Implied Volatility

0.1 Calculating implied volatility

The Black-Scholes formula gives us the value of an option as a function of
the following inputs:
- The current price of the underlying St
- The time to maturity τ = T − t
The risk-free interest rate r
- The volatility σ
How do we choose the value of r and σ ?

For r
The central bank sets a ’risk free’ interest rate, which might suggest a good
value of r. However, this rate is not representative of the cost of borrowing
(as we cannot borrow at the rate r in reality).
Practically, to get a good value of r from market data, we can look at the
forward price for S. Remember that the forward price (assuming constant
interest) is given by Ft = er(T −t) St . Hence log (St /Ft ) = r(T − t) which gives
us a representative interest rate for participants in the forward market.

For σ
One idea: estimate from historical data. Split the interval [t − N δ, t] with a
partition {t − N δ = t0 , . . . , tN = t} with |tn − tn+1 | = δ. Then an unbiased
and efficient estimator of σ 2 i
N 2
Sti+1 − Sti

2 1 X
σ =
(N − 1)δt Sti
i=1

This is natural. However this method has certain drawbacks due to the
fact that, in reality, σ varies significantly over time, and it is rather hard to
capture its ” most recent” value: the estimator becomes less reliable as we
decrease t, while, if we increase it, we obtain an ”averaged” value of σ over
time. It also has the problem that it only looks at the past, while the price
should reflect what the market expects to happen through the future.

Another way is to deduce σ from the prices currently observed in the market.
This gives rise to Implied Volatility:

1
Given a market price of an option V mktt , the [implied volatility] σimp is
the value such that
Vtmkt = V BS (St , t; σimp )

Implied volatility is very common for call and put option (in many cases,
markets quote prices in terms of implied volatility, rather than the price of
the option).
Implied Volatility is well defined, provided
◦ Vtcall,mkt is not impossible, that is,
 + 
Vtcall,mkt ∈ S − Ke−r(T −t)
,S

∂ call,BS
◦ And ∂σ V does not change its sign, as a function of σ.

The first condition is satisfied in practice (up to model error), since oth-
erwise there is a model-independent arbitrage. The second condition is satis-
fied, since, as we’ll see, the BS Vega is always nonnegative. This also makes
a simple numerical search algorithm work well.

0.2 Implied Smile and volatility surfaces

If the BS model was true, there would exist one value of implied volatility
σimp for call options of all strikes and maturities. However, this is not true
in practice. Typically, for each pair (T, K), we have a different value of
implied vol σimp (T, K). Plotted as a function of negative log-moneyness
x = log(K/S), this function is typically convex around x = 0 and, hence, is
often referred to as the [implied smile].
In equities (where S is the price of a stock or stock index), the implied
smile typically has a [negative skew], assigning higher values to negative
x = log(K/S) (i.e. K < S ). The implied volatility is often higher for long
and short maturity options, and lower for intermediate maturities.
By plotting the implied volatility for all traded values of (T, K), we obtain the
implied volatility surface. This is commonly used to give an understanding
of market perspectives on the riskiness of options trades, and as an input to
risk management and hedging decisions.

2
Greeks
Assume the market is described by a BS model, and denote the price function
of an option by V (S, t). Sensitivities of the option price V with respect to the
input variables (S and t) and parameters (σ and r) are called the Greeks.
These sensitivities are very important for hedging and risk management, as
they show how the value of the option changes with small changes in the
uncertain input!

0.3 ∆ (Delta)
Delta is defined as
∂
∆= V
∂S
and it is the primary sensitivity, as, even if the model is true, the value of
underlying will change, and its change is likely to be of a higher magnitude
than the time increment. Notice that
√
St+δt ≈ St µδt + St σξ δt

where ξ is a standard normal.

