Black–Scholes intuition

~10 min read

🎯 Interview frequent

Why it matters

Black–Scholes–Merton is the benchmark language of options: even when the market doesn’t obey it, people quote prices as implied vols in a BS formula. Interviewers almost never want a PDE derivation from scratch; they want replication, risk-neutral pricing, what \(N(d_1)\) and \(N(d_2)\) mean, and which assumptions break in practice (see Volatility & smiles).

Core concepts

The product is still just a payoff. A European call is \(\max(S_T-K,0)\). BS is one way to compute the discounted risk-neutral expectation of that payoff when \(S\) is geometric Brownian motion with constant \(\sigma\).
Replication, not forecasting. If you can dynamically trade the stock and a money-market account to match the option payoff, the option’s price is the cost of that replicating portfolio. Otherwise there would be arbitrage. The stock’s real-world drift \(\mu\) drops out because it is already in the stock you use to hedge.
Risk-neutral pricing is the other face of the same coin. Change to a measure \(Q\) where the discounted (dividend-adjusted) asset is a martingale, i.e. the drift becomes \(r-q\). Then \(V_0 = e^{-rT} E^Q[\text{payoff}]\). You are not assuming people are risk-neutral; you are using the hedge to justify discounting at \(r\).
The formula (non-dividend European call): \[ C = S_0 N(d_1) - K e^{-rT} N(d_2), \] with \[ d_{1,2} = \frac{\ln(S_0/K) + (r \pm \tfrac12\sigma^2)T}{\sigma\sqrt{T}}, \quad d_2 = d_1 - \sigma\sqrt{T}. \] Put: \(P = K e^{-rT} N(-d_2) - S_0 N(-d_1)\). Put-call parity still holds (see Options payoff & Greeks).
What \(N(d_2)\) and \(N(d_1)\) are. \(N(d_2) = Q(S_T > K)\), the risk-neutral probability of finishing ITM. \(N(d_1)\) is the call’s delta (no dividends): the number of shares in the replicating portfolio. So the formula reads: shares \(\times\) stock minus borrowed cash \(\times\) bond.
Inputs. \(S, K, r, T, q\) are (mostly) observed. \(\sigma\) is not — it is the only free parameter, which is why the market inverts the formula for implied vol.
Assumptions that actually matter. GBM with constant \(\sigma\); continuous trading / no jumps; no transaction costs; constant \(r\); European exercise. Reality: vol is not constant (smiles), markets jump, you rehedge discretely (see Hedging & Greeks in practice).
Limits. \(\sigma \to 0\) or \(T \to 0\): price \(\to\) discounted intrinsic (forward). \(\sigma \to \infty\): call \(\to S_0 e^{-qT}\). More vol helps the long option because payoff is convex (Jensen).

Mental model

  Two equivalent stories:

  (A) Hedge              (B) Expectation
  ---------------------  -----------------------------
  sell 1 call            C = e^{-rT} E^Q [max(S_T-K,0)]
  hold  N(d1) shares
  borrow K e^{-rT} N(d2)
  rebalance as S,t move

  If (A) is possible, (B) must equal the cost of (A).
  mu never appears: it is already in the stock you hold.

  S_T under Q:  log S_T ~ N( ln S0 + (r-q-σ²/2)T , σ² T )

Think of BS as “the unique no-arbitrage price in a complete GBM world,” not as a statistical forecast of \(S_T\).

Interview questions

Why doesn’t \(\mu\) appear in the Black–Scholes formula? Doesn’t a higher-drift stock make a call more valuable? Answer: A higher \(\mu\) also makes the stock more expensive to short/long in the hedge. In the replicating argument the two effects cancel. Under \(Q\), drift is replaced by \(r-q\). A higher real-world drift does make the call more likely to finish ITM in the \(P\)-measure, but that extra chance is exactly the equity risk premium already priced in the stock.
Interpret \(N(d_1)\) and \(N(d_2)\) in one sentence each. Answer: \(N(d_2)\) is the risk-neutral probability the call finishes in the money. \(N(d_1)\) is the delta: the number of shares in the replicating portfolio (no dividends). The difference in the \(d\)’s is \(\sigma\sqrt{T}\), the lognormal width.
A 1-year ATM-forward call with \(\sigma=20\%\), \(r=0\). Roughly, is the call worth 8% of spot, 20%, or 50%? How do you think? Answer: ATM-forward, \(r=q=0\), \(C \approx S_0 \sigma \sqrt{T/2\pi} \approx 0.08\, S_0\) (the Brenner–Subrahmanyam / ATM approximation \(0.4\,S\sigma\sqrt{T}\) is the same order). So ~8% of spot, not 20% (that’s vol, not price) and not 50% (that’s a digital, or a deep-ITM delta).
List three BS assumptions that fail in equity markets, and what we do instead. Answer: (1) Constant vol → implied-vol smile/surface and local/stochastic vol models. (2) Continuous paths → jumps, fat tails, OTM puts expensive. (3) Continuous frictionless hedging → discrete rehedging, transaction costs, borrow fees. Practitioners still quote in BS implied vol.
How is BS the continuum limit of a binomial tree? Answer: Add more steps, set \(u/d\) from \(\sigma\sqrt{\Delta t}\), risk-neutral \(p^*\) from the drift \(r-q\). The binomial distribution of \(\log S_T\) converges to the BS lognormal, and the discounted expected payoff converges to the BS formula (see Binomial trees).

Watch

Introduction to the Black-Scholes formula — Khan Academy. Dissects the formula and why vol is the special input.
19. Black-Scholes Formula, Risk-neutral Valuation — MIT 18.S096 (Vasily Strela). Risk-neutral pricing from a one-step binomial through to BS.
Implied volatility — Khan Academy. Invert BS to read the vol the market is implying.