PARAMETER RISK IN THE BLACK AND SCHOLES

MODEL

MARC HENRARD

Abstract. We study parameter or estimation risk in the hedging

of options. We suppose that the world is such that the price of an
asset follows a stochastic differential equation. The only unknown
is the (future) volatility of the asset. Options are priced and hedged
according to the Black and Scholes formula. We describe the dis-
tribution of the profit and loss of the hedging activity when the
volatility of the underlying is misestimated. A financial interpre-
tation of the results is provided. Analytical bounds and numerical
results for call, put, and portfolios complete our description.

1. Introduction
We suppose that we are in a world with Black and Scholes belief.
By this we mean that there exists an asset that follows the stochastic
equation
dSt = g(t, ω)St dt + v(t, ω)St dWt
where g, v : R+ × Ω → R and Wt is a Brownian motion. We suppose
that v is positive and g and v are regular enough for the existence of
a solution of the equation. We do not explicitly give such conditions;
it can be some standard Lipschitz condition (to satisfy Ito theorem [7,
Chap. 5]) or a more specific one. The volatility can depend on the time,
the level of the asset but also of an independant random component.
But people act as if the asset was following a pure geometric Brow-
nian motion
dSt = µ(t)St dt + σ(t)St dWt
with µ, σ : R+ → R. We suppose also that continuous rebalancing
is possible. People think they know the volatility, sell options at the
price given by the Black and Scholes formula and hedge it using the
∆-neutral self-financing strategy and hope that the profit at maturity
(or at any time between the initial transaction and maturity) will be
0. It means that not only one sells the option at a wrong price but also
that the hedging used will be ∆-neutral with respect to an incorrect

Date: First version: August 18, 2001; This version: December 11, 2001.
Key words and phrases. Black and Scholes model, option, parameter risk, profit
distribution.
1
2 M. HENRARD

∆. So the final profit will be non zero, not only due to the initial error
but also due to accumulating errors during the life of the option.
In this paper we study the theoretical profit. We compute the profit
using continuous rebalancing and asset movement given by the stochas-
tic differential equation. Study of the profit using discrete rebalancing,
best estimate of volatility and real data can be found in Figlewski and
Green [3]. Our approach is very different to this in the sense that we
study the error coming from the misestimation of the parameter of the
model not from the inadequacy of the model with reality.
We give a description of the distribution of this profit. Then we
obtain lower and upper bounds in the case of call and put. Numerical
results describing the influence of different parameters are then pro-
vided. We develop the result for the case where the asset is paying
continuously a dividend.
This note doesn’t present the more complex model. But it includes
the Hull and White model for zero coupon bonds. Its goal is to present
one type of risk that appears when trading options. In particular the
parameters that influence the risk are underlined.
All the parameters of the model influence the distribution of the
profit. This list includes a parameter that is not present in the pricing
formula: the growth rate µ.
We insist also on the fact that the profit does not only comes from the
misprizing of the option trades but also from all the hedging process.
Even if all the option trades are done internally within a bank and each
component of the bank believes that its risk is non-existent, the total
profit of the bank can be non zero (see 6.4).
We also present some results on portfolios of options. A small change
of the pay-off profile can lead to large differences of the profit distri-
bution. This comes from the difficulty to hedge some positions with
instable hedging strategies (large gamma).

2. Notation an preliminary remarks

As we have two parameters for the volatility (v and σ), we will use
a double notation for all the discussion. Any variable with a bar (H̄,
V̄ , π̄, . . . ) will refer to the value using the estimated volatility σ.
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 3

The following variables are used:

