⚫ Example on replicating an option (p.
3 of Shreve 2004):
For the one-step tree, let 𝑆0 = 4, 𝑢 = 2, 𝑑 = 1/2, 𝑟 = 1/4 and 𝐾 = 5 is the strike
price of a European call option.
So 𝑆0 𝑢 = 8 and 𝑆0 𝑑 = 2. If we begin with an initial wealth 𝑊0 = 1.20, and would
like to buy ∆0 = 1/2 shares of stock at 𝑡 = 0; this means that we need to borrow an
additional 0.80 to make this investment. As a result, our cash position is:
𝑊0 − ∆0 𝑆0 = −0.80
At 𝑡 = 1 (or after one year), this debt will become:
1
−0.80(1 + 𝑟) = −0.80 (1 + ) = −1.0
4
On the other hand, at 𝑡 = 1 our stock investment will be worth:
1
( )∙8=4
2
1
{ 2) ∙ 2 = 1
(
Taken together the two positions, we have either (4 − 1) = 3 or (1 − 1) = 0, and
these exactly match the payoff of the call option at 𝑡 = 1.
1
So what is the price of this call option at 𝑡 = 0?
It turns out the option price at 𝑡 = 0 must be 1.20, because this is the price at which
there is no arbitrage opportunity:
➢ If the option price is higher than 1.20, say, 1.22, then anyone will sell the option
for 1.22 at 𝑡 = 0, use 1.20 to “manufacture” this call option and save 0.02 in a
bank account. At 𝑡 = 1, if stock price goes up to 8, the option holder will exercise
and demand 3 as profit; if stock price goes down to 2, then the option holder will
not exercise (demand 0 profit). In either case the demand can be met by the
portfolio set up by 1.20. But the remaining 0.02 will earn risk-free interest rate
and become 0.025, and this is the arbitrage profit.
➢ If the option price is higher than 1.20, say, 1.18, then anyone will buy the option
at 𝑡 = 0 and “reverse-manufacture” the portfolio – short sell 1/2 shares of the
stock to receive 2, save 0.80 in a bank account and save the remaining (2 −
1.18 − 0.80) = 0.02 in another bank account. At 𝑡 = 1, if stock price goes up to
8, she needs 4 to buy back the stock, which can be exactly financed by the profit
1
from option 3 and the 0.80 (1 + 4) from first bank account. If stock price goes
down to 2 , she needs 1 to buy back the stock, which again is financed by
1
0.80 (1 + 4) and the option goes expired. In either case, the 0.02 in the second
bank account will earn risk-free interest rate and become 0.025.
With the above argument, and the possibility of replicating an option (which is to
determine the number of shares ∆0 and the cash position), we can obtain (by means of
Itô’s Lemma) the following Black-Scholes-Merton partial differential equation (PDE),
where the stock price 𝑋(𝑡) follows a GBM:
2
The BSM partial differential equation (PDE) is:
𝜕𝑐 𝜕𝑐 1 2 2 𝜕 2 𝑐
+ 𝑟𝑋 + 𝜎 𝑋 = 𝑟𝑐
𝜕𝑡 𝜕𝑋 2 𝜕𝑋 2
, where 𝑐 is the price of a European call option, 𝑟 is the risk-free rate and the
underlying stock price 𝑋(𝑡) follows GBM:
d𝑋(𝑡) = 𝜇𝑋(𝑡)d𝑡 + 𝜎𝑋(𝑡)d𝐵(𝑡).
We do not give a proof here, but it can be found in Hull (2012) and in Mikosch (1998).
Also, there is another way to derive the BSM PDE which is through the Feynman-Kac
Theorem (Chapter 6 of Shreve 2004).
Also, many texts do not provide a detailed solution to the PDE, since it is indeed
quite a long derivation 1 . However, here we can verify the BSM formula for 𝑐 is
indeed the solution to the BSM PDE:
The BSM formula for a European call option at 𝑡 = 0 is:
𝑐(0, 𝑋(0)) = 𝑋(0)𝑁(𝑑1 ) − 𝐾𝑒 −𝑟𝑇 𝑁(𝑑2 )
ln(𝑋(0)/𝐾)+(𝑟+𝜎2 /2)𝑇
, where 𝑑1 = , 𝑑2 = 𝑑1 − 𝜎√𝑇 and 𝑁(𝑥) = 𝐶𝐷𝐹𝑁(0,1) (𝑥).
