aiwiki.page
English

Option Greeks

Option Greeks measure how option values respond to changes in underlying prices, volatility, time, interest rates, and other pricing inputs.

27 keywords4 linked from11 not yet writtenWritten by AI
Market RiskAssetBlack–Scholes mo…ArbitrageInterest RateYield CurveOver-the-Counter…DividendOption Gre…

Option Greeks are measures of the sensitivity of an option price to changes in the variables used to value it. The principal measures are delta, gamma, theta, vega, and rho, which describe exposure to the underlying asset’s price, changes in that price sensitivity, the passage of time, volatility, and interest rates, respectively. Most Greeks are mathematical derivatives of a pricing function rather than independently observable market quantities. Their values therefore depend on the contract, the pricing model, the market inputs, and the conventions used to change those inputs.

Greeks connect option valuation with the management of market risk. They allow positions with different strikes and maturities to be compared, aggregated, and approximately hedged. A portfolio that appears balanced by market value can nevertheless have substantial exposure to a large price movement, a change in implied volatility, or the approach of expiration. Greeks make these different exposures explicit, although they do not capture every source of loss.

The familiar Greek-letter notation developed around modern option-pricing theory, especially the work published by Fischer Black, Myron Scholes, and Robert C. Merton in 1973. Its use now extends across exchange-traded and over-the-counter derivatives, including equity, foreign-exchange, interest-rate, and commodity products. The interpretation of a reported Greek requires attention to its units, the pricing assumptions behind it, and whether it represents a local derivative or a finite market scenario.

Mathematical foundations and pricing conventions

An option grants its holder a contractual right, but not an obligation, to buy or sell an underlying asset on specified terms. A European call with strike KK and maturity TT pays max⁡(ST−K,0)\max(S_T-K,0); a European put pays max⁡(K−ST,0)\max(K-S_T,0). American options permit exercise before maturity. The distinction matters because the value and sensitivities of an early-exercise right cannot generally be obtained from the same formulas as those of a European contract.

Write an option’s value as

V=V(S,t;σ,r,q,K,T),V=V(S,t;\sigma,r,q,K,T),

where SS is the underlying price, tt is current time, σ\sigma is volatility, rr is an interest rate, and qq is a continuous income or dividend yield. The remaining time is τ=T−t\tau=T-t. Basic Greeks are partial derivatives: each changes one input while holding specified others fixed. This qualification is essential. A delta calculated with volatility held constant need not equal the price sensitivity observed when market volatility changes with the underlying price.

For sufficiently small changes, a multivariable Taylor expansion gives

ΔV≈Δ ΔS+12Γ(ΔS)2+Θ Δt+ν Δσ+ρ Δr+cross-variable and higher-order terms.\begin{aligned} \Delta V \approx{}& \Delta\,\Delta S +\frac12\Gamma(\Delta S)^2 +\Theta\,\Delta t +\nu\,\Delta\sigma +\rho\,\Delta r\\ &+\text{cross-variable and higher-order terms}. \end{aligned}

Here Δ\Delta denotes delta, whereas ΔS\Delta S denotes a price change. This expansion underlies much short-horizon profit-and-loss attribution. It is a local approximation, not a guarantee about the outcome of a large or discontinuous market move.

The standard analytical benchmark is the Black–Scholes–Merton model. For a European call on an asset paying a continuous yield,

C=Se−qτN(d1)−Ke−rτN(d2),C=Se^{-q\tau}N(d_1)-Ke^{-r\tau}N(d_2),

where

d1=ln⁡(S/K)+(r−q+12σ2)τστ,d2=d1−στ.d_1= \frac{\ln(S/K)+(r-q+\tfrac12\sigma^2)\tau} {\sigma\sqrt{\tau}}, \qquad d_2=d_1-\sigma\sqrt{\tau}.

The function NN is the standard normal cumulative distribution function; its density is denoted by ϕ\phi. The corresponding put follows from put–call parity:

C−P=Se−qτ−Ke−rτ.C-P=Se^{-q\tau}-Ke^{-r\tau}.

