2.9 priced the option. This page asks what happens when the inputs move. A Greek is a partial derivative of the price with respect to one input, and answers the trader's question: "if this input changes and everything else stays fixed, how much does my book move?"

For a developer, these five derivatives are as important as the price itself. They drive hedging, risk reports, desk screens and pricer tests, and they are the quantities the golden file carries and the C++ in Part 3 must reproduce to its tolerances.

What this page shows, and how

A · Five numbers. Compute the call and put Greeks at the base case, relate them to one-unit moves in stock, volatility, rates and time, and fix the units that the oracle and implementation use.

B · Bump-and-revalue. Check every analytic Greek against finite differences at the base case and across the 216-point grid, and show why choosing the bump size is its own problem.

C · Shapes. Study each Greek against spot and expiry: where the curves peak, where they vanish, and how the expiry singularities appear.

D · Theta pays for gamma. Re-read 2.7's PDE as a P&L statement, attribute every step of the hedged book to gamma, theta and carry, and connect the result to realised versus implied variance.

E · The put. Differentiate put–call parity and obtain the put's Greeks from the call's.

F · The oracle. Check all seven columns of the golden file and QuantLib against the closed forms across the grid and edge cases.

The experiment — what was done

The Black–Scholes closed forms for Δ\Delta, Γ\Gamma, vega, Θ\Theta, ρ\rho and dividend-rho were implemented and evaluated at the base case. Each analytic derivative was then checked against bump-and-revalue finite differences at the base point and across the same 216-point grid used in 2.9. The delta-hedged book from 2.7 was re-run over 4,000 paths, with every step attributed to gamma gain, theta cost and carry, so that the PDE could be read directly as a P&L decomposition. The Greeks were then mapped across spot and expiry, put parity identities checked across the grid, and all seven Greek columns verified against both QuantLib and the 7,234-row oracle.

·
·

Five numbers

At the base case (S0=K=100S_0 = K = 100, r=0.05r = 0.05, σ=0.4\sigma = 0.4, T=1T = 1) the call has

Δ=0.627Γ=0.00946vega=37.8Θ=−9.80ρ=44.7\begin{aligned} \Delta &= 0.627 \\ \Gamma &= 0.00946 \\ \text{vega} &= 37.8 \\ \Theta &= -9.80 \\ \rho &= 44.7 \end{aligned}

All five are raw partials — vega per 1.00 of σ\sigma, theta per year, rho per 1.00 of rr — the units the golden file stores and the C++ returns. A desk scales at display: ÷100 per vol point, ÷365 per day, ÷10 000 per basis point. The put at the same strike has Δ=−0.373\Delta = -0.373, the same Γ\Gamma and vega, Θ=−5.05\Theta = -5.05 and ρ=−50.4\rho = -50.4; E says why.

Delta measures the slope of the price against stock and is the hedger's Greek. Gamma measures how quickly delta changes and is the curvature the hedge leaves behind. Vega measures sensitivity to volatility and is the vol trader's Greek. Theta measures the cost of time passing with nothing else moving — the holder's daily bill. Rho measures sensitivity to rates: the rates desk's Greek, and a small one for a one-year equity option.

The five Greeks, base case

S0=K=100S_0 = K = 100, r=0.05r = 0.05, σ=0.4\sigma = 0.4, T=1T = 1, q=0q = 0, from the committed experiments/2_10_greeks.json: results.base_case.{call,put}.{delta,gamma,vega,theta,rho}, results.base_case.{d1,d2,phi_d1}, results.base_case.theta_pieces.{gamma_term,carry_term} and results.base_case.desk_units.{vega_per_vol_point,theta_per_day,rho_per_basis_point}. The two figures after theta's closed form are its two pieces, the gamma term and the carry. The last column is the raw partial divided once, for display.