T Maturity of the option.
X Strike of the option.
r0 The (short term) interest rate (continuously com-
pounded).
r1 Rate of the dividend (continously compounded).
Vt Value of the replication portfolio in t.
Ht The hedging ratio (delta) in t.
Ht0 The quantity of cash in the hedging strategy.
St The value of the underlying. Rt
StD
Discounted value of St : e− R 0 r0 (s)ds St .
t
VtD Discounted value of Vt : e− 0 r0 (s)ds Vt .
π(t, S D ) Discounted price of the option in t when the discounted
price of the underlying is S D using the Black and Sc-
holes formula. The usual formula for the price of an
option is in term of its price (not discounted price). If
weRdenote by P (t, S) the usual price, we have π(t, S D ) =
t Rt
e− 0 r0 (s)ds P (t, e 0 r0 (s)ds S D ).
For the explanations, we consider that the dividend is continuously
paid in the asset. This means that if we have one asset at some stage,
after a period of t we have er1 t assets. If the dividend is paid in local
currency instead, from a theoretical point of view nothing is changed.
As we can continuously rebalance, we can buy the quantity of assets
we need. This last type of payment is usually the case for gold options
in USD, where the interest on the gold deposits are paid in USD.
The model we present includes the Garman-Kohlhagen model for
currency options if one considers the interest on the foreign currency
as a dividend. The meaning of local and foreign currencies is not related
to the citizenship of the trader or its base currency. The local currency
is the one in which the premium is payed, the foreign currency is the
other one. For example for a USD based trader which sell a EUR/CHF
call, receiving the premium in CHF, the local currency is the CHF. So if
he delta hedges this position, he will buy ∆ EUR and sell the equivalent
in CHF. Its position in CHF will be −∆St + P .

3. The profit
Consider a derivative with pay-off P (T, ST ) at time T . If the value
of this derivative exists between 0 and T and depend only on t and St ,
we denote it P (t, St ). If the price doesn’t exists between 0 an T , the
formulae containing P (t, St ) have to be considered only for t = T .
4 M. HENRARD

The profit1 at any moment between the initial transaction and the
maturity of the option is
PnLt = V̄t − P (t, St )
where V̄t is the value of the hedging portfolio using the estimated pa-
rameter σ. To simplify the writing, we will use the discounted values
in the computations. The value of the hedging portfolio is
V̄tD = H̄t0 + H̄t StD
where H̄t = D2 π̄(t, StD ). Note that for call and put, D2 π̄ is bounded
and C ∞ in (t, S), so regular enough for the integration. We suppose
that the strategy we use is self-financing:
Z t Z t
D D D
V̄t = V̄0 + r1 (θ)H̄θ Sθ dθ + H̄θ dSθD
0 0

In the case where v is different from σ the profit will come in H̄t0 that
will no longer be the one predicted.
The estimated price π̄(t, StD ) satisfies the partial differential equation
1
D1 π̄(t, x) − r1 (t)xD2 π̄(t, x) + x2 σ 2 (t)D22 π̄(t, x) = 0.
2
Applying the Itô formula to Πt = π̄(t, StD ) and this last result, we have
1
dΠt = D1 π̄(t, StD )dt + D2 π̄(t, StD )dStD + D22 π̄(t, StD )v 2 (t, ω)StD dt
2
D D D
= D2 π̄(t, St )(r1 (t)St dt + dSt )
1 2
+ StD (v 2 (t, ω) − σ 2 (t))D22 π̄(t, StD )dt
2
and then
Z t Z t
D D D
(1) H̄θ dSθ = π̄(t, St ) − π̄(0, S0 ) − D2 π̄(θ, SθD )r1 (t)SθD dθ
0 0
1 t
Z
2
− D22 π̄(θ, SθD )SθD (v 2 (θ, ω) − σ 2 (θ))dθ.
2 0
Combining (1) with the self-financing strategy and the fact that the
portfolio is ∆-hedged (H̄θ = D2 π̄), we have
PnLD
t = V̄0D − π̄(0, S0D )
+π̄(t, StD ) − π(t, StD )
1 t D2 2
Z
+ S (σ (θ) − v 2 (θ, ω))D22 π̄(θ, SθD )dθ.
2 0 θ
1
We call profit the value of the strategy. This value can be positive or negative.
So a sentence like “profit accumulate slowly” should be interpreted as “the value
change slowly”. It can also be a loss accumulating. So profit can be a very bad
thing!
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 5

