The Deriving of Black-Scholes Equation

Basic Wiener Process

dz=εdt\mathrm{d}z = \varepsilon \sqrt{\mathrm{d}t}

where ε\varepsilon has a standard normal distribution ϕ(0,1)\phi(0, 1).

Generalized Wiener Process

A generalized Wiener process for a variable xx can be defined in terms of dx\mathrm{d} x

dx=adt+bdz\mathrm{d} x = a \mathrm{d} t + b\mathrm{d}z

where a and b are constants.

Ito Process

An Ito process for a variable dx\mathrm{d}x can be written as

dx=a(x,t)dt+b(x,t)dz\mathrm{d}x = a( x, t) \mathrm{d}t + b( x, t )\mathrm{d}z

where a(x,t)a(x,t) and b(x,t)b(x, t) are functions of the value of the underlying variable x and time t, and dz is a Wiener process.

Ito's Lemma

Suppose that the value of a variable x\mathrm{x} follows the Ito process, Ito's lemma shows that a function G of x and t follows the process

dG=(Gxa(x,t)+Gt+122Gx2b(x,t)2)dt+Gxb(x,t)dzdG = \left ( \frac{\partial{G}}{\partial{x}}a(x,t)+\frac{\partial{G}}{\partial{t}}+\frac{1}{2} \frac{\partial^2{G}}{\partial{x^2}} b(x,t)^2 \right )\mathrm{d}t+\frac{\partial{G}}{\partial{x}}b(x,t)\mathrm{d}z

Note, G also follows an Ito process.

Proof

First of all, let's solve the mean and variance of YYgiven Y=X2Y=X^2where Xϕ(0,σ2)X \sim \phi(0, \sigma^2).

Since σ2=E(X2)E(X)2\sigma^2 = E(X^2) - E(X)^2, and E(X)=0E(X) = 0, we get

E(X2)=σ2E(X^2) = \sigma^2

And becuaseVarX2=EX4(EX2)2VarX^2 = EX^4 - (EX^2)^2, and

E(X4)=x42πσex22σ2dx=σ2x32πσex22σ2dx22σ2=σ2x32πσdex22σ2=12πσ(σ2x3ex22σ2++3σ2x2ex22σ2dx)=12πσ(03σ4xex22σ2++3σ4ex22σ2dx)=3σ4\begin{aligned} E(X^4) &= \int \frac{x^4}{\sqrt{2\pi}\sigma}e^{-\frac{x^2}{2\sigma^2}}dx \\ &=\int -\frac{\sigma^2x^3}{\sqrt{2\pi}\sigma}e^{-\frac{x^2}{2\sigma^2}}d-\frac{x^2}{2\sigma^2} \\&=\int -\frac{\sigma^2x^3}{\sqrt{2\pi}\sigma}de^{-\frac{x^2}{2\sigma^2}} \\&= \frac{1}{\sqrt{2\pi}\sigma}\left ( -\sigma^2x^3e^{-{\frac{x^2}{2\sigma^2}}}|_{-\infty }^{+\infty } + 3\sigma^2\int x^2e^{-{\frac{x^2}{2\sigma^2}}}dx \right ) \\ &= \frac{1}{\sqrt{2\pi}\sigma} \left ( 0 - 3\sigma^4xe^{-{\frac{x^2}{2\sigma^2}}}|_{-\infty }^{+\infty } + 3\sigma^4\int e^{-{\frac{x^2}{2\sigma^2}}}dx \right ) \\ &= 3\sigma^4 \end{aligned}

Actually, the generalized form of E[X2n]E[X^{2n}]is,

E[X2n]=(2n1)!!σ2nE\left [ X^{2n}\right ] = (2n - 1)!!\sigma^{2n},

where!!!! is the double factorial.

