1Price an option as the cost of a copy made of the index and cash, in a one-step model: the desk's put is worth 195.53 points on a ±10% tree.
2Compute the hedge ratio and the risk-neutral probability, and show that the discounted risk-neutral average gives the same price as the copy.
3Explain why the real-world probability and the index's expected return drop out of the price, and why discounting a real-world average at the risk-free rate is wrong.
4State the model-free no-arbitrage bounds on calls and puts, and spot a tree that allows an arbitrage.
Why it matters
RiskAtoms Bank plc's desk holds 1,000 puts on the UK index and marks them at £948,965. Nobody on the desk knows whether the index will rise or fall in the next three months, and no two economists would agree on the chances. Yet banks that hold the same puts mark them close to one another. The reason is that the puts can be copied with things that trade, the index and cash, and the copy has one cost that needs no forecast. This module prices the desk's puts every way a bank does; this lesson starts with the idea under all of them.
A copy, its cost, and the probabilities that fall out
Illustrative dataRiskAtoms Bank plc's UK index puts, as booked
The trading desk of M3.2: Market risk holds 1,000 European puts on the UK index, struck at 7,600, expiring in 3 months, at £10 a point. It marks them with Black-Scholes at an index level of 8,000, one implied volatility of 16%, a rate of 4% and a dividend yield of 3.5%, all continuously compounded: £948,965 at 31 December 2025. The bank, the desk and its levels are made up and stay as booked in Levels 3 and 4. Module 5.1 prices the same puts with each new model and explains any difference.
The real market that day, for context only. The FTSE 100 closed 2025 at 9,931.38 (FTSE Russell). The Bank of England had cut Bank Rate to 3.75% on 18 December 2025, so SONIA was 3.73% and the 3-month SONIA swap rate about 3.72% (derived from MNI's meeting-dated OIS strip). None of these enters the desk's figures.
This page adds a one-step model with an up move to 8,800 and a down move to 7,200 in three months, round numbers for the arithmetic, and real-world chances that are pure assumptions. as at 31 Dec 2025
The idea from Level 3, made exact
L3.1.12: Arbitrage pricing theory priced a portfolio by building a copy of it with the same factor exposures: if the copy is cheaper, buy it and sell the original. An option can be copied the same way, and here the copy is exact, not approximate. L1.7.09: What drives an option’s price promised that the “special chances” used to price options would be explained later. They come out of the copy.
One step, two states
Strip the world down to its simplest form. The index stands at 8,000 today. In three months it will be at 8,800 or 7,200, nothing else. One of the desk's puts, struck at 7,600, then pays 0 points or 400 points. Cash earns 4% a year and the index pays a dividend yield of 3.5%, reinvested.
Now build a copyCopy (replicating portfolio): a mix of traded assets, here the index and cash, that pays exactly what the option pays in every state. of the put from two things anyone can trade: some units of the index, and some cash. We need it to pay 0 in the up state and 400 in the down state. That is two equations in two unknowns (Math Atom M-A02: Rearranging equations). The index moves 1,600 points between the states and the put's payoff moves 400 the other way, so the copy needs −0.25 units of the index at expiry: a short position. Dividends reinvested turn each unit bought today into 1.00879 units, so today the copy sells 0.2478 units short. That number is the hedge ratioHedge ratio: the units of the underlying in the copy, (Vu − Vd)/(Su − Sd) adjusted for dividends. It is the option's delta in the one-step model., the one-step version of the delta of L3.2.06: Option Greeks across a book. The cash makes up the rest: lend 2,178.11 points today.
The copy costs −0.2478 × 8,000 + 2,178.11 = 195.53 points. Whatever happens, it pays exactly what the put pays. So 195.53 is the only price for the put that leaves no free money on the table.
ExploreIf the put traded at 200 points instead
Click each stage of the trade.
↓→
↓→
↓→
Today
Sell the put for 200. Buy the copy for 195.53: sell 0.2478 units of the index short and lend 2,178.11. Keep the difference, 4.47 points, or £44,667 on the desk's 1,000 contracts.
The same price as an average
Write the copy's cost another way and a probability appears. The price equals the payoffs averaged with weight 0.50625 on the up state and 0.49375 on the down state, then discounted at the risk-free rate:
e−0.04×0.25×(0.50625×0+0.49375×400)=195.53
That weight is the risk-neutral probabilityRisk-neutral probability p*: the weight on the up state that makes the index, with its dividends, earn the risk-free rate on average. It prices; it does not forecast. p*. Under it the index is expected at 8,010.01 in three months, which is 8,000 grown at the carryCarry: what holding the index earns against cash over the period, here the rate less the dividend yield: a growth factor of e^((r − q)T)., the rate less the dividend yield: with dividends added back, the index earns exactly the risk-free rate. A risk-neutral investor would demand no more, hence the name. The average is an expectation (Math Atom M-C02: Expectation), but under made-up chances chosen so that every traded asset earns the risk-free rate.
