Value at Risk (VaR 95%)

What is Value at Risk (VaR 95%)?

Value at Risk answers a deceptively simple question: how bad is a bad period, where "bad" is defined by a confidence level you choose. At 95% confidence over a monthly horizon, the VaR is the return level that the strategy underperforms in only the worst 5% of months. If a strategy's VaR 95% is negative 7 percent, that means roughly 1 month in 20 was worse than a 7 percent loss, and the other 19 were better. It is a single number that puts a floor on routine downside, the kind of loss you should plan to see periodically rather than the once-a-decade catastrophe.

VaR became a standard risk measure after JPMorgan published its RiskMetrics methodology in 1994, and it remains embedded in bank regulation, fund risk reports, and portfolio dashboards. It is defined by two parameters: the confidence level (95% here, though 99% is also common) and the horizon (one month in this context). Both matter. A 99% VaR is deeper than a 95% VaR because it reaches further into the tail, and a one-year VaR is far larger than a one-month VaR because losses compound over longer windows.

VaR is best understood as the edge of normal, not the worst case. It marks the boundary of the routine 95% of outcomes but says nothing about what happens beyond it. Two strategies with identical VaR 95% can have completely different tail behavior past the threshold, one capping its bad months near the VaR level and another occasionally plunging far below it. That blind spot is why VaR is almost always read alongside Conditional VaR, which averages the losses inside the worst 5%.

Formula

There are two standard ways to compute VaR, and they can disagree meaningfully. The historical method makes no assumption about the shape of returns: it simply sorts every observed periodic return from worst to best and reads off the value at the 5th percentile. The parametric method assumes returns are normally distributed and computes the threshold as the mean minus 1.645 standard deviations, where 1.645 is the point on the standard normal curve below which 5% of the mass falls. The parametric form is cleaner but unreliable for real return series, which have fatter tails than a normal distribution and therefore experience extreme losses more often than the bell curve predicts.

SledgeKey reports VaR 95% using the historical method on the strategy's monthly returns over the test window: it ranks the monthly returns and reports the 5th-percentile value as a monthly figure. Using the empirical distribution rather than the Gaussian assumption means the number reflects the actual fat tails in the data instead of pretending returns are normal. The figure is expressed as a monthly return, so a VaR 95% of negative 8 percent reads directly as "1 month in 20 was worse than down 8 percent."

Why VaR 95% Matters in Backtesting

VaR informs position sizing and expectation setting. Volatility tells you the average dispersion of returns, but most people do not think in standard deviations; they think in "how bad can a typical bad month get." VaR 95% answers exactly that, in the units returns are already quoted in. If you know that 1 month in 20 is likely to be worse than a given loss, you can size the position so that such a month is survivable rather than account-ending, and you can warn yourself ahead of time so the loss does not feel like a surprise when it arrives.

The failure mode of ignoring VaR is mistaking a smooth average for a calm experience. A strategy with an attractive annual return and moderate volatility can still have a left tail that produces a punishing month roughly twice a year. Reading VaR 95% surfaces that routine downside before you commit capital, and it makes strategies comparable on downside terms: two with the same Sharpe ratio but very different VaR 95% differ in how fat their left tail is.

VaR is most useful as the first half of a pair. On its own it marks the edge of the worst 5% but is silent about the severity beyond it. Read next to Conditional VaR (the average loss inside that worst 5%) and Maximum Drawdown (the worst compounded peak-to-trough event), VaR becomes one coordinate in a fuller map of downside rather than a single misleading reassurance.

How SledgeKey Implements VaR 95%

VaR 95% appears on the backtest results page as a single monthly figure within the risk metrics, computed from the same monthly returns the rest of the results use. It is the historical 5th percentile: the monthly returns are ranked from worst to best, and the value at the 5th percentile is reported. Because it is empirical rather than parametric, it reflects the real shape of the return series, including any fat-tailed months, instead of assuming a normal distribution.

The benchmark's VaR 95% is computed the same way over the identical window, so the comparison is the useful reading. A strategy with a shallower (less negative) VaR than the benchmark put you through milder routine bad months than the index; a deeper VaR means larger periodic hits. Because the figure is a monthly return at the 5th percentile, it is most stable over long windows; over a short one, only a handful of observations sit in the worst 5% and the estimate becomes noisy. It measures routine downside, not the worst case, so read it together with the maximum drawdown and the worst single month.

Common Pitfalls

The first pitfall is reading VaR as a worst-case loss. It is not. VaR 95% is the boundary of the routine 95% of outcomes, and it says nothing about how bad the worst 5% can get. A strategy can show a comfortable VaR 95% of negative 6 percent and still have suffered a negative 30 percent month that lives beyond the threshold. The famous critique of VaR, sharpened during the 2008 crisis, is exactly this: institutions reported reassuring VaR figures while the losses that actually hurt them came from the tail VaR ignores by construction. Always pair VaR with Conditional VaR to see inside that tail.

The second pitfall is confusing the two computation methods. A parametric VaR that assumes normal returns will almost always look milder than the historical VaR for the same data, because real returns have fatter tails than the bell curve. If you compare a VaR from one source against a VaR from another without checking the method, you may conclude one strategy is safer when the difference is purely the assumption. Confirm both numbers use the same method and the same horizon before drawing conclusions.

The third pitfall is horizon and confidence-level confusion. A monthly VaR 95% is a completely different number from an annual VaR 95% or a monthly VaR 99%. Quoting "the VaR" without stating both the horizon and the confidence level is meaningless. Scaling a one-month VaR to one year by multiplying by the square root of 12 assumes independent, identically distributed returns, an approximation that breaks down precisely in the stressed markets where you care about the answer most.

Related Concepts