implementation-shortfall

Execution research platform: Maslov LOB simulator, HMM/KMeans regime detection, VWAP/POV scheduling, and a paired-significance harness with shuffled-label negative controls. Finds no measurable benefit from regime conditioning on this data.

GPU-batched reinforcement learning for limit-order-book optimal execution with IPPO, trained against Bybit BTCUSDT L2 order-book data.

MBA dissertation: can reinforcement learning reduce implementation shortfall? PPO and DQN against TWAP, VWAP and Almgren-Chriss in a literature-calibrated multi-venue simulator, with Newey-West HAC t-statistics. No proprietary or real tick data.