Value at risk (VaR) explained
Value at risk (VaR) estimates the most you'd expect to lose over a set period at a given confidence level. A one-day 95% VaR of $1,000 means: on a typical day, your loss shouldn't exceed $1,000 — but on the worst 5% of days, it can be worse, and VaR doesn't tell you how much worse. It's a useful summary of routine risk and a dangerous one if mistaken for a worst case.
What VaR actually says
value at risk"1-day 95% VaR = $1,000"
# Read as:
on ~95% of days, loss ≤ $1,000
on ~5% of days, loss could be MORE # VaR is silent on how much
VaR has three ingredients: a time horizon (one day, ten days), a confidence level (95%, 99%) and a loss amount. Change any one and the number changes.
Three ways to compute it
- Historical — rank your strategy's past daily returns and read off the 5th percentile. Simple, makes no distribution assumption, handles fat tails — but assumes the future resembles the past.
- Parametric (variance-covariance) — assume returns are normal; VaR ≈ mean − z × standard deviation. Fast, but understates tail risk because real returns are fatter-tailed than normal.
- Monte Carlo — simulate thousands of return paths and read the percentile. Flexible but only as good as the model you feed it. See our Monte Carlo guide.
VaR tells you a threshold, not the depth of the tail beyond it. In 2008, losses blew through 99% VaR estimates repeatedly because the normal model badly underestimates extreme moves. Conditional VaR (expected shortfall) fixes this by averaging the losses that occur beyond the VaR point — which is why regulators increasingly prefer it.
Using VaR with a trading bot
- Compute a simple historical VaR on your strategy's daily returns as a routine-risk monitor.
- Never treat it as a worst case — pair it with maximum drawdown for the tail.
- Combine with risk of ruin to judge survivability, not just a typical bad day.
- Re-estimate regularly; VaR drifts as volatility regimes change.
The U.S. SEC and the Basel banking framework both use VaR-style measures, while acknowledging the tail limitation; Investopedia's value at risk overview covers the same three methods.
Frequently asked questions
What is value at risk (VaR)?
Value at risk estimates the most you would expect to lose over a set period at a chosen confidence level. A one-day 95% VaR of $1,000 means that on 95% of days your loss should not exceed $1,000 — but on the worst 5% of days it could be more. It is a single-number summary of routine downside risk.
How is VaR calculated?
Three common methods: the historical method (rank your actual past returns and read off the chosen percentile), the parametric method (assume a normal distribution and use mean and standard deviation), and Monte Carlo simulation (generate many random return paths). Each has trade-offs; historical and Monte Carlo handle fat tails better than the normal-distribution assumption.
What is the biggest weakness of VaR?
VaR tells you a threshold but not how bad things get beyond it — it is silent about the size of losses in the tail. The 2008 crisis exposed this: 'once-in-a-lifetime' breaches happened repeatedly because markets have fatter tails than the normal model assumes. Conditional VaR (expected shortfall) addresses this by averaging the losses beyond the VaR point.
Is VaR useful for retail traders and bots?
Yes, as a sanity check on routine downside, but treat it as a floor on bad days, not a worst case. Pair it with maximum drawdown and risk of ruin, and never assume the VaR number is a hard ceiling — real markets regularly exceed it. For a trading bot, a simple historical VaR on your strategy's daily returns is a reasonable monitor.