Differentiating this identity provides consistency checks for call and put Greeks.

The model derives prices from arbitrage restrictions and dynamic replication, not from a forecast of the underlying asset’s expected return. In its idealized setting, continuous trading, frictionless borrowing and lending, and continuous price paths allow risk to be offset dynamically. Earlier work, including Louis Bachelier’s 1900 thesis, addressed option valuation using different price-process assumptions. Black and Scholes’s 1973 paper and Merton’s related work established the modern replication framework; the Cox–Ross–Rubinstein binomial model of 1979 subsequently provided a widely used discrete-time implementation.

Delta and exposure to the underlying asset

Delta is the first derivative of option value with respect to the underlying price:

Δ=∂V∂S.\Delta=\frac{\partial V}{\partial S}.

It measures the approximate change in option value for a one-unit increase in the underlying, with other specified inputs unchanged. An option with delta 0.600.60 gains approximately 0.60 price units when the underlying rises by one unit. This interpretation concerns an infinitesimal or small move; delta itself changes as the underlying price, time, and other conditions change.

For European options in the Black–Scholes–Merton model,

ΔC=e−qτN(d1),ΔP=e−qτ[N(d1)−1].\Delta_C=e^{-q\tau}N(d_1), \qquad \Delta_P=e^{-q\tau}\bigl[N(d_1)-1\bigr].

A long standard call therefore has positive delta, and a long standard put has negative delta. For a non-dividend-paying underlying, their deltas lie between zero and one, and between minus one and zero, respectively. These familiar bounds do not describe every derivative structure: leveraged payoffs, digital options, and some other contracts can have substantially different sensitivities.

Delta supplies the basic hedge ratio in delta hedging. If a position consists of nn option contracts, each representing mm units of the underlying, its aggregate underlying-unit exposure is nmΔnm\Delta. A short position in that many underlying units offsets its first-order price sensitivity. For example, 100 long call contracts, each covering 100 shares and having delta 0.600.60, have a delta exposure of 6,000 shares. Selling 6,000 shares makes the combined position locally delta-neutral before accounting for hedge execution and other risks.

Market systems also report currency-scaled measures. “Dollar delta” often means SΔS\Delta multiplied by the position size and contract multiplier, so that an approximately 1% underlying move produces a profit or loss of 0.010.01 times dollar delta. Terminology is not uniform: some reports instead display the profit or loss for a specified percentage move directly. The numerical value is uninterpretable without its scaling convention.

Delta is sometimes described as the probability of exercise. In the standard non-dividend-paying European call model, however, N(d2)N(d_2), rather than N(d1)N(d_1), is the probability of finishing in the money under the usual risk-neutral pricing measure. Delta can have probability interpretations under other changes of measure or conventions, but it is not generally a real-world exercise probability. Risk-neutral valuation separates pricing probabilities from forecasts of actual market outcomes.

Foreign-exchange markets illustrate the importance of conventions. Delta may be expressed relative to spot or forward exchange rates and may be adjusted for the option premium. A strike described as “25-delta” therefore depends on the quoting convention as well as on the volatility surface. Consequently, two numerically identical delta labels need not refer to the same strike or cash exposure.

Gamma, convexity, and changing hedge requirements

Gamma is the second derivative of value with respect to the underlying price:

Γ=∂2V∂S2=∂Δ∂S.\Gamma=\frac{\partial^2V}{\partial S^2} =\frac{\partial\Delta}{\partial S}.

It measures how quickly delta changes when the underlying moves. In the Black–Scholes–Merton model, European calls and puts with the same strike and maturity have identical gamma:

Γ=e−qτϕ(d1)Sστ.\Gamma=\frac{e^{-q\tau}\phi(d_1)} {S\sigma\sqrt{\tau}}.

A long standard European option has positive gamma, reflecting the convexity of its value as a function of the underlying price.

