Concept: The Black-Scholes Model
The Full Mathematical Derivation of the Call Formula
To truly understand Black-Scholes, we must walk step-by-step from the assumption of the asset’s random walk to the final pricing formula. We do this by deriving Itô’s Lemma, establishing the Partial Differential Equation (PDE), transforming it into an expected value, and integrating.
1. Introduction: Starting with a stochastic model of the underlying asset price
Suppose we have a stochastic process \(X_t\) that models our asset price, following a generalized Itô drift-diffusion process:
\[ dX_t = \mu(X_t, t)dt + \sigma(X_t, t)dW_t \]
Where:
* \(\mu(X_t, t)\) is the predictable, deterministic drift.
* \(\sigma(X_t, t)\) is the volatility (diffusion) scaling factor.
* \(dW_t\) is the increment of a standard Wiener process (Brownian motion), representing random market shocks.
Now that we have a function that describes the movement of the underlying asset price, we want to have a function that describes the movement of the option price on top of it. Therefore, working with this function of the underlying stock, we want to find the option pricing function on top of this, more specifically, the differential \(df\) of a scalar function \(f(X_t, t)\), that is at least twice-differentiable in \(x\) and once-differentiable in \(t\). The option pricing function needs to be at least twice-differentiable in \(x\) because Ito Calculus shows that the second-order term in the Taylor expansion does not disappear like in traditional calculus.
To find this differential, we can’t simply reach for the standard chain rule of classical calculus. Because our underlying price \(X_t\) is driven by the ‘infinite zig-zags’ of Brownian motion, it is too jagged to be handled by ordinary derivatives. To build a bridge between the movement of the asset and the movement of the function, we must return to first principles: the Taylor Series expansion. As we are about to see, the rules for what we consider ‘negligible’ are fundamentally rewritten when the underlying world refuses to sit still.
2. The Divergence from Ordinary Calculus (Taylor Series Expansion)
We begin by examining the change in our function \(f\) over an infinitesimally small time increment \(\Delta t\). We use a two-variable Taylor expansion:
\[ \Delta f \approx \frac{\partial f}{\partial t}\Delta t + \frac{\partial f}{\partial x}\Delta X + \frac{1}{2}\frac{\partial^2 f}{\partial x^2}(\Delta X)^2 + \frac{1}{2}\frac{\partial^2 f}{\partial t^2}(\Delta t)^2 + \frac{\partial^2 f}{\partial x \partial t}(\Delta X \Delta t) + \dots \]
The Core Difference: In standard, deterministic calculus, as \(\Delta t \to 0\), any terms of order higher than \(\Delta t\) (like \(\Delta t^2\) or \(\Delta x^2\), the 3rd and 4th term in the above equation) shrink to zero so rapidly that they become negligible. We simply drop them.
However, in the stochastic world, the path of \(X_t\) is driven by the random walk of \(W_t\). Brownian motion is continuous but nowhere differentiable—it is infinitely jagged.
Because of this extreme jaggedness, the squared variation \((\Delta X)^2\) does not shrink to zero faster than \(\Delta t\). It is large enough that it must be kept in our equation.
3. Analyzing \((\Delta X)^2\) and the Rules of Stochastic Calculus
To see why the second-order term survives, we substitute our definition of \(\Delta X\) into the squared term:
\[ (\Delta X)^2 = (\mu \Delta t + \sigma \Delta W)^2 \] \[ (\Delta X)^2 = \mu^2 (\Delta t)^2 + 2\mu\sigma (\Delta t \Delta W) + \sigma^2 (\Delta W)^2 \]
Now, we take the limit as \(\Delta t \to 0\) and apply the formal multiplication rules of stochastic calculus (Itô multiplication table):
* \(dt \cdot dt = 0\) (Shrinks too fast)
* \(dt \cdot dW_t = 0\) (Shrinks too fast, scales as \(dt^{3/2}\))
* \(dW_t \cdot dW_t = dt\) (The critical survival)
Deep Dive: Why does \((dW_t)^2 = dt\)?