𝜎√𝑇
To begin with, note that the BSM PDE is for a general time 𝑡, so
1“We do not state or solve the PDE here, because we prefer to emphasize the derivation of option
prices from risk-neutral price dynamics. (p.373 Taylor 2005)”;
“Solving the Black-Scholes partial differential equation is not always that easy. However, in some cases
it is possible to evaluate explicitly the above expected value in the risk-neutral pricing formula. (p.31
Schoutens 2003)”;
“… rather than showing how to solve the equation, we shall simply present the solution and check that
it works; … we present a derivation of this solution based on probability theory. (p.158, Shreve 2004)”.
3
𝑐(𝑡, 𝑋(𝑡)) = 𝑋(𝑡)𝑁(𝑑1 ) − 𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁(𝑑2 )
ln(𝑋(𝑡)/𝐾)+(𝑟+𝜎2 /2)(𝑇−𝑡)
with 𝑑1 = and 𝑑2 = 𝑑1 − 𝜎√𝑇 − 𝑡. The verification can be
𝜎√𝑇−𝑡
done in three steps.
Step 1
Show that 𝑋(𝑡)𝑁′(𝑑1 ) = 𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁′(𝑑2 ):
1 𝑑12
⟹ 𝑋(𝑡)𝑁′(𝑑1 ) = 𝑋(𝑡) exp (− )
√2𝜋 2
2
1 (𝑑2 + 𝜎√𝑇 − 𝑡)
= 𝑋(𝑡) exp (− )
√2𝜋 2
1 𝑑22 −2𝑑2 𝜎√𝑇 − 𝑡 − 𝜎 2 (𝑇 − 𝑡)
= 𝑋(𝑡) exp (− ) exp ( )
√2𝜋 2 2
−2𝑑2 𝜎√𝑇 − 𝑡 − 𝜎 2 (𝑇 − 𝑡)
= 𝑋(𝑡) exp ( ) 𝑁′(𝑑2 )
2
We can simplify the above expression by (Please verify yourself!):
ln(𝑋(𝑡)/𝐾) + (𝑟 + 𝜎 2 /2)(𝑇 − 𝑡)
𝑑1 =
𝜎√𝑇 − 𝑡
Thus,
−2𝑑2 𝜎√𝑇 − 𝑡 − 𝜎 2 (𝑇 − 𝑡)
𝑋(𝑡)𝑁′(𝑑1 ) = 𝑋(𝑡) exp ( ) 𝑁′(𝑑2 )
2
= 𝑋(𝑡) exp(ln 𝐾 − ln 𝑋(𝑡) − 𝑟(𝑇 − 𝑡)) 𝑁′(𝑑2 ) = 𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁′(𝑑2 )
We need this result in Step 2.
4
Step 2
∂𝑐 𝜕𝑑1 𝜕𝑑2
= 𝑿(𝒕)𝑵′(𝒅𝟏 ) − (𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁′(𝑑2 ) + 𝑟𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁(𝑑2 ))
∂𝑡 𝜕𝑡 𝜕𝑡
𝜕𝑑2 𝜎 1 𝜕𝑑2
= 𝑋(𝑡)𝑁′(𝑑1 ) ( − (𝑇 − 𝑡)−2 ) − (𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁′(𝑑2 ) + 𝑟𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁(𝑑2 ))
𝜕𝑡 2 𝜕𝑡
𝜕𝑑2 𝜎 1
= 𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁′(𝑑2 ) ( − (𝑇 − 𝑡)−2 )
𝜕𝑡 2
𝜕𝑑2
− (𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁′(𝑑2 ) + 𝑟𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁(𝑑2 ))
𝜕𝑡
𝜎
= −𝑟𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁(𝑑2 ) − 𝑋(𝑡)𝑁′(𝑑1 ) (𝑇 − 𝑡)−1/2
2
Step 3
∂𝑐
= 𝑁(𝑑1 )
∂𝑋
and
∂2 𝑐 1
2
= 𝑁′(𝑑1 )
∂𝑋 𝑋𝜎√𝑇 − 𝑡
Now we can verify the BSM formula is the solution to BSM PDE:
𝜕𝑐 𝜕𝑐 1 2 2 𝜕 2 𝑐
𝐿𝐻𝑆 = + 𝑟𝑋 + 𝜎 𝑋
𝜕𝑡 𝜕𝑋 2 𝜕𝑋 2
𝜎
= −𝑟𝐾𝑒 −𝑟(𝑇−𝑡) 𝑁(𝑑2 ) − 𝑋(𝑡)𝑁′(𝑑1 ) (𝑇 − 𝑡)−1/2 + 𝑟𝑋𝑁(𝑑1 )
2
1 1
+ 𝜎 2 𝑋 2 𝑁′(𝑑1 ) = 𝑟𝑐 = 𝑅𝐻𝑆
2 𝑋𝜎√𝑇 − 𝑡
□
5
Finally, we show how to obtain the BSM formula with risk-neutral pricing. But before
that I want to share a story, which is taken from the book:
春夜十話 - 數學與情緒 (1963), 岡 潔 (Kiyoshi OKA)
It is a beautiful story (in my humble opinion).
