Calibration risk under probabilistic parameter dependencies and model output effects

We propose a novel regression-based framework to model calibration risk in asset price models. Traditionally, models are calibrated to liquid contract quotes by minimizing error. Calibration risk relates to uncertainty in parameter estimates (inputs), which is transferred to other contracts (outputs), but is often overlooked. Our probabilistic modelling of parameter estimates and their dependencies allows systematic detection and alleviation of this and its effects on ultimate outputs. Studying the global sensitivity of contracts’ values to model parameters enables us to rank them by influence of their knowledge on maximizing the increase in likelihood of profit or loss of an investor’s position.