The job-finding rate of Unemployment Insurance (UI) recipients declines in the initial months of unemployment and then exhibits a spike at the benefit exhaustion point. A range of theoretical explanations have been proposed, but those are hard to disentangle using data on job finding alone. To better understand the underlying mechanisms, we conducted a large text-message-based survey of unemployed workers in Germany. We surveyed 6,349 UI recipients twice a week for four months about their job search effort. The panel structure allows us to observe how search effort evolves within individuals over the unemployment spell. We provide three key facts: 1) search effort is flat early on in the UI spell, 2) search effort exhibits an increase up to UI exhaustion and a decrease thereafter, 3) UI recipients do not appear to time job start dates to coincide with the UI exhaustion point. A standard search model with unobserved heterogeneity struggles to explain the second fact, and a model of storable offers is not consistent with the third fact. The patterns are well captured by a model of reference-dependent job search or by a model with duration dependence in search cost.

中文翻译：

---

来自德国失业工人调查的求职模型证据*

失业保险 (UI) 领取者的求职率在失业的最初几个月下降，然后在福利用尽点出现飙升。已经提出了一系列理论解释，但仅使用与找工作有关的数据很难解开这些解释。为了更好地了解潜在机制，我们对德国的失业工人进行了一项基于短信的大型调查。我们在四个月内每周两次对 6,349 名 UI 接受者进行了关于他们求职努力的调查。小组结构使我们能够观察在失业期间个人内部的搜索努力如何演变。我们提供了三个关键事实：1) 在 UI 拼写的早期搜索工作量是持平的，2) 搜索工作量在 UI 耗尽之前会增加，然后会减少，3) UI 接收者似乎不会将工作开始日期与 UI 耗尽点一致。具有未观察到的异质性的标准搜索模型难以解释第二个事实，而可存储报价模型与第三个事实不一致。这些模式可以通过依赖于参考的工作搜索模型或搜索成本中具有持续时间依赖性的模型来很好地捕获。