
The price of VIX futures with maturity T is then determined by
 
T
Q
t T
Q
t T
Q
t
T
t
BV B E VIX E VXB E F + − × = × = = θ ) 1 ( 1000 ) 10 ( ) (
T t T T
dV V V f BV B )  ( ) 1 ( 1000
0
× + − × =
∫
+∞
θ . (7)
Equations (6) and (7) determine the VIX futures price,
T
t
F , as a function of its underlying,
t
VXB (
t
VIX × 10 ) and time to maturity Tt with three parameters, κ , θ and
V
σ as follows
) , , ; , (
V t
T
t
T
t
t T VXB F F σ θ κ − = . (8)
Given the values of parameters, κ , θ ,
V
σ and current level of
t
VXB , we can back out
t
V from equation (6) and then calculate with equation (7) the current VIX futures price,
T
t
F ,
for different maturities T. Equation (7) can be regarded as a closedform formula for the
VIX futures price.
12
To further the relation between
T
t
F and
t
VXB , we expand
T
BV B + − θ ) 1 ( with Taylor
expansion near the point of ) (
T
Q
t
V E and obtain
        ) ( ) ( ) 1 (
2
1
) ( ) 1 ( ) 1 (
2 / 1 2 / 1
2 / 1
T
Q
t T T
Q
t T
Q
t T
V E V B V BE B V BE B BV B − + − + + − = + −
−
θ θ θ
   
2
2
2 / 3
) ( ) ( ) 1 (
8
1
T
Q
t T T
Q
t
V E V B V BE B − + − −
−
θ
      ( )
4 3
3
2 / 5
) ( ) ( ) ( ) 1 (
16
1
T
Q
t T T
Q
t T T
Q
t
V E V O V E V B V BE B − + − + − +
−
θ ,
where O(.) stands for the higher order terms. Taking expectation in the riskneutral measure
gives an approximate formula for the VIX futures price
 
 
( )
 
( ) ( )
) (
1
2
1 1
2
3
) 1 (
16
2
1 1
) 1 (
8
) 1 (
1000
6
2
3
) (
2
2
) (
) ( 3
2 / 5
) ( ) (
4
2
) ( ) (
) ( 2
2 / 3
) ( ) (
2
2 / 1
) ( ) (
V
t T t T
t T
t
t T
t
t T V
t T t T
t T
t
t T
t
t T V
t T
t
t T
T
t
O
e e
e V B Be V Be
e e
e V B Be V Be
Be V Be
F
σ
κ
θ
κ
θ
σ
κ
θ
κ
θ
σ
θ
κ κ
κ κ κ
κ κ
κ κ κ
κ κ
+
(
(
¸
(
¸
−
+
−
+ − +
(
(
¸
(
¸
−
+
−
+ − −
+ − =
− − − −
− −
−
− − − −
− − − −
− −
−
− − − −
− − − −
(9)
Numerical results show that the approximate formula is very accurate for reasonable set of
parameter values. Evaluating VIX futures price with the approximate formula is much
faster. This proves to be very important in the model calibration exercises when large
numbers of price calculation are required.
4.2. Estimating model parameters from VIX values
Given the physical process of the instantaneous variance we can obtain the likelihood
function of the variance over one day. In equation (6) we relate the VIX index to the
instantaneous variance; hence we would be able to evaluate the likelihood of the VIX over
13
the day. By maximizing the likelihood function over the last 42 trading days, we can
estimate the four parameters,
P
κ ,
P
θ , λ and
V
σ for each day
4
. The daily estimated
parameters are presented in Figure 5. Their average values are 6.9671, 0.02492, 1.02046
and 0.3501 respectively; their standard deviations are 0.9914, 0.01298, 0.7882 and 0.1375
respectively. The negative value of λ is consistent with the empirical phenomenon of
negative volatility/variance risk premium
5
.
We then price VIX futures by using equation (7) with the estimated parameters. The
model price and market price are presented in Figure 6. In general, the model price seems to
be highly correlated with the market price, but there is a non trivial gap between the prices
The root mean squared error is about 1015% of the VIX futures price.
4.3. Calibrating the VIX futures price model
We then used the term structure of VIX futures to get parameter estimates that may
provide better model prices. For example, on 26 March 2004, the market prices are given by
0
VXB =173.3, (
1
T ,
1
0
T
mkt
F ) = (30/365, 190.2), (
2
T ,
2
0
T
mkt
F ) = (60/365, 202.8) and (
3
T ,
3
0
T
mkt
F ) =
(90/365, 201.5), we solve the following optimization problem
( )
∑
=
−
3
1
2
0 0 0
) , , (
) , , ; , ( min
i
T
mkt i
T
mdl
V
i i
F V T VXB F σ θ κ
σ θ κ
and obtain the following values of the three parameters
6
4
Zhang and Zhu (2006) use the same method to estimate the parameters for the whole 1990 to 2005 period.
5
Coval and Shumway (2001) report a negative return of zerobeta, atthemoney straddle positions. Bakshi
and Kapadia (2003) examine the negative market volatility risk premium by using Deltahedged option
portfolios. Carr and Wu (2003) study the variance risk premia on indexes and individual stocks by using the
return of variance swaps.
6
It takes 420 seconds for one calibration exercise if we use the exact formula (7), however it takes only one
second if we use the approximate formula (9). The results from two formulas are very close each other. For
efficiency, we will use the approximate formula in our calibration from now on.
14
= κ 7.6246, = θ 0.04396, =
V
σ 0.2005.
Figure 7 shows the daily calibrated set of parameters. When we do the calibration we set
bounds for the three parameters 8 4 ≤ ≤ κ , 25 . 0 01 . 0 ≤ ≤ θ , and 8 . 0 2 . 0 ≤ ≤
V
σ . The
calibrated values of κ and
V
σ are somehow randomly distributed within the bounds. The
calibrated values of θ seem to be more stable than the other two parameters. The average
values of the three parameters κ , θ and
V
σ are 5.5805, 0.03259 and 0.5885 respectively.
Their standard deviations are 1.5246, 0.00909 and 0.2398 respectively.
With the daily calibrated set of parameters we can recover the latent instantaneous
variance/volatility variable by using formula (6). The result is shown in Figure 8 together
with the VIX index. The two time series seem to be highly correlated.
With the set of parameters and market value of VXB on day t1 we can compute the
model price of VIX futures on the next day, t. We then compare the market prices with the
model prices. Figure 9 shows that the model prices are remarkably close to the market prices
of 30, 60 and 90day VIX futures. The root of mean squared error between the two prices
is around 2.3 dollars, which is less than 2% of the average VIX futures price of around 150.
Our simple meanreverting variance model captures the dynamics of VIX futures price very
well.
4.4. Volatility of the VIX index and VIX futures
From equations (6) and (4), applying Ito’s Lemma gives us a stochastic process for the
VIX index as follows
Q
t t V
t
t V
t
t
t t
t
dB V B
VIX
dt V B
VIX
V B
VIX VIX
dVIX
2
2
2 2
4 2
100
2
1 100
8
1
) (
100
2
1
σ σ θ κ


.

\

+
(
(
¸
(
¸


.

\

− −


.

\

= .
15
The volatility of the VIX index (or VXB) is then given by
t V
t
VIX
V B
VIX
σ σ
2
100
2
1


.

