In the old joke, an economist walking with a friend refuses to pick up a $20 bill on the pavement: "if it were real, someone would have taken it already." The joke is on the economist — but the instinct behind it is the most productive idea in all of finance.

An arbitrage is a free lunch made precise: a self-financing trading strategy that costs nothing today, can never lose money, and makes money with positive probability. Formally, a portfolio with value V_t is an arbitrage on [0, T] if

You put in nothing, you are guaranteed to walk away no worse off, and sometimes you walk away richer — a riskless profit conjured from nowhere. The no-arbitrage principle is the assumption that no such opportunity persists: if one appeared, traders would pile in until prices adjusted and the gap closed. Notice what the axiom does not say. It does not claim markets are wise, prices right, or investors rational. It says something far more modest and far more believable: $20 bills don't lie on the pavement for long. Free lunches get eaten — usually within milliseconds.

From this one modest axiom flows essentially all of derivative pricing. Every pricing formula you will meet from here on — the forward price F = S_0e^{rT}, put–call parity, the Black–Scholes equation itself — is secretly a sentence ending the same way: "…the price must be this, or else there's an arbitrage." Learn to hear that clause and the subject snaps into focus: we never predict where prices will go; we work out where they must already be so that nobody can print money.

Three conditions, each doing real work

The definition looks terse, but each clause carries weight, and each rules out a whole family of impostors. Unpack them one at a time.

V_0 = 0 — it costs nothing. The strategy is self-financing from the first instant: every dollar spent on one leg is raised by another (typically by shorting or borrowing at the risk-free rate). This is what makes an arbitrage a money pump rather than a good investment: profit at zero stake can be run at any scale. Double every position and you double the profit, still risk-free — and that unboundedness is exactly why arbitrages cannot survive: someone will scale into the gap until it closes.

\mathbb{P}(V_T \ge 0) = 1 — it can never lose. Not "rarely loses", not "loses less than it gains on average" — never, in every state of the world the model admits. This clause disqualifies almost everything the financial press calls arbitrage: a bet that wins 99.9% of the time is not an arbitrage, it is a bet. Sell enough lottery-style insurance and your expectation is handsome right up until the hurricane. No-arbitrage arguments never touch probabilities of gain or expected returns; they use only the iron statement that a payoff is non-negative everywhere.

\mathbb{P}(V_T > 0) > 0 — it sometimes wins. Without this clause the empty strategy (do nothing, hold nothing) would qualify: zero cost, never loses. The third condition demands genuine free money in at least one state — and note how weak it is: one state of strictly positive profit, however unlikely, is enough. "Can't lose, might win" already breaks the market, because a rational trader takes an infinite amount of it.

A useful equivalent form: being paid to enter (V_0 < 0) with V_T \ge 0 always is also an arbitrage — bank the entry payment and you have rebuilt the standard form. The constructions below all arrive in this "pocket cash now, owe nothing later" shape.

The simplest arbitrage: one asset, two prices

Strip everything away and the primal example is an asset quoted at two prices at the same moment. Suppose the same share trades at \$100 in New York and \$101 in London, with no fees, no delay, and the two listings genuinely interchangeable. The trade writes itself: buy in New York, simultaneously sell in London, pocket \$1 per share. Your net position in the share is zero — one bought, one sold — so you carry no market risk whatsoever. Repeat for a million shares: a riskless \$1{,}000{,}000 at zero cost.

Turn the argument around and it becomes a statement about prices: since such a trade cannot persist, the two venues must quote the same price. That is the law of one price in its street-level form. The deeper form is about portfolios: any two baskets whatsoever that deliver the same payoff in every state of the world must agree in price — a bond plus an option here, a stock plus some borrowing there, however different they look. Let us derive that by contradiction, in the same buy-cheap-short-dear style.

Suppose two portfolios A and B have the same terminal value in every state of the world,

but trade at different prices today. Without loss of generality, say A is the cheaper one: P_A < P_B. We build a free lunch in three lines.

Step 1 — buy the cheap, short the dear. Today, go long A (pay P_A) and short B (receive P_B). The net cash you collect at t = 0 is

Step 2 — bank the surplus risklessly. Set that strictly positive amount aside (in the risk-free account, where it can only grow). Your net position now costs nothing — every dollar to open it came from the short — yet you are holding a guaranteed cushion.

Step 3 — unwind at maturity. At time T you owe the short position V_T^B and you collect V_T^A from the long. Because the payoffs were assumed identical, these cancel exactly:

Step 4 — read off the contradiction. You opened the trade for zero net cost, the maturity cashflows wash out to zero, and yet you pocketed P_B - P_A > 0 at the start with certainty. That is precisely an arbitrage — V_0 = 0, V_T \ge 0 always, and a sure positive profit. No-arbitrage forbids it, so the premise P_A \ne P_B is impossible. Hence

Pause on how strange and powerful this is. At no point did we ask what the payoff V_T actually is, how likely any state is, or what anyone expects the market to do — the argument is state-by-state accounting plus one behavioural axiom. That is the template for this whole course: to price a derivative, replicate its payoff with things whose prices you know, then invoke the law of one price.

