Adjusting Beliefs via Transformed Fuzzy Priors

Abstract

© Published under licence by IOP Publishing Ltd. Instead of leaving a decision to a pure data-driven system, intervention and collaboration by human would be preferred to fill the gap that machine cannot perform well. In financial applications, for instance, the inference and prediction during structural changes by critical factors; such as market conditions, administrative styles, political policies, etc.; have significant influences to investment strategies. With the conditions differing from the past, we believe that the decision should not be made by only the historical data but also with human estimation. In this study, the updating process by data fusion between expert opinions and statistical observations is thus proposed. The expert's linguistic terms can be translated into mathematical expressions by the predefined fuzzy numbers and utilized as the initial knowledge for Bayesian statistical framework via the possibility-to-probability transformation. The artificial samples on five scenarios were tested in the univariate problem to demonstrate the methodology. The results showed the shifts and variations appeared on the parameters of the distributions and, as a consequence, adjust the degrees of belief accordingly.