Skip to content
QuantCalc Tools
Directory

Parametric Var Expected Shortfall

Lognormal vs Arithmetic Parametric VaR: Multi-Month Horizon Math Guide

Linear arithmetic VaR reports absurd 116% losses on unlevered 50%-volatility positions over 1-year horizons; exact lognormal transformation bounds losses below 100% while accounting for Ito drift.

By Elena Rostova, FRM, CQFReview not recorded
Lognormal vs Arithmetic Parametric VaR: Multi-Month Horizon Math Guide — QuantCalc Tools 2026 quantitative & supervisory data card
QuantCalc Tools (Parametric VaR, Expected Shortfall & Fat-Tail Risk Calibration): Ito -0.5*sigma^2 drift, <100% long wealth bounds, and exponential short-squeeze VaR.Source: QuantCalc Tools Institutional Research DeskLicense: CC-BY-4.0

Plug an unlevered $$10,000,000$ long position in a $55\%$-annual-volatility biotechnology or digital-asset basket into a naive 1-year linear arithmetic 99% VaR formula ($VaR = W \cdot z{0.99} \cdot \sigmaa$), and the spreadsheet outputs $VaR_{99} = \$10\text{M} \times 2.32635 \times 0.55 = \$12,794,925$—a 127.95% loss on an unlevered cash asset that cannot legally drop below $$0$! That embarrassing mathematical impossibility arises whenever risk analysts conflate continuously compounded log returns $r{log} = \ln(ST / S0)$ with discrete percentage wealth changes $R{simple} = (ST - S0)/S0 = e^{r{log}} - 1$. Compare 1-day linear approximations in our Value at Risk (VaR) calculator, inspect long-horizon compounding in our CAGR calculator, and review our Parametric VaR & Expected Shortfall calibration guide.

Exact Geometric Brownian Motion Lognormal VaR & Expected Shortfall#

Suppose asset price $St$ follows Geometric Brownian Motion $dSt / St = \mu\,dt + \sigma\,dWt$, where $\mu$ is annualized arithmetic drift and $\sigma$ is annualized log-return volatility. By Itô's Lemma, terminal log return over horizon $T$ (in years) is normally distributed with variance-drained geometric mean $(\mu - \frac{1}{2}\sigma^2)T$: $$\ln\!\left(\frac{ST}{S0}\right) \sim \mathcal{N}\!\Big(\big(\mu - \tfrac{1}{2}\sigma^2\big)T,\;\; \sigma^2 T\Big)$$ Exponentiating the left-tail quantile $q{log} = (\mu - \frac{1}{2}\sigma^2)T - z\alpha\sigma\sqrt{T}$ yields Exact Lognormal Relative VaR (strictly bounded inside $[0, W)$ for all $\sigma, T < \infty$), and integrating the lognormal lower tail yields Exact Closed-Form Lognormal Expected Shortfall:

$$VaR{\alpha}^{log} = W\left[1 - \exp\!\Big(\big(\mu - \tfrac{1}{2}\sigma^2\big)T - z\alpha \sigma\sqrt{T}\Big)\right]$$

$$ES{\alpha}^{log} = W\left[1 - \frac{e^{\mu T}}{1 - \alpha}\,\Phi\!\left(-z\alpha - \sigma\sqrt{T}\right)\right]$$

