Importance Sampling Enhanced with the COS Method for the Portfolio Risk Allocation

Battle Count: 76

The paper addresses credit portfolio risk measurement and allocation, which is relevant to credit risk management in trading books and regulatory capital. However, it is not directly about trading strategies, market microstructure, or asset pricing. The importance sampling and rare-event simulation techniques could be adapted for tail-risk estimation in trading portfolios, but the primary application is credit portfolio risk allocation.

Implementation requires: (1) multi-factor credit model with conditional independence structure, (2) closed-form conditional characteristic function computation, (3) filtered COS expansion with proper lattice handling and clipping, (4) weighted Gaussian and inverse-Gamma moment fitting, (5) covariance regularisation (ridge/shrinkage), (6) production importance sampling with factor and conditional-default likelihood ratios, (7) saddle-point optimization for conditional Bernoulli twisting. The homogeneous block grouping reduces computational cost but adds implementation complexity. Proper handling of discrete loss support, left-limit conventions, and second-moment diagnostics requires careful numerical work.

The paper provides detailed experimental design including specific seeds (42, 420, 421), exact parameter values, COS specifications (EXP-4 filter, mode counts), portfolio structure (Glasserman Example 4), and implementation details (batch sizes, ridge regularization, homogeneous block grouping). The benchmark portfolio is publicly available from Glasserman [4]. However, no code repository is explicitly linked, and timing results lack hardware/software metadata.