Risk-Neutral Measure

Motivation: why this matters in quant finance

Every derivative price a quant computes is an expectation under the risk-neutral measure

\mathbb{Q}

V_0 = e^{-rT}\,\mathbb{E}^{\mathbb{Q}}[\text{payoff}]

The superscript $\mathbb{Q}$ is not cosmetic. Replacing it with the real-world measure $\mathbb{P}$ would give the wrong answer — catastrophically wrong for anything beyond the simplest payoffs. Risk-neutral valuation is the single most important idea in derivatives pricing: it reduces the problem "what is this contingent claim worth?" to a linear expectation, at the cost of replacing the real-world measure with a fictitious one in which investors behave as if they were indifferent to risk.

Understanding $\mathbb{Q}$ answers three questions that every practitioner must be able to answer:

Why can we price by taking expectations when real investors demand a risk premium?
What does it mean for discounted asset prices to be martingales under $\mathbb{Q}$ , and why does that make pricing tractable?
How does this connect to no-arbitrage, change of measure via Girsanov's theorem, and the Black-Scholes formula?

This note defines

\mathbb{Q}

formally, explains how it emerges from no-arbitrage via the first fundamental theorem of asset pricing, walks through the one-period binomial example end-to-end, and shows the link to the Black-Scholes setting. Every later quant-finance lesson in this vault — implied volatility, the derivation of the Black-Scholes formula, martingale-i, stochastic volatility models — assumes the reader understands what

\mathbb{Q}

is.

The informal idea

Imagine a market with a single risky asset $S$ and a risk-free account growing at rate $r$ . Real-world investors, operating under $\mathbb{P}$ , demand a return premium on $S$ above the risk-free rate — that's what makes equities risky assets rather than bonds. Under $\mathbb{P}$ , the stock has drift $\mu > r$ (typically), reflecting the equity risk premium $\mu - r$ .

Now imagine re-weighting the probability of each outcome so that the stock's expected return equals

r

— exactly the risk-free rate. Under this new weighting, call it

\mathbb{Q}

, nobody is compensated for bearing the stock's risk beyond what a bond would give them. A risk-neutral investor — one who is indifferent between a certain return and a risky return with the same expectation — would value both assets identically.

The miracle of derivative pricing: in a complete, arbitrage-free market, the price of any derivative equals the expectation of its discounted payoff under exactly this

\mathbb{Q}

. Real investors aren't risk-neutral — but the replicating portfolio that reproduces the derivative's cash flows doesn't care what investors' risk preferences are. No-arbitrage alone pins down the price, and that price happens to equal

e^{-rT}\mathbb{E}^{\mathbb{Q}}[\text{payoff}]

Three key clarifications up front:

$\mathbb{Q}$ is not the "true" probability of anything. It is a pricing tool. Historical frequencies obey $\mathbb{P}$ ; betting odds in an arbitrage-free derivatives market obey $\mathbb{Q}$ .
$\mathbb{P}$ and $\mathbb{Q}$ agree on impossible events. Any event with $\mathbb{P}$ -probability zero also has $\mathbb{Q}$ -probability zero and vice versa. This is the equivalence of the two measures, and it is what makes the Radon-Nikodym derivative $d\mathbb{Q}/d\mathbb{P}$ well-defined.
Multiple $\mathbb{Q}$ 's may exist. In an incomplete market (stochastic volatility, jump models), there's no unique risk-neutral measure — a fact with deep consequences for pricing exotic and non-replicable contingent claims.

Formal definition

Setup

Fix a probability space

(\Omega, \mathcal{F}, \mathbb{P})

with a filtration

(\mathcal{F}_t)_{t \geq 0}

. A market consists of:

A risk-free numeraire (bank account) $B_t = e^{rt}$ for a constant rate $r$ (more general numeraires are allowed)
A risky asset $S_t$ adapted to the filtration

Call any

\mathcal{F}_t

-adapted trading strategy

(\phi_t, \psi_t)

(holdings in stock and bond) a portfolio. It is self-financing if value changes only due to price moves, not cash injections.

