The Black-Scholes option pricing model has been the foundation of option pricing analysis. Yet as well known as the model itself, its empirical deficiencies are also well documented. Option prices generated by the Black-Scholes formula are often found to systematically differ from observed prices. The patterns of mispricing are generally believed to result from violations of one or more assumptions underlying the Black-Scholes option pricing model, such as the natural logarithm of the underlying stock price following a normal distribution with a variance that increases exactly linearly with time. This dissertation concerns an evaluation of the Principle of Maximum Entropy as a method for recovering a probability density function from stock index option prices. Theoretically, the resulting probability density is "the least prejudiced estimate since it is maximally noncommittal with respect to missing or unknown information." Empirically, this dissertation demonstrates that entropy valuation gives much stronger performance than does the Black-Scholes model in pricing stock index options on the S & P 500 and on the Dow Jones Industrial Average.

Option price contains information on the distribution of the underlying asset. Under insufficient condition we employ the maximum entropy principle to estimate the probability density of the asset price. The problem is equivalent to finding the Lagrange multipliers of a linear functional defined by entropy and payoff functions. Buchen and Kelly proved that the maximum entropy distribution recovered from observed option prices is quite similar with the original asset distribution. In this article we apply a similar method to recover the probability density function of an asset from given option prices for binary options and European options.

This paper suggests a new method of implementing the principle of maximum entropy to retrieve the risk neutral density of future stock, or any other asset, returns from European call and put prices. Instead of options prices used by previous studies the method maximizes the entropy measure subject to values of the risk neutral moments. These moments can be retrieved from market option prices in a first step, at each point of time. Compared to other existing methods of retrieving the risk neutral density based on the principle of maximum entropy, the benefits of the method that the paper suggests is the use of all the available information provided by the market more sufficiently. To evaluate the performance of the suggested method, the paper compares the new method proposed to other risk neutral density estimation techniques based on a number of simulation and empirical exercises.

Entropy pricing applies notions of information theory to derive the theoretical value of options. This paper employs the maximum entropy formulation of option pricing, given risk-neutral moment constraints computed directly from the observed prices. First, higher-order moments are used to generate option prices. Then a generalization of Shannon entropy, known as Renyi entropy, is studied to account for extreme events. This maximum entropy problem provides a class of heavy-tailed distributions. Examples and Monte Carlo simulations are provided to examine the effects of moment constraints on option prices. The call option values are then constructed using daily S&P 500 index options. The findings suggest that entropy pricing with higher-order moment constraints provides higher forecasting accuracy.

This paper describes the application of the Principle of Maximum Entropy to the estimation of the distribution of an underlying asset from a set of option prices. The resulting distribution is least committal with respect to unknown or missing information and is hence the least prejudiced. The maximum entropy distribution is the only information about the asset that can be inferred from the price data alone. An extension to the Principle of Minimum Cross-Entropy allows the inclusion of prior knowledge of the asset distribution. We show that the maximum entropy distribution is able to accurately fit a known density, given simulated option prices at different strikes.