Replication & Self-Financing Portfolios
How do you price a derivative? Here is the deep idea of Black, Scholes, and Merton,
and it fits in one sentence: to price something, build it. Suppose you can
assemble a portfolio of just the stock and cash, and — rebalancing as prices move, but never
adding or removing a penny — steer it so that at expiry it holds exactly the option's
payoff in every possible scenario. Then the option and the portfolio are the same asset
wearing different clothes, and the
law of one price
does the rest: the option must cost today whatever the portfolio costs today, or there is a
free lunch. This is replication, and it is the engine under every
no-arbitrage price in this course — the Black–Scholes formula is, at heart, the running cost
of a construction project.
Concretely, consider a portfolio holding \Delta_t shares of the
stock S_t and a bond position worth B_t
in the risk-free account. Its value is
V_t = \Delta_t S_t + B_t.
Both holdings are allowed to change through time — we rebalance, buying and selling
shares as the price moves. The crucial discipline is self-financing: after the
initial investment, no cash is ever added or withdrawn. Every purchase of stock is funded by
selling bonds and vice versa. The portfolio's value changes only because prices change,
never because we top it up. Without that discipline, "replication" would be a cheat: anyone can
end up holding the payoff if they're allowed to wire in fresh money along the way.
Worked example: replication in a two-scenario world
Before continuous time, solve a market small enough to do by hand. Today the stock is
S_0 = 100. Tomorrow it will be worth either
S_u = 120 or S_d = 80 — just two
scenarios. The bank pays zero interest (a bond worth 1 today is
worth 1 tomorrow; a nonzero rate only sprinkles discount factors
on the same argument). We want to price a call with strike K = 100,
which pays C_u = 20 in the up scenario and
C_d = 0 in the down one.
Step 1 — set up the replicating portfolio. Buy
\Delta shares and put B in the bank
(negative B means borrowing). To replicate, the portfolio must
match the option's payoff in both scenarios — two equations, two unknowns:
\begin{aligned} 120\,\Delta + B &= 20 \quad &\text{(up)} \\ 80\,\Delta + B &= 0 \quad &\text{(down)} \end{aligned}
Step 2 — solve. Subtract the second equation from the first:
40\,\Delta = 20, so \Delta = \tfrac{1}{2}.
Substitute back: 80 \cdot \tfrac{1}{2} + B = 0, so
B = -40. The recipe: buy half a share, borrow 40.
Step 3 — check it really replicates. Up:
\tfrac{1}{2}\cdot 120 - 40 = 20 ✓. Down:
\tfrac{1}{2}\cdot 80 - 40 = 0 ✓. The portfolio pays exactly what
the call pays, whichever way the coin lands.
Step 4 — read off the price. Building the portfolio today costs
V_0 = \Delta S_0 + B = \tfrac{1}{2}\cdot 100 - 40 = 10,
so the call is worth 10. Not "about 10", not "10 if the market is
rational about probabilities" — exactly 10, enforced by arbitrage. If it traded at
12, you would sell the call (collect 12), spend 10 building the
portfolio, pocket 2 today, and at expiry hand over the portfolio's payoff — which is the
call's payoff to the penny. Risk-free profit; the mispricing cannot survive.
Step 5 — notice what's missing. At no point did we use the
probability of the up-move. Whether the stock rises with chance 90% or 10%, the
replication equations — and hence the price — are identical, because matching the payoff in
every state makes the odds of each state irrelevant. This is the great surprise of
derivative pricing, and it foreshadows why the Black–Scholes formula contains the volatility
but not the stock's expected return. In general, the share holding that matches the spread of
payoffs to the spread of prices is
\Delta = \frac{C_u - C_d}{S_u - S_d} = \frac{20 - 0}{120 - 80} = \frac{1}{2},
the discrete ancestor of the continuous-time delta \partial V/\partial S
you'll meet below.
The self-financing condition, line by line
Now let time flow continuously. We want to say precisely what "value changes only from price
moves" means as a stochastic differential. In the literature you will often see the holdings
written \varphi_t (shares) and \psi_t
(bond units), with V_t = \varphi_t S_t + \psi_t B_t and the
self-financing condition dV_t = \varphi_t\,dS_t + \psi_t\,dB_t; we
keep the equivalent \Delta_t-and-bond-value notation. Write the
bond as a risk-free account accruing continuously at rate r,
dB_t = r B_t\,dt,
which just says cash in the bank grows at the risk-free rate. Now differentiate the portfolio
value.
Step 1 — write the value and its general differential. From
V_t = \Delta_t S_t + B_t, a naïve product-rule differential would
track changes in both the holdings and the prices:
dV_t = \underbrace{\Delta_t\,dS_t + dB_t}_{\text{prices move}} \;+\; \underbrace{S_t\,d\Delta_t + (\text{rebalancing of the bond})}_{\text{holdings move}}.
