Exploratory Mean-Variance with Jumps: An Equilibrium Approach

Battle Count: 136

Highly relevant to quantitative trading. The paper directly addresses continuous-time portfolio optimization with realistic market dynamics (jump-diffusion), provides an implementable RL algorithm that learns without knowing market parameters, and demonstrates practical profitability on 24 years of real S&P500 data (13/14 tests profitable). The equilibrium policy framework addresses time-inconsistency, a critical practical issue for investors. The Actor-Critic algorithm with only 3 parameters is computationally efficient and interpretable.

Moderate to high complexity. Requires understanding of stochastic calculus (Itô's formula for jump-diffusions), time-inconsistent control theory, EHJB equations, and RL algorithms. However, the final RL model is remarkably simple with only 3 parameters and a straightforward gradient update rule (Algorithm 1). The main complexity lies in the theoretical derivation and the discretization of the continuous-time SDE for simulation. The jump-diffusion simulation and MLE fitting add additional implementation steps.

The paper provides detailed mathematical derivations, explicit algorithm (Algorithm 1), parameter tables, and simulation settings. However, no code repository is mentioned. The methodology relies on publicly available S&P500 and Treasury Bill data from Yahoo Finance. The analytical solutions are fully specified, enabling theoretical reproduction, but the RL training details (learning rates, number of epochs beyond 550) could be more precisely specified.