| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 980.17 KB | Adobe PDF |
Orientador(es)
Resumo(s)
O reconhecimento e a quantificação da incerteza na determinação do preço das opções representam um desafio critico na área das finanças que tem sido largamente subexplorado. O presente estudo propõe-se quantificar a incerteza na determinação do preço das opções recorrendo à metodologia de previsão conformalizada, procurando assim colmatar uma lacuna existente na literatura. A previsão conformalizada é uma técnica que permite construir intervalos de previsão com cobertura válida em amostras finitas, sem fazer suposições sobre a distribuição da amostra. Recorrendo a uma vasta base de dados de opções no S&P500, realizamos um estudo empírico para avaliar o desempenho de intervalos de previsão conformalizados para gradient boosting machines. Os resultados empíricos indicam que os intervalos atingem uma cobertura empírica igual à cobertura nominal, o que não é assegurado em metodologias não conformes. Além disso, observamos variações sistemáticas na largura dos intervalos consoante as características das opções. Nomeadamente, as opções out-of-the-money e as opções com um curto período até à maturidade têm intervalos de previsão relativamente mais largos, o que sugere uma maior incerteza na determinação do preço. Observamos também que as calls com curto período até à maturidade têm intervalos mais largos do que as puts com curto período até à maturidade, devido às diferenças inerentes de retorno. Em suma, os resultados validam a utilização da previsão conformalizada no âmbito da previsão de preços de opções financeiras e salientam o seu valor prático para a tomada de decisões financeiras em condições de incerteza.
The acknowledgement and quantification of uncertainty in option pricing represents a critical challenge in the field of finance that has been largely underexplored. The present study proposes to quantify the uncertainty of option pricing by using conformal prediction, thus aiming to fill a gap in existent literature. Conformal prediction is a technique for constructing prediction intervals with valid coverage in finite samples without making distributional assumptions. Using a large dataset of call and put options on the S&P500, we conduct an empirical study to evaluate the performance of conformal prediction intervals for gradient boosting machines. The empirical results indicate that the prediction intervals reach an empirical coverage equal to the nominal target, which is not observed in non-conformal methodologies. Furthermore, we observe systematic variations in the width of the intervals across option characteristics. Notably, out-of-the-money options and options with a short time-to-maturity have relatively wider prediction intervals, suggesting higher pricing uncertainty. We also observe that short-term put options have wider intervals than short-term call options, due to their inherent payoff differences. Overall, the findings validate the use of conformal prediction in the field of option pricing and highlight its practical value for financial decision-making under uncertainty.
The acknowledgement and quantification of uncertainty in option pricing represents a critical challenge in the field of finance that has been largely underexplored. The present study proposes to quantify the uncertainty of option pricing by using conformal prediction, thus aiming to fill a gap in existent literature. Conformal prediction is a technique for constructing prediction intervals with valid coverage in finite samples without making distributional assumptions. Using a large dataset of call and put options on the S&P500, we conduct an empirical study to evaluate the performance of conformal prediction intervals for gradient boosting machines. The empirical results indicate that the prediction intervals reach an empirical coverage equal to the nominal target, which is not observed in non-conformal methodologies. Furthermore, we observe systematic variations in the width of the intervals across option characteristics. Notably, out-of-the-money options and options with a short time-to-maturity have relatively wider prediction intervals, suggesting higher pricing uncertainty. We also observe that short-term put options have wider intervals than short-term call options, due to their inherent payoff differences. Overall, the findings validate the use of conformal prediction in the field of option pricing and highlight its practical value for financial decision-making under uncertainty.
Descrição
Trabalho Final de Mestrado, Finance, ISEG, 2025
Palavras-chave
Conformal Prediction Option Pricing Machine Learning Quantile Regression LightGBM Previsão Conformalizada Precificação de opções Machine Learning Regressão de Quantis LightGBM
