VIX, Variance Swap, and Volatility Products

Min Dai
National University of Singapore

Contents
1 VIX, Variance Swap, and Volatility Products 1
1.1 Estimate of historical volatility . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Implied volatility and old VIX . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Variance swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.1 Realized volatility and volatility products . . . . . . . . . . . . . . . . 3
1.3.2 Continuous-time approximation to realized volatility . . . . . . . . . 3
1.3.3 Variance swap rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 New VIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.1 Definition of VIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.2 Linkage between VIX and variance swap rate . . . . . . . . . . . . . 6
1.4.3 How to use VIX or variance/volatility swap? . . . . . . . . . . . . . . 7
1.5 Pricing models for general volatility products . . . . . . . . . . . . . . . . . . 7
1.5.1 A local volatility model . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5.2 Discrete monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1 VIX, Variance Swap, and Volatility Products

Volatility is a measure for variation of price of a financial instrument over time.

1.1 Estimate of historical volatility

Recall the price process of an underlying asset in the BS model:
dSt
= µdt + σdBt . (1)
St
A historical (constant) volatility is derived from time series of past market prices of
the underlying under the assumption of (1) which is rewritten as

σ2
 
d ln St = µ − dt + σdBt .
2

1
 Denote ti = iδt, i = 0, 1, ..., N and Si := Sti . Consider the discretization of the above
SDE:
σ2
 
Si
ln = ln Si − ln Si−1 ≈ µ − δt + σφδt1/2 , i = 1, 2, ..., N,
Si−1 2
where E (φ) = 0, V ar (φ) = E (φ2 ) = 1. It follows
2  2
σ2
 
Si 1/2
ln = µ− δt + σφδt ≈ σ 2 φ2 δt,
Si−1 2

namely, " 2 #
Si
E ln ≈ σ 2 δt.
Si−1
Then we can estimate the historical volatility by
v
u
u1 1 X N  2
S i
σ=t ln . (2)
δt N i=1 Si−1

1.2 Implied volatility and old VIX

Consider a call option on the underlying. We define σimp as

CM KT (S, 0; K, T ) = C(S, 0; K, T )
≡ Se−qT N (d1 (σimp )) − Ke−rT N (d2 (σimp )) ,

where  
S σ2
1
Z x
ξ2
ln + r − q ± K 2
T
N (x) = √ e− 2 dξ, d1,2 (σ) = √ .
2π −∞ σ T
Note that the implied volatility is well defined because
∂C
> 0.
∂σ
However, if the market price is beyond the range of the theoretical price, then you would be
unable to find an implied volatility.
Theoretically, due to the put-call parity, the implied volatility computed from the call
should be the same as that from the put with the same strike price and maturity. However,
it may not the case in practice.
The definition of the old VIX is based on the implied volatility of at-the-money options
on S&P100.

1.3 Variance swap

The new VIX is linked to the variance swap which is a volatility product.

2
1.3.1 Realized volatility and volatility products
(2) motives us to define the realized volatility from time 0 to T = N ∆t :
v
u
u 252 X N  2
Si
RealizedVolatility = t ln ,
N i=1 Si−1

where we use daily prices Si (so δt = 1/252), and N is number of business days from and
including effective date to maturity date.
In terms of the realized volatility, we can define the payoff of
• variance swap:
Notional Amount × (RealizedVolatility2 − FixedVol2 )
N  2 !
252 X Si
= Notional Amount × ln − Xvar
N i=1 Si−1

• volatility swap:
Notional Amount × (RealizedVolatility − FixedVol)
v 
u
u 252 XN  2
Si
= Notional Amount × t ln − Xvol 
N i=1 Si−1

Without loss of generality we will assume that the notional amount is 1.

