In this talk, I will discuss a framework for predictive models based on interpretations. Interpretations take two forms.
They can be projections onto a subset of variables or they can be clustered groupings of attributes. Projection interpretations
with non overlapping support produce negatively correlated signals. Thus, the interpretation framework that I construct
calls into question one of the standard assumptions in signaling models -- namely independence based on the state of the world.
I will also discuss the implications of these results for auctions and voting.
|