Surface Calibration Practice

Calibration is where theory meets noisy data. A model is fitted by minimizing a weighted sum of squared pricing errors against quoted volatilities, and every choice carries weight: vega or bid-ask weighting respects where liquidity actually sits, penalty terms pull parameters toward stability, and bounds keep the optimizer in economically meaningful territory. A fit that nails today’s quotes with unstable parameters is worthless, because tomorrow’s recalibration jumps and exotic prices whipsaw. Sound practice combines good starting values, multi-start search, regularization, and out-of-sample checks across days — the same craft whether you are fitting SABR slices, a Heston set, or a full SVI surface.