ivanzoccolan/va-riders

VA Riders — Nested Stochastic Simulation with AAD Greeks

Simulation dataset for variable annuity (VA) guarantee pricing and Greek estimation, generated with ActuaLib and test data for Sobolev-trained neural network surrogates.

Contents

Each HDF5 file corresponds to one contract of a nested Monte Carlo simulation:

• fmv_per_path [n_outer] — fair market value per outer scenario
• delta [n_outer × 5] — equity Gr)
• rho [n_outer × 10] — interest rate Greeks (AAD, 10 yield-curve knots)
• Outer market state: spot prices,

Structure

| Split | Outer paths | Inner paths |-------|------------|-------------|----------|-------------| | test/{rule} | 100,000 | 1,000 | | | trainJ/{rule} | 20,000 | 100 | 180 | Coverage-first training tier |

{rule} ∈ {rop, ratchet, rollup} (benefit-base update rule). Each split contains 180 policies: 20 per rider × 9 riders.

Riders: GMAB,GMDB,GMDB_AB,GMDB_IB,GMDB_MB,GMDB_WB,GMIB,GMMB,GMWB

Benefit-base rules: Return of P(step-up), annual roll-up

Simulation Engine

ActuaLib — C++17, reverse-mode AAD,er model: Multivariate Black-Scholes+ Hull-White 1-factor short rate. Outer model: regime-switching GBM (equity) + AIRG 3-factor log-normal (rates, AAA ESWG mandate

RNG: Sobol quasi-random sequencrainJ 20260402 (independent to prevent data leakage).

Integrity

SHA-256 manifest at sha256_manifest.txt — verify with:

sha256sum --check sha256_manifest.txt

Companion Models

ivanzoccolan/va-surrogates