Historical Simulation VaR
Historical Simulation VaR (HS-VaR) is a non-parametric method for estimating Value at Risk that calculates the potential loss of a portfolio by applying historical return scenarios — drawn from an actual historical time series of asset price changes — to the current portfolio, then reading off the loss at a specified confidence level from the resulting empirical P&L distribution. Unlike parametric VaR, it makes no distributional assumptions about returns.
Key takeaways
- Historical simulation avoids the normality assumption of parametric VaR, naturally capturing fat tails, skewness, and non-linear payoffs present in historical data.
- The method is computationally intensive for large portfolios because it requires repricing all instruments using each historical scenario.
- The quality of HS-VaR estimates depends critically on the length and relevance of the historical window; a window that does not include relevant stress events understates tail risk.
- Weighted historical simulation variants (e.g., filtered historical simulation or age-weighted scenarios) address the stale-data problem by giving more weight to recent observations.
- HS-VaR can understate risk during novel stress scenarios not present in the historical window — a significant weakness during unprecedented events like the COVID-19 pandemic.
Explanation
Historical Simulation VaR became the dominant industry methodology for VaR estimation in the 2000s and 2010s, largely displacing the earlier variance-covariance (parametric) approach because of its ability to capture non-normal return distributions without requiring explicit specification of distributional parameters. The method was mandated for internal models-based VaR under the Basel II Market Risk framework and remains widely used under Basel III/IV.
The methodology proceeds in several steps. First, a historical window of daily (or other frequency) percentage changes in market risk factors is assembled — typically 250 trading days (one year) for the standard Basel window, though risk managers often use 500-1,000 days or more to capture a wider range of market conditions. Second, each historical scenario is applied to the current portfolio: if day t had a 2% decline in the S&P 500, a 10 basis point rise in 10-year yields, and a 1% depreciation of the EUR, these changes are applied to the current portfolio holdings to compute a hypothetical one-day P&L for that scenario. Third, after computing P&L for each historical scenario (250 or more), the results are sorted from worst to best. The VaR at a 99% confidence level is the 2.5th worst scenario (for 250 scenarios) or, more precisely, the loss at the (1 − confidence level) × n percentile of the empirical distribution.
Historical simulation's non-parametric nature is its greatest advantage and simultaneously its greatest limitation. The advantage is the automatic capture of all market phenomena present in the historical data: volatility clustering, fat tails, skewness, cross-asset correlations, and non-linear payoffs from options. No assumptions are needed about the shape of the return distribution. The limitation is that the method can only generate scenarios that have historically occurred; it is incapable of producing scenarios for risks that are genuinely novel. A portfolio of pandemic-sensitive equities managed through 2019 with a 500-day historical window would have had no COVID-19 related scenarios in its VaR calculation, making the estimate meaningless for that specific tail risk.
Regulatory implementations of HS-VaR have evolved in response to observed failures. During the 2007-2009 financial crisis, many institutions' historical windows were dominated by 2003-2006 data — an unusually low-volatility period — resulting in severely underestimated VaR. The Basel 2.5 reforms required institutions to use a stressed VaR measure computed over a one-year historical window that includes a period of significant financial stress (specifically, roughly the period of August 2007 through December 2008 for most institutions). The Fundamental Review of the Trading Book (FRTB), implemented under Basel IV, replaces VaR with Expected Shortfall (ES) at the 97.5% confidence level over a stressed observation period, addressing some of HS-VaR's known deficiencies.
Formula
HS-VaR_{α} = -P&L_{(1-α)·n}, where scenarios are sorted from worst to best and α is the confidence level; ES = E[Loss | Loss > VaR]
Example
A risk manager at a hedge fund calculates 1-day 99% Historical Simulation VaR for a $500 million long/short equity portfolio using 500 trading days of historical data (approximately 2 years). The calculation yields 500 historical P&L scenarios. After sorting, the 5th worst scenario (500 × 1% = 5 scenarios in the left tail) shows a P&L of -$18.5 million. The HS-VaR is therefore $18.5 million, interpreted as: there is a 99% probability that the portfolio will not lose more than $18.5 million in a single trading day, based on the range of market outcomes observed over the past 2 years. During COVID-19 volatility in March 2020, the actual daily loss reached $42 million on several days, exceeding the HS-VaR by 2.3x — a VaR breach that correctly triggered review of the historical window's coverage of stress scenarios.
Related terms
Basel Iii Basel Iv Basis Covariance Drawdown Equity Expected Shortfall Fat Tails Financial Crisis Hedge Fund Marginal Var Market Risk