How Value at Risk is calculated

Every VaR calculation starts from two things: the current positions (the live book from trade capture) and market data , prices, forward curves, volatilities, and their historical behaviour. The method is how you turn those into a loss distribution.

Take the actual market moves from a historical window (say the last 250 or 500 trading days), apply each day’s moves to today’s positions, and record the resulting P&L. Sort those simulated P&Ls and read the loss at your percentile, the 5th percentile for a 95% VaR. It assumes little about distributions but is limited by the history you feed it.

Assume returns are (approximately) normal, estimate the volatility of each risk factor and the correlations between them, and compute VaR analytically from the portfolio’s combined volatility. It is fast and closed-form, but the normality assumption understates fat tails.

Generate thousands of random market scenarios consistent with estimated volatilities and correlations, revalue the book under each, and read the loss percentile from the simulated distribution. It is the most flexible, handling optionality and non-linear payoffs, and the most compute-intensive.

Whatever the method, the output is a loss distribution and a single number at a chosen confidence and horizon. The Risk module computes these against live positions, and VaR explained covers what the number does and does not tell you.