Greekdefinitioncallputclosed formdesk scaling
Δ\Delta — delta∂V/∂S\partial V/\partial S0.6274−0.3726N(d1)N(d_1); put N(d1)−1N(d_1) - 1—
Γ\Gamma — gamma∂2V/∂S2=∂Δ/∂S\partial^2 V/\partial S^2 = \partial \Delta/\partial S0.0094600.009460φ(d1)/(S0σT)\varphi(d_1)/(S_0 \sigma\sqrt{T}), with φ(d1)=0.3784\varphi(d_1) = 0.3784—
vega∂V/∂σ\partial V/\partial \sigma, per 1.00 of σ\sigma37.84237.842S0φ(d1)TS_0 \varphi(d_1) \sqrt{T}0.378 per vol point (÷100)
Θ\Theta — theta∂V/∂t\partial V/\partial t at fixed SS, per year−9.804−5.048−12σ2S02Γ−rKe−rTN(d2)=−7.568−2.236-\tfrac12\sigma^2 S_0^2 \Gamma - r K e^{-rT} N(d_2) = -7.568 - 2.236−0.0269 per day (÷365)
ρ\rho — rho∂V/∂r\partial V/\partial r, per 1.00 of rr44.718−50.405KTe−rTN(d2)K T e^{-rT} N(d_2); put −KTe−rTN(−d2)-K T e^{-rT} N(-d_2)0.0045 per basis point (÷10 000)

The one-unit moves make the interpretation concrete. A stock move from 100 to 101 changes the option by 0.632, not the 0.627 predicted by delta alone. Delta supplies the first-order term and gamma supplies the missing curvature term 12Γ(δS)2\tfrac12\Gamma(\delta S)^2. A one-vol-point rise adds about 0.378, a one-rate-point rise adds about 0.45 (ρ×0.01=0.447\rho \times 0.01 = 0.447), and one day costs about 0.027.

One-unit moves against the Greeks

Each row moves one input from the base case and reprices with the closed form, from results.moves.{spot_100_to_101,spot_100_to_99,vol_0.40_to_0.41,rate_0.05_to_0.06,one_day_passes}.{actual_change,first_order}, with second_order for the two moves of the stock, and the prices after each move, C_at_101, C_at_99, C_at_0.41, C_at_0.06 and C_at_364_days. "Greek × move" is the first-order estimate, and the last column adds the gamma term to it.

moveCC beforeCC afteractual changeGreek × move+12Γ(δS)2+ \tfrac12\Gamma(\delta S)^2
S0S_0 100 → 10118.023018.6551+0.6321+0.6274+0.6321
S0S_0 100 → 9918.023017.4003−0.6227−0.6274−0.6227
σ\sigma 0.40 → 0.4118.023018.4013+0.3783+0.3784—
rr 0.05 → 0.0618.023018.4726+0.4497+0.4472—
one day passes18.023017.9961−0.0269−0.0269—
D6Black–Scholes pricer and Greeks
price
18.0230
payoff now
0.0000
price − payoff
18.0230
moneyness S/K
1.000
Greekraw ∂as quoted
delta ∂V/∂S0.6274090.6274 per 1.00 of spot
gamma ∂²V/∂S²0.0094600.0095 delta per 1% of spot
vega ∂V/∂σ37.8419830.3784 per vol point
theta ∂V/∂t−9.804296−0.0269 per day
rho ∂V/∂r44.7179950.004472 per basis point
∂V/∂q−62.740946−0.006274 per basis point
Type

The curve meets the payoff only at expiry. Shorten the expiry and watch it fold onto the kink; that folding, per day, is theta. The curve's slope at the marked spot is delta (0.627) and its curvature is gamma; the two are why a hedged book is never quite hedged between rebalances, which is what D5 measures. The price sits 18.02 above the payoff — what the remaining optionality is worth. Every number on this page comes from the same code the library's C++ mirrors, held to the QuantLib golden file to 1e−10.

Bump-and-revalue

A bump Greek replaces differentiation by repricing. Move an input by a small amount hh, revalue the option, and estimate the derivative from the resulting price difference.

Two errors compete. If hh is too large, Taylor terms omitted by the finite-difference formula dominate and the truncation error grows. If hh is too small, subtracting nearly equal prices amplifies floating-point noise. The page therefore states one sensible bump size for each Greek and verifies agreement with the analytic formula to 10−710^{-7}. The systematic study of optimal bump sizes and Monte Carlo Greeks belongs to 3.5.

