Home / Wiki / Value at Risk

Risk & AnalyticsReviewed 2026-07-21

Value at Risk

A statistical estimate of the loss a portfolio is not expected to exceed over a given time horizon at a given confidence level, under normal market conditions.

Definition

Value at Risk (VaR) estimates the loss threshold a portfolio is not expected to exceed over a specific time horizon, at a specific confidence level, under normal market conditions. A "1-day 95% VaR of $500,000" means the portfolio is expected to lose no more than $500,000 over one day on 95% of days — equivalently, a loss larger than $500,000 is expected on roughly 1 day in 20.

VaR is deliberately silent about what happens in the remaining 5% (or 1%) of cases — it is a threshold, not a worst case. It also says nothing about *how* the loss unfolds, only its magnitude at the stated confidence level. Both of these are common criticisms of VaR: it can understate tail risk in fat-tailed markets like crypto, and two portfolios with identical VaR can have very different loss profiles beyond that threshold.

The two most common approaches are parametric VaR (assumes returns follow a known distribution, typically normal, and derives the loss threshold from volatility and a z-score) and historical simulation (reprices the current book through actual past days without assuming any distribution). Parametric VaR is fast to compute and easy to scale across horizons; historical simulation captures real fat tails and skew at the cost of being anchored to whatever history happens to be available.

Why it matters

VaR gives managers and LPs a single, comparable number for "how much risk is this book carrying" without requiring either party to read a full position list and reason through correlations themselves. It is the most widely quoted risk metric in institutional due diligence precisely because it compresses a portfolio's risk into one figure at one horizon.

For crypto funds specifically, VaR needs to be read with more caution than in equities: crypto return distributions have fatter tails and higher realized correlation spikes during stress than a normal-distribution assumption implies, so a parametric VaR can meaningfully understate the loss actually experienced in a genuine crisis — which is exactly why VaR is typically paired with stress testing rather than used alone.

Parametric VaR (1-day, scaled to h-day)

VaR(c%, 1d) = z(c) × σ(1d) × Portfolio Value;  VaR(c%, hd) = VaR(c%, 1d) × √h

z(c)

The z-score for confidence level c — 1.645 for 95%, 2.326 for 99% (one-tailed normal)

σ(1d)

The portfolio's estimated 1-day return volatility (standard deviation), as a decimal

Portfolio Value

The current NAV or dollar value being measured for risk

√h

The square-root-of-time scaling factor from a 1-day horizon to an h-day horizon

The √h scaling ("square-root-of-time rule") assumes returns are independent and identically distributed day to day — it is a convenience, not a law, and breaks down when volatility clusters (a common feature of crypto markets, where a volatile day tends to be followed by another volatile day).

A parametric VaR at two confidence levels and two horizons

A fund holds a NAV of $10,000,000 with an estimated 1-day portfolio return volatility of 2.0%. Using the standard one-tailed z-scores, the 95% 1-day VaR is 1.645 × 0.02 × $10,000,000 = $329,000, and the 99% 1-day VaR is 2.326 × 0.02 × $10,000,000 = $465,200.

Scaling the 95% figure to a 10-day horizon with the square-root-of-time rule: $329,000 × √10 ≈ $329,000 × 3.16228 ≈ $1,040,389. In plain terms: this portfolio is estimated to lose no more than about $329,000 on 95% of individual days, and no more than about $1.04 million over 95% of any given 10-day stretch — with the important caveat that the remaining 5% of cases, and any single extreme event, are exactly what this number does not describe.

Confidence	Horizon	z-score	VaR
95%	1-day	1.645	$329,000
99%	1-day	2.326	$465,200
95%	10-day	1.645 (×√10)	≈ $1,040,389

Common mistakes

Treating VaR as a worst-case loss — it explicitly excludes the tail beyond its confidence level, which is precisely where the largest losses occur.
Scaling VaR across horizons with √h without checking whether volatility clustering makes that independence assumption unreasonable — crypto volatility tends to cluster, which the square-root-of-time rule does not account for.
Using a parametric (normal-distribution) VaR on assets with known fat tails or jump risk without cross-checking against historical simulation or stress testing — a normal assumption systematically understates tail risk for crypto.
Comparing VaR figures across funds or periods without checking they use the same confidence level, horizon, and methodology — a 99% 10-day VaR and a 95% 1-day VaR are not comparable numbers even when they look similar in size.
Reporting VaR in isolation, with no accompanying stress test or drawdown history — VaR describes routine risk under normal conditions and is a poor substitute for scenario analysis of genuine tail events.

In practice

Crypto portfolios routinely see realized volatility several multiples of equity-market norms, and correlations between assets that are usually loosely correlated can spike toward 1 during a broad risk-off event (a correlation regime shift) — both of these erode the reliability of a parametric VaR built on calmer historical data, which is why VaR figures for crypto books are best read as a floor on routine risk, not a ceiling on possible loss.

Nyx Fund's risk engine computes parametric VaR and Expected Shortfall at 95% and 99% confidence, at 1-day and 10-day horizons, using sample or exponentially-weighted covariance over log returns of the fund's actual live positions — and pairs it with a Stress Lab of historical crisis scenarios precisely because VaR alone does not capture true tail risk.

The free Fund Health Check does not output a VaR figure, but its gross leverage and cash-buffer checks flag the same exposure a VaR model would price as risky before you run any statistics on it.

Try it free →

Questions, answered

What does a 95% 1-day VaR of $500,000 actually mean?

It means the portfolio is expected to lose no more than $500,000 over one day on roughly 95% of days. On the remaining 5% of days, a larger loss can occur, and VaR says nothing about how much larger it might be.

Why do funds use both 95% and 99% VaR?

The two confidence levels give a sense of how quickly the risk profile deteriorates further into the tail. If 95% and 99% VaR are close together, losses beyond the routine threshold grow slowly; a large gap suggests the tail is fatter than a normal-distribution model assumes.

Is VaR a good measure of risk for a crypto fund?

It is a useful, standard measure but an incomplete one for crypto. Crypto returns have fatter tails and more volatility clustering than the normal-distribution assumption behind parametric VaR typically captures, so VaR for crypto books is usually paired with historical simulation and stress testing rather than used alone.

How is VaR scaled from a 1-day to a 10-day horizon?

The common approach multiplies the 1-day VaR by the square root of the number of days (the "square-root-of-time rule"), which assumes returns are independent day to day. It is a convenient approximation, not an exact result, and is less reliable when volatility clusters.