Thus,

VarX2=EX4(EX2)2=3σ4σ4=2σ4\begin{aligned} VarX^2 &= EX^4 - (EX^2)^2 \\ &= 3\sigma^4 - \sigma^4 \\ &= 2\sigma^4 \end{aligned}

Consider a continuous and differentiable function G of two variables x and t, a Taylor series expansion of ΔG\Delta G is

ΔG=GxΔx+GtΔt+122Gx2Δx2+2GxtΔxΔt+122Gt2Δt2+\Delta{G}=\frac{\partial{G}}{\partial{x}}\Delta{x} + \frac{\partial{G}}{\partial{t}}\Delta{t} + \frac{1}{2}\frac{\partial^2G}{\partial{x}^2}\Delta{x}^2 + \frac{\partial^2{G}}{\partial{x}\partial{t}}\Delta{x}\Delta{t}+\frac{1}{2}\frac{\partial^2{G}}{\partial{t}^2}\Delta{t}^2+\dots

discretize the Ito process to

Δx=a(x,t)Δt+b(x,t)Δz=a(x,t)Δt+b(x,t)εΔt\begin{aligned} \mathrm{\Delta}x &= a( x, t) \mathrm{\Delta}t + b( x, t )\mathrm{\Delta}z \\ &= a(x, t)\Delta{t}+b(x,t)\varepsilon\sqrt{\Delta{t}} \end{aligned}

then we have

Δx2=b2(x,t)ε2Δt+terms of higher order in Δt\Delta{x}^2 = b^2(x, t)\varepsilon^2\Delta{t} + \text{terms of higher order in } \Delta{t}

It shows that the term involving Δx2\Delta{x}^2 in the Taylor series expansion cannot be ignored, and

E(X2)=σ2 and Var(X2)=2σ4E(X^2) = \sigma^2 \text{ and }Var{(X^2)} = 2\sigma^4

Since the variance of a standard normal distribution is 1, E(ε2Δt)=ΔtE(\varepsilon^2\Delta{t}) = \Delta{t} and Var(ε2Δt)=2Δt2.Var(\varepsilon^2\Delta{t}) = 2\Delta{t}^2.

We know that the variance of the change in a stochastic variable in time Δt\Delta{t} is proportional to Δt\Delta{t}, not Δt2\Delta{t}^2. The variance of ε2Δt\varepsilon^2\Delta{t} is therefore too small for it to have a stochastic component. As a result, we can treat it as nonstochastic and equal to its expected value, Δt\Delta{t}, as Δt\Delta{t} tends to zero.

Taking limits as Δx\Delta{x} and Δt\Delta{t} tend to zero, and using dx2=b2(x,t)dtdx^2 = b^2(x,t)dt, we obtain

dG=Gxdx+Gtdt+122Gx2b2(x,t)dtdG = \frac{\partial{G}}{\partial{x}}dx + \frac{\partial{G}}{\partial{t}}dt + \frac{1}{2}\frac{\partial^2{G}}{\partial{x}^2}b^2(x,t)dt

if we substitute the dx with dx=a(x,t)dt+b(x,t)dz\mathrm{d}x = a( x, t) \mathrm{d}t + b( x, t )\mathrm{d}z, equation becomes

dG=(Gxa(x,t)+Gt+122Gx2b2(x,t))dt+Gxb(x,t)dzdG = \left ( \frac{\partial{G}}{\partial{x}}a(x,t) + \frac{\partial{G}}{\partial{t}} + \frac{1}{2}\frac{\partial^2G}{\partial{x}^2} b^2(x,t)\right )dt + \frac{\partial{G}}{\partial{x}}b(x,t)dz

Property of Stock Prices

The Process for A Stock Return

The most widely used model of stock return is

dSS=μdt+σdz and ΔSSϕ(μΔt,σ2Δt)\frac{dS}{S} = \mu \mathrm{d}t + \sigma \mathrm{d}z \text{ and } \frac{\Delta{S}}{S} \sim \phi(\mu\Delta{t}, \sigma^2\Delta{t})

where μ\mu is the stock's expected rate of return, and σ\sigma is the volatility of the stock price. It represents the stock return process in the real world. In a risk-neutral world, μ\mu equals the risk-free rate rr.

The Log Return

The Taylor series expansion of ln(1+x)\ln{(1+x)} is

ln(1+x)=0+11+0×Δx121(1+0)2×Δx2+x\ln{(1+x)} = 0 + \frac{1}{1+0}\times \Delta{x} - \frac{1}{2}\frac{1}{(1+0)^2} \times \Delta{x}^2 + \dots \approx x

for small Δx\Delta{x} and ignore the terms of higher order in Δx\Delta{x}.

Thus, lnStS0=ln(1+StS0S0)ΔSS0 ONLY for small ΔS\ln{\frac{S_t}{S_0}} = \ln{(1 + \frac{S_t-S_0}{S_0})} \approx \frac{\Delta{S}}{S_0} \text{ ONLY for small } \Delta{S}, also note that StS_t here is instantaneous price.