Bumps against the closed forms

The closed forms against central differences of the same closed form, from results.bump.{h_rel,base.{analytic,bump},grid.{n_points,max_abs_error_delta,max_abs_error_gamma,max_abs_error_vega,max_abs_error_theta,max_abs_error_rho,max_abs_error_dividend_rho}} and results.bump.gamma_bump_study[*].{h_rel,abs_error}, against the tolerance params.bump_tolerance, 10−710^{-7} for every Greek. The grid is 2.9's 216 points of (S0,K,r,q,σ,T)(S_0, K, r, q, \sigma, T). The steps are relative to the input in SS, σ\sigma and TT, and absolute in rr and qq, which may be zero. The last row is gamma's error at six of the eight steps the study tried. The worst is over call and put. The errors at the three smallest steps, and Γ\Gamma's +2×10−10+2 \times 10^{-10} above them, are round-off: they are this run's and differ by machine. The floor at 10−4×S10^{-4} \times S does not.

Greekstep hhanalytic (call)bump − analytic, base caseworst over 216 points
Δ\Delta5×10−5×S5 \times 10^{-5} \times S0.62740946−7×10−10-7 \times 10^{-10}1.1×10−81.1 \times 10^{-8}
Γ\Gamma (second difference)10−4×S10^{-4} \times S0.00946050+2×10−10+2 \times 10^{-10}3.2×10−93.2 \times 10^{-9}
vega5×10−5×σ5 \times 10^{-5} \times \sigma37.841983−1×10−9-1 \times 10^{-9}3.0×10−83.0 \times 10^{-8}
Θ\Theta (in TT, sign flipped)5×10−5×T5 \times 10^{-5} \times T−9.804296−3×10−9-3 \times 10^{-9}5.9×10−95.9 \times 10^{-9}
ρ\rho10−510^{-5}44.717995−2×10−9-2 \times 10^{-9}6.7×10−86.7 \times 10^{-8}
∂V/∂q\partial V/\partial q10−510^{-5}−62.740946−2×10−9-2 \times 10^{-9}5.9×10−85.9 \times 10^{-8}
Γ\Gamma's step, varied10−110^{-1} · 10−210^{-2} · 10−310^{-3} · 10−410^{-4} · 10−510^{-5} · 10−610^{-6} × SSerror 1×10−51 \times 10^{-5} · 9×10−89 \times 10^{-8} · 9×10−109 \times 10^{-10} · 2×10−102 \times 10^{-10} · 1×10−81 \times 10^{-8} · 2×10−62 \times 10^{-6}truncation falls as h2h^2, round-off rises as 1/h21/h^2; the floor is at 10−4×S10^{-4} \times S

Shapes

Delta is a smoothed step from 0 to 1 that sharpens as expiry approaches. Gamma is the bump underneath that step. For long maturities its peak sits below the strike because Γ=φ(d1)/(SσT)\Gamma = \varphi(d_1)/(S\sigma\sqrt{T}): the 1/S1/S factor pulls the maximum to S=Ke−(r+3σ2/2)TS = K e^{-(r + 3\sigma^2/2)T} (from 87.8, where d1=0d_1 = 0, to 74.8 at T=1T = 1). As expiry approaches, that location moves back towards the strike and the peak grows like 1/T1/\sqrt{T}.

Vega is gamma reweighted by S2σTS^2\sigma T. It remains largest near the strike, but unlike gamma it grows with maturity because more time gives volatility more opportunity to affect the terminal distribution. At the money, vega scales like T\sqrt{T} and therefore vanishes at expiry.

Theta is most negative near the strike. Deep in the money it approaches the forward's carry term; deep out of the money it tends towards zero. Near expiry and at the money, gamma and theta blow up like 1/T1/\sqrt{T}, while vega collapses like T\sqrt{T}.

Greeks against spot, four expiries

The call's Greeks against spot at four times to expiry τ\tau, with K=100K = 100, r=0.05r = 0.05, σ=0.4\sigma = 0.4, on a grid of spot from 40 to 200 in steps of 1, from results.shapes_against_S.curves.tau_*.{gamma_peak.{S_on_grid,S_closed_form,gamma_at_peak,gamma_at_strike},vega_peak.{S_on_grid,S_closed_form,vega_at_strike},theta_trough.{S_on_grid,theta_at_trough,theta_at_strike}}. A column headed by a formula is that formula's value; the spots beside them are read off the grid, so they are whole numbers.