The free lunch, column by column

Professionals check an arbitrage the way an accountant checks a ledger: write every leg of the trade as a row, every date as a column, and demand a non-negative bottom line in every column and every state. Lay out the cashflows of "buy A, short B" this way — today and maturity. The two maturity legs cancel because the payoffs are equal; the two opening legs leave a strictly positive P_B - P_A in your pocket. Step through to watch the riskless profit emerge with nothing owed at T.

This tabular habit is worth acquiring now: every no-arbitrage proof in this course — the forward price below, put–call parity, the binomial hedge — is at heart one of these tables, with a zero (or better) in every future column and a plus sign today. When a claimed arbitrage confuses you, draw its table: either a column refuses to be non-negative (it wasn't an arbitrage) or the mispricing stares back from the today column.

Worked example: the currency triangle

Here is an arbitrage you can run around the world in three trades. Suppose a bank quotes these exchange rates, all live at the same instant, with no fees:

Start with €1{,}000{,}000 and chase it around the loop:

You end where you started, in the currency you started with, holding €10{,}000 more than you began — a riskless 1\% profit in the time it takes to click three times. Check the definition: the trade is self-financing (each leg spends exactly what the previous leg produced), the loop closes in euros so there is no residual currency risk, and the profit is certain the moment the three quotes are locked. A genuine arbitrage, limited in size only by how much the bank will let you trade.

Now find the condition that kills it. Chasing one euro around the loop multiplies it by the product of the three rates, so the loop is arbitrage-free exactly when that product is 1:

In our example the product is 1.25 \times 0.80 \times 1.01 = 1.01 — the tell-tale 1\%. The quote that would close the door is

i.e. exactly one euro per pound. (If the product came out below 1, you would run the loop the other way — EUR → GBP → USD → EUR — and profit again; only a product of exactly 1 is safe in both directions.) This is the law of one price wearing a different hat: the "two portfolios with the same payoff" are a euro held still and a euro sent around the triangle. In real FX markets this consistency condition pins every cross rate: quote EUR/USD and GBP/USD, and the EUR/GBP rate is no longer anyone's opinion — it is determined, to within the bid–ask spread, by the other two.

The canonical example: pricing a forward by no-arbitrage

Everything so far removed mispricings that already existed. The real power of the axiom is that it generates prices for new contracts. Take the simplest derivative of all: a forward contract, an agreement today to buy one share of a stock at time T for a price F fixed now. No money changes hands today. The stock trades at S_0, and cash can be borrowed or lent at the continuously compounded risk-free rate r. What must F be?

A first instinct says: surely F should be where we expect the stock to be at T — a forecast, a growth rate, an opinion. No-arbitrage says: none of that. You can manufacture "a share delivered at T" today, at a known cost, and the forward must match that cost or someone eats a free lunch. The recipe is called cash-and-carry: borrow S_0, buy the share now, and carry it to T, at which point you hold the share and owe S_0e^{rT} on the loan — a terminal all-in cost of S_0e^{rT}, whatever the stock does in between. Now squeeze the forward price from both sides.

If F > S_0e^{rT} — sell the dear forward, carry the cheap stock. Today: sell the forward (costs nothing), borrow S_0, buy the share. Net outlay: zero. At T: deliver the share into the forward, receive F, repay the loan S_0e^{rT}. You bank

in every state of the world — the stock price at T never entered the ledger, because you owned the very share you had promised to deliver. Zero cost, sure profit: arbitrage.

If F < S_0e^{rT} — buy the cheap forward, short the dear stock. Today: buy the forward (costs nothing), short-sell the share for S_0, invest the proceeds at r. Net outlay: zero. At T: your deposit has grown to S_0e^{rT}; pay F through the forward to receive a share, hand it back to close the short. You bank S_0e^{rT} - F > 0, again in every state. Arbitrage in the other direction (the trade desks call it reverse cash-and-carry).

Both inequalities are forbidden, so exactly one price survives:

Numbers. With S_0 = \$100, r = 5\% and T = 1 year, the only arbitrage-free forward price is F = 100e^{0.05} \approx \$105.13. If a dealer quotes \$107, the cash-and-carry table locks in 107 - 105.13 = \$1.87 per share, risklessly, at zero cost — and arbitrageurs will hit that quote until it moves.

Sit with what just happened: we priced a contract about the future without a single probability, forecast, or opinion about the stock — its expected growth appears nowhere in F = S_0e^{rT}. That eerie disappearance of expectations is the signature of every no-arbitrage price, and exactly the phenomenon the risk-neutral measure will later explain. The forward is the canonical example because the replication is so clean; Black–Scholes is the same argument with a replication that must be adjusted continuously.

See it explained