Endogenous Learning in Multi-Sector Economies

Consider a multi-sector general equilibrium model where firms have incomplete information about the returns to scale of their production and where that information is sequentially updated once real production is observed. What is the impact of these learning dynamics on the market-wise equilibrium objects? Under which conditions are firms able to efficiently learn their actual returns to scale? At which rate does this learning happen? In this work, we analyze endogenous learning mechanisms and their implications for the market-wise equilibrium objects in the multi-sector model. Our results shed light on how idiosyncratic shocks translate into the learning dynamics of firms returns to scale. Particularly, we uncover the advantages and disadvantages of the maximum a-posteriori estimation as a learning approach and we observe that all the relevant information in the learning dynamics is encoded in the input decisions and the manner in which input decisions are taken. We deduce conditions under which firms are able to learn the actual returns to scale. Using the notion of centrality in the multi-sector network, we uncover a price mechanism which is consistent not only with the correct knowledge of the returns to scale, but also with any converging sequence of belief on the returns to scale. On the empirical side, the proposed analysis of the endogenous learning dynamics is complemented with a statistical approach that allows testing the presence and level of learning using available input-output data. The empirical figures reveal the presence of sizable learning processes (driven by underestimations and overestimations of the returns to scale parameters) in different sectors.