Physical vs Risk-Neutral World¶
Every derivative price in this book ultimately rests on a single formula: \(V_0 = \mathbb{E}^{\mathbb{Q}}[e^{-rT}X_T]\). The expectation is taken not under the probability measure that governs real-world returns, but under a different measure \(\mathbb{Q}\) constructed specifically for pricing. The distinction between these two measures---the physical measure \(\mathbb{P}\) and the risk-neutral measure \(\mathbb{Q}\)---is the most important conceptual boundary in quantitative finance. Confusing them leads to pricing errors, flawed risk estimates, and deep misunderstandings of what models actually say.
This page owns the conceptual boundary: what each measure is, why they differ, and how they relate. The algebraic mechanism that effects the change is developed in § Risk Premium Decomposition; the pricing formula it makes possible is developed in § Risk-Neutral Valuation Principle.
The Physical Measure ℙ¶
When we observe market prices, estimate returns from historical data, or simulate future portfolio values for risk management, we are working under the physical measure \(\mathbb{P}\). This is the measure that governs how asset prices actually evolve in the real world.
Under \(\mathbb{P}\), a stock modeled by geometric Brownian motion satisfies
where \(\mu\) is the physical drift (expected return) and \(\sigma\) is the volatility. The drift \(\mu\) reflects economic fundamentals: growth prospects, risk premia demanded by investors, and the compensation required for bearing uncertainty. Different assets carry different drifts because investors require different compensation for different risks.
The physical measure is the right tool for:
- Forecasting: estimating where prices will be in the future
- Risk management: computing Value-at-Risk, expected shortfall, and stress-test scenarios
- Portfolio optimization: maximizing expected utility of terminal wealth
What \(\mathbb{P}\) is not suited for is pricing derivatives. The reason is subtle but fundamental: a derivative's price must be consistent with no-arbitrage, and no-arbitrage imposes constraints that the physical measure does not automatically satisfy.
The Risk-Neutral Measure ℚ¶
The risk-neutral measure \(\mathbb{Q}\) is not a description of how prices actually evolve---it is a mathematical device engineered so that derivative prices can be computed as discounted expectations. Its construction is the central achievement of the no-arbitrage pricing theory.
Recall (see § Risk-Neutral Valuation Principle): \(\mathbb{Q}\) is defined by the martingale property \(\mathbb{E}^{\mathbb{Q}}[S_T/B_T \mid \mathcal{F}_t] = S_t/B_t\), which yields the pricing formula \(V_0 = \mathbb{E}^{\mathbb{Q}}[e^{-rT}X_T]\) for any contingent claim. The entire problem of pricing a derivative reduces to computing an integral under the right measure.
Under \(\mathbb{Q}\), the stock dynamics become
Every traded asset drifts at the risk-free rate \(r\), regardless of its physical drift \(\mu\). The volatility \(\sigma\) is unchanged. This is not a statement about investor beliefs---it is a consequence of the no-arbitrage requirement built into the measure.
Same Paths, Different Weights¶
A crucial point is that \(\mathbb{P}\) and \(\mathbb{Q}\) are equivalent measures: they agree on which events are possible and which are impossible. Formally, \(\mathbb{P}(A) = 0\) if and only if \(\mathbb{Q}(A) = 0\) for every measurable event \(A\). The set of sample paths \(\omega \in \Omega\) is the same under both measures---the stock can reach the same prices, follow the same trajectories, and produce the same extreme outcomes.
What changes is the weighting of those paths. Recall (see § Girsanov's Theorem): the Radon--Nikodym derivative \(Z_T = \exp(-\theta W_T^{\mathbb{P}} - \tfrac{1}{2}\theta^2 T)\) with \(\theta = (\mu - r)/\sigma\) reweights each path: high-\(W_T^{\mathbb{P}}\) paths (good outcomes) are downweighted, low-\(W_T^{\mathbb{P}}\) paths (bad outcomes) are upweighted. The effect is precisely to remove the risk premium from the drift.