\

= .
With a reasonable set of parameters, for example, κ = 8.0,
V
σ = 0.3501,
t
VIX = 13.55,
t
V =
0.01489, we obtain a volatility of 85.2%, which is close to that computed from the market
values of the VIX index.
We now define volatility hedge ratio, ∆, to be the sensitivity of the VIX futures price
with respect to VXB, i.e.,
t
t
T
t
VXB
t T VXB F
∂
− ∂
= ∆
) , (
. (10)
With the average
0
VXB = 135.5 and average values of the calibrated parameters κ =
5.5805, θ = 0.03259 and
V
σ = 0.5885, we may obtain ∆ as a function of maturity T. As
shown in Figure 10, the volatility hedge ratio is a decreasing function of time to maturity. It
starts from 1 for the VXB, goes to zero for the VIX futures with a very long maturity.
The volatilities of the VIX futures and VXB are related by
VXB T
t
t
F
F
VXB
σ σ ∆ = . (11)
Because
T
t
F and
t
VXB are very close to each other and 1 < ∆ the volatility of the VIX
futures is smaller than the volatility of VXB (VIX). For example, if
VXB
σ = 0.844,
0
VXB =
135.5, κ = 5.5805, θ = 0.03259 and
V
σ = 0.5885, we may compute the volatility of 30day
VIX futures price as follows
16
30 F
σ = 0.5321 844 . 0
8 . 144
5 . 135
× × = 0.409,
60 F
σ = 0.3042 844 . 0
8 . 152
5 . 135
× × = 0.228,
90 F
σ = 0.1819 844 . 0
8 . 157
5 . 135
× × = 0.132,
which is close to the value, 0.357, 0.300 and 0.258 computed from the market prices. Our
simple model also explains the downward sloping term structure of the volatility of VIX
futures.
5. Conclusion
With the enormous increase in derivatives trading and the focus on volatility came the
realization that stochastic volatility is an important risk factor affecting pricing and hedging.
A new asset class, volatility instruments, is emerging and markets that trade these
instruments are created. The first exchange traded instrument is, VIX futures. It has been
trading on the CBOE Futures Exchange since 26 March 2004.
In this paper, we first study the behavior of VIX futures prices using the market data
from March 2004 to November 2006. We observe three stylized facts:
1. The index, VIX, and the 3 VIX futures prices are negatively correlated with the S&P
500 index. VIX and VIX futures with three different maturities are almost perfectly
correlated.
2. The term structure of the average VIX futures price is upward sloping.
3. The volatility term structure of VIX futures is downward sloping.
17
The first fact is not surprising. Traders often take long position in volatility derivatives to
hedge the risk in stock position. The second fact shows that the long term mean level of
volatility is higher than the current level. The third observation is not wellknown.
In the second part of the paper we use a simple meanreverting variance model to
establish the theoretical relationship between VIX futures prices and its underlying spot
index. Using the variance parameters calibrated from our model with the market data at t1,
we can price VIX futures at time t conditional on VIX at time t. An empirical study over the
whole sample period shows that our model provides prices that are very close to the market
prices. The root mean squared error between the market and model prices is around 2.3
dollars, which is less than 2% of the average VIX futures price of around 150. Our simple
meanreverting variance model captures the dynamics of VIX futures price very well.
References:
1. Bakshi, Gurdip, and Nikunji Kapadia, 2003, Deltahedged gains and the negative market
volatility risk premium, Review of Financial Studies 16, 527566.
2. Brenner, Menachem, and Dan Galai, 1989, New Financial Instruments for Hedging
Changes in Volatility, Financial Analyst Journal, July/August, 6165.
3. Carr, Peter, and Dilip Madan, 1998, Towards a theory of volatility trading. In Robert
Jarrow (Ed.), Volatility estimation techniques for pricing derivatives, London: Risk
Books, pp. 417427.
4. Carr, Peter, and Liuren Wu, 2003, Variance risk premia, Working paper, City University
of New York and Bloomberg L. P.
5. Carr, Peter, and Liuren Wu, 2006, A tale of two indices, Journal of Derivatives 13, 13
29.
18
6. Corrado, Charles J., and Thomas W. Miller, Jr., 2005, The forecast quality of CBOE
implied volatility indexes, Journal of Futures Markets 25, 339373.
7. Coval, Joshua D., and Tyler Shumway, 2001, Expected option returns, Journal of
Finance 56, 9831009.
8. Cox, John C., Jonathan E. Ingersoll, Jr., and Stephen A. Ross, 1985, A theory of the
term structure of interest rates, Econometrica 53, 385407.
9. Demeterfi, Kresimir, Emanuel Derman, Michael Kamal, and Joseph Zou, 1999, A guide
to volatility and variance swaps, Journal of Derivatives 6, 932.
10. Heston, Steven L., 1993, A closedform solution for options with stochastic volatility
with applications to bond and currency options, Review of Financial Studies 6, 327343.
11. Merton, Robert C., 1973, Theory of Rational Option Pricing. Bell Journal of Economics
and Management Science 4, 141183.
12. Zhang, Jin E., and Yingzi Zhu, 2006, VIX Futures, Journal of Futures Markets 26, 521
531.
19
Table 1: Summary Statistics for VIX Futures Contracts from 26 March 2004 to 21
November 2006
Code Contract
No. of observations
Period covered
VIX Futures prices Open interest
Volume
Start End Mean Std Mean Mean
K4 May 04
38 03/26/04 05/19/04 185.6 9.2 1166 148
M4 Jun 04
56 03/26/04 06/16/04 186.8 14.3 1256 170
N4 July 04
35 05/21/04 07/14/04 163.5 30.9 1926 186
Q4 Aug 04
100 03/26/04 08/18/04 193.2 12.3 1697 135
U4 Sep 04
42 07/19/04 09/15/04 178.6 22.8 1533 158
V4 Oct0 4
37 08/20/04 10/13/04 152.7 29.0 1226 124
X4 Nov 04
164 03/26/04 11/17/04 190.1 25.6 2794 145
F5 Jan 05
62 10/21/04 01/19/05 146.1 11.6 1210 75
G5 Feb 05
168 06/18/04 02/16/05 171.2 32.0 3074 124
H5 Mar 05
37 01/24/05 03/16/05 127.4 7.9 1181 154
K5 May 05
168 09/20/04 05/18/05 156.5 17.2 2629 157
M5 June 05
61 03/21/05 06/15/05 142.9 12.4 1178 138
Q5 Aug 05
186 11/19/04 08/17/05 148.6 18.9 4403 172
V5 Oct 05
86 06/20/05 10/19/05 143.0 5.4 826 66
X5 Nov 05
188 02/22/05 11/16/05 150.6 8.3 3303 129
Z5 Dec 05
42 10/21/05 12/21/05 123.7 23.0 928 84
F6 Jan 06
40 11/18/05 01/18/06 122.1 20.9 305 52
G6 Feb 06
185 05/23/05 02/15/06 150.7 11.6 3416 146
H6 Mar 06
43 01/20/06 03/22/06 122.8 20.6 677 64
J6 Apr 06
42 02/17/06 04/19/06 122.1 19.7 648 61
K6 May 06
187 08/19/05 05/17/06 148.4 19.1 5627 281
M6 Jun 06
43 04/21/06 06/14/06 146.2 32.1 1905 279
N6 Jul 06
42 05/22/06 07/19/06 154.5 27.5 1293 174
Q6 Aug 06
164 12/21/06 08/16/06 150.4 16.4 10487 467
U6 Sep 06
42 07/24/06 09/20/06 140.8 12.9 3788 412
V6 Oct 06
43 08/18/06 10/18/06 131.9 23.9 8220 685
X6 Nov 06
178 03/08/06 11/15/06 147.5 19.3 14957 604
Z6 Dec 06
45 09/21/06 12/20/06 134.2 15.2 6849 408
F7 Jan 07
24 10/20/06 01/17/07 124.7 27.1 406 72
G7 Feb 07
182 03/08/06 02/21/07 156.7 15.2 3839 190
H7 Mar 07
24 10/20/06 03/21/07 136.1 29.5 508 28
J7 Apr 07
24 10/20/06 04/18/07 140.5 30.3 36 7
K7 May 07
172 03/22/06 05/16/07 161.3 13.9 1183 49
Q7 Aug 07
109 06/21/06 08/15/07 164.1 4.1 177 6
Average
150.5 2784 181
Note: The futures contract code is the expiration month code followed by a digit
representing the expiration year. The expiration month codes follow the convention for all
commodities futures, which is defined as follows: JanuaryF, FebruaryG, MarchH, AprilJ,
MayK, JuneM, JulyN, AugustQ, SeptemberU, OctoberV, NovemberX and December
Z.
20
Table 2: The correlation matrix between daily continuously compounded returns of S&P
500 index, VXB and fixed maturity VIX futures computed based on the market data from 26
March 2004 to 21 November 2006. The fixed maturity VIX futures prices are constructed
by using the market data of available contracts with linear interpolation technique. The daily
continuously compounded return is defined as the logarithm of the ratio between the price
on next day and the price on current day. The upper triangle is the same as the lower triangle
because the matrix is symmetric.
S&P 500
index
VXB 30day VIX
futures
60day VIX
futures
90day VIX
Futures
S&P 500
index
1
VXB
0.7974 1
30day VIX
futures
0.7748 0.8140 1
60day VIX
futures
0.6827 0.6918 0.8420 1
90day VIX
futures
0.6975 0.6715 0.8075 0.8249 1
21
0
50
100
150
200
250
300
2
0
0
4

