.
.
. ..
.
.
An Analytical Approximation for Pricing
VWAP Options
Hideharu Funahashi and Masaaki Kijima
Graduate School of Social Sciences, Tokyo Metropolitan University
September 4, 2015
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 1 / 24
This Talk is Based on
...1 Funahashi, H. and Kijima, M. (2015c), “An analytical approximation
for pricing VWAP options,” Working Paper.
...2 Funahashi, H. and Kijima, M. (2015b), “A unified approach for the
pricing of options related to averages,” Working Paper.
...3 Funahashi, H. and Kijima, M. (2014), “An extension of the chaos
expansion approximation for the pricing of exotic basket options,”
Applied Mathematical Finance, 21 (2), 109–139.
...4 Funahashi, H. and Kijima, M. (2015a), “A chaos expansion approach
for the pricing of contingent claims,” Journal of Computational
Finance, 18 (3), 27–58.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 1 / 24
Formulation
Let St and vt be the time-t price and trading volume, respectively, of
the underlying asset.
VWAP (volume weighted average price) is determined by
MT =
∫ T0 vtStdt∫ T0 vtdt
,
where MT is called the VWAP of the time interval [0, T ].
The standard definition of a continuous VWAP call option is given by
VC(S, v,K, T ) = e−rTE[(MT − K)+]
where S = S0 is the initial price, v = v0 is the initial trading
volume, K is a strike, T is a maturity, and r is the short rate, E is
the risk-neutral expectation operator.
Existing papers try to approximate the distribution of MT directly.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 2 / 24
Motivation
VWAP Options are becoming increasingly popular in options markets,
since they can help corporate firms hedge risks arising from market
disruption when entering large buy or sell orders.
Their prices assign more weight to periods of high trading than to
periods of low trading in its calculation.
Hence, VWAP options differ conceptually from Asian options because
the resulting payoff is not a linear combination of underlying prices.
As a result, the pricing of VWAP options is significantly more difficult
than Asians and few pricing models have been proposed in the
literature, despite their popularity in practice.
See, e.g., Buryak and Guo (2014), Novikov et al. (2013) for details.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 3 / 24
Literature Review
It is common to model the underlying St by using the geometric
Brownian motion (GBM) for simplicity.
For the trading volume vt, Stace (2007) proposes a mean-reverting
process; Novikov et al. (2013) use a squared Ornstein–Uhlenbeck
(OU) process; and Buryak and Guo (2014) suggest a simple gamma
process, respectively, for the trading volume process.
Under the GBM assumption, these papers produce approximated
pricing formulas by utilizing the moment-matching technique for MT .
On the other hand, Novikov and Kordzakhia (2013) derive very tight
upper and lower bounds for the price of VWAP options.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 4 / 24
In This Talk
Other than the GBM case, these approaches seem difficult to apply
for deriving an approximation formula of VWAP options.
Funahashi and Kijima (2015b) apply the chaos expansion technique
to derive a unified approximation method for pricing any type of
Asian options when the underling process follows a diffusion.
In this talk, not of the VWAP itself, but we try to approximate the
distribution of MT ,
MT (x) =
∫ T
0vtStdt − x
∫ T
0vtdt,
when the underling asset price and trading volume processes follow a
local volatility model and a mean-reverting model, respectively.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 5 / 24
The Setup
We assume that the price St and the trading volume vt of the
underlying asset are modeled by the following SDE:dSt
St= r(t)dt + σ(St, t)dWt
dvt = (θ(t) − κ(t)vt)dt + γ(vt)dWvt
under the risk-neutral measure Q, where {Wt} and {W vt } are the
standard Brownian motions with correlation dWtdWvt = ρdt.
The volatility functions σ(S, t) and γ(v) are sufficiently smooth with
respect to (S, t) and v, respectively.
r(t), θ(t) and κ(t) are some deterministic functions of time t.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 6 / 24
Key Observation
Denote the cumulative distribution function (CDF) of MT by
FM(x) = Q{MT ≤ x}, x > 0.