4. Interpretation
The profit is
Rt
r0 (s)ds


PnLt = e 0 V̄0 − P̄ (0, S0 )
+P̄ (t, St ) − P (t, St )
1 t 2 2
Z
+ Sθ (σ (θ) − v 2 (θ, ω))D22 P̄ (θ, Sθ )dθ
2 0
and is made of three parts.
The first one is the (actualized) difference between the premium paid
for the option and its value using the estimated volatility σ. It can be
interpreted as the initial margin made on the trade or the fees. This
is a fixed amount decided by the trader. There is no (model) risk
associated to this amount. So we usually suppose that it is zero.
The second one is the difference, at the moment of the computation
of the profit, between the value computed using the estimated volatility
of the option and the true one. It is the profit that appears when the
real volatility is discovered. At the maturity of the option, as the value
of the option is its intrinsic value whatever the volatility is, this part
is zero. If the option is never hedged by another option and never
marked-to-market with its true value, this part is never known.
The third one is due to the incorrect ∆ used for the ∆-hedging. It is
the integral of the gamma of the option (second derivative of the price
with respect to the underlying) times the square of the price multiplied
by the difference between the square of the expected volatility and the
true one. This is the most insidious part of the risk. If the position is
held to maturity this profit accumulate slowly and this source of profit
remains unknown.
This last part can be interpreted as the cumulative cost of the non-
self-financing part of the theoretical strategy. It is the amount that has
to be injected in the portfolio to maintain the theoretical quantity of
asset and cash. In [4, Section 7.1.1], it is obtained for constant σ using
this idea, even it is not obvious that this number does not only include
the necessary injections but also the financing cost of such injections.

4.1. Convexity. When the price is convex in St , Γ is always positive

and thus an overestimation of the volatility gives always a positive
profit even if the wrong volatility is used for the hedging. In the case
were the payoff is a convex function of the price, the price of the option
is itself convex ([7, Section 4.5, p. 83]). The call and the put are
particular examples of this.

4.2. Gamma reduction. Suppose that one is hedging a call. Look-

ing at the profit, it can be tempting to try to reduce Γ to reduce the
importance of the third part. This can be done making σ close to
0. Then Γ converges to 0 almost everywhere. From there one can be
6 M. HENRARD

tempted to deduce that when σ tend to 0 the risk disappears. From a

financial point of view this would mean that you can hedge the option
without risk by choosing a volatility of (very close to) 0. This corre-
spond to a stop-loss strategy (see [6, Section 13.3, p. 309]). From a
technical point of view this is not correct. It can not be deduced from
the fact that Γ tend to 0 almost everywhere that its integral tends
to 0. Lebesgue dominated convergence theorem can not be used here
because the functions are not dominated when S = Xe−r(T −t) (at the
money option). This strategy is cost-free except when the option is at
the money (which will be the case with a non-zero probability). In this
case there is an infinite number of rebalancing of all the portfolio (∆
jump from 0 to 1 and inversely, Γ is infinite) and the cost is high.
Note nevertheless that the gamma used is the one with respect to
the estimated Black and Scholes model, so it is known. If the option is
chosen to reduce the gamma, the risk can be reduced (see Section 6.6).

4.3. Structural robustness. At maturity, the value of the option is

model independent. The only part left is the third one. In particular
we don’t even need the existence of a true price of the option between
the initial transaction and the maturity to have this result. The only
thing we need is the existence of the price of the asset. The volatility
can be very wild (stochastic, time dependant, depending of S) and the
result still hold. So even if the true equation of the underlying is not
as nice as described by the Black and Scholes model, some bound can
be obtain for the profit. It is enough to know an upper bound of the
volatility to obtain surely a positive profit. This structural robustness is
very important from a practical point of view. Without such a property
a trader would have to be convinced that the Black and Scholes model
he uses to trade and hedge options is perfect and he knows the exact
volatility. With the property it is enough that he believes that he can
estimate well enough the volatility to obtain an upper bound or at least
not to be too below the reality. It is strange (as noted in [2]) that such
a robustness property which is very important for the practicality of
the technic is rarely mentioned in standard books on the subject.

4.4. Growth rate. Even if it doesn’t appear explicitely in the formula,

the growth rate µ plays a role through St . It is not use for the pricing
and the rebalancing but influences the final profit and thus the risk.
This can be viewed more explicitely by writing the stochastic derivative
of PnLD (see also [5]):

¯ t S D dt + ∆
d PnLD = r1 (t)∆ ¯ t dS D − (r1 (t)∆t S D dt + ∆t dS D )
t t t

t
¯

D
= ∆t − ∆t St (µ − r0 (t) + r1 (t)) dt + σ dWt .
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 7

5. Bound on the profit

In this section we describe a bound on the profit for calls (or puts)
under certain restrictions on the volatility and the rates. The estimated
equation for the underlying is a pure geometric Brownian motion (σ
constant). The true volatility is bounded by two constants, i.e. there
exists σ− > 0 and σ+ such that

σ− ≤ v(t, ω) ≤ σ+ .

