Duration Models - Centro de Estudios Monetarios y Financieros持续时间模型，货币与金融研究中心.pdf

Duration Models Class Notes Manuel Arellano November 1989 Revised: December 3, 2008 1 Duration data Duration data appear in a diversity of situations in economics, but I will often refer to unemployment duration to provide a focus for the presentation of the material. Suppose we select a random sample of N persons entering unemployment and wait until they find jobs. Then we record the number of weeks each person has been unemployed: t , t , ..., t . We have 1 2 N observations on the duration of a spell of unemployment for each of the individuals in the sample. In practice we are more likely to have censored duration data. That is, for some individuals we do not observe ti , possibly because they have not found a job at the time of the interview, so that all we know is that ti t for some particular t, or they have found a job between selection and interview and all we know is that ti lies within a certain interval t ti t. There are enormous variations in the duration of spells of unemployment from one individual to the next (from a few weeks to five years or more) and it is important to model the causes for these differences. In particular it is important to know how the re-employment probability changes over the period of the spell and what is the impact of the level of unemployment benefits on these probabilities. More generally, duration data measure how long individuals remain in a certain state. Other examples include: job turnover, marital instability, time to transactions in stock marke