Worked 1-Year Distressed-Vol VaR Audit: Linear >100% Bug vs Exact Lognormal#

Formula or model from the articleNot independently verified
Worked 1-Year 99% VaR & 97.5% ES: Linear Arithmetic vs Exact Lognormal ($10,000,000 Unlevered Position)
  • Inputs: W = $10,000,000, Horizon T = 1.00 Year, Annualized Volatility sigma = 55.0% (0.55)
  • Arithmetic Expected Return mu = +12.0% => Geometric Log Drift = 0.12 - 0.5*(0.55)^2 = -0.03125 (-3.125%!)
  • 1) Naive Linear Arithmetic 99% VaR (Broken Formula: W * (z_0.99 * sigma - mu)):
  • VaR_99(linear) = $10,000,000 * (2.32635 * 0.55 - 0.12) = $11,594,925 (115.95% LOSS — IMPOSSIBLE!)
  • Zero-Mean Linear 99% VaR = $12,794,925 (127.95% LOSS — IMPOSSIBLE!)
  • 2) Exact Lognormal 99.0% VaR (z_0.99 = 2.32635):
  • Log Quantile q_0.01 = -0.03125 - 2.32635 * 0.55 = -0.03125 - 1.27949 = -1.31074
  • Terminal Wealth Ratio ST / S0 = exp(-1.31074) = 0.26962 (Asset falls to 26.96 cents on the dollar)
  • Exact Lognormal VaR_99 = $10,000,000 * (1 - 0.26962) = $7,303,799 (73.04% Loss — Physically Valid!)
  • 3) Exact Lognormal 97.5% Expected Shortfall (z_0.975 = 1.95996, 1 - alpha = 0.025):
  • Shifted Normal Argument = -1.95996 - 0.55 = -2.50996 => Phi(-2.50996) = 0.0060373
  • ES_97.5(lognormal) = $10,000,000 * [ 1 - (exp(0.12) / 0.025) * 0.0060373 ]
  • = $10,000,000 * [ 1 - 45.0999 * 0.0060373 ] = $7,277,185 (72.77% Tail Mean Loss!)

Look at the beauty of that closed-form Lognormal ES equation $W[1 - \frac{e^{\mu T}}{1 - \alpha}\Phi(-z\alpha - \sigma\sqrt{T})]$: notice it uses the exact same Girsanov measure-change shift ($-z\alpha - \sigma\sqrt{T}$) as $d1$ vs $d2$ inside the Black-Scholes-Merton pricing formula! And notice how $-\frac{1}{2}\sigma^2 = -15.125\%$ variance drag flipped a $+12.0\%$ arithmetic expected return into a $-3.125\%$ median geometric log drift!

Decision and conversion workflow between arithmetic 1-day VaR and exact Ito lognormal multi-horizon VaR
Decision and conversion workflow between arithmetic 1-day VaR and exact Ito lognormal multi-horizon VaRVerified Exhibit

The Flip Side: Linear VaR Massively UNDERSTATES Short-Position Squeeze Risk!#

While linear arithmetic VaR overstates long-only losses ($1 - e^{-x} < x$ for $x > 0$), Taylor concavity flips in the exact opposite direction for short sellers ($e^{+x} - 1 > x$)! Suppose a hedge fund shorts $$10\text{M}$ of that same $55\%$-volatility biotech stock ($\mu = 0$):

  • Linear 1-Year 99% Short VaR: $2.32635 \times 55\% = \mathbf{127.95\%}$ ($$12.79\text{M}$ loss).
  • Exact Lognormal 1-Year 99% Short VaR: $\exp(-0.5(0.55)^2 + 2.32635(0.55)) - 1 = \exp(1.1282) - 1 = \mathbf{209.02\%}$ ($$20.90\text{M}$ loss — $$8.11\text{M}$ larger than linear VaR!)!

Arithmetic Linear VaR vs Exact Lognormal VaR Across Volatility Horizons#

Quantitative Distribution · $x = z\sigma\sqrt{T}$
$x = z\sigma\sqrt{T}$Naive Linear VaR ($x$)
011-Day G10 FX ($\sigma_a = 1…$1.465\%$0210-Day Equity Index ($\sigm…$10.228\%$0360-Day HY Credit / Oil ($\s…$43.249\%$041-Year Biotech / Crypto ($\…$127.949\%$051-Year Venture / Token ($\s…$209.372\%$
Horizon & Volatility Scenario ($z_{0.99} = 2.32635, \mu = 0$)$x = z\sigma\sqrt{T}$Naive Linear VaR ($x$)Exact Long Lognormal VaR ($1 - e^{-0.5\sigma^2 T - x}$)Exact Short Lognormal VaR ($e^{-0.5\sigma^2 T + x} - 1$)
1-Day G10 FX ($\sigma_a = 10\%$)$1.465\%$$1.47\%$$1.46\%$ ($-0.7\%$ diff)$1.48\%$ ($+0.7\%$ diff)
10-Day Equity Index ($\sigma_a = 22\%$)$10.228\%$$10.23\%$$9.81\%$ ($-4.1\%$ diff)$10.65\%$ ($+4.1\%$ diff)
60-Day HY Credit / Oil ($\sigma_a = 38\%$)$43.249\%$$43.25\%$$36.22\%$ ($-16.3\%$ diff)$51.49\%$ ($+19.1\%$ diff)
1-Year Biotech / Crypto ($\sigma_a = 55\%$)$127.949\%$$127.95\%$ (Impossible)$76.09\%$ (Bounded $<100\%$)$209.02\%$ ($+63.4\%$ squeeze!)
1-Year Venture / Token ($\sigma_a = 90\%$)$209.372\%$$209.37\%$ (Impossible)$91.78\%$ (Bounded $<100\%$)$441.25\%$ ($+110.8\%$ squeeze!)
Divergence between linear arithmetic VaR, exact long lognormal VaR, and exact short-seller lognormal VaR at 55% volatility
Divergence between linear arithmetic VaR, exact long lognormal VaR, and exact short-seller lognormal VaR at 55% volatilityVerified Exhibit