The interest rates (r0 and r1 ) are constant2 We denote by X the strike
price of the option.
In this case

N 0 (d1 )e−r1 (T −t)

Γ= √
Sσ T − t

where

ln(S/X) + (r0 − r1 + σ 2 /2)(T − t)

d1 = √ .
σ T −t

The first bound we can obtain has already been described in the
previous section. As the payoff of the call is convex, when σ ≥ σ+ , the
profit is positive and when σ ≤ σ− , the profit is negative.
To present the second bound we need a longer argument. To obtain
the bound, we maximize It (S) = Γ(t, S)S 2 for a fixed t. For any t, it
can be proved that It ≥ 0 (due to convexity), limS→0 It (S) = 0 and
limS→+∞ It (S) = 0 (details are given in Appendix A). Then there exists
a maximum of It in [0, +∞). We denote it Smax . As It is differentiable
in the interval, this maximum is such that It0 (Smax ) = 0. A strait-
forward computation gives

2 /2)(T −t)
Smax = Xe−(r0 −r1 −σ .
√
In Smax , we have d1 = σ T − t and thus
√
N 0 (σ T − t) X e−(r0 −r1 )(T −t)
It (Smax (t)) = √ Smax = √ √ .
σ T −t 2πσ T −t

2
The same construction is possible if σ, σ− , σ+ , r0 and r1 are time dependent,
but then we don’t have the explicit formula anymore.
8 M. HENRARD

From this we obtain the bound on the profit. When r0 6= r1 , we have

PnLt −(P̄ − P )
Z t
1 2 2
≤ (σ − σ− ) Iθ (Smax (θ))dθ
2 0
1 X t e−(r0 −r1 )(T −θ)
Z
1 2 2
= (σ − σ− ) √ √ dθ
2 2π σ 0 T −θ
√ Z √2(r0 −r1 )√T −t
1 2 1 X 2 1 2
= (σ − σ−2
)√ √ √ √
e− 2 ξ dξ
2 2π σ r0 − r1 2(r0 −r1 ) T

1 2 2 X  p √ p √ 
= √ (σ − σ− ) √ N ( 2(r0 − r1 ) T ) − N ( 2(r0 − r1 ) T − t)
2 σ r0 − r1
So if we denote by f (X, T, r0 , r1 ) the two factors on the right, we
have
1 1
√ (σ 2 − σ+2
)f (X, T, r0 , r1 ) ≤ PnLt −(P̄ − P ) ≤ √ (σ 2 − σ− 2
)f (X, T, r0 , r1 ).
2 2
In particular if the volatility is overestimated (σ ≥ σ+ ), the profit is
always positive.
When r0 = r1 , we have
Z t
1 2 2
PnLt −(P̄ − P ) ≤ (σ − σ− ) Iθ (Smax (θ))dθ
2 0
1 X t
Z
1 2 1
= 2
(σ − σ− ) √ (T − θ)− 2 dθ
2 2π σ 0
√ √ 
= 2(σ 2 − σ− 2
) T − T −t

This bound can not be improved as St = Smax (t) is a possible path.

It is a (too) smooth growth at rate σ 2 /2 − r0 + r1 to the strike price.

6. Numerical results
As the profit depends in a non-trivial way of all the path of the
price (and not only its final value), an analytical description of its
distribution seems out of reach.
We give some numerical results3 in the case where the true equation
is a pure geometric Brownian motion (v constant). The methodology
we use to obtain the results is the following. For each result, we have
simulated a large number of path of the underlying (typically 10,000).
Each path is made of 100 or more time-steps. We used those simulated
paths to numerically compute the integral.
In the different figures, we have represented the minimum, the 25%,
50%, the 75% percentile, the average, and the maximum of the differ-
ent profit realizations. The minimum and maximum realizations are
3
Matlab code available
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 9