The jump to \((dW_t)^2 = dt\) is the most unintuitive part of stochastic calculus. How can a random variable squared equal a deterministic, predictable passage of time? We can break this down statistically, behaviorally, and heuristically.
1. The Statistical Properties
Here is the exact step-by-step logic chain to prove this equivalence:
* Definition: By definition, an increment of a Wiener process follows a normal distribution centered at zero: \(\Delta W \sim \mathcal{N}(0, \Delta t)\).
* Variance Formula: From fundamental statistics, the variance of any random variable is \(Var[\Delta W] = E[(\Delta W)^2] - (E[\Delta W])^2\).
* Plug in the numbers: Since the mean expected value \(E[\Delta W]\) is exactly \(0\), squaring it \((E[\Delta W])^2\) is also \(0\).
* The Reduction: This leaves us with a perfectly simplified equation: \(Var[\Delta W] = E[(\Delta W)^2]\).
* The Result: Because the variance of a Wiener process over time \(\Delta t\) is defined as \(\Delta t\), it must be mathematically true that the expected value of the squared increment is the time step itself: \(E[(\Delta W)^2] = \Delta t\).2. The Law of Large Numbers (Quadratic Variation)
One might naturally ask: “If \(E[(\Delta W)^2] = \Delta t\), isn’t \((\Delta W)^2\) still a random variable around that expected value?”
Yes, for a single, isolated step, it is. However, Itô’s Lemma deals with continuous time, where the limit as \(\Delta t \to 0\). If we divide a time interval \([0, T]\) into \(n\) microscopic pieces, the total sum of these squared increments is called the Quadratic Variation: \[[W, W]_T = \lim_{n \to \infty} \sum_{i=1}^{n} (W_{t_i} - W_{t_{i-1}})^2\]
As \(n\) becomes infinitely large, we must look at the variance of this sum. For a normal distribution, the fourth moment allows us to calculate the variance of the square: \[Var[(\Delta W)^2] = E[(\Delta W)^4] - (E[(\Delta W)^2])^2 = 3(\Delta t)^2 - (\Delta t)^2 = 2(\Delta t)^2\]
Because the variance of \((\Delta W)^2\) is proportional to \((\Delta t)^2\), it shrinks to zero vastly faster than the mean (\(\Delta t\)). In the continuous limit, the randomness completely “washes out,” leaving only the deterministic expected value. The sum of squares behaves perfectly like a ticking clock.
3. The Heuristic Comparison
To visualize why this is unique to random walks, compare it to a standard smooth function, \(g(t) = t\): * Standard Calculus: \(\Delta g = \Delta t\). Therefore, \((\Delta g)^2 = (\Delta t)^2\). As \(\Delta t \to 0\), this squared term essentially vanishes to zero. * Stochastic Calculus: \(\Delta W \approx \sqrt{\Delta t}\). Therefore, \((\Delta W)^2 \approx \Delta t\).
Because Brownian motion is inherently “square-root-of-time” risky, its square is of the exact same “size” as time itself.