τ\tauΓ\Gamma peaks at SS=Ke−(r+3σ2/2)τ= K e^{-(r + 3\sigma^2/2)\tau}Γ\Gamma at the peak · at KKvega peaks at SS=Ke(σ2/2−r)τ= K e^{(\sigma^2/2 - r)\tau}vega at KKΘ\Theta most negative at SSΘ\Theta there · at KK
17574.80.0123 · 0.0095103103.037.84114−10.20 · −9.80
12\tfrac128786.50.0157 · 0.0137102101.527.47107−13.61 · −13.32
14\tfrac149393.00.0210 · 0.0197101100.819.69103−18.35 · −18.14
0.059998.60.0450 · 0.0445100100.28.90101−38.12 · −38.05
Greeks at the money against T

At the money, S0=K=100S_0 = K = 100, r=0.05r = 0.05, σ=0.4\sigma = 0.4, from five of the nine rows of results.shapes_against_T.rows[*].{T,gamma,vega,theta,gamma_sqrtT,vega_over_sqrtT,theta_sqrtT} and the three limits in results.shapes_against_T.limits. The last three columns rescale each Greek by the power of TT it moves with, so each settles on its limit.

at the money, TTΓ\GammavegaΘ\ThetaΓT\Gamma\sqrt{T}vega/T\text{vega}/\sqrt{T}ΘT\Theta\sqrt{T}
20.0063550.76−7.150.0089735.90−10.11
10.0094637.84−9.800.0094637.84−9.80
14\tfrac140.0196919.69−18.140.0098439.37−9.07
0.010.09973.99−82.20.0099739.87−8.22
10−610^{-6}9.970.040−79810.00997439.894−7.981
limit T→0T \to 0∞\infty0−∞-\inftyφ(0)/(S0σ)=0.009974\varphi(0)/(S_0 \sigma) = 0.009974S0φ(0)=39.894S_0 \varphi(0) = 39.894−S0σφ(0)/2=−7.979-S_0 \sigma\varphi(0)/2 = -7.979

Theta pays for gamma

2.7 derived the PDE

Θ+12σ2S2Γ+rSΔ−rV=0.\Theta + \tfrac12\sigma^2 S^2 \Gamma + rS\Delta - rV = 0.

This page reinterprets it as a P&L identity. On each hedge step the option's change is split into a theta term, a delta term and a gamma term. The hedge removes the delta term and the cash leg adds the carry, leaving three named pieces — theta cost, gamma gain and carry — plus a second-order remainder that vanishes with the step.

Summed across a path, the PDE states that what theta costs, net of carry, exactly equals the gamma gain the pricing model assumes will be earned at the implied volatility. The holder therefore pays theta in exchange for owning gamma. When realised volatility equals implied volatility, the gamma gain matches the theta bill net of carry. When realised volatility is higher, gamma earns more than was paid for; when lower, it earns less.

2.7 measured the totals: −3.76 ± 0.02, 0.00 ± 0.003 and +3.74 ± 0.03 at σr=0.3\sigma_r = 0.3, 0.4, 0.5. This page attributes them step by step: at σr=σi\sigma_r = \sigma_i the gamma gain, 7.58 ± 0.05, is the theta cost net of carry, 9.71 − 2.14.

That is what traders mean by "long gamma, short theta": the book benefits from movement in either direction through 12Γ(δS)2\tfrac12\Gamma(\delta S)^2, but pays a steady decay bill while the market sits still.

Hedged P&L attributed, three realised vols

Means ± s.e. over 4,000 exact-GBM paths with μ=0\mu = 0, drawn at 4096 steps and hedged at every one of them, the call priced and hedged at σi=0.4\sigma_i = 0.4, from results.attribution.sigma_r_*.per_N[1].{gamma_gain,theta_cost,carry,gamma_dt,hedged_pnl_discounted}.{mean,se_of_mean} and .{mean_predicted_by_vol_gap,gamma_integral_mean}. Every sum is discounted to time 0. On every path, theta cost + carry + the assumed gain is 0 to 5×10−135 \times 10^{-13} (pde_identity_max_abs), and the three pieces leave a second-order remainder whose rms is 0.013 to 0.018 (remainder.rms). The last column is 2.7's committed result at the same 4096 steps, seed 20261101, recorded in params.cited_from_2_7 and not measured here.