Risk-neutral measure

A probability measure

\mathbb{Q}

(\Omega, \mathcal{F})

is called a risk-neutral measure (or equivalent martingale measure, EMM) if:

Equivalence: $\mathbb{Q} \sim \mathbb{P}$ , meaning both measures have the same null events.
Discounted asset is a $\mathbb{Q}$ -martingale: the process $\widetilde{S}_t = S_t / B_t = e^{-rt}S_t$ satisfies

\mathbb{E}^{\mathbb{Q}}[\widetilde{S}_T \mid \mathcal{F}_t] = \widetilde{S}_t \quad \text{for all } 0 \leq t \leq T

Equivalently,

\widetilde{S}_t

is a martingale with respect to

(\mathcal{F}_t)

under

\mathbb{Q}

. The drift of the discounted price is zero under

\mathbb{Q}

The requirement that $\widetilde{S}_t$ — not $S_t$ itself — be a martingale is crucial: the stock grows on average at $r$ under $\mathbb{Q}$ , so discounting by the risk-free rate produces a flat (zero-drift) process.

The first fundamental theorem of asset pricing

Theorem (FTAP, loose statement). A market is free of arbitrage if and only if there exists at least one risk-neutral measure $\mathbb{Q}$ equivalent to $\mathbb{P}$ .

Proof techniques vary with the market model (hyperplane separation in discrete models, Hilbert-space duality in continuous models). The practical takeaway: no-arbitrage and the existence of $\mathbb{Q}$ are the same statement. Quantifying arbitrage opportunities is equivalent to checking that the set of risk-neutral measures is non-empty.

The second fundamental theorem of asset pricing

Theorem (second FTAP). An arbitrage-free market is complete (every contingent claim can be replicated by a self-financing portfolio) if and only if the risk-neutral measure is unique.

In a complete market, the pricing formula is unambiguous: any admissible

\mathbb{Q}

gives the same price. In incomplete markets (stochastic vol, jumps), many risk-neutral measures exist and every choice corresponds to a different pricing rule — the market's "choice" of

\mathbb{Q}

is usually calibrated from observed implied volatility surfaces.

Risk-neutral pricing formula

Let $X$ be the payoff at time $T$ of a derivative (an $\mathcal{F}_T$ -measurable random variable). In an arbitrage-free, complete market the fair price at time $t$ is:

V_t = B_t \cdot \mathbb{E}^{\mathbb{Q}}\!\left[\frac{X}{B_T} \,\Big|\, \mathcal{F}_t\right] = e^{-r(T-t)}\,\mathbb{E}^{\mathbb{Q}}[X \mid \mathcal{F}_t]

At $t = 0$ with $\mathcal{F}_0$ trivial this reduces to the canonical formula $V_0 = e^{-rT}\mathbb{E}^{\mathbb{Q}}[X]$ .

Why this is a pricing formula, not a forecast. The expectation

\mathbb{E}^{\mathbb{Q}}[X]

is not the expected payoff in any real-world sense — it is the quantity that, when discounted at the risk-free rate, equals the cost of the replicating portfolio. Replication is what assigns the price; the expectation is the bookkeeping that tracks the replication cost.

Worked examples

Example 1: one-period binomial — $\mathbb{P}$ vs $\mathbb{Q}$ explicitly

Take a single period of length $T$ . A stock worth $S_0 = 100$ today will be worth either $S_T = 110$ (up) or $S_T = 90$ (down). The continuously compounded risk-free rate is $r = 0$ (so $B_T = 1$ ) for simplicity.

