Introduction to dynamic belief elicitation

We present a protocol for eliciting dynamic beliefs from forecasters. At time t = 0, forecasters hold beliefs about a random variable of interest that will realize publicly at time t = 1. Between t = 0 and t = 1, forecasters observe private information that impacts their beliefs. We design a class of protocols that, at the outset, elicit forecasters' beliefs about the random variable and elicit beliefs about any private information they expect to receive over time, and then elicit the private information that forecasters receive as they receive it. We show that any alternative elicitation mechanism can be approximated by protocols in our class. The information elicited can be used to solve optimally an arbitrary dynamic decision problem.