The VWAP call price can be written as
VC(S, v,K, T ) = e−rT
∫ ∞
K(1 − FM(x))dx
For each x > 0, let FM,x
(y) be the CDF of the random variable
MT (x) =
∫ T
0vtStdt − x
∫ T
0vtdt
It follows from the definition of VWAP that
FM(x) = FM,x
(0), x > 0
Therefore, it suffices to know the CDF FM,x
(y) at y = 0.
To this end, we apply the chaos expansion approach to approximate
the distribution of the random variable MT (x).
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 7 / 24
Approximation 1
Employing the same idea as in Theorem 3.1 of Funahashi and Kijima
(2015a), St is approximated by the following formula:.Lemma..
.
. ..
.
.
Let F (0, t) = Se∫ t0 r(u)du be the forward price of the underlying asset
with delivery date t. Then,
St ≈ F (0, t)
[1 +
∫ t
0
p1(s)dWs +
∫ t
0
p2(s)
(∫ s
0
σ0(u)dWu
)dWs
+
∫ t
0
p3(s)
(∫ s
0
σ0(u)
(∫ u
0
σ0(r)dWr
)dWu
)dWs
+
∫ t
0
p4(s)
(∫ s
0
p5(u)
(∫ u
0
σ0(r)dWr
)dWu
)dWs
]
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 8 / 24
Approximation 1, Continued
where
p1(s) := σ0(s) + F (0, s)σ′0(s)
(∫ s
0σ20(u)du
)+
1
2F 2(0, s)σ′′
0 (s)
(∫ s
0σ20(u)du
)p2(s) := σ0(s) + F (0, s)σ′
0(s)
p3(s) := σ0(s) + 3F (0, s)σ′0(s) + F 2(0, s)σ′′
0 (s)
p4(s) := σ0(s) + F (0, s)σ′0(s)
p5(s) := F (0, s)σ′0(s)
with σ′0(t) := ∂xσ(x, t)|x=F (0,t) and σ′′
0 (t) := ∂xxσ(x, t)|x=F (0,t)
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 9 / 24
Approximation 2
Employing the successive substitution used in Funahashi (2014), we obtain
the following result. Let E(0, t) = e∫ t0 κ(u)du, E(t) = 1/E(t), and
Vt = E(t)
(v0 +
∫ t
0
E(s)θ(s)ds
).Lemma..
.
. ..
.
.
The trading volume vt is approximated as vt ≈
V (0, t) + E(t)∫ t
0p6(s)dW
vs + E(t)
∫ t
0γ′0(s)
(∫ s
0E(u)γ0(u)dW
vu
)dW v
s
+ E(t)∫ t
0E(s)γ′′
0 (s)(∫ s
0E(u)γ0(u)
(∫ u
0E(r)γ0(r)dW
vr
)dW v
u
)dW v
s
+ E(t)∫ t
0γ′0(s)
(∫ s
0γ′0(u)
(∫ u
0E(r)γ0(r)dW
vr
)dW v
u
)dW v
s ,
where γ0(t) := γ(Vt), γ′0(t) := ∂xγ(x)|x=Vt , γ
′′0 (t) := ∂xxγ(x)|x=Vt , and
p6(t) := E(t)γ0(t) +1
2E(t)γ′′
0 (t)
(∫ t
0
E2(s)γ20(s)ds
)
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 10 / 24
Approximation 3
Using the approximation results for St and vt, we obtain the following..Lemma..
.
. ..
.
.
The traded value vtSt is approximated as
vtSt ≈ I0(t) + I1(t) + I2(t) + I3(t),
where I0(t) = VtF (0, t) + ρF (0, t)E(t)∫ t0 p6(s)p1(s)ds,
I1(t) = V (0, t)F (0, t)
∫ t
0
p1(s)dWs + F (0, t)E(t)
∫ t
0
p10(t, s)dWs
+F (0, t)E(t)
∫ t
0
(p6(s) + p9(t, s)) dWvs ,
and others (omitted).