Why the real-world drift drops out
Suppose the desk thinks the up move has a real-world probabilityReal-world probability: the chance an investor actually believes the up move has. It sets the expected return, not the price of a copyable option. of 0.6. Then the index is expected at 8,160, a return of 11.4% a year with dividends: a healthy risk premium over cash. The put's expected payoff is 160. Discount that at 4% and you get 158.41, not 195.53. Believe 0.9 instead and you get 39.60. Both are wrong, and the copy shows why: it pays the same in each state whatever the chances, so its cost cannot depend on them.
Where did the risk premium go?
When a tree breaks
The argument needs the index to beat cash in one state and lose to it in the other: d < e(r−q)T < u, which is the same as 0 < p* < 1. Draw a tree where the index only rises to 8,008 and that fails: the carry is 0.125% in three months, the up move only 0.1%. Sell the index short and lend the 8,000: at expiry the loan returns 8,080.40 and buying back the short (with its dividends) leaves 2.02 in the up state and 817.12 in the down state. Free money in both, so p* comes out at 1.0025 and the put's “price” at −0.98. Every tree in this module is built to pass this test.
Bounds that need no model
Some limits hold whatever the model, because breaking them allows an arbitrage like the one above (Merton 1973). A European put can never be worth more than the present value of its strike, K e−rT, and never less than K e−rT − S e−qT when that is positive. For the desk's put the no-arbitrage boundsNo-arbitrage bounds: limits on an option's price that hold under any model, because breaking them would allow an arbitrage (Merton 1973). are 0 and 7,524.38 points; for the call on the same strike, 405.93 and 7,930.31. They are wide: a model narrows them down to one number. The one-step prices also keep put-call parity from L1.7.08: Put-call parity: 601.46 − 195.53 = 405.93 = 8,000e−qT − 7,600e−rT.
The formal version
The one-period binomial model
Today's index S0; at time T it is Su=uS0 or Sd=dS0. The claim pays Vu or Vd. The continuously compounded rate is r and the dividend yield q, reinvested in the index. The copy holds Δ units of the index and B in cash. E∗ is the expectation under the risk-neutral probability p∗:
Sources: Cox, Ross and Rubinstein (1979), section 3, where p* = (r − d)/(u − d) with r one plus the rate, and the observation that “the probability q does not appear in the formula”; Merton (1973) for the bounds and the continuous dividend yield. The general result, that no arbitrage means prices are discounted expectations under some probability, goes back to Harrison and Kreps (1979). In this model the market is completeComplete market: one where every payoff can be copied with traded assets. The one-step tree with two states and two assets is one.: two states, two assets, so every payoff has a copy.
Try it
Start with the page's tree: a put struck at 7,600, up 10% and down 10%. Slide the real-world chance from 0.05 to 0.95 and watch which numbers move. Switch to the call and check parity. Then shrink the up move below 0.125% and see the tool flag the arbitrage.
Try itCopy an option in one step
Option
Index 8,000 today; dividend yield 3.5%, reinvested in the index.
Units of the index in the copy−0.2478
Cash in the copy (points)2,178.11
Price = cost of the copy195.53
Risk-neutral average, discounted195.53
Real-world average, discounted at r158.41
The copy pays 0.00 if the index rises and 400.00 if it falls, whatever the chances. Move the real-world chance: the price does not change, but the real-world average does, so it cannot be the price.
Worked example
The desk's put in one step
One put, 8,000 → 8,800 or 7,200 in 3 months, index points
No rule prescribes a pricing model, but the accounting and capital rules lean on this lesson's idea. Fair value under IFRS 13 is an exit price built from market inputs (L4.7.08: Fair value measurement): a price from a copy made of traded instruments is exactly that. Where the copy is inexact, because the model is uncertain or a hedge is costly, the prudent valuation rules ask for an adjustment (L4.7.09: Prudent valuation). The difference between real-world and risk-neutral probabilities returns in risk modelling: VaR engines simulate real-world moves (M6.2: Market risk modelling), and exposure simulation for XVA uses both (M6.5: Counterparty risk and XVA).
Spark Atoms
✦Spark AtomNo risk premium needed+
When the Royal Swedish Academy of Sciences announced the 1997 economics prize for Robert Merton and Myron Scholes on 14 October 1997, its press release singled out this point: in their method “it is in fact not necessary to use any risk premium when valuing an option.” Fischer Black, who built the formula with Scholes, had died in 1995.