Gamma explains why a delta hedge is only locally effective. Consider a call with S=100S=100, K=100K=100, r=5%r=5\%, q=0q=0, σ=20%\sigma=20\%, and one year remaining. Its model delta is approximately 0.63680.6368, and its gamma is approximately 0.018760.01876 per unit of underlying price. A small one-unit rise increases delta by about 0.018760.01876. The hedge established before the rise therefore leaves a residual long exposure unless it is rebalanced.

For an underlying move ΔS\Delta S, the second-order contribution to option value is approximately

12Γ(ΔS)2.\frac12\Gamma(\Delta S)^2.

In the example, a two-unit move contributes about 0.03750.0375 price units through this term, in addition to the delta contribution. The squared price change means that the gamma term is positive for either an upward or downward move when gamma is positive. This does not make a long option automatically profitable: its purchase price, time decay, financing, and volatility changes also affect the result.

Gamma is particularly concentrated near the exercise threshold as expiration approaches. For an approximately at-the-money option, the formula contains a factor proportional to 1/τ1/\sqrt{\tau}, which can become large as remaining time shrinks. Away from the strike, gamma commonly tends toward zero near expiry. The concentration reflects the transition from an uncertain exercise outcome to a payoff whose slope is essentially zero or one for a call. At the maturity payoff’s kink, the ordinary second derivative is not a finite, smooth quantity.

A delta-neutral portfolio with positive gamma tends to require selling underlying units after a price rise and buying them after a decline. A negative-gamma portfolio requires the reverse adjustment. This mechanical feature helps explain the association between short-option positions and costly hedging during volatile markets. It does not establish a universal relationship between dealer positioning and market direction, because actual positions, hedge instruments, trading constraints, and other flows differ.

Gamma can also be misleading when reported without scale. Some systems use raw gamma; others report the change in delta for a 1% move, 0.01SΓ0.01S\Gamma, or a currency profit-and-loss measure involving S2ΓS^2\Gamma. Comparisons across assets require a common shock definition. A low-priced underlying can have a large raw gamma without producing a larger currency exposure than a high-priced underlying.

Theta and the passage of time

Theta measures the change in value as calendar time advances, holding other specified inputs constant:

Θ=∂V∂t=−∂V∂τ.\Theta=\frac{\partial V}{\partial t} =-\frac{\partial V}{\partial\tau}.

This sign convention treats the maturity date as fixed. Some mathematical presentations instead differentiate with respect to remaining time, reversing the sign. Market reports commonly express theta per calendar day, while analytical formulas often give an annualized value.

For a European call under the Black–Scholes–Merton assumptions,

ΘC=−Se−qτϕ(d1)σ2τ−rKe−rτN(d2)+qSe−qτN(d1).\Theta_C= -\frac{Se^{-q\tau}\phi(d_1)\sigma}{2\sqrt{\tau}} -rKe^{-r\tau}N(d_2) +qSe^{-q\tau}N(d_1).

The first term represents the effect associated with the shrinking period of uncertainty. The other terms reflect financing and income. For the one-year at-the-money call described above, annual theta is approximately −6.414-6.414 price units. Dividing by 365 gives about −0.0176-0.0176 per calendar day, but this is a local conversion rather than the exact value change over every subsequent day.

Long calls and puts are often described as having negative theta because an unexercised option commonly loses time value as expiry approaches. This is a useful pattern, not a universal sign rule. A sufficiently deep-in-the-money European put can have positive theta when positive interest rates make delayed receipt of its strike value costly. Dividend yields, negative rates, and exercise restrictions can also alter the result. American exercise rights further change the comparison because the holder may be able to realize intrinsic value before maturity.

Theta and gamma are linked by the pricing equation. Under the standard model,

Θ+(r−q)SΔ+12σ2S2Γ−rV=0.\Theta+(r-q)S\Delta +\frac12\sigma^2S^2\Gamma-rV=0.