3

2
6
2
0
0
4

5

2
6
2
0
0
4

7

2
6
2
0
0
4

9

2
6
2
0
0
4

1
1

2
6
2
0
0
5

1

2
6
2
0
0
5

3

2
6
2
0
0
5

5

2
6
2
0
0
5

7

2
6
2
0
0
5

9

2
6
2
0
0
5

1
1

2
6
2
0
0
6

1

2
6
2
0
0
6

3

2
6
2
0
0
6

5

2
6
2
0
0
6

7

2
6
2
0
0
6

9

2
6
Trading Volume
VXB
30day VIX Futures
60day VIX Futures
90day VIX Futures
Figure 1. VXB and VIX futures price with three fixed timetomaturities between 26
March 2004 and 21 November 2006. The VXB time series is from the CBOE. The fixed
maturity VIX futures prices are constructed by using the market data of available contracts
with linear interpolation technique. The bar chart shows the trading volume (normalized by
100 contracts) of futures of all maturity on the day.
0
50
100
150
200
250
0 50 100 150 200 250 300
VXB
3
0

d
a
y
V
I
X
F
u
t
u
r
e
s
Figure 2. The relation between 30day VIX futures and VXB (March 26 2004 – 21
November 2006). The VXB time series is from the CBOE. The 30day VIX futures prices
are constructed by using the market data of available contracts with linear interpolation
technique.
22
(90,157.8)
(60,152.8)
(30,144.8)
(0,135.5)
120
125
130
135
140
145
150
155
160
0 30 60 90
Time to Maturity
A
v
e
r
a
g
e
V
I
X
F
u
t
u
r
e
s
P
r
i
c
e
Figure 3. The term structure of average VIX futures price. The average VIX futures
price is computed based on the fixed maturity data from 26 March 2004 to 21 November
2006. The fixed maturity VIX futures prices are constructed by using the market data of
available contracts with linear interpolation technique.
0.844
0.357
0.300
0.258
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0 30 60 90
Time to Maturity
V
o
l
a
t
i
l
i
t
y
o
f
V
I
X
F
u
t
u
r
e
s
P
r
i
c
e
Figure 4. The annualized volatility of VIX futures price. The annualized volatility is
computed based on the fixed maturity data from 26 March 2004 to 21 November 2006. The
fixed maturity VIX futures prices are constructed by using the market data of available
contracts with linear interpolation technique.
23
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
4
5
6
7
8
9
10
κ
P
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
0.02
0.04
0.06
0.08
θ
P
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
3
2
1
0
1
λ
24
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
0.2
0.4
0.6
0.8
1
σ
V
Figure 5. The variance parameters estimated from the historical VIX index of the last
42 trading days. The average values of the estimated four parameters
P
κ ,
P
θ , λ and
V
σ
are 6.9671, 0.02492, 1.02046 and 0.3501 respectively. Their standard deviations are
0.9914, 0.01298, 0.7882 and 0.1375 respectively.
25
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
50
100
150
200
250
0
3

y
a
d
X
I
V
s
e
r
u
t
u
F
e
c
i
r
P
d
n
a
B
X
V
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
50
100
150
200
250
0
6

y
a
d
X
I
V
s
e
r
u
t
u
F
e
c
i
r
P
d
n
a
B
X
V
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
50
100
150
200
250
0
9

y
a
d
X
I
V
s
e
r
u
t
u
F
e
c
i
r
P
d
n
a
B
X
V
Figure 6. A comparison of model predicted prices and market prices of 30, 60 and 90
day VIX futures. The upper solid line is the market price. The dots are model predicted
prices based on the parameters estimated from the VIX index of the last 42 trading days.
The lower solid line is the VXB value shifted downward by 90 for the purpose of an easy
comparison. The average prices of 30, 60 and 90day VIX futures are (144.8, 152.8,
157.8). The roots of mean squared error between model predicted price and market price are
(14.89, 21.97, 26.18).
26
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
4
5
6
7
8
9
10
κ
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
0.02
0.03
0.04
0.05
0.06
0.07
0.08
θ
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
0.2
0.4
0.6
0.8
1
σ
V
Figure 7. The model parameter values calibrated from the market prices of 30, 60
and 90day VIX futures. The average values of the three parameters κ , θ and
V
σ are
5.5805, 0.03259 and 0.5885 respectively. Their standard deviations are 1.5246, 0.00909 and
0.2398 respectively.
27
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
0
0.05
0.1
0.15
0.2
0.25
è ! !!!
V
t
VIX
t
100
−0.09
Figure 8. The instantaneous volatility,
t
V , recovered from VIX index and daily
calibrated set of parameters. The upper line is the instantaneous variance, the lower line is
VIX index divided by 100 and shifted down by 9% for the purpose of an easy comparison.
28
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
50
100
150
200
0
3

y
a
d
X
I
V
s
e
r
u
t
u
F
d
n
a
B
X
V
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
50
100
150
200
0
6

y
a
d
X
I
V
s
e
r
u
t
u
F
d
n
a
B
X
V
01Jun04 01Dec04 01Jun05 01Dec05 01Jun06
50
100
150
200
0
9