Under the real-world measure

\mathbb{P}

: say

\mathbb{P}(\text{up}) = 0.7

\mathbb{P}(\text{down}) = 0.3

. Real-world expected stock price:

\mathbb{E}^{\mathbb{P}}[S_T] = 0.7 \cdot 110 + 0.3 \cdot 90 = 104

. The real-world drift is

\mu = (\ln 104 - \ln 100)/T \approx 3.9\%/T

— an equity risk premium over the risk-free rate

r = 0

Under the risk-neutral measure

\mathbb{Q}

: define

q

so that the discounted stock is a martingale:

\mathbb{E}^{\mathbb{Q}}[e^{-rT}S_T] = e^{-r \cdot 0}S_0 \quad\Longrightarrow\quad q \cdot 110 + (1-q) \cdot 90 = 100 \quad\Longrightarrow\quad q = 0.5

So $\mathbb{Q}(\text{up}) = \mathbb{Q}(\text{down}) = 0.5$ . Note $q \neq p$ : the risk-neutral probability differs from the real-world probability, and neither is "right" — they serve different purposes.

Now price a call with strike $K = 100$ , payoff $X = (S_T - 100)^+$ . Payoff is $10$ in the up state, $0$ in the down state.

V_0 = e^{-rT}\,\mathbb{E}^{\mathbb{Q}}[X] = 1 \cdot (0.5 \cdot 10 + 0.5 \cdot 0) = 5

If you instead used

\mathbb{P}

: you'd get

\mathbb{E}^{\mathbb{P}}[X] = 0.7 \cdot 10 = 7

. Using $\mathbb{P}$ overprices the call by 40%. A competitor selling at

\

and hedging with

\Delta = 0.5$ shares would lock in risk-free profit at your expense.

Why $\mathbb{P}$ gives the wrong answer. The replicating portfolio's cost depends only on the two possible stock prices and the risk-free rate. The real-world probability

p

never enters the replication calculation — it is irrelevant to the hedge. The risk-neutral probability

q = 0.5

is the unique weight that reproduces this arbitrage-free price as an expectation.

Example 2: the Black-Scholes setting

Under

\mathbb{P}

, the stock price follows geometric Brownian motion:

dS_t = \mu S_t\,dt + \sigma S_t\,dW_t^{\mathbb{P}}

Under the risk-neutral measure $\mathbb{Q}$ , the same stock has drift $r$ instead of $\mu$ :

dS_t = r S_t\,dt + \sigma S_t\,dW_t^{\mathbb{Q}}

Notice the volatility

\sigma

is unchanged between

\mathbb{P}

and

\mathbb{Q}

— a direct consequence of Girsanov's theorem, which absorbs the drift difference into the Brownian motion's shift:

dW_t^{\mathbb{Q}} = dW_t^{\mathbb{P}} + \theta\,dt, \qquad \theta = \frac{\mu - r}{\sigma}

The quantity

\theta

is the market price of risk (Sharpe-like ratio). Changing measure removes it. This is why the Black-Scholes formula does not involve

\mu

— the real-world drift is irrelevant to the option's no-arbitrage price.

Under

\mathbb{Q}

S_T = S_0 \exp\bigl((r - \tfrac{1}{2}\sigma^2)T + \sigma W_T^{\mathbb{Q}}\bigr)

is log-normal with parameters

(\ln S_0 + (r - \tfrac{1}{2}\sigma^2)T,\, \sigma^2 T)

. Plugging this into

V_0 = e^{-rT}\mathbb{E}^{\mathbb{Q}}[(S_T - K)^+]

and evaluating the Gaussian integral yields the Black-Scholes call price.

