value-at-risk

# Value at Risk (VaR) **Canonical Source**: JP Morgan RiskMetrics (1989-1994) **Practitioner**: Sir Dennis Weatherstone (JP Morgan Chairman) **Domain**: Finance, Risk Management, Decision Analysis **Introduced**: 1989 (internal), 1994 (public RiskMetrics) ## One-Line Summary Statistical measure estimating the maximum potential loss in portfolio value over a defined time period at a given confidence level under normal market conditions. ## Core Concept Value at Risk (VaR) answers: "What is the most I can lose with X% confidence over Y time period?" It translates complex portfolio risk into a single, interpretable number that executives and regulators can use for decision-making and capital allocation. **The Innovation**: Before VaR, firms lacked a unified risk metric. Sir Dennis Weatherstone demanded a daily one-page report showing firm-wide risk exposure. JP Morgan's response became RiskMetrics—the industry standard for market risk measurement. ## When to Use **Ideal Scenarios**: - Setting risk limits for trading desks or investment portfolios - Determining capital reserves required to cover potential losses - Comparing risk across different asset classes or strategies - Regulatory compliance (Basel Accords require VaR reporting) - Communicating risk to non-technical executives - Stress-testing portfolios under normal market volatility **Not Suitable For**: - Black swan events (VaR assumes normal market conditions) - Illiquid assets with unreliable pricing data - When tail risk beyond VaR threshold is the primary concern - Very short time horizons (<1 day) with high-frequency data gaps ## Execution Steps ### 1. Define VaR Parameters - **Confidence level**: Typically 95% or 99% (higher = more conservative) - **Time horizon**: 1 day, 10 days, 1 month (match to liquidity/holding period) - **Currency**: Portfolio denomination currency *Output*: VaR specification (e.g., "95% 1-day VaR in USD") ### 2. Gather Historical Data - Collect price/return data for all portfolio positions - Minimum 1-2 years of daily data (250-500 observations) - Ensure data quality (adjust for splits, dividends, corporate actions) - Include correlations between assets *Output*: Clean historical return series ### 3. Calculate Portfolio Returns Distribution - **Historical Simulation**: Use actual past returns - **Variance-Covariance**: Assume normal distribution, calculate portfolio volatility - **Monte Carlo**: Generate thousands of scenarios based on statistical properties *Output*: Distribution of possible portfolio value changes ### 4. Identify VaR Threshold - Sort potential outcomes from worst to best - Find the loss value at the (100 - confidence level) percentile - For 95% VaR with 250 days of data: 13th worst outcome - For 99% VaR: 3rd worst outcome *Output*: VaR amount (e.g., "$2.5M at 95% confidence") ### 5. Interpret and Communicate - "We are 95% confident we will not lose more than $2.5M tomorrow" - Or inversely: "There is a 5% chance we could lose more than $2.5M" - Report alongside portfolio value for context (VaR as % of total) *Output*: Executive-ready risk summary ### 6. Backtest and Validate - Track actual daily losses vs. VaR predictions - Count "VaR breaches" (actual loss > VaR estimate) - Expected breach rate should match confidence level (5% for 95% VaR) - If breaches >> expected, recalibrate model *Output*: Model validation report, adjustments to methodology ### 7. Update Daily/Regularly - Recalculate VaR as portfolio composition changes - Update historical data window (rolling window approach) - Monitor for regime changes (volatility spikes, market structure shifts) *Output*: Current VaR estimate for active risk management ## Common Pitfalls **"VaR Covers Everything" Fallacy** VaR does NOT predict losses during market crashes. It measures normal volatility. 2008 showed portfolios losing 10x their VaR in a single day. **Solution**: Supplement with stress testing, scenario analysis, and Conditional VaR (CVaR/Expected Shortfall). **Model Risk Blind Spots** Different VaR methods (historical vs. parametric vs. Monte Carlo) can produce wildly different results for the same portfolio. **Solution**: Calculate VaR using multiple methods, understand assumptions, report ranges. **Correlation Breakdown** VaR models assume historical correlations persist. In crises, correlations spike toward 1.0 (everything falls together). **Solution**: Stress-test correlation assumptions, use conservative estimates during volatile periods. **Data Quality Issues** Garbage in = garbage out. Stale prices, missing data, or survivorship bias corrupt VaR estimates. **Solution**: Rigorous data validation, mark-to-market verification, independent pricing sources. ## Key Insights **Regulatory Standardization**: Basel II/III require banks to hold capital equal to 10-day 99% VaR. This standardization made VaR the lingua franca of financial risk. **The 5% You Ignore**: A 95% VaR says nothing about the OTHER 5% of outcomes. That tail could be -$3M or -$300M. Always ask "What happens when VaR is breached?" **JP Morgan's Open-Source Gambit**: By releasing RiskMetrics data/methodology for free in 1994, JP Morgan shaped industry standards and positioned themselves as thought leaders. The methodology eventually became a $1.55B acquisition (MSCI bought RiskMetrics Group in 2010). **VaR ≠ Maximum Loss**: Common misinterpretation. VaR is the threshold, not a ceiling. Actual losses can far exceed VaR during tail events. ## Real-World Application **Trading Desk Limits**: A prop trading desk might set a $500K daily VaR limit. If VaR exceeds this, they must reduce positions before market close. **Bank Capital Requirements**: Under Basel III, a bank calculates 10-day 99% VaR across its trading book. Regulators require capital reserves of at least 3x this VaR amount. **Pension Fund Monitoring**: A pension fund uses VaR to ensure they can meet liabilities. If 1-month 95% VaR exceeds 5% of fund value, they trigger risk mitigation protocols. ## Related Frameworks - **Conditional VaR (CVaR/Expected Shortfall)**: Measures average loss BEYOND VaR threshold (captures tail risk) - **Stress Testing**: Scenario-based analysis for extreme events VaR doesn't cover - **Monte Carlo Simulation**: Primary method for calculating VaR with complex portfolios - **Sharpe Ratio**: Risk-adjusted return metric (uses volatility, not VaR) - **Maximum Drawdown**: Worst peak-to-trough decline (historical measure, not probabilistic) ## Anti-Patterns **VaR as the Only Risk Metric** Relying solely on VaR creates blind spots to tail risk, liquidity risk, and operational risk. **Overfitting to Recent History** Using only 1-3 months of data makes VaR hypersensitive to recent volatility spikes. **Ignoring Non-Normal Distributions** Assuming returns are normally distributed when they have fat tails underestimates extreme risk. ## Score Justification **Framework Assessment**: 42/50 (Tier 1 - Canonical) - **Practitioner Weight (9/10)**: Created by JP Morgan for internal risk management, became industry standard. Proven in production across every major financial institution. - **Clarity & Executability (8/10)**: Clear mathematical definition, multiple calculation methods. However, technical expertise required for proper implementation. - **Proven ROI (8/10)**: Prevented countless over-leveraged positions. Regulatory adoption proves value. But 2008 crisis showed limitations. - **Novelty (7/10)**: Revolutionary in 1989 by distilling complex portfolio risk to a single number. Now ubiquitous in finance. - **Cross-Domain Applicability (10/10)**: Used in banking, insurance, asset management, corporate treasury, energy trading, and project risk management. **Notable**: RiskMetrics became a $1.55B company (acquired by MSCI). VaR is mandated by Basel Accords for global banks. The JP Morgan technical document from 1994 became the de facto industry reference.