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 11 / 24
Approximation 4
By changing the order of integration, MT can be approximated by a
truncated sum of iterated Ito stochastic integrals as follows..Lemma..
.
. ..
.
.
For each x > 0, the random variable MT can be approximated as
MT =
∫ T
0vtStdt − x
∫ T
0vtdt
≈ J0(x, T ) + J1(x, T ) + J2(x, T ) + J3(x, T ),
where
J0(x, T ) =
∫ T
0Vt (F (0, t) − x) dt
+ ρ
∫ T
0F (0, t)E(t)
(∫ t
0p6(s)p1(s)ds
)dt,
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 12 / 24
Approximation 4, Continued
J1(x, T ) =
∫ T
0
p1(t)
(∫ T
t
V (0, s)F (0, s)ds
)dWt
+
∫ T
0
(∫ T
t
p10(s, t)F (0, s)E(s)ds
)dW v
t
+
∫ T
0
(∫ T
t
(p6(t) + p9(s, t)F (0, s)E(s)
)ds
)dW v
t
−x
∫ T
0
p6(t)
(∫ T
t
E(s)ds
)dW v
t ,
J2(x, T ) =
∫ T
0
r1(t)
(∫ t
0
σ0(s)dWs
)dWt +
∫ T
0
r2(t)
(∫ t
0
p6(s)dWvs
)dWt
+
∫ T
0
r3(t)
(∫ t
0
p1(s)dWs
)dW v
t +
∫ T
0
r4(x, t)
(∫ t
0
E(s)γ0(s)dWvs
)dW v
t ,
and others (omitted).
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 13 / 24
Option Pricing Formula
Let us define Yt = Mt − J0(x, t), and denote its probability density
function (PDF) by fYT ,x(y).
We can obtain the PDF by applying the following lemma.
.Lemma..
.
. ..
.
.
The PDF of YT is approximated as
fYT ,x(y) ≈ n (y; 0, Vx(T )) − ∂
∂y{E[J2(x, T )|J1(x, t) = y]n (y; 0, Vx(T ))}
− ∂
∂y{E[J3(x, t)|J1(x, t) = y]n (y; 0, Vx(T ))}
+1
2
∂2
∂y2
{E[J2(x, t)
2|J1(x, t) = y]n (y; 0, Vx(T ))},
where n(y; a, b) denotes the normal density with mean a and variance b.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 14 / 24
Option Pricing Formula, Continued
The conditional expectations can be evaluated explicitly.
Using the approximated density function of fYT ,x(y), we can
approximate the CDF FYT ,x(y) of YT .
But, from the relation FM,x
(y) = FYT ,x(y − J0(x, T )), we have
FM(x) = FM,x
(0) = FYT ,x(−J0(x, T ))
It follows that the VWAP call option price can be approximated as
VC(S, v,K, T ) ≈ e−rT
∫ ∞
K(1 − FYT ,x(−J0(x, T )))dx
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 15 / 24
Numerical Examples: CEV Case
We suppose that the volatilities are specified as
σ(S, t) = σSβ−1, γ(v) = ν vλ−1,
where σ, β, ν, and λ are some constants.
The base-case parameters are set to be S = 100, K = 100,
T = 1, r(t) = 3.0%, v = 100, and ρ = 0.3.
Also, we set θ(t) = 10 and κ(t) = 0.1, i.e., the long-run average of
the trading volume is 100.
As to the volatilities, we consider (H) high and (L) low volatility cases
in which we set σSβ−1 = 30% and ν vλ−1 = 30% for case (H)
and σSβ−1 = 15% and ν vλ−1 = 15% for case (L), respectively.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 16 / 24
Numerical Examples; Accuracy Check
We consider two cases; (1) log-normal case (β = 1 and λ = 1), and
(2) square-root case (β = 0.5 and λ = 0.5).
