Risk-Neutral Pricing: The Fundamental Theorem of Asset Pricing Made Concrete

The Question Risk-Neutral Pricing Answers

Suppose you are offered a contract that pays £100 if a coin lands heads and £0 if it lands tails, one second from now. What is a fair price?

The naive answer is £50 — the expected payoff. That is right if the coin is fair, if you are risk-neutral, and if there is no time value to money over the interval. Change any of those assumptions and the naive answer breaks. Change the underlying to a stock, the payoff to a call option, and the horizon to a year, and the pricing problem becomes non-trivial: real investors are risk-averse, the underlying has a real-world drift that includes a risk premium, and the discount rate is not obvious.

The remarkable result of modern derivatives theory is that, under the right assumptions, you can price a derivative by taking an expectation that ignores investor preferences. You replace the real-world probability measure (\mathbb{P}) with an artificial measure (\mathbb{Q}) under which every asset earns the risk-free rate, take the expected payoff, and discount at the risk-free rate. This is risk-neutral pricing, and it is arguably the single most important idea in quantitative finance. The modern formulation is due to Harrison & Kreps (1979) and Harrison & Pliska (1981); textbook expositions are given in Björk (2020), Baxter & Rennie (1996), Joshi (2003), and Hull (2021).

The purpose of this article is to explain why this works, when it works, and how to actually use it.

---

The No-Arbitrage Principle

The starting point is a simple economic principle. In a well-functioning market, there should be no way to make a riskless profit with zero net investment. If two portfolios generate the same cash flows in every future state of the world, they must have the same price today. Otherwise, a trader could buy the cheaper one, short the more expensive one, and pocket the difference immediately with no residual risk.

The law of one price is a corollary: identical cash flows must have identical prices. Risk-neutral pricing is the mathematical machinery that turns this economic principle into an operational recipe for computing prices.

---

A One-Period Binomial Example

The simplest setting where risk-neutral pricing appears cleanly is a one-period binomial model.

A stock trades at (S_0 = 100). Over one period, it either moves up to (uS_0 = 110) or down to (dS_0 = 90). The one-period risk-free rate is (r), so £1 held risk-free grows to (e^r) (or (1 + r) in the simple compounding version — take (r = 5%) for concreteness).

Consider a European call with strike £100. Its payoff is £10 in the up state and £0 in the down state.

The Replicating Portfolio

Look for a portfolio of (\Delta) shares and (B) pounds of the risk-free asset that reproduces the option payoff in both states:

• Up state: (110 \Delta + e^r B = 10)
• Down state: (90 \Delta + e^r B = 0)

Subtracting: (20 \Delta = 10), so (\Delta = 0.5). Substituting: (45 + e^r B = 0), so (B = -45 / e^r).

The cost today of assembling this portfolio is:

Portfolio cost = 100 * 0.5 + (-45 / e^r) = 50 - 45 e^{-r}

With (r = 5%): cost = (50 - 45 \cdot e^{-0.05} \approx 50 - 42.80 = 7.20).

The call must trade for £7.20. If it traded higher, sell the call and buy the portfolio for a riskless profit. If it traded lower, buy the call and short the portfolio. No arbitrage locks the price.

The Risk-Neutral Probability

Now do the same calculation a different way. Suppose there exists a probability (q) such that the expected discounted stock price equals today's price:

S_0 = e^{-r} * (q * u S_0 + (1 - q) * d S_0)

Solving for (q):

q = (e^r - d) / (u - d)

With (u = 1.1), (d = 0.9), (r = 0.05): (q = (e^{0.05} - 0.9) / (1.1 - 0.9) \approx (1.0513 - 0.9) / 0.2 = 0.7564).

Now compute the discounted expected payoff using this (q) instead of the true probability:

Call price = e^{-r} * (q * 10 + (1 - q) * 0) = e^{-0.05} * 0.7564 * 10 ≈ 7.20

Same answer. The trick is that (q) is chosen precisely so that the stock earns the risk-free rate in expectation, and every derivative on that stock inherits the same discount structure.

