GENERATIVE ADVERSARIAL NETWORKS FOR GASOLINE CRACK SPREAD RISK ANALYSIS

Managing risks associated with commodities is crucial to ensure that business operations lead
to favorable financial results and reduce the risk of short-term financeability problems. To
achieve this, it is necessary to create scenarios for commodity prices that accurately reflect their
probability distributions. This paper presents an implementation of the well-established
TimeGAN architecture for generating multiple scenarios of gasoline crack spread, with the
objective of supporting risk management and business decisions. This approach offers a
complementary approach to traditional stochastic models based on Stochastic Differential
Equations for time series simulation and risk analysis. It leverages the powerful capabilities of
Generative Adversarial Networks (GANs) to produce realistic scenarios, particularly in
capturing complex probabilistic distributions without needing any assumptions about the data
distribution. By accurately modeling the probabilistic distribution of critical risk factors, the
GAN framework enables more reliable estimation of their potential impact on business
performance, making it a robust tool for financial risk assessment.