For a call option,
∆call = e−q(T −t) N (d+ )

By put-call parity,

∆put = ∆call − e−q(T −t) = −e−q(T −t) N (−d+ )

The ∆-hedging portfolio, consisting of ∆call units of S and Vt − ∆call S

units in bonds is an ”instantaneous perfect hedge’ of a short position in the
option, assuming the Black-Scholes Model. In more complicated models, and
in practice, it is a reasonably good hedge.
Example Suppose we are short a call option with r = 0.05, σ = 0.2, T =
1, K = S0 = 100.= Then

V call = 100 ∗ N ((0.05 + 0.02)/0.2) − 100 ∗ N ((0.05 − 0.02)/0.2) = 10.4506,

∆call = N ((0.05 + 0.02)/0.2) = 0.63683

We buy a portfolio consisting of ∆S = $63.683 invested in stocks (buy ∆ =

3
0.63683 stocks), −Vt = $10.4506 in the option (our short position), Vt −∆S =
−$53.2325 in bonds (i.e. borrow $53.2325 risk free). Our total position is
zero.
Suppose the next day S = 101. Now T ≈ 249/250, and the option price is
$11.0708. Our borrowing is now $53.2325∗e0.05/250 = 53.2431 due to interest,
so our total portfolio is worth

0.63683 × 101 − 11.0708 − 53.2431 = 0.006

By using this ‘∆-neutral’ porfolio, we have eliminated the loss due to the
increase in the option price.
Suppose instead the next day S = 110. Now T ≈ 249/250, and the option
price is $17.6365. Our borrowing is still now $53.2431 due to interest, so our
total portfolio is worth

0.63683 × 110 − 17.6365 − 53.2431 = −0.8283

We have partly minimized the loss due to the increase in the option price,
but not as effectively as for a smaller change.

We have so far assumed we can trade continuously, but in practice, we can

only trade finitely often. As a result, we encounter the discretization error -
the price of the hedging (replicating) portfolio no longer coincides with the
option price at all times.
In particular, we cannot keep both positions - in the stock and in bonds
- as prescribed by the Black-Scholes model. Therefore, at each moment of
rebalancing, we have to choose whether
- we keep ∆ (the amount of shares of stock) as prescribed by the model,
and invest the rest of the available capital in bonds (or borrow the required
amount by shorting),
- or keep the amount of money in bonds as prescribed by the model, and
invest the rest in the stock.

Typically traders choose to keep the value of ∆ as prescribed by the model,

because changes in the stock price are more significant than changes in the
value of bonds. This strategy is called ∆-hedging.

4
0.4 Γ (Gamma)
Gamma is the sensitivity of ∆ with respect to changes in S :

∂ ∂2
Γ= ∆= V
∂S ∂S 2

Γ measures how fast the hedging weight ∆ changes with the changes in the
underlying.
This is important since, as mentioned before, in practice we only trade at
discrete times. The smaller the curvature of the price of an option, as a func-
tion of S, the smaller the error of the discretization hedge - the difference
between the price function and its tangent line around the point of tangency.
Thus, Gamma measures the discretization error.

Let’s make this statement more precise. Assume we have short-sold an option
and set up the hedging portfolio at time t :

∂
γt = Vt − ∆t St , ∆t = V (St , t)
∂S

Then the hedging error at time t + δt is given by

γt Bt+δt + ∆t St+δt − V (St+δt , t + δt)

= (V (St , t) − ∆t St ) (1 + rδt) + ∆t St+δt − V (St+δt , t + δt)

Recall that St+δt ≈ St + rδtSt + σSt (Wt+δt − Wt ) and that (Wt+δt − Wt )2 ≈

δt.
Expanding using Taylor series, our error becomes
 
∂ ∂
Vt −St Vt + rδt Vt − St Vt
∂S ∂S
∂ ∂
− Vt − δt Vt − (St+δt − St ) Vt
∂t ∂S
1 ∂2 ∂
− σ 2 St2 (Wt+δt − Wt )2 2
Vt + St+δt Vt + o(δt)
2 ∂S ∂S
1 ∂2  
= − σ 2 St2 2 Vt (Wt+δt − Wt )2 − δt + o(δt)
2 ∂S
1 2 2  
= − σ St Γt (Wt+δt − Wt )2 − δt + o(δt)
2

5
where o(δt) is a function satisfying

o(δt)
→ 0, as δt → 0
δt

Notice that (δWt )2 − δt is a random variable with zero mean and variance
2(δt)2 .