Example 3: the Radon-Nikodym derivative in the binomial model

In Example 1, the measures

\mathbb{P}

(

p = 0.7

) and

\mathbb{Q}

(

q = 0.5

) are equivalent. The Radon-Nikodym derivative is the random variable:

Z = \frac{d\mathbb{Q}}{d\mathbb{P}} = \begin{cases} q / p = 0.5 / 0.7 = 5/7 & \text{on the up state} \\ (1 - q)/(1 - p) = 0.5 / 0.3 = 5/3 & \text{on the down state} \end{cases}

Check: $\mathbb{E}^{\mathbb{P}}[Z] = 0.7 \cdot 5/7 + 0.3 \cdot 5/3 = 0.5 + 0.5 = 1$ — as it must be, since $\mathbb{Q}$ is a probability measure. And for any payoff $X$ :

\mathbb{E}^{\mathbb{Q}}[X] = \mathbb{E}^{\mathbb{P}}[Z X]

This is the measure-change machinery in its simplest form. Girsanov's theorem is the continuous-time analogue, where

Z = \exp\bigl(-\theta W_T^{\mathbb{P}} - \tfrac{1}{2}\theta^2 T\bigr)

Common confusions and pitfalls

"The risk-neutral probability is what you should actually use to estimate anything." No.

\mathbb{Q}

is a pricing tool, not a statistical model. For historical VaR, backtesting, or estimating the actual distribution of returns, always use

\mathbb{P}

. The distinction between "forecasting" (use

\mathbb{P}

) and "pricing" (use

\mathbb{Q}

) is the cleanest single test of whether an analysis is risk-neutral or real-world.

" $\mathbb{P}$ is obsolete once you have $\mathbb{Q}$ ." Not remotely.

\mathbb{Q}

is used only for pricing contingent claims under no-arbitrage. Risk management, statistical arbitrage, portfolio construction, regulatory VaR and expected shortfall, and anything involving the actual distribution of market outcomes all run under

\mathbb{P}

. A quant moves fluently between the two.

" $\mathbb{Q}$ is the 'right' probability of default because that's what CDS spreads imply." The probability of default implied by a CDS spread is the

\mathbb{Q}

-probability, which incorporates both the real-world default probability and a risk premium compensating the seller for bearing default risk. Historical default rates give the

\mathbb{P}

-probability. The gap between the two is the credit risk premium — it is economically meaningful but does not mean either is "wrong."

"Volatility is invariant under change of measure, so historical vol equals implied vol." The instantaneous volatility

\sigma

of the process is invariant under Girsanov (the measure change affects only the drift). But "historical vol" is usually realised vol, computed from a specific sample path under

\mathbb{P}

, while implied volatility is the

\sigma

that makes the Black-Scholes formula match an observed

\mathbb{Q}

-price. Even when the model is correctly specified they disagree because of sampling error, and in mis-specified models (e.g. when true dynamics are stochastic-vol) they disagree systematically. See the implied volatility note for more.

"Every arbitrage-free market has a unique $\mathbb{Q}$ ." Only complete markets do. In stochastic-vol models, jump models, and any market where claims can't be perfectly replicated, infinitely many risk-neutral measures exist. The practitioner picks one by calibrating to observed option prices — effectively choosing which members of the family of

\mathbb{Q}

's matches the data — and prices under that calibrated

\mathbb{Q}

. See stochastic volatility models.

Where this goes next

Change of Measure: The Radon-Nikodym derivative $d\mathbb{Q}/d\mathbb{P}$ and Girsanov's theorem — the machinery that constructs $\mathbb{Q}$ from $\mathbb{P}$ in continuous time.
Derivation of the Black-Scholes Formula: The risk-neutral pricing formula $V_0 = e^{-rT}\mathbb{E}^{\mathbb{Q}}[X]$ evaluated for a European call under log-normal $\mathbb{Q}$ -dynamics.
Martingale I: The martingale property of discounted asset prices under $\mathbb{Q}$ is the object's defining feature — this note develops the full martingale framework.
Implied Volatility: The $\sigma$ that, when plugged into Black-Scholes under the risk-neutral measure, reproduces the observed market option price. Implied vol is the market's verdict on $\mathbb{Q}$ .
Binomial Tree Model: The discrete-time workhorse that makes every risk-neutral concept concrete — the $q$ computation in Example 1 extends to multi-period trees.
Stochastic Volatility Models: The canonical setting where $\mathbb{Q}$ is not unique, and calibration to the vol surface selects a pricing measure.