← Courses / Quantitative Finance · Pricing Core

Black-Scholes

Published in 1973 by Black, Scholes, and Merton — the cornerstone of modern derivatives pricing, even though everyone knows it is wrong. This course derives everything from A to Z: Itô's lemma, the PDE via delta-hedging, the closed-form formula via risk-neutral expectation, every Greek, and why the model carries within it the proof of its own limitations — the volatility smile.

Quantitative Finance Pricing Core Quant · Trading · Structuring Itô · PDE · Greeks · IV 15 sections · ~105 min

C = SN(d₁)−Ke⁻ʳᵀN(d₂)

The Formula

Θ ≈ −½σ²S²Γ

Theta-Gamma Relationship

Vega = ΓS²σT

Gamma-Vega Identity

σ(K,T)

The smile — proof that BS is wrong

// Section 1 — Foundations

The Model's Assumptions

In 1973, Black, Scholes, and Merton made a radical bet: rather than trying to faithfully describe the full complexity of real markets, they chose a deliberately simplified set of assumptions that makes the pricing problem analytically solvable in closed form. This choice has a cost — each assumption is, to varying degrees, contradicted by market data — but it is precisely this compromise between realism and tractability that allowed the model to become, fifty years later, the common language of every trading floor in the world. Before writing a single formula, we must therefore understand what the model assumes, why each assumption is necessary for the proofs in the following sections, and how its violation manifests concretely on a trader's book.

// Why start with the assumptions

Knowing the Black-Scholes assumptions means knowing exactly when the model applies — and why it systematically breaks down in practice. This is not an academic exercise: an options desk constantly adjusts its prices and hedges to compensate for the gap between BS's idealized world and the real market, and knowing precisely which gap dominates at any given moment is what distinguishes a junior trader from a senior one. It is also the most common interview question: "Which assumptions does BS violate, and what are the precise consequences?" A vague answer like "the model is unrealistic" immediately signals a candidate who memorized the formula without understanding its structure — the right answer names the assumption, quantifies the effect, and proposes the correction.

Assumption	Statement	Empirical violation
Log-normal diffusion	\(dS_t=\mu S_t\,dt+\sigma S_t\,dW_t\)	Fat tails (kurtosis > 3) — BS underprices deep OTM options
Constant volatility	σ known and fixed over the entire life of the option	IV varies by strike (skew/smile) and by maturity — Section 12
Constant risk-free rate	r known, identical for borrowing and lending	Rates vary over time; real bid-offer spread on borrowing/lending
Continuous, costless trading	Continuous and free rehedging	Discrete hedging in practice — generates stochastic hedge P&L (see Hedging)
No dividends (base version)	Merton extension: constant \(q\), replace \(S\) with \(Se^{-qT}\)	Handled natively in this course (Sections 5, 8)
No jumps	Continuous path of \(S_t\)	Opening gaps, crises — motivates jump models (Merton, Kou)

These six assumptions are not an arbitrary list — they combine to define an idealized "complete" market, in the sense that any contingent payoff can be replicated exactly by a dynamic portfolio consisting solely of stock and cash (the central argument of Section 2). It is this market completeness, more than the accuracy of any individual assumption, that constitutes the model's true theoretical contribution: it establishes that a unique, arbitrage-free option price exists in this idealized world, even if that price remains only an approximation — sometimes excellent, sometimes crude — once assumptions are relaxed one by one in the following sections and in the courses that extend this one.

// The right interview answer is not "the model is wrong"

A strong candidate never simply says that BS is unrealistic — they name the precise assumption being violated, the quantitative impact on the price (underpricing of OTMs, poor gamma risk management...), and the standard extension that corrects that specific point (Dupire, Heston, SABR, jump models — Section 13). It is this structure — name, quantify, correct — and not the mere observation that the model is imperfect, that distinguishes a strong candidate from one who recites a generic critique heard in class without ever having connected it to a concrete market mechanism.

// Section 2 — From hedging to the partial differential equation

Itô's Lemma and the Hedging Argument — Deriving the PDE

// The central idea

If one can exactly replicate the payoff of an option with a dynamic portfolio of stock and cash, the option price is determined by no-arbitrage — without ever needing to know investors' return expectations. This is the founding argument of Black, Scholes, and Merton, and the same principle underlying the entire Hedging — Derivatives Hedging course. In practice, this means we never need to guess whether the market will go up or down to price an option: it suffices to know how to construct, at every instant, the right mix of stock and cash that tracks the option price step by step. It is this idea of dynamic replication, rather than any assumption about future investor behavior, that makes the following argument entirely mechanical once its logic is grasped.

Itô's Lemma

Before building the hedging portfolio, we need a tool to manipulate functions of a stochastic process — here, the option price \(V(S,t)\) is a function of the stock price \(S\), which is itself random. In ordinary differential calculus, one would simply write \(dV=(\partial V/\partial t)dt+(\partial V/\partial S)dS\) (a first-order Taylor expansion). Itô's lemma is the correct version of this formula when \(S\) follows a Brownian motion: the Taylor expansion must be pushed to second order in \(dS\), because a term one would normally neglect — \((dS)^2\) — does not tend to zero fast enough to be ignored.

If dS = μS dt + σS dW and V(S,t) is twice differentiable

\[dV = \left(\frac{\partial V}{\partial t} + \mu S\frac{\partial V}{\partial S} + \frac{1}{2}\sigma^2 S^2 \frac{\partial^2 V}{\partial S^2}\right)dt + \sigma S\frac{\partial V}{\partial S}\,dW\]

The term \(\tfrac12\sigma^2S^2\,\partial^2V/\partial S^2\) is the Itô correction — absent from ordinary differential calculus, it reflects \((dW)^2=dt\): to first order, the square of a Brownian increment is not negligible, unlike a deterministic increment where \((dt)^2\) vanishes much faster than \(dt\). It is this difference in regime (a random increment of standard deviation \(\sqrt{dt}\), hence square of order \(dt\), versus a deterministic increment of order \(dt\)) that fundamentally distinguishes stochastic calculus from classical calculus, and which will play a decisive role in constructing the risk-free portfolio below.

Constructing the risk-free portfolio

The idea of Black, Scholes, and Merton is to combine the option and a carefully chosen quantity of stock so that the result no longer depends on chance — that is, the \(dW\) term disappears completely from the expression of \(d\Pi\). Once this portfolio is made "risk-free" at every instant, the simplest possible no-arbitrage argument applies: a risk-free asset can only earn the risk-free rate, otherwise a guaranteed profit with no capital is created.

Deriving the PDE via Δ-neutral hedging

Portfolio \(\Pi = V - \Delta\cdot S\) (long option, \(\Delta\) shares sold short). \(\Delta\) is for now an arbitrary number to be determined — it is precisely the choice of \(\Delta\) that does all the work in the next step.

\(d\Pi = dV - \Delta\,dS\). Expanding \(dV\) via Itô's lemma and collecting the \(dW\) terms: the coefficient of \(dW\) in \(d\Pi\) is \(\bigl(\sigma S\,\partial V/\partial S - \Delta\sigma S\bigr)\). By choosing \(\Delta=\partial V/\partial S\), this coefficient vanishes exactly — no source of randomness remains in \(d\Pi\), the portfolio becomes instantaneously risk-free regardless of the future movement of \(S\).

A risk-free portfolio must earn the risk-free rate: \(d\Pi = r\Pi\,dt = r(V-\Delta S)\,dt\). If this were not the case — if \(\Pi\) earned more or less than \(r\) — one could borrow (or lend) at \(r\) to fund \(\Pi\) and extract a certain profit, risk-free and with no initial capital: that is exactly the definition of an arbitrage, which the market cannot sustain.