The Two Interpretations

Two facts to sit with:

1. (q) is not the true probability. The true probability of the up state is some (p) that depends on investor beliefs about the stock, and it will generally not equal 0.7564. (q) is a mathematical construct chosen to make the pricing formula work.
2. The call price does not depend on (p). Whether investors believe the stock will rise with probability 0.5 or 0.95, the option is worth £7.20. The replicating-portfolio argument uses only the up and down levels, not their probabilities.

The second point is the key insight of modern derivatives pricing. Once the underlying's dynamics and the risk-free rate are specified, the derivative's price is determined by no-arbitrage — without any reference to what the market thinks about direction.

---

From Binomial to Continuous Time

The binomial result generalises to continuous time under the fundamental theorem of asset pricing (FTAP) (Harrison & Pliska, 1981; Delbaen & Schachermayer, 1994):

A market model is arbitrage-free if and only if there exists a probability measure Q, equivalent to the physical measure P, under which every discounted asset price is a martingale.

Two key ideas are packed into that sentence.

Equivalent Measures

Two probability measures (\mathbb{P}) and (\mathbb{Q}) are equivalent if they agree on which events have zero probability. Concretely, an event that is impossible under (\mathbb{P}) must also be impossible under (\mathbb{Q}), and vice versa. Equivalent measures can assign different probabilities to the same event, but never zero versus positive. This equivalence guarantees that when you change measure to price, you have not fabricated events that could not really happen.

Martingale Property Under Q

Under (\mathbb{Q}), the discounted price of every tradeable asset is a martingale — its conditional expectation given today equals today's discounted price. Formally, if (B_t = e^{rt}) is the risk-free money-market account and (S_t) is an asset price, the discounted price (S_t / B_t) satisfies:

E^Q[ S_T / B_T | F_t ] = S_t / B_t

This is the mathematical statement that "assets earn the risk-free rate on average under (\mathbb{Q})".

The Pricing Formula

For any European derivative with payoff (H(S_T)), the no-arbitrage price at time (t) is:

V_t = e^{-r(T - t)} * E^Q[ H(S_T) | F_t ]

That single formula is the foundation of essentially all option pricing. The discount is at the risk-free rate; the expectation is taken under (\mathbb{Q}).

---

Girsanov's Theorem: How to Change Measure

For diffusion models such as geometric Brownian motion, Girsanov's theorem (Girsanov, 1960; see also Shreve, 2004, ch. 5) specifies exactly how the measure change works.

Under (\mathbb{P}), suppose (S_t) satisfies:

dS_t = mu * S_t dt + sigma * S_t dW_t^P

where (W_t^P) is a Brownian motion under (\mathbb{P}). Define:

W_t^Q = W_t^P + ((mu - r) / sigma) * t

Girsanov's theorem states that (W_t^Q) is a Brownian motion under (\mathbb{Q}), and (S_t) under (\mathbb{Q}) satisfies:

dS_t = r * S_t dt + sigma * S_t dW_t^Q

Two things stayed the same and one thing changed:

• Volatility (\sigma) is unchanged. This is why implied volatility is a legitimate market parameter — it lives in (\mathbb{Q}) but has the same value as it would in (\mathbb{P}).
• Brownian motion is still Brownian motion, just with a shifted drift.
• The drift changed from (\mu) to (r). The Sharpe ratio ((\mu - r) / \sigma) is subtracted from the Brownian drift to produce a process that no longer contains a risk premium.

This is the whole engine. Every risk premium in the physical world is stripped out by the change of measure, leaving a world where every asset earns the risk-free rate.

---

Deriving the Black-Scholes Formula

Applying the pricing formula to a European call with payoff ((S_T - K)^+), under (\mathbb{Q}) with (S_T = S_0 e^{(r - \sigma^2/2) T + \sigma W_T^Q}), and evaluating the resulting log-normal integral gives:

C = S_0 * N(d_1) - K * e^{-rT} * N(d_2)

with

d_1 = ( ln(S_0 / K) + (r + sigma^2 / 2) T ) / (sigma * sqrt(T))

d_2 = d_1 - sigma * sqrt(T)

This is the Black-Scholes formula. Every step — the SDE under (\mathbb{Q}), the log-normal distribution of (S_T), the closed-form integral — is a consequence of the risk-neutral pricing recipe.