We conclude Γt scales the main term in the discretization error of the hedge.
If Γt < 0 (short Gamma), the hedged portfolio benefits from large market
moves, and loses on small ones. If Γt > 0 (long Gamma) - vice versa. If we
hedge a long position in the option, the opposite conclusions hold.
The Γ of a call is 1 2
call e− 2 d+
Γ = p
Sσ 2π(T − t)
∂2
As t → T, Γcall → ∂S 2
(S − K)+ = δ(S − K). Due to put-call parity, the put
Γ is the same.

Gamma-hedging
We can reduce the discretization error of the hedge over the first time step
by Gamma-hedging. We cancel the current (instantaneous) Gamma of our
option V by opening a position in another option V 1 . Typically, we hedge
an exotic option with underlying and a vanilla call or put. The current value
of the resulting portfolio is given by

−Vt + ∆1 Vt1 + ∆St + γBt = 0

due to self-financing. We would like it to stay close to zero at time t + δt.

The above portfolio is Gamma-neutral if

∂2 1 ∂
2
V t − ∆ V1 =0
∂S 2 ∂S 2 t

So
∂2
1 V
∂S 2 t Γt
∆ = =
∂2
V1 Γ1t
∂S 2 t

6
Thus, we obtain a new option which is a linear combination of V and V 1 .
We then Delta-hedge this new option:

∂ ∂ 1
∆= V t − ∆1 V
∂S ∂S t

As before, this is only an instantaneous hedge, as ∆ and Γ will change through

time.
Example Suppose S = 100, σ = 0.3, T = 1, r = q = 0 We wish to hedge
a short position in a put option with K = 100, using the stock and a call
option with K = 80 We have the initial prices, ∆s and Γs

Price ∆ Γ
Stock 100 1 0
Put 11.9235 −0.4404 0.01315
Call 23.5344 0.8143 0.0089

To get a Γ -neutral portfolio, as we are short one put option, we need to

purchase
0.01315
∆1 = = 1.4743
0.0089
call options. To get a ∆-neutral portfolio, we then need to purchase

∆ = −0.4404 − 1.4743 × 0.8143 = −1.6409

stocks (i.e. a short position). These trades give us 11.9235−1.4743∗23.5344+

1.6409 ∗ 100 = 141.3170 in cash, which we invest.

0.5 ν (Vega)
Volatility σ is the only parameter in the Black-Scholes model that is not
directly observed in the market. It is, therefore, important to be able to
evaluate the dependence of option price on volatility.
Vega (commonly written ν), is the sensitivity of the option price to changes
in the volatility σ.
∂
ν= V
∂σ
In the BS model, σ is constant, so hedging with respect to changes in σ
doesn’t make sense. However, one can ask: what if my estimate of σ is
wrong? To estimate how far off, in this case, the computed option price is

7
from the “true” price, we need to find Vega.
For a call option, we have
r 2
call T − t − d+
ν = Se 2
2π

And it tends to zero as t → T , since the payoff is independent of σ It is the

same for a put, due to put-call parity.

Vega-hedging
Vega-hedging can be defined in the same way as Gamma-hedging, however,
its purpose is different:
- rather than reducing the discretization error, it is meant to reduce the
model
error - a misspecification of σ.
Given an additional derivative with price V 1 , the Vega-hedge of a short po-
sition in the original option prescribes to hold ∆1 units of V 1 and ∆ units of
S.
In order to make the portfolio instantaneously Vega-neutrat, we need
∂
∂ ∂ ∂σ Vt νt
− Vt + ∆1 Vt1 = 0 ⇒ ∆1 = =
∂σ ∂σ ∂
∂σ Vt
1 νt1

As before, ∆ is determined as the corresponding S-derivative of the portfolio

of options V − ∆1 V 1 , assuming ∆1 is fixed:

∂ ∂ 1
∆= V t − ∆1 V
∂S ∂S t

Of course, in order to keep the portfolio Vega-neutral at the next moment in

time t+δt, the value of ∆1 ( as well as ∆) will need to be changed at that time.

