Validation & Diagnostics - Model Accuracy Testing
Last reviewed: by Options Analysis Suite Research.
Validation & Diagnostics
The Options Analysis Suite includes comprehensive validation and diagnostic tools to ensure numerical accuracy and identify potential pricing errors. These institutional-grade checks run automatically in the background and surface alerts when anomalies are detected.
PDE Residual Tracking
When using PDE methods, the platform monitors discretization errors at multiple grid points to ensure numerical solutions satisfy the Black-Scholes PDE within acceptable tolerances.
- Real-Time Monitoring: Residuals computed at 5-10 representative grid points during each time step
- Residual Formula: |∂V/∂t + 0.5σ²S²∂²V/∂S² + (r-q)S∂V/∂S - rV| < ε
- Tolerance Thresholds: Default ε = 1e-6 for price units, adjustable per user requirements
- Grid Points Tracked: ATM, ±1σ, ±2σ strikes plus boundaries for comprehensive coverage
- Convergence Diagnostics: Residual decay tracked across grid refinement to verify O(Δt²) + O(Δx²) convergence
Call-Put Parity Verification
The fundamental arbitrage relationship C - P = S - K·exp(-rT) must hold for European options. Violations indicate numerical errors, data issues, or model misspecification.
- Automatic Checks: Parity validated across 20+ strike/volatility combinations for each pricing run
- Tolerance Gates: |C - P - (S - K·exp(-rT))| < $0.01 for typical parameters
- American Options: Relaxed bounds account for early exercise premium: C - P ≤ S - K·exp(-rT) + ε
- Diagnostic Output: Maximum parity violation reported in model diagnostics panel
- Cross-Model Validation: Compare parity errors across models to identify model-specific issues
Greeks Monotonicity and Sign Checks
Vanilla options satisfy theoretical properties (e.g., call delta ∈ [0,1], gamma ≥ 0, vega ≥ 0 for long positions), and a vanilla violation indicates numerical instability or an edge case. Exotics can break them legitimately: a long at-the-money up-and-out call with its barrier 10% above spot has negative delta, gamma and vega (-0.0102, -0.0006 and -0.016 per share in the API reference's barrier example), so a validator warning on such a model flags a value to read, not an error.
- Delta Bounds: Vanilla call delta ∈ [0,1], put delta ∈ [-1,0], with warnings for violations
- Gamma Positivity: Γ ≥ 0 for long vanilla options, validated at all grid points
- Theta Sign: not enforced: Θ is negative for most long calls and puts but positive for a deep in-the-money European put (and a call under a high dividend yield), so a positive value is not by itself an error
- Vega Positivity: ν ≥ 0 for long vanilla positions (a barrier or other exotic can be negative)
- Convexity Checks: Γ ≥ 0 for long vanilla options; Vomma takes either sign (negative near the money, where d1 and d2 straddle zero), so it is not sign-checked
- ATM Special Cases: ATM options (S ≈ K) have Delta ≈ 0.5, maximum Gamma, maximum Vega
American Exercise Boundary Validation
For American options, the optimal exercise boundary must satisfy smooth paste and continuity conditions.
- Boundary Extraction: Identify S*(t) where V(S*,t) = intrinsic value at each time step
- Smooth Paste: Verify ∂V/∂S continuous across exercise boundary (delta matching)
- Monotonicity: Exercise boundary S*(t) must be monotonic in time (non-decreasing for calls with q < r)
- Early Exercise Premium: American price - European price ≥ 0 with equality for deep OTM
- Dividend Effects: High-dividend stocks show larger early exercise regions for puts
Monte Carlo Convergence Diagnostics
- Standard Error Tracking: Real-time σ/√N displayed alongside price estimates
- Confidence Intervals: 95% CI = [Price - 1.96σ/√N, Price + 1.96σ/√N]
- Convergence Rate: Plot error vs. path count to verify O(N^(-1/2)) convergence
- Variance Reduction Effectiveness: Compare raw vs. variance-reduced estimator standard errors
- Histogram Analysis: Terminal price distributions visualized to detect fat tails, skewness, jumps
Cross-Model Consistency Checks
Compare pricing across models to identify potential calibration issues or model misspecification.
- Black-Scholes Baseline: All models should converge to BS in the limit of no jumps, constant vol
- ATM Consistency: Different models should produce similar ATM prices when volatility matched
- Greeks Ranking: Model-specific Greeks ordered consistently (e.g., Jump Diffusion Vega > BS Vega)
- Arbitrage Bounds: All prices respect no-arbitrage bounds: S - K ≤ Call ≤ S, K·exp(-rT) - S ≤ Put ≤ K·exp(-rT)
Data Quality Alerts
- Implied Volatility Failures: Alert when IV solver fails to converge (illiquid strikes, stale quotes)
- Negative Prices: Flag any model producing negative option values (numerical overflow)
- Extreme Greeks: Warn when Greeks exceed reasonable bounds (e.g., |Vega| > 100 for typical equity options)
- Stale Data Detection: Market data timestamps checked against current time
Performance Benchmarks
- Timing Diagnostics: Model computation time displayed for performance tuning
- GPU vs CPU Comparison: Speed ratios reported when WebGPU active
- Greeks Overhead: Incremental cost of computing Greeks beyond base price
- Calibration Duration: Time required for automated model parameter fitting
This page is part of the Options Analysis Suite documentation hub. Browse the glossary for term definitions.