Risk-Neutral Valuation Principle¶
In the unifying framework of this section, risk-neutral valuation is the payoff of achieving control — once local martingales have been upgraded to true martingales under \(\mathbb{Q}\), pricing reduces to computing expectations.
The risk-neutral valuation principle is the central pricing formula of mathematical finance. It states that the price of any derivative equals the discounted expected payoff under the risk-neutral measure, not the physical measure.
Prerequisites
This section assumes familiarity with:
The Fundamental Pricing Formula¶
This formula works because control succeeded: local martingales have been upgraded to true martingales under \(\mathbb{Q}\), so the conditional expectation below is well-defined and yields a unique price. In an arbitrage-free market, the time-\(t\) price of a contingent claim with payoff \(\Phi(X_T)\) at maturity \(T\) is:
where:
- \(\mathbb{Q}\) is the risk-neutral (equivalent martingale) measure
- \(r_s\) is the instantaneous risk-free rate
- \(\mathcal{F}_t\) is the information available at time \(t\)
For constant interest rate \(r\):
Foundation: The Fundamental Theorem of Asset Pricing¶
The risk-neutral valuation principle rests on the Fundamental Theorem of Asset Pricing (FTAP):
First Fundamental Theorem
A market is arbitrage-free if and only if there exists an equivalent martingale measure \(\mathbb{Q}\) under which discounted prices of traded assets are martingales.
Second Fundamental Theorem
An arbitrage-free market is complete (every contingent claim can be replicated) if and only if the equivalent martingale measure \(\mathbb{Q}\) is unique.
Key implications:
- The existence of \(\mathbb{Q}\) is equivalent to no-arbitrage—it's not an assumption but a consequence
- In complete markets, there is exactly one no-arbitrage price for every derivative
- In incomplete markets, multiple equivalent martingale measures exist, leading to price bounds rather than unique prices
See Fundamental Theorem of Asset Pricing for proofs and details.
Why Risk-Neutral, Not Physical?¶
Under the physical measure \(\mathbb{P}\):
- Expected returns reflect risk preferences
- Different investors may assign different values
- No unique price emerges from expectations alone
Under the risk-neutral measure \(\mathbb{Q}\):
- All assets earn the risk-free rate in expectation
- Prices are determined by no-arbitrage alone
- Unique prices emerge (in complete markets)
Key insight: Risk-neutral pricing is not about beliefs—it's about the price required to prevent arbitrage. The measure \(\mathbb{Q}\) encodes the prices of traded options, not anyone's subjective probabilities.
Derivation from No-Arbitrage¶
Step 1: FTAP Gives the Martingale Property¶
By the First Fundamental Theorem, no-arbitrage implies existence of \(\mathbb{Q}\) such that discounted prices of traded assets are \(\mathbb{Q}\)-martingales.
For the money market account \(B_t = e^{\int_0^t r_s\,ds}\) as numeraire, the discounted stock price:
is a \(\mathbb{Q}\)-martingale.
Step 2: Derivative Prices Must Also Be Martingales¶
Let \(V_t\) be the price of a derivative with payoff \(\Phi(X_T)\) at time \(T\). If the derivative can be replicated by trading in the underlying assets, its discounted price must also be a \(\mathbb{Q}\)-martingale (otherwise an arbitrage exists between the derivative and its replicating portfolio).
Therefore:
is a \(\mathbb{Q}\)-martingale.
Step 3: Apply the Martingale Property¶
By the martingale property:
Step 4: Use the Terminal Condition¶
At maturity, \(V_T = \Phi(X_T)\), so:
Step 5: Rearrange¶
Multiplying both sides by \(e^{\int_0^t r_s\,ds}\):
The Meaning of Risk Neutrality¶
Under \(\mathbb{Q}\), investors behave as if they are indifferent to risk:
| Aspect | Physical World (\(\mathbb{P}\)) | Risk-Neutral World (\(\mathbb{Q}\)) |
|---|---|---|
| Stock drift | \(\mu\) (includes risk premium) | \(r\) (risk-free rate) |
| Risk premium | \(\mu - r > 0\) (typically) | \(0\) |
| Interpretation | Reflects beliefs and preferences | Reflects no-arbitrage prices |
Important: Risk-neutral pricing does not assume investors are actually risk-neutral. It's a mathematical transformation that simplifies pricing by absorbing risk preferences into the measure change.
