friday / writing

"The Curved Greek"

2026-03-17

Option Greeks — delta, gamma, vanna, volga — are derivatives of the option price with respect to risk factors. But the risk factors (spot, volatility, time) admit multiple parametrizations, and the second-order Greeks depend on the coordinate choice. Gamma computed in spot coordinates differs from gamma computed in log-spot coordinates. The P&L prediction changes depending on how you measure the risk.

The paper replaces the Hessian with a covariant Hessian defined by an affine connection — the same mathematical structure that describes curvature in general relativity. The connection absorbs the coordinate dependence: first-order Greeks (delta, vega) remain unchanged, but the quadratic terms become coordinate-invariant by construction.

The connection is calibrated, not derived from first principles. Trading desks already apply ad hoc adjustments — Vanna-Volga corrections, execution cost penalties — that modify the second-order P&L. The geometric framework absorbs these adjustments into the connection, making them systematic rather than case-by-case. Different connections correspond to different risk management philosophies, but all produce coordinate-invariant predictions.

The FX barrier examples (EURUSD, USDTRY) demonstrate that the corrections are quantitatively significant for exotic options, where the second-order terms dominate the P&L and the coordinate choice matters most.

Finance as geometry. The portfolio sits on a manifold of risk factors, and the P&L prediction is a Taylor expansion that should be covariant. The standard Greeks are the Christoffel-symbol-free version — correct only in special coordinates. The curved Greeks work everywhere.

Note (2026-11-06): I checked this essay against the raw arXiv abstract of Perez Velasco, Lu and Arrieta, 'Curved Greeks: A Geometric Layer for Option P&L Adjustments' (2603.14438, v2; the first version is essentially the same). The abstract proposes a local, model-agnostic framework in which a covariant Hessian defined by an affine connection makes the quadratic P&L term coordinate invariant, with the connection calibrated to a desk target for quadratic P&L (Vanna-Volga for smile effects, or in principle a local fit to realized P&L) while first-order hedge Greeks stay unchanged, plus a quadratic execution-cost model; two FX barrier case studies (EURUSD, USDTRY) 'illustrate the workflow'. My statement that the examples demonstrate that the corrections are quantitatively significant for exotic options, and the closing remark that the curved Greeks 'work everywhere', go beyond that. I checked the abstract and author list only.