Vega hedging makes sense if one believes that the ”true” volatility is con-
stant, but we may be mistaken about its true value. However, sometimes,
you may see a description of Vega-hedging as “hedging the non-constant
volatility”. If the volatility is believed to be changing dynamically, thenmore
complicated stochastic volatility models have to be used.
Using a constant volatility model, to design a hedge against stochastic volatil-
ity is, clearly, self-contradictory. It may sometimes be used in practice, if

8
other options are too hard to implement, however, then, one has to be very
careful and make sure that the side effects of such hedging do not overweigh
the positive impact.

There is a universal relation between Vega and Gamma which holds

for all European options, because their prices are functions of time and the
value of the underlying and these functions satisfy the BSPDE.
Consider the BSPDE

∂ 1 ∂2 ∂
LBS V = V + σ 2 S 2 2 V + rS V − rV = 0
∂t 2 ∂S ∂S

and differentiate it with respect to σ.

∂ ∂2 ∂
V = ν, V = ν, ···
∂σ ∂σ∂t ∂t

As a result, we obtain

∂ 1 ∂2 ∂ ∂2
LBS ν = ν + σ 2 S 2 2 ν + rS ν − rν = −σS 2 2 V
∂t 2 ∂S ∂S ∂S
= −σS 2 Γ
ν(S, T ) = 0

In PDE language, the above equation means that “−σS 2 Γ00 is a source for
∂2
ν. Notice also that σS 2 Γ = σS 2 ∂S 2 V satisfies the BSPDE (recall homework

exercise).
Therefore, it is easy to check that

ν(S, t) = (T − t)σS 2 Γ(S, t)

satisfies the desired PDE:

LBS (T − t)σS 2 Γ(S, t) = −σS 2 Γ + (T − t)LBS σS 2 Γ(S, t)

   

= −σS 2 Γ

This is a useful trick, and a good example of how the PDE techniques may
help in establishing certain non-trivial relations between various quantities
in mathematical finance.

9
From the fact that
ν(S, t) = (T − t)σS 2 Γ(S, t)

we see that for European options, Γ-hedging and Vega-hedging amount to the
same thing. The motivations however, are quite different! This relationship
does not necessarily hold for other types of options.

0.6 Other Greeks

There are also sensitivities to other parameters:
◦ Theta Θ = ∂V /∂t is sensitivity with respect to t, and measures the
maturity
sensitivity of our portfolio.
◦ Rho ρ = ∂V /∂r is sensitivity with respect to the interest rate r, and
measures
the potential impact of interest rate changes on the value of the portfolio
(more
important over the long-term).
◦ Epsilon  = ∂V /∂q is the sensitivity with respect to the dividend yield.

You can have higher order sensitivities:

◦ Vanna ∂∆/∂σ = ∂ 2 V /∂σ∂S is the sensitivity of ∆ with respect to
changes in
volatility σ. This is a measure of model dependence of the Delta-hedging
strategy itself.
◦ Charm ∂∆/∂t = ∂Θ/∂S = ∂ 2 V /∂t∂S is the sensitivity of ∆ to time
◦ Vomma ∂ν/∂σ = ∂ 2 V /∂σ∂σ is the second order sensitivity to the
volatility.
◦ Veta −∂ν/∂t
◦ Vera ∂ρ/∂σ = ∂ 2 V /∂σ∂r
◦ Speed ∂Γ/∂S = ∂ 3 V /∂S 3
◦ Zomma ∂Γ/∂σ
◦ Color −∂Γ/∂t

Given enough trading instruments (assets, options), we can cancel the higher
order sensitivities as well. However, decreasing the risk associated with a
wrong choice of parameters can increase model risk: our family of models

10
(the Black-Scholes models, parameterized by r and σ) is not the true model!
We also may have worse performance for larger moves (as we are using local
sensitivities). Therefore, one should find an optimal trade-off, and shouldn’t
go too far with “matching the Greeks”.

11