represented by the dashed line. The 25, 50 and 75% percentile are in
dotted line. The average is in continuous thin line.
We have also computed the profit that would be realized if the trans-
action is done at the estimated volatility and the hedging is done from
the start with the correct volatility. This correspond the initial differ-
ence of value with the estimated volatility and the true one actualized
(at the constant rate r0 ). This profit is called the initial profit and in
the figures it is represented by the thick line. The theoretical maximum
and minimum are also represented, in continuous thin lines.
We describe the influence of several parameters on the profit.
Volatility error: For a fixed v, how the profit changes with σ.
True volatility: For a given volatility error, how the profit changes
with v.
Time: How the profit evolves between the initial transaction and
the maturity of the option.
Growth rate: The growth rate µ does not influence the price and
the hedging of the option, but influences the final profit. This
show that risk management is more difficult than pricing.
Portfolio: The maximum profit on a portfolio of options is not the
sums of the maximum profits on the components. We give a
example of reduction and increase of profit.
The maturity of the option can be viewed as a scaling factor. All the
options we study have a maturity of 1. An option with a maturity of
0.5 and a volatility of σ is just an option with maturity 1 and volatility
σ/4 in a world with a different unit of time.
6.1. Distribution of the error. Figure 1 represents the distribution
of the profit due to pricing and hedging of an option with a wrong
volatility. The initial asset price is 100. The strike price are respectively
90 and 115. The true volatility is 20% and the estimated one is 22%.
The growth rate is 10%, the risk free rate and the dividend rate are
5% and the time to maturity is 1 year. The sample consist of 20,000
simulations of 200 time-steps.
The distribution is not symmetric around the average. The shape of
the distribution depends of the parameters and is discussed in the rest
of this section.
6.2. Volatility error. Figure 2 represents the profit due to pricing
and hedging of an option with a wrong volatility for different errors.
The strike price is 90 and the time to maturity is 1 year. The true
volatility is 20%, the growth rate is 5% and the dividend rate 3%. The
initial price is 100. The sample consists of 10,000 simulations of 200
steps.
As expected the profit depends strongly on the volatility error. The
profit has the same sign as σ−v. The average profit is almost symmetric
around the true volatility. This is also the case for the initial profit.
10 M. HENRARD

1100 700

1000
600
900

800 500

700

400
600

500
300

400

200
300

200
100
100

0 0
0 1 2 3 4 5 6 7 8 9 10 0 5 10 15 20 25 30 35 40 45 50

(a) Strike price of 90 (b) Strike price of 115

Figure 1. Numerical simulation of the distribution of

the profit (as percentage of the true price).

Nevertheless the extreme profits are larger for underestimations of the

volatility than for overestimation. This non-symmetry can be observe
in the bound on the profit in the factor 1/σ.
6.3. True volatility. If we suppose that the error is fixed (absolute
or relative), the profit changes with the true volatility.
Figure 3 shows the profit for an error of 1% of the annual volatility.
The other parameters are: µ = 0.10, r0 = 0.05, r1 = 0.03, X = 100,
T = 1, and S0 = 80. The simulation uses 10,000 iterations of 200 steps.
When the true volatility is 10%, the average profit is 0.1362; when the
volatility is 30%, the average profit is 0.3163. It means the for an equal
absolute error, the profit changes from the simple to more than the
double. When the volatility is high, delta-hedging is riskier; but also
the same volatility margin is generating a larger profit.
Figure 4 shows the profit for an error of one tenth of the volatility.
The other parameters are: µ = 0.10, r0 = 0.05, r1 = 0.03, X = 90,
T = 1, and S0 = 1000. Then the result is even worse; the average
profit range from 0.0880 when v is 10% to 0.9628 when v is 30%, which
is more than 10 times higher.
6.4. Growth rate. Figure 5 represent the profit for different growth
rates. The option we used is a call with a strike of 100 and a maturity
of one year. The true volatility of the geometric Brownian motion is
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 11

−2

−4

−6

−8

−10

0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

Figure 2. Numerical simulation of the profit (as per-

centage of the true price) for different values of the esti-
mated volatility σ.

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

Figure 3. Numerical simulation of the profit for differ-

ent values of the true volatility and an error of 1% of
annual volatility.

20%. The true initial price of this option is 8.4333. All the values we
use in this subsection are percentages of this value.
12 M. HENRARD

1.5

0.5

0
0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

Figure 4. Numerical simulation of the profit for differ-

ent values of the true volatility and an error of one tenth
of the true volatility.

20

18

16

14

12

10

0
−0.1 −0.05 0 0.05 0.1 0.15

Figure 5. Numerical simulation of the profit (as per-

centage of the true price) for different values of the
growth rate µ.

