[pt] Na tentativa de melhor capturar fatos estilizados do comportamento dos preços de opções financeiras, em especial para tratar a questão do sorriso da volatilidade, modelos de volatilidade estocástica têm sido objeto de estudo em diversos mercados. Neste contexto, o principal objetivo deste trabalho é avaliar os modelos de volatilidade estocástica de Heston (1993), Bates (1996) e Double Heston (2009) junto ao método de Lewis (2000) para precificar opções de ações no mercado brasileiro de derivativos, caracterizados por serem de curto prazo. Para isto foram precificadas opções de compra da Petrobrás e Vale. Os modelos foram comparados de acordo com a qualidade do ajuste aos dados in-sample e a capacidade preditiva com dados out-of-sample. Ademais, buscou-se verificar a volatilidade implícita gerada por cada um dos modelos. Ao fim, identificou-se que considerar a volatilidade como estocástica, mesmo quando é descrita por apenas um processo estocástico, é a decisão mais importante a ser tomada a fim de melhorar o apreçamento das opções. Além disso, adicionar saltos a um modelo de volatilidade estocástica parece ser mais relevante do que adicionar um segundo processo estocástico para modelar a volatilidade na precificação de opções de curto prazo. / [en] In an attempt to better capture stylized facts about financial option prices behavior, especially to address the issue of volatility smile, stochastic volatility models have been the object of study in several markets. In this context, the main purpose of this work is to assess the stochastic volatility models of Heston (1993), Bates (1996) and Double Heston (2009) along with the Lewis method (2000) for stock option pricing in Brazilian derivative market, featured by being short-term. Therefore, Petrobrás and Vale s call options were priced. The models were compared according to the in-sample fit skill and the out-of-sample forecasting power. Furthermore, it was verified the implied volatility begot by each model. In the end, it was figured out that consider the volatility as stochastic even when it is described by only one stochastic process is the preeminent matter to do in order to improve option pricing. Plus, adding jumps in a stochastic volatility model seems to be more important than adding a second stochastic process to model the volatility in short-term option pricing.

Black and Scholes have proposed a model for pricing European options where the underlying asset follows a so-called geometric Brownian motion which assumes constant volatility. The proposed Black-Scholes model has an exact solution. However, it has been shown that such an assumption of constant volatility is not realistic, and numerous extensions have been developed. In addition, models usually do not have a closed-form solution which makes pricing a challenging task. The thesis focuses on pricing Bermudan options under two stochastic volatility Heston-type models using an Almost-Exact scheme for simulation. Namely, we focus on deriving the Almost-Exact scheme for Heston and Double Heston model and numerically study the behaviour of the scheme. We show that the AES works well when the number of simulated steps is equal to the number of exercise dates which makes it efficient.