y
a
d
X
I
V
s
e
r
u
t
u
F
d
n
a
B
X
V
Figure 9. A comparison on model predicted prices and market prices of 30, 60 and
90day VIX futures. The upper solid line is the market price. The dots are model predicted
prices based on the parameters calibrated from VIX futures prices on the previous day. The
lower solid line is the VXB value shifted downward by 90 for the purpose of an easy
comparison. The average prices of 30, 60 and 90day VIX futures are (144.8, 152.8,
157.8). The roots of mean squared error between model price and market price are (2.56,
2.34, 2.27).
29
0 20 40 60 80
Time to maturity HdayL
0.2
0.4
0.6
0.8
1
e
g
d
e
H
o
i
t
a
r
Figure 10. The volatility hedge ratio as a function of time to maturity.
0
VXB = 135.5,
the set of parameters is taken to be the average calibrated values κ = 5.5805, θ = 0.03259
and
V
σ = 0.5885.
The Market for Volatility Trading; VIX Futures
Abstract This paper analyses the new market for trading volatility; VIX futures. We first use market data to establish the relationship between VIX futures prices and the index itself. We observe that VIX futures and VIX are highly correlated; the term structure of VIX futures price is upward sloping while the term structure of VIX futures volatility is downward sloping. To establish a theoretical relationship between VIX futures and VIX, we model the instantaneous variance using a simple square root meanreverting process. Using daily calibrated variance parameters and VIX, the model gives good predictions of VIX futures prices. These parameter estimates could be used to price VIX options.
1. Introduction
Stochastic volatility was ignored for many years by academics and practitioners. Changes in volatility were usually assumed to be deterministic (e.g. Merton (1973)). The importance of stochastic volatility and its potential effect on asset prices and hedging/investment decisions has been recognized after the crash of ’87. The industry and academia have started to examine it in the late 80s, empirically as well as theoretically. The need to hedge potential volatility changes which would require a reference index has been first presented by Brenner and Galai (1989). In 1993 the Chicago Board Options Exchange (CBOE) has introduced a volatility index based on the prices of index options. This was an implied volatility index based on option prices of the S&P100 and it was traced back to 1986. Until about 1995 the index was not a good predictor of realized volatility. Since then its forecasting ability has improve markedly (see Corrado and Miller (2005)) though it is biased upwards. Although many market participants considered the index to be a good predictor of short term volatility, daily or even intraday, it took many years for the market to introduce volatility products, starting with over the counter products like variance swaps. The first exchange traded product, VIX futures, was introduced in March 2004 followed by VIX options in February 2006. These volatility derivatives use the VIX index as their underlying. The current VIX is based on a different methodology than the previous VIX, renamed VXO, and uses the S&P500 European style options rather than the S&P100 American style options. Despite these two major differences the correlation between the levels of the two indices is about 98%. (see Carr and Wu (2006)).
2
com/micro/vix/vixwhite. K 0 is the first strike that is below the forward index level. T is the 30 day volatility estimate.VIX is computed from the option quotes of all available calls and puts on the S&P500 (SPX) with a nonzero bid price (see the CBOE white paper1) using following formula 2 1 F ∆K ∑ K 2i e RT Q( K i ) − T K − 1 . In September 2003. and Demeterfi et al (1999) developed the original idea of replicating realized variance by a portfolio of European options. equation (1) becomes 2 σ = e RT T 2 K0 +∞ 2 F F 1 1 RT p ( K )dK + e ∫ 2 c( K )dK + ln − − 1 . T i i 0 2 σ2 = (1) where the volatility σ times 100 gives the value of the VIX index level. Thus. In practice options with 30day maturity might not exist. the CBOE used their theory to design a new methodology to compute VIX.cboe. We now briefly review the theory behind equation (1). are combined to obtain the 30day variance. the variances of the two nearterm options. hence F − 1 is very small but always positive. Q( K i ) is the midpoint of the bidask spread of each option with strike K i . K 0 is very close to F. Carr and Madan (1998). If we assume that the strike price is distributed continuously from 0 to + ∞ and neglect the discretizing error. ∆K i is the interval between two strikes. ∫ K2 K T K0 K0 0 K0 (2) By construction. K i is the strike price of ith outofmoney options. R is the riskfree rate to expiration.pdf 3 . with at least 8 days left to expiration. F is the implied forward index level derived from the nearest to the money index option prices by using putcall parity. With K0 a Taylor series expansion we obtain 1 The CBOE white paper can be retrieved from http://www.
ST ST = K − 1 − K0 0 K0 3 ln 1 1 ∫ K 2 max(K − ST . 0 0 +∞ the riskneutral expectation of the log of the terminal stock price over strike K 0 is RT S F Q E0 ln T = K − 1 − e K0 0 Hence equation (2) can be written as K0 1 1 RT ∫ K 2 p( K )dK − e K∫ K 2 c( K )dK . 4 . K0 0 0 0 0 2 3 F − 1 . It should be noted that the realized variance can be replicated by a portfolio of all outofmoney calls and puts but the VIX index itself cannot be replicated by a portfolio of options because the computation of the VIX involves a square root operation against the price of a portfolio of options and the square root function is nonlinear. T 0 0 St T where the last equal sign is due to Ito’s Lemma d (ln St ) = dSt 1 2 − σ t dt . 2 In practice. 0 0 +∞ σ2 = S 2 F S 2 F Q Q − E0 ln T = ln − E0 ln T ln K T S T K0 0 0 S0 T 1 QT 2 2 Q dSt = E0 ∫ − d (ln St ) = E0 ∫ σ t dt .0)dK − K∫ K 2 max(ST − K .0)dK . O K 0 equation (1). σ t . the last term of equation (2) becomes that of By omitting the third order terms. Carr and Madan (1998) and Demeterfi et al (1999) show that due to the following mathematical identity. the variance swap rate is quoted as volatility instead of variance. under the St 2 assumption that the SPX index follows a diffusion process. dSt = µSt dt + σ t St dBt with a general stochastic volatility process. F F 1 F F F ln = ln 1 + K − 1 = K − 1 − 2 K − 1 + O K − 1 . So VIX 2 represents 30day S&P 500 variance swap rate2.