Result

Equating the two expressions for \(d\Pi\) — the one obtained via Itô's lemma (from which only the \(dt\) term remains, once the random term is eliminated in step 2) and the one imposed by no-arbitrage — yields an equation containing no randomness, only derivatives of \(V\): the Black-Scholes partial differential equation.

The Black-Scholes PDE

\[\frac{\partial V}{\partial t} + rS\frac{\partial V}{\partial S} + \frac{1}{2}\sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} = rV\]

Boundary conditions: \(V(S,T)=\max(S-K,0)\) for a call, \(\max(K-S,0)\) for a put. The Black-Scholes formula (Section 4) is the analytical solution of this PDE under these conditions — any European derivative whose payoff depends only on \(S_T\) satisfies the same equation; only the terminal condition changes. This generality is what makes the PDE so central: it unifies the pricing of all European payoffs under a single equation.

// The drift μ has disappeared

The expected return \(\mu\) of the stock appears nowhere in the PDE — it was eliminated during the construction of the Δ-neutral portfolio in step 2 above. This is exactly the same principle as the risk-neutral probability in the binomial tree (see Binomial Model): the price of an option never depends on the real drift of the underlying, only on σ, r, and the contractual characteristics. This may seem counterintuitive — two traders who completely disagree on the future direction of the market, and therefore have very different subjective \(\mu\) values, must nonetheless agree on the same option price, provided they share the same estimate of σ. This is precisely because the option price does not compensate for a directional bet, but for the cost of dynamic replication itself — a purely mechanical operation, independent of anyone's convictions.

// Section 3 — The missing ingredient for solving the PDE

The Log-Normal Distribution of S_T

Section 2 established the Black-Scholes PDE, but solving it directly requires fairly heavy PDE techniques. Section 4 takes a more direct route: computing explicitly the risk-neutral expectation of the payoff, \(C=e^{-rT}\mathbb{E}^{\mathbb{Q}}[\max(S_T-K,0)]\) (a result that can be established independently, via the Feynman-Kac theorem, as the probabilistic solution of the same PDE). To evaluate this expectation, we first need the precise distribution of \(S_T\) under the risk-neutral measure \(\mathbb{Q}\) — that is the sole object of this section. Under \(\mathbb{Q}\), we replace the true drift \(\mu\) with the risk-free rate \(r\) (a direct consequence of the fact that \(\mu\) disappeared from the PDE in Section 2): \(dS_t = rS_t\,dt+\sigma S_t\,dW_t^{\mathbb{Q}}\).

Applying Itô's lemma to f(S) = ln S

We apply Itô's lemma not to \(V(S,t)\) as in Section 2, but to the function \(f(S)=\ln S\) — a choice that turns out to be particularly convenient because it transforms the multiplicative dynamics of \(S\) into additive dynamics, much simpler to integrate. \(f'(S)=1/S\), \(f''(S)=-1/S^2\). Itô's lemma gives: \(d(\ln S) = \dfrac{1}{S}dS - \dfrac{1}{2S^2}(dS)^2\).

With \((dS)^2=\sigma^2S^2\,dt\) (since \((dW)^2=dt\), exactly the same Itô correction as in Section 2): \(d(\ln S) = \dfrac1S(rS\,dt+\sigma S\,dW^{\mathbb{Q}}) - \dfrac12\sigma^2\,dt = \left(r-\dfrac{\sigma^2}{2}\right)dt+\sigma\,dW^{\mathbb{Q}}\). Notice that the \(S\) terms have cancelled — \(d(\ln S)\) no longer depends on the level of \(S\), only on constants and the Brownian motion. This is exactly why one speaks of "percentage returns" being independent of the price level.

This process is deterministic + standard Brownian, i.e. of the form \(d(\ln S) = a\,dt + b\,dW^{\mathbb{Q}}\) with \(a,b\) constant — the simplest possible case in stochastic calculus, which can be integrated term by term exactly like an ordinary function. Integrating directly from 0 to T: \(\ln S_T - \ln S_0 = \left(r-\dfrac{\sigma^2}{2}\right)T+\sigma W_T^{\mathbb{Q}}\).

Result

A standard Brownian motion evaluated at \(T\) follows a normal distribution with variance \(T\): \(W_T^{\mathbb{Q}}=\sqrt{T}Z\), \(Z\sim\mathcal{N}(0,1)\). Substituting and taking the exponential of both sides to go from \(\ln S_T\) back to \(S_T\):

Risk-neutral distribution of S_T

\[S_T = S_0\exp\!\left[\left(r-\frac{\sigma^2}{2}\right)T+\sigma\sqrt{T}\,Z\right], \qquad Z\sim\mathcal{N}(0,1)\]

The \(-\tfrac{\sigma^2}{2}\) term is the Itô correction. Without stochastic calculus, one would naively write \(S_T=S_0e^{rT+\sigma\sqrt{T}Z}\) — wrong, and incompatible with \(\mathbb{E}^{\mathbb{Q}}[S_T]=S_0e^{rT}\) (the martingale property of \(e^{-rt}S_t\) under \(\mathbb{Q}\), a sine qua non for no-arbitrage: under the risk-neutral measure, the discounted price of every traded asset must be a martingale, otherwise one could construct an arbitrage). Direct verification: \(\mathbb{E}^{\mathbb{Q}}[S_T]=S_0e^{(r-\sigma^2/2)T}\mathbb{E}[e^{\sigma\sqrt{T}Z}]=S_0e^{(r-\sigma^2/2)T}\cdot e^{\sigma^2T/2}=S_0e^{rT}\) — exact thanks to \(-\sigma^2/2\), which precisely cancels the \(e^{\sigma^2T/2}\) factor generated by the expectation of the exponential of a normal variable (\(\mathbb{E}[e^{\sigma\sqrt T Z}]=e^{\sigma^2T/2}\), a standard result of the normal moment generating function). Without this correction the model would admit a trivial arbitrage — showing that the Itô correction is not a technical detail but a condition of economic consistency.

// Section 4 — The derivation few candidates truly know

Full Derivation — From the Risk-Neutral Expectation to the Formula

The result of Section 3 allows us to attack the price calculation directly, without formally solving the PDE. The idea is simple to state but requires some technique to execute: the payoff \(\max(S_T-K,0)\) is zero when \(S_T\le K\) and equals \(S_T-K\) otherwise, so the expectation reduces to an integral over only the region where the option ends in the money. This section unrolls that calculation completely, without skipping a step, precisely because it is the derivation that most clearly separates a candidate who has understood the model from one who only memorized the final formula.

Starting point — the fundamental theorem of asset pricing

\[C = e^{-rT}\,\mathbb{E}^{\mathbb{Q}}\!\left[\max(S_T-K,0)\right]\]

This result — that today's price is the discounted expectation of the payoff under the risk-neutral measure — is a general consequence of no-arbitrage, independent of Black-Scholes; it appears identically in the binomial tree (Binomial Model) and in any replication-based pricing model. We compute this integral explicitly using \(S_T=S\,e^{(r-\sigma^2/2)T+\sigma\sqrt{T}Z}\) proved in Section 3, which turns the expectation over \(S_T\) into an explicit integral over the standard normal variable \(Z\).

Step 1 — Exercise condition

Threshold

We begin by translating the condition "the option ends in the money" (\(S_T>K\)) into a condition directly on the standard normal variable \(Z\), the sole source of randomness in the expression for \(S_T\). \(S_T>K \iff \sigma\sqrt{T}Z > \ln(K/S)-(r-\tfrac{\sigma^2}{2})T \iff Z>-d_2\), with \(d_2=\dfrac{\ln(S/K)+(r-\tfrac12\sigma^2)T}{\sigma\sqrt{T}}\) — \(d_2\) emerges naturally as the standardized exercise threshold: the larger \(d_2\), the more likely the option is to be exercised, since a less and less extreme draw of \(Z\) is needed to exceed \(-d_2\).

