Abstract:This paper applies classic linear inverse theory to the estimation of the implied risk neutral probability density function (PDF) from option prices. To overcome non-uniqueness and instability inherent in the option inverse problem, smoothness requirement for the shape of a PDF and an initial model are introduced by a penalty function. Positivity constraints are included as a hard bond on the PDF values. Then the option inverse problem becomes a non-negative least-squares problem which can be solved by the classic methods such as the non-negative least squares program of Lawson and Hanson (1974). The best solution is not the solution that gives best fit, but the solution that gives the optimal trade-off between the goodness of fit and smoothness of the estimated risk natural PDF. The proposed inversion technique is compared to the models of Black-Scholes (BS), a mixture of two lognormals (MLN), Jarrow and Rudd's Edgeworth expansion (JR), and jump diffusion (JD) for the estimation of the PDF from the option prices associated with the September 2007 NYMEX natural gas futures. It is found that the inversion technique not only gives best goodness of fit, but also the significantly better model resolution. BS, JD and MLN models basically cannot resolve the densities far away from the strikes where option prices are observed and can resolve long wavelength features of the densities inside the strikes where option prices are observed. On the other hand, the inversion model can resolve not only the significant details of the densities inside the strikes where option prices are observed, but also the long wavelength features of the densities away from the strikes where option prices are observed. The empirical study for the last three months of the September 2007 futures contract shows that the shapes of the estimated PDFs become more symmetric as the futures contract is closer to the expiration date. The dispersion of the estimated PDFs decreases with decreasing the time to expire, indicating the resolution of uncertainty with time.

Recovering Risk-Neutral Probability Density Functions Using Gaussian Mixture Distributions

Recovering implied minimum distance risk-neutral probability measure using GMD

Risk-neutral density extraction from option prices: improved pricing with mixture density networks

Inversion of Option Prices for Implied Risk Neutral Probability Density Functions: General Theory and Its Applications to the Natural Gas Market

Inversion of Option Prices for Implied Risk-Neutral Probability Density Functions: General Theory and Its Applications to the Natural Gas Market

Learning Probability Density Functions from Marginal Distributions with Applications to Gaussian Mixtures

Matching distributions: Recovery of implied physical densities from option prices

Recovering Implied Risk-Neutral Probability Density Function Using SVR

Risk-Neutral Generative Networks

Probability Weighting Functions Obtained from Hong Kong Index Option Market

Differentiability and Approximation of Probability Functions under Gaussian Mixture Models: A Bayesian Approach

Gaussian mixture Taylor approximations of risk measures constrained by PDEs with Gaussian random field inputs

Parsimonious Generative Machine Learning for Non-Gaussian Tail Modeling and Risk-Neutral Distribution Extraction

Biology-inspired geometric representation of probability and applications to completion and options' pricing

Numerical characteristics and parameter estimation of finite mixed generalized normal distribution

Inverse Gaussian quadrature and finite normal-mixture approximation of the generalized hyperbolic distribution

A Generalized Finite Mixture Model for Asymmetric Probability Distribution Observations

Value at Risk and Expected Shortfall for Normal Variance Mean Mixtures of Finite Weighted Inverse Gaussian Distributions

A Specialized Probability Density Function for the Input of Mixture of Gaussian Processes.

Pricing Cdo Tranches with Stochastic Correlation and Random Factor Loadings in A Mixture Copula Model

An Option Pricing Model with Probability Measure Ambiguity