Volatility Studio

AMERICAN OPTIONS / RESEARCH STUDIO

Every smile. A clearer view.

Calibrate your market. Explore the fit. Understand the risk.

MARKET DATA

Market data

CSV INPUT

Choose a single smile or a multi-expiration sample. Uploaded CSVs may contain up to 128 expirations. Calibration follows the selected model maturity treatment.

Market sample CSV ↓

View & edit source CSV
CSV format & column requirements

CSV up to 16 MB / 100,000 rows; one underlying and valuation date per run, up to 128 expirations. Required columns: underlying, valuation_date, expiry (YYYY-MM-DD), spot, strike, option_type (Call/Put), market_price. Optional: currency (USD), exercise_style (American). Set dividend yield and discount factor using the controls below; defaults are 0 and 1. Provide 5–80 distinct strikes spanning spot per expiration. Prices must have positive time value.

HistoricalData.net · sample source ↗
StrikeOptionMarket price

Dividend & discount curves Optional
Dividend curve · optional
Sample dividend curve ↓
View & edit curve CSV

Upload a dated dividend curve or choose a constant. Defaults to 0.

Discount curve · optional
Sample discount curve ↓
View & edit curve CSV

Upload a dated discount curve or choose a constant. Defaults to 1.

Choose Constant or CSV curve for each input. Uploading a file selects CSV curve automatically. American pricing uses the full dated dividend and discount curves throughout the exercise period. Dividend yields interpolate linearly; discount factors interpolate log-linearly.

CALIBRATION

Model & calibration

PRICING SETUP
Model
Model & calibration method

Default interpolation

Market volatility: natural cubic splines across strikes; linear interpolation of total variance (volatility² × time) between expirations. Before the first expiration, the first smile is held constant. A single-expiration smile is held constant through time.

Grid preparation: missing strike nodes are filled by linear interpolation of variance in log-strike, using the nearest value outside the quoted range.

Carry curves: dividend yields interpolate linearly; discount factors interpolate log-linearly from a valuation-date factor of 1.

Calibrated Dupire surface: local variance interpolates linearly across spot nodes and stays constant between calibration dates.

American quotes: American implied volatilities and European-equivalent prices are derived before constructing the market surface.

Selects which option sides are used for calibration, pricing and plotting.

Pricing engine
About this pricing engine

BUILD / VALIDATE

Pricing run

Connecting to the pricing engine…

ANALYSIS / MODEL OUTPUT

Calibration overview

READY

Upload and validate a CSV to begin.

CALIBRATION QUOTES—
CALIBRATED EXPIRATIONS—
PRICE RMSE · SELECTED EXPIRY—

Option prices

Market quotes & model fit

Filters prices and Greeks. All expirations stay calibrated.

Call quotesPut quotesCall modelPut model

Calibrate and price to see option prices.

Dots: market quotes. Dashed curves: model prices. Only the displayed expiration and selected option types are shown.

Prices by strike

Selected expiration
StrikeOptionCSV market priceModel priceDifferenceBid / askMC standard errorEuropean reference

RISK / SENSITIVITIES

Option Greeks

READY

Choose an expiration to view its Greeks.

Available for every supported model. Each shock reprices the saved fitted state for every expiration. The expiration selectors filter the saved results. Values are frozen-model Greeks.

Sensitivities & bump sizes

Every shock reuses the fitted parameters and dated diffusion/leverage coefficients without market recalibration. Black vega adds fitted volatility points; other models scale equity diffusion by 1 + bump / base ATM volatility, keeping jump and funding laws fixed. American exercise policies train again for each shock.

A full calculation reprices twelve scenarios. Monte Carlo uses wider default bumps to reduce policy noise; check convergence with explicit settings.

Complete calibration and pricing to calculate sensitivities.

CallsPuts

No Greeks calculated.

StrikeOptionDeltaGammaVega / vol ptTheta / dayRho / bpDividend rho / bpDiagnostics

Values are per option in the CSV price units. Monte Carlo noise, exercise decisions and calibration changes affect the differences. Missing values mean a scenario was unavailable.

Scenario calibration diagnostics

Each shock uses native American exercise and independent training/pricing samples. Random blocks stay aligned across date rolls. Black uses an exact terminal-payoff control with a simulated American premium; other models use Brownian bridges and their existing controls. Exercise decisions respect valid model holding bounds. MC delta and gamma use the full spot bump; the half bump checks stability. Exercise-boundary, calibration and convergence warnings remain visible. No European put–call parity is imposed.

Calibration details
Calibration & exercise method

Regression Monte Carlo uses the same valuation paths for this comparison. Black uses its exact European terminal value as a control; its American exercise premium remains simulated. Other deterministic-carry models use terminal-stock control. PDE and terminal-only pricing use a separate same-model calculation. MC standard error describes the continuation estimate before the valuation-date intrinsic floor and excludes exercise-policy and calibration bias.

Black–Scholes fits one American PDE volatility per contract. Shared models use approximate de-Americanization before native smile calibration; the final residuals always compare against original American quotes.

No European put–call parity is applied to American prices. Under deterministic nonnegative rates and nonpositive dividend yield, calls have no optimal early exercise and use their terminal payoff directly. Other contracts use in-the-money continuation regression with independent training and valuation draws and a quadratic basis in normalized spot. With stochastic volatility or funding it omits variance, rate and borrow factors from the exercise regression, which may cause policy bias. Sampling error does not measure that bias. A finite-parameter shared model need not fit every quote exactly.