Connection to PDE Pricing¶
Risk-neutral valuation is equivalent to solving the pricing PDE via the Feynman–Kac theorem.
The Pricing PDE¶
If \(V(t,x)\) is the price as a function of time \(t\) and underlying value \(x\), it satisfies:
where \(\mathcal{L}^{\mathbb{Q}}\) is the infinitesimal generator under \(\mathbb{Q}\).
Generator for Geometric Brownian Motion¶
For a stock following \(dS_t = rS_t\,dt + \sigma S_t\,dW_t^{\mathbb{Q}}\) under \(\mathbb{Q}\):
Note that the drift is \(r\), not \(\mu\)—this is the risk-neutral drift.
The Black–Scholes PDE¶
Substituting into the pricing PDE:
This is the Black–Scholes PDE.
Feynman–Kac Connection¶
The Feynman–Kac theorem states that the solution to the PDE equals the risk-neutral expectation:
Two equivalent computational approaches:
- Solve the PDE (analytical methods, finite differences)
- Compute the expectation (Monte Carlo simulation)
See Feynman–Kac Formula for the rigorous statement and proof.
Examples¶
Example 1: European Call Option¶
Under \(\mathbb{Q}\), the stock follows:
The call price is:
Evaluating the expectation: Under \(\mathbb{Q}\), the terminal stock price is:
where \(Z \sim N(0,1)\) under \(\mathbb{Q}\). The expectation of \((S_T - K)^+\) for lognormal \(S_T\) can be evaluated explicitly (see Black–Scholes Formula), yielding:
where:
Example 2: Zero-Coupon Bond¶
A bond paying \(1\) at time \(T\) has price:
For constant \(r\): \(P(t,T) = e^{-r(T-t)}\).
For stochastic \(r_t\), the expectation depends on the short-rate model (Vasicek, CIR, etc.).
Example 3: Digital (Binary) Option¶
A digital call paying \(1\) if \(S_T > K\) has price:
Under the Black–Scholes model:
where \(d_2\) is as defined above, and \(\Phi\) is the standard normal CDF.
What Changes Under Q?¶
| Quantity | Under \(\mathbb{P}\) | Under \(\mathbb{Q}\) |
|---|---|---|
| Drift of \(S\) | \(\mu\) | \(r\) |
| Volatility of \(S\) | \(\sigma\) | \(\sigma\) (unchanged) |
| Brownian motion | \(W^{\mathbb{P}}\) | \(W^{\mathbb{Q}} = W^{\mathbb{P}} + \int_0^\cdot \theta_s\,ds\) |
| Probabilities | Reflect beliefs | Reflect market prices |
Volatility Invariance¶
The instantaneous (diffusion) volatility \(\sigma(t, S_t)\) is invariant under equivalent measure changes—only drift changes. This follows from Girsanov's theorem: the measure change affects the drift of the Brownian motion but not its quadratic variation.
Clarification: This refers to the model volatility parameter. Implied volatility (extracted from option prices) and realized volatility (computed from historical data) are market/statistical quantities that don't "change under measure"—they are what they are.
Generalization: Change of Numéraire¶
The risk-neutral measure uses the money market account \(B_t = e^{\int_0^t r_s\,ds}\) as numéraire. More generally, for any positive tradeable asset \(N_t\) as numéraire:
where \(\mathbb{Q}^N\) is the measure under which prices divided by \(N_t\) are martingales.
Common choices:
- \(N_t = B_t\) (money market): gives the standard risk-neutral measure \(\mathbb{Q}\)
- \(N_t = P(t,T)\) (zero-coupon bond): gives the \(T\)-forward measure \(\mathbb{Q}^T\)
- \(N_t = S_t\) (stock): gives the stock measure
See Numéraire and Measure Change for details.