Step 2 — Decomposition and the K term

Decomposition

On the region \(Z>-d_2\) where the option is exercised, the payoff is simply \(S_T-K\) — the expected payoff therefore splits naturally into two pieces, the expectation of \(S_T\) (restricted to this region) minus \(K\) times the probability of this region: \(\mathbb{E}^{\mathbb{Q}}[(S_T-K)^+] = \underbrace{\mathbb{E}^{\mathbb{Q}}[S_T\mathbf{1}_{Z>-d_2}]}_{I_1} - \underbrace{K\cdot\mathbb{Q}(Z>-d_2)}_{I_2}\). The second term, \(I_2\), is immediate since it is just a probability; the first, \(I_1\), requires a bit more work because \(S_T\) itself depends on \(Z\) exponentially.

I₂

\(I_2 = K\cdot\mathbb{Q}(Z\le d_2) = K\,N(d_2)\) by symmetry of the standard normal (\(\mathbb{Q}(Z>-d_2)=\mathbb{Q}(Z\le d_2)\) since the density of \(Z\) is symmetric around 0). Discounted contribution: \(K\,e^{-rT}N(d_2)\) — exactly the second term of the final formula, already obtained at this stage.

Step 3 — Computing I₁ by completing the square

We substitute the explicit expression for \(S_T\) into \(I_1\) — the expectation becomes an ordinary integral against the normal density of \(Z\): \(I_1 = S\,e^{(r-\sigma^2/2)T}\displaystyle\int_{-d_2}^{\infty} e^{\sigma\sqrt{T}z}\dfrac{e^{-z^2/2}}{\sqrt{2\pi}}\,dz\). The challenge is that the integrand combines a linear exponential in \(z\) and a quadratic one — they must be merged into a single quadratic exponential to recognize a normal density.

This is exactly what completing the square achieves: \(\sigma\sqrt{T}z-\tfrac{z^2}{2} = -\tfrac12(z-\sigma\sqrt{T})^2+\tfrac{\sigma^2T}{2}\) — the first term is the density of a normal distribution recentered at \(\sigma\sqrt T\) rather than 0. The constant factor \(e^{\sigma^2T/2}\), which does not depend on \(z\), factors out and combines with \(e^{(r-\sigma^2/2)T}\) to give \(e^{rT}\) — recovering, in slightly different form, exactly the computation \(\mathbb{E}^{\mathbb{Q}}[S_T]=S_0e^{rT}\) verified in Section 3.

What remains is an integral of a shifted normal density — it is reduced to a standard normal CDF by the change of variable \(\tilde z=z-\sigma\sqrt{T}\), which also shifts the lower integration bound: new bound \(-d_2-\sigma\sqrt{T}=-d_1\) with \(d_1=d_2+\sigma\sqrt{T}\). It is this shift of the bound, caused by the change of variable, that makes \(d_1\) emerge as a distinct object from \(d_2\).

Result

After the change of variable, the remaining integral is simply the CDF of a standard normal: \(I_1 = S\,e^{rT}\displaystyle\int_{-d_1}^{\infty}\dfrac{e^{-\tilde z^2/2}}{\sqrt{2\pi}}\,d\tilde z = S\,e^{rT}N(d_1)\). Discounted contribution: \(S\,N(d_1)\) — the first term of the final formula.

Result — the Black-Scholes formula, derived end to end

\[\boxed{C = S\,N(d_1) - K\,e^{-rT}\,N(d_2)}\] \[d_1 = \frac{\ln(S/K)+\bigl(r+\tfrac12\sigma^2\bigr)T}{\sigma\sqrt{T}}, \qquad d_2 = d_1-\sigma\sqrt{T}\]

Combining the discounted contributions of steps 2 and 3 — \(S\,N(d_1)\) and \(-K\,e^{-rT}N(d_2)\) — we recover exactly the Black-Scholes formula, this time obtained without ever invoking a hedging argument, purely by direct expectation calculation. Origin of \(+\tfrac12\sigma^2\) in \(d_1\): completing the square (step 3). Origin of \(-\tfrac12\sigma^2\) in \(d_2\): the Itô correction of Section 3. The difference \(d_1-d_2=\sigma\sqrt T\) is therefore the sum of these two distinct contributions, appearing at completely different stages of the calculation — it is only in hindsight that one realizes they combine so simply.

// Why d₁ ≠ d₂ — a hidden change of measure in the calculation

\(N(d_2)=\mathbb{Q}(S_T>K)\) is the direct risk-neutral exercise probability — obtained as-is at step 2, without any further manipulation. \(N(d_1)\), on the other hand, emerges only after the change of variable in step 3, a shift that is not merely a computational trick: it corresponds mathematically to a change of probability measure. We move from \(\mathbb{Q}\) (numeraire = cash \(e^{rT}\), the measure under which we usually discount) to the stock measure (numeraire = \(S_t\) itself — value is measured not in discounted dollars but in units of stock). Under this new measure, scenarios where the stock is worth more are weighted more heavily, so \(S_T>K\) is always at least as likely as under \(\mathbb{Q}\), hence \(N(d_1)\ge N(d_2)\) always — an inequality we will encounter again whenever we compare Delta (Section 8) to the exercise probability.

// Section 5 — Complete formulas and interpretation

The Formulas — and What d₁, d₂ Really Mean

Section 4 derived the formula without dividends. In practice, the vast majority of stocks and indices pay dividends, and Merton's extension — replacing \(S\) everywhere with \(Se^{-qT}\), where \(q\) is the continuous dividend yield — fits directly into the same derivation: a shareholder who holds the stock receives a continuous dividend flow, which reduces by that amount the value of the "bare" stock that must be held to replicate the option. The formula below is therefore the general version, valid whether \(q\) is zero (no-dividend stock) or not.

	Formula (with continuous dividend q)
Call	\(C = S\,e^{-qT}N(d_1) - K\,e^{-rT}N(d_2)\)
Put	\(P = K\,e^{-rT}N(-d_2) - S\,e^{-qT}N(-d_1)\)
d₁, d₂	\(d_1=\dfrac{\ln(S/K)+(r-q+\tfrac12\sigma^2)T}{\sigma\sqrt T}, \quad d_2=d_1-\sigma\sqrt T\)

The formula itself is easy to memorize — it is the interpretation of \(d_1\) and \(d_2\) that distinguishes a candidate who recites it from one who understands it. They are two quantities that look like probabilities (both pass through the normal CDF \(N(\cdot)\)), but answer two very different economic questions, as the derivation in Section 4 revealed.

	N(d₂) — risk-neutral ITM probability	N(d₁) — call delta
Definition	\(N(d_2)=\mathbb{Q}(S_T>K)\)	\(N(d_1)=\mathbb{Q}^{S}(S_T>K)\), stock measure
Role	The "true" exercise probability under ℚ	The hedge ratio — number of shares in the replicating portfolio
Comparison	\(N(d_1) \ge N(d_2)\) always (since \(\sigma\sqrt T \ge 0\)) — equality only if \(\sigma=0\) or \(T=0\)

This distinction between the two quantities allows the entire formula to be read as a natural economic decomposition, rather than an arbitrary algebraic combination of four terms:

Economic decomposition of C = Se⁻qᵀN(d₁) − Ke⁻ʳᵀN(d₂)

Term 1

\(S\,e^{-qT}N(d_1)\) = present value of the asset received if exercised, weighted by its (stock-measure) probability of exercise. This term answers the question: "on average, under the stock measure, how much is the stock I will receive if I exercise worth today?"

