Question 1

How is Value at Risk read off a backtest?

Accepted Answer

The historical VaR takes every daily return of the backtest, lines them up from worst to best, and reads the return at the 5% mark. That loss, written as a positive number, is the historical VaR: no assumption is made about the shape of the returns, so whatever the strategy actually lived through — calm years, crashes, fat tails — is in the figure. Its weakness is the sample itself: a backtest of a few hundred days rests its 5% mark on a handful of days, so the figure moves noticeably when the history is short.

Question 2

How is the Gaussian VaR calculated?

Accepted Answer

The Gaussian VaR assumes the daily returns follow a normal distribution with the backtest's own average and volatility, and reads the 5% mark off that curve instead of off the history: $$ 	ext{VaR}_{95\%} = 1.645\,\sigma - \mu $$ where: $\sigma$ is the daily volatility, $\mu$ is the average daily return, and 1.645 is the number of standard deviations that leaves 5% of a normal distribution below it. In words: start from the average day, go 1.645 standard deviations down, and the distance below zero is the Gaussian VaR.

Question 3

Why do the historical and the Gaussian VaR differ?

Accepted Answer

The two agree when the strategy's returns really were close to normal, and part ways when they were not. A normal curve makes very large days rare, and markets produce them more often than it expects, so a historical VaR clearly above the Gaussian one means the strategy's bad days were worse than a normal distribution predicts — its losses have fat tails, and the Gaussian figure understates them. A historical VaR below the Gaussian one is rarer and usually means a short or unusually calm history.

Question 4

How does Fincanva handle it?

Accepted Answer

Daily, at 95%. Every VaR in the app is a one-day loss at the 95% level, written as a positive number: "2.1%" means a loss of 2.1% in one day, and a larger number is a larger loss. Performance Metrics shows both forms in the Worst days · 95% group, as Daily VaR · historical and Daily VaR · Gaussian — see what every number in Performance Metrics means. Contribution analytics shows the Combined's Daily VaR 95% · Gaussian above the table and splits it among the strategies in the VaR 95% contribution column — see contribution analytics. The Stress test sets the Daily VaR 95% of the strategy's historical days beside the same figure under the scenario — see stress test. A simulation computed before these rows existed does not carry them, and shows "n/a" rather than 0%: a 0% loss would claim a safety the figure does not show.

Question 5

What does it look like in practice?

Accepted Answer

A strategy's Performance Metrics read Daily VaR · historical 2.4% and Daily VaR · Gaussian 1.8%. On 95% of its trading days it lost less than 2.4% or gained; on roughly one day in twenty it lost 2.4% or more. The normal-curve estimate of the same threshold is 1.8%, so the strategy's bad days were deeper than its volatility alone suggests — a fat left tail. A second strategy with the same volatility but both figures near 1.8% had bad days that behaved as a normal curve expects.

Question 6

What counts as a good value?

Accepted Answer

Lower is a smaller loss on a bad day, but VaR is only comparable between strategies over the same period: a history that includes a crash has a higher VaR than one that does not, whatever the strategy. Read it beside volatility and the Conditional Value at Risk: VaR marks where the bad days begin, CVaR says how bad they are once there, and neither is a limit on what a single future day can lose.