Equivalence is essential
If \(\mathbb{Q}\) assigned zero probability to an event that \(\mathbb{P}\) considers possible, a contingent claim paying off only in that event would be priced at zero yet could deliver a positive payoff---a free lunch. The fundamental theorem of asset pricing requires equivalence to prevent such arbitrage.
Why Drift Disappears¶
The conceptual takeaway is simple: the physical drift \(\mu\) enters the Radon--Nikodym derivative but cancels out of every \(\mathbb{Q}\)-expectation, so derivative prices depend on \(r\), \(\sigma\), \(S_0\), \(K\), and \(T\)---never on \(\mu\). The algebraic mechanism by which the risk premium \(\mu - r\) is absorbed into the measure change is developed in § Risk Premium Decomposition, and the pricing formula that uses these \(\mathbb{Q}\)-dynamics is developed in § Risk-Neutral Valuation Principle.
Why ℚ Is Not "Believed"¶
The risk-neutral measure is one of the most misunderstood objects in quantitative finance. Because every asset earns the risk-free rate under \(\mathbb{Q}\), it is tempting to interpret \(\mathbb{Q}\) as describing a world of risk-neutral investors. This interpretation is misleading in several ways.
\(\mathbb{Q}\) is not a forecast. Risk-neutral probabilities do not predict future prices. If a stock has a 60% physical probability of rising and a 45% risk-neutral probability of rising, the 45% figure carries no information about what will actually happen.
\(\mathbb{Q}\) does not reflect beliefs. No market participant believes that all assets earn \(r\). The measure \(\mathbb{Q}\) is a mathematical construction---a change of variables that converts pricing into an expectation calculation.
\(\mathbb{Q}\) does encode risk preferences---indirectly. The Radon--Nikodym derivative \(Z_T = d\mathbb{Q}/d\mathbb{P}\) depends on the market price of risk \(\theta\), which measures the compensation investors demand for bearing uncertainty. In a truly risk-neutral world, \(\mathbb{P}\) would already satisfy the martingale property and \(\mathbb{Q} = \mathbb{P}\). The fact that \(\mathbb{Q} \neq \mathbb{P}\) is precisely the signature of risk aversion. At a deeper level, the measure change is equivalent to introducing a stochastic discount factor that reweights outcomes by marginal utility---see Stochastic Discount Factor.
The correct framing: \(\mathbb{Q}\) is the unique measure (in complete markets) that makes discounted prices martingales and thereby enforces no-arbitrage pricing. It is a pricing tool, not a worldview.
Practical Implications¶
The P/Q distinction is not merely philosophical---it has direct consequences for how quantitative work is organized. Using the wrong measure for a given task produces systematic errors that no amount of calibration can fix.
| Task | Measure | Why |
|---|---|---|
| Derivative pricing | \(\mathbb{Q}\) | No-arbitrage requires martingale discounting |
| Hedging ratios (Greeks) | \(\mathbb{Q}\) | Greeks are derivatives of \(\mathbb{Q}\)-prices |
| Value-at-Risk | \(\mathbb{P}\) | VaR measures real-world loss probabilities |
| Expected shortfall | \(\mathbb{P}\) | Tail risk is a physical-measure concept |
| Portfolio optimization | \(\mathbb{P}\) | Expected returns matter for allocation |
| Model calibration | \(\mathbb{Q}\) | Market prices are \(\mathbb{Q}\)-expectations |
| Backtesting | \(\mathbb{P}\) | Historical returns are \(\mathbb{P}\)-realizations |
A common error
Using risk-neutral probabilities for risk management overstates the likelihood of adverse outcomes (because \(\mathbb{Q}\) upweights bad states). Using physical probabilities for pricing violates no-arbitrage consistency. The two measures answer different questions and must not be interchanged.
Exercises¶
Exercise 1. A stock follows geometric Brownian motion under \(\mathbb{P}\) with \(\mu = 0.12\), \(\sigma = 0.30\), and the risk-free rate is \(r = 0.04\). Compute the market price of risk \(\theta\), write down the stock dynamics under \(\mathbb{Q}\), and verify explicitly that the risk premium \(\mu - r\) is absorbed into the measure change.
