PIVOT: Bridging Black-Scholes Implied-Volatility and Price Objectives via Differentiable J\"ackel Operator
Researchers have developed PIVOT, a novel system designed to bridge the gap between price and implied volatility (IV) objectives in option-learning models. PIVOT utilizes a differentiable Jäckel operator to maintain the efficiency of existing solvers while providing a backward pass that handles the singularity in the low-vega regime. This approach significantly improves accuracy, as demonstrated by a 43.4% reduction in held-out price MAE and a 21.3% joint improvement in price and IV MAE on SPX OptionMetrics data. AI