The Black–Scholes Market Model
The Black–Scholes market is a small world with sharp rules. It contains exactly
two tradeable assets — one risky stock following geometric Brownian motion, one riskless
bond — and trading in them is frictionless, continuous and unlimited. Nothing else exists: no
second stock, no dividends, no fees, no gaps, no defaults. It is unmistakably a toy. But inside
this toy world something remarkable happens: every option has exactly one fair
price — not a range, not an estimate, but a single number forced on all traders by
the rules of the world itself, whatever they each believe about the stock's future. That toy,
published in 1973, turned out to run a multi-trillion-dollar industry: the formulas priced on
the world's derivatives desks today are all descendants of this two-asset model.
Here are the rules. The bond (the money-market account) grows deterministically at the
risk-free rate r,
dB_t = r B_t\,dt, \qquad B_0 = 1 \;\Rightarrow\; B_t = e^{rt},
while the stock follows
geometric Brownian motion,
dS_t = \mu S_t\,dt + \sigma S_t\,dW_t, \qquad S_0 > 0,
with constant drift \mu and constant volatility
\sigma. Six standing assumptions pin the model down:
constant r and \sigma,
no dividends, no transaction costs,
continuous trading, and no arbitrage. Every one of them is
knowingly false — and each buys something specific. The rest of this page walks the world's
two assets, stress-tests the assumptions one by one, and computes the single quantity
(the discounted stock) on which the whole pricing theory turns.
What the stock actually does: the GBM solution
The stock SDE has an explicit solution — one of the few SDEs that does. Applying
Itô's lemma
to \ln S_t (the second-derivative term
-\tfrac{1}{2S^2}\,(dS)^2 = -\tfrac{\sigma^2}{2}\,dt is where the
correction enters) gives
S_T = S_0\,\exp\!\Big(\big(\mu - \tfrac{\sigma^2}{2}\big)T + \sigma W_T\Big).
Since W_T \sim N(0, T), the log-price is normal and
S_T itself is
lognormal
— always positive, skewed to the right. Two summary numbers matter constantly, and they are
not the same:
\mathbb{E}[S_T] = S_0\,e^{\mu T}, \qquad \operatorname{median}(S_T) = S_0\,e^{(\mu - \sigma^2/2)T}.
Why the -\sigma^2/2? The exponential is convex, so
upside wiggles help more than downside wiggles hurt: a few lucky paths that compound enormously
drag the mean up to S_0 e^{\mu T}, while the
typical path — the median — grows at only \mu - \sigma^2/2.
This gap is volatility drag. Concretely, take
\mu = 10\% and \sigma = 50\%: the mean
grows at 10\% a year, but
\mu - \sigma^2/2 = 0.10 - 0.125 = -0.025 — the typical
path loses about 2.5\% a year even while the average gains. A GBM
can have a rising mean and a falling median at the same time; more than half of all paths lose
money while the average portfolio "does great". Forgetting the drag is one of the most
expensive arithmetic slips in finance.
The two assets, side by side
The two engines of the model on one set of axes: the deterministic bond
B_t = e^{rt} climbing smoothly, and a representative stock path
S_t = S_0\,e^{(\mu - \frac12\sigma^2)t + \sigma W_t} jittering
around it. The Brownian path underneath is frozen, so the sliders reshape the same
randomness rather than rolling new dice — which makes the parameters' separate jobs visible.
Push \sigma up and the stock's wiggles amplify (and, thanks to the
-\sigma^2/2 drag, the path's overall level sags even though
\mu hasn't moved — watch for it at high
\sigma). Push \mu up and the whole
stock path tilts upward without getting any noisier. Push r up and
the bond curve steepens — the hurdle the stock must beat. Tune all three and ask: which asset
wins, and how sure are you?
The assumptions, stress-tested one by one
A model earns its keep by what its assumptions buy. Here is each Black–Scholes
assumption, what it purchases for the mathematics, and where reality breaks it:
-
Constant volatility \sigma.
Buys: a lognormal S_T with a single uncertainty
parameter, hence a closed-form price. Reality: volatility clusters (calm weeks,
violent weeks), and options of different strikes trade at different implied
volatilities — the volatility smile. Markets price deep out-of-the-money
options as if extreme moves were far more likely than the lognormal allows. This is the most
observably false assumption in the list, and the whole implied-volatility industry exists to
manage the failure.
-
Continuous paths (no jumps). Buys: the delta-hedge works — over an
instant dt the option and the hedge move together, so
risk can be cancelled exactly. Reality: prices gap on earnings, on news, on crashes;
on 19 October 1987 the S&P fell about 20% in a day, through every hedge in its path. Jump
and Lévy models add back the discontinuities Black–Scholes smooths over.
-
Constant risk-free rate r. Buys: a
deterministic bond e^{rt} and a clean discount factor.
Reality: rates move daily and have a random term structure of their own. For a
three-month equity option the error is small; for long-dated or interest-rate products it is
the whole story, and stochastic-rate models take over.
-
No dividends. Buys: the stock's only reward is price growth, so
holding a share and holding exposure to S_T are the same thing.