4. Final Substitution
Now we return to our original Taylor expansion. Knowing which terms survive the limit as \(\Delta t \to 0\), we truncate the series, keeping only the terms of order \(dt\) or \(dW_t\):
\[ df = \frac{\partial f}{\partial t}dt + \frac{\partial f}{\partial x}dX_t + \frac{1}{2}\frac{\partial^2 f}{\partial x^2}(dX_t)^2 \]
Next, we substitute our initial definitions:
* \(dX_t = \mu dt + \sigma dW_t\)
* \((dX_t)^2 = \sigma^2 dt\)
\[ df = \frac{\partial f}{\partial t}dt + \frac{\partial f}{\partial x}(\mu dt + \sigma dW_t) + \frac{1}{2}\frac{\partial^2 f}{\partial x^2}(\sigma^2 dt) \]
5. The Result: Itô’s Lemma
Finally, we group the deterministic \(dt\) terms together and separate the stochastic \(dW_t\) term. This yields the foundational equation of stochastic calculus—Itô’s Lemma:
\[ df = \left( \frac{\partial f}{\partial t} + \mu \frac{\partial f}{\partial x} + \frac{1}{2} \sigma^2 \frac{\partial^2 f}{\partial x^2} \right) dt + \sigma \frac{\partial f}{\partial x} dW_t \]
(Note for Black-Scholes: In our options pricing model, \(f\) is the option price \(V(S, t)\), the asset \(X_t\) is the stock price \(S_t\), the drift \(\mu\) becomes \(\mu S_t\), and the diffusion \(\sigma\) becomes \(\sigma S_t\).)
Step 2: Constructing the Riskless Portfolio (Delta Hedging)
Now that we have Itô’s Lemma to describe how the option price \(V(S, t)\) changes over time, we need a way to find its fair value. Fischer Black and Myron Scholes realized that if you hold the right combination of the option and the underlying stock, you can completely cancel out the unpredictable market shocks (\(dW_t\)).
We construct a hypothetical portfolio, denoted as \(\Pi\). In this portfolio, we hold exactly one long position in our option, and we short sell a specific number of shares of the underlying stock. We will call this number of shares \(\Delta\) (not to be confused with a time step \(\Delta t\)).
The value of our portfolio at any given time is: \[ \Pi = V - \Delta S \]
We want to observe how the value of this portfolio changes over a microscopic time step \(dt\). By applying basic differentiation: \[ d\Pi = dV - \Delta dS \]
Now, we substitute the two massive equations we have built so far:
1. \(dV\): The change in the option price (from Itô’s Lemma).
2. \(dS\): The change in the stock price (from our Geometric Brownian Motion assumption: \(dS = \mu S dt + \sigma S dW_t\)).
Substituting these into our portfolio equation yields: \[ d\Pi = \left[ \left( \frac{\partial V}{\partial t} + \mu S \frac{\partial V}{\partial S} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} \right) dt + \sigma S \frac{\partial V}{\partial S} dW_t \right] - \Delta \left[ \mu S dt + \sigma S dW_t \right] \]
This looks messy, but we can organize it by grouping all the deterministic parts (attached to \(dt\)) and all the random, volatile parts (attached to \(dW_t\)):
\[ d\Pi = \left( \frac{\partial V}{\partial t} + \mu S \frac{\partial V}{\partial S} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} - \Delta \mu S \right) dt + \left( \sigma S \frac{\partial V}{\partial S} - \Delta \sigma S \right) dW_t \]
Eliminating the Randomness
The second set of parentheses contains the \(dW_t\) term. This is the sole source of risk in our portfolio. If the market suddenly crashes or spikes, this is the term that dictates our losses or gains.
The brilliance of the Black-Scholes model lies in a simple algebraic observation: we can force that entire term to equal zero. We do this by continuously adjusting the number of shares we short (\(\Delta\)).
If we deliberately choose: \[ \Delta = \frac{\partial V}{\partial S} \]
Then the random term becomes: \[ \left( \sigma S \frac{\partial V}{\partial S} - \frac{\partial V}{\partial S} \sigma S \right) dW_t = 0 \cdot dW_t = 0 \]
By setting our stock position equal to the partial derivative of the option with respect to the stock (which traders now call the option’s “Delta”), the random shocks completely cancel out. The portfolio becomes completely immune to small movements in the underlying asset.
What we are left with is a portfolio whose change in value is entirely deterministic, driven only by the passage of time: \[ d\Pi = \left( \frac{\partial V}{\partial t} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} \right) dt \]
Step 3: The Principle of No-Arbitrage and the Black-Scholes PDE
At the end of Step 2, we successfully engineered a portfolio \(\Pi\) that is completely insulated from the random shocks of the stock market. Because we chose to hold \(\Delta = \frac{\partial V}{\partial S}\) shares of stock short against our long option, the \(dW_t\) term was mathematically forced to zero.