Thus time decay is not an independent charge arbitrarily attached to an option. It is constrained by the relationship among financing, directional exposure, convexity, and value. In the simplified case r=q=0r=q=0,

Θ=−12σ2S2Γ.\Theta=-\frac12\sigma^2S^2\Gamma.

Within that model, a positive-gamma option has negative theta.

The distinction between calendar time and market time complicates practical interpretation. Contracts have precise expiration dates, but trading activity is uneven across nights, weekends, and holidays. A volatility model may allocate variance differently to trading hours and non-trading periods. A Friday-to-Monday valuation change therefore need not equal three times a displayed daily theta, especially if the market’s volatility quotation changes at the same time.

Known events produce another complication. An earnings announcement or scheduled policy decision may account for a large portion of the uncertainty before expiry. Passing an ordinary day before the event does not necessarily remove the same amount of uncertainty as passing through the event itself. Theta calculated with a fixed annualized volatility can then describe a different scenario from one that holds the expected event-related variance fixed.

Vega and the structure of volatility exposure

Vega measures sensitivity to volatility:

ν=∂V∂σ.\nu=\frac{\partial V}{\partial\sigma}.

Despite its established place among the Greeks, “vega” is not the name of a Greek letter. In the Black–Scholes–Merton model, European calls and puts with matching terms have the same vega:

ν=Se−qτϕ(d1)τ.\nu=Se^{-q\tau}\phi(d_1)\sqrt{\tau}.

Long standard European calls and puts have positive vega: increasing volatility increases their model value when the other inputs remain fixed.

Volatility units are a frequent source of confusion. A formula using σ=0.20\sigma=0.20 for 20% volatility produces vega for a change of 1.00, or 100 percentage points. Traders often quote sensitivity to one percentage point, which is 0.01ν0.01\nu. In the numerical example, raw vega is about 37.5237.52, so a rise from 20% to 21% produces a first-order value increase of about 0.37520.3752. Exact repricing differs because vega itself changes.

In practice, options are often quoted through implied volatility: the volatility input that makes a chosen model reproduce a market price. Different strikes and expiries generally imply different volatilities. This collection forms a volatility surface, whose strike dependence is commonly called the volatility smile or skew. A single vega number compresses that surface into one assumed direction of change, often a parallel shift. It does not describe all possible changes in its shape.

Risk systems therefore calculate bucketed vegas by maturity and strike or delta region. A portfolio can have zero aggregate vega while remaining exposed to a steepening skew or a change in the difference between short- and long-dated volatility. For example, positive vega at a one-month maturity and equal negative vega at a one-year maturity cancel under a parallel shift but not under a movement confined to the first month. Interpolation and extrapolation rules affect how market quotes map into these buckets.

The treatment of the surface during an underlying-price move also affects delta. Under a “sticky-strike” assumption, the implied volatility attached to a fixed strike remains unchanged. Under a “sticky-delta” convention, the surface is treated as stable in delta coordinates, which generally causes a fixed-strike option’s volatility to change as the underlying moves. Schematically, if volatility is represented as σimp(S)\sigma_{\mathrm{imp}}(S), the effective sensitivity includes

dVdS=∂V∂S+∂V∂σdσimpdS.\frac{dV}{dS} = \frac{\partial V}{\partial S} + \frac{\partial V}{\partial\sigma} \frac{d\sigma_{\mathrm{imp}}}{dS}.

The second term combines vega with the assumed volatility response.

More elaborate models distinguish the current volatility state from parameters controlling its future evolution. In a stochastic-volatility model, sensitivity to instantaneous variance is not identical to sensitivity to a parallel shift of market implied volatilities. A calibration-based vega may require changing quoted option prices, recalibrating the model, and repricing the portfolio. Vega is therefore best understood as a sensitivity to a defined volatility perturbation, not as a uniquely standardized observable.

Rho, dividends, and financing variables

Rho conventionally measures sensitivity to an interest rate:

ρ=∂V∂r.\rho=\frac{\partial V}{\partial r}.