Figure 1 shows option prices for (L) with short maturity (T = 0.5)
when (1), whereas Figure 2 depicts for (2).
Through the numerical experiments, it is observed that the effect of
volatility and maturity appears only around ATM (K = 100), and
the volatility effect is stronger than the maturity effect.
As to the accuracy of our approximation, we find that the difference
between our approximation and the Monte Carlo result are very small.
The error becomes slightly larger for long maturity and high volatility
cases; however, for practical uses, the errors are sufficiently small.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 17 / 24
Figure 1 (GBM, low vol, T = 0.5)
0
5
10
15
20
25
80 90 100 110 120-0.1
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1O
ptio
n P
rice
Dif
f
Strike
MCWIC (2nd)
Diff
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 18 / 24
Figure 2 (Square-Root, low vol, T = 0.5)
0
5
10
15
20
25
80 90 100 110 120-0.1
-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1O
ptio
n P
rice
Dif
f
Strike
MCWIC (2nd)
Diff
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 19 / 24
Other Findings; Effect of Correlation
The effect of correlation gets bigger as κ, the speed of mean
reversion, becomes smaller.
When κ is large, the trading volume vt sticks around the long-run
average so as to behave as if it were uncorrelated to the stock price.
Stace (2007) sets κ = 100 under the assumption ρ = 0.
Our result suggests that, when κ = 100, the impact of correlation on
the VWAP call option prices is negligible.
This result may be an important message for practitioners, because it
is in general very difficult to estimate the correlation accurately.
The effect of correlation gets bigger as the maturity T becomes
longer and the volatility σ of the asset price becomes larger.
These results can be understood by the fact that the effect of
correlation is bigger as more uncertainty is involved.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 20 / 24
Other Findings; Effect of Model Choice
Recall that
σ(S, t) = σSβ−1, γ(v) = ν vλ−1
The effect of β gets bigger as κ becomes smaller, the maturity
becomes longer and the asset volatility becomes larger.
These results can be explained by the exactly same reason as above.
The effect of λ gets bigger as κ becomes smaller, and has less impact
on the others.
Compared with the impact of the underlying asset price, the maturity
as well as the volatility of trading volume has less impact on the
VWAP option prices.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 21 / 24
Some Extensions
Squared OU Model for Trading Volume as in Novikov et al. (2014).
vt = X2t + γ, dXt = (θ − κXt)dt + βdW v
t ,
where θ, κ, γ and β are some constants. In this case, we have
X2t = V 2
t + 2VtE(t)
∫ t
0
βE(s)dW vs + E2(t)
(∫ t
0
β2E2(s)ds
)+2E2(t)
(∫ t
0
βE(s)
(∫ s
0
βE(u)dW vu
)dW v
s
)
Generalized VWAP MT =∫ T0 w1
t vtStdt∫ T0 w2
t vtdt, where wi
t is a deterministic
function of time t. Consider a floating-strike VWAP option defined by
VC(S, v,K, T ) = e−rTE[(MT − ST )+]
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 22 / 24
Conclusion
In this talk, we develop a unified approximation method for options
whose payoff depends on a volume weighted average price (VWAP).
Compared to the previous works, our method is applicable to the local
volatility model, not just for the geometric Brownian motion case.
Moreover, our method can be used for any special type of VWAP
option, including ordinary Asian and Australian options, with
fixed-strike, floating-strike, continuously sampled, discretely sampled,
forward starting, and in-progress transactions.
Through numerical examples, we show that the accuracy of the
second-order approximation is high enough for practical use.
Our approximation get slightly worse for long maturity and high
volatility case; in such a case, 3rd-order may be required.
Kijima (TMU) Pricing of VWAP Options AMMF2015 @ 04/09/15 23 / 24