We were left with a portfolio whose change in value is entirely deterministic, driven only by the smooth passage of time: \[ d\Pi = \left( \frac{\partial V}{\partial t} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} \right) dt \]
Now, we must ask an economic question: If a portfolio has absolutely zero risk, how much return should it generate?
The Financial Law of Gravity: No-Arbitrage
In an efficient market, there is a fundamental law known as the Principle of No-Arbitrage. An “arbitrage” is a guaranteed, risk-free profit that requires zero net investment.
If our completely riskless portfolio \(\Pi\) earned a return higher than a risk-free government bond (yielding the risk-free rate, \(r\)), an arbitrageur could borrow infinite amounts of money at rate \(r\) from the bank, invest it into our portfolio \(\Pi\), and pocket the difference completely risk-free. Conversely, if \(\Pi\) earned less than \(r\), an arbitrageur would short our portfolio and put the cash into the bank, again locking in a risk-free profit.
Because market forces instantly exploit and close these loopholes, a fundamental truth emerges: Any portfolio with zero risk must earn exactly the risk-free interest rate.
Mathematically, the growth of our portfolio over the time step \(dt\) must equal the continuously compounded risk-free rate \(r\) multiplied by the current value of the portfolio: \[ d\Pi = r \Pi dt \]
Equating the Math and the Economics
We now have two completely different ways to describe the change in our portfolio (\(d\Pi\)).
1. The Calculus View (from Itô’s Lemma): \(d\Pi = \left( \frac{\partial V}{\partial t} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} \right) dt\)
2. The Economics View (from No-Arbitrage): \(d\Pi = r \Pi dt\)
Because these describe the exact same portfolio, we can set them equal to each other: \[ \left( \frac{\partial V}{\partial t} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} \right) dt = r \Pi dt \]
Recall from Step 2 that the total value of our portfolio is \(\Pi = V - \Delta S\), and we defined \(\Delta = \frac{\partial V}{\partial S}\). Let’s substitute that back in: \[ \left( \frac{\partial V}{\partial t} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} \right) dt = r \left( V - \frac{\partial V}{\partial S} S \right) dt \]
Since both sides are simply multiplying a rate by the time step \(dt\), we can divide both sides by \(dt\) to remove it entirely: \[ \frac{\partial V}{\partial t} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} = rV - rS \frac{\partial V}{\partial S} \]
Finally, let’s rearrange this equation by moving everything to the left side so that it equals zero.
\[ \frac{\partial V}{\partial t} + rS \frac{\partial V}{\partial S} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} - rV = 0 \]
The Grand Reveal: Deconstructing the PDE
This is it. This is the Black-Scholes Partial Differential Equation. It is one of the most famous equations in finance, and every single term has a physical, trading interpretation, often mapped directly to the “Greeks”:
- \(\frac{\partial V}{\partial t}\) (Theta - Time Decay): This is negative for options. As time ticks closer to expiration, the option loses value because there is less time for the stock to make a favorable move.
- \(rS \frac{\partial V}{\partial S}\) (Risk-Neutral Drift): This term represents the cost of financing the Delta-hedged position in the stock.
- \(\frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2}\) (Gamma - Convexity): This is the engine of the option’s value. Because the second derivative (\(\Gamma\)) is positive, volatility (\(\sigma^2\)) directly adds value to the option. This perfectly offsets the bleed from time decay (\(\Theta\)).
- \(- rV\) (Discounting): This simply discounts the final payoff of the option back to present value at the risk-free rate.
The beauty of the Black-Scholes PDE is that the expected return of the stock (\(\mu\)) has completely vanished! Because we can hedge away the directional risk of the stock, the option’s price depends only on the stock’s volatility (\(\sigma\)), not its expected direction.