1.3.2 Continuous-time approximation to realized volatility

In reality, the volatility is not constant. In what follows we always assume
dSt
= µt dt + σt dBt ,
St
where σt is an adaptive process. Again,
σt2
 
d ln St = µt − dt + σt dBt .
2
Then
N  2 N
1 T 2
Z
252 X Si 1 X 2
ln ≈ σ δt → σ dt
N i=1 Si−1 N δt i=1 ti−1 T 0 t
as N → +∞.
Recall the payoff of a variance swap
N  2
252 X Si
ln − Xvar ,
N i=1 Si−1
whose continuous time limit is Z T
1
σt2 dt − Xvar .
T 0

3
1.3.3 Variance swap rate
Now let us examine the variance swap rate which is the fair value of Xvar such that the
variance swap has a zero value at time 0,
 Z T 
−rT b 1 2
e E σ dt − Xvar = 0,
T 0 t

namely,  Z T 
Xvar b 1
=E σt2 dt . (3)
T 0

In the risk neutral world,

σt2
 
d ln St = r − dt + σt dB
bt .
2

It follows
T T T T T
σt2
Z Z Z Z Z
bt − ST
dt = rdt + σ t dB d ln St = rT − ln + σt dB
bt
0 2 0 0 0 S0 0
Z T Z T
ST ST
= − ln + σt dBt =: − ln
b + σ t dBbt ,
S0 erT 0 F 0

where F = S0 erT is the forward price.

Therefore, we can rewrite the payoff of the variance swap as

1 T 2
Z  Z T 
2 ST
σ dt = − ln + σt dBt .
b (4)
T 0 t T F 0
RT
Since 0 σt dB bt is a martingale in the risk-neutral world, the payoff of the variance swap
is equivalent to a log payoff
ST
− ln .
F
The log payoff has the following decomposition:
ST ST − F
− ln = − (forward contract)
F Z F +∞
1
+ (ST − K)+ dK (out of money call)
F
K2
Z F
1
+ 2
(K − ST )+ dK (out of money put) (5)
0 K

4
Proof of Identity (5): For any smooth function f (x), one has
Z x
0
f (x) = f (a) + f (a)(x − a) + f 00 (K)(x − K)dK
Za x
= f (a) + f 0 (a)(x − a) + f 00 (K) (x − K)+ − (K − x)+ dK
 

Za x Z a
0 00
= f (a) + f (a)(x − a) + +
f (K)(x − K) dK + f 00 (K)(K − x)+ dK
Za +∞ x
Z a
0 00
= f (a) + f (a)(x − a) + +
f (K)(x − K) dK + f 00 (K)(K − x)+ dK.
a 0

The desired result follows by choosing f (x) = − ln Fx , a = F, and x = ST . This completes

the proof.

Remark: The above result implies that a variance swap can be replicated by vanilla options.

The combination of (3)-(5) yields

Z +∞ Z F 
2 1 b + 1 b +
Kvar = E (ST − K) dK + E (K − ST ) dK
T F K2 0 K
2
Z +∞ Z F 
2 1 rT 1 rT
= e C(S0 , 0; K, T )dK + e P (S0 , 0; K, T )dK , (6)
T F K2 0 K
2

where C and P are the prices of European call and put, respectively.

1.4 New VIX

1.4.1 Definition of VIX
The (new) VIX index is a volatility index for S&P 500, which is defined as
s  
365 N2 − 30 30 − N1
V IX = 100 T1 VT1 + T2 VT2 ,
30 N2 − N1 N2 − N1

where  2
2 X ∆Ki rT 1 F
VT = e Q (Ki ) − −1 , (7)
T i Ki2 T K0
and

(i) N1 and N2 denote the number of actual days to expiration for the two nearest maturities
(When the nearest time to maturity is 8 days or fewer, the CBOE switches to the next-
nearest maturity in order to avoid microstructure effects).

(ii) Q(Ki ) : The midpoint of the bid-ask spread for each out of money option with strike
Ki .

5
(iii) F : the forward S&P 500 index level (F = S0 erT by assuming constant interest rate in
our model). In reality, we use

F = erT [C(K) − P (K)] + K,

where we choose a pair of put and call options with prices that are closest to each
other.

(iv) K0 : the first strike below F, namely K0 ≤ F.