Limitations and Extensions¶
Incomplete Markets¶
When markets are incomplete (unhedgeable risks exist):
- Multiple equivalent martingale measures exist
- Risk-neutral pricing gives a range of arbitrage-free prices, not a unique price
- Additional criteria are needed to select a price:
- Utility-based pricing: maximize expected utility
- Variance-optimal measure: minimize hedging error variance
- Minimal entropy measure: closest to \(\mathbb{P}\) in relative entropy
- Superhedging: worst-case bound
See Incomplete Markets and Pricing Bounds.
Model Dependence¶
Risk-neutral pricing requires:
- A model for the underlying dynamics (GBM, stochastic vol, jumps, etc.)
- Calibration to market prices of liquid instruments
- The derivative price is only as good as the model
Model risk is the risk that the model is misspecified.
Path-Dependent Options¶
For options whose payoff depends on the entire path \(\{X_s\}_{t \leq s \leq T}\):
Examples include:
- Asian options: payoff depends on average price
- Barrier options: payoff depends on whether price crosses a level
- Lookback options: payoff depends on maximum or minimum price
These expectations typically require Monte Carlo simulation or specialized PDE methods.
Summary¶
| Aspect | Description |
|---|---|
| What it says | Price = discounted expected payoff under \(\mathbb{Q}\) |
| Foundation | Fundamental Theorem of Asset Pricing (no-arbitrage ⟺ \(\mathbb{Q}\) exists) |
| Why \(\mathbb{Q}\) | Ensures no-arbitrage; unique prices in complete markets |
| What changes | Drift becomes \(r\); volatility unchanged |
| PDE equivalence | Same as solving Black–Scholes PDE (Feynman–Kac) |
| Limitations | Incomplete markets → non-unique prices; model dependence |
Key Takeaway
The risk-neutral valuation principle transforms the economic problem of pricing into a mathematical problem of computing expectations. The measure \(\mathbb{Q}\) is not about beliefs—it encodes the no-arbitrage constraints from traded asset prices. In complete markets, this gives unique derivative prices; in incomplete markets, it gives bounds.
Exercises¶
Exercise 1. A stock follows \(dS_t = rS_t\,dt + \sigma S_t\,dW_t^{\mathbb{Q}}\) under the risk-neutral measure with \(r = 0.05\), \(\sigma = 0.30\), \(S_0 = 100\), and \(T = 0.5\). Compute the risk-neutral price of a European put option with strike \(K = 95\) using the Black-Scholes formula. Verify that put-call parity holds.
Solution to Exercise 1
With \(r = 0.05\), \(\sigma = 0.30\), \(S_0 = 100\), \(T = 0.5\), \(K = 95\):
The call price is:
The put price via put-call parity is:
Verification of put-call parity: \(C - P = 12.30 - 4.93 = 7.37\) and \(S_0 - Ke^{-rT} = 100 - 92.63 = 7.37\). The parity holds.
Exercise 2. A digital (binary) call option pays \(\$1\) if \(S_T > K\) and nothing otherwise. Derive its price under risk-neutral valuation and show that it equals \(e^{-rT}\Phi(d_2)\). What happens to the price as \(\sigma \to 0\)?
Solution to Exercise 2
The digital call pays \(\Phi_{\text{digital}} = \mathbf{1}_{S_T > K}\). By risk-neutral valuation:
Under \(\mathbb{Q}\), \(\ln S_T = \ln S_0 + (r - \sigma^2/2)T + \sigma\sqrt{T}\,Z\) where \(Z \sim N(0,1)\). Therefore:
Hence:
As \(\sigma \to 0\): The stock becomes deterministic with \(S_T = S_0 e^{rT}\). If \(S_0 e^{rT} > K\) (i.e., the option is in-the-money at the forward), then \(d_2 \to +\infty\) and \(\Phi(d_2) \to 1\), giving \(V_0 \to e^{-rT}\). If \(S_0 e^{rT} < K\), then \(d_2 \to -\infty\) and \(V_0 \to 0\). If \(S_0 e^{rT} = K\), \(d_2 \to 0\) and \(V_0 \to e^{-rT}/2\). The price converges to the discounted deterministic payoff.
Exercise 3. Explain why the risk-neutral valuation formula uses \(\mathbb{Q}\) rather than \(\mathbb{P}\). Specifically, show that computing \(e^{-rT}\mathbb{E}^{\mathbb{P}}[\Phi(S_T)]\) does not in general yield a no-arbitrage price. Under what special condition on the payoff \(\Phi\) would the two computations agree?
