Class ComponentVar

java.lang.Object
com.quantfinlib.risk.ComponentVar
public final class ComponentVar extends Object
COMPONENT VaR — the answer to the risk committee's actual question. A portfolio VaR of $10m is a fact; "which desk owns how much of it?" is the decision. Under the delta-normal model the Euler allocation makes the split exact and additive: 
   sigma_p     = sqrt(w' Σ w)
   marginal_i  = (Σw)_i / sigma_p          d sigma_p / d w_i
   component_i = w_i · marginal_i            and Σ_i component_i = sigma_p  (exact)

Scaled by the same z-score, component VaRs SUM EXACTLY to portfolio VaR — no "diversification residual" bucket to argue over. The three numbers answer three different questions:

  • component VaR — how much of today's risk this position owns (the risk-budget number);
  • marginal VaR — how fast VaR moves per unit added (the "should the next dollar go here?" number);
  • incremental VaR — how much VaR disappears if the position is CLOSED entirely (a full re-computation without it — NOT component VaR, and the difference is the whole point: a large position that hedges the book has POSITIVE size, NEGATIVE component, and closing it RAISES VaR).

Delta-normal only, stated: allocations inherit the model's assumptions (linear positions, normal returns). For the portfolio-level numbers under fatter models, see VarEngine; the Euler split of a Monte Carlo ES is a different (kernel-estimation) exercise this class does not pretend to do. Sign convention: VaR is reported positive; components carry sign (a hedge's component is negative). Research lane, deterministic.

  • Method Details

    • allocate

      public static ComponentVar.Allocation allocate(double[] weights, double[][] covariance, double confidence)
      Euler allocation of delta-normal VaR.
      Parameters:
      weights - position exposures (currency units), signed
      covariance - return covariance matrix, symmetric, n×n
      confidence - e.g. 0.99; z = Φ⁻¹(confidence)

    • incremental

      public static double incremental(double[] weights, double[][] covariance, double confidence, int i)
      Incremental VaR of position i: portfolio VaR now minus VaR with the position closed (weight zeroed). Positive means closing it REDUCES risk; negative means the position is a hedge and closing it raises VaR.