6
By risk-neutral pricing, we mean: “An option’s price at time zero equals the present
value of the expectation of its future price in a risk-neutral world. (p.376 Taylor 2005)”
We thus have:
𝑐 = 𝑒 −𝑟𝑇 𝐸 𝑄 [max(𝑋(𝑇) − 𝐾, 0)]
∞ ∞
−𝑟𝑇
=𝑒 ∫ max(𝑋 − 𝐾, 0) 𝑓𝑄 (𝑋)d𝑋 = 𝑒 −𝑟𝑇 ∫ (𝑋 − 𝐾) 𝑓𝑄 (𝑋)d𝑋
0 𝐾
, where 𝐸 𝑄 [∙] denotes the expectation taken under the risk-neutral measure 𝑄 and
𝑓𝑄 (𝑋) is the risk-neutral density for the terminal stock price 𝑋(𝑇) . We use a
probability measure 𝑄 to calculate probabilities of events in the risk-neutral world, as
opposed to the usual probability measure 𝑃 for the real world. The derivation of
option price is then straightforward.
In the BSM model, we have seen that stock price follows GBM which means
under risk-neutral measure 𝑄, we have:
ln𝑋(𝑇) ~ 𝑁(ln𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇, 𝜎 2 𝑇), 𝑋(𝑇) > 0
This result is derived using eq. (33) from p. 14 of lecture notes Week 8-12 II, by
replacing 𝜇 with risk-free rate 𝑟. As a result, 𝑋(𝑇) has a lognormal distribution and
therefore its probability density function (PDF) under 𝑄, 𝑓𝑄 (𝑋), can be expressed as:
1
𝑓𝑄 (𝑋) = 𝑓𝑁(0,1) (𝑧)
𝑋(𝑇)𝜎√𝑇
7
, where 𝑓𝑁(0,1) (𝑧) denotes the PDF of standard normal distribution:
1 1
𝑓𝑁(0,1) (𝑧) = exp (− 𝑧 2 )
√2𝜋 2
and 𝑧 is a value from the standardized variable 𝑍:
ln𝑋(𝑇) − ln𝑋(0) − (𝑟 − 𝜎 2 /2)𝑇
𝑍= ~𝑁(0, 1)
𝜎√𝑇
Me again!