M4. Since VIX is not a traded asset. The underlying value of the VIX futures contract is VIX times 10 under the symbol “VXB” VXB = 10 × VIX. The VIX futures data. close and settle prices. the term structure of VIX futures prices and to estimate the volatility of VIX futures prices. a futures contract written on the VIX index. high. First. we use the daily VIX index and VIX futures data provided by the CBOE. The first letter indicates the expiration month followed by the expiration year. 26 March 2004. using a simple stochastic volatility model. one cannot replicate a VIX futures contract using the VIX and a risk free asset. to use market data to analyse empirically the relationship between VIX futures prices and VIX. Data In this paper. low and close levels. the newly created CBOE Futures Exchange (CFE) started to trade an exchange listed volatility product. are available from January 2 1990 to the present. 2. are available from March 26 2004 to the present.On March 26. The VIX index data. including open. (3) 5 . June. Q4 and X4 were traded which stand for the following futures expiration: May. It is cash settled with the VIX. For example. to find parameter estimates. high. trading volume and open interest. Second. four contracts K4. on the first day of the listing. that best describe the empirical relationships and could possibly be used to price VIX futures and options. low. 2004. Thus a costofcarry relationship between VIX futures and VIX cannot be established. Our objective is two fold. For each day we have four futures contracts: two near term and two additional months on the February quarterly cycle. including open. VIX futures. August and November 2004 respectively.
1 for contracts that will mature in August 2007 (the average was taken for samples up to the maturity date or November 21 2006.6 for contracts that matured in May 2004 to 164. 3 Using the average VIX futures prices 135. Empirical evidence 3.5 million dollars. The shortest contract lasted 35 days. the VXB would be 173. The relation between VIX futures and VIX (VXB) Because the underlying variable of VIX futures. arbitrage free. In general. is not a traded asset. between the futures price. VXB. and its underlying. That is.90 on November 21 2006. we compute the market value as 135. For example.5 × 100 × 2784 = 37. we are not able to obtain a simple costofcarry relationship.3 and the contract size would be $17. The average daily trading volume for each contract is 181.1.5. Our empirical study covers the period of two years and eight months from March 16 2004 to November 21 2006. while the longest 188 days.33 on 26 March 2004.e. while the VIX level ranged from 17. The average futures price for each contract changed from 185.The contract size is $100 times VXB. 6 . The average open interest for each contract is 2784. FtT . which corresponds to a market value of 38 million dollars3. FtT ≠ VXBt e r (T −t ) .330. which corresponds to 2. VXBt . i.33 on March 26 2004 to 9. The settlement date is usually the Wednesday prior to the third Friday of the expiration month.723. 3.200. Table 1 provides a summary statistics of all of them. whichever is earlier). the market expected future volatility decreased during this period. with a VIX value of 17. within which there were 34 contract months traded all together.
Table 2 presents the correlation matrix between the returns of S&P 500 index. VXB and VIX futures with three different maturities are almost perfectly correlated. and T is the maturity. we have four kind of VIX futures with maturities. The 90day futures price is computed with June and August futures. Figure 2 shows the relationship between 30day VIX futures and VXB for the market data from 26 March 2004 to 21 November 2006. 142 and 231 days. that could be used to price volatility derivatives. We calculate these fixed timetomaturity futures price on each day and obtain three time series of 30. the VIX futures price is an increasing function of VXB. in a stochastic volatility model. 60 and 90day futures prices by a linear interpolation technique.where r is the interest rate. and 90day futures prices. August and November. we have gone to the data to see what we can learn about the relationship between VIX futures prices and VXB. Figure 1 shows the time series of VXB and VIX futures for three fixed timetomaturities. 60. For example. Intuitively the four time series are highly correlated. We use this relationship to estimate the parameters. We construct 30. The return is computed as the logarithm of the price relative on two consecutive ends of day prices. June. The 60day futures price is computed by using the market data of May and June futures. All of the four series are negatively correlated with the S&P 500 index. on 26 March 2004. The higher the VIX (VXB). Thus. In general. 80. For example. which corresponds to times to maturity of 53. the 30day futures price is computed by using the market data of VXB and May futures on 26 March 2004. There are four futures contracts available on a typical day. 7 . VXB and VIX futures. the higher is the price of VIX futures with a given maturity. Figure 1 also shows that the trading volume of VIX futures has been gradually increasing. in May.
30. while the volatilities of VIX futures price are 35. The phenomenon of downward sloping VIX futures volatility is consistent with the meanreverting feature of the volatility. 152.3. During the two year period of our study we estimated the volatility of VXB to be 84. the average VXB was 135. longtenor VIX futures should be less volatile than shorttenor ones. the lower is the volatility of volatility.8.8% for 30. The term structure of the average VIX futures price is upward sloping. The average VIX futures prices were 144.8 and 157.3.7%. 3.2. 60 and 90 day maturities respectively. which is demonstrated graphically in Figure 3.5.4%.and 90day maturities respectively. The volatility of VXB and VIX futures With the time series of VXB and fixed maturity VIX futures price. 60. we compute the standard deviation of daily log price (index) relatives to obtain estimates of the volatility of these four series.8 for 30. The term structure of VIX futures volatility is downward sloping. Since the longterm volatility approaches to a fixed level.0% and 25. 8 . The upward sloping VIX futures term structure indicates that the current level of volatility is relatively low compared with the longterm mean level and that the volatility is increasing to the longterm high level. The longer the maturity. assuming that these series follow a lognormal process. The term structure of VIX futures price Over the period of March 26 2004 to 21 November 2006. Figure 4 shows the volatility of VXB and fixed maturity VIX futures price.
σ V measures the volatility of variance. t Q dVt = [κ Pθ P − (κ P + λ )Vt ]dt + σ V Vt dB2t . 9 . where µ is the expected return. By changing probability measure from P to Q as follows. and λ is the market price of variance risk. They are assumed to be correlated with a constant coefficient. dB1P and dB2Pt are increments of two Brownian motions that describe the random noises in t SPX index return and variance. θ P is longterm mean level of the instantaneous variance. We then test the extent to which model prices can explain market prices.1. κ P is the meanreverting speed of the variance. Q dB2Pt = dB2t − λ Vt dt . S t . A Theoretical Model of VIX Futures price 4. VIX futures price We now use a simple theoretical model to price the futures contracts using parameter estimates obtained from market data. In the physical measure. ρ . t dVt = κ P (θ P − Vt )dt + σ V Vt dB2Pt . we obtain the dynamics of the SPX index in the riskneutral measure dS t = rS t dt + Vt S t dB1Q . the SPX index. is assumed to follow Heston (1993) stochastic variance model dS t = µS t dt + Vt S t dB1P .4. σV where r is the riskfree rate. dB1P = dB1Q − t t µ −r Vt dt .
