Uncertain Volatility Model

Black-Scholes assumes you know the volatility of the underlying exactly, and that it stays constant. In reality nobody knows the future volatility, and it moves around. The Uncertain Volatility Model (UVM) refuses to pretend otherwise: instead of a single number, it takes a band — a minimum and maximum volatility the underlying could plausibly realize — and asks, over that whole band, what is the worst-case cost of hedging this option?

The mechanics work by choosing, at every point in the option's life, whichever volatility inside the band is most punishing for whoever is short the option: high volatility when the option's curvature (its gamma) is positive, low volatility when it is negative. This produces a nonlinear pricing equation — the Black-Scholes-Barenblatt equation — whose solution is not one price but an upper and lower bound: the most a rational seller should charge, and the least a rational buyer should pay, to be safe against any volatility path inside the assumed range.

A simple case makes the intuition concrete. Suppose a trading desk believes volatility on an underlying will sit somewhere between 15% and 25% over the next month, but does not know exactly where. A single long call has positive gamma everywhere, so its worst case for the seller is volatility pinned at the top of the band, 25%, for the whole period — the model prices it as if realized volatility were the maximum, giving a conservative (higher) premium than a naive Black-Scholes calculation using, say, the midpoint of 20%. A short position in the same option, by contrast, is worst-cased at the low end of the band, because low realized volatility hurts a different leg of a hedged book.

This matters most for portfolios with mixed long and short gamma, where no single constant volatility is conservative for every position simultaneously — the UVM's worst-case choice can flip from one option to the next inside the same book, which is exactly why banks use it for exotic-option risk limits rather than for quoting vanilla screen prices, where a single implied volatility is simpler and good enough.