➢ Proof
Let 𝑌~𝑙𝑜𝑔𝑛𝑜𝑟𝑚𝑎𝑙 (𝜇, 𝜎 2 ), and so ln 𝑌 ~𝑁(𝜇, 𝜎 2 ); we have:
d d
𝑓𝑙𝑜𝑔𝑛𝑜𝑟𝑚𝑎𝑙 (𝑦) = Pr(𝑌 ≤ 𝑦) = Pr(ln 𝑌 ≤ ln 𝑦)
d𝑦 d𝑦
d ln 𝑌 − 𝜇 ln 𝑦 − 𝜇 d ln 𝑦 − 𝜇 d ln 𝑦 − 𝜇
= Pr ( ≤ )= Pr (𝑍 ≤ )= 𝐹𝑁(0,1) ( )
d𝑦 𝜎 𝜎 d𝑦 𝜎 d𝑦 𝜎
1 ln 𝑦 − 𝜇
=( ) 𝑓𝑁(0,1) ( )
𝜎𝑦 𝜎
1 1 1 ln 𝑦 − 𝜇 2
=( ) exp (− ( ) )
𝜎𝑦 √2𝜋 2 𝜎
8
We can then evaluate the integral for 𝐸 𝑄 [max(𝑋(𝑇) − 𝐾, 0)]:
∞ ∞ ∞
∫ (𝑋 − 𝐾) 𝑓𝑄 (𝑋)d𝑋 = ∫ 𝑋 𝑓𝑄 (𝑋)d𝑋 − 𝐾 ∫ 𝑓𝑄 (𝑋) d𝑋
𝐾 𝐾 𝐾
= 𝑋(0)𝑒 𝑟𝑇 𝑁(𝑑1 ) − 𝐾𝑁(𝑑2 )
⟹ 𝑐 = 𝑋(0)𝑁(𝑑1 ) − 𝐾𝑒 −𝑟𝑇 𝑁(𝑑2 )
➢ Proof
We first consider the second integral:
∞
∫ 𝑓𝑄 (𝑋) d𝑋
𝐾
[Link]
This is the area under the lognormal PDF and to the right of 𝐾, which is simple the
probability -under the risk-neutral measure 𝑄- of (𝑋 > 𝐾). Thus, we have:
∞
∫ 𝑓𝑄 (𝑋) d𝑋 = 𝑄(𝑋 > 𝐾) = 𝑄(ln 𝑋 > ln 𝐾)
𝐾
⚫ A short question: why it is okay to have ln 𝑋 > ln 𝐾 from 𝑋 > 𝐾?
9
ln 𝑋 − (ln𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇) ln 𝐾 − (ln𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇)
= 𝑄( > )
𝜎√𝑇 𝜎√𝑇
ln 𝐾 − (ln𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇)
= 𝑄 (𝑍 > )
𝜎√𝑇
ln𝑋(0) − ln 𝐾 + (𝑟 − 𝜎 2 /2)𝑇
= 𝑄 (𝑍 ≤ ) = 𝑁(𝑑2 )
𝜎√𝑇
Recall that:
ln(𝑋(0)/𝐾) + (𝑟 + 𝜎 2 /2)𝑇 ln 𝑋(0) − ln 𝐾 + (𝑟 + 𝜎 2 /2)𝑇
𝑑1 = =
𝜎√𝑇 𝜎√𝑇
𝑑2 = 𝑑1 − 𝜎√𝑇
Now we turn to the first integral and 𝑁(𝑑1 ). The first integral is related to the
following conditional expectation of 𝑋 under 𝑄:
∞
𝐸 𝑄 [𝑋𝟏{𝑿>𝑲} ]
∫ 𝑋 𝑓𝑄 (𝑋)d𝑋 = 𝐸 𝑄 [𝑋|𝑋 > 𝐾] =
𝐾 𝑄(𝑋 > 𝐾)
, where 𝟏{𝑿>𝑲} is an indicator function for the event {𝑋 > 𝐾}:
1, 𝑖𝑓 𝑋(𝑇) > 𝐾
𝟏{𝑿>𝑲} = {
0, , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Thus, we can evaluate the integral as:
∞ ∞
∫ 𝑋 𝑓𝑄 (𝑋)d𝑋 = ∫ 𝑋 𝟏{𝑿>𝑲} 𝑓𝑄 (𝑋)d𝑋 = 𝐸 𝑄 [𝑋𝟏{𝑿>𝑲} ]
𝐾 0
Obviously, we can use simulation to find out this expected value. But there exists an
analytical solution. The key is to deal with the indicator function in the integral.
