Reading the Market's Mind: Breeden and Litzenberger Extract Probabilities From Option Prices

Look at a screen of listed option prices. Calls at every strike from far below the current price to far above, all expiring on the same day. It looks like a price list.

Breeden and Litzenberger showed in 1978 that it is not a price list. It is a probability distribution, written in code. Every belief the market holds about where the index might be in three months, the chance of a crash, the chance of a melt-up, the fat tails, all of it is sitting there in plain sight, and you can decode it with a piece of arithmetic that a competent teenager could perform.

This is one of the most quietly useful results in all of finance, and it is used every day by central banks, regulators and trading desks.

The problem: prices are visible, beliefs are not

Suppose you want to know what the market thinks. Not what one analyst thinks, not what a survey says, but what the collective, money-on-the-line consensus is about the probability that the S&P 500 falls more than 20 percent in the next year.

There is no market where you can look this up directly. What you can see is the price of a great many options. The question Breeden and Litzenberger asked is whether those prices contain enough information to reconstruct the underlying probabilities, and the answer turned out to be a clean and complete yes.

The key idea via analogy: the butterfly that pays off in one narrow place

Start by imagining the ideal instrument for this job. It is a contract that pays exactly one dollar if the index lands in a very narrow band, say between 5000 and 5010, and pays nothing otherwise. Economists call this an Arrow-Debreu security or a state price. Its price tells you, directly, what the market is willing to pay for a dollar delivered in that particular state of the world, which is essentially the market's probability of that state (discounted for time).

If you could observe the price of one of these for every possible level of the index, you would have the whole distribution. But nobody trades them.

Here is the trick. You can build one out of ordinary call options.

Buy a call struck at 4990. Sell two calls struck at 5000. Buy a call struck at 5010. This is a standard structure called a butterfly spread, and its payoff diagram is a narrow triangular tent centred on 5000. It pays nothing if the index ends below 4990 or above 5010, and it pays a small amount, peaking in the middle, if the index ends inside the window. Narrow the strikes and the tent gets tighter, until in the limit it is exactly the "pays one dollar if we land right here" contract we wanted.

So: the price of a butterfly spread is the market's price of that outcome. Buy butterflies at every strike, and you have read off the entire distribution.

The mathematical version, which is the same thing

Written more formally, the observation is this. The price of a butterfly, as the strikes get closer together, is precisely the second derivative of the call price with respect to the strike price. Take the call prices, differentiate twice with respect to strike, and out comes the risk-neutral probability density of the underlying at expiry.

That is the entire result. Two derivatives. The market's implied probability distribution is hiding in the curvature of the call-price curve as you move across strikes.

There is a first derivative result worth knowing too: differentiate call price once with respect to strike, and you get (minus) the discounted probability that the index finishes above that strike. So the slope of the call curve gives you the cumulative distribution, and its curvature gives you the density. The option chain is the market's forecast, differentiated.

Why it mattered

• It gives you the whole distribution, not one number. Implied volatility, the usual summary of an option's price, compresses everything into a single figure and assumes a bell curve. Breeden-Litzenberger throws away that assumption and hands you the full, lumpy, skewed, fat-tailed shape that the market actually believes in. If the market is pricing a fat left tail, you will see the fat left tail.
• Central banks use it, routinely. The Bank of England, the Federal Reserve and many others publish option-implied probability distributions for interest rates, exchange rates and equity indices, derived exactly this way. When a policymaker says "markets assign a 30 percent probability to a rate cut," this is very often the machinery behind the statement.
• It underpins local volatility. Dupire's local volatility formula, the workhorse of exotic options pricing, is essentially Breeden-Litzenberger plus one more step. If you know the implied distribution at every maturity, you can back out the volatility function that would produce it. The 1994 breakthrough rests on this 1978 foundation.
• It explains the volatility smile from the other direction. A smile in implied volatility is just what a non-normal implied distribution looks like when you insist on describing it with a single volatility number. Breeden-Litzenberger says: stop insisting, and look at the distribution directly.
• It priced variance swaps. The result that a variance swap can be replicated by a strip of options across all strikes, the core of the Demeterfi-Derman-Kamal-Zou paper and of the VIX itself, is a direct application of the same "any payoff can be built from a portfolio of options" logic.

The honest limitations

• This is the risk-neutral distribution, and it is not a forecast. This is the single most important and most frequently violated caveat. The distribution you extract is the market's pricing distribution, which blends genuine beliefs with the risk premium people pay for insurance. Investors will pay well over fair odds for crash protection, so the implied distribution shows a much fatter left tail than the real-world probability of a crash. Reading it as a prediction systematically overstates disaster risk, sometimes by a factor of several.
• You need a continuum of strikes, and you do not have one. The formula wants call prices at every conceivable strike. Real markets quote a few dozen, spaced apart, and the far tails are illiquid or absent. Everything outside the traded range has to be extrapolated, and the tails, the part you usually care about most, are precisely where the data is worst.
• Differentiating twice amplifies noise viciously. Any small error in a quoted price, a stale mid, a wide spread, gets magnified enormously by taking second derivatives. Raw application to raw data can produce densities that go negative, which is nonsense. Practitioners must first smooth or fit the implied volatility curve, and the choice of smoothing method visibly changes the answer. The result is exact in theory and delicate in practice.
• Bad interpolation creates fake arbitrage. If your fitted call-price curve is not properly convex in strike, the implied density comes out negative, meaning your fit is asserting that some outcomes have negative probability. Ensuring an arbitrage-free fit is a real technical discipline of its own.
• It gives you a snapshot, not a dynamic. You learn the distribution at one expiry date. You learn nothing about the path taken to get there, which is what path-dependent options actually depend on.

The one-line takeaway

Breeden and Litzenberger showed that the market's entire implied probability distribution is encoded in the curvature of option prices across strikes, recoverable by a butterfly spread or, equivalently, by differentiating call prices twice, which is why central banks can read a probability of recession off an option screen, and why every practitioner must also remember that the distribution they are reading is priced, not predicted.