The fat-tailed model
Crash risk lives on the downside. This model keeps a Gaussian upside of scale a and gives the downside a heavy, polynomial tail whose shape is set by the circangle modulus k. The skew is not fitted in: it comes out of the model. Ξ, the sensitivity to the tail shape, is a Greek of its own.
Brent example: underlying 110, at-the-money, three months, rate 4%, upside 22%, k² = 0.5, halves joined, which gives an at-the-money implied volatility near 55%. Illustrative inputs, not live market data.
Central differences, in the book’s units: Vega per 1.00 of volatility, Theta per year, Rho per 1.00 of rate, Ξ per 1.00 of k². When the halves are joined, b follows a in Vega and k in Ξ.
The skew it produces
The tails, on a log scale
| Strike | Price | Implied vol | Delta | Gamma | Vega | Theta | Rho | Ξ |
|---|
The model
The downside comes from the circangle function sc: with u uniform on (−K(k), K(k)), x = sc(u, k) has polynomial tails, and at k = 0 it is the Cauchy distribution. Joined to a Gaussian upside, which keeps E[eX] finite, the log-return has density
with . The halves meet with the same height, and with they also meet with the same curvature, so the joined curve is smooth to third order and the model has two free parameters, a and k. Prices use ST = S0erTeX/E[eX]; the upside integrals are in closed form and the downside is integrated numerically.
The tail is heavy: with the halves joined at a = 20% over a year, the 5% quantile is a fall of about 91%. The sc downside suits far out-of-the-money puts, where crash risk is priced, more than a whole year’s returns. The derivation and full tables are in Circangle: π is Obsolete, Part 4.
This is a proof of concept. It has not been calibrated to market quotes or checked for arbitrage across a full volatility surface. Use of the model is acceptance that it is for educational purposes only. See the disclaimer.