Term 2

\(K\,e^{-rT}N(d_2)\) = present value of the strike paid, weighted by its (risk-neutral) probability of exercise. This term answers the symmetric question: "how much is the price I will have to pay if I exercise worth today, in risk-neutral expectation?"

∴

Call = [expected asset received | ITM]×P(ITM) − [expected price paid | ITM]×P(ITM) = discounted risk-neutral expectation of the payoff, exactly the starting definition (Section 4). The formula is therefore nothing other than "what I receive in expectation" minus "what I pay in expectation", each term computed under the probability measure that makes it most natural to express.

// Section 6 — An identity that depends on no model

Call-Put Parity

// A relationship more fundamental than Black-Scholes itself

Call-put parity is proved by a static replication argument with no assumptions about the underlying's dynamics — no log-normality, no constant volatility, not even the existence of a closed-form formula for any individual option price. It must hold under any model consistent with no-arbitrage — Heston, Dupire, a jump model, or even a model not yet invented — making it a universal consistency test between a call price and a put price with the same K, T. It is a pure accounting relationship, not a modelling one: it says nothing about the individual price of the call or the put, only about their difference.

The proof relies on a technique very different from Sections 2 and 4: instead of constructing a continuously rebalanced portfolio (dynamic hedging), we build two static portfolios, bought once at inception and never touched, and show that they have rigorously the same payoff at maturity. This is the distinction between static and dynamic replication discussed in Section 8 of the Hedging course — here, static suffices.

Proof by static replication

Portfolio A: one European call + K zero-coupon bonds of face value 1 (worth \(Ke^{-rT}\) today, since a zero-coupon bond of face value 1 pays exactly 1 at maturity). At maturity, the portfolio is worth \(\max(S_T-K,0)+K\). If the call is exercised (\(S_T>K\)), this equals \((S_T-K)+K=S_T\); otherwise, simply \(K\). In both cases: \(\max(S_T-K,0)+K = \max(S_T,K)\).

Portfolio B: one European put + one share (with dividend reinvestment, costing \(Se^{-qT}\) today to end up with exactly 1 share at maturity). At maturity: if the put is exercised (\(S_Texactly the same result as Portfolio A, state by state, under any scenario.

The two portfolios have exactly the same payoff in every state of the world at maturity — not just on average or in expectation, but scenario by scenario. By the law of one price (see Hedging §1), two portfolios with identical payoffs in every state must have the same price today: otherwise one could buy the cheaper one, sell the more expensive one, and pocket a guaranteed risk-free profit at maturity — a pure arbitrage.

Call-Put Parity

\[C + K\,e^{-rT} = P + S\,e^{-qT} \qquad\Longleftrightarrow\qquad C - P = S\,e^{-qT} - K\,e^{-rT}\]

The first form directly equates the prices of portfolios A and B constructed above. The second, obtained by simple algebraic rearrangement, is the most useful in practice: it directly gives the expected call-put spread from observable market data (spot, strike, rate, dividend), with no need to know implied vol or any other model parameter.

Numerical verification

ATM example from this course — S=100, K=100, T=1, r=3%, σ=20%, q=0

Prices

\(C=9.4134\), \(P=6.4579\) (Section 5 / calculator Section 7)

Spread

\(C-P = 9.4134-6.4579 = \mathbf{2.9554}\)

Target

\(S-Ke^{-rT} = 100-100\,e^{-0.03} = 100-97.0446 = \mathbf{2.9554}\) — exact.

// Practical use: detecting a mispricing

If the market prices of a call and a put with the same K, T do not satisfy parity (after adjusting for transaction costs and discrete dividends), there exists an arbitrage independent of any model — no belief in Black-Scholes required to exploit it. It is the very first check an options desk runs on any new price source.

// Section 7 — Seeing the formula in action

Interactive Price Calculator

The previous four sections derived the formula step by step, term by term — it is time to see it come alive. Adjust the parameters below and watch the price update in real time, computed exactly by the formulas derived in Sections 4–5 (no approximations, no precomputed tables: every slider move reruns the full calculation of \(d_1\), \(d_2\), \(N(d_1)\) and \(N(d_2)\)). This is also an opportunity to develop purely visual intuition before tackling the Greeks in Section 8: observe for instance how the price curve approaches the intrinsic value (the dotted line) as \(T\) shrinks, or how it pulls away when \(\sigma\) increases — the "time value" of the option is nothing other than this gap between the Black-Scholes price and the payoff one would get from immediate exercise.

Option parameters

Spot S 100

Strike K 100

Maturity T (years) 1.00

Volatility σ 20%

Rate r 3.0%

Dividend q 0.0%

—

Call Price

—

Put Price

—

d₁

—

d₂

// Section 8 — Not just formulas: derivations

The Greeks — Fully Derived

The Greeks are the partial derivatives of the option price with respect to each of its parameters — the daily vocabulary of any options desk, well before anyone talks about "price" in the strict sense. A trader never asks "how much is this option worth?" without immediately asking "and if the spot moves, if vol moves, if time passes, by how much does my P&L move?" — those are exactly the questions Delta, Gamma, Vega, Theta, and Rho answer. The Greeks in Practice course covers their day-to-day use on a book; this section focuses on something else, rarely shown in detail: where these formulas come from, and why they take precisely this form.

// The key identity that simplifies all derivations

Each Greek is obtained by differentiating \(C=Se^{-qT}N(d_1)-Ke^{-rT}N(d_2)\) with respect to the relevant parameter. The apparent difficulty: \(d_1\) and \(d_2\) themselves depend on \(S\) (and on \(\sigma\), and on \(T\)), so the chain rule systematically produces extra terms of the type \(\partial d_1/\partial S\) and \(\partial d_2/\partial S\), which one would in principle need to compute and combine each time. The good news: these extra terms always simplify and vanish completely, regardless of the parameter being differentiated — thanks to one single algebraic identity, proved once and for all below, then reused without re-proving for each Greek.

Proof of the identity S·e⁻qᵀ·φ(d₁) = K·e⁻ʳᵀ·φ(d₂)

Start from the definition of \(\varphi\) (the standard normal density) applied to \(d_2\), and use \(d_2=d_1-\sigma\sqrt T\) to re-express everything in terms of \(d_1\): \(\varphi(d_2)=\varphi(d_1-\sigma\sqrt T)=\dfrac{1}{\sqrt{2\pi}}e^{-\frac12(d_1-\sigma\sqrt T)^2}\). Expanding the square in the exponent, this rewrites as \(\varphi(d_1)\cdot e^{d_1\sigma\sqrt T-\frac12\sigma^2T}\) — \(\varphi(d_2)\) is expressed as \(\varphi(d_1)\) multiplied by a correction factor that remains to be simplified.

This correction factor simplifies remarkably well: going back directly to the definition of \(d_1\), we have \(d_1\sigma\sqrt T=\ln(S/K)+(r-q+\tfrac12\sigma^2)T\) — a pure algebraic rearrangement of the \(d_1\) formula, with no approximation. Therefore \(d_1\sigma\sqrt T-\tfrac12\sigma^2T=\ln(S/K)+(r-q)T\): the \(\sigma^2\) term has vanished, leaving only a logarithm and a linear term in \(T\).

Result

Taking the exponential of this simplified term: \(\varphi(d_2)=\varphi(d_1)\cdot\dfrac{S}{K}e^{(r-q)T}\), which rearranges to \(K\,e^{-rT}\varphi(d_2)=S\,e^{-qT}\varphi(d_1)\) — verified numerically to \(10^{-9}\) on any parameter set. Intuitively, this identity says that the "probability mass" at the exercise threshold, seen from both sides of the balance sheet (asset received and price paid), is always exactly balanced — this balance is what cancels the chain-rule terms in all the Greeks that follow.