is v f (VT  Vt ) = ce −u −v u where c= q/2 I q 2 uv . Ingersoll and Ross (1985). χ 2 (2v. Then the riskneutral dynamics of the instantaneous variance can be written as Q dVt = κ (θ − Vt )dt + σ V Vt dB2t . The distribution function is the noncentral chisquare. q= 2κθ 2 σV −1. E Q t [(V − E s Q t (Vs ) ) ]= σ 2 3 −κ ( s −t ) 2 V t Ve 1 − e −κ ( s −t ) κ +σ 2 V (1 − e θ −κ ( s −t ) 2 2κ ) . E Vs − E (Vs ) Q t Q t [( )] 3 4 1 − e −κ ( s − t ) = σ V Vt e −κ ( s − t ) κ2 2 ( ) 2 1 4 1 − e −κ ( s − t ) + σ Vθ κ2 2 ( ) 3 . κP +λ κ =κP +λ. ( ) (5) 2κ . σ [1 − e −κ (T − t ) ] 2 V u = cVt e −κ (T − t ) . we can evaluate the first three conditional moments of the future variance. Vt .Q where dB1Q and dB2t are increments of two Q Brownian motions with the correlation ρ . With the riskneutral dynamics of the variance. 10 .2u ) . Vs . as follows EtQ (Vs ) = θ + (Vt − θ )e −κ ( s −t ) . and I q (⋅) is the modified Bessel function of the first kind of order q.2q + 2. (4) The transition probability density. where EtQ stands for the conditional expectation in the riskneutral measure. given by Cox. v = cVT . t We define riskneutral longterm mean level θ and meanreverting speed κ as θ= κ Pθ P . with 2q + 2 degrees of freedom and parameter of noncentrality 2u proportional to the current variance. 0 < t < s .
The price of VIX futures with maturity T is then determined by FtT = EtQ (VXBT ) = EtQ (10 × VIX T ) = 1000 × EtQ +∞ [ (1 − B )θ + BVT ] (7) = 1000 × ∫ 0 (1 − B)θ + BVT × f (VT  Vt )dVT . VXBt ( 10 × VIX t ) and time to maturity Tt with three parameters.The VIX index at current time t is defined as the variance swap rate over the next 30 calendar days. does not enter into the VIX formula. hence the VIX values do not capture the skewness of stock return. Equation (7) can be regarded as a closedform formula for the VIX futures price. 2 VIX t Q 1 = Et 100 τ 0 t +τ 0 1 Vs ds = ∫ τ0 t t +τ 0 t +τ 0 ∫E t Q t (Vs )ds = 1 τ0 ∫ [θ + (V t t − θ )e −κ ( s −t ) ds = (1 − B)θ + BVt . κ . It is equal to the riskneutral expectation of the future variance over the period of 30 days from t to t + τ 0 with τ 0 = 30 / 365 . (8) Given the values of parameters.σ V ) . for different maturities T. 11 . θ . as a function of its underlying.θ . FtT . FtT . ] (6) where B = 1 − e −κτ 0 κτ 0 VIX t is a number between 0 and 1. Notice that the correlation.κ . κ . σ V and current level of VXBt . T − t . Equations (6) and (7) determine the VIX futures price. θ and σ V as follows FtT = FtT (VXBt . Hence is the weighted average 100 2 between longterm mean level θ and instantaneous variance Vt with B as the weight. ρ . we can back out Vt from equation (6) and then calculate with equation (7) the current VIX futures price.
16 [ ] [ ] ([ ]) where O(.) stands for the higher order terms. 4. This proves to be very important in the model calibration exercises when large numbers of price calculation are required.2. Estimating model parameters from VIX values Given the physical process of the instantaneous variance we can obtain the likelihood function of the variance over one day.To further the relation between FtT and VXBt . we expand expansion near the point of EtQ (VT ) and obtain (1 − B)θ + BVT with Taylor [(1 − B)θ + BVT ]1/ 2 = [(1 − B)θ + BEtQ (VT )]1/ 2 + 1 [(1 − B)θ + BEtQ (VT )]−1/ 2 B[VT − EtQ (VT )] 2 − −3 / 2 2 2 1 (1 − B )θ + BEtQ (VT ) B VT − EtQ (VT ) 8 [ ] [ ] + −5 / 2 3 3 4 1 (1 − B)θ + BEtQ (VT ) B VT − EtQ (VT ) + O VT − EtQ (VT ) . hence we would be able to evaluate the likelihood of the VIX over 12 . In equation (6) we relate the VIX index to the instantaneous variance. Taking expectation in the riskneutral measure gives an approximate formula for the VIX futures price FtT = θ (1 − Be −κ (T −t ) ) + Vt Be −κ (T −t ) 1000 − + 2 σV [ ] 1/ 2 8 4 σV [θ (1 − Be −κ (T −t ) ) + Vt Be −κ (T −t ) ) + Vt Be ] −3 / 2 1 − e −κ (T −t ) 1 − e −κ (T −t ) B 2 Vt e −κ (T −t ) +θ 2κ κ 3 1 − e −κ (T −t ) B Vt e −κ (T −t ) κ2 2 3 ( ) 2 [θ (1 − Be 16 −κ (T −t ) −κ (T −t ) −5 / 2 ] ( ) 2 1 1 − e −κ (T −t ) + θ 2 κ2 ( ) + O(σ 3 6 V ) (9) Numerical results show that the approximate formula is very accurate for reasonable set of parameter values. Evaluating VIX futures price with the approximate formula is much faster.
By maximizing the likelihood function over the last 42 trading days.7882 and 0. their standard deviations are 0. Their average values are 6.θ . ( T1 . Bakshi and Kapadia (2003) examine the negative market volatility risk premium by using Deltahedged option portfolios. Coval and Shumway (2001) report a negative return of zerobeta. The model price and market price are presented in Figure 6. however it takes only one second if we use the approximate formula (9).1375 respectively.02046 and 0. In general. For example. F0Tmkt ) = (60/365. the model price seems to be highly correlated with the market price. 190. but there is a non trivial gap between the prices The root mean squared error is about 1015% of the VIX futures price. θ P . κ P . ( T2 . It takes 420 seconds for one calibration exercise if we use the exact formula (7). 0. we will use the approximate formula in our calibration from now on.2).02492. we can estimate the four parameters. 0.9914. 6 13 .the day. We then price VIX futures by using equation (7) with the estimated parameters.01298. 201. 1. atthemoney straddle positions. F0Tmkt ) = (90/365. the market prices are given by 3 1 2 VXB0 =173. The results from two formulas are very close each other. Calibrating the VIX futures price model We then used the term structure of VIX futures to get parameter estimates that may provide better model prices.σV ) − F0Tmkt ) 2 and obtain the following values of the three parameters6 4 5 Zhang and Zhu (2006) use the same method to estimate the parameters for the whole 1990 to 2005 period.σV ) ∑ (F 3 i =1 Ti 0 mdl i (VXB0 . For efficiency. The daily estimated parameters are presented in Figure 5. 0.8) and ( T3 . F0Tmkt ) = (30/365.5). The negative value of λ is consistent with the empirical phenomenon of negative volatility/variance risk premium5.3.κ . 202. 4. we solve the following optimization problem min (κ .3501 respectively. λ and σ V for each day 4 . on 26 March 2004. Carr and Wu (2003) study the variance risk premia on indexes and individual stocks by using the return of variance swaps.3. Ti .θ .9671.