✦Spark AtomA suggestion from William Sharpe+
The tree in this lesson was published in 1979 by John Cox, Stephen Ross and Mark Rubinstein, who thanked William Sharpe for first suggesting the discrete-time approach to them. In the same year Richard Rendleman and Brit Bartter published “Two-state option pricing”, a similar formulation found independently.
Key takeaways
An option that can be copied with traded assets is worth what the copy costs; any other price allows an arbitrage.
In one step the copy holds Δ = e^(−qT) × (Vu − Vd)/(Su − Sd) units of the index and some cash: for the desk's put on a ±10% tree, −0.2478 units and 2,178.11 points, a price of 195.53.
The same price is the payoff averaged under the risk-neutral probability p* = (e^((r − q)T) − d)/(u − d) = 0.50625 and discounted at r: under p* the index earns the risk-free rate.
The real-world chances and the index's expected return never enter. Discounting a real-world average at r gives the wrong price.
No arbitrage needs d < e^((r − q)T) < u. Model-free bounds put the desk's put between 0 and Ke^(−rT) = 7,524.38 points; a model picks one number inside.
Reaction
Check what stuck
Question 1 · objective 1
In one step the index goes from 8,000 to 8,800 or 7,000. A put struck at 7,600 pays 0 or 600. Ignoring dividends, how many units of the index does the copy hold? (A short position is negative.)
units
Question 2 · objective 2
The index can rise 10% or fall 10% in 3 months; the rate is 4% and the dividend yield 3.5%, both continuous. What is the risk-neutral probability of the up move?
Question 3 · objective 3
An economist convinces the desk that the up move is 90% likely, not 60%. What happens to the one-step price of the put?
Question 4 · objective 2
On the same tree (p* = 0.50625), a call struck at 7,600 pays 1,200 if the index rises and nothing if it falls. What is it worth today, in points?
points
Question 5 · objective 4
Someone draws a tree in which the index goes to 8,008 or 7,200 in 3 months, with r = 4% and q = 3.5%. What is wrong with it?
Glossary
Copy (replicating portfolio)
Copy (replicating portfolio): a mix of traded assets, here the index and cash, that pays exactly what the option pays in every state.
Hedge ratio
Hedge ratio: the units of the underlying in the copy, (Vu − Vd)/(Su − Sd) adjusted for dividends. It is the option's delta in the one-step model.
Arbitrage
Arbitrage: a trade that costs nothing, can never lose and may gain. Prices that allow one cannot last.
Risk-neutral probability
Risk-neutral probability p*: the weight on the up state that makes the index, with its dividends, earn the risk-free rate on average. It prices; it does not forecast.
Real-world probability
Real-world probability: the chance an investor actually believes the up move has. It sets the expected return, not the price of a copyable option.
Carry
Carry: what holding the index earns against cash over the period, here the rate less the dividend yield: a growth factor of e^((r − q)T).
No-arbitrage bounds
No-arbitrage bounds: limits on an option's price that hold under any model, because breaking them would allow an arbitrage (Merton 1973).
Complete market
Complete market: one where every payoff can be copied with traded assets. The one-step tree with two states and two assets is one.
John C. Hull, Options, Futures, and Other Derivatives, 11th edition (Pearson, 2021): the chapter on binomial trees. Intermediate.
Steven Shreve, Stochastic Calculus for Finance I: The Binomial Asset Pricing Model (Springer, 2004), chapter 1. Advanced.
Origins
John C. Cox, Stephen A. Ross and Mark Rubinstein, “Option pricing: a simplified approach”, Journal of Financial Economics 7(3), 1979, pp. 229–263.
Robert C. Merton, “Theory of rational option pricing”, Bell Journal of Economics and Management Science 4(1), 1973, pp. 141–183.
J. Michael Harrison and David M. Kreps, “Martingales and arbitrage in multiperiod securities markets”, Journal of Economic Theory 20(3), 1979, pp. 381–408.
Richard J. Rendleman and Brit J. Bartter, “Two-state option pricing”, Journal of Finance 34(5), 1979, pp. 1093–1110.
Royal Swedish Academy of Sciences, press release on the 1997 prize in economic sciences, 14 October 1997.
FTSE Russell, FTSE 100 year-end close, 31 December 2025 (reprinted by Mondo Visione).
Bank of England, Monetary Policy Summary, December 2025 (published 18 December 2025).
Bank of England Database, SONIA (series IUDSOIA), 31 December 2025: 3.7257%.
MNI, “STIR: Just under 40bp of BoE cuts priced for ’26”, 2 January 2026 (the OIS strip).
Watch this lesson
Video version
Prefer to watch? The same lesson as a 14-minute video will sit here, made from this page's script.