Delta — spot sensitivity

Delta answers the most immediate question a trader asks: if the underlying moves by one dollar, by how much does my option price move? It is also, as Section 2 showed, exactly the number of shares to hold to build the risk-free hedging portfolio — Delta is therefore not merely a sensitivity, it is literally an ingredient in the model's proof itself.

∂C/∂S

Differentiate \(C=Se^{-qT}N(d_1)-Ke^{-rT}N(d_2)\) term by term with respect to \(S\), remembering that \(d_1\) and \(d_2\) themselves depend on \(S\) (chain rule, applied to the two composite functions \(N(d_1(S))\) and \(N(d_2(S))\)): \(\dfrac{\partial C}{\partial S} = e^{-qT}N(d_1) + S e^{-qT}\varphi(d_1)\dfrac{\partial d_1}{\partial S} - Ke^{-rT}\varphi(d_2)\dfrac{\partial d_2}{\partial S}\). The first term is the "direct" term (S appears explicitly in front of \(N(d_1)\)); the next two are the "indirect" terms from the hidden dependence of \(d_1,d_2\) on \(S\).

Since \(d_2=d_1-\sigma\sqrt T\) and \(\sigma\sqrt T\) does not depend on \(S\) (it is constant from this derivative's point of view), both partial derivatives are rigorously equal: \(\partial d_1/\partial S=\partial d_2/\partial S=1/(S\sigma\sqrt T)\) — a direct calculation from \(d_1=[\ln(S/K)+\ldots]/(\sigma\sqrt T)\), where only the \(\ln S\) term contributes to the derivative with respect to \(S\).

Since the two derivatives are identical, the two "indirect" terms from step 1 can be factored together: \(\left[Se^{-qT}\varphi(d_1)-Ke^{-rT}\varphi(d_2)\right]\cdot\dfrac{1}{S\sigma\sqrt T}\). The bracket is exactly zero, by the identity proved just above — the two indirect terms therefore cancel against each other, leaving only the direct term from step 1. This is the general mechanism that will reappear for Vega and Theta below: chain-rule terms never cancel individually, but always in pairs, thanks to this same identity.

Delta

\[\Delta_{call}=e^{-qT}N(d_1) \qquad \Delta_{put}=-e^{-qT}N(-d_1)=e^{-qT}(N(d_1)-1)\]

The hedge ratio: buying 1 call ⟺ holding \(\Delta\) shares as a hedge, exactly the \(\Delta\) introduced in Section 2 to build the risk-free portfolio — this is not a notational coincidence, it is literally the same object. \(\Delta_{call}\in[0,e^{-qT}]\), \(\Delta_{put}\in[-e^{-qT},0]\): a call behaves, in terms of exposure, between "hold nothing" (deep OTM) and "hold the stock itself" (deep ITM), never beyond.

Gamma — convexity

Gamma

\[\Gamma = \frac{\partial\Delta}{\partial S} = \frac{e^{-qT}\varphi(d_1)}{S\sigma\sqrt T}\]

Gamma measures how fast Delta itself changes as the spot moves — it is therefore the second derivative of the price with respect to \(S\), and the direct source of the convexity that makes a long options position fundamentally different from a stock position. It is obtained by differentiating \(\Delta_{call}=e^{-qT}N(d_1)\) directly with respect to \(S\) (using \(\partial d_1/\partial S=1/(S\sigma\sqrt T)\) established above) — a one-step calculation, much shorter than Delta's, precisely because the heavy lifting (establishing the identity, computing \(\partial d_1/\partial S\)) has already been done. Gamma is identical for call and put (since \(\Delta_{put}=\Delta_{call}-e^{-qT}\), and the derivative of a constant with respect to \(S\) is zero — differentiating a second time therefore erases the difference between the two Deltas). It is maximum at the money, where uncertainty about the option's final outcome (exercised or not) is greatest, and therefore where Delta changes fastest from one day to the next.

Vega — volatility sensitivity

Vega

\[\mathcal V = \frac{\partial C}{\partial\sigma} = S\,e^{-qT}\sqrt T\,\varphi(d_1)\]

Vega answers the question that defines the entire vol trading profession: by how much does the option price change if the market revises its estimate of future volatility up (or down)? The derivation follows exactly the same approach as for Delta: \(d_1\) and \(d_2\) both depend on \(\sigma\), the chain rule produces two indirect terms in \(\partial d_1/\partial\sigma\) and \(\partial d_2/\partial\sigma\), and these two terms combine and cancel by the same identity \(Se^{-qT}\varphi(d_1)=Ke^{-rT}\varphi(d_2)\), leaving only the direct term — here, there is not even an explicit "direct" term as there was for Delta (neither \(S\) nor \(K\) literally contains \(\sigma\)), so Vega comes entirely from this cancellation of indirect terms, making it the most revealing demonstration of the power of the initial identity. Vega is identical for call and put, a consequence of parity detailed in Section 9.

Remarkable identity — Vega as a function of Gamma

\[\mathcal V = \Gamma\cdot S^2\,\sigma\,T\]

A direct consequence of the two formulas above: they share the same factor \(e^{-qT}\varphi(d_1)\); dividing one by the other immediately reveals this proportionality. Useful in trading: knowing a position's Gamma immediately gives its Vega, with no recalculation — the two Greeks are strictly proportional at fixed \(S,\sigma,T\), which explains why a trader managing Gamma is, de facto, managing a large part of Vega simultaneously (and vice versa): the two risks are never fully independent on a vanilla options book.

Theta — time decay

Theta (call)

\[\Theta_{call} = -\frac{S\,e^{-qT}\varphi(d_1)\,\sigma}{2\sqrt T} - rK e^{-rT}N(d_2) + qS e^{-qT}N(d_1)\]

Theta measures the loss of option value as time passes, with spot and volatility unchanged — the "decay" every option buyer suffers, day after day, even if the market does not move. It is obtained by differentiating \(C\) with respect to \(T\): this is the densest of the five Greek derivations, since \(T\) appears both in \(d_1,d_2\) (via \(\sigma\sqrt T\) and the \((r-q)T\) term) and directly in the two discount factors \(e^{-rT}\) and \(e^{-qT}\). The first term of the formula, involving \(\varphi(d_1)\), again comes from the simplification via the Delta identity — exactly the same pairwise cancellation mechanism as for Vega; the other two terms, however, come solely from the direct dependence of the discount factors on \(T\), with no counterpart in the Delta or Vega derivations. Theta is almost always negative for a long position: every day that passes without the underlying moving brings the option closer to its intrinsic value, stripping away a little of its time value.

Rho — rate sensitivity

Rho (call)

\[\rho_{call} = \frac{\partial C}{\partial r} = K\,T\,e^{-rT}N(d_2)\]

Rho is by far the simplest of the five Greeks to derive: \(r\) appears only in the second term of the formula (via \(e^{-rT}\) and, indirectly, in \(d_1,d_2\)), and — exactly as for Vega — the indirect terms from \(\partial d_1/\partial r\) and \(\partial d_2/\partial r\) cancel by the same identity, leaving only the direct derivative of the discount factor \(Ke^{-rT}\). Positive for a call: a higher rate increases the opportunity cost of holding the asset rather than the option (money tied up in stock does not earn the risk-free rate), making the option position relatively more attractive and pushing up its price. Often neglected for short-dated options (the \(T\) factor in front of the formula makes it mechanically small), Rho becomes significant beyond 2 years — a point that long-rate desks and long-maturity product structurers monitor closely.

