computational finance
What is the fair price today of the right to buy a stock for 100 dollars three months from now? Nobody knows where the stock will be, so the price hinges on the whole cloud of possible futures and their probabilities. Computational finance is the use of numerical methods to answer questions like this — pricing options and other derivatives, measuring risk, and optimizing portfolios — when no simple formula will do.
Two big computational engines dominate. The first is the partial-differential-equation route: under standard assumptions the fair price of an option obeys the Black-Scholes PDE, a diffusion-like equation in price and time. You solve it numerically with the same finite-difference machinery used for the heat equation — set up a grid in stock price and time, and march backward from the known payoff at expiry to today's price, usually with an implicit scheme like Crank-Nicolson for stability. The second engine is Monte Carlo simulation: simulate thousands or millions of random price paths drawn from the assumed model, compute the option's payoff along each, and average them; by the law of large numbers that average converges to the price. Monte Carlo shines for complex, path-dependent, or high-dimensional products (many underlying assets) where a PDE grid would suffer the curse of dimensionality.
These methods price the derivatives traded in trillions daily, compute risk measures like value-at-risk, and stress-test portfolios. The honesty here is sharp and was written in blood during financial crises: the numbers are only as sound as the model behind them. Black-Scholes assumes a tidy random walk with constant volatility — markets jump, crash, and have fat tails it ignores. Monte Carlo error falls only like 1 over the square root of the number of paths, so a hundred times more computation buys just ten times more accuracy, which is why variance-reduction tricks matter. A precise price from a wrong model is precisely wrong.
To price a 'basket' option that pays off based on the average of five stocks, a PDE would need a five-dimensional grid — far too large. Instead a quant runs a Monte Carlo: simulate one million joint price paths for the five stocks, compute the payoff on each, and average. The standard error shrinks like 1 over the square root of a million, giving a price to about three digits.
Monte Carlo prices a five-asset option where a PDE grid would hit the curse of dimensionality.
A precise price is not a correct one. Black-Scholes assumes constant volatility and no jumps — real markets have fat tails and crashes — so model risk, not arithmetic, is usually the dominant error in finance.