Solution to Exercise 3
Under \(\mathbb{P}\), \(S_T = S_0 \exp((\mu - \sigma^2/2)T + \sigma W_T^{\mathbb{P}})\), so:
uses the physical drift \(\mu\), not \(r\), in the distribution of \(S_T\). If \(\Phi(x) = x\) (the payoff is the stock itself), then:
unless \(\mu = r\). This violates no-arbitrage since the discounted stock price should have expectation \(S_0\) under the pricing measure, not \(S_0 e^{(\mu-r)T}\).
The fundamental issue is that \(\mathbb{P}\)-expectations do not account for the market price of risk. The \(\mathbb{Q}\)-expectation correctly prices by eliminating risk premia from the drift.
The two computations agree when \(\Phi\) is a constant payoff: \(\Phi(S_T) = c\). Then \(e^{-rT}\mathbb{E}^{\mathbb{Q}}[c] = e^{-rT}\mathbb{E}^{\mathbb{P}}[c] = ce^{-rT}\), since a constant is independent of the measure. More generally, the two agree if \(\Phi(S_T)\) is \(\mathcal{F}_0\)-measurable (deterministic payoff), so the expectation does not depend on the probability measure at all.
Exercise 4. Derive the risk-neutral valuation formula from no-arbitrage in five steps: (i) apply the FTAP to get the martingale property of discounted prices, (ii) extend to derivative prices, (iii) apply the martingale property, (iv) use the terminal condition, and (v) rearrange. At which step does the assumption of market completeness enter?
Solution to Exercise 4
(i) The FTAP states that NFLVR holds if and only if there exists \(\mathbb{Q} \sim \mathbb{P}\) such that the discounted traded asset \(\tilde{S}_t = e^{-\int_0^t r_s\,ds}S_t\) is a \(\mathbb{Q}\)-martingale.
(ii) If the derivative with price \(V_t\) can be replicated by a self-financing strategy in the traded assets, then \(\tilde{V}_t = e^{-\int_0^t r_s\,ds}V_t\) must also be a \(\mathbb{Q}\)-martingale. If it were not, an arbitrage would exist between the derivative and its replicating portfolio.
(iii) The martingale property gives: \(\tilde{V}_t = \mathbb{E}^{\mathbb{Q}}[\tilde{V}_T \mid \mathcal{F}_t]\).
(iv) At maturity \(V_T = \Phi(X_T)\), so \(\tilde{V}_T = e^{-\int_0^T r_s\,ds}\Phi(X_T)\). Substituting: \(\tilde{V}_t = \mathbb{E}^{\mathbb{Q}}[e^{-\int_0^T r_s\,ds}\Phi(X_T) \mid \mathcal{F}_t]\).
(v) Since \(\tilde{V}_t = e^{-\int_0^t r_s\,ds}V_t\), multiply both sides by \(e^{\int_0^t r_s\,ds}\):
Market completeness enters at step (ii): the argument that \(\tilde{V}_t\) is a \(\mathbb{Q}\)-martingale requires the derivative to be replicable. In an incomplete market, the claim may not be replicable, and different equivalent martingale measures give different prices. Completeness (equivalently, uniqueness of \(\mathbb{Q}\)) ensures a unique no-arbitrage price.
Exercise 5. Consider a stochastic interest rate model where \(r_t\) follows the Vasicek process. The price of a zero-coupon bond is \(P(t, T) = \mathbb{E}^{\mathbb{Q}}[\exp(-\int_t^T r_s\,ds) | \mathcal{F}_t]\). Explain why this is a direct application of the risk-neutral valuation principle with payoff \(\Phi = 1\). Why is the discount factor inside the expectation rather than outside when \(r_t\) is stochastic?
Solution to Exercise 5
A zero-coupon bond pays \(\Phi = 1\) at time \(T\). The risk-neutral valuation formula gives:
This is indeed the risk-neutral valuation formula with constant payoff \(\Phi = 1\).