Interlude: The Fork in the Road (How do we solve this PDE?)
At the end of Step 3, we successfully derived the Black-Scholes PDE: \[ \frac{\partial V}{\partial t} + rS \frac{\partial V}{\partial S} + \frac{1}{2} \sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} - rV = 0 \]
This equation dictates exactly how the option price must behave to prevent arbitrage. But an equation is not a price. To get a dollar value, we have to solve this PDE subject to a boundary condition: at maturity (\(T\)), a call option is worth exactly \(\max(S_T - K, 0)\).
From here, quantitative finance splits into two distinct paths: Numerical Methods and Analytical Methods.
Path A: The Numerical Approach (Finite Difference Methods)
Instead of looking for a perfect algebraic formula, Numerical Methods use brute-force computation. We chop time and asset prices into a massive grid (a matrix). We know the exact payoff at expiration (the rightmost edge of the grid), and we mathematically step backward through time, cell by cell, until we reach “today” (the leftmost edge).
To do this, we approximate the continuous derivatives from our PDE using discrete “differences” on our grid. Let \(V_i^j\) be the option price at stock price node \(i\) and time step \(j\).
* Delta (First Derivative of Space): \(\frac{\partial V}{\partial S} \approx \frac{V_{i+1}^j - V_{i-1}^j}{2\Delta S}\)
* Gamma (Second Derivative of Space): \(\frac{\partial^2 V}{\partial S^2} \approx \frac{V_{i+1}^j - 2V_i^j + V_{i-1}^j}{(\Delta S)^2}\)
* Theta (First Derivative of Time): \(\frac{\partial V}{\partial t} \approx \frac{V_i^{j+1} - V_i^j}{\Delta t}\)
By substituting these grid approximations into the Black-Scholes PDE, we can build a programmatic solver. There are three main ways to construct this grid, trading off between computational speed and mathematical stability.
1. The Explicit Scheme
The Explicit method is the simplest. To find the unknown option price at a specific node one step back in time (\(V_i^j\)), it relies only on the already-known information from the nodes immediately ahead of it in the future (\(V_{i-1}^{j+1}, V_i^{j+1}, V_{i+1}^{j+1}\)).
- Pros: It is incredibly easy to code. The calculation for each node is a simple linear equation.
- Cons: It is conditionally stable. If your time steps (\(\Delta t\)) are too large compared to your price steps (\(\Delta S\)), the mathematical errors compound exponentially. The grid “blows up,” and your computer will output an option price of infinity.
2. The Implicit Scheme
The Implicit method flips the logic. Instead of expressing the current time step as a function of the future, it expresses the future known boundary conditions as a function of the current unknown time step.
Because multiple unknown nodes depend on each other simultaneously, you cannot calculate them one by one. You must solve the entire column of the grid at the exact same time using a system of linear equations (specifically, by inverting a tridiagonal matrix).
- Pros: It is unconditionally stable. No matter how large you make your time steps, the math will never blow up.
- Cons: It is computationally expensive. Solving a massive matrix at every single time step requires significantly more processing power.
3. The Crank-Nicolson Scheme (The Industry Standard)
In 1947, John Crank and Phyllis Nicolson realized that if you take an exact 50/50 average of the Explicit and Implicit methods, magical things happen to the error terms.
The Crank-Nicolson method centers the time difference exactly halfway between step \(j\) and step \(j+1\). Like the Implicit method, it requires solving a tridiagonal matrix (often using a lightning-fast technique called the Thomas Algorithm).
- The Verdict: This is the gold standard for PDE options pricing. It is unconditionally stable (like Implicit) but converges on the true price much faster and with significantly higher accuracy than the other two methods.
Why use Finite Difference Methods? If we just want to price a standard European option, FDM is overkill. But for American Options, which can be exercised early, FDM is mandatory. Because we step backward through time cell by cell, we can stop at every single node and ask the computer: “Is the theoretical option price here lower than the immediate payoff of early exercise? If yes, replace the theoretical price with the exercise value.”