(v) Ki : Strike price of the ith out-of-the money option; a call if Ki > K0 ; and a put if
Ki < K0 ; both put and call if Ki = K0 , that is, Q(K0 ) represents the average of the
call and put option prices at this strike.

(vi) ∆Ki = (Ki+1 − Ki−1 )/2; ∆K for the lowest strike is simply the difference between the
lowest strike and the next higher strike. Likewise, ∆K for the highest strike is the
difference between the highest strike and the next lower strike.

1.4.2 Linkage between VIX and variance swap rate

If K0 = F, then (7) is a straightforward discretization of (6). If K0 < F, then replacing F
with K0 in (5) gives
+∞ K0
ST − K0
Z Z
ST 1 1
− ln =− + 2
(ST − K)+ dK + 2
(K − ST )+ dK.
K0 K0 K0 K 0 K

Then
   
2 b ST 2 b ST F
Kvar = E − ln = E − ln + ln
T F T K0 K0
 
2 b ST − K0 F
= E − + ln
T K0 K0
Z +∞ Z K0 
2 1 rT 1
+ e C(S0 , 0; K, T )dK + P (S0 , 0; K, T )dK
T K0 K 2 0 K2

Note that
 
S T − K0 F F − K0 F
E −
b + ln = − + ln
K0 K0 K0 K0
   2
F − K0 F 1 F F
= − + −1 − − 1 + O( − 1)3
K0 K0 2 K0 K0
 2
1 F F
= − − 1 + O( − 1)3 ,
2 K0 K0

which leads to the desired result by omitting O( KF0 − 1)3 .

6
1.4.3 How to use VIX or variance/volatility swap?
The presumption is that as equity markets fall volatility tends to rise.

• Going long a variance/volatility swap or VIX futures can provide an offset for a long-
only fund in falling market conditions.
• Investors seeking to hedge against decreased liquidity since liquidity tends to decrease
during increased levels of volatility
• Insurance companies that might like to hedge some of their underlying business expo-
sures to volatility in the marketplace.
• Option trading firms may want to use VIX futures or variance/volatility swaps to offset
their exposure to market fluctuations. Additionally, option trading firms typically
benefit from higher volatility since it correlates to increased trading activity.

1.5 Pricing models for general volatility products

1.5.1 A local volatility model
The simplest model that is internally consistent is to just use local volatility (deterministic)
σt = σ(St , t).
dSt
= µdt + σ(St , t)dBt .
St
where
σ (., .) is a deterministic function,
Then v
u N 
u 252 X 2 s Z T
Si 1
t ln = σ 2 (St , t)dt.
N i=1 Si−1 T 0
The payoff of a volatility product is then written as

f (IT )

where
s Z
1 t
It = σ(Sτ , τ )2 dτ .
t 0
Recall the pricing of Asian options:
 Z T +
1
Sτ dτ − K .
T 0
Introduce a path-dependent variable:
Z t
1
At = Sτ dτ.
t 0

7
Here
St − At
dAt = dt.
t
The option value V = V (St , At , t) satisfies

∂V 1 ∂ 2V ∂V S − A ∂V
+ σ 2 (S, t) S 2 2 + rS − rV + =0
∂t 2 ∂S ∂S t ∂A
in S > 0, A > 0, t ∈ [0, T ) with

V (S, A, t) = (A − K)+ .

For the current product,

V = V (St , It , t)
∂V 1 ∂ 2V ∂V σ 2 − I 2 ∂V
+ σ 2 (S, t) S 2 2 + rS − rV + =0
∂t 2 ∂S ∂S 2tI ∂I
because
− 21 2 Rt 2
1 1 t −
 Z
σ t σ dτ
dIt = σ(Sτ , τ )2 dτ 2
0
dt
2 t 0 t
1 σ 2 t − tI 2
= dt
2 It2
σ2 − I 2
= dt.
2tI