The estimated volatility used for pricing and hedging the option is
22%. If the option had been sold with a price corresponding to this
volatility of 22% but hedge (perfectly) with the true volatility, the profit
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 13

would have been 9.4485% of the true initial price. As the estimated
volatility is higher than the true one, the minimum profit is zero (see
Section 5). The maximum profit is given by the horizontal thin line
and is 18.0021%.
We have estimated the values of the profit of different realizations
of the stochastic equation and the hedging. This has been done for 21
different growth rates (µ) between -15% and 15%.
The average and the percentile are influenced by the growth rate.
The growth of the smooth path for which the profit is maximum (see
Section 5) is given by the vertical dotted line. When the actual growth
rate is close to this rate, on average the paths will be closer to this
profit-maximizer path. That is why the average profit is higher when
the growth rate is close to profit-maximizer rate. In the numerical
results we obtain, the difference between the highest average (9.4276%)
and the minimum (8.3179%) reaches 1.1097%.
This parameter, which is usually not estimated when dealing with
options using the Black and Scholes model, can change the average
profit by about 1% of the value of the product sold. This is about
11.5% of the initial margin that could have been obtained by hedging
with the true volatility.
The difference between the average profit and the initial one depends
also of the strike price. Figure 6 shows the absolute profit when µ =
−0.1 and µ = 0.1 for strikes between 80 and 120 and S0 = 100. The
figures were done using 17 different strikes; for each of them 10,000
simulations with 200 steps were used. As can be seen from the graphs,
the average is above or below the initial profit depending of the strike
and this relation depends itself from the growth rate.
Suppose that in a bank two traders (A and B) have a superior pre-
dictive ability with respect to the market to estimate the volatilities.
Suppose also there is a set of independent underlying (or the same un-
derlying on several periods). The volatility for each of them is 20 or
21% and for each of them the market price is 20 or 21%. By saying that
the hedgers as a superior predictive ability with respect to the market,
we mean that when the market estimation is correct, the hedgers are
correct also and when the market is wrong, the hedgers can be wrong
like the market or correct. The market price can be the price available
outside the bank or any price that is used for internal transactions.
The traders are trading options among themselves at the market
price and delta hedge them with the outside market. All the option
transactions are done internally. So if the price is not perfect, it is
only an internal transfer of profit. The market price is only used to
compute the profit of the individual traders. For the total of the bank,
the market price is irrelevant. The only thing that change the total
14 M. HENRARD

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
80 85 90 95 100 105 110 115 120

(a) µ = −0.1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
80 85 90 95 100 105 110 115 120

(b) µ = +0.1

Figure 6. Numerical simulation of the profit for differ-

ent values of the strike price.
profit is the fact that one trader is correctly estimating the volatility
and the other not4 .
4
If both traders are wrong, but one less than the other, the result of this section
stays similar.
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 15

If both traders have the same estimates, they will always have oppo-
site position, not only in term of options but also in term of hedging.
So the total profit for the bank will always be zero. If they don’t have
the same estimates, one is correct and one is wrong. As the traders
are superior to the market, the market is wrong. The possibilities are
summarized in the following table.

Trader A / Trader B
Market – True 20/21 21/20
20 – 21 short/long short/long
-average/+initial +initial/-average
21 – 20 long/short long/short
+initial/-average -average/+initial
Let’s comment on the first line; the second is just the symmetric
situation. If the market price is 20 then B, the trader estimating the
volatility at 21, perceives the price as underestimated and so will be
long. As he is correct in his estimate and uses it for hedging, he will
make the initial profit. The other trader will hedge the opposite po-
sition with the wrong hedging and will make the average profit (if we
suppose that there are enough independent positions to use the aver-
age).
We apply this to the case describe in Figure 6 for calls largely out-
of-the-money with a trike price of 115. If µ = 0.1, the profit will
be −0.044: 0.3373 of initial profit minus 0.3813 of average profit. If
µ = −0.1, then the profit will be +0.0748: 0.3373 of initial profit minus
0.2625 of average profit. Note also that in the case where µ = −0.1,
the initial profit is at the level of the 75% percentile, i.e. 75% of the
time the profit will be below the profit that would have been obtained
by hedging the option correctly.
This means that in the situation described here, even if the traders
have a superior predictive ability, a risk manager should not allow the
traders to trade internally. That type of position will generate losses for
the bank when the growth rate is positive. This further emphasizes the
fact that risk management methods are not only pricing methods. The
internal trades are not only transfers of profit, but also transfer of risk
from traders with superior predictive ability to less capable traders.