Solution to Exercise 1
The market price of risk is
Girsanov's theorem defines \(W_t^{\mathbb{Q}} = W_t^{\mathbb{P}} + \theta t\), so \(dW_t^{\mathbb{P}} = dW_t^{\mathbb{Q}} - \theta\,dt\). Substituting into the \(\mathbb{P}\)-dynamics:
Since \(\sigma\theta = 0.30 \times 0.2667 = 0.08 = \mu - r\), the drift becomes
The risk-neutral dynamics are
The risk premium \(\mu - r = 0.08\) has been entirely absorbed into the measure change via the term \(\sigma\theta\,dt\). The volatility \(\sigma = 0.30\) is unchanged because quadratic variation is invariant under equivalent measure changes.
Exercise 2. A European call option has \(S_0 = 100\), \(K = 105\), \(T = 0.5\), \(r = 0.03\), and \(\sigma = 0.20\). Compute the Black--Scholes price \(C_0\). Then verify the consistency check \(C_0 \to S_0\) as \(K \to 0\).
Solution to Exercise 2
First compute \(d_1\) and \(d_2\):
Using the standard normal CDF:
The Black--Scholes price is
For the consistency check: as \(K \to 0\), the call becomes certain to finish in the money. We have \(d_1 = [\ln(S_0/K) + (r + \sigma^2/2)T]/(\sigma\sqrt{T}) \to +\infty\) and \(d_2 \to +\infty\), so \(N(d_1) \to 1\) and \(N(d_2) \to 1\). Then
confirming the formula correctly prices a zero-strike call at \(S_0\).
Exercise 3. Prove that the volatility \(\sigma\) is the same under \(\mathbb{P}\) and \(\mathbb{Q}\). Specifically, show that the quadratic variation \(\langle \ln S \rangle_t = \sigma^2 t\) is invariant under the measure change from \(\mathbb{P}\) to \(\mathbb{Q}\).
Solution to Exercise 3
Under \(\mathbb{P}\), Ito's formula gives
Under \(\mathbb{Q}\), the same calculation gives
The quadratic variation of \(\ln S\) is determined by the diffusion coefficient alone:
under both measures. The key insight is that quadratic variation is a pathwise property---it depends only on the realized sample path \(\omega\), not on how paths are weighted. Formally, quadratic variation is defined as
where the limit is taken along partitions of \([0, t]\). Since \(\mathbb{P}\) and \(\mathbb{Q}\) are equivalent measures (they agree on which paths have probability zero), the limit is the same path-by-path, and hence the quadratic variation is identical under both measures.
This is why derivative prices depend on \(\sigma\) (a path property) but not on \(\mu\) (a measure-dependent property): \(\sigma\) is observable from the path, while \(\mu\) is absorbed into the measure change. \(\square\)
Exercise 4. A risk manager computes that a portfolio has a 3% physical probability of losing more than $2 million over the next quarter. A pricing quant reports that the risk-neutral probability of the same event is 8%. Explain why these numbers differ. If a third analyst uses the 8% figure in a VaR report, what error does this introduce and in which direction?
Solution to Exercise 4
The two probabilities differ because they are computed under different measures that serve fundamentally different purposes.
The physical probability \(\mathbb{P}(\text{loss} > \$2\text{M}) = 3\%\) reflects actual market dynamics---empirical return distributions, risk premia, and historical correlations. This is the correct input for risk management.
The risk-neutral probability \(\mathbb{Q}(\text{loss} > \$2\text{M}) = 8\%\) comes from the pricing measure, which reweights scenarios to enforce the martingale property. The measure change from \(\mathbb{P}\) to \(\mathbb{Q}\) removes the positive risk premium from asset drifts, effectively lowering expected returns. This tilts the distribution toward adverse outcomes: losses become more probable under \(\mathbb{Q}\) than under \(\mathbb{P}\).
The third analyst's error is using the risk-neutral probability for risk management. Since \(\mathbb{Q}\) upweights bad states (to compensate for the removed risk premium), the 8% figure overstates the actual likelihood of the loss event. The VaR threshold would be set too conservatively: the analyst would report that the portfolio exceeds the $2M loss with probability 8% rather than the true 3%, leading to excessive capital reserves or unnecessarily restrictive position limits.