The log rate of return is also called the continuously compounded rate of return.

The Lognormal Property of Stock Prices

Define

G=lnSG = \ln{S}

Where S follows dSS=μdt+σdz\frac{dS}{S} = \mu \mathrm{d}t + \sigma \mathrm{d}z.

Since

GS=1S,2GS2=1S2,Gt=0\frac{\partial{G}}{\partial{S}} = \frac{1}{S}, \frac{\partial^2G}{\partial{S^2}} = - \frac{1}{S^2}, \frac{\partial{G}}{\partial{t}} = 0

Using Ito's Lemma, we can get

dG=(μσ22)dt+σdzdG = \left ( \mu - \frac{\sigma^2}{2}\right)dt + \sigma dz

Since μ\muand σ\sigmaare constant, this equation indicates that G=lnSG = \ln{S}follows a generalized Wiener process. It has constant drift rate μσ2/2\mu - \sigma^2/2and constant variance rate σ2\sigma^2. The change in lnS\ln{S}between time 0 and some future time TTis therefore normally distributed, with mean (μσ2/2)\left( \mu-\sigma^2/2\right)and variance σ2T\sigma^2T. This means that

lnSTS0ϕ[(uσ2/2)T,σ2T]\ln\frac{S_T}{S_0} \sim \phi{\left [\left ( u - \sigma^2/2 \right )T, \sigma^2T \right ]}

Where STS_Tis the stock price at time T, S0S_0is the stock price at time 0.

Black-Scholes Equation

c=S0N(d1)KerTN(d2)c = S_0N(d_1) - Ke^{-rT}N(d_2)
p=KerTN(d2)S0N(d1)p = Ke^{-rT}N(-d_2)-S_0N(-d_1)

Where

d1=ln(S0/K)+(r+σ2/2)TσTd_1 = \frac{\ln{\left ( S_0 / K \right )} + \left ( r + \sigma^2/2\right )T}{\sigma \sqrt{T}}
d2=ln(S0/K)+(rσ2/2)TσT=d1σTd_2 = \frac{\ln{\left ( S_0/K\right )} +\left ( r - \sigma^2/2\right )T}{\sigma \sqrt T} = d_1 - \sigma \sqrt{T}

Proof

Key Result

If V is lognormally distributed, and the mean of lnV\ln{V}is mm and the standard deviation of lnV\ln{V}is ww, then

E(max(VK,0))=E(V)N(d1)KN(d2)E( \max(V - K, 0)) = E(V)N(d_1) - KN(d_2)

Where

d1=ln[E(V)/K]+w2/2wd_1 = \frac{\ln{\left [ E(V)/K\right ]} + w^2/2}{w}
d2=ln[E(V)/K]w2/2wd_2 = \frac{\ln \left [ E(V) / K\right ] - w^2/2}{w}

Note that as to the call option,

  • N(d2)N(d_2)is the probability of exercise,

  • S0erTN(d1)/N(d2)S_0e^{rT}N(d_1)/N(d_2)is the expected stock price in a risk-neutral world if option is exercised

Similarly, as to the put option

  • N(d2)N(-d_2)is the probability of exercise,

  • S0N(d1)/N(d2)S_0N(-d_1)/N(-d_2)is the expected stock price in a risk-neutral world if option is exercised

Proof of Key Result

Define g(V)g(V)as the probability density function of V. It follows that

E[max(VK,0)]=K(VK)g(V)dVE\left [ \max(V-K, 0) \right ] = \int_{K}^{\infty} (V-K)g(V)dV

From the properties of the lognormal distribution,

E(V)=em+12σ2E(V)=e^{m+\frac{1}{2}\sigma^2}
m=lnE(V)w2/2m = \ln{E(V)} - w^2/2

Define a new variable

Q=f(V)=lnVmwQ = f(V) = \frac{\ln{V} - m}{w}

It follows standard normal distribution,

Denote the density function of QQby h(Q)h(Q)so thath(Q)=12πeQ2/2h(Q) = \frac{1}{\sqrt{2\pi}}e^{-Q^2/2},

And since f(V)f(V)is strictly increasing, we can use the formula for the density of a strictly increasing function.