Step 2 — impose self-financing. The defining requirement is that the
holdings-move bracket contributes nothing to the value: any cash freed by changing
\Delta_t is exactly absorbed by the bond, and conversely. Killing
that bracket leaves only the price-driven terms,
dV_t = \Delta_t\,dS_t + dB_t.
This is the self-financing condition in differential form: the value moves only
as the assets you already hold are repriced. To feel what the condition permits, watch one
rebalance up close. Suppose at some instant you hold 20 shares at £40 and decide to lighten up:
you sell 5 shares, receiving £200, and immediately park the £200 in bonds. Your share leg drops
by £200, your bond leg rises by £200 — the portfolio's value is unchanged at the instant of
the trade. Rebalancing pays for itself. That is all self-financing demands: trades shuffle
value between the legs, they never create or destroy it.
Step 3 — substitute the bond dynamics. Using
dB_t = r B_t\,dt,
dV_t = \Delta_t\,dS_t + r B_t\,dt.
The first term is your exposure to the stock — \Delta_t shares times
the price move dS_t — and the second is the risk-free interest your
bond holding earns. That is the complete law of motion of a self-financing stock-and-bond
portfolio. Contrast a strategy that is not self-financing: a manager who wires in
£10 of fresh cash per day has
dV_t = \Delta_t\,dS_t + r B_t\,dt + 10\,dt.
That extra 10\,dt is exogenous money, and it wrecks the pricing
logic: a topped-up portfolio can be steered to any payoff whatsoever, so its terminal
value tells you nothing about what the payoff is worth. Only when the funding term is zero
does "what it ends with" pin down "what it costs".
Step 4 — replicate, then price by no-arbitrage. Suppose we can choose the
rebalancing rule \Delta_t (and the bond position) so that the
terminal value matches a target option payoff H in every state,
V_T = H = (S_T - K)^+ \quad \text{(say, for a call).}
Then the portfolio and the option have identical payoffs, so by the law of one price they must
have identical prices at every earlier time — in particular today:
\text{option price at } t = 0 \;=\; V_0.
If the option traded for anything other than V_0, you would buy the
cheaper of {option, replicating portfolio}, sell the dearer, and harvest a riskless profit — the
contradiction is exactly the arbitrage argument from before. So the price of any replicable
derivative is the cost of the self-financing portfolio that reproduces it.
Let a portfolio hold \Delta_t shares and a bond
B_t with dB_t = r B_t\,dt, value
V_t = \Delta_t S_t + B_t. Then:
-
The portfolio is self-financing exactly when its value moves only through
prices: dV_t = \Delta_t\,dS_t + r B_t\,dt (no cash injected or
withdrawn).
-
If a self-financing portfolio replicates an option payoff,
V_T = H in every state, then no-arbitrage forces the option's
price to equal the replicating cost V_0 at every time.
-
The replication price is unique: two self-financing portfolios with the
same terminal value in every state must have the same value at every earlier time.
Which \Delta_t replicates? Intuitively, hold exactly as many shares
as the option's sensitivity to the stock,
\Delta_t = \frac{\partial V}{\partial S}(t, S_t),
the option's delta. Then a small move dS_t changes
the share leg by \Delta_t\,dS_t, matching the option's own
first-order move and cancelling its randomness — the position is instantaneously
hedged. Because delta itself drifts as S_t and time change,
the hedge must be continuously rebalanced: this is delta-hedging, and the
requirement of continuous trading is exactly what makes the Black–Scholes machinery a
continuous-time theory. Substituting \Delta = \partial V/\partial S
into the self-financing equation and matching terms with
Itô's lemma
is precisely how one derives the
Black–Scholes PDE.
Replication is a promise that a self-financing \Delta_t
exists for the payoff in hand — and that promise is not automatic. It is guaranteed by
the martingale representation theorem: in a market driven by a single Brownian
motion, every square-integrable payoff can be written as a constant plus a stochastic integral
against that Brownian motion,
H = \mathbb{E}[H] + \int_0^T \phi_t\,dW_t.
The integrand \phi_t is the hedge — read off, it tells you
how many shares to hold at each instant. A market in which every payoff is replicable this way
is called complete, and completeness is what makes the no-arbitrage price not
merely bounded but unique. The Black–Scholes market, with one stock and one Brownian
driver, is complete — which is why every European option there has a single fair price.
-
Self-financing is a constraint, not a consequence. Writing
V_t = \varphi_t S_t + \psi_t B_t is pure bookkeeping — it holds
for any pair of processes (\varphi_t, \psi_t), including
a gambler who tops the account up hourly. It does not follow that
dV_t = \varphi_t\,dS_t + \psi_t\,dB_t; the honest differential
carries the extra S_t\,d\varphi_t + B_t\,d\psi_t (plus
cross-variation) terms, and self-financing is the demand that those terms cancel.
