A new paper by Julien Guyon and Léo Parent addresses a familiar source of model discomfort: parameters fitted to historical returns under the physical measure can be inconsistent with those needed to reproduce option prices under the risk-neutral measure. Their proposed mixed calibration works inside a path-dependent-volatility framework and tries to make the two views of the same process coexist rather than maintaining separate, quietly contradictory fits. Risk.net’s paper page, September 16
That is attractive because the division between P and Q is operationally porous. Forecasting, scenario generation and risk attribution draw on the historical world; pricing and hedging draw on the market-implied one. A joint fit may reduce the need for ad hoc reconciliation between them.
The paper’s title promises calibration robustness, not a free lunch. The tests worth looking for are stability through regime changes, the weights assigned to time-series and option-surface errors, and whether the mixed calibration improves hedging out of sample rather than merely producing a more harmonious in-sample score. The full technical detail is behind Risk.net’s access wall, so this is a paper to retrieve before forming a view on the implementation.