Skip to content
QuantCalc Tools
Directory

Parametric Var Expected Shortfall

Garman-Klass & Yang-Zhang OHLC Volatility: Efficient VaR Input Math Guide

Close-to-close volatility treats an intraday +/-4% V-reversal ending unchanged as zero variance; Yang-Zhang OHLC estimators achieve up to 14x higher statistical efficiency while capturing overnight gaps.

By Elena Rostova, FRM, CQFReview not recorded
Garman-Klass & Yang-Zhang OHLC Volatility: Efficient VaR Input Math Guide — QuantCalc Tools 2026 quantitative & supervisory data card
QuantCalc Tools (Parametric VaR, Expected Shortfall & Fat-Tail Risk Calibration): Parkinson, Garman-Klass, Rogers-Satchell, and Yang-Zhang 14x efficient VaR volatility.Source: QuantCalc Tools Institutional Research DeskLicense: CC-BY-4.0

Suppose an equity index opens at $5,000$, plunges $-4.2\%$ intraday to $4,790$ on a macro headline, rallies $+4.4\%$ after central-bank reassurance, and closes at $5,002$: a traditional close-to-close volatility estimator records $r_t = +0.04\%$, misclassifying a violent intraday liquidation whipsaw as the calmest day of the year! Replacing close-to-close variance with Open-High-Low-Close (OHLC) range-based volatility estimators—specifically Parkinson (1980), Garman-Klass (1980), Rogers-Satchell (1991), and Yang-Zhang (2000)—slashes estimation error variance by up to $14\times$ and feeds dramatically cleaner $\sigma$ inputs into our Value at Risk (VaR) calculator and Black-Scholes-Merton implied volatility comparisons. Review how volatility estimators integrate into our Parametric VaR & Expected Shortfall calibration guide.

Exact Mathematical Hierarchy of OHLC Volatility Estimators#

Define normalized daily log prices relative to yesterday's close $C{t-1}$ and today's open $Ot$:

  • Overnight jump log return: $ot = \ln(Ot / C_{t-1})$
  • Open-to-close intraday log return: $ct = \ln(Ct / O_t)$
  • Normalized intraday log High and Low: $ut = \ln(Ht / Ot) \ge 0$ and $dt = \ln(Lt / Ot) \le 0$
  1. Parkinson High-Low Range ($5.2\times$ Efficiency, Zero Drift, No Overnight Jumps):

$$\hat{\sigma}{P}^2 = \frac{1}{4T\ln 2}\sum{t=1}^T \big(\ln(Ht / Lt)\big)^2 \approx \frac{0.36067}{T}\sum{t=1}^T (ut - d_t)^2$$

  1. Garman-Klass ($7.4\times$ Efficiency, Zero Drift, Continuous Trading):

$$\hat{\sigma}{GK}^2 = \frac{1}{T}\sum{t=1}^T \left[\frac{1}{2}(ut - dt)^2 - (2\ln 2 - 1)\,c_t^2\right], \qquad (2\ln 2 - 1 \approx 0.38629)$$

  1. Rogers-Satchell ($6.0\times$ Efficiency, Drift-Independent for Trending Markets!):

$$\hat{\sigma}{RS}^2 = \frac{1}{T}\sum{t=1}^T \big[ut(ut - ct) + dt(dt - ct)\big]$$

  1. Yang-Zhang ($14\times$ Max Efficiency, Unbiased Under BOTH Drift AND Overnight Gaps!):

$$\hat{\sigma}{YZ}^2 = \hat{\sigma}{o}^2 + k\,\hat{\sigma}{c}^2 + (1 - k)\,\hat{\sigma}{RS}^2, \qquad k = \frac{0.34}{1.34 + \frac{T + 1}{T - 1}}$$ where $\hat{\sigma}o^2$ is the sample variance of overnight jumps $ot$ and $\hat{\sigma}c^2$ is the sample variance of open-to-close returns $ct$.

