Price expectations drive traders' buy, hold, and sell decisions. They are often estimated by surveying investors; however, verbal accounts may differ from latent expectations. In this paper, we propose how to infer traders' price expectations from trading data instead. We assume traders' goal is maximizing final earnings by sequentially buying, holding, or selling shares. Due to sequentiality, trading is represented as a Partially Observable Markov Decision Process solved with Deep Reinforcement Learning. This model follows an approximately optimal trading policy with respect to price paths used in training. Meanwhile, we assume traders choose optimal trading actions given their price expectations. Therefore, price paths characterized by trend and volatility parameters are assumed to approximate expectations. We then infer which values of these parameters produce a human-like trading policy. While this approach achieves a good model fit with worst-performing traders in an empirical study, our results are more ambiguous for top traders.