Path B: The Analytical Approach
While grids are powerful, for European options (which cannot be exercised early), we do not need to step through time mechanically. We can use advanced probability theory to solve the PDE perfectly, instantly, and in one continuous leap.
This brings us back to our derivation and the genius of Richard Feynman and Mark Kac.
Step 4: The Risk-Neutral Expectation (Feynman-Kac)
Solving the Black-Scholes PDE directly requires complex algebraic substitutions to strip away the drift and discounting until it looks like a standard heat equation. Instead, we use the Feynman-Kac theorem. This theorem acts as a mathematical bridge between local differential equations and global probability.
It states that the solution to this PDE, subject to the European Call boundary condition at maturity \(V(S, T) = \max(S_T - K, 0)\), can be written simply as a discounted expected value under the risk-neutral measure \(\mathbb{Q}\):
\[ C(S,t) = e^{-r(T-t)} \mathbb{E}^{\mathbb{Q}}[\max(S_T - K, 0)] \]
The Risk-Neutral Measure (\(\mathbb{Q}\)) and the “Fair Game”
Notice the \(\mathbb{Q}\) above the expectation operator. This indicates that we are calculating the expectation under the Risk-Neutral Measure, not the real-world probability measure (\(\mathbb{P}\)).
Why does this work? Remember when we delta-hedged our portfolio in Step 2? The real-world expected return of the stock (\(\mu\)) completely disappeared from our math because the PDE forces it to zero. By stripping away the real-world trend, the discounted option price becomes a Martingale (a mathematically “fair game”).
In a Martingale, the process has zero drift. The expected future value is simply the current value plus random noise. Because the option price does not care about the stock’s true expected growth, we can pretend we live in a simplified “risk-neutral” universe. In this universe, all assets (including the risky stock) are expected to grow at exactly the risk-free rate \(r\).
So, we rewrite our stock’s stochastic differential equation (SDE), replacing \(\mu\) with \(r\): \[ dS_t = r S_t dt + \sigma S_t dW^{\mathbb{Q}}_t \]
Solving for the Terminal Stock Price (\(S_T\)) and the Normal Distribution
To evaluate our expected payoff, we need an explicit formula for \(S_T\). At expiration, \(S_T - K\) is a solid number. However, looking forward from today, \(S_T\) is not a single number; it is a probability density function.
We solve the risk-neutral SDE by applying Itô’s Lemma to the natural log of the stock price, \(f(S_t) = \ln(S_t)\). We use the natural log because stock prices compound geometrically and cannot fall below zero.
Using Itô’s Lemma on \(\ln(S_t)\) yields: \[ d(\ln S_t) = \left( r - \frac{1}{2}\sigma^2 \right) dt + \sigma dW^{\mathbb{Q}}_t \]
(Notice the \(-\frac{1}{2}\sigma^2\) term appearing! This is the exact mathematical origin of the volatility drag we discussed in Section 1. It acts as a mathematical tax where the variance physically pulls down the expected geometric growth rate.)
Because the drift and volatility terms are now constants, we can cleanly integrate both sides from today (\(t\)) to expiration (\(T\)). Let \(\tau = T - t\) (the time to maturity). \[ \ln(S_T) - \ln(S_t) = \left( r - \frac{1}{2}\sigma^2 \right)\tau + \sigma (W_T - W_t) \]
When we integrate the Wiener process \(dW_t\), we accumulate all the random increments over time \(\tau\). By definition, the change in a Wiener process over time \(\tau\) generates a normal distribution with a mean of 0 and a variance of \(\tau\). This creates a standard normal random variable \(Z \sim \mathcal{N}(0,1)\) scaled by \(\sqrt{\tau}\).
