Abstracts Category : Other

Riders on the storm : a risk preference profiling on investors who rode the subprime financial storm

by Gonalo Saramago Fernandes

This dissertation estimates risk-neutral densities from 3-week contracts on the S&P 500 index inan attempt to characterize the underlying index in a risk-neutral environment through the statisticsderived from the implied distributions for two samples: pre-subprime-crisis and crisis. Thedistributions are estimated using mixture of lognormal densities, generalized beta distribution ofthe second kind and lognormal-polynomials. The mean values are similar in the three methodsemployed, along with the standard deviation. Moreover, the distributions tend to be negativelyskewed and leptokurtic for both samples.The constant relative risk aversion coefficient is estimated for both samples assuming the powerutility is well representative of investors behavior. The method employed was the mixture oflognormal distributions under both expected utility (EU) and rank-dependent utility assumptions(RDEU). The obtained coefficients for the pre-crisis sample were: 2,81 (EU) and 4,41 (RDEU)while in the crisis sample, the coefficients obtained were: 0,47 (EU) and -1,94 (RDEU). In linewith literature, by applying the real-world transformation (RDEU) to the mixture of lognormaldistribution estimated RND produced distributions with higher mean, lower standard deviation,less negatively skewed and with lower Kurtosis. Nesta dissertao, estimam-se risk-neutral densities a partir de derivados sobre o ndice S&P500 para dois perodos: pr-crise e crise. Com o objetivo de descrever o impacto da crise dosubprime no mercado americano. As distribuies so extradas usando trs mtodos diferentes:mistura de distribuies log-normais, distribuio beta generalizada do segundo tipo e lognormal-polinomiais. Da aplicao das trs metodologias obtm-se mdias e desvios-padrosemelhantes. As distribuies obtidas tendem a ter skewness negativa e um valor de kurtosissuperior a 3.O coeficiente de averso ao risco estimado para ambos os perodos assumindo que a funo deutilidade representativa a power utility. O mtodo utilizado foi a mistura de log-normaisassumindo Expected utility (EU) e rank-dependent expected utility (RDEU). Os coeficientesestimados foram de 2,81 (EU) e 4,41 (RDEU) para o perodo pr-crise. Para o perodo de criseos valores obtidos foram de 0,47 (EU) e -1,94 (RDEU). Finalmente, ao transformar as Riskneutraldensities em Real-world densities, obtem-se distribuies com mdia mais elevada,desvio-padro mais baixo, skewness menos negativa e kurtosis mais baixa, estando estasconcluses de acordo com o exposto na literatura.Advisors/Committee Members: Guedes, Jos Filipe Garcia Corra.