Detailed derivations: V = V (St , It , t), and

2
 
∂V 1 2 2∂ V ∂V ∂V
dV = + σ (S, t)S dt + dS + dIt
∂t 2 ∂S 2 ∂S ∂I
2
σ 2 − I 2 ∂V
 
∂V 1 2 2∂ V ∂V
= + σ (S, t)S dt + dS + dt
∂t 2 ∂S 2 ∂S 2tI ∂I

The pricing model is

∂V 1 ∂ 2V ∂V σ 2 − I 2 ∂V
+ σ 2 (S, t)S 2 2 + rS − rV + = 0,
∂t 2 ∂S ∂S 2tI ∂I
in t ∈ [0, T ), S > 0, I > 0,
V |t=T = f (I)

A note: we can use another path-dependent variable

Z t
Jt = σ 2 (Sτ , τ )dτ,
0
U = U (St , Jt , t)

8
 2
∂U 1 2 2∂ U ∂U 2 ∂U
+ σ (S, t)S + rS − rU + σ (S, t) =0
∂t 2 ∂S 2 ∂S ∂J
in t ∈ [0, T ), S > 0, J > 0,
J
U |t=T = f ( )
T
Equivalence between two PDE models:
r
J
U (S, J, t) = V (S, I, t), I =
t

∂U ∂V ∂V ∂I
= +
∂t ∂t ∂I ∂t
2
∂U ∂V ∂ U ∂ 2V
= ; = ;
∂S ∂S ∂S 2 ∂S 2
∂U ∂V ∂I
=
∂J ∂I ∂J

1.5.2 Discrete monitoring

Consider the variance swap whose payoff is
N  2
252 X Si
ln − Kvar
N i=1 Si−1

Assume
dS
= µdt + σdBt .
S
In contrast, the payoff of Asian options with discrete monitoring:
N
!+
1 X
Si − K .
N i=1

Apparently, we need to introduce a path-dependent variable:

X 2
Sj
Jt = ln .
t ≤t
Sj−1
j

Note that
X  2
Sj
Jti + = ln
tj ≤ti +
Sj−1
X  2  2
Sj Si
= ln + ln
tj ≤ti −
Sj−1 Si−1
 2
Si
= Jti − + ln
Si−1

9
 It should be emphasized that

V = V (St , yt , Jt , t)
where
yt = Si−1 , for t ∈ [ti−1 , ti ).
In this way, we have the connection condition:
 2
S
V (S, y, J, ti −) = V (S, S, J + ln , ti +).
y
For t ∈ (ti−1 , ti ),
1 2 2 ∂ 2V
 
∂V ∂V
dV (St , yt , Jt , t) = + σ S dt + dS
∂t 2 ∂S 2 ∂S
∂V ∂V
+ dy + dJ
∂y ∂J
1 2 2 ∂ 2V
 
∂V ∂V
= + σ S dt + dS
∂t 2 ∂S 2 ∂S
because

dJt = 0;
dyt = 0;

It follows
LBS V = 0
for any y > 0, J > 0, S > 0, t ∈ (ti−1 , ti ).
The terminal condition is
252
V (S, y, J, T + ) = J − Kvar
N
Dimension reduction
S
V (S, y, J, t) = U (x, J, t), x = .
y
Then the connection condition and the terminal condition become

U (x, J, ti −) = u(1, J + (ln x)2 , ti +)

and
252
U (x, J, T + ) = J − Kvar ,
N
respectively.
The original PDE
∂V 1 2 2 ∂ 2V ∂V
+ σ S 2
+ rS − rV = 0
∂t 2 ∂S ∂S
10
in (ti−1 , ti ) . Note that

∂V ∂U
=
∂t ∂t
∂V ∂U ∂x 1 ∂U
= =
∂S ∂x ∂S y ∂x
∂ 2V 1 ∂ 2 U ∂x 1 ∂ 2U
 
∂ 1 ∂U
= = = .
∂S 2 ∂S y ∂x y ∂x2 ∂S y 2 ∂x2

So,
∂U 1 ∂ 2U ∂U
+ σ 2 x2 2 + rx − rU = 0
∂t 2 ∂x ∂x
in t ∈ (ti−1 , ti ) , x > 0, J > 0.
Exercise: Variance products: exchange between the realized volatilities of two stocks
S1 and S2 .

11