Portfolio Caveat: Weighted Sum of Exponentials Is Not Exact Lognormal#

One final mathematical nuance for multi-asset risk engines: while log returns add linearly across time ($r{0 \to T} = \sum{t=1}^T rt$), simple arithmetic returns add linearly across portfolio weights ($RP = \sum{i=1}^N wi (e^{r_i} - 1)$). Over 60-to-250-day horizons (such as 120-day buckets in Basel FRTB liquidity horizon bucketing or Solvency II 1-year insurer SCR models), approximate portfolio sums via Fenton-Wilkinson moment matching or simulate correlated log returns directly inside Cholesky covariance Monte Carlo VaR. Explore the full cluster on the Parametric VaR & Expected Shortfall Topic Hub.

Frequently asked questions

When is it safe to use simple arithmetic VaR instead of lognormal VaR?

On 1-day horizons for assets with daily volatility below 1.5% (so z_0.99 * sigma <= 3.5%), 1 - exp(-0.035) = 3.44% sits within 1.7% of 3.50%. Once holding periods reach 10+ days or annualized volatility exceeds 30%, switch to exact lognormal formulas.

Why must -0.5 * sigma^2 * T be subtracted inside the lognormal VaR exponent?

By Ito’s Lemma (and Jensen’s inequality E[e^X] = exp(m + 0.5*v)), if an asset’s arithmetic expected return is mu, its continuously compounded log-return mean is mu - 0.5*sigma^2. Omitting -0.5*sigma^2 overstates median wealth and understates long-horizon VaR.

Why does linear VaR dangerously underestimate risk for short positions?

Long losses are bounded at -100% as S_T -> 0, whereas short losses grow exponentially without bound as S_T = S_0 * exp(r_log) rallies. Replacing exp(+x) - 1 with linear +x drops all positive quadratic and cubic compounding terms.

What is the difference between Absolute Lognormal VaR and Mean-Relative Lognormal VaR?

Absolute VaR measures shortfall relative to today’s initial wealth W_0 (loss = W_0 - W_T). Mean-Relative VaR measures shortfall relative to expected terminal wealth E[W_T] = W_0 * exp(mu*T), canceling drift mu*T in Solvency II-style economic capital.

Publication record

Published
14 November 2026
Updated
14 November 2026
Author
Elena Rostova, FRM, CQF
Review
Review not recorded
Automated fact check
Passed in the stored pipeline record
Source URLs
3 recorded

AI assistance recorded: anthropic/claude-opus-4.6. This field does not establish human review or source verification.

Academic & Institutional Attribution

Cite This Research & Embed Live Data Card

Open Attribution License (CC BY 4.0)

Standard academic & institutional research paper citation format.

Elena Rostova, FRM, CQF. (2026, November 14). Lognormal vs Arithmetic Parametric VaR: Multi-Month Horizon Math Guide. QuantCalc Tools. https://quantcalctools.com/guides/lognormal-vs-arithmetic-return-var-large-drawdown-horizon-error
Explore Parametric Var Expected Shortfall Hub (20) →

QuantCalc Tools Data, Benchmarks & Contributor Records

Lognormal vs Arithmetic Parametric VaR Formula Guide | QuantCalc Tools