Directedness in Search

We present a tractable, hybrid framework that nests random and perfectly directed search, in which workers are more likely to direct their search toward submarkets with higher returns, while still searching in inferior submarkets with positive probability. The choice of submarket is governed by a logit choice model with noise parameter μ ∈ [0,∞). In the respective limits, search becomes either completely random or perfectly directed. We characterize the model equilibrium and show that even the perfectly directed search limit is inefficient, in contrast to its otherwise close cousin, competitive search. We proceed to quantify the extent of directedness on Danish matched employer-employee data. Identification relies on the insight that the two benchmark models differ qualitatively in their implications for job-to-job worker reallocation. We find evidence of substantial directedness in search. Finally, we study the implications for underinvestment due to holdup problems and show that the observed degree of directedness substantially reduces underinvestment relative to a setting with random search.