Overcoming Execution Drag in Volatility Harvesting: Hysteresis and Partial PCA Orthogonalization

Abstract

Retail quantitative algorithms frequently fail in live markets because alpha in statistical arbitrage decays at the speed of light. This paper bridges the divide between theoretical volatility modeling and the physics of market microstructure by documenting the architectural evolutionofa high-frequency trading engine on the Nifty50. In the initial baseline, we construct a Volatility Risk Premium (VRP) extraction model utilizing a decoupled EGARCH(1,1) volatility forecaster, an agile rolling-window Gaussian Hidden Markov Model (GHMM) for regime detection, and a bounded discrete-time Kalman Filter for dynamic delta hedging. While this ensemble successfully isolates pricing inefficiencies to generate a 17.85% gross return and a1.84 Sharpe ratio, we empirically demonstrate that continuous delta-hedging friction and execution slippage violently invert this into a-18.50%terminal net loss.To salvage the theoretical edge, we implement a beta hysteresis deadband to throttle rebalancing drag and deploy Partial Principal Component Analysis (PCA) to orthogonalize the index. By specifically neutralizing the first principal component (PC1), we eliminate systemic macro-directional shocks while preserving tradableidiosyncratic volatility. This structural sterilization successfully reverses the microstructure decay, yielding a friction-aware 11.44% net return and a 2.03 Sharpe ratio.Finally, we extend the architecture to isolate the pure Correlation Risk Premium (CRP), formalized via a continuous-time Ito calculus framework. We resolve the "PCA Paradox" by introducing a Dual-Horizon Architecture utilizing daily spectral decomposition to isolate top-quintile, high Idiosyncratic Variance Ratio (IVR) constituents for sparse options dispersion (enforcing index weight constraints), while simultaneously deploying intraday Euclidean distance matching on PCA factor loadings to execute factor-neutral statistical arbitrage. The resulting framework proves that institutional-grade structural alpha can be successfully harvestedwhen mathematical models are explicitly constrained by execution latency