6.5. Time. To study the effect of time on profit, we change slightly

our approach. We suppose that a call is sold at the true volatility (21%)
but hedge with a lower volatility (20%). This represents the situation
where someone sell an option above its estimated volatility, expecting
to make a profit, but the price was the correct one and the hedging is
wrong. For that scenario we compute two profits: the true one and the
one estimated by the trader. For the last, the option position is valued
using the estimated volatility (not the true one).
16 M. HENRARD

The parameters used are the following: growth rate of 10%; risk free
rate of 5%, dividend rate of 3%; initial value of 100; strike price of 100;
time to maturity of 1. The simulation consists of 500 path of 100 steps.
As we have one computation of the profit for each time step and not
only the final one, the computer time is larger for the results of this
section.
The convention for the results given in Figure 7 is the same that
in the previous sections. Here we have two sets of results. The one
starting at 0 is the true profit, the one starting at about 0.4 is the
estimated one.

0.4

0.3

0.2

0.1

−0.1

−0.2

−0.3

−0.4

−0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 7. Numerical simulation of the profit between

the initial transaction and the expiry of the option.

As the final value of the option is not parameter dependent, both

converge to the same values. The thick almost horizontal line represents
the expected profit at the start of the deal. This is the value of the
position in 0 using the estimated volatility for the valuation accruing at
the risk free rate. This value represents the maximum possible profit
(both for the estimated and the true profit). This means that the
profit expected by the dealer will never be achieved. The minimum
value for the estimated profit is given by the initial profit plus the
bound obtained in Section 5. The minimum value for the true profit
required an extra-computation. The term P̄ (t, St ) − P (t, St ) has also
to be estimated. The minimum for that term is
P̄ (t, St ) − P (t, St ) = Xe−r0 θ (N (d2 ) − N (−d2 ))
√
with d2 = 12 (σ − v) θ.
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 17

6.6. Portfolio. We take two portfolios of calls that we compare with

a call of strike 100. The pay-off functions of the portfolios
√ are equal
for values of the underlying between 90 and 100 + 5 2. Between those
values the pay-off of the portfolio are modified: one to reduce the
parameter risk and the second one to increase it.
The first
√ portfolio is composed of two options: long a call with strike
√
100 − 5 2 on 1/2 asset and long another one with strike 100 + 5 2
on 1/2 asset. The second portfolio is composed of three options: long
a call with strike 90 on one asset, short a call with strike 95 on 2
assets and long a call with strike 100 on 2 assets. Those two portfolios
are chosen such that the average of the difference between the pay-off
functions is 0. The pay-off functions are represented in Figure 8.

12

10

0
85 90 95 100 105 110

Figure 8. Pay-off for the portfolios. The curves rep-

resent the value of the portfolio 1 week, 1 month and 6
months before maturity. The dotted lines represent the
call, the dashed-dotted lines represent the first portfolio
and the dashed lines represent the second one. The cir-
cles represent the points where the Gamma is maximum
or minimum.

For the rest of this section, we use a initial value of the asset of
100. The true volatility is 10%, the estimated one is 11%. The time
to maturity is 0.5. Note that the initial vega (derivative of the option
price with respect to the volatility) is of the same order for the different
portfolios: 27.4026 for the call, 18.2937 for the reducing-risk portfolio,
and 24.0470 for the increasing-risk portfolio. Note in particular that the
increasing-risk portfolio has a lower vega than the call. This number is
often used as a way to measure the risk with respect to change or error
18 M. HENRARD

in volatility estimates. But it measures the instantaneous risk, not the

risk during the life of the position. A deal can have little risk today
but creates significantly more in the future (see comments of Donna
Howe in the framework of Raroc [1]).
Figure 9 shows the distribution of profit for the 3 portfolios. The
distribution is constructed with 10,000 samples of 100 time-steps. The
vertical dotted lines represent the average profit; the continuous lines
represents the theoretical extremum for the profit. The average profit
in the sample are 0.2594 for the call, 0.1793 for the first portfolio and
0.2580 for the second. The maxima are 0.5285 for the call, 0.2941 for
the first portfolio and 0.7824 for the second one. The call and the
first portfolios are convex, so the minima are 0. The minimum for the
second portfolio is -0.4646.

10

0
−0.4 −0.2 0 0.2 0.4 0.6 0.8

Figure 9. Simulated distribution for the call and the

two portfolios. The dotted lines represent the call, the
dashed-dotted lines represent the first portfolio and the
dashed lines represent the second one. The vertical lines
are the theoretical extremum.

As it is clear from Figure 9, the distributions are quite different.