10
Since the lognormal density 𝑓𝑄 (𝑋) is proportional to the standard normal PDF
1
𝑓𝑁(0,1) (𝑧) by a factor 𝑋(𝑇)𝜎√𝑇, we can rewrite the integral as:
∞ ∞ ∞
1
∫ 𝑋 𝑓𝑄 (𝑋)d𝑋 = ∫ 𝑋 𝟏{𝑿>𝑲} 𝑓𝑄 (𝑋)d𝑋 = ∫ 𝑋 𝟏{𝑿>𝑲} 𝑓𝑁(0,1) (𝑍)d𝑋
𝐾 0 𝑋𝜎√𝑇 0
∞
1 1 1
= ∫ 𝑋𝟏{𝑿>𝑲} exp (− 𝑍 2 ) d𝑋
𝑋𝜎√𝑇 0 √2𝜋 2
, where
ln 𝑋(𝑇) − (ln 𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇)
𝑍= ~ 𝑁(0, 1)
𝜎√𝑇
We then rewrite the integral in terms of 𝑍 via the following conversions:
1
d𝑍 = d𝑋 ⟹ d𝑋 = 𝑋𝜎√𝑇d𝑍
𝑋𝜎√𝑇
2 /2)𝑇
𝑋(𝑇) = exp(𝑍𝜎√𝑇 + ln 𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇) = 𝑋(0)𝑒 𝑍𝜎√𝑇+(𝑟−𝜎
ln 𝐾 − (ln 𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇)
{𝑋 > 𝐾} = {𝑍 > }≡𝐵
𝜎√𝑇
In addition, the lower and upper bounds of the integral will become:
lower bound: 𝑋(𝑇) = 0 ⟹ 𝑍 = ln 𝑋(𝑇) → −∞
upper bound: 𝑋(𝑇) = ∞ ⟹ 𝑍 = ln 𝑋(𝑇) → ∞
As a result:
∞ ∞
2 /2)𝑇 1 1
⟹ ∫ 𝑋 𝑓𝑄 (𝑋)d𝑋 = ∫ 𝑋(0)𝑒 𝑍𝜎√𝑇+(𝑟−𝜎 𝟏𝑩 exp (− 𝑍 2 ) d𝑍
𝐾 −∞ √2𝜋 2
11
∞
1 1
= 𝑋(0)𝑒 𝑟𝑇 ∫ 𝟏𝑩 exp (− 𝑍 2 + 𝑍𝜎√𝑇 − 𝜎 2 𝑇/2) d𝑍
−∞ √2𝜋 2
∞
1 1
= 𝑋(0)𝑒 𝑟𝑇 ∫ 𝟏𝑩 exp (− (𝑍 2 − 2𝑍𝜎√𝑇 + 𝜎 2 𝑇)) d𝑍
−∞ √2𝜋 2
∞
𝑟𝑇
1 1 2
= 𝑋(0)𝑒 ∫ 𝟏𝑩 exp (− (𝑍 − 𝜎√𝑇) ) d𝑍
−∞ √2𝜋 2
At this stage, we can take 𝑍 − 𝜎√𝑇 = 𝑍′ and the integral becomes:
∞
1 1
∫ 𝟏{𝒁′ <𝒅𝟏 } exp (− (𝑍′)2 ) d𝑍 ′
−∞ √2𝜋 2
, since d𝑍 = d𝑍′ and the event 𝐵 becomes {𝑍 ′ < 𝑑1 }, because:
ln 𝐾 − (ln 𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇)
𝐵 = {𝑍 > }
𝜎√𝑇
ln 𝐾 − (ln 𝑋(0) + (𝑟 − 𝜎 2 /2)𝑇)
= {𝑍 − 𝜎√𝑇 > − 𝜎√𝑇}
𝜎√𝑇
ln 𝐾 − ln 𝑋(0) − (𝑟 + 𝜎 2 /2)𝑇
= {𝑍′ > } = {𝑍′ > −𝑑1 } = {𝑍 ′ < 𝑑1 }
𝜎√𝑇
As a result, we show that:
∞ ∞
1 1
∫ 𝑋 𝑓𝑄 (𝑋)d𝑋 = 𝑋(0)𝑒 𝑟𝑇 ∫ 𝟏{𝒁′ <𝒅𝟏 } exp (− (𝑍′)2 ) d𝑍 ′
𝐾 −∞ √2𝜋 2
= 𝑋(0)𝑒 𝑟𝑇 𝑁(𝑑1 )
12
Dear all,
Thank you for choosing this course, and for all the efforts you made in it. I appreciate
your contributions during the past few months which greatly enrich the content of the
lectures.
“Now this is not the end. It is not even the beginning of the end. But it is, perhaps, the
end of the beginning.” By Winston Churchill.
Hope one day when you have an opportunity to go back to these lecture notes,
you will see through these equations, theorems and proofs with a different eye and
find them making more sense to you – in a delightful way.
Vincent
13