Value-at-Risk (VaR) answers: 'what is the most I'd expect to lose with X% confidence over a Y-day horizon?' A 1-day, 95% VaR of $10,000 means there's roughly a 5% probability of losing more than $10k tomorrow under historical conditions.
Three common methods to compute it: historical (look at the worst N% of your actual past daily returns), parametric (assume normal distribution, scale by standard deviation), and Monte Carlo (simulate many possible paths). Historical is the most honest for traders with enough sample; parametric is the fastest but breaks badly during regime changes.
Institutional desks compute VaR every evening on the entire book and report it to the risk committee. Position size limits, leverage caps, and pre-trade checks are usually framed in VaR terms. If you've ever wondered why a bank lets some traders run a billion-dollar book and others a million-dollar book, the difference is usually their VaR allocation rather than the headline notional.
The famous critique: VaR tells you nothing about *how bad* the bad days are beyond the threshold. A 95% VaR ignores the worst 5% entirely — and in 2008, the worst 5% included the actual blowups. Conditional VaR (CVaR) or 'expected shortfall' fixes this by averaging the losses *beyond* the VaR threshold, which is the metric most post-2008 risk committees actually rely on. VaR is the headline number; CVaR is the one that keeps people up at night.