By taking the exponential of both sides, we get the explicit, log-normal distribution for the stock price at maturity: \[ S_T = S_t \exp\left( \left(r - \frac{\sigma^2}{2}\right)\tau + \sigma \sqrt{\tau} Z \right) \]
Step 5: Splitting and Solving the Integrals (The Balance Sheet)
Because \(S_T\) is driven by the normally distributed variable \(Z\), we can now rewrite our expectation as an integral over the standard normal Probability Density Function (PDF), which is \(\phi(z) = \frac{1}{\sqrt{2\pi}}e^{-z^2/2}\):
\[ C(S,t) = e^{-r\tau} \int_{-\infty}^{\infty} \max(S_T(z) - K, 0) \phi(z) dz \]
Finding the Boundary Condition (Dropping the Max Function)
The option only has a payoff when \(S_T > K\). Because we do not care about scenarios where the option expires worthless, we need to find the critical value of our normal variable \(Z\) where the stock price exactly equals the strike price (\(S_T = K\)):
\[ S_t \exp\left( \left(r - \frac{\sigma^2}{2}\right)\tau + \sigma \sqrt{\tau} Z \right) > K \] Take the natural log of both sides: \[ \left(r - \frac{\sigma^2}{2}\right)\tau + \sigma \sqrt{\tau} Z > \ln\left(\frac{K}{S_t}\right) \] Isolate \(Z\): \[ Z > \frac{\ln(K/S_t) - (r - \frac{\sigma^2}{2})\tau}{\sigma \sqrt{\tau}} \]
Let’s define this entire right-side boundary threshold as \(-d_2\). Thus, the option is “in-the-money” and pays off whenever \(Z > -d_2\). Because we know the exact lower bound where the value becomes positive, we can drop the \(\max()\) function and restrict our integral bounds strictly to the winning scenarios:
\[ C(S,t) = e^{-r\tau} \int_{-d_2}^{\infty} (S_T(z) - K) \phi(z) dz \]
Splitting the Integral: Cash vs. Asset
We expand the terms and split this into two separate integrals. This perfectly reflects the dual nature of an option exercise: the expected benefit of receiving the stock minus the expected cost of paying the strike. \[ C(S,t) = \underbrace{e^{-r\tau} \int_{-d_2}^{\infty} S_T(z) \phi(z) dz}_{\text{Part A (The Asset Component)}} \quad - \quad \underbrace{K e^{-r\tau} \int_{-d_2}^{\infty} \phi(z) dz}_{\text{Part B (The Cash Component)}} \]
Solving Part B (The Cash Component)
Part B is incredibly straightforward. It represents a “Cash-or-Nothing” bet. The cost of \(K\) is flat, meaning it does not matter if the stock finishes \(\$1\) in the money or \(\$1,000\) in the money; you still only pay exactly \(K\).
Therefore, we only care about the base probability that \(Z > -d_2\). Because the standard normal distribution is perfectly symmetric, the probability of being greater than \(-d_2\) is identical to the probability of being less than \(d_2\). Let \(N(\cdot)\) denote the standard normal Cumulative Distribution Function (CDF). \[ P(Z > -d_2) = P(Z < d_2) = N(d_2) \]
So, Part B simplifies beautifully to the expected, discounted cost of cash: \[ K e^{-r\tau} N(d_2) \]
Solving Part A (The Asset Component and the Mean Shift)
Part A is trickier because it represents an “Asset-or-Nothing” bet. Unlike the flat cost of cash, the value of the stock you receive does depend on how deep in the money you are. If the stock finishes much higher, the payoff is much larger.
