Haoran Wang

This paper studies how imperfect communication within firms shapes managerial expectations, input choices, and aggregate productivity. I develop a dynamic model of team production in which capital and labor managers share the firm’s objective but observe segmented information about productivity. The model yields a novel empirical test using expectation data: investment forecasts underreact to hiring when capital managers fail to incorporate labor managers’ private information. Using managerial forecast data of US public firms, I find that investment forecasts simultaneously underreact to hiring and overreact to investment, a pattern inconsistent with common-information models. Quantitatively, communication friction substantially weakens investment-hiring comovement and generates capital-labor-ratio dispersion equal to about 39% of its observed magnitude. The resulting MRPK and MRPN dispersions equal 27% and 19% of their observed counterparts, and the associated aggregate TFP loss is approximately 5% relative to the frictionless allocation. This productivity loss is limited relative to the marginal-product dispersion because segmented information induces a negative covariance between MRPK and MRPN.

I propose an associative-learning framework to study how firms innovate when they are uncertain about the data generating process of growth opportunities in technologies. Firms hold Gaussian-process beliefs about the payoffs over technological space and update them using market feedback to their patents. Because feedback in one field informs the firm about the values of technologically related fields, learning generates conditional path dependence: even holding accumulated technological knowledge fixed, different signal histories lead firms to pursue different innovation paths. Using patent-level market responses and citation-based measures of technological proximity for U.S. public firms, I document that firms direct subsequent patenting toward fields associated with more favorable signals, are more likely to enter new fields near higher-signal experience, and avoid entering the technological neighborhoods of fields with unfavorable signals. Embedding the learning mechanism in a quantitative endogenous-innovation environment calibrated to U.S. patent data, I show that field-specific market optimism can generate a persistent reallocation of innovation toward high-sentiment technologies. Most of this response comes from firms intensifying innovation in fields where they are already active, rather than from encouraging other firms entering those fields.