Reality: most large stocks pay dividends, which pull the forward price down. This
one is easy to patch (a dividend-yield term), which is why it's listed last among the
"harmless" idealisations.
-
Frictionless, continuous trading. Buys: the replication argument —
the hedge portfolio can be rebalanced at every instant for free, so the option's
payoff can be manufactured exactly. Reality: rebalancing costs money and happens in
discrete time; a real hedge is adjusted daily or hourly, not continuously, so a perfect
delta-hedge is unattainable. Transaction-cost models accept a little hedging error in
exchange for far less trading. (Note what continuous trading does not require:
nobody prices by actually trading infinitely often — the limit is a mathematical device that
identifies the fair price.)
-
No arbitrage. Buys: uniqueness. If two portfolios have identical
payoffs, they must have identical prices — this is the axiom that converts "we can replicate
the option" into "therefore the option's price is the replication cost". Of the six
assumptions this is the one closest to true: visible
arbitrage
in liquid markets is hunted to extinction within seconds, precisely because everyone is
looking for it.
The pattern to take away: each assumption removes a source of risk or friction so that
exactly one source of randomness remains — the Brownian motion
W_t — and one risky instrument to trade against it. One risk, one
hedge: that is the balance which makes the price unique.
The discounted stock, line by line
The single most important quantity in the model is the discounted stock
\tilde{S}_t = e^{-rt} S_t = \frac{S_t}{B_t},
the stock priced in units of the bond — what a share is worth after stripping out the
risk-free growth that any idle cash would have earned anyway. Its dynamics set the entire stage
for risk-neutral pricing, so we compute d\tilde S_t carefully with the
Itô
product rule.
Step 1 — name the two factors. Write
\tilde S_t = f(t)\,S_t with the deterministic discount factor
f(t) = e^{-rt}, whose ordinary derivative is
df(t) = -r\,e^{-rt}\,dt.
Step 2 — apply the product rule. For a deterministic factor times an Itô
process there is no cross-variation term (df carries no
dW, so df\,dS = 0), and the Itô product
rule reduces to the ordinary one:
d\tilde S_t = df(t)\,S_t + f(t)\,dS_t = -r\,e^{-rt}S_t\,dt + e^{-rt}\,dS_t.
Step 3 — substitute the GBM dynamics
dS_t = \mu S_t\,dt + \sigma S_t\,dW_t:
d\tilde S_t = -r\,e^{-rt}S_t\,dt + e^{-rt}\big(\mu S_t\,dt + \sigma S_t\,dW_t\big).
Step 4 — factor out e^{-rt}S_t. Every term carries
the common factor e^{-rt}S_t = \tilde S_t:
d\tilde S_t = e^{-rt}S_t\big(-r\,dt + \mu\,dt + \sigma\,dW_t\big) = \tilde S_t\big((\mu - r)\,dt + \sigma\,dW_t\big).
Step 5 — read off the discounted dynamics. Distributing
\tilde S_t gives the SDE the whole theory turns on:
d\tilde S_t = (\mu - r)\,\tilde S_t\,dt + \sigma\,\tilde S_t\,dW_t.
Discounting has done one beautifully simple thing: it replaced the drift
\mu by the excess drift
\mu - r, the stock's growth over and above the risk-free
rate, while leaving the volatility \sigma untouched. The discounted
stock is itself a geometric Brownian motion — with drift \mu - r.
Here is the punchline that the next lessons cash in. If only the drift were
zero, the discounted stock would be a
martingale
— a fair game — and prices would be plain expectations. The excess drift
\mu - r is the only thing in the way, and it is exactly what a
change of measure (Girsanov) will rotate away, turning \mu - r
into 0 and \tilde S_t into a martingale
under the risk-neutral measure. Step 5 is the doorway to
risk-neutral pricing.
The model consists of two assets and a set of standing assumptions:
-
Bond: dB_t = r B_t\,dt, hence
B_t = e^{rt} — riskless growth at rate
r.
-
Stock:
dS_t = \mu S_t\,dt + \sigma S_t\,dW_t — geometric Brownian
motion with drift \mu, volatility
\sigma, solved by
S_T = S_0 e^{(\mu - \sigma^2/2)T + \sigma W_T}.
-
Discounted stock:
d\tilde S_t = (\mu - r)\,\tilde S_t\,dt + \sigma\,\tilde S_t\,dW_t,
where \tilde S_t = e^{-rt}S_t — discounting trades
\mu for the excess drift \mu - r.
-
Assumptions: constant r and
\sigma, no dividends, no transaction costs, continuous trading,
no arbitrage.
A startling feature of the model: the option price will not depend on the stock's drift
\mu at all. Step 5 foreshadows why. Pricing is done by
replication, which only ever touches the discounted stock, and the change of measure
that makes \tilde S_t a martingale erases the excess drift
\mu - r entirely — the stock is forced to drift at the risk-free
rate under the pricing measure. What remains is the volatility
\sigma and the rate r. Two investors who
violently disagree about \mu (will the stock soar or stagnate?) must
nonetheless agree on every option price — a fact that still surprises newcomers, and the single
deepest idea in derivative pricing.