---

Risk-Neutral Pricing by Monte Carlo

For payoffs without closed-form solutions, the pricing formula is evaluated numerically:

V_0 = e^{-rT} * E^Q[ H(S_T) ] ≈ e^{-rT} * (1/N) * sum_{i=1..N} H(S_T^{(i)})

where the paths (S_T^{(i)}) are simulated under (\mathbb{Q}).

import numpy as np


def price_call_monte_carlo(
    s0: float,
    K: float,
    r: float,
    sigma: float,
    T: float,
    n_paths: int = 200_000,
    seed: int | None = None,
) -> dict:
    """Price a European call by risk-neutral Monte Carlo under GBM."""
    rng = np.random.default_rng(seed)
    z = rng.standard_normal(n_paths)
    s_T = s0 * np.exp((r - 0.5 * sigma**2) * T + sigma * np.sqrt(T) * z)
    payoff = np.maximum(s_T - K, 0.0)
    discounted = np.exp(-r * T) * payoff
    price = discounted.mean()
    stderr = discounted.std(ddof=1) / np.sqrt(n_paths)
    return {"price": price, "stderr": stderr}


from math import log, sqrt, exp
from statistics import NormalDist


def black_scholes_call(s0, K, r, sigma, T):
    d1 = (log(s0 / K) + (r + 0.5 * sigma**2) * T) / (sigma * sqrt(T))
    d2 = d1 - sigma * sqrt(T)
    N = NormalDist().cdf
    return s0 * N(d1) - K * exp(-r * T) * N(d2)


mc = price_call_monte_carlo(s0=100, K=100, r=0.05, sigma=0.20, T=1.0, seed=1)
bs = black_scholes_call(100, 100, 0.05, 0.20, 1.0)
print(f"Monte Carlo price: {mc['price']:.4f} +/- {1.96 * mc['stderr']:.4f}")
print(f"Black-Scholes:     {bs:.4f}")

The Monte Carlo estimate should agree with the closed-form Black-Scholes price to within the reported confidence interval. If not, either the number of paths is too small or the simulator is not running under (\mathbb{Q}).

---

When Risk-Neutral Pricing Fails

The FTAP requires two economic conditions and one modelling condition.

The Market Must Be Arbitrage-Free

If arbitrage exists in the assumed model, no equivalent martingale measure (\mathbb{Q}) exists and risk-neutral pricing does not apply. Care must be taken in model specification to avoid inadvertently introducing arbitrages.

The Market Must Be Complete

A market is complete if every contingent claim can be replicated by a self-financing trading strategy in the underlying assets. Completeness is what makes (\mathbb{Q}) unique. Incomplete markets — jump-diffusion, stochastic volatility, real-world markets with transaction costs and trading restrictions — have infinitely many equivalent martingale measures. Prices are no longer determined by no-arbitrage alone; a choice of measure (or of preferences) is required.

Practically, in incomplete-market models the risk-neutral measure is chosen to match observed market prices of certain liquid instruments. In the Heston model, for example, the risk-neutral drift of the volatility process is calibrated to fit the implied volatility surface — it is not a physical quantity but a market-consistent choice.

The Underlying Must Be Tradeable

The replicating-portfolio argument requires that the underlying can actually be bought and sold. For non-tradeable underlyings (weather indices, mortality, sports outcomes) risk-neutral pricing does not apply in its pure form; some auxiliary hedging instrument or actuarial assumption is required.

---

The Q-vs-P Distinction in Practice

Risk-neutral pricing does not say the real world behaves as if under (\mathbb{Q}). It says that arbitrage-free derivative prices can be expressed as expectations under (\mathbb{Q}). The two probability worlds coexist:

• Under (\mathbb{P}), stocks earn a risk premium and expected returns exceed the risk-free rate. This is the world of portfolio construction, alpha estimation, and CAPM.
• Under (\mathbb{Q}), all assets earn the risk-free rate. This is the world of derivatives pricing, hedging, and replication.

Mixing them up is a classic mistake. Do not price options using the physical expected return of the underlying. Do not use option-implied probabilities as forecasts of future physical outcomes without a risk-premium adjustment.