Worked Overnight-Gap OHLC Volatility Audit: Close-to-Close vs Yang-Zhang#

Formula or model from the articleNot independently verified
Single-Session Whipsaw & 20-Day VaR Input Audit ($20,000,000 NASDAQ-100 Futures Desk)
  • Yesterday Close C{t-1} = 20,000 | Today Open Ot = 19,600 (Overnight Gap o_t = -2.0203%)
  • Today High Ht = 20,190 (ut = +2.9659%) | Today Low Lt = 19,115 (dt = -2.5056%) | Close C_t = 19,980
  • Close-to-Close Return: ln(19,980 / 20,000) = -0.1001% | Open-to-Close c_t = +1.9202%
  • 1) Single-Day Squared Variance Contribution Comparison:
  • Close-to-Close r_t^2 ............. (-0.001001)^2 = 0.00000100 -> Implied Daily Vol = 0.10% (BLIND!)
  • Parkinson Range: 0.36067 * (0.054715)^2 ......... = 0.00107978 -> Implied Daily Vol = 3.29%
  • Garman-Klass (No overnight gap):
  • 0.5(0.054715)^2 - 0.38629(0.019202)^2 ....... = 0.00135443 -> Implied Daily Vol = 3.68%
  • Rogers-Satchell Intraday Drift-Free Term:
  • 0.029659(0.010457) + (-0.025056)(-0.044258) . = 0.00141907 -> Intraday Vol = 3.77%
  • Yang-Zhang Full 24h Variance (Overnight ot^2 + 0.164*ct^2 + 0.836*RS):
  • 0.00040816 + 0.164(0.0003687) + 0.836(0.0014191) = 0.0016550 -> Full Daily Vol = 4.07%!
  • 2) Impact on 20-Day Rolling 99% Parametric VaR ($20M Notional, z = 2.32635):
  • 20d Close-to-Close Vol = 1.18% => VaR_99 = $548,995
  • 20d Yang-Zhang Vol = 1.69% => VaR_99 = $786,306 (+43.23% hidden gap + whipsaw protection!)

That single trading day illustrates why close-to-close VaR fails during earnings seasons and FOMC weeks: while close-to-close sees $-0.10\%$, Parkinson misses the $-2.02\%$ overnight gap, Garman-Klass is biased upward by nonzero trend drift, and Yang-Zhang cleanly captures both the $-2.02\%$ overnight jump AND the $5.47\%$ intraday range ($4.07\%$ realized daily volatility)!

Yang-Zhang OHLC range-volatility filtration pipeline feeding EWMA and GARCH parametric VaR engines
Yang-Zhang OHLC range-volatility filtration pipeline feeding EWMA and GARCH parametric VaR enginesVerified Exhibit

Microstructure Discretization Bias: When NOT to Use Raw High-Low Range#

Range estimators rest on continuous-time Brownian path extremes where expected High-Low range scales continuously. Two real-world microstructure effects require care on institutional execution desks (such as platforms running MT5 or FIX API):

  1. Discrete Tick / Sparse Trade Downward Bias: On illiquid small-cap equities or off-the-run corporate bonds trading only 30 times a day, observed $\max(P_i)$ sits strictly below the continuous Brownian maximum, biasing Parkinson and Garman-Klass downward by 10% to 25%!
  2. Bid-Ask Bounce / Bad Print Upward Bias: Conversely, if an retail CFD feed prints synthetic wide bid-ask spikes at the 17:00 New York rollover, a single bad $Ht$ or $Lt$ tick inflates $(ut - dt)^2$ quadratically! Always compute OHLC bars strictly from mid-quotes during active session hours before running Yang-Zhang.

