The existing literature assumes that unemployment insurance (UI) a↵ects the labor market through the job finding rate of eligible workers. I argue that this focus is too narrow. I show evidence for UI e↵ects through three other margins: (i) search externalities; (ii) takeup of other welfare state programs; and (iii) job separations. This suggests that the analysis of optimal UI should take a more comprehensive view of how UI a↵ects the labor market.Keywords: Unemployment insurance, extended benefits, unemployment duration, unemployment inflow, layo↵s, disability insurance, optimal benefits. JEL Codes: J65, J63, J64. * Address: University of Zurich, CH-8000 Zurich, Switzerland; email josef.zweimueller@econ.uzh.ch. This paper was written for the Adam Smith Lecture at EALE 2017, St.Gallen, Switzerland. I want to thank the organizers for inviting me to give this lecture and for the opportunity to present my work in front of this great audience. The research discussed here would not have happened without the energy and enthusiasm of my co-authors. First and foremost, I am indebted to my long-term collaborator Rafael Lalive, who provided invaluable input for this research. With him I started working on the universe of the Austrian social security data (ASSD) in the late 1990s, a data set that many other researchers are using now. Special thanks go to Stefan Staubli with whom I have enjoyed working over the last years and whose expertise and work attitude were of tremendous help in all our projects. The research discussed here benefitted from the energetic input of Camille Landais, who initiated the market externality project. I am indebted to Lukas Inderbitzin who co-authored the paper on program substitution. The paper on UI and job separations is work in progress and I have the pleasure to work with Simon Jäger and Benjamin Schoefer on this project. Over the years, I could rely on the support of engaged and talented research assistants. Particular thanks go to Andreas Kettemann and Philippe Ruh, who were extremely helpful over the last years and who provided decisive input for the empirical evidence presented here. Thanks also to Damian Osterwalder and Dominik Eglo↵ for putting together the tables and graphs of the paper. I also thank Andreas Steinhauer and Andreas Haller for comments on an earlier version of this paper and Johanna Posch for checking the final version.