[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] In this study, I examine effects of financial incentives and non-financial incentives on public school teachers' retirement in Missouri. For financial incentives, existing studies commonly model the retirement in two types, static reduced-form and structural-form. There are two main types of the reduced-form models, linear probability model (e.g., Costrell and McGee, 2010; Brown, 2013), probit or logit model (e.g., Furgeson et al., 2006; Asch et al., 2005; Coile and Gruber, 2007). A prominent structural form model is the Stock and Wise (1990) option value model. In addition, some researchers (Brown and Laschever, 2012; Gustman and Steinmeier, 2000; Dwyer and Mitchell, 1999) suggest that non-financial incentives, such as colleagues, health status, and family, are also important factors to explain retirement decisions. Among these non-financial incentives, peer effects are the focus of my study. The first chapter analyzes peer effects on public school teachers' retirement timing decisions. I use an administrative dataset of the full population of late career Missouri public school teachers during academic years 1994-2007 to construct a model incorporating both financial incentives and non-financial incentives when teachers make retirement decisions. My study mainly differs from previous literature in the controls of financial incentives and the use of data. For controls of financial incentives, instead of using pension wealth and peak value, I propose the simulated "financial benefit of postponing retirement" based on Stock-Wise (hereafter, SW) option value model as a new financial incentive proxy. Besides, I use a much richer dataset so that more detailed information about teachers are available and the schools are more diverse in poverty and urbanicity. Empirically, when the retirement rate of peers in previous year increases 1%, the probability of retirement increases approximately 0.128 percentage points. In addition, I find evidence of heterogeneous peer effects. For example, teachers with different education levels respond to peers retirement differently. Because the reduced-form parameters generally depend on the pension rules, the reduced-form models are unreliable to predict the effects of changes in the pension rules. In contrast, the structural model parameters are independent of pension rules, which makes them useful for policy analysis. The second chapter uses an economic structural model to fit late career Missouri public school teachers' data and study the effects of changing pension rules on the timing of retirement. Ni and Podgursky (2016) show that the SW option value model produces good fit for the Missouri public school teachers during 2002-2008. There are three limitations in their study and other applications of the SW model. First, the SW option value model has only been applied to retirement data when the pension rules are fixed. Second, there is a selection bias in the sample of senior teachers: among the retirement-eligible teachers only the "stayers" are in the sample, while the "early leavers" are absent. This bias has not been adjusted in the previous studies. Third, the SW model is estimated based on the likelihood of panel data of individuals (teachers or salesmen). The cost of computing the likelihood increases in the number of teachers and the length of the sample period. For a larger state, such as Texas, and/or a longer sample period, the computation cost may be prohibitively high. The second chapter solves these problems. First, I model expectations of pension rule changes. Second, I adjust the sample selection bias by using distribution of initial preference errors. Third, I group individual teachers into a fixed number of (age, experience) cells to reduce the time of evaluating the likelihood. I estimate the model under different expectation assumptions and compare their performance by simulation. The estimated SW model exhibits good in-sample and out-of-sample fits. Counterfactual analysis shows that pension enhancements reduce the retirement age by around 0.4 years for the 1994 cohort teachers and by more than one year for teachers in the steady state. Finally, I compare SW option value model with an alternative widely applied structural model of retirement, dynamic programming model. Compared to dynamic programming model, the SW model tends to predict early retirement.
Modeling time-variation over the business cycle (1960–2017): an international perspective
