Pérez-Mendoza, Carlos Octavio Hedging targeted risks with reinforcement learning : application to life insurance contracts with embedded guarantees / Carlos Octavio Pérez-Mendoza and Frédéric Godin Sumario: The article presents a risk hedging framework based on deep reinforcement learning applied to life insurance contracts with embedded financial guarantees, such as variable annuities. The methodology relies on Shapley value decompositions to attribute the contributions of different risk sources to the contract's cash flows. Building on this decomposition, a targeted hedging strategy is developed that allows only selected risks to be mitigated while preserving exposure to the remaining ones. The model further incorporates a local risk penalization based on Conditional Value-at-Risk (CVaR) to enhance hedging stability. Numerical results demonstrate a significant improvement over traditional methods such as delta hedging and standard deep hedging En: Astin bulletin. - Belgium : ASTIN and AFIR Sections of the International Actuarial Association = ISSN 0515-0361. - 20/04/2026 Volumen 56 Número 2 - abril 2026 , 27 p. 1. Cálculo actuarial . 2. Seguros de vida riesgo . 3. Gerencia de riesgos . 4. Renta vitalicia . 5. Inteligencia artificial . 6. Simulación Monte Carlo . I. Godin, Fréderic . II. International Actuarial Association . III. Título.