Numerical verification of all formulas

Greek	Formula (S=100,K=95,T=0.75,r=3%,q=1%,σ=22%)	Finite differences
Delta	0.666179	0.666184
Gamma	0.018838	0.018837
Vega	31.0833	31.0832
Theta	−5.5640	−5.5639
Rho	41.7811	41.7814

Interactive visualization

Parameters

Greek displayed

Maturity T 1.0

Volatility σ 20%

Strike K 100

// Section 9 — The shortcut parity offers

Put Greeks and Parity Relations

// No need to re-derive from scratch

Since \(P = C - Se^{-qT}+Ke^{-rT}\) (parity, Section 6), each put Greek is obtained by differentiating this relation term by term — much faster than a full re-derivation repeating all the chain-rule work of Section 8. This is an algebraic shortcut, but above all an illustration of the power of parity: it does not merely relate the prices of the call and the put, it relates their entire sensitivity structure, without needing to go back to the explicit put formula each time.

Greek	Call	Put (derived from parity)	Relation
Delta	\(e^{-qT}N(d_1)\)	\(e^{-qT}(N(d_1)-1)\)	\(\Delta_{put}=\Delta_{call}-e^{-qT}\)
Gamma	\(\dfrac{e^{-qT}\varphi(d_1)}{S\sigma\sqrt T}\)	same formula	\(\Gamma_{put}=\Gamma_{call}\)
Vega	\(Se^{-qT}\sqrt T\varphi(d_1)\)	same formula	\(\mathcal V_{put}=\mathcal V_{call}\)
Theta	see Section 8	\(\Theta_{call}+rKe^{-rT}-qSe^{-qT}\)	different (the put "re-discounts" differently)
Rho	\(KTe^{-rT}N(d_2)\)	\(-KTe^{-rT}N(-d_2)\)	opposite sign, consistent with inverse rate sensitivity

Proof — put Delta by differentiating parity

Start from \(P=C-Se^{-qT}+Ke^{-rT}\) and differentiate each term separately with respect to \(S\): \(Se^{-qT}\) has a trivial derivative \(e^{-qT}\) (affine in \(S\)), and \(Ke^{-rT}\) is a constant with respect to \(S\), so its derivative is zero. What remains is \(\dfrac{\partial P}{\partial S}=\dfrac{\partial C}{\partial S}-e^{-qT}=\Delta_{call}-e^{-qT}\) — the put Delta is obtained by subtracting a simple constant from the call Delta, without ever having to go back through the chain rule on \(d_1,d_2\).

Substituting the known formula for \(\Delta_{call}\): \(\Delta_{call}-e^{-qT}=e^{-qT}N(d_1)-e^{-qT}=e^{-qT}(N(d_1)-1)=-e^{-qT}N(-d_1)\) (using the symmetry property of the normal distribution \(N(x)-1=-N(-x)\), which simply says the probability of exceeding \(x\) is the complement of the probability of staying below \(-x\)).

Numerical check

On the example S=100,K=95,T=0.75,r=3%,q=1%,σ=22% (same parameter set as the verification table of Section 8): put Delta by finite differences = \(-0.326345\), formula \(\Delta_{call}-e^{-qT}=-0.326349\) — consistent to numerical precision.

// Why Gamma and Vega are identical for call and put

\(Se^{-qT}\) and \(Ke^{-rT}\) are both affine in S and independent of σ — their second derivatives in S (the Gamma, if one treated them as "prices" in their own right) and their derivatives in σ (the Vega) are therefore zero: a straight line has no curvature, and a quantity that does not contain σ cannot be sensitive to σ. Since parity is a purely affine relation between \(P\) and \(C\) — i.e., \(P\) minus \(C\) is a combination of affine terms in \(S\), independent of σ — any derivative of order ≥ 2 in \(S\), or any derivative in \(\sigma\), is automatically identical on both sides of the equation. This is why a delta-neutral book combining calls and puts only needs to track Gamma and Vega once, regardless of the call/put split: whether a position is built with 100 calls or 100 puts of the same K, T, its Gamma and Vega risk profile is rigorously identical — only the net Delta differs, and in a perfectly predictable way (a constant gap of \(e^{-qT}\)).

// Section 10 — The engine of options trading

The Theta-Gamma Relationship

Sections 8 and 9 derived each Greek in isolation — as if they were five independent sensitivities. They are not: the Black-Scholes PDE, established back in Section 2, is in fact an equation that ties Theta, Gamma, Delta, and the price itself together in a single constraint. This section exploits that constraint to extract the most useful relationship in all of options trading — the one that concretely explains why buying options costs money even when one is not betting on the direction of the market.

Directly from the Black-Scholes PDE (Section 2)

\[\Theta + \frac12\sigma^2S^2\Gamma + rS\Delta - rV = 0 \quad\Longrightarrow\quad \Theta \approx -\frac12\sigma^2S^2\Gamma \;\;\text{(delta-hedged portfolio, } rV, rS\Delta \text{ small)}\]

The PDE itself — recalled here in its original form — exactly links these four quantities at every point \((S,t)\). For a properly delta-hedged portfolio (where the \(rS\Delta\) term largely offsets the financing of the stock position, and where \(rV\) remains small for short maturities or moderate rates), the two dominant remaining terms are Theta and Gamma — hence the approximation \(\Theta \approx -\tfrac12\sigma^2S^2\Gamma\). Long gamma (Γ>0) → theta is paid every day. This is not an empirical coincidence observed after the fact from market data, it is a no-arbitrage constraint directly inscribed in the PDE: it is mathematically impossible to have positive Gamma without the negative Theta that accompanies it, given a zero expected hedge P&L.

// The P&L of a delta-hedged book — the link with the Hedging course

This PDE constraint takes its full economic meaning once transposed into the daily P&L of a trader who has delta-hedged their position:

\[\text{P\&L} \approx \frac12\Gamma(\Delta S)^2+\Theta\,\Delta t = \frac12\Gamma S^2(\sigma_{realized}^2-\sigma_{impl}^2)\,\Delta t\] Long gamma: one earns \(\tfrac12\Gamma(\Delta S)^2\) regardless of the direction of the move (the term is a square, always positive — the famous "gamma scalping", profiting from movement irrespective of direction), and pays \(|\Theta|\Delta t\) every day with no movement, much like an insurance premium paid continuously. The trade is globally profitable if realized vol exceeds the implied vol paid at inception — not if the market "goes up" or "goes down", only if it moves more (or less) than what the option price had priced in. It is this shift — from a directional bet to a bet on the magnitude of movement — that fundamentally defines the vol trader's profession, as opposed to a classical directional trader. The full derivation of this formula, with Monte Carlo verification, is in Hedging §3.

Numerical example — ATM delta-hedged call (S=100,K=100,T=1,r=3%,σ=20%)

Greeks

Using the ATM example from throughout this course (Sections 6, 7). Price = 9.4134, Delta = 0.5987, Gamma = 0.01933, Theta = −0.01474/day — these four numbers suffice to reconstruct the daily P&L of the position, day by day, without ever recomputing the full option price.

Day with movement

Market rises +2% (\(\Delta S=+2\)): Gamma P&L \(\approx\tfrac12\times0.01933\times4=+0.0387\) (the trader buys back shares at a higher price than they sold them at the last rebalancing — but the positive Gamma means this catch-up always works in their favor). Theta P&L \(=-0.01474\) (the fixed cost paid regardless of scenario). Net \(\approx+0.024\) — a winning day for the long gamma position.

Quiet day

\(\Delta S=0\): Gamma P&L = 0 (nothing to harvest if the market does not move). Theta P&L \(=-0.01474\) (paid regardless). Net \(=-0.01474\) — a losing day, the scenario every option buyer dreads in a market without catalysts.