The road from model to formula
This page sets the stage; the derivation itself unfolds across the next lessons. Here is the
whole route in one view, so each step lands in context when you get there:
-
Set up the hedge portfolio. Hold the
option
and trade against it a
self-financing portfolio
of \Delta_t shares and some bond — a portfolio that never needs
fresh cash after time zero.
-
Apply Itô. Write the option value as a smooth function
V(t, S_t) and expand dV with
Itô's lemma
— the \tfrac12 \sigma^2 S^2 V_{SS}\,dt term appears here, and it
is the term that separates option pricing from ordinary calculus.
-
Cancel the risk, invoke no-arbitrage. Choosing
\Delta_t = \partial V/\partial S kills the
dW term: the hedged position is momentarily riskless, so
no-arbitrage forces it to earn exactly r. That equality is the
Black–Scholes PDE
— and, true to the vignette above, \mu has already cancelled out
of it.
-
Attach boundary conditions. The PDE alone fits every derivative on this
stock; the contract enters only through its terminal payoff — for a call,
V(T, S) = \max(S - K, 0) — plus the behaviour at
S = 0 and S \to \infty.
-
Solve. A change of variables turns the PDE into the heat equation, and out
comes the closed-form
Black–Scholes formula,
C = S_0\,\Phi(d_1) - K e^{-rT}\,\Phi(d_2). Its sensitivities —
the Greeks
— are what a trading desk actually watches.
There is a second, parallel route — price as a discounted risk-neutral
expectation and evaluate the lognormal integral — and the
Feynman–Kac theorem
is the bridge proving the PDE route and the expectation route give the same answer. The model
on this page is the ground both routes stand on.
What the model's prices look like
To feel what the machinery outputs, run the finished formula on a concrete contract — a
one-year at-the-money call, S_0 = K = 100,
r = 5\% — under two volatilities:
-
At \sigma = 20\% the model says the call is worth about
\$10.45.
-
At \sigma = 40\% — same stock price, same strike, same rate — it
says about \$18.02.
Read what just happened: nothing observable about today changed. The entire
\$7.57 difference is a statement about the width of the future —
more volatility means more chance of finishing deep in the money, while the loss on the
downside is capped at the premium. An option is convex, so it loves volatility, and in
this model \sigma is the only free parameter carrying that
information: \mu appears nowhere in either number. This is why
options traders describe their job as trading volatility — the direction of
the stock is hedged away; the width of its distribution is the product. It is also why the
market runs the formula backwards: given an option's traded price, solve for the
\sigma that reproduces it. That number — the implied
volatility — is how option prices are quoted the world over.
Black–Scholes does not say that markets follow geometric Brownian motion. It
says: if a stock followed GBM with constant \sigma in a
frictionless market, then each option on it would have this one arbitrage-free price.
It is a conditional, an implication — and attacking the model for the falsity of its
hypothesis misses how it is actually used. Three things to hold straight:
-
The constant-σ assumption fails observably, and everyone knows it. If the
model were literally true, every strike and maturity would imply the same
\sigma. They don't — the implied-volatility surface smiles and
skews, steepening dramatically after the 1987 crash. The market is openly pricing fatter
tails than the lognormal.
-
The model survives as a quoting convention, not a truth. Traders quote
options in implied vol precisely because the formula is a clean, monotone,
universally-agreed map between price and \sigma. "I'll pay 21 vol"
is meaningful from desk to desk in a way a dollar price is not. The formula became the
market's shared language — arguably a deeper victory than being right.
-
μ's absence is the deepest feature, not an oversight. A newcomer scans the
formula for the expected return and concludes something must be missing. Nothing is: the
hedge eliminates directional risk, so directional opinion cannot enter the price. If option
prices did depend on \mu, anyone with a different
forecast could be arbitraged. The model's most counterintuitive property is its most robust
one.
Fischer Black and Myron Scholes could barely get the paper published. The Journal of
Political Economy rejected it; so did the Review of Economics and Statistics —
too specialised, not enough economics. Only after Merton Miller and Eugene Fama leaned on the
JPE's editors did "The Pricing of Options and Corporate Liabilities" appear, in 1973. The
timing was uncanny: the Chicago Board Options Exchange opened the same year —
the first marketplace for standardised options — and suddenly thousands of traders needed
exactly the number the rejected paper computed. Within a few years Texas Instruments was
shipping a handheld calculator module with the Black–Scholes formula built in; Black joked
that he received no royalties. Robert Merton, working in parallel, recast the argument in
continuous time — the elegant replication-and-Itô derivation this course follows is really
the Black–Scholes–Merton model. In 1997 Scholes and Merton received the Nobel Prize
in economics for it; Fischer Black had died in 1995, and the Nobel is never awarded
posthumously — the committee took the unusual step of naming him in the citation. From twice
rejected to a Nobel, an exchange, and a pocket calculator: not bad for a toy world with two
assets.
See it explained