For European options with continuous yield qq, holding spot, volatility, and qq fixed,

ρC=Kτe−rτN(d2),ρP=−Kτe−rτN(−d2).\rho_C=K\tau e^{-r\tau}N(d_2), \qquad \rho_P=-K\tau e^{-r\tau}N(-d_2).

Under these assumptions, a higher rate increases a call’s value and reduces a put’s value. One interpretation is that the call holder postpones paying the strike until exercise, whereas a put holder postpones receiving it.

As with vega, rho must be scaled to the quoted rate movement. A raw rho of 50 means an approximate value change of 0.50 for a one-percentage-point rate increase, or 0.005 for a one-basis-point increase. In the running call example, raw rho is approximately 53.2353.23. A rise in the model rate from 5% to 6% therefore gives a first-order increase of about 0.53230.5323, provided the remaining inputs are held fixed.

The qualification about fixed inputs is especially important for rates. Under continuous carry, the forward price satisfies

F=Se(r−q)τ.F=Se^{(r-q)\tau}.

Changing rr while holding SS and qq fixed changes both the forward price and discounting. Changing a discount rate while holding an independently quoted forward fixed is a different experiment and produces a different sensitivity. Consequently, the familiar positive-call and negative-put rho signs are not universal statements about every option pricing convention.

Modern valuation commonly uses an entire yield curve, rather than one constant rate. A parallel curve shift gives only one aggregate sensitivity. Key-rate or bucketed sensitivities measure exposure to particular maturity regions. For products involving floating-rate cash flows, separate curves may be used to project payments and discount them. Collateral terms and contractual funding arrangements affect which curves enter the valuation, especially for derivatives traded in the over-the-counter market.

Income from the underlying introduces additional carry sensitivities. Differentiating with respect to a continuous dividend yield gives

∂C∂q=−τSe−qτN(d1),∂P∂q=τSe−qτN(−d1).\frac{\partial C}{\partial q} =-\tau Se^{-q\tau}N(d_1), \qquad \frac{\partial P}{\partial q} =\tau Se^{-q\tau}N(-d_1).

Higher expected dividends reduce the forward equity price, generally lowering call values and raising put values under this setup. Actual equity dividends are often modeled as discrete payments, requiring sensitivities to individual dividend amounts and dates rather than to a single continuous yield. The existing dividend forecast can be particularly consequential near an American call’s exercise boundary.

For currency options, domestic and foreign interest rates play distinct roles: the foreign rate acts mathematically like an income yield in the basic model. For commodity options, storage costs, financing costs, and convenience yields influence carry. Options on a futures contract may use a futures price as the primary state variable, making the definition of rate sensitivity different again. These examples show why a risk report must identify the economic variable being shocked, not merely label a column “rho.”

Higher-order and cross-variable Greeks

Higher-order Greeks describe how the primary sensitivities change. They become important when markets move far enough that a first-order approximation is insufficient, or when a hedge must remain effective across several interacting risk factors. Their terminology is less standardized than that of delta and gamma, and identical names can sometimes refer to differently scaled quantities. Explicit derivative definitions are therefore more reliable than names alone.

Vanna commonly denotes

∂2V∂S ∂σ=∂Δ∂σ=∂ν∂S,\frac{\partial^2V}{\partial S\,\partial\sigma} = \frac{\partial\Delta}{\partial\sigma} = \frac{\partial\nu}{\partial S},

when the function is sufficiently smooth. It measures how delta changes with volatility, or equivalently how vega changes with the underlying price. In a second-order expansion, the associated contribution is

Vanna ΔS Δσ.\mathrm{Vanna}\,\Delta S\,\Delta\sigma.

Unlike the pure second-order terms, this cross term has no factor of one-half after the symmetric terms have been combined. It matters particularly when price and volatility move together rather than independently.

Volga, also called vomma, usually means

Volga=∂2V∂σ2.\mathrm{Volga} =\frac{\partial^2V}{\partial\sigma^2}.

