Compare Two Equity Risk Models

# Compare Two Equity Risk Models Two candidate equity risk models provide factor exposures, factor covariance, and specific risk. Design an out-of-sample evaluation that determines which model is more useful for portfolio construction. Include predicted-versus-realized volatility, residual correlation, stability, coverage, and behavior across market regimes. ### Constraints & Assumptions - Model versions are reconstructed using only information available at each historical prediction date. - Realized risk is noisy and must be measured over a future horizon. - The production portfolios may differ from broad benchmark portfolios. - A model can look accurate in aggregate while failing for important sectors, styles, or stress periods. ### Clarifying Questions to Ask - Which portfolios, holding horizons, and risk decisions will use the model? - What minimum coverage and latency are required? - Is the goal absolute risk, active risk, marginal risk, or all three? - How costly are turnover, unstable exposures, and systematic residuals? ### What a Strong Answer Covers - Point-in-time rolling or walk-forward evaluation with representative portfolios - Calibration, ranking, and loss metrics for predicted versus realized risk - Residual covariance and unintended factor structure - Exposure, covariance, coverage, and turnover stability - Regime analysis, statistical uncertainty, and economic impact ### Follow-up Questions 1. How would you compare models when realized covariance is itself estimated noisily? 2. What does systematic correlation among residuals reveal? 3. When might a slightly less accurate but more stable model be preferable? ```hint Evaluate the decision, not only the covariance matrix Backtest risk forecasts on portfolios resembling production and measure whether the model improves sizing, constraints, and realized drawdown control. ```