Regression-based modeling of market option prices: With application to S&P500 options
Abstract
This paper presents a simple empirical approach to modeling and forecasting market option prices using localized option regressions (LOR). LOR projects market option prices over localized regions of their state space and is robust to assumptions regarding the underlying asset dynamics (e.g. log-normality) and volatility structure. Our empirical study using three years of daily S&P500 options shows that LOR yields smaller out-of-sample pricing errors (e.g. 32% one-day-out) relative to an efficient benchmark from the literature and produces option prices free of the volatility smile. In addition to being an efficient and robust option modeling and valuation tool for large option books, LOR provides a simple to implement empirical benchmark for evaluating more complex risk-neutral models.