Daily break-even

By equating the Gamma gain to the Theta cost, one can calculate exactly the daily move magnitude needed for the position to break even: \(\tfrac12\Gamma(\Delta S_{break})^2=|\Theta|\,dt \Rightarrow \Delta S_{break}=\sigma_{impl}\cdot S\cdot\sqrt{dt}\) — an elegant result, since this threshold is nothing other than the "average" daily move implied by the volatility \(\sigma_{impl}\) itself. If the actual move repeatedly exceeds this threshold over the option's lifetime: the long delta-hedged gamma position is cumulatively profitable, even if individual days are losing ones.

// Section 11 — Inverting the formula

Implied Volatility

// The "price" markets actually quote

No desk quotes a price in dollars — they quote a volatility. Throughout this course, we have always used the Black-Scholes formula in its forward direction: given \(S,K,T,r,\sigma\), we deduce the price. Implied volatility inverts this logic — it starts from the observable market price and deduces which value of σ, once plugged into the formula, would exactly reproduce that price. The implied volatility \(\sigma_{IV}\) is therefore the unique value of σ such that \(C_{BS}(\sigma_{IV})\) exactly reproduces the observed market price. It exists and is unique because \(\text{Vega}=\partial C/\partial\sigma>0\) always (Section 8) — \(C_{BS}\) is strictly increasing in σ, hence bijective: every market price corresponds to one and only one implied volatility, never two, never none (as long as the price stays within the economically valid bounds established by parity and no-arbitrage).

Newton-Raphson algorithm

\[\sigma_{n+1} = \sigma_n - \frac{C_{BS}(\sigma_n)-C_{market}}{\mathcal V(\sigma_n)}\]

The Newton-Raphson idea is geometric: starting from an estimate \(\sigma_n\), one locally approximates the curve \(C_{BS}(\sigma)\) by its tangent at \(\sigma_n\) (with slope exactly the Vega derived in Section 8) and steps to the point where this tangent crosses the target price \(C_{market}\). Since the true curve \(C_{BS}(\sigma)\) is very close to a straight line over a reasonable interval around the solution, this local approximation is excellent, and convergence is quadratic — the number of correct digits roughly doubles each iteration, thanks to a Vega that is always well-defined and strictly positive, preventing the method from stalling or diverging (unlike other functions where Newton-Raphson can misbehave if the derivative vanishes or changes sign near the solution).

Verified example — recovering σ=28% from the price alone

Target

We simulate a trader who observes only a market price, without knowing the vol that generated it. S=100, K=105, T=0.5, r=2.5%, σ_true=28% → call price = 6.317721 (computed once, to fabricate the "target" the algorithm must recover without ever seeing 28% directly).

Iteration 0

Initial estimate \(\sigma_0=20\%\) — a reasonable starting point but deliberately far from the true value (28%), to illustrate convergence speed: price error = −2.237, meaning the Black-Scholes price at 20% vol is still well below the target.

Iteration 1

A single Newton-Raphson correction brings the estimate very close to the target: \(\sigma_1=28.07\%\): error = 0.0187 — already within 0.07 vol points, starting from 8 points away.

Iterations 2–3

Quadratic convergence does the rest very quickly: \(\sigma_2=28.0000\%\), error \(=5.5\times10^{-7}\); \(\sigma_3=28.0000\%\), error \(=7\times10^{-13}\) — machine-precision convergence in just 3 iterations, which explains why production pricing systems can recompute tens of thousands of implied volatilities per second with no perceptible latency.

Interactive calculator — live inversion

Enter a market price, recover the implied volatility

Market call price 10.00

Strike K 100

—

Implied σ

—

Newton-Raphson Iterations

—

Reconstructed Price

—

Residual Error

S=100, T=0.5, r=2.5%, q=0 fixed for this example.

// Implied volatility vs realized volatility

Two notions of "volatility" coexist, and they must never be confused. Realized vol \(\sqrt{\tfrac{252}{N}\sum(\ln(S_i/S_{i-1}))^2}\) measures what actually happened — it is computed after the fact from historical underlying data, and has nothing to do with option prices. Implied vol, conversely, is what the market anticipates today, embedded in the prices of listed options — it is a purely forward-looking object, extracted by the inversion demonstrated above. Historically, IV > RV on average for equity indices (a phenomenon called the variance risk premium) — option sellers are structurally compensated for bearing variance risk, much as an insurer charges a premium above the average statistical cost of claims. The IV−RV difference is, to first order, the P&L of a delta-hedged vol seller (Section 10) — it is exactly this gap that every vol desk continuously hunts.

// Section 12 — The most direct empirical proof

The Volatility Smile

All previous sections treated σ as a single parameter, valid for any strike and any maturity of the same underlying — exactly the "constant volatility" assumption stated in Section 1. Section 11 just showed how to extract, for one given option, the volatility implied by its market price. There is now only one step left to reveal the model's most visible flaw: repeat this extraction for all listed options on the same underlying, on the same date, and compare the results.

// If BS were true, this chart would be a flat horizontal line

Under BS, all options on the same underlying, regardless of strike, should display the same implied volatility — since σ is supposed to be a single parameter of the underlying, not of the option itself. This is never the case — IV varies systematically by strike (called skew or smile depending on the shape) and by maturity (the vol term structure). This is the most direct and most cited empirical refutation of the model, precisely because it requires no sophisticated statistical test to observe: simply plot IV against strike to see it with your own eyes.

	Skew (equity indices)	Smile (FX)
Shape	Downward sloping — OTM puts more expensive	U-shaped — both wings more expensive than ATM
Main cause	Post-1987 crash-phobia, leverage effect, structural flows	Symmetric fat tails, no structural directional bias

// This course intentionally stops here

The complete anatomy of the smile — professional quoting in Risk Reversal/Butterfly, SVI parametrization, Gatheral's no-arbitrage conditions, sticky strike vs sticky delta, interactive 3D Bloomberg-style surface — is the subject of a dedicated course: The Volatility Surface. Using BS with a flat σ systematically underprices or overprices options depending on the smile shape; in production, desks use BS as a quoting language (one IV per strike) while pricing with models that reproduce the smile — Dupire (local volatility) or Heston (stochastic volatility), covered in detail in their respective courses.

// Section 13 — What practitioners use instead

Model Limitations and Extensions

This course has followed the model from its assumptions (Section 1) to its most direct empirical refutation (Section 12, the smile). It remains to close the loop: for each of the initial assumptions, which standard industry model relaxes it, and what does one gain in return? This final table serves as a navigation map to the rest of the catalogue — each row corresponds to a full course on this site, dedicated to exploring one of these extensions in depth.

Violated assumption	Standard extension	What it provides
Constant volatility	Dupire (1994) — local volatility	A unique surface σ(S,t) that exactly reproduces all vanilla market prices
No vol dynamics	Heston (1993), SABR (Hagan et al. 2002)	Volatility itself is a stochastic process, correlated to the underlying
No jumps	Merton (1976), Kou	Poisson process superimposed on Brownian motion — fatter tails, natural short-term smile
Continuous costless hedging	Hedging — Leland (1985)	Volatility adjusted to integrate the real cost of discrete rebalancing
No counterparty risk	XVA — CVA/DVA/FVA	Price adjustments for counterparty default risk on OTC trades

// BS is not used because it is true

It is used because it is convenient and universal — a common quoting language (IV), instantaneous analytical Greeks, an intuition shared by the entire industry for 50 years. Every desk knows it is wrong; what matters, in interviews as in practice, is being able to articulate precisely where it is wrong and which extension corrects which flaw.

