Work
Individual ProjectDerivatives · Numerical Methods · Risk

Derivatives Pricing & Risk Engine

Different pricing methods should converge on the same economics when their assumptions align. The project tests that agreement, then explores what changes when volatility and hedging become more realistic.

Central findingClosed-form, tree and simulation methods reconcile closely on shared benchmarks, while variance reduction and more frequent hedging materially improve numerical precision and risk control.

Evidence

The result in context

Black–Scholes pricing error
<3e−5
Put–call parity error
0
Control-variate standard-error reduction
61.75%
Heston calibration RMSE
3.12e−05
On this page

Question

Can several pricing methods be implemented, cross-checked and used to explain how option value responds to market inputs?

A pricing and risk system spanning closed-form options, trees, simulation, stochastic volatility, calibration and hedging.

An illustrative volatility surface shows implied volatility changing across strike and maturity; the project uses synthetic option data rather than a live market feed.

System

One engine, several views of option value

Black–Scholes provides an analytical benchmark; a Cox–Ross–Rubinstein tree makes early-exercise logic and convergence visible; Monte Carlo supports flexible simulation; Heston introduces stochastic volatility.

Greeks measure how an option’s value responds to changes in the market or model inputs. The explorer below calculates price, Delta, Gamma and Vega locally from the tested Black–Scholes implementation.

Local calculation

Option sensitivity explorer

Greeks measure how an option’s value responds to market and model inputs.

Call price
£10.45
Put price
£5.57
Call Delta
0.6368
Put Delta
-0.3632
Gamma
0.0188
Vega · per 1 vol point
0.3752
Call value as spot price changesA sensitivity curve calculated from the current strike, volatility, maturity and rate inputs.

Call value across 65%–135% of the selected spot.

Black–Scholes assumes constant volatility and frictionless trading. The calculation is an explanatory benchmark, not a market quote.

Validation

Methods are checked against each other

Black–Scholes pricing error is below 3e−5 with zero put–call parity error. A 1,000-step tree comes within approximately 0.002 of the analytical benchmark.

Cross-method reconciliation is useful because a plausible-looking option price is not, by itself, evidence that an implementation is correct.

Interactive evidence

Monte Carlo error contracts as the path count increases

Absolute pricing error versus simulation paths; logarithmic path axis.

Read: At 50,000 paths the absolute error falls to £0.0141; sampling noise means the 100,000-path point is not mechanically the minimum.

Data table · 5 verified rows
Complete dataset
PathsEstimateAbsolute ErrorStandard ErrorCi LowerCi Upper
1,00010.9688690.5182850.48553810.01723211.920505
5,00010.6338410.1832570.21314710.2160811.051602
10,00010.3186530.1319310.14622510.03205610.605249
50,00010.4646570.0140730.06605210.33519710.594116
100,00010.4799670.0293840.04657310.38868510.571249

Simulation

Variance reduction turns computation into precision

Antithetic sampling reduces standard error by 29.16%; a control variate reduces it by 61.75%. The objective is not a more impressive simulation count, but a tighter estimate for the same computational budget.

Risk

Hedging frequency leaves a measurable residual

Daily rebalancing produces hedge-error standard deviation of 0.4257 versus 1.9104 for monthly rebalancing in the tested setup. The comparison makes discrete hedging risk tangible without implying frictionless implementation.

More frequent rebalancing reduces hedge-error dispersion in an idealised Black–Scholes benchmark with zero transaction costs.

Limitations

What this evidence does not establish

  • Black–Scholes Greeks inherit the model’s assumptions, including constant volatility and frictionless trading.
  • Heston calibration uses a deterministic synthetic option chain; calibration quality is conditional on that surface and objective function.
  • The hedge comparison is an idealised Black–Scholes benchmark with zero transaction costs and a shared random seed, not live-trading evidence.

Source and reproducibility

Trace the evidence

Source code, evaluation outputs and supporting material are available in the repository.

View repository
  1. Volatility surfacedocs/images/dashboard_volatility_surface.pngCommit / evidence ID: 60c6c85a25df2291dba13ee6de53652d6a711d19
  2. Numerical convergence resultsreports/numerical_convergence_results.csvCommit / evidence ID: 60c6c85a25df2291dba13ee6de53652d6a711d19
  3. Delta-hedging simulator artefactdocs/images/dashboard_delta_hedging_simulator.pngCommit / evidence ID: 60c6c85a25df2291dba13ee6de53652d6a711d19