Support RQ={q=f(v):vRV}R_{Q} = \left \{ q = f(v) : v \in R_{V}\right \},

h(q)={g(f1(q))df1(q)dq,if qRQ0,ifqRQ\begin{aligned} h(q) = \left \{ \begin{aligned} &g(f^{-1}(q))\frac{df^{-1}(q)}{dq}, && \text{if}\ q \in R_{Q}\\ &0, && \text{if} q \notin R_{Q} \end{aligned} \right. \end{aligned}
E[max(VK,0)]=K(VK)g(V)dv=f(K)[f1(Q)K]g(f1(Q))d(f1(Q))dQdQ=(lnKm)/w(eQw+mK)h(Q)dQ=(lnKm)/weQw+mh(Q)dQK(lnKm)/2h(Q)d(Q)\begin{aligned} E\left [ \max(V-K, 0) \right ] &= \int_{K}^{\infty} (V-K)g(V)dv \\ &= \int_{f(K)}^{\infty} \left [f^{-1}(Q) - K\right ]g(f^{-1}(Q))\frac{d(f^{-1}(Q))}{dQ}dQ \\ &= \int_{(\ln{K}-m)/w}^{\infty}\left( e^{Qw+m} - K \right )h(Q)dQ \\ &= \int_{(\ln{K} -m)/w}^{\infty}e^{Qw+m}h(Q)dQ-K\int_{(\ln{K} - m)/2}^{\infty}h(Q)d(Q) \end{aligned}

Now

eQw+mh(Q)=12πe(Q2+2Qw+2m)/2=12πe[(Qw)2+2m+w2]/2=em+w2/22πe[(Qw)2]/2=em+w2/2h(Qw)\begin{aligned} e^{Qw+m}h(Q) &= \frac{1}{\sqrt{2\pi}}e^{(-Q^2+2Qw+2m)/2} \\ &=\frac{1}{\sqrt{2\pi}}e^{\left [ -(Q-w)^2+2m+w^2 \right ]/2} \\ &=\frac{e^{m+w^2/2}}{\sqrt{2\pi}}e^{\left [ -(Q-w)^2\right ]/2} \\ &= e^{m+w^2/2}h(Q-w) \end{aligned}

This means that

E(max(VK,0))=em+w2/2(lnKm)/wh(Qw)dQK(lnKm)/wh(Q)dQ=em+w2/2(lnKm)/wwh(Qw)d(Qw)K(lnKm)/wh(Q)dQ=em+w2/2[1N(lnKmww)]K[1N(lnKmw)]=em+w2/2N(lnK+mw+w)KN(lnK+mw)=E(V)N(lnK+lnE(V)w2/2w+w)KN(lnK+lnE(V)w2/2w)=E(V)N(lnE(V)K+w2/2w)KN(lnE(V)Kw2/2w)=E(V)N(d1)KN(d2)\begin{aligned} E(max(V-K, 0)) &= e^{m+w^2/2}\int_{(\ln{K}-m)/w}^{\infty}h(Q-w)dQ - K\int_{(\ln{K}-m)/w}^{\infty}h(Q)dQ \\ &= e^{m+w^2/2}\int_{(\ln{K}-m)/w-w}^{\infty}h(Q-w)d(Q-w) - K\int_{(\ln{K}-m)/w}^{\infty}h(Q)dQ \\ &= e^{m+w^2/2}\left[1 - N\left (\frac{\ln{K}-m}{w}-w\right )\right] - K\left[1 - N\left(\frac{\ln{K-m}}{w}\right) \right] \\ &=e^{m+w^2/2} N\left (\frac{-\ln{K}+m}{w}+w \right) - K N\left ( \frac{-\ln{K}+m}{w}\right ) \\ &= E(V) N\left (\frac{-\ln{K}+\ln{E(V)}-w^2/2}{w}+w \right) - K N\left ( \frac{-\ln{K}+\ln{E(V)}-w^2/2}{w}\right ) \\ &= E(V) N\left (\frac{\ln{\frac{E(V)}{K}}+w^2/2}{w} \right) - K N\left ( \frac{\ln{\frac{E(V)}{K}}-w^2/2}{w}\right ) \\ &= E(V) N\left (d_1\right) - K N\left ( d_2\right ) \end{aligned}

Where

d1=lnE(V)K+w2/2wd_1 = \frac{\ln{\frac{E(V)}{K}}+w^2/2}{w}
d2=lnE(V)Kw2/2wd_2 = \frac{\ln{\frac{E(V)}{K}}-w^2/2}{w}