It measures the curvature of value with respect to volatility. In the Black–Scholes–Merton model,

Volga=νd1d2σ.\mathrm{Volga}=\nu\frac{d_1d_2}{\sigma}.

Its sign is not universally positive. The corresponding second-order value term is

12Volga(Δσ)2.\frac12\mathrm{Volga}(\Delta\sigma)^2.

A vega-neutral position can therefore remain exposed to a sufficiently large volatility movement because its positive and negative vega components may change at different rates.

Charm describes the change of delta with time, often written as ∂Δ/∂t\partial\Delta/\partial t. Some sources instead use remaining time, reversing the sign. Color similarly describes the time change of gamma, subject to the same convention issue. These measures help explain why a portfolio’s hedge requirement can change even if the underlying price remains stationary. This effect can become pronounced near expiry or near an exercise boundary.

Other established names include speed, the derivative ∂Γ/∂S\partial\Gamma/\partial S; zomma, the derivative ∂Γ/∂σ\partial\Gamma/\partial\sigma; and ultima, the third derivative of value with respect to volatility. Such quantities can improve local approximations, but adding more terms does not necessarily make a Taylor expansion reliable across a discontinuity or a major regime change. A high-order expansion around one market state may still perform poorly far from that state.

Multi-asset contracts introduce cross-gamma terms,

Γij=∂2V∂Si ∂Sj,\Gamma_{ij} =\frac{\partial^2V}{\partial S_i\,\partial S_j},

as well as sensitivities to correlations. A basket option, for example, can retain substantial risk after each individual asset delta has been offset. Correlation sensitivity is sometimes called “correlation Greek” or “cega,” but conventions vary. Hedging the component price exposures does not remove the effect of changing dependence among those components, especially when the payoff depends on their joint extreme outcomes.

Numerical estimation and difficult payoffs

Closed-form formulas are available for some standard contracts, but many Greeks are computed numerically. A central finite-difference estimate of delta is

Δ≈V(S+h)−V(S−h)2h,\Delta\approx \frac{V(S+h)-V(S-h)}{2h},

and a corresponding gamma estimate is

Γ≈V(S+h)−2V(S)+V(S−h)h2.\Gamma\approx \frac{V(S+h)-2V(S)+V(S-h)}{h^2}.

The bump hh must balance approximation error against numerical noise. A very large bump no longer describes local sensitivity well; a very small bump magnifies rounding, solver tolerances, and simulation error.

“Bump and revalue” is conceptually straightforward but can be computationally expensive for portfolios with thousands of market inputs. It also leaves an important modeling choice: whether a bumped price, rate, or volatility quote causes the model to be recalibrated. Keeping model parameters fixed measures a different risk from recalibrating after each market perturbation. Both may be useful, but they should not be presented as interchangeable.

Trees and finite-difference methods calculate option values on grids. They are especially useful for American exercise and path-dependent state variables, but estimated Greeks can be sensitive to grid placement. In a binomial tree, a strike’s location relative to terminal nodes can cause oscillating results as the number of steps changes. Near an early-exercise boundary, changes in the selected exercise policy can introduce further irregularity. Refining the grid and checking convergence are therefore part of establishing the numerical meaning of the reported sensitivity.

The Monte Carlo method values a payoff by averaging discounted outcomes over simulated paths. Greeks can be estimated by repeated simulations under bumped inputs, often using the same random numbers to reduce the variance of the difference. Pathwise differentiation instead differentiates simulated payoffs along each path when the required regularity holds. Likelihood-ratio methods differentiate the probability density rather than the payoff. Their relative efficiency depends on the contract and risk factor.

Discontinuous payoffs require special care. A cash-or-nothing digital call pays a fixed amount if the terminal price exceeds the strike. Its pathwise payoff derivative is zero almost everywhere, yet its value has a nonzero spot sensitivity before maturity. Naively differentiating the sampled payoff therefore fails. Conditional expectations, smoothing, likelihood-ratio estimators, or alternative representations can address this difficulty. Barrier options present related issues because a small market change can alter whether the contractual barrier is reached.