σrealised\sigma_{\text{realised}} (σimplied=0.4)(\sigma_{\text{implied}} = 0.4)gamma gain ∑e−rt12Γ(δS)2\sum e^{-rt} \tfrac12\Gamma(\delta S)^2theta cost ∑e−rtΘ δt\sum e^{-rt} \Theta \,\delta tcarry ∑e−rt(rSΔ−rV) δt\sum e^{-rt} (rS\Delta - rV) \,\delta tthe gain the price assumed, ∑e−rt12σi2S2Γ δt\sum e^{-rt} \tfrac12\sigma_i^2 S^2 \Gamma \,\delta thedged P&L = gain + cost + carry(σr2−σi2)×(\sigma_r^2 - \sigma_i^2) \times the gamma integral2.7's measurement, cited
0.3+4.89 ± 0.03−10.87 ± 0.05+2.18 ± 0.02+8.69 ± 0.04−3.80 ± 0.02−0.07 × 54.3 = −3.80−3.76 ± 0.02
0.4+7.58 ± 0.05−9.71 ± 0.05+2.14 ± 0.02+7.57 ± 0.05+0.003 ± 0.00300.00 ± 0.003
0.5+10.34 ± 0.07−8.69 ± 0.05+2.06 ± 0.02+6.62 ± 0.04+3.72 ± 0.03+0.09 × 41.4 = +3.73+3.74 ± 0.03
D5Delta-hedging simulator

seed 20261101 · 4,000 paths · μ = 0 · σ_r = 0.4 · σ_i = 0.4 · S₀ = K = 100 · r = 0.05

V₀, at σ_i
18.02
fitted slope, N ≥ 4
−0.488
hedged mean, N = 256
−0.01 ± 0.01
hedged sd, N = 256
0.874sd × √N = 14.0
unhedged mean
−3.11 ± 0.47closed form −3.10
unhedged sd
29.6closed form 29.8
discounted hedged mean
−0.01 ± 0.01(σ_r² − σ_i²) × gamma integral: +0.00 × 46.8 = +0.00
forward: max |P&L|
2e−12Δ = 1, N = 256
live sample: hedged mean
+0.01 ± 0.03781 paths, this site’s generator
live sample: hedged sd
0.8870 off the bins
committed bars
N = 2560 past ±20
unhedged, past the axis
5.0 %beyond 60
Source
Drift μ
Realised vol σ_r

The experiment’s first five paths, hedged at N = 256, at μ = 0 and σ_r = σ_i = 0.4, whatever is set above.

path 1: S_T
75.11
path 1: P&L at T
+1.37
path 2: P&L at T
+0.14
path 3: P&L at T
+1.07
path 4: P&L at T
−0.46
path 5: P&L at T
−0.74

The ladder, the bars and the readouts above the live sample are the committed experiment: 4,000 paths drawn once at 4,096 steps and hedged at each N, P&L at T unless marked discounted. The experiment kept its histograms at N = 1, 256 and 4096; the line is a fresh sample on this site’s generator, which is not numpy’s. The fitted slope is −0.488 against a theoretical −½.

The put and the oracle

Put–call parity,

C−P=S0e−qT−Ke−rT,C - P = S_0 e^{-qT} - K e^{-rT},

is an identity in every input, so it can be differentiated. Delta becomes ΔP=ΔC−e−qT\Delta_P = \Delta_C - e^{-qT}, gamma and vega are unchanged, theta gains the forward's carry term, and rho loses the forward's entire rate sensitivity. Nothing is re-derived; the put's Greeks fall directly out of parity.

The put's Greeks through parity

At the base case, where q=0q = 0 and so e−qT=1e^{-qT} = 1, from results.parity_greeks.base_case.{call,put} and results.parity_greeks.base_case.{forward_theta,forward_rho}, with the grid's results.parity_greeks.grid.{n_points,max_abs_error_*}. The JSON records the forward's theta and rho as put minus call, +4.756 and −95.123; the column here is the forward's own, call minus put. The last column is the worst gap from the identity across 2.9's 216 points.

