Quant Finance Foundations, Part 8. Previously: Random Walks, Brownian Motion and Itô’s Lemma.
The Black-Scholes formula is treated as either sacred or discredited depending on who is talking. Neither framing is useful. In reality it is a tool with five inputs and a set of assumptions, and knowing what each input does — and which assumptions fail — is worth far more than reproducing the derivation.
Consequently, this post does no derivation at all. The posts on binomial tree option pricing and Brownian motion already established where it comes from: replication, translated to continuous time.
The formula
For a European call:
C = S·N(d₁) − K·e−rT·N(d₂)
where d₁ = [ln(S/K) + (r + σ²/2)T] / (σ√T), and d₂ = d₁ − σ√T
N(·) is the cumulative standard normal distribution — the probability a standard normal draw falls below a given value.
The structure is more readable than it looks. It is (what you receive) minus (what you pay), each weighted by a probability-like term:
- K·e−rT is the strike, discounted to today. Part 2 material.
- N(d₂) is the risk-neutral probability the option finishes in the money. So the second term is the present value of paying the strike, weighted by the chance you actually pay it.
- N(d₁) is the option’s delta — the hedge ratio. The first term is the value of the shares you would hold in the replicating portfolio.
Compare that to the binomial result from Part 6: hold Δ shares, borrow cash. Identical architecture. Black-Scholes is the same portfolio with the step count taken to infinity.
The five Black-Scholes inputs, one at a time
| Input | Effect on a call | Effect on a put |
|---|---|---|
| Spot price | Higher | Lower |
| Strike price | Lower | Higher |
| Time to expiry | Higher | Usually higher |
| Interest rate | Higher | Lower |
| Volatility | Higher | Higher |
Spot price (S)
Higher spot, more valuable call, less valuable put. The rate of change is delta, between 0 and 1 for a call. Deep out-of-the-money, delta is near zero and the option barely responds to the underlying. Deep in-the-money, delta approaches 1 and the option tracks the stock nearly one-for-one.
Strike price (K)
Higher strike, cheaper call, more expensive put. This is the only input fixed in the contract — the other four move.
Time to expiry (T)
More time, more valuable — for calls almost always, for puts usually. The important detail is the shape of the decay. Time value does not bleed away linearly; it decays roughly with √T, which means it falls slowly at first and then collapses in the final weeks. A one-year option loses far less value in its first month than a one-month option loses in its only month.
Interest rate (r)
Higher rates raise call values and lower put values. The mechanism is financing: the replicating portfolio for a call involves borrowing, so a higher borrowing cost raises the option’s value. In most environments this is the least influential of the five, but for long-dated options it matters considerably.
Volatility (σ)
The one that dominates everything. Higher volatility raises both calls and puts, for the asymmetry reason established in Part 5 — more upside, downside already capped.
It is also the only input you cannot look up. Spot, strike, time and rate are all observable. Volatility is a forecast of the future, and every argument about an option’s price is fundamentally an argument about σ. Roughly speaking, an at-the-money option’s value scales with σ√T, so a doubling of volatility roughly doubles the premium.
A rough calibration
Some numbers to anchor the intuition. Stock at 100, strike 100, one year, rate 5%:
| Volatility | Approximate call value |
|---|---|
| 10% | 6.8 |
| 20% | 10.4 |
| 30% | 14.2 |
| 40% | 18.0 |
Nearly proportional in σ over this range, which is the σ√T scaling in action. Note also that at 30% volatility an at-the-money option costs about 14% of the stock price — options are not cheap, and the retail intuition that they are is a pricing illusion created by buying far out of the money.
The assumptions, and how they fail
The model’s honest reading is: this is the price if the following were true. None of them quite is.
Constant, known volatility. The most consequential failure. Volatility varies over time, clusters (turbulent days follow turbulent days), and — as observed in market prices — differs by strike and by maturity. That last point is the volatility smile, and it is direct market evidence against the model’s own assumption.
Continuous trading, no transaction costs. Replication requires continuous rebalancing. Real hedgers rebalance discretely and pay costs each time, producing hedging error that scales with how often they trade and how much they pay.
Lognormal returns. The model assumes returns are normally distributed in log terms. Actual returns have fat tails — extreme moves happen far more often than a normal distribution allows. The October 1987 crash was a move that Black-Scholes assigns a probability so small it should not have occurred in the age of the universe. It occurred.
No jumps. Prices are assumed continuous. Earnings announcements, central bank surprises and defaults produce genuine gaps that no continuous path can represent.
One rate for borrowing and lending, no dividends, European exercise. The easy ones. Each has a standard extension, and none is conceptually difficult.
Why it survives anyway
Given a list of failures that long, the formula’s continued centrality needs explaining. Three reasons.
It is a common language. Nobody quotes a volatility believing it is constant. They quote it because inverting the formula converts a price into a single comparable number, and that number can be compared across strikes, maturities and underlyings. The formula functions as a unit of account.
It is transparent. Five inputs, closed form, instant sensitivities. When a price looks wrong you can find out why in seconds.
It is the right baseline. Local volatility, stochastic volatility, jump-diffusion — every subsequent model is described in terms of which Black-Scholes assumption it relaxes. You cannot follow that conversation without knowing the base case.
The failure mode is not using the model. It is forgetting that you are.
Try it yourself
Build the formula and then invert it, in this order:
- Implement Black-Scholes in a spreadsheet — NORM.S.DIST handles N(·).
- Build a sensitivity table varying one input at a time.
- Confirm each behaviour described above.
- Take a real market option price and use Goal Seek to solve for the volatility that reproduces it.
- Repeat for several strikes on the same expiry and plot the results.
Then do the inversion that matters most in practice: take a real market option price and solve for the volatility that reproduces it, using Goal Seek. That number is the implied volatility, and it is what traders actually quote and trade. Do it for several strikes on the same underlying and expiry, and plot the results.
The line will not be flat. Why it bends is the subject of a later post in this series.
Frequently asked questions
Is the Black-Scholes model still used?
Yes, though mostly as a quoting convention rather than a description of reality. It converts prices into implied volatilities that can be compared across strikes and maturities.
Which Black-Scholes assumption fails worst?
Constant, known volatility. Markets price different volatilities at different strikes, which is direct evidence against the model’s own assumption.
Why is volatility the only input you cannot look up?
Spot, strike, time and rate are all observable. Volatility, by contrast, is a forecast, so every argument about an option’s price is ultimately an argument about it.
What does N(d₂) actually mean?
It is the risk-neutral probability that the option finishes in the money. Meanwhile N(d₁) is the option’s delta, the hedge ratio in the replicating portfolio.
Next in this series: the Greeks in plain English — what a hedger actually watches on a live book. After that, why the volatility smile exists.