The classic hand-waving error in Black–Scholes derivations is to differentiate the value,
silently drop the holdings-move terms, and present the self-financing equation as if the
product rule delivered it. It didn't — you imposed it. Whenever you see
dV = \varphi\,dS + \psi\,dB, read it as "…and we require the
strategy to be self-financing".
-
Replication needs enough instruments. Matching a payoff state-by-state is a
system of equations, and you need as many independent hedging assets as there are independent
sources of randomness. One Brownian driver, one traded stock: every payoff is attainable —
the market is complete, and prices are pinned uniquely. But add a second source of
randomness the stock alone cannot span — stochastic volatility, jump risk, a second
imperfectly correlated Brownian motion — and some payoffs become unhedgeable:
replication fails, no-arbitrage only brackets the price inside an interval, and choosing a
number inside that interval requires extra assumptions (a volatility risk premium, a chosen
pricing measure). In the binomial example this is vivid: two scenarios, two instruments
(stock and bond), two equations, a unique solution. A three-scenario tree with the
same two instruments is generically unsolvable — three equations, two unknowns.
Why the price is unique (and what breaks if it isn't)
The theorem above claims more than "a replicating cost is a fair price" — it claims
the price is the only one. Here is the argument, and it is pure arbitrage. Suppose two
self-financing portfolios V and
\widetilde V both replicate the same payoff,
V_T = \widetilde V_T = H in every state, but cost different amounts
today, say V_0 < \widetilde V_0. Go long the cheap one and short
the dear one. The combined position is again self-financing (a difference of self-financing
strategies is self-financing — no cash crosses the boundary of either), it banks
\widetilde V_0 - V_0 > 0 at inception, and at expiry it holds
V_T - \widetilde V_T = 0: nothing owed, nothing owned. Free money
today, zero obligation tomorrow — an arbitrage. So in an arbitrage-free market the replicating
cost is unique, and the derivative's price is not a matter of opinion, expectation, or
risk appetite: it is forced.
Notice how the two hypotheses divide the labour. Replication makes the
terminal values agree; self-financing guarantees that agreement at
T propagates backwards to every earlier time — because the
only way two self-financing value processes with equal endpoints could differ mid-flight is by
someone slipping money in or out, which the condition forbids. Drop either hypothesis and the
conclusion collapses: without replication the payoffs differ somewhere and the comparison is
meaningless; without self-financing the £10-a-day strategy "replicates" everything and prices
nothing. What remains — and it is the business of the
next page
— is to actually find the replicating strategy for a call and compute
V_0. The answer is a partial differential equation.
Replication is not just a proof device: it is the day job of every option desk. When a bank
sells you a call it does not pray the market falls — it manufactures the payoff it
just sold, running exactly the recipe on this page: hold \Delta_t
shares against the book, finance them at the money-market rate, and rebalance as the delta
drifts (in practice a few times a day, within tolerance bands, since each trade costs
commission and spread). On this view an option's premium is its manufacturing
cost — the raw materials are shares and borrowed cash, and the cost of the recipe
turns out to depend on how violently the ingredients move: the volatility. Desks even speak of
their "hedging P&L": if realised volatility comes in below what the premium assumed, the
manufacturing ran cheap and the desk keeps the difference.
The dark twin of this story is 19 October 1987. Through the mid-1980s,
"portfolio insurance" funds sold institutions a synthetic put — not a real option, but a
replication strategy for one, run in reverse: as the market falls, a put's delta says
sell more stock; as it rises, buy back. Fine for one fund. But by 1987 an estimated
$60–100 billion followed the same rule, and the rule has a feedback loop built in: prices fall
→ every insurer's model orders selling → selling pushes prices lower → the models order more
selling. On Black Monday the Dow fell 22.6% in a single day — still the worst
one-day percentage loss in its history — with the insurers' program trades a major
amplifier. The moral is not that replication is wrong; it is that replication assumes you can
trade continuously at posted prices without moving them. When half the market runs
the same self-financing strategy, that assumption devours itself.
Watch the portfolio shadow the option
Below, a random stock path S_t (geometric Brownian motion) runs from
0 to maturity T. A delta-hedged
self-financing portfolio value V_t is rebalanced along the way so that
at T it lands exactly on the call payoff
(S_T - K)^+ — marked on the right against the dashed strike
K. The portfolio never receives a cash injection; it just rides the
price.
Three things to look for. First, V_0 starts above the
intrinsic value (S_0 - K)^+ — that gap is the option's time value,
the cost of the hedging campaign still to come. Second, V_t wiggles
with the stock but more gently: the portfolio holds only
\Delta_t < 1 shares, so it feels a damped copy of every stock
move — that is the hedge working. Third, press Refresh a few times and watch
the endings: on paths that finish in the money, V_T lands on
S_T - K; on paths that die below the strike, it glides down to
exactly 0. One self-financing recipe, every scenario covered —
which is precisely the property that lets V_0 serve as the price.