Figure 7 shows the daily calibrated set of parameters.5805.25 . When we do the calibration we set bounds for the three parameters 4 ≤ κ ≤ 8 .and 90day VIX futures.03259 and 0. Figure 9 shows that the model prices are remarkably close to the market prices of 30. Our simple meanreverting variance model captures the dynamics of VIX futures price very well. We then compare the market prices with the model prices. 0. The average values of the three parameters κ .8 .κ = 7.00909 and 0.01 ≤ θ ≤ 0. The calibrated values of κ and σ V are somehow randomly distributed within the bounds.2 ≤ σ V ≤ 0.2005. The root of mean squared error between the two prices is around 2. The two time series seem to be highly correlated. applying Ito’s Lemma gives us a stochastic process for the VIX index as follows dVIX t 1 100 = VIX t 2 VIX t 14 1 100 Bκ (θ − Vt ) − 8 VIX t 2 4 2 2 1 100 B σ V Vt dt + 2 VIX t Q Bσ V Vt dB2t . 4. With the daily calibrated set of parameters we can recover the latent instantaneous variance/volatility variable by using formula (6). t. σ V = 0. The result is shown in Figure 8 together with the VIX index.2398 respectively. 0.5885 respectively. The calibrated values of θ seem to be more stable than the other two parameters.3 dollars.6246. θ = 0. 2 . With the set of parameters and market value of VXB on day t1 we can compute the model price of VIX futures on the next day. and 0.4.04396. which is less than 2% of the average VIX futures price of around 150. Their standard deviations are 1. 60. Volatility of the VIX index and VIX futures From equations (6) and (4). θ and σ V are 5. 0.5246.
∆ .5. the volatility hedge ratio is a decreasing function of time to maturity. VIX t = 13. It starts from 1 for the VXB.e.03259 and σ V = 0. ∆= ∂Ft T (VXBt .. we may compute the volatility of 30day VIX futures price as follows 15 . θ = 0. σ V = 0. For example. ∂VXBt (10) With the average VXB0 = 135. κ = 5.2%. we obtain a volatility of 85. We now define volatility hedge ratio. if σ VXB = 0. i.5885.5805. for example.03259 and σ V = 0. we may obtain ∆ as a function of maturity T.0. 2 σ VIX With a reasonable set of parameters.5885. T − t ) .01489. to be the sensitivity of the VIX futures price with respect to VXB. As shown in Figure 10. The volatilities of the VIX futures and VXB are related by σF = ∆ VXBt σ VXB . κ = 8. VXB0 = 135.5805.5 and average values of the calibrated parameters κ = 5.The volatility of the VIX index (or VXB) is then given by 1 100 = 2 VIX t Bσ V Vt . Ft T (11) Because FtT and VXBt are very close to each other and ∆ < 1 the volatility of the VIX futures is smaller than the volatility of VXB (VIX).844.3501. goes to zero for the VIX futures with a very long maturity.55. which is close to that computed from the market values of the VIX index. θ = 0. Vt = 0.
VIX and VIX futures with three different maturities are almost perfectly correlated. 157.844 = 0. 0.3042 × σ F 90 = 0. 144. volatility instruments. we first study the behavior of VIX futures prices using the market data from March 2004 to November 2006.132. 5. 16 .σ F 30 = 0. is emerging and markets that trade these instruments are created.5 × 0. Conclusion With the enormous increase in derivatives trading and the focus on volatility came the realization that stochastic volatility is an important risk factor affecting pricing and hedging.844 = 0.5321 × 135. A new asset class. 3.5 × 0.300 and 0.844 = 0. VIX futures.357. 0. In this paper.228.8 135. We observe three stylized facts: 1.5 × 0. Our simple model also explains the downward sloping term structure of the volatility of VIX futures. The first exchange traded instrument is.1819 × which is close to the value. VIX. and the 3 VIX futures prices are negatively correlated with the S&P 500 index. It has been trading on the CBOE Futures Exchange since 26 March 2004. The index.8 σ F 60 = 0.8 135. 152.258 computed from the market prices.409. The term structure of the average VIX futures price is upward sloping. The volatility term structure of VIX futures is downward sloping. 2.
pp. Bakshi. 417427. P. 6165. 2003. Peter. Carr. we can price VIX futures at time t conditional on VIX at time t. Peter. 1989. Working paper. Carr. References: 1. Carr. and Liuren Wu.). and Liuren Wu. In the second part of the paper we use a simple meanreverting variance model to establish the theoretical relationship between VIX futures prices and its underlying spot index. July/August. Menachem. Review of Financial Studies 16. The second fact shows that the long term mean level of volatility is higher than the current level. 17 . which is less than 2% of the average VIX futures price of around 150. 1998. The root mean squared error between the market and model prices is around 2. 5. Journal of Derivatives 13. Gurdip. City University of New York and Bloomberg L. and Dan Galai. 2006. Brenner. In Robert Jarrow (Ed. Volatility estimation techniques for pricing derivatives. London: Risk Books. Peter. Our simple meanreverting variance model captures the dynamics of VIX futures price very well. New Financial Instruments for Hedging Changes in Volatility. Financial Analyst Journal.3 dollars. and Nikunji Kapadia. A tale of two indices. Towards a theory of volatility trading. 3. and Dilip Madan. Deltahedged gains and the negative market volatility risk premium. 2.The first fact is not surprising. 1329. Traders often take long position in volatility derivatives to hedge the risk in stock position. 2003. 4. Variance risk premia. The third observation is not wellknown. 527566. Using the variance parameters calibrated from our model with the market data at t1. An empirical study over the whole sample period shows that our model provides prices that are very close to the market prices.
Theory of Rational Option Pricing. 8. Jr. Steven L. Miller. Jin E. 385407. Ross. and Thomas W. 339373. The forecast quality of CBOE implied volatility indexes. 7. Journal of Derivatives 6. 2006. Demeterfi. A closedform solution for options with stochastic volatility with applications to bond and currency options. Econometrica 53. A theory of the term structure of interest rates. VIX Futures.. and Tyler Shumway. Michael Kamal. Merton. Journal of Futures Markets 25. Zhang. Jonathan E. and Stephen A. Coval. 18 . Emanuel Derman. Cox.. 11. 2005. 327343. Charles J. 1993. 9831009. Corrado... Journal of Futures Markets 26. 932. Expected option returns... 2001. and Yingzi Zhu. Bell Journal of Economics and Management Science 4.6. Review of Financial Studies 6. 1985. Ingersoll. Robert C. 1999. 1973. Journal of Finance 56. 141183. John C. 521531. and Joseph Zou.. Heston. 12. 9.. Jr. 10. Joshua D. A guide to volatility and variance swaps. Kresimir.
8 29.4 18.5 193.4 156.9 12.9 147.2 124.2 127.7 156. of observations 38 56 35 100 42 37 164 62 168 37 168 61 186 86 188 42 40 185 43 42 187 43 42 164 42 43 178 45 24 182 24 24 172 109 Period covered Start 03/26/04 03/26/04 05/21/04 03/26/04 07/19/04 08/20/04 03/26/04 10/21/04 06/18/04 01/24/05 09/20/04 03/21/05 11/19/04 06/20/05 02/22/05 10/21/05 11/18/05 05/23/05 01/20/06 02/17/06 08/19/05 04/21/06 05/22/06 12/21/06 07/24/06 08/18/06 03/08/06 09/21/06 10/20/06 03/08/06 10/20/06 10/20/06 03/22/06 06/21/06 End 05/19/04 06/16/04 07/14/04 08/18/04 09/15/04 10/13/04 11/17/04 01/19/05 02/16/05 03/16/05 05/18/05 06/15/05 08/17/05 10/19/05 11/16/05 12/21/05 01/18/06 02/15/06 03/22/06 04/19/06 05/17/06 06/14/06 07/19/06 08/16/06 09/20/06 10/18/06 11/15/06 12/20/06 01/17/07 02/21/07 03/21/07 04/18/07 05/16/07 08/15/07 VIX Futures prices Mean 185.0 7.1 146.5 150.6 11.2 178.2 29.3 22.6 143.9 23.9 5.1 27.3 15.1 150.4 8. which is defined as follows: JanuaryF. FebruaryG.6 20.8 131.1 140.9 4.8 163. 