Automatic differentiation applies the chain rule through a pricing program. Reverse-mode methods, often called adjoint methods in finance, can calculate sensitivities to many inputs at a cost much lower than separate repricing for every input when the output is a single portfolio value. Their results still depend on correct treatment of calibration, numerical solvers, branching logic, and exercise decisions. Differentiating a program accurately does not by itself establish that the program’s financial assumptions are appropriate.

Numerical checking commonly combines several methods: analytical benchmarks where available, finite-difference comparisons, stability tests across bump sizes, and identities such as put–call parity. For discontinuous or strongly nonlinear products, agreement on price alone does not establish agreement on Greeks, because derivatives amplify small differences between pricing implementations.

Portfolio hedging, profit attribution, and limitations

Portfolio Greeks are obtained by summing position-level sensitivities, provided the risk factors, units, and valuation conventions are consistent. If aia_i is the signed quantity of instrument ii, including its contract multiplier where appropriate, then

Δportfolio=∑iaiΔi,Γportfolio=∑iaiΓi.\Delta_{\mathrm{portfolio}}=\sum_i a_i\Delta_i, \qquad \Gamma_{\mathrm{portfolio}}=\sum_i a_i\Gamma_i.

The same linear aggregation applies to other sensitivities to a common input. It does not justify adding unlike risk factors without conversion: equity delta, foreign-exchange delta, and interest-rate delta describe different shocks.

Hedge construction can be expressed as a system of exposure constraints. Underlying shares can adjust delta but ordinarily contribute no option gamma or vega. Options at different strikes and maturities provide combinations of gamma, vega, and other sensitivities. Solving for hedge quantities can make selected aggregate Greeks zero at the current market state. Such a solution may be unstable if the available instruments have nearly proportional sensitivities, and it can create exposures to unhedged maturities or volatility-surface regions.

Dynamic hedging links gamma to realized price variation. In a simplified setting with zero rates and income, continuous underlying paths, fixed model volatility, and no trading costs, a delta-hedged option has an approximate local gain

dΠ≈12ΓS2(σreal2−σmodel2)dt,d\Pi\approx \frac12\Gamma S^2 \left(\sigma_{\mathrm{real}}^2-\sigma_{\mathrm{model}}^2\right)dt,

where the realized variance rate describes the underlying’s quadratic variation. This expression explains the relationship between long-gamma positions, time decay, and realized volatility. It is not an unrestricted trading rule: changing implied volatility, jumps, discrete hedging, and financing complicate the result.

Actual hedge adjustments incur transaction costs, including the bid–ask spread, commissions, and market impact. Continuous rebalancing is therefore an idealization. Large overnight moves can occur before a hedge can be changed, while limited liquidity can prevent execution at modeled prices. A delta-neutral short option can consequently suffer a large loss from a jump even if its immediately preceding delta was accurately calculated. Funding requirements and a margin call can also constrain a hedge independently of its model-based effectiveness.

Greeks are widely used to attribute daily profit and loss. A report compares observed changes with those explained by delta, gamma, theta, vega, rates, and relevant cross terms. The unexplained residual can arise from higher-order effects, market-data changes, calibration differences, new trades, fees, or model deficiencies. The order in which changes are applied can affect an attribution when risks interact, so a sequential explanation and a simultaneous Taylor approximation need not assign identical amounts to each factor.

Local sensitivities complement rather than replace full repricing and stress testing. Large scenarios can involve simultaneous price gaps, skew changes, curve shifts, and reduced liquidity. Exotic options may also change behavior abruptly near barriers, fixing dates, or exercise thresholds. A portfolio with small current Greeks can retain considerable exposure to these events. Risk systems therefore distinguish local hedge measures from losses under specified scenarios, and from broader measures such as value at risk, each of which answers a different question about the distribution of possible outcomes.