The correct practice: use \(\mathbb{P}\) for risk (VaR, expected shortfall, stress testing) and \(\mathbb{Q}\) for pricing (derivative valuation, calibration). The two measures answer different questions and must not be interchanged.
Exercise 5. Suppose \(\mathbb{Q} = \mathbb{P}\) for a particular asset (the physical and risk-neutral measures coincide). What does this imply about the asset's expected return under \(\mathbb{P}\)? What does it imply about investor risk preferences with respect to this asset? Prove your claims using the relationship \(\theta = (\mu - r)/\sigma\).
Solution to Exercise 5
If \(\mathbb{Q} = \mathbb{P}\), the Radon--Nikodym derivative is \(Z_T = d\mathbb{Q}/d\mathbb{P} = 1\) almost surely. Since
this requires \(\theta = 0\) (since \(W_T^{\mathbb{P}}\) is a non-degenerate random variable, the exponent must vanish identically). From \(\theta = (\mu - r)/\sigma = 0\), we conclude
The asset's expected return under \(\mathbb{P}\) equals the risk-free rate. Investors demand no compensation beyond \(r\) for holding this asset, meaning the risk premium is zero.
This implies that investors are risk-neutral with respect to this asset---they are indifferent between the risky asset and a risk-free bond with the same expected return. In a world where all investors are truly risk-neutral (not just with respect to one asset, but universally), \(\mathbb{Q} = \mathbb{P}\) for all traded assets, and the physical measure itself would satisfy the martingale property for discounted prices.
The contrapositive is equally important: whenever \(\mu \neq r\) (which is the empirically relevant case for equities), we have \(\theta \neq 0\) and \(\mathbb{Q} \neq \mathbb{P}\). The gap \(\mu - r\) is the risk premium, and the fact that \(\mathbb{Q} \neq \mathbb{P}\) is the mathematical signature of risk aversion. \(\square\)
Exercise 6. A colleague claims: "Under \(\mathbb{Q}\), all stocks earn the risk-free rate, so a stock with \(\mu = 0.15\) is overpriced relative to its \(\mathbb{Q}\)-return of \(r = 0.04\)." Identify every error in this statement and explain the correct interpretation of risk-neutral expected returns.
Solution to Exercise 6
The colleague's reasoning contains several fundamental errors.
Error 1: Confusing \(\mathbb{Q}\)-returns with valuation signals. Under \(\mathbb{Q}\), every traded asset has an expected return of \(r\). This is a mathematical property of the risk-neutral measure by construction---it is not a signal of mispricing. If this logic were correct, every stock in the market would be "overpriced," which is absurd.
Error 2: Comparing \(\mu\) against \(r\) in the wrong framework. The physical expected return \(\mu = 0.15\) and the risk-neutral expected return \(r = 0.04\) are computed under different measures. Comparing them directly is meaningless---they answer different questions. The quantity \(\mu - r = 0.11\) is the risk premium: the compensation investors earn for bearing the stock's risk. It is a feature of the market, not evidence of mispricing.
Error 3: Treating \(\mathbb{Q}\) as a belief or forecast. The risk-neutral measure is not a prediction about future returns. It is a mathematical tool that reweights probabilities to enforce the martingale property. Saying the stock "earns only \(r\) under \(\mathbb{Q}\)" is like saying a rotated coordinate system assigns different coordinates to the same point---it is a change of representation, not a change of reality.
Correct interpretation: The fact that \(\mu > r\) means investors require a positive risk premium to hold the stock, reflecting risk aversion. The measure change from \(\mathbb{P}\) to \(\mathbb{Q}\) removes this premium by downweighting favorable outcomes and upweighting unfavorable ones. The resulting \(\mathbb{Q}\)-expectation gives the correct no-arbitrage price, which is \(S_0\)---exactly the current market price. The stock is fairly priced by construction; the role of \(\mathbb{Q}\) is to ensure pricing consistency, not to assess whether assets are cheap or expensive.