To solve this mathematically, we are multiplying the normal distribution by the exponentially growing stock price. Let’s substitute \(S_T(z)\) and \(\phi(z)\) into the integral: \[ e^{-r\tau} \int_{-d_2}^{\infty} S_t e^{(r - \sigma^2/2)\tau + \sigma \sqrt{\tau} z} \left( \frac{1}{\sqrt{2\pi}} e^{-z^2/2} \right) dz \]
We pull the constants out of the integral and combine the \(e\) exponents: \[ S_t e^{-r\tau} e^{r\tau - \sigma^2\tau/2} \int_{-d_2}^{\infty} \frac{1}{\sqrt{2\pi}} e^{-\frac{1}{2}z^2 + \sigma \sqrt{\tau} z} dz \]
The \(e^{-r\tau}\) and \(e^{r\tau}\) cancel out outside the integral. Inside the integral, we must complete the square in the exponent to turn it back into a recognizable normal distribution: \[ -\frac{1}{2}z^2 + \sigma \sqrt{\tau} z = -\frac{1}{2}(z^2 - 2\sigma \sqrt{\tau} z) \] \[ (z^2 - 2\sigma \sqrt{\tau} z) = (z - \sigma \sqrt{\tau})^2 - \sigma^2 \tau \]
Substituting this completed square back into the exponent: \[ S_t e^{-\sigma^2\tau/2} \int_{-d_2}^{\infty} \frac{1}{\sqrt{2\pi}} e^{-\frac{1}{2}(z - \sigma \sqrt{\tau})^2 + \sigma^2\tau/2} dz \]
The \(e^{-\sigma^2\tau/2}\) on the outside and \(e^{\sigma^2\tau/2}\) on the inside cancel perfectly! We are left with: \[ S_t \int_{-d_2}^{\infty} \frac{1}{\sqrt{2\pi}} e^{-\frac{1}{2}(z - \sigma \sqrt{\tau})^2} dz \]
Notice what happened here. By weighting the probability by the actual asset value (an exponential function), this integral has become a new standard normal distribution that is shifted to the right by a mean of \(\sigma \sqrt{\tau}\).
The “center of gravity” for the expected stock value moves forward by exactly one standard deviation of the randomness. Let’s make a final substitution: let \(y = z - \sigma \sqrt{\tau}\), so \(dy = dz\).
When \(z = -d_2\), our new lower bound becomes \(y = -d_2 - \sigma \sqrt{\tau}\). By mathematical definition, we call this new boundary \(-d_1\). Therefore: \[ d_1 = d_2 + \sigma \sqrt{\tau} \]
The integral becomes: \[ S_t \int_{-d_1}^{\infty} \frac{1}{\sqrt{2\pi}} e^{-y^2/2} dy \]
Just like in Part B, this is simply \(S_t \times P(Y > -d_1)\). Due to symmetry, this equals the expected, weighted value of the asset received: \(S_t N(d_1)\).
Step 6: The Final Black-Scholes Formula
By recombining our solved integrals (Part A minus Part B), we finally arrive at the glorious, closed-form Black-Scholes formula for a European Call option:
\[ C(S, t) = S_t N(d_1) - K e^{-r\tau} N(d_2) \]
Where: \[ d_1 = \frac{\ln(S_t/K) + (r + \frac{\sigma^2}{2})\tau}{\sigma \sqrt{\tau}} \] \[ d_2 = d_1 - \sigma \sqrt{\tau} = \frac{\ln(S_t/K) + (r - \frac{\sigma^2}{2})\tau}{\sigma \sqrt{\tau}} \]
Financial Intuition of the Formula
The math is beautiful, but what does it actually mean for a trader? * \(N(d_2)\) is the literal risk-neutral probability that the option will expire “in-the-money” (that \(S_T > K\)). Therefore, \(K e^{-r\tau} N(d_2)\) is simply the expected cost of having to pay the strike price at expiration, discounted to today. * \(N(d_1)\) is the option’s Delta (\(\Delta\)). It represents the probability-weighted expected value of the stock, conditional on the option expiring in the money. It is shifted higher than \(N(d_2)\) because it accounts for the magnitude of the win. Higher stock prices have more mathematical weight, pulling the expectation upward. * Therefore, the formula is simply a logical balance sheet: Call Price = (Expected Benefit of receiving the stock) minus (Expected Cost of paying the strike).