Greekput = call − the forward'scallthe forward'sputworst over 216 points
Δ\DeltaΔC−e−qT\Delta_C - e^{-qT}0.62741−0.37261×10−161 \times 10^{-16}
Γ\Gammaequal0.00946000.0094600
vegaequal37.842037.8420
Θ\ThetaΘC+rKe−rT−qS0e−qT\Theta_C + r K e^{-rT} - q S_0 e^{-qT}−9.804−4.756−5.0483×10−153 \times 10^{-15}
ρ\rhoρC−KTe−rT\rho_C - K T e^{-rT}44.71895.123−50.4053×10−143 \times 10^{-14}

The resulting formulas were checked against both QuantLib and the oracle. All seven columns — price, delta, gamma, vega, theta, rho and dividend-rho — agree across the full 7,234-row dataset and the page's live grid to the stated tolerances. This is the first study page to read the oracle's complete Greek surface rather than price and delta alone.

Price and six Greeks against QuantLib

The closed forms of this page against QuantLib, from the committed experiments/2_10_greeks.json: results.oracle.live.{version,n_points,max_abs_error_*} and results.oracle.golden.{oracle.version,cases,edge_cases,tolerance_rule}. The first row's errors are absolute; the golden rows use the file's own rule, absolute below 1 and relative above it. Both sides are raw partials, the units the golden file stores.

comparison with the oraclerowsworst error: price · Δ\Delta · Γ\Gamma · vega · Θ\Theta · ρ\rho · ∂V/∂q\partial V/\partial qtolerance
QuantLib 1.43 BlackCalculator at the page's grid, call and put216 × 23.6×10−143.6 \times 10^{-14} · 1.1×10−151.1 \times 10^{-15} · 1.2×10−161.2 \times 10^{-16} · 4.6×10−144.6 \times 10^{-14} · 4.9×10−144.9 \times 10^{-14} · 9.9×10−149.9 \times 10^{-14} · 1.1×10−131.1 \times 10^{-13}10−1010^{-10}
golden/bs_vanilla.json grid cases, generated by QuantLib 1.437,2001.2×10−131.2 \times 10^{-13} · 4.3×10−154.3 \times 10^{-15} · 1.4×10−151.4 \times 10^{-15} · 4.5×10−144.5 \times 10^{-14} · 9.8×10−149.8 \times 10^{-14} · 2.5×10−132.5 \times 10^{-13} · 2.8×10−132.8 \times 10^{-13}the file's: 10−1010^{-10}, relative above 1
its edge cases (one-day, near-zero and very high vol, deep in and out of the money, negative rate, ten-year, tiny and large spot)341.7×10−141.7 \times 10^{-14} · 1.0×10−141.0 \times 10^{-14} · 6.2×10−166.2 \times 10^{-16} · 2.1×10−142.1 \times 10^{-14} · 1.6×10−111.6 \times 10^{-11} · 1.4×10−141.4 \times 10^{-14} · 1.3×10−141.3 \times 10^{-14}the file's: 10−810^{-8}

Into 2.11

Everything here assumes a single volatility σ\sigma. The market does not. Options with different strikes and expiries trade at different implied volatilities, even when written on the same stock.

The Greeks survive that transition, but their role changes. They become the language used to describe how prices react as the volatility surface moves, twists and reshapes. 2.11 leaves the world of one σ\sigma and enters the world of volatility smiles, where vanna, volga and the other higher-order risks begin to matter.

Check

Δ = 0.627 here. The stock moves from 100 to 101 and the call price moves by:

With a year to run (S₀ = K = 100, σ = 0.4), gamma is largest:

A hedger is long one call and short Δ shares. The stock moves by δS in either direction, and nothing else happens. The hedged book:

Theta for the one-year call at σ = 0.4 is −9.80 per year at the money. It is most negative:

Vega = S₀φ(d₁)√T = 37.8 per unit of σ here. As expiry T grows, vega at the money:

ρ = ∂C/∂r for the call. In C = S₀N(d₁) − Ke^(−rT)N(d₂), r appears in the discount factor and inside d₁ and d₂. Rho is: