The experimental ‘learning-to-forecast’ literature finds that subjects use simple linear backward-looking models when forecasting in environments with little to no inout the economic framework. We study the formation of expectations in a laboratory economy of monopolistic firms and labour unions with almost complete knowledge of the model. We observe simple backward-looking rules, but also a considerable share of model-based expectations using information on the economic structure. At least for some subjects, expectations are informed by theory. As in the previous literature, we find individual prediction rules to be heterogeneous.