// Section 14 — Interview questions

Interview Questions

Derive the Black-Scholes PDE using a hedging argument, without invoking the risk-neutral measure.

Form \(\Pi=V-\Delta S\) with \(\Delta=\partial V/\partial S\). Itô's lemma gives \(dV=\Theta dt+\Delta dS+\tfrac12\Gamma\sigma^2S^2dt\), so \(d\Pi=dV-\Delta dS=(\Theta+\tfrac12\sigma^2S^2\Gamma)dt\) — the stochastic term cancels exactly. A risk-free portfolio must earn \(r\) by no-arbitrage: \(d\Pi=r(V-\Delta S)dt\). Equating: \(\Theta+\tfrac12\sigma^2S^2\Gamma+rS\Delta-rV=0\), the Black-Scholes PDE — obtained without ever mentioning probability.

What do N(d₁) and N(d₂) precisely mean?

N(d₂) is the risk-neutral probability that the call ends in the money: \(\mathbb{Q}(S_T>K)\). N(d₁) is the call delta — a probability under a different measure (the stock measure, where the numeraire is \(S_t\) rather than cash). N(d₁) ≥ N(d₂) always, the gap \(d_1-d_2=\sigma\sqrt T\) arising from both the Itô correction in the distribution of \(S_T\) and the completing-the-square step needed to compute \(\mathbb{E}[S_T | \text{exercise}]\).

Prove call-put parity without using the Black-Scholes formula.

Portfolio A: 1 call + K zero-coupons (worth \(Ke^{-rT}\)) → payoff at maturity \(\max(S_T-K,0)+K=\max(S_T,K)\). Portfolio B: 1 put + 1 share (worth \(Se^{-qT}\)) → payoff \(\max(K-S_T,0)+S_T=\max(S_T,K)\). Both portfolios have the same payoff in every state of the world, so the same price today by the law of one price: \(C+Ke^{-rT}=P+Se^{-qT}\). No assumption about \(S\)'s dynamics is needed — the relation holds under any coherent model.

Why are Gamma and Vega identical for a call and a put with the same K, T?

By parity, \(P=C-Se^{-qT}+Ke^{-rT}\). The terms \(Se^{-qT}\) and \(Ke^{-rT}\) are affine in \(S\) and independent of σ — their second derivative in \(S\) and their derivative in σ are zero. Therefore \(\partial^2P/\partial S^2=\partial^2C/\partial S^2\) (identical Gamma) and \(\partial P/\partial\sigma=\partial C/\partial\sigma\) (identical Vega). This is not a numerical coincidence but a direct, general consequence of the affine structure of parity.

What is the relationship between Theta and Gamma for a delta-hedged portfolio, and why?

\(\Theta\approx-\tfrac12\sigma^2S^2\Gamma\) — directly from the Black-Scholes PDE applied to a delta-neutral portfolio. Being long gamma costs theta every day; being short gamma earns it. This is not an empirical observation but a no-arbitrage constraint: the PDE itself imposes this relationship, derived from the hedging argument of Section 2.

How is the implied volatility of an option computed? Why does the method converge so fast?

By numerical inversion — no closed form exists. Newton-Raphson: \(\sigma_{n+1}=\sigma_n-(C_{BS}(\sigma_n)-\text{Market Price})/\mathcal V(\sigma_n)\). Convergence is quadratic (correct digits roughly double each iteration) because Vega, always strictly positive, guarantees that \(C_{BS}\) is a well-conditioned (increasing, no plateau) function of σ — 3 to 5 iterations typically suffice for \(10^{-8}\) precision.

Why is Vega maximum for an ATM option, and what happens as T→0?

\(\mathcal V=Se^{-qT}\sqrt T\varphi(d_1)\), maximum when \(\varphi(d_1)\) is maximum, i.e. \(d_1=0\) (ATM-forward option) — uncertainty about the option's outcome (ITM or OTM) is greatest there, so a rise in σ changes the price the most. For a deep ITM/OTM option, the outcome is nearly certain, a rise in σ changes almost nothing. As \(T\to0\): Vega \(\to0\) (no time value), but Gamma \(\to\infty\) for an ATM option (φ(d₁) concentrates) and Theta \(\to-\infty\) — the Theta-Gamma relationship remains verified at the limit.

A client asks where the volatility smile comes from and why OTM index puts are so expensive — explain without mathematics.

BS assumes "symmetric" markets — frequent small moves, rare large moves in both directions. In reality, equity markets fall faster and harder than they rise. OTM puts serve as insurance against those falls — demand is structurally strong (pension funds, portfolio managers seeking protection), natural supply is limited. Result: these options cost more than a simple log-normal model predicts, which translates into a higher implied vol at low strikes — full details in The Volatility Surface.

What is the difference between buying an option and buying volatility?

Buying an option (e.g. a call) is a combined directional and vol position — long delta (profits from a rise in the underlying) and long vega (profits from a rise in implied vol). "Buying volatility" means buying the option and then continuously delta-hedging it: the directional risk is eliminated (net delta zero), leaving only pure exposure to the gap between realized and implied vol — exactly the P&L derived in Section 10. Confusing the two is the classic beginner's mistake in options trading.

BS assumes constant vol, but every strike has its own IV. Isn't that contradictory?

Yes, conceptually — if BS were the right model, a single σ should apply to all strikes and maturities, which is never observed (the smile, Section 12). In practice, desks use BS as a parametrization tool: a different IV σ(K,T) per option, allowing consistent prices to be quoted across the entire surface and communication in vol rather than raw price. This is not an error but a conscious usage — the real smile modelling (Dupire, Heston, SABR) is used for pricing products that explicitly depend on the smile's shape (exotics, barrier products).

Prove that the Black-Scholes formula converges to intrinsic value as T→0.

As \(T\to0\) with \(S\ne K\): if \(S>K\), \(d_1,d_2\to+\infty\) (the term \(\ln(S/K)/(\sigma\sqrt T)\) dominates and diverges to \(+\infty\)), so \(N(d_1),N(d_2)\to1\) and \(C\to S-K\). If \(S

// Section 15 — Sources

References — Books and Academic Papers

This course draws on the founding texts and standard reference textbooks of quantitative finance. The citations below allow each point to be explored beyond what an online course can cover, and form the bibliographic foundation that every serious quant/trading interview candidate should have encountered at least once.

Founding papers

Reference	Title	Relevance to this course
Black, F. & Scholes, M. (1973)	The Pricing of Options and Corporate Liabilities, Journal of Political Economy, 81(3), 637–654	The founding paper — original PDE derivation via hedging argument (Section 2)
Merton, R.C. (1973)	Theory of Rational Option Pricing, Bell Journal of Economics and Management Science, 4(1), 141–183	Extension with continuous dividends (Section 5) and rigorous treatment of boundary conditions
Gatheral, J. & Jacquier, A. (2014)	Arbitrage-Free SVI Volatility Surfaces, Quantitative Finance, 14(1), 59–71	Reference on smile modelling discussed in Section 12 — see The Volatility Surface

Reference textbooks

Author	Title	What it adds
Hull, J.C.	Options, Futures, and Other Derivatives, Pearson	The universal reference textbook — Black-Scholes, Greeks, and implied volatility with a pedagogy close to this course
Shreve, S.	Stochastic Calculus for Finance II: Continuous-Time Models, Springer	Rigorous treatment of Itô's lemma and the risk-neutral measure (Sections 2–4)
Wilmott, P.	Paul Wilmott on Quantitative Finance, Wiley	Desk/trading perspective on the Greeks and the Theta-Gamma tradeoff (Sections 8–10)
Gatheral, J.	The Volatility Surface: A Practitioner's Guide, Wiley	Deep dive into the volatility smile (Section 12) — see The Volatility Surface