Even if the initial price, the initial vega and the pay-off are somehow
similar. The difference lies not only in the average but mainly in the
extreme profits. The reducing risk portfolio reduce the distribution
of risk by about a factor of 2. The increasing risk portfolio doesn’t
change much the average risk but change drastically the distribution of
it. Even if at the start we are globally long of a call and we overestimate
the volatility, the profit can be negative!
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 19

Appendix A. Details of the computations of the upper

bound on the profit
We prove that limS→0 It (S) = 0 and limS→+∞ It (S) = 0.
The first one is simple as when S → 0, d1 → −∞, so d21 → +∞ and
0
N (d1 ) → 0. This shows that It (S) = S 2 Γ(t, S) → 0.
For the second, we need some extra computation. We have for S >
X,
2
ln(S/X) + (r + σ 2 /2)(T − t)

1
d21= √ ≥ 2 (ln(S/X))2 .
σ T −t σ (T − t)
So by posing S/X = y, it is enough to prove that limy→+∞ y exp(−α(ln y)2 ) =
0 for α > 0. As this last limit can be written as exp(−α(ln y)2 + ln y)
and that the argument of the exponential tend to −∞, the result is
proved.

References
[1] O. Bennett. Reinventing Raroc. Risk, 14(9):112–113, September 2001. 18
[2] M. Davis. Mathematics of financial markets. In B. Engquist and W. Schmid, ed-
itors, Mathematics Unlimited–2001 and beyond, pages 361–380. Springer, 2000.
6
[3] S. Figlewski and C. T. Green. Market risk and model risk for financial institu-
tions writing options. Journal of Finance, 54(4):1465–1499, 1999. Reprinted in
Model Risk: concepts, calibration and pricing, Risk books, 2000. 2
[4] J.-P. Fouque, G. Papanicolaou, and K. R. Sircar. Derivatives in financial mar-
kets with stochastic volatility. Cambridge University Press, 2000. 5
[5] R. Gibson, L. S. Lhabitant, N. Pistre, and D. Talay. Interest rate model risk:
an overview. The Journal of Risk, 1(3):37–62, 1999. 6
[6] J. C. Hull. Options, futures, and other derivatives. Prentice Hall, fourth edition,
2000. 6
[7] D. Lamberton and B. Lapeyre. Introduction au calcul stochastique appliqué à la
finance. Ellipses, 1997. 1, 5

Contents
1. Introduction 1
2. Notation an preliminary remarks 2
3. The profit 3
4. Interpretation 5
4.1. Convexity 5
4.2. Gamma reduction 5
4.3. Structural robustness 6
4.4. Growth rate 6
5. Bound on the profit 7
20 M. HENRARD

6. Numerical results 8
6.1. Distribution of the error 9
6.2. Volatility error 9
6.3. True volatility 10
6.4. Growth rate 10
6.5. Time 15
6.6. Portfolio 17
Appendix A. Details of the computations of the upper bound
on the profit 19
References 19
List of Figures 20

List of Figures

1 Numerical simulation of the distribution of the profit

(as percentage of the true price). 10
2 Numerical simulation of the profit (as percentage of
the true price) for different values of the estimated
volatility σ. 11
3 Numerical simulation of the profit for different values
of the true volatility and an error of 1% of annual
volatility. 11
4 Numerical simulation of the profit for different values
of the true volatility and an error of one tenth of the
true volatility. 12
5 Numerical simulation of the profit (as percentage of
the true price) for different values of the growth rate
µ. 12
6 Numerical simulation of the profit for different values
of the strike price. 14
7 Numerical simulation of the profit between the initial
transaction and the expiry of the option. 16
8 Pay-off for the portfolios. The curves represent the
value of the portfolio 1 week, 1 month and 6 months
before maturity. The dotted lines represent the call,
the dashed-dotted lines represent the first portfolio
and the dashed lines represent the second one. The
circles represent the points where the Gamma is
maximum or minimum. 17
9 Simulated distribution for the call and the two
portfolios. The dotted lines represent the call, the
dashed-dotted lines represent the first portfolio and
 PARAMETER RISK IN THE BLACK AND SCHOLES MODEL 21

the dashed lines represent the second one. The

vertical lines are the theoretical extremum. 18

c 2001 by Marc Henrard, Leimenstrasse 30, CH 4051 Basel, Switzer-

land
E-mail address: Marc.Henrard@advalvas.be
URL: http://www.dplanet.ch/users/marc.henrard