Statistical Efficiency & Drift/Gap Robustness of OHLC Estimators#

Quantitative Distribution · Relative Efficiency vs Close-to-Close
Relative Efficiency vs Close-to-CloseIdeal Asset Class for VaR
01Close-to-Close Sample SD$1.0\times$ (Ba…02Parkinson (1980) High-Low$5.2\times$03Garman-Klass (1980)$7.4\times$04Rogers-Satchell (1991)$6.0\times$05Yang-Zhang (2000)$8\times$ to $1…
Volatility EstimatorRelative Efficiency vs Close-to-CloseHandles Strong Drift $\mu \ne 0$?Handles Overnight Jumps?Ideal Asset Class for VaR
Close-to-Close Sample SD$1.0\times$ (Baseline)Yes (Demeaned)YesIlliquid daily NAV funds / bonds
Parkinson (1980) High-Low$5.2\times$Biased upwardNo (Misses $Ot/C{t-1}$)Spot Gold / 24h FX (low drift)
Garman-Klass (1980)$7.4\times$Biased upwardNo (GK-YZ adds $o_t^2$)24/5 G10 FX & Crypto perpetuals
Rogers-Satchell (1991)$6.0\times$Yes (Exact for all $\mu$)No (Intraday session only)Intraday desk stop-loss VaR
Yang-Zhang (2000)$8\times$ to $14\times$Yes (Exact for all $\mu$)Yes (Unbiased $o_t$ + intraday)Cash Equities, ETFs & Futures
Effective sample size equivalent of 20 trading days across Close-to-Close, Parkinson, Rogers-Satchell, Garman-Klass, and Yang-Zhang
Effective sample size equivalent of 20 trading days across Close-to-Close, Parkinson, Rogers-Satchell, Garman-Klass, and Yang-ZhangVerified Exhibit

Feeding Range Variance Into CARR / Range-GARCH Models#

Because a 20-day Yang-Zhang window achieves the statistical precision of 170+ days of close-to-close returns, risk officers can shorten lookback windows to react $8\times$ faster to regime changes without suffering noisy parameter jitter! Better still, Chou's (2005) Conditional Autoregressive Range (CARR) and Brandt-Jones Range-EGARCH feed daily Garman-Klass-Yang-Zhang single-day proxy $vt^{GKYZ} = ot^2 + \frac{1}{2}(ut - dt)^2 - (2\ln 2 - 1)c_t^2$ directly into the innovation slot of EWMA lambda 0.94 or GARCH(1,1)! Compare intraday stop-out hazards with Christoffersen Markov breach clustering and explore the Parametric VaR & Expected Shortfall Topic Hub.

Frequently asked questions

Why does Parkinson or Garman-Klass underestimate cash equity volatility?

Parkinson and original 1980 Garman-Klass assume continuous 24-hour Brownian trading without market closes. On NYSE or LSE equities where 30% to 45% of total variance arrives between 16:00 close and 09:30 open, omitting overnight jumps understates VaR.

What is the Garman-Klass-Yang-Zhang (GKYZ) single-day variance proxy?

Adding squared overnight log return ln(O_t / C_{t-1})^2 directly onto each day’s Garman-Klass range term creates a non-negative daily variance proxy v_t that can be smoothed via RiskMetrics EWMA (lambda = 0.94) on a day-by-day basis.

Why does Rogers-Satchell remain unbiased during strong bull or bear market trends?

Parkinson and Garman-Klass assume zero drift (mu = 0); a steady +0.3%/day drift stretches High-Low range even at low volatility, biasing them upward. Rogers-Satchell cross-multiplies u(u - c) + d(d - c), canceling drift algebraically for any mu.

Can a single day’s Garman-Klass variance term ever evaluate to a negative number?

Never! Because H_t >= max(O_t, C_t) and L_t <= min(O_t, C_t), u_t - d_t >= |c_t|, so 0.5*(u_t - d_t)^2 - 0.38629*c_t^2 >= (0.50 - 0.38629)*c_t^2 = 0.1137*c_t^2 >= 0 for every valid OHLC bar.

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). Garman-Klass & Yang-Zhang OHLC Volatility: Efficient VaR Input Math Guide. QuantCalc Tools. https://quantcalctools.com/guides/intraday-garman-klass-yang-zhang-range-volatility-var-inputs
Explore Parametric Var Expected Shortfall Hub (20) →

QuantCalc Tools Data, Benchmarks & Contributor Records

Garman-Klass & Yang-Zhang OHLC Volatility VaR Formulas | QuantCalc Tools