19 . The expiration month codes follow the convention for all commodities futures.5 16.7 136.5 142.0 25.9 11.1 150.1 Open interest Mean 1166 1256 1926 1697 1533 1226 2794 1210 3074 1181 2629 1178 4403 826 3303 928 305 3416 677 648 5627 1905 1293 10487 3788 8220 14957 6849 406 3839 508 36 1183 177 2784 Volume Mean 148 170 186 135 158 124 145 75 124 154 157 138 172 66 129 84 52 146 64 61 281 279 174 467 412 685 604 408 72 190 28 7 49 6 181 Note: The futures contract code is the expiration month code followed by a digit representing the expiration year.3 30. JuneM.7 190.1 15.0 20. MayK.5 161.7 19. SeptemberU.2 14.Table 1: Summary Statistics for VIX Futures Contracts from 26 March 2004 to 21 November 2006 Code K4 M4 N4 Q4 U4 V4 X4 F5 G5 H5 K5 M5 Q5 V5 X5 Z5 F6 G6 H6 J6 K6 M6 N6 Q6 U6 V6 X6 Z6 F7 G7 H7 J7 K7 Q7 Contract May 04 Jun 04 July 04 Aug 04 Sep 04 Oct0 4 Nov 04 Jan 05 Feb 05 Mar 05 May 05 June 05 Aug 05 Oct 05 Nov 05 Dec 05 Jan 06 Feb 06 Mar 06 Apr 06 May 06 Jun 06 Jul 06 Aug 06 Sep 06 Oct 06 Nov 06 Dec 06 Jan 07 Feb 07 Mar 07 Apr 07 May 07 Aug 07 Average No.5 134.9 19.4 140. OctoberV.1 148.6 19. JulyN. AugustQ.9 17. NovemberX and DecemberZ. AprilJ.2 154.5 Std 9.0 150.3 164.7 122.7 122.2 12.1 32.4 146.5 30.3 23.4 12.9 148.1 171. MarchH.8 122.2 27.6 32.3 13.6 123.6 186.6 152.
Table 2: The correlation matrix between daily continuously compounded returns of S&P 500 index. The daily continuously compounded return is defined as the logarithm of the ratio between the price on next day and the price on current day.8249 1 VXB 30day VIX futures 60day VIX futures 90day VIX Futures 20 . S&P 500 index S&P 500 index VXB 30day VIX futures 60day VIX futures 90day VIX futures 1 0.6918 0.8420 0.7748 0. VXB and fixed maturity VIX futures computed based on the market data from 26 March 2004 to 21 November 2006.8075 1 0.6827 0.8140 0.7974 0. The fixed maturity VIX futures prices are constructed by using the market data of available contracts with linear interpolation technique.6715 1 0.6975 1 0. The upper triangle is the same as the lower triangle because the matrix is symmetric.
250 200 30day VIX Futures 150 100 50 0 0 50 100 150 VXB 20051126 200 250 2006926 300 Figure 2. The fixed maturity VIX futures prices are constructed by using the market data of available contracts with linear interpolation technique. The 30day VIX futures prices are constructed by using the market data of available contracts with linear interpolation technique. VXB and VIX futures price with three fixed timetomaturities between 26 March 2004 and 21 November 2006. The VXB time series is from the CBOE. The relation between 30day VIX futures and VXB (March 26 2004 – 21 November 2006). The VXB time series is from the CBOE.300 250 200 150 100 50 0 2004326 2004526 2004726 2004926 2005126 2005326 2005526 Trading Volume VXB 30day VIX Futures 60day VIX Futures 90day VIX Futures 2005726 2005926 2006126 2006326 2006526 2006726 20041126 Figure 1. The bar chart shows the trading volume (normalized by 100 contracts) of futures of all maturity on the day. 21 .
4 0.8) (90.8) 90 Time to Maturity Figure 3.5) (30. The annualized volatility is computed based on the fixed maturity data from 26 March 2004 to 21 November 2006.1 0.7 0.258 30 60 90 Time to Maturity Figure 4.0 0 0.844 0.3 0.135.8 0.6 0. The average VIX futures price is computed based on the fixed maturity data from 26 March 2004 to 21 November 2006.5 0. The fixed maturity VIX futures prices are constructed by using the market data of available contracts with linear interpolation technique.9 Volatility of VIX Futures Price 0.160 Average VIX Futures Price 155 150 145 140 135 130 125 120 0 30 60 (0.152. The fixed maturity VIX futures prices are constructed by using the market data of available contracts with linear interpolation technique.300 0.8) (60. 22 .2 0. The annualized volatility of VIX futures price.357 0.157. 0. The term structure of average VIX futures price.144.
08 θP 0.04 0.06 0.02 01Jun04 1 01Dec04 01Jun05 01Dec05 01Jun06 λ 0 1 2 3 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 23 .10 9 8 7 6 5 4 κP 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 0.
8 0.2 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 Figure 5.1375 respectively. λ and σ V are 6.6 0.7882 and 0. The average values of the estimated four parameters κ P . θ P . The variance parameters estimated from the historical VIX index of the last 42 trading days.3501 respectively.9914.4 0.1 σV 0.02492. 1. Their standard deviations are 0. 24 . 0. 0. 0.02046 and 0.9671.01298.
18). 60. 26. 21.8.8. A comparison of model predicted prices and market prices of 30. The upper solid line is the market price.97. The dots are model predicted prices based on the parameters estimated from the VIX index of the last 42 trading days.89.8). The average prices of 30. The lower solid line is the VXB value shifted downward by 90 for the purpose of an easy comparison. 152. The roots of mean squared error between model predicted price and market price are (14.and 90day VIX futures.and 90day VIX futures are (144. 60.30day VIX Futures Price and VXB 250 200 150 100 50 01Jun04 250 200 150 100 50 01Dec04 01Jun05 01Dec05 01Jun06 60day VIX Futures Price and VXB 01Jun04 250 200 150 100 50 01Dec04 01Jun05 01Dec05 01Jun06 90day VIX Futures Price and VXB 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 Figure 6. 157. 25 .
26 .2 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 Figure 7.5805.5885 respectively. 0. The average values of the three parameters κ .03 0.02 01Jun04 1 01Dec04 01Jun05 01Dec05 01Jun06 θ 01Dec04 01Jun05 01Dec05 01Jun06 σV 0.00909 and 0. The model parameter values calibrated from the market prices of 30.03259 and 0.04 0.10 9 8 7 6 5 4 κ 01Jun04 0.2398 respectively.5246. 0.4 0.07 0.08 0. 60and 90day VIX futures. Their standard deviations are 1.6 0. θ and σ V are 5.06 0.8 0.05 0.
0.2 0.09 100 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 Figure 8. the lower line is VIX index divided by 100 and shifted down by 9% for the purpose of an easy comparison.25 0.1 0. The upper line is the instantaneous variance.15 0. Vt . recovered from VIX index and daily calibrated set of parameters. The instantaneous volatility. 27 .05 0 è ! !!! Vt VIXt −0.
and 90day VIX futures. 60. 152. The average prices of 30. 28 .27).and 90day VIX futures are (144.8).200 30day VIX Futures and VXB 150 100 50 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 200 60day VIX Futures and VXB 150 100 50 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 200 90day VIX Futures and VXB 150 100 50 01Jun04 01Dec04 01Jun05 01Dec05 01Jun06 Figure 9. 2. 157.8.34. 60. The upper solid line is the market price.8. The roots of mean squared error between model price and market price are (2. The dots are model predicted prices based on the parameters calibrated from VIX futures prices on the previous day. A comparison on model predicted prices and market prices of 30.56. The lower solid line is the VXB value shifted downward by 90 for the purpose of an easy comparison. 2.
03259 and σ V = 0.1 0. The volatility hedge ratio as a function of time to maturity. 29 . θ = 0. VXB0 = 135. the set of parameters is taken to be the average calibrated values κ = 5.2 0 20 40 60 Tim to m e aturity HdayL 80 Figure 10.8 Hedge ratio 0.5.5885.6 0.4 0.5805.