Statistical Inference with Ambiguity

Topic:

Statistical Inference with Ambiguity

Time&Date: 

12:00-13:30pm, 2018/8/31

Venue:

Room A619, Teaching A

Speaker:

Prof. Xiaoquan (Michael) Zhang (Chinese University of Hong Kong)

Statistical inference is a fundamental tool for studying uncertain events. However, in social and economic settings, on top of outcome uncertainty, there often exists distribution uncertainty, or ambiguity. We develop and solve a model of ambiguity. We show that uncertainty can be decomposed into pure risk, pure ambiguity, and their interactions. Even when risk can be perfectly hedged, ambiguity still exists. With moderate levels of ambiguity, confidence levels can quickly drop, making classical statistical inference unreliable. Considering ambiguity can be more fruitful than obtaining a larger sample. Our results have implications on black swan events and fat-tailed distributions.