The discount factor \(e^{-\int_t^T r_s\,ds}\) must be inside the expectation because it is random when \(r_t\) is stochastic. The integral \(\int_t^T r_s\,ds\) depends on the future path of \(r_s\), which is unknown at time \(t\). If we wrote \(e^{-r(T-t)}\mathbb{E}^{\mathbb{Q}}[1]\), we would be treating the discount factor as deterministic, which is only valid when \(r\) is constant.
Under the Vasicek model, \(r_t\) is an Ornstein–Uhlenbeck process under \(\mathbb{Q}\), and \(\int_t^T r_s\,ds\) is Gaussian (as a linear functional of a Gaussian process). The expectation can be computed in closed form, yielding the affine bond price formula \(P(t,T) = \exp(A(T-t) - B(T-t)r_t)\) for explicit functions \(A\) and \(B\).
Exercise 6. In an incomplete market, two equivalent martingale measures \(\mathbb{Q}_1\) and \(\mathbb{Q}_2\) both exist. For a non-traded claim with payoff \(\Phi(X_T)\), show that the two measures can give different prices \(V_0^{(1)} \neq V_0^{(2)}\), both consistent with no-arbitrage. Explain the economic meaning of the pricing interval \([\underline{V}, \overline{V}]\).
Solution to Exercise 6
Let \(\mathbb{Q}_1\) and \(\mathbb{Q}_2\) be two equivalent martingale measures. The two prices are:
Both are valid no-arbitrage prices because both measures make discounted traded-asset prices into martingales. However, since \(\Phi(X_T)\) is a non-traded claim, different measures assign different expectations.
To see \(V_0^{(1)} \neq V_0^{(2)}\) in general, note that \(\mathbb{Q}_1 \neq \mathbb{Q}_2\) means they differ on some events. If \(\Phi\) is non-constant and depends on the unhedgeable risk, then the difference in measure weighting leads to different expectations.
The pricing interval \([\underline{V}, \overline{V}]\) is defined as:
where \(\mathcal{M}\) is the set of all equivalent martingale measures. Any price within \([\underline{V}, \overline{V}]\) is consistent with no-arbitrage. The bounds \(\underline{V}\) and \(\overline{V}\) correspond to superhedging prices: \(\overline{V}\) is the minimum cost to super-replicate \(\Phi(X_T)\) (seller's price), and \(\underline{V}\) is the maximum price a buyer would pay while maintaining no-arbitrage (buyer's price).
Exercise 7. Show that the Black-Scholes PDE
is equivalent to the risk-neutral valuation formula via the Feynman-Kac theorem. Identify the generator \(\mathcal{L}^{\mathbb{Q}}\), the discount rate, and the terminal condition. Explain why the PDE contains \(r\) (not \(\mu\)) in the drift term.
Solution to Exercise 7
The Black-Scholes PDE is:
with terminal condition \(V(T,x) = \Phi(x)\).
Identifying components for Feynman–Kac:
- Generator: \(\mathcal{L}^{\mathbb{Q}} = rx\frac{\partial}{\partial x} + \frac{1}{2}\sigma^2 x^2\frac{\partial^2}{\partial x^2}\). This is the infinitesimal generator of geometric Brownian motion under \(\mathbb{Q}\): \(dS_t = rS_t\,dt + \sigma S_t\,dW_t^{\mathbb{Q}}\).
- Discount rate: \(r\) (the coefficient of \(-V\) in the PDE).
- Terminal condition: \(V(T,x) = \Phi(x)\).
The Feynman–Kac theorem states that the solution to \(\frac{\partial V}{\partial t} + \mathcal{L}^{\mathbb{Q}}V - rV = 0\) with \(V(T,x) = \Phi(x)\) is:
where \(S\) follows the \(\mathbb{Q}\)-dynamics. This is exactly the risk-neutral valuation formula.
The PDE contains \(r\) (not \(\mu\)) because the generator \(\mathcal{L}^{\mathbb{Q}}\) corresponds to the risk-neutral dynamics \(dS = rS\,dt + \sigma S\,dW^{\mathbb{Q}}\). The measure change from \(\mathbb{P}\) to \(\mathbb{Q}\) via Girsanov replaces the physical drift \(\mu\) with \(r\). The PDE is derived by requiring that the discounted option price be a \(\mathbb{Q}\)-martingale, which forces the drift to be \(r\).