---

Applications

• Vanilla option pricing — Black-Scholes for European, tree methods for American.
• Exotic option pricing — Monte Carlo simulation under (\mathbb{Q}) for barrier, Asian, lookback, and basket options.
• Interest-rate derivatives — bond pricing under the risk-neutral forward measure; swaptions under the swap-annuity measure.
• Credit derivatives — CDS pricing as expectations under a survival-adjusted (\mathbb{Q}).
• Volatility surface calibration — the model's risk-neutral parameters (e.g. Heston's mean, vol-of-vol, correlation) are chosen to match market prices, not to describe physical dynamics.
• XVA calculations — CVA, DVA, and FVA all involve expectations under specific choices of (\mathbb{Q}) that reflect counterparty risk, funding costs, and capital constraints.

---

References

• Baxter, M., & Rennie, A. (1996). Financial Calculus: An Introduction to Derivative Pricing. Cambridge University Press.
• Björk, T. (2020). Arbitrage Theory in Continuous Time (4th ed.). Oxford University Press.
• Delbaen, F., & Schachermayer, W. (1994). A general version of the fundamental theorem of asset pricing. Mathematische Annalen, 300, 463–520.
• Girsanov, I. V. (1960). On transforming a certain class of stochastic processes by absolutely continuous substitution of measures. Theory of Probability and Its Applications, 5(3), 285–301.
• Harrison, J. M., & Kreps, D. M. (1979). Martingales and arbitrage in multiperiod securities markets. Journal of Economic Theory, 20(3), 381–408.
• Harrison, J. M., & Pliska, S. R. (1981). Martingales and stochastic integrals in the theory of continuous trading. Stochastic Processes and Their Applications, 11(3), 215–260.
• Hull, J. C. (2021). Options, Futures, and Other Derivatives (11th ed.). Pearson.
• Joshi, M. S. (2003). The Concepts and Practice of Mathematical Finance. Cambridge University Press.
• Shreve, S. E. (2004). Stochastic Calculus for Finance II: Continuous-Time Models. Springer.

---

Frequently Asked Questions

What is the difference between the real-world (P) and risk-neutral (Q) measures?

Under (\mathbb{P}), asset returns include a risk premium — investors demand more than the risk-free rate to hold risky assets. Under (\mathbb{Q}), the risk premium has been priced out and every asset earns the risk-free rate in expectation. (\mathbb{P}) is used for forecasting, portfolio construction, and risk management. (\mathbb{Q}) is used for derivatives pricing. The two measures are equivalent — they agree on which events are possible — but assign different probabilities to those events.

Does the physical probability of a stock going up matter for option pricing?

No, remarkably. As the binomial example shows, the option's no-arbitrage price depends only on the up and down levels, the interest rate, and the volatility — not on how likely each state is under the physical measure. This independence is the reason two investors with different views on the stock's direction still agree on the option's fair price.

Why does the drift change from mu to r under Q?

Because (\mathbb{Q}) is chosen to make every discounted asset price a martingale. If the asset itself grows at rate (r), then (S_t / B_t = S_t / e^{rt}) has zero drift and is a martingale. Girsanov's theorem provides the explicit change of Brownian motion that achieves this drift shift while preserving volatility.

What is the market price of risk?

The market price of risk is the ratio ((\mu - r) / \sigma) — the excess return per unit of risk — which is precisely the quantity subtracted from the Brownian drift under Girsanov's theorem. It is also the Sharpe ratio of the asset. Under (\mathbb{Q}) the market price of risk is zero by construction, which is another way of saying that risk is unpriced under the risk-neutral measure.

Why do incomplete markets have multiple risk-neutral measures?

Because in an incomplete market there are contingent claims that cannot be exactly replicated by trading in the underlying. No replication means the price cannot be pinned down by no-arbitrage alone. Multiple risk-neutral measures correspond to different possible prices, all consistent with no-arbitrage among the traded instruments. In practice, one measure is selected by calibrating to observed market prices of hedge instruments.

Is the risk-free rate under Q the same as the observed short rate?

Yes, in the standard formulation. Under (\mathbb{Q}), the drift of every tradeable asset is the short rate. If the short rate itself is stochastic (as in Vasicek, CIR, Hull-White), the discount factor becomes (e^{-\int_0^T r_s ds}) inside the expectation, and the appropriate risk-neutral measure is the money-market account measure. Working with forward measures, swap measures, or other numeraires produces different but equally valid pricing formulas.