The Black-Scholes Result

We now consider a call option on a non-dividend-paying stock maturing at time T. The strike price is K, the risk-free rate is r, the current stock price is S0S_0, the stock price at time T is STS_{T} and the volatility is σ\sigma. The call price is given by

c=erTEˆ(max(STK,0))c = e^{-rT}\^{E}(\max(S_{T}-K, 0))

Where Eˆ\^{E}denotes the expectation in a risk-neutral world. As mentioned above, under the stochastic process assumed by Black-Scholes, STS_{T}is lognormal. Based on the deduction above, this means that

lnSTS0ϕ[(rσ2/2)T,σ2T]\ln\frac{S_T}{S_0} \sim \phi{\left [\left ( r - \sigma^2/2 \right )T, \sigma^2T \right ]}

Based on the property of lognormal, Eˆ(STS0)=erT\^{E}(\frac{S_T}{S_0})=e^{rT}, and Var(STS0)=e2rT(eσ2T1)Var\left( \frac{S_T}{S_0}\right)=e^{2rT}\left(e^{\sigma^2T}-1 \right), from the key result just proved, it implies

c=erT[S0erTN(d1)KN(d2)]=S0N(d1)KerTN(d2)c = e^{-rT}\left [ S_{0}e^{rT}N(d_1)-KN(d_2) \right ] = S_0N(d_1)-Ke^{-rT}N(d_2)

Where

d1=ln[Eˆ(ST)/K]+σ2T/2σT=ln(S0/K)+(r+σ2/2)TσTd_1 = \frac{\ln{\left [ \^{E}(S_T)/K\right ]+\sigma^2T/2}}{\sigma \sqrt{T}} = \frac{\ln(S_0/K)+(r+\sigma^2/2)T}{\sigma \sqrt{T}}
d2=ln[Eˆ(ST)/K]σ2T/2σT=ln(S0/K)+(rσ2/2)TσTd_2 = \frac{\ln{\left [ \^{E}(S_T)/K\right ]-\sigma^2T/2}}{\sigma \sqrt{T}} = \frac{\ln(S_0/K)+(r-\sigma^2/2)T}{\sigma \sqrt{T}}

Proof of the Put Option

We can do it in the same way as above. However, here we will do it in another way.

We know that as put option, N(d2)N(-d_2) is the probability of exercise,

Eˆ(max(KST,0))=d2(KS0e[(r12σ2)T+σTε])f(ε)dε=KN(d2)d2S0erT12πe(εσT)22dε=KN(d2)d2σTS0erT12πe(εσT)22d(εσT)=KN(d2)d1S0erT12πe(εσT)22d(εσT)=KN(d2)S0erTN(d1)\begin{aligned} \^{E}(\max(K-S_T, 0))&=\int_{-\infty}^{-d_2}\left ( K - S_0e^{\left [(r-\frac{1}{2}\sigma^2)T+\sigma \sqrt{T}\varepsilon\right ]}\right )f(\varepsilon)d\varepsilon \\ &=KN(-d_2) - \int_{-\infty}^{-d_2}S_0e^{rT}\frac{1}{\sqrt{2\pi}}e^{-\frac{(\varepsilon-\sigma \sqrt{T})^2}{2}}d\varepsilon \\ &=KN(-d_2) - \int_{-\infty}^{-d_2-\sigma \sqrt{T}}S_0e^{rT}\frac{1}{\sqrt{2\pi}}e^{-\frac{(\varepsilon-\sigma \sqrt{T})^2}{2}}d(\varepsilon-\sigma \sqrt{T}) \\ &= KN(-d_2) - \int_{-\infty}^{-d_1}S_0e^{rT}\frac{1}{\sqrt{2\pi}}e^{-\frac{(\varepsilon-\sigma \sqrt{T})^2}{2}}d(\varepsilon-\sigma \sqrt{T}) \\ &= KN(-d_2)-S_0e^{rT}N(-d_{1}) \end{aligned}

where

  • ε\varepsilon has a standard normal distribution ϕ(0,1)\phi(0, 1)

  • f(ε)=12πeε22f(\varepsilon) = \frac{1}{\sqrt{2 \pi}}e^{-\frac{\varepsilon^2}{2}}


Strictly speaking, this is not a deduction of the Black-Scholes equation, because we were using the result from the equation that N(d2)N(-d_2)is the probability of exercise. And